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 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp);
  69
  70 const nfs4_stateid zero_stateid = {
  71         { .data = { 0 } },
  72         .type = NFS4_SPECIAL_STATEID_TYPE,
  73 };
  74 static DEFINE_MUTEX(nfs_clid_init_mutex);
  75
  76 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  77 {
  78         struct nfs4_setclientid_res clid = {
  79                 .clientid = clp->cl_clientid,
  80                 .confirm = clp->cl_confirm,
  81         };
  82         unsigned short port;
  83         int status;
  84         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  85
  86         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  87                 goto do_confirm;
  88         port = nn->nfs_callback_tcpport;
  89         if (clp->cl_addr.ss_family == AF_INET6)
  90                 port = nn->nfs_callback_tcpport6;
  91
  92         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  93         if (status != 0)
  94                 goto out;
  95         clp->cl_clientid = clid.clientid;
  96         clp->cl_confirm = clid.confirm;
  97         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98 do_confirm:
  99         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 100         if (status != 0)
 101                 goto out;
 102         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 103         nfs4_schedule_state_renewal(clp);
 104 out:
 105         return status;
 106 }
 107
 108 /**
 109  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 110  *
 111  * @clp: nfs_client under test
 112  * @result: OUT: found nfs_client, or clp
 113  * @cred: credential to use for trunking test
 114  *
 115  * Returns zero, a negative errno, or a negative NFS4ERR status.
 116  * If zero is returned, an nfs_client pointer is planted in
 117  * "result".
 118  *
 119  * Note: The returned client may not yet be marked ready.
 120  */
 121 int nfs40_discover_server_trunking(struct nfs_client *clp,
 122                                    struct nfs_client **result,
 123                                    struct rpc_cred *cred)
 124 {
 125         struct nfs4_setclientid_res clid = {
 126                 .clientid = clp->cl_clientid,
 127                 .confirm = clp->cl_confirm,
 128         };
 129         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 130         unsigned short port;
 131         int status;
 132
 133         port = nn->nfs_callback_tcpport;
 134         if (clp->cl_addr.ss_family == AF_INET6)
 135                 port = nn->nfs_callback_tcpport6;
 136
 137         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 138         if (status != 0)
 139                 goto out;
 140         clp->cl_clientid = clid.clientid;
 141         clp->cl_confirm = clid.confirm;
 142
 143         status = nfs40_walk_client_list(clp, result, cred);
 144         if (status == 0) {
 145                 /* Sustain the lease, even if it's empty.  If the clientid4
 146                  * goes stale it's of no use for trunking discovery. */
 147                 nfs4_schedule_state_renewal(*result);
 148
 149                 /* If the client state need to recover, do it. */
 150                 if (clp->cl_state)
 151                         nfs4_schedule_state_manager(clp);
 152         }
 153 out:
 154         return status;
 155 }
 156
 157 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 158 {
 159         struct rpc_cred *cred = NULL;
 160
 161         if (clp->cl_machine_cred != NULL)
 162                 cred = get_rpccred(clp->cl_machine_cred);
 163         return cred;
 164 }
 165
 166 static void nfs4_root_machine_cred(struct nfs_client *clp)
 167 {
 168         struct rpc_cred *cred, *new;
 169
 170         new = rpc_lookup_machine_cred(NULL);
 171         spin_lock(&clp->cl_lock);
 172         cred = clp->cl_machine_cred;
 173         clp->cl_machine_cred = new;
 174         spin_unlock(&clp->cl_lock);
 175         if (cred != NULL)
 176                 put_rpccred(cred);
 177 }
 178
 179 static struct rpc_cred *
 180 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 181 {
 182         struct rpc_cred *cred = NULL;
 183         struct nfs4_state_owner *sp;
 184         struct rb_node *pos;
 185
 186         for (pos = rb_first(&server->state_owners);
 187              pos != NULL;
 188              pos = rb_next(pos)) {
 189                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 190                 if (list_empty(&sp->so_states))
 191                         continue;
 192                 cred = get_rpccred(sp->so_cred);
 193                 break;
 194         }
 195         return cred;
 196 }
 197
 198 /**
 199  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 200  * @clp: client state handle
 201  *
 202  * Returns an rpc_cred with reference count bumped, or NULL.
 203  * Caller must hold clp->cl_lock.
 204  */
 205 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 206 {
 207         struct rpc_cred *cred = NULL;
 208         struct nfs_server *server;
 209
 210         /* Use machine credentials if available */
 211         cred = nfs4_get_machine_cred_locked(clp);
 212         if (cred != NULL)
 213                 goto out;
 214
 215         rcu_read_lock();
 216         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 217                 cred = nfs4_get_renew_cred_server_locked(server);
 218                 if (cred != NULL)
 219                         break;
 220         }
 221         rcu_read_unlock();
 222
 223 out:
 224         return cred;
 225 }
 226
 227 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 228 {
 229         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 230                 spin_lock(&tbl->slot_tbl_lock);
 231                 nfs41_wake_slot_table(tbl);
 232                 spin_unlock(&tbl->slot_tbl_lock);
 233         }
 234 }
 235
 236 static void nfs4_end_drain_session(struct nfs_client *clp)
 237 {
 238         struct nfs4_session *ses = clp->cl_session;
 239
 240         if (clp->cl_slot_tbl) {
 241                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 242                 return;
 243         }
 244
 245         if (ses != NULL) {
 246                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 247                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 248         }
 249 }
 250
 251 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 252 {
 253         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 254         spin_lock(&tbl->slot_tbl_lock);
 255         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 256                 reinit_completion(&tbl->complete);
 257                 spin_unlock(&tbl->slot_tbl_lock);
 258                 return wait_for_completion_interruptible(&tbl->complete);
 259         }
 260         spin_unlock(&tbl->slot_tbl_lock);
 261         return 0;
 262 }
 263
 264 static int nfs4_begin_drain_session(struct nfs_client *clp)
 265 {
 266         struct nfs4_session *ses = clp->cl_session;
 267         int ret = 0;
 268
 269         if (clp->cl_slot_tbl)
 270                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 271
 272         /* back channel */
 273         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 274         if (ret)
 275                 return ret;
 276         /* fore channel */
 277         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 278 }
 279
 280 #if defined(CONFIG_NFS_V4_1)
 281
 282 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 283 {
 284         int status;
 285         struct nfs_fsinfo fsinfo;
 286         unsigned long now;
 287
 288         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 289                 nfs4_schedule_state_renewal(clp);
 290                 return 0;
 291         }
 292
 293         now = jiffies;
 294         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 295         if (status == 0) {
 296                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 297                 nfs4_schedule_state_renewal(clp);
 298         }
 299
 300         return status;
 301 }
 302
 303 static void nfs41_finish_session_reset(struct nfs_client *clp)
 304 {
 305         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 306         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 307         /* create_session negotiated new slot table */
 308         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 309         nfs41_setup_state_renewal(clp);
 310 }
 311
 312 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 313 {
 314         int status;
 315
 316         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 317                 goto do_confirm;
 318         status = nfs4_proc_exchange_id(clp, cred);
 319         if (status != 0)
 320                 goto out;
 321         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 322 do_confirm:
 323         status = nfs4_proc_create_session(clp, cred);
 324         if (status != 0)
 325                 goto out;
 326         if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R))
 327                 nfs4_state_start_reclaim_reboot(clp);
 328         nfs41_finish_session_reset(clp);
 329         nfs_mark_client_ready(clp, NFS_CS_READY);
 330 out:
 331         return status;
 332 }
 333
 334 /**
 335  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 336  *
 337  * @clp: nfs_client under test
 338  * @result: OUT: found nfs_client, or clp
 339  * @cred: credential to use for trunking test
 340  *
 341  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 342  * If NFS4_OK is returned, an nfs_client pointer is planted in
 343  * "result".
 344  *
 345  * Note: The returned client may not yet be marked ready.
 346  */
 347 int nfs41_discover_server_trunking(struct nfs_client *clp,
 348                                    struct nfs_client **result,
 349                                    struct rpc_cred *cred)
 350 {
 351         int status;
 352
 353         status = nfs4_proc_exchange_id(clp, cred);
 354         if (status != NFS4_OK)
 355                 return status;
 356
 357         status = nfs41_walk_client_list(clp, result, cred);
 358         if (status < 0)
 359                 return status;
 360         if (clp != *result)
 361                 return 0;
 362
 363         /*
 364          * Purge state if the client id was established in a prior
 365          * instance and the client id could not have arrived on the
 366          * server via Transparent State Migration.
 367          */
 368         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
 369                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
 370                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 371                 else
 372                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 373         }
 374         nfs4_schedule_state_manager(clp);
 375         status = nfs_wait_client_init_complete(clp);
 376         if (status < 0)
 377                 nfs_put_client(clp);
 378         return status;
 379 }
 380
 381 #endif /* CONFIG_NFS_V4_1 */
 382
 383 /**
 384  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 385  * @clp: client state handle
 386  *
 387  * Returns an rpc_cred with reference count bumped, or NULL.
 388  */
 389 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 390 {
 391         struct rpc_cred *cred;
 392
 393         spin_lock(&clp->cl_lock);
 394         cred = nfs4_get_machine_cred_locked(clp);
 395         spin_unlock(&clp->cl_lock);
 396         return cred;
 397 }
 398
 399 static struct nfs4_state_owner *
 400 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 401 {
 402         struct rb_node **p = &server->state_owners.rb_node,
 403                        *parent = NULL;
 404         struct nfs4_state_owner *sp;
 405
 406         while (*p != NULL) {
 407                 parent = *p;
 408                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 409
 410                 if (cred < sp->so_cred)
 411                         p = &parent->rb_left;
 412                 else if (cred > sp->so_cred)
 413                         p = &parent->rb_right;
 414                 else {
 415                         if (!list_empty(&sp->so_lru))
 416                                 list_del_init(&sp->so_lru);
 417                         atomic_inc(&sp->so_count);
 418                         return sp;
 419                 }
 420         }
 421         return NULL;
 422 }
 423
 424 static struct nfs4_state_owner *
 425 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 426 {
 427         struct nfs_server *server = new->so_server;
 428         struct rb_node **p = &server->state_owners.rb_node,
 429                        *parent = NULL;
 430         struct nfs4_state_owner *sp;
 431         int err;
 432
 433         while (*p != NULL) {
 434                 parent = *p;
 435                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 436
 437                 if (new->so_cred < sp->so_cred)
 438                         p = &parent->rb_left;
 439                 else if (new->so_cred > sp->so_cred)
 440                         p = &parent->rb_right;
 441                 else {
 442                         if (!list_empty(&sp->so_lru))
 443                                 list_del_init(&sp->so_lru);
 444                         atomic_inc(&sp->so_count);
 445                         return sp;
 446                 }
 447         }
 448         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 449         if (err)
 450                 return ERR_PTR(err);
 451         rb_link_node(&new->so_server_node, parent, p);
 452         rb_insert_color(&new->so_server_node, &server->state_owners);
 453         return new;
 454 }
 455
 456 static void
 457 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 458 {
 459         struct nfs_server *server = sp->so_server;
 460
 461         if (!RB_EMPTY_NODE(&sp->so_server_node))
 462                 rb_erase(&sp->so_server_node, &server->state_owners);
 463         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 464 }
 465
 466 static void
 467 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 468 {
 469         sc->create_time = ktime_get();
 470         sc->flags = 0;
 471         sc->counter = 0;
 472         spin_lock_init(&sc->lock);
 473         INIT_LIST_HEAD(&sc->list);
 474         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 475 }
 476
 477 static void
 478 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 479 {
 480         rpc_destroy_wait_queue(&sc->wait);
 481 }
 482
 483 /*
 484  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 485  * create a new state_owner.
 486  *
 487  */
 488 static struct nfs4_state_owner *
 489 nfs4_alloc_state_owner(struct nfs_server *server,
 490                 struct rpc_cred *cred,
 491                 gfp_t gfp_flags)
 492 {
 493         struct nfs4_state_owner *sp;
 494
 495         sp = kzalloc(sizeof(*sp), gfp_flags);
 496         if (!sp)
 497                 return NULL;
 498         sp->so_server = server;
 499         sp->so_cred = get_rpccred(cred);
 500         spin_lock_init(&sp->so_lock);
 501         INIT_LIST_HEAD(&sp->so_states);
 502         nfs4_init_seqid_counter(&sp->so_seqid);
 503         atomic_set(&sp->so_count, 1);
 504         INIT_LIST_HEAD(&sp->so_lru);
 505         seqcount_init(&sp->so_reclaim_seqcount);
 506         mutex_init(&sp->so_delegreturn_mutex);
 507         return sp;
 508 }
 509
 510 static void
 511 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 512 {
 513         /* This state_owner is no longer usable, but must
 514          * remain in place so that state recovery can find it
 515          * and the opens associated with it.
 516          * It may also be used for new 'open' request to
 517          * return a delegation to the server.
 518          * So update the 'create_time' so that it looks like
 519          * a new state_owner.  This will cause the server to
 520          * request an OPEN_CONFIRM to start a new sequence.
 521          */
 522         sp->so_seqid.create_time = ktime_get();
 523 }
 524
 525 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 526 {
 527         nfs4_destroy_seqid_counter(&sp->so_seqid);
 528         put_rpccred(sp->so_cred);
 529         kfree(sp);
 530 }
 531
 532 static void nfs4_gc_state_owners(struct nfs_server *server)
 533 {
 534         struct nfs_client *clp = server->nfs_client;
 535         struct nfs4_state_owner *sp, *tmp;
 536         unsigned long time_min, time_max;
 537         LIST_HEAD(doomed);
 538
 539         spin_lock(&clp->cl_lock);
 540         time_max = jiffies;
 541         time_min = (long)time_max - (long)clp->cl_lease_time;
 542         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 543                 /* NB: LRU is sorted so that oldest is at the head */
 544                 if (time_in_range(sp->so_expires, time_min, time_max))
 545                         break;
 546                 list_move(&sp->so_lru, &doomed);
 547                 nfs4_remove_state_owner_locked(sp);
 548         }
 549         spin_unlock(&clp->cl_lock);
 550
 551         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 552                 list_del(&sp->so_lru);
 553                 nfs4_free_state_owner(sp);
 554         }
 555 }
 556
 557 /**
 558  * nfs4_get_state_owner - Look up a state owner given a credential
 559  * @server: nfs_server to search
 560  * @cred: RPC credential to match
 561  *
 562  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 563  */
 564 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 565                                               struct rpc_cred *cred,
 566                                               gfp_t gfp_flags)
 567 {
 568         struct nfs_client *clp = server->nfs_client;
 569         struct nfs4_state_owner *sp, *new;
 570
 571         spin_lock(&clp->cl_lock);
 572         sp = nfs4_find_state_owner_locked(server, cred);
 573         spin_unlock(&clp->cl_lock);
 574         if (sp != NULL)
 575                 goto out;
 576         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 577         if (new == NULL)
 578                 goto out;
 579         do {
 580                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 581                         break;
 582                 spin_lock(&clp->cl_lock);
 583                 sp = nfs4_insert_state_owner_locked(new);
 584                 spin_unlock(&clp->cl_lock);
 585         } while (sp == ERR_PTR(-EAGAIN));
 586         if (sp != new)
 587                 nfs4_free_state_owner(new);
 588 out:
 589         nfs4_gc_state_owners(server);
 590         return sp;
 591 }
 592
 593 /**
 594  * nfs4_put_state_owner - Release a nfs4_state_owner
 595  * @sp: state owner data to release
 596  *
 597  * Note that we keep released state owners on an LRU
 598  * list.
 599  * This caches valid state owners so that they can be
 600  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 601  * It also pins the uniquifier of dropped state owners for
 602  * a while, to ensure that those state owner names are
 603  * never reused.
 604  */
 605 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 606 {
 607         struct nfs_server *server = sp->so_server;
 608         struct nfs_client *clp = server->nfs_client;
 609
 610         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 611                 return;
 612
 613         sp->so_expires = jiffies;
 614         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 615         spin_unlock(&clp->cl_lock);
 616 }
 617
 618 /**
 619  * nfs4_purge_state_owners - Release all cached state owners
 620  * @server: nfs_server with cached state owners to release
 621  * @head: resulting list of state owners
 622  *
 623  * Called at umount time.  Remaining state owners will be on
 624  * the LRU with ref count of zero.
 625  * Note that the state owners are not freed, but are added
 626  * to the list @head, which can later be used as an argument
 627  * to nfs4_free_state_owners.
 628  */
 629 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 630 {
 631         struct nfs_client *clp = server->nfs_client;
 632         struct nfs4_state_owner *sp, *tmp;
 633
 634         spin_lock(&clp->cl_lock);
 635         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 636                 list_move(&sp->so_lru, head);
 637                 nfs4_remove_state_owner_locked(sp);
 638         }
 639         spin_unlock(&clp->cl_lock);
 640 }
 641
 642 /**
 643  * nfs4_purge_state_owners - Release all cached state owners
 644  * @head: resulting list of state owners
 645  *
 646  * Frees a list of state owners that was generated by
 647  * nfs4_purge_state_owners
 648  */
 649 void nfs4_free_state_owners(struct list_head *head)
 650 {
 651         struct nfs4_state_owner *sp, *tmp;
 652
 653         list_for_each_entry_safe(sp, tmp, head, so_lru) {
 654                 list_del(&sp->so_lru);
 655                 nfs4_free_state_owner(sp);
 656         }
 657 }
 658
 659 static struct nfs4_state *
 660 nfs4_alloc_open_state(void)
 661 {
 662         struct nfs4_state *state;
 663
 664         state = kzalloc(sizeof(*state), GFP_NOFS);
 665         if (!state)
 666                 return NULL;
 667         atomic_set(&state->count, 1);
 668         INIT_LIST_HEAD(&state->lock_states);
 669         spin_lock_init(&state->state_lock);
 670         seqlock_init(&state->seqlock);
 671         return state;
 672 }
 673
 674 void
 675 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 676 {
 677         if (state->state == fmode)
 678                 return;
 679         /* NB! List reordering - see the reclaim code for why.  */
 680         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 681                 if (fmode & FMODE_WRITE)
 682                         list_move(&state->open_states, &state->owner->so_states);
 683                 else
 684                         list_move_tail(&state->open_states, &state->owner->so_states);
 685         }
 686         state->state = fmode;
 687 }
 688
 689 static struct nfs4_state *
 690 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 691 {
 692         struct nfs_inode *nfsi = NFS_I(inode);
 693         struct nfs4_state *state;
 694
 695         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 696                 if (state->owner != owner)
 697                         continue;
 698                 if (!nfs4_valid_open_stateid(state))
 699                         continue;
 700                 if (atomic_inc_not_zero(&state->count))
 701                         return state;
 702         }
 703         return NULL;
 704 }
 705
 706 static void
 707 nfs4_free_open_state(struct nfs4_state *state)
 708 {
 709         kfree(state);
 710 }
 711
 712 struct nfs4_state *
 713 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 714 {
 715         struct nfs4_state *state, *new;
 716         struct nfs_inode *nfsi = NFS_I(inode);
 717
 718         spin_lock(&inode->i_lock);
 719         state = __nfs4_find_state_byowner(inode, owner);
 720         spin_unlock(&inode->i_lock);
 721         if (state)
 722                 goto out;
 723         new = nfs4_alloc_open_state();
 724         spin_lock(&owner->so_lock);
 725         spin_lock(&inode->i_lock);
 726         state = __nfs4_find_state_byowner(inode, owner);
 727         if (state == NULL && new != NULL) {
 728                 state = new;
 729                 state->owner = owner;
 730                 atomic_inc(&owner->so_count);
 731                 list_add(&state->inode_states, &nfsi->open_states);
 732                 ihold(inode);
 733                 state->inode = inode;
 734                 spin_unlock(&inode->i_lock);
 735                 /* Note: The reclaim code dictates that we add stateless
 736                  * and read-only stateids to the end of the list */
 737                 list_add_tail(&state->open_states, &owner->so_states);
 738                 spin_unlock(&owner->so_lock);
 739         } else {
 740                 spin_unlock(&inode->i_lock);
 741                 spin_unlock(&owner->so_lock);
 742                 if (new)
 743                         nfs4_free_open_state(new);
 744         }
 745 out:
 746         return state;
 747 }
 748
 749 void nfs4_put_open_state(struct nfs4_state *state)
 750 {
 751         struct inode *inode = state->inode;
 752         struct nfs4_state_owner *owner = state->owner;
 753
 754         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 755                 return;
 756         spin_lock(&inode->i_lock);
 757         list_del(&state->inode_states);
 758         list_del(&state->open_states);
 759         spin_unlock(&inode->i_lock);
 760         spin_unlock(&owner->so_lock);
 761         iput(inode);
 762         nfs4_free_open_state(state);
 763         nfs4_put_state_owner(owner);
 764 }
 765
 766 /*
 767  * Close the current file.
 768  */
 769 static void __nfs4_close(struct nfs4_state *state,
 770                 fmode_t fmode, gfp_t gfp_mask, int wait)
 771 {
 772         struct nfs4_state_owner *owner = state->owner;
 773         int call_close = 0;
 774         fmode_t newstate;
 775
 776         atomic_inc(&owner->so_count);
 777         /* Protect against nfs4_find_state() */
 778         spin_lock(&owner->so_lock);
 779         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 780                 case FMODE_READ:
 781                         state->n_rdonly--;
 782                         break;
 783                 case FMODE_WRITE:
 784                         state->n_wronly--;
 785                         break;
 786                 case FMODE_READ|FMODE_WRITE:
 787                         state->n_rdwr--;
 788         }
 789         newstate = FMODE_READ|FMODE_WRITE;
 790         if (state->n_rdwr == 0) {
 791                 if (state->n_rdonly == 0) {
 792                         newstate &= ~FMODE_READ;
 793                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 794                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 795                 }
 796                 if (state->n_wronly == 0) {
 797                         newstate &= ~FMODE_WRITE;
 798                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 799                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 800                 }
 801                 if (newstate == 0)
 802                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 803         }
 804         nfs4_state_set_mode_locked(state, newstate);
 805         spin_unlock(&owner->so_lock);
 806
 807         if (!call_close) {
 808                 nfs4_put_open_state(state);
 809                 nfs4_put_state_owner(owner);
 810         } else
 811                 nfs4_do_close(state, gfp_mask, wait);
 812 }
 813
 814 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 815 {
 816         __nfs4_close(state, fmode, GFP_NOFS, 0);
 817 }
 818
 819 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 820 {
 821         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 822 }
 823
 824 /*
 825  * Search the state->lock_states for an existing lock_owner
 826  * that is compatible with either of the given owners.
 827  * If the second is non-zero, then the first refers to a Posix-lock
 828  * owner (current->files) and the second refers to a flock/OFD
 829  * owner (struct file*).  In that case, prefer a match for the first
 830  * owner.
 831  * If both sorts of locks are held on the one file we cannot know
 832  * which stateid was intended to be used, so a "correct" choice cannot
 833  * be made.  Failing that, a "consistent" choice is preferable.  The
 834  * consistent choice we make is to prefer the first owner, that of a
 835  * Posix lock.
 836  */
 837 static struct nfs4_lock_state *
 838 __nfs4_find_lock_state(struct nfs4_state *state,
 839                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 840 {
 841         struct nfs4_lock_state *pos, *ret = NULL;
 842         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 843                 if (pos->ls_owner == fl_owner) {
 844                         ret = pos;
 845                         break;
 846                 }
 847                 if (pos->ls_owner == fl_owner2)
 848                         ret = pos;
 849         }
 850         if (ret)
 851                 atomic_inc(&ret->ls_count);
 852         return ret;
 853 }
 854
 855 /*
 856  * Return a compatible lock_state. If no initialized lock_state structure
 857  * exists, return an uninitialized one.
 858  *
 859  */
 860 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 861 {
 862         struct nfs4_lock_state *lsp;
 863         struct nfs_server *server = state->owner->so_server;
 864
 865         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 866         if (lsp == NULL)
 867                 return NULL;
 868         nfs4_init_seqid_counter(&lsp->ls_seqid);
 869         atomic_set(&lsp->ls_count, 1);
 870         lsp->ls_state = state;
 871         lsp->ls_owner = fl_owner;
 872         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 873         if (lsp->ls_seqid.owner_id < 0)
 874                 goto out_free;
 875         INIT_LIST_HEAD(&lsp->ls_locks);
 876         return lsp;
 877 out_free:
 878         kfree(lsp);
 879         return NULL;
 880 }
 881
 882 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 883 {
 884         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 885         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 886         kfree(lsp);
 887 }
 888
 889 /*
 890  * Return a compatible lock_state. If no initialized lock_state structure
 891  * exists, return an uninitialized one.
 892  *
 893  */
 894 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 895 {
 896         struct nfs4_lock_state *lsp, *new = NULL;
 897
 898         for(;;) {
 899                 spin_lock(&state->state_lock);
 900                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 901                 if (lsp != NULL)
 902                         break;
 903                 if (new != NULL) {
 904                         list_add(&new->ls_locks, &state->lock_states);
 905                         set_bit(LK_STATE_IN_USE, &state->flags);
 906                         lsp = new;
 907                         new = NULL;
 908                         break;
 909                 }
 910                 spin_unlock(&state->state_lock);
 911                 new = nfs4_alloc_lock_state(state, owner);
 912                 if (new == NULL)
 913                         return NULL;
 914         }
 915         spin_unlock(&state->state_lock);
 916         if (new != NULL)
 917                 nfs4_free_lock_state(state->owner->so_server, new);
 918         return lsp;
 919 }
 920
 921 /*
 922  * Release reference to lock_state, and free it if we see that
 923  * it is no longer in use
 924  */
 925 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 926 {
 927         struct nfs_server *server;
 928         struct nfs4_state *state;
 929
 930         if (lsp == NULL)
 931                 return;
 932         state = lsp->ls_state;
 933         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 934                 return;
 935         list_del(&lsp->ls_locks);
 936         if (list_empty(&state->lock_states))
 937                 clear_bit(LK_STATE_IN_USE, &state->flags);
 938         spin_unlock(&state->state_lock);
 939         server = state->owner->so_server;
 940         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 941                 struct nfs_client *clp = server->nfs_client;
 942
 943                 clp->cl_mvops->free_lock_state(server, lsp);
 944         } else
 945                 nfs4_free_lock_state(server, lsp);
 946 }
 947
 948 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 949 {
 950         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 951
 952         dst->fl_u.nfs4_fl.owner = lsp;
 953         atomic_inc(&lsp->ls_count);
 954 }
 955
 956 static void nfs4_fl_release_lock(struct file_lock *fl)
 957 {
 958         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 959 }
 960
 961 static const struct file_lock_operations nfs4_fl_lock_ops = {
 962         .fl_copy_lock = nfs4_fl_copy_lock,
 963         .fl_release_private = nfs4_fl_release_lock,
 964 };
 965
 966 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 967 {
 968         struct nfs4_lock_state *lsp;
 969
 970         if (fl->fl_ops != NULL)
 971                 return 0;
 972         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 973         if (lsp == NULL)
 974                 return -ENOMEM;
 975         fl->fl_u.nfs4_fl.owner = lsp;
 976         fl->fl_ops = &nfs4_fl_lock_ops;
 977         return 0;
 978 }
 979
 980 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 981                 struct nfs4_state *state,
 982                 const struct nfs_lock_context *l_ctx)
 983 {
 984         struct nfs4_lock_state *lsp;
 985         fl_owner_t fl_owner, fl_flock_owner;
 986         int ret = -ENOENT;
 987
 988         if (l_ctx == NULL)
 989                 goto out;
 990
 991         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 992                 goto out;
 993
 994         fl_owner = l_ctx->lockowner;
 995         fl_flock_owner = l_ctx->open_context->flock_owner;
 996
 997         spin_lock(&state->state_lock);
 998         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
 999         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1000                 ret = -EIO;
1001         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1002                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1003                 ret = 0;
1004         }
1005         spin_unlock(&state->state_lock);
1006         nfs4_put_lock_state(lsp);
1007 out:
1008         return ret;
1009 }
1010
1011 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1012 {
1013         const nfs4_stateid *src;
1014         int seq;
1015
1016         do {
1017                 src = &zero_stateid;
1018                 seq = read_seqbegin(&state->seqlock);
1019                 if (test_bit(NFS_OPEN_STATE, &state->flags))
1020                         src = &state->open_stateid;
1021                 nfs4_stateid_copy(dst, src);
1022         } while (read_seqretry(&state->seqlock, seq));
1023 }
1024
1025 /*
1026  * Byte-range lock aware utility to initialize the stateid of read/write
1027  * requests.
1028  */
1029 int nfs4_select_rw_stateid(struct nfs4_state *state,
1030                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1031                 nfs4_stateid *dst, struct rpc_cred **cred)
1032 {
1033         int ret;
1034
1035         if (!nfs4_valid_open_stateid(state))
1036                 return -EIO;
1037         if (cred != NULL)
1038                 *cred = NULL;
1039         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1040         if (ret == -EIO)
1041                 /* A lost lock - don't even consider delegations */
1042                 goto out;
1043         /* returns true if delegation stateid found and copied */
1044         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1045                 ret = 0;
1046                 goto out;
1047         }
1048         if (ret != -ENOENT)
1049                 /* nfs4_copy_delegation_stateid() didn't over-write
1050                  * dst, so it still has the lock stateid which we now
1051                  * choose to use.
1052                  */
1053                 goto out;
1054         nfs4_copy_open_stateid(dst, state);
1055         ret = 0;
1056 out:
1057         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1058                 dst->seqid = 0;
1059         return ret;
1060 }
1061
1062 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1063 {
1064         struct nfs_seqid *new;
1065
1066         new = kmalloc(sizeof(*new), gfp_mask);
1067         if (new == NULL)
1068                 return ERR_PTR(-ENOMEM);
1069         new->sequence = counter;
1070         INIT_LIST_HEAD(&new->list);
1071         new->task = NULL;
1072         return new;
1073 }
1074
1075 void nfs_release_seqid(struct nfs_seqid *seqid)
1076 {
1077         struct nfs_seqid_counter *sequence;
1078
1079         if (seqid == NULL || list_empty(&seqid->list))
1080                 return;
1081         sequence = seqid->sequence;
1082         spin_lock(&sequence->lock);
1083         list_del_init(&seqid->list);
1084         if (!list_empty(&sequence->list)) {
1085                 struct nfs_seqid *next;
1086
1087                 next = list_first_entry(&sequence->list,
1088                                 struct nfs_seqid, list);
1089                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1090         }
1091         spin_unlock(&sequence->lock);
1092 }
1093
1094 void nfs_free_seqid(struct nfs_seqid *seqid)
1095 {
1096         nfs_release_seqid(seqid);
1097         kfree(seqid);
1098 }
1099
1100 /*
1101  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1102  * failed with a seqid incrementing error -
1103  * see comments nfs4.h:seqid_mutating_error()
1104  */
1105 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1106 {
1107         switch (status) {
1108                 case 0:
1109                         break;
1110                 case -NFS4ERR_BAD_SEQID:
1111                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1112                                 return;
1113                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1114                                         " sequence-id error on an"
1115                                         " unconfirmed sequence %p!\n",
1116                                         seqid->sequence);
1117                 case -NFS4ERR_STALE_CLIENTID:
1118                 case -NFS4ERR_STALE_STATEID:
1119                 case -NFS4ERR_BAD_STATEID:
1120                 case -NFS4ERR_BADXDR:
1121                 case -NFS4ERR_RESOURCE:
1122                 case -NFS4ERR_NOFILEHANDLE:
1123                 case -NFS4ERR_MOVED:
1124                         /* Non-seqid mutating errors */
1125                         return;
1126         };
1127         /*
1128          * Note: no locking needed as we are guaranteed to be first
1129          * on the sequence list
1130          */
1131         seqid->sequence->counter++;
1132 }
1133
1134 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1135 {
1136         struct nfs4_state_owner *sp;
1137
1138         if (seqid == NULL)
1139                 return;
1140
1141         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1142         if (status == -NFS4ERR_BAD_SEQID)
1143                 nfs4_reset_state_owner(sp);
1144         if (!nfs4_has_session(sp->so_server->nfs_client))
1145                 nfs_increment_seqid(status, seqid);
1146 }
1147
1148 /*
1149  * Increment the seqid if the LOCK/LOCKU succeeded, or
1150  * failed with a seqid incrementing error -
1151  * see comments nfs4.h:seqid_mutating_error()
1152  */
1153 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1154 {
1155         if (seqid != NULL)
1156                 nfs_increment_seqid(status, seqid);
1157 }
1158
1159 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1160 {
1161         struct nfs_seqid_counter *sequence;
1162         int status = 0;
1163
1164         if (seqid == NULL)
1165                 goto out;
1166         sequence = seqid->sequence;
1167         spin_lock(&sequence->lock);
1168         seqid->task = task;
1169         if (list_empty(&seqid->list))
1170                 list_add_tail(&seqid->list, &sequence->list);
1171         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1172                 goto unlock;
1173         rpc_sleep_on(&sequence->wait, task, NULL);
1174         status = -EAGAIN;
1175 unlock:
1176         spin_unlock(&sequence->lock);
1177 out:
1178         return status;
1179 }
1180
1181 static int nfs4_run_state_manager(void *);
1182
1183 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1184 {
1185         smp_mb__before_atomic();
1186         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1187         smp_mb__after_atomic();
1188         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1189         rpc_wake_up(&clp->cl_rpcwaitq);
1190 }
1191
1192 /*
1193  * Schedule the nfs_client asynchronous state management routine
1194  */
1195 void nfs4_schedule_state_manager(struct nfs_client *clp)
1196 {
1197         struct task_struct *task;
1198         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1199
1200         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1201                 return;
1202         __module_get(THIS_MODULE);
1203         atomic_inc(&clp->cl_count);
1204
1205         /* The rcu_read_lock() is not strictly necessary, as the state
1206          * manager is the only thread that ever changes the rpc_xprt
1207          * after it's initialized.  At this point, we're single threaded. */
1208         rcu_read_lock();
1209         snprintf(buf, sizeof(buf), "%s-manager",
1210                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1211         rcu_read_unlock();
1212         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1213         if (IS_ERR(task)) {
1214                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1215                         __func__, PTR_ERR(task));
1216                 if (!nfs_client_init_is_complete(clp))
1217                         nfs_mark_client_ready(clp, PTR_ERR(task));
1218                 nfs4_clear_state_manager_bit(clp);
1219                 nfs_put_client(clp);
1220                 module_put(THIS_MODULE);
1221         }
1222 }
1223
1224 /*
1225  * Schedule a lease recovery attempt
1226  */
1227 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1228 {
1229         if (!clp)
1230                 return;
1231         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1232                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1233         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1234                         clp->cl_hostname);
1235         nfs4_schedule_state_manager(clp);
1236 }
1237 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1238
1239 /**
1240  * nfs4_schedule_migration_recovery - trigger migration recovery
1241  *
1242  * @server: FSID that is migrating
1243  *
1244  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1245  * value is returned.
1246  */
1247 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1248 {
1249         struct nfs_client *clp = server->nfs_client;
1250
1251         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1252                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1253                                 clp->cl_hostname);
1254                 return -NFS4ERR_IO;
1255         }
1256
1257         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1258                 return -NFS4ERR_IO;
1259
1260         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1261                         __func__,
1262                         (unsigned long long)server->fsid.major,
1263                         (unsigned long long)server->fsid.minor,
1264                         clp->cl_hostname);
1265
1266         set_bit(NFS_MIG_IN_TRANSITION,
1267                         &((struct nfs_server *)server)->mig_status);
1268         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1269
1270         nfs4_schedule_state_manager(clp);
1271         return 0;
1272 }
1273 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1274
1275 /**
1276  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1277  *
1278  * @clp: server to check for moved leases
1279  *
1280  */
1281 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1282 {
1283         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1284                 __func__, clp->cl_clientid, clp->cl_hostname);
1285
1286         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1287         nfs4_schedule_state_manager(clp);
1288 }
1289 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1290
1291 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1292 {
1293         int res;
1294
1295         might_sleep();
1296
1297         atomic_inc(&clp->cl_count);
1298         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1299                                  nfs_wait_bit_killable, TASK_KILLABLE);
1300         if (res)
1301                 goto out;
1302         if (clp->cl_cons_state < 0)
1303                 res = clp->cl_cons_state;
1304 out:
1305         nfs_put_client(clp);
1306         return res;
1307 }
1308
1309 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1310 {
1311         unsigned int loop;
1312         int ret;
1313
1314         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1315                 ret = nfs4_wait_clnt_recover(clp);
1316                 if (ret != 0)
1317                         break;
1318                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1319                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1320                         break;
1321                 nfs4_schedule_state_manager(clp);
1322                 ret = -EIO;
1323         }
1324         return ret;
1325 }
1326
1327 /*
1328  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1329  * @clp: client to process
1330  *
1331  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1332  * resend of the SETCLIENTID and hence re-establish the
1333  * callback channel. Then return all existing delegations.
1334  */
1335 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1336 {
1337         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1338         nfs_expire_all_delegations(clp);
1339         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1340                         clp->cl_hostname);
1341 }
1342
1343 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1344 {
1345         nfs40_handle_cb_pathdown(clp);
1346         nfs4_schedule_state_manager(clp);
1347 }
1348
1349 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1350 {
1351
1352         if (!nfs4_valid_open_stateid(state))
1353                 return 0;
1354         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1355         /* Don't recover state that expired before the reboot */
1356         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1357                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1358                 return 0;
1359         }
1360         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1361         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1362         return 1;
1363 }
1364
1365 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1366 {
1367         if (!nfs4_valid_open_stateid(state))
1368                 return 0;
1369         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1370         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1371         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1372         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1373         return 1;
1374 }
1375
1376 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1377 {
1378         struct nfs_client *clp = server->nfs_client;
1379
1380         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1381                 return -EBADF;
1382         nfs_inode_find_delegation_state_and_recover(state->inode,
1383                         &state->stateid);
1384         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1385                         clp->cl_hostname);
1386         nfs4_schedule_state_manager(clp);
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1390
1391 static struct nfs4_lock_state *
1392 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1393                 const nfs4_stateid *stateid)
1394 {
1395         struct nfs4_lock_state *pos;
1396
1397         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1398                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1399                         continue;
1400                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1401                         return pos;
1402         }
1403         return NULL;
1404 }
1405
1406 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1407                 const nfs4_stateid *stateid)
1408 {
1409         bool found = false;
1410
1411         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1412                 spin_lock(&state->state_lock);
1413                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1414                         found = true;
1415                 spin_unlock(&state->state_lock);
1416         }
1417         return found;
1418 }
1419
1420 void nfs_inode_find_state_and_recover(struct inode *inode,
1421                 const nfs4_stateid *stateid)
1422 {
1423         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1424         struct nfs_inode *nfsi = NFS_I(inode);
1425         struct nfs_open_context *ctx;
1426         struct nfs4_state *state;
1427         bool found = false;
1428
1429         spin_lock(&inode->i_lock);
1430         list_for_each_entry(ctx, &nfsi->open_files, list) {
1431                 state = ctx->state;
1432                 if (state == NULL)
1433                         continue;
1434                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1435                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1436                         found = true;
1437                         continue;
1438                 }
1439                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1440                     nfs4_state_mark_reclaim_nograce(clp, state))
1441                         found = true;
1442         }
1443         spin_unlock(&inode->i_lock);
1444
1445         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1446         if (found)
1447                 nfs4_schedule_state_manager(clp);
1448 }
1449
1450 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1451 {
1452         struct inode *inode = state->inode;
1453         struct nfs_inode *nfsi = NFS_I(inode);
1454         struct nfs_open_context *ctx;
1455
1456         spin_lock(&inode->i_lock);
1457         list_for_each_entry(ctx, &nfsi->open_files, list) {
1458                 if (ctx->state != state)
1459                         continue;
1460                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1461         }
1462         spin_unlock(&inode->i_lock);
1463 }
1464
1465 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1466 {
1467         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1468         nfs4_state_mark_open_context_bad(state);
1469 }
1470
1471
1472 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1473 {
1474         struct inode *inode = state->inode;
1475         struct nfs_inode *nfsi = NFS_I(inode);
1476         struct file_lock *fl;
1477         struct nfs4_lock_state *lsp;
1478         int status = 0;
1479         struct file_lock_context *flctx = inode->i_flctx;
1480         struct list_head *list;
1481
1482         if (flctx == NULL)
1483                 return 0;
1484
1485         list = &flctx->flc_posix;
1486
1487         /* Guard against delegation returns and new lock/unlock calls */
1488         down_write(&nfsi->rwsem);
1489         spin_lock(&flctx->flc_lock);
1490 restart:
1491         list_for_each_entry(fl, list, fl_list) {
1492                 if (nfs_file_open_context(fl->fl_file)->state != state)
1493                         continue;
1494                 spin_unlock(&flctx->flc_lock);
1495                 status = ops->recover_lock(state, fl);
1496                 switch (status) {
1497                 case 0:
1498                         break;
1499                 case -ESTALE:
1500                 case -NFS4ERR_ADMIN_REVOKED:
1501                 case -NFS4ERR_STALE_STATEID:
1502                 case -NFS4ERR_BAD_STATEID:
1503                 case -NFS4ERR_EXPIRED:
1504                 case -NFS4ERR_NO_GRACE:
1505                 case -NFS4ERR_STALE_CLIENTID:
1506                 case -NFS4ERR_BADSESSION:
1507                 case -NFS4ERR_BADSLOT:
1508                 case -NFS4ERR_BAD_HIGH_SLOT:
1509                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1510                         goto out;
1511                 default:
1512                         pr_err("NFS: %s: unhandled error %d\n",
1513                                         __func__, status);
1514                 case -ENOMEM:
1515                 case -NFS4ERR_DENIED:
1516                 case -NFS4ERR_RECLAIM_BAD:
1517                 case -NFS4ERR_RECLAIM_CONFLICT:
1518                         lsp = fl->fl_u.nfs4_fl.owner;
1519                         if (lsp)
1520                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1521                         status = 0;
1522                 }
1523                 spin_lock(&flctx->flc_lock);
1524         }
1525         if (list == &flctx->flc_posix) {
1526                 list = &flctx->flc_flock;
1527                 goto restart;
1528         }
1529         spin_unlock(&flctx->flc_lock);
1530 out:
1531         up_write(&nfsi->rwsem);
1532         return status;
1533 }
1534
1535 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1536 {
1537         struct nfs4_state *state;
1538         struct nfs4_lock_state *lock;
1539         int status = 0;
1540
1541         /* Note: we rely on the sp->so_states list being ordered
1542          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1543          * states first.
1544          * This is needed to ensure that the server won't give us any
1545          * read delegations that we have to return if, say, we are
1546          * recovering after a network partition or a reboot from a
1547          * server that doesn't support a grace period.
1548          */
1549         spin_lock(&sp->so_lock);
1550         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1551 restart:
1552         list_for_each_entry(state, &sp->so_states, open_states) {
1553                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1554                         continue;
1555                 if (!nfs4_valid_open_stateid(state))
1556                         continue;
1557                 if (state->state == 0)
1558                         continue;
1559                 atomic_inc(&state->count);
1560                 spin_unlock(&sp->so_lock);
1561                 status = ops->recover_open(sp, state);
1562                 if (status >= 0) {
1563                         status = nfs4_reclaim_locks(state, ops);
1564                         if (status >= 0) {
1565                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1566                                         spin_lock(&state->state_lock);
1567                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1568                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1569                                                         pr_warn_ratelimited("NFS: "
1570                                                                             "%s: Lock reclaim "
1571                                                                             "failed!\n", __func__);
1572                                         }
1573                                         spin_unlock(&state->state_lock);
1574                                 }
1575                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1576                                         &state->flags);
1577                                 nfs4_put_open_state(state);
1578                                 spin_lock(&sp->so_lock);
1579                                 goto restart;
1580                         }
1581                 }
1582                 switch (status) {
1583                         default:
1584                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1585                                         __func__, status);
1586                         case -ENOENT:
1587                         case -ENOMEM:
1588                         case -EACCES:
1589                         case -EROFS:
1590                         case -EIO:
1591                         case -ESTALE:
1592                                 /* Open state on this file cannot be recovered */
1593                                 nfs4_state_mark_recovery_failed(state, status);
1594                                 break;
1595                         case -EAGAIN:
1596                                 ssleep(1);
1597                         case -NFS4ERR_ADMIN_REVOKED:
1598                         case -NFS4ERR_STALE_STATEID:
1599                         case -NFS4ERR_OLD_STATEID:
1600                         case -NFS4ERR_BAD_STATEID:
1601                         case -NFS4ERR_RECLAIM_BAD:
1602                         case -NFS4ERR_RECLAIM_CONFLICT:
1603                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1604                                 break;
1605                         case -NFS4ERR_EXPIRED:
1606                         case -NFS4ERR_NO_GRACE:
1607                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1608                         case -NFS4ERR_STALE_CLIENTID:
1609                         case -NFS4ERR_BADSESSION:
1610                         case -NFS4ERR_BADSLOT:
1611                         case -NFS4ERR_BAD_HIGH_SLOT:
1612                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1613                                 goto out_err;
1614                 }
1615                 nfs4_put_open_state(state);
1616                 spin_lock(&sp->so_lock);
1617                 goto restart;
1618         }
1619         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1620         spin_unlock(&sp->so_lock);
1621         return 0;
1622 out_err:
1623         nfs4_put_open_state(state);
1624         spin_lock(&sp->so_lock);
1625         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1626         spin_unlock(&sp->so_lock);
1627         return status;
1628 }
1629
1630 static void nfs4_clear_open_state(struct nfs4_state *state)
1631 {
1632         struct nfs4_lock_state *lock;
1633
1634         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1635         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1636         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1637         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1638         spin_lock(&state->state_lock);
1639         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1640                 lock->ls_seqid.flags = 0;
1641                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1642         }
1643         spin_unlock(&state->state_lock);
1644 }
1645
1646 static void nfs4_reset_seqids(struct nfs_server *server,
1647         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
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                 sp->so_seqid.flags = 0;
1660                 spin_lock(&sp->so_lock);
1661                 list_for_each_entry(state, &sp->so_states, open_states) {
1662                         if (mark_reclaim(clp, state))
1663                                 nfs4_clear_open_state(state);
1664                 }
1665                 spin_unlock(&sp->so_lock);
1666         }
1667         spin_unlock(&clp->cl_lock);
1668 }
1669
1670 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1671         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1672 {
1673         struct nfs_server *server;
1674
1675         rcu_read_lock();
1676         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1677                 nfs4_reset_seqids(server, mark_reclaim);
1678         rcu_read_unlock();
1679 }
1680
1681 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1682 {
1683         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1684         /* Mark all delegations for reclaim */
1685         nfs_delegation_mark_reclaim(clp);
1686         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1687 }
1688
1689 static int nfs4_reclaim_complete(struct nfs_client *clp,
1690                                  const struct nfs4_state_recovery_ops *ops,
1691                                  struct rpc_cred *cred)
1692 {
1693         /* Notify the server we're done reclaiming our state */
1694         if (ops->reclaim_complete)
1695                 return ops->reclaim_complete(clp, cred);
1696         return 0;
1697 }
1698
1699 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1700 {
1701         struct nfs_client *clp = server->nfs_client;
1702         struct nfs4_state_owner *sp;
1703         struct rb_node *pos;
1704         struct nfs4_state *state;
1705
1706         spin_lock(&clp->cl_lock);
1707         for (pos = rb_first(&server->state_owners);
1708              pos != NULL;
1709              pos = rb_next(pos)) {
1710                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1711                 spin_lock(&sp->so_lock);
1712                 list_for_each_entry(state, &sp->so_states, open_states) {
1713                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1714                                                 &state->flags))
1715                                 continue;
1716                         nfs4_state_mark_reclaim_nograce(clp, state);
1717                 }
1718                 spin_unlock(&sp->so_lock);
1719         }
1720         spin_unlock(&clp->cl_lock);
1721 }
1722
1723 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1724 {
1725         struct nfs_server *server;
1726
1727         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1728                 return 0;
1729
1730         rcu_read_lock();
1731         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1732                 nfs4_clear_reclaim_server(server);
1733         rcu_read_unlock();
1734
1735         nfs_delegation_reap_unclaimed(clp);
1736         return 1;
1737 }
1738
1739 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1740 {
1741         const struct nfs4_state_recovery_ops *ops;
1742         struct rpc_cred *cred;
1743         int err;
1744
1745         if (!nfs4_state_clear_reclaim_reboot(clp))
1746                 return;
1747         ops = clp->cl_mvops->reboot_recovery_ops;
1748         cred = nfs4_get_clid_cred(clp);
1749         err = nfs4_reclaim_complete(clp, ops, cred);
1750         put_rpccred(cred);
1751         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1752                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1753 }
1754
1755 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1756 {
1757         nfs_mark_test_expired_all_delegations(clp);
1758         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1759 }
1760
1761 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1762 {
1763         switch (error) {
1764                 case 0:
1765                         break;
1766                 case -NFS4ERR_CB_PATH_DOWN:
1767                         nfs40_handle_cb_pathdown(clp);
1768                         break;
1769                 case -NFS4ERR_NO_GRACE:
1770                         nfs4_state_end_reclaim_reboot(clp);
1771                         break;
1772                 case -NFS4ERR_STALE_CLIENTID:
1773                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1774                         nfs4_state_start_reclaim_reboot(clp);
1775                         break;
1776                 case -NFS4ERR_EXPIRED:
1777                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1778                         nfs4_state_start_reclaim_nograce(clp);
1779                         break;
1780                 case -NFS4ERR_BADSESSION:
1781                 case -NFS4ERR_BADSLOT:
1782                 case -NFS4ERR_BAD_HIGH_SLOT:
1783                 case -NFS4ERR_DEADSESSION:
1784                 case -NFS4ERR_SEQ_FALSE_RETRY:
1785                 case -NFS4ERR_SEQ_MISORDERED:
1786                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1787                         /* Zero session reset errors */
1788                         break;
1789                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1790                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1791                         break;
1792                 default:
1793                         dprintk("%s: failed to handle error %d for server %s\n",
1794                                         __func__, error, clp->cl_hostname);
1795                         return error;
1796         }
1797         dprintk("%s: handled error %d for server %s\n", __func__, error,
1798                         clp->cl_hostname);
1799         return 0;
1800 }
1801
1802 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1803 {
1804         struct nfs4_state_owner *sp;
1805         struct nfs_server *server;
1806         struct rb_node *pos;
1807         LIST_HEAD(freeme);
1808         int status = 0;
1809
1810 restart:
1811         rcu_read_lock();
1812         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1813                 nfs4_purge_state_owners(server, &freeme);
1814                 spin_lock(&clp->cl_lock);
1815                 for (pos = rb_first(&server->state_owners);
1816                      pos != NULL;
1817                      pos = rb_next(pos)) {
1818                         sp = rb_entry(pos,
1819                                 struct nfs4_state_owner, so_server_node);
1820                         if (!test_and_clear_bit(ops->owner_flag_bit,
1821                                                         &sp->so_flags))
1822                                 continue;
1823                         if (!atomic_inc_not_zero(&sp->so_count))
1824                                 continue;
1825                         spin_unlock(&clp->cl_lock);
1826                         rcu_read_unlock();
1827
1828                         status = nfs4_reclaim_open_state(sp, ops);
1829                         if (status < 0) {
1830                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1831                                 nfs4_put_state_owner(sp);
1832                                 status = nfs4_recovery_handle_error(clp, status);
1833                                 return (status != 0) ? status : -EAGAIN;
1834                         }
1835
1836                         nfs4_put_state_owner(sp);
1837                         goto restart;
1838                 }
1839                 spin_unlock(&clp->cl_lock);
1840         }
1841         rcu_read_unlock();
1842         nfs4_free_state_owners(&freeme);
1843         return 0;
1844 }
1845
1846 static int nfs4_check_lease(struct nfs_client *clp)
1847 {
1848         struct rpc_cred *cred;
1849         const struct nfs4_state_maintenance_ops *ops =
1850                 clp->cl_mvops->state_renewal_ops;
1851         int status;
1852
1853         /* Is the client already known to have an expired lease? */
1854         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1855                 return 0;
1856         spin_lock(&clp->cl_lock);
1857         cred = ops->get_state_renewal_cred_locked(clp);
1858         spin_unlock(&clp->cl_lock);
1859         if (cred == NULL) {
1860                 cred = nfs4_get_clid_cred(clp);
1861                 status = -ENOKEY;
1862                 if (cred == NULL)
1863                         goto out;
1864         }
1865         status = ops->renew_lease(clp, cred);
1866         put_rpccred(cred);
1867         if (status == -ETIMEDOUT) {
1868                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1869                 return 0;
1870         }
1871 out:
1872         return nfs4_recovery_handle_error(clp, status);
1873 }
1874
1875 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1876  * and for recoverable errors on EXCHANGE_ID for v4.1
1877  */
1878 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1879 {
1880         switch (status) {
1881         case -NFS4ERR_SEQ_MISORDERED:
1882                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1883                         return -ESERVERFAULT;
1884                 /* Lease confirmation error: retry after purging the lease */
1885                 ssleep(1);
1886                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1887                 break;
1888         case -NFS4ERR_STALE_CLIENTID:
1889                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1890                 nfs4_state_start_reclaim_reboot(clp);
1891                 break;
1892         case -NFS4ERR_CLID_INUSE:
1893                 pr_err("NFS: Server %s reports our clientid is in use\n",
1894                         clp->cl_hostname);
1895                 nfs_mark_client_ready(clp, -EPERM);
1896                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1897                 return -EPERM;
1898         case -EACCES:
1899         case -NFS4ERR_DELAY:
1900         case -ETIMEDOUT:
1901         case -EAGAIN:
1902                 ssleep(1);
1903                 break;
1904
1905         case -NFS4ERR_MINOR_VERS_MISMATCH:
1906                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1907                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1908                 dprintk("%s: exit with error %d for server %s\n",
1909                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1910                 return -EPROTONOSUPPORT;
1911         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1912                                  * in nfs4_exchange_id */
1913         default:
1914                 dprintk("%s: exit with error %d for server %s\n", __func__,
1915                                 status, clp->cl_hostname);
1916                 return status;
1917         }
1918         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1919         dprintk("%s: handled error %d for server %s\n", __func__, status,
1920                         clp->cl_hostname);
1921         return 0;
1922 }
1923
1924 static int nfs4_establish_lease(struct nfs_client *clp)
1925 {
1926         struct rpc_cred *cred;
1927         const struct nfs4_state_recovery_ops *ops =
1928                 clp->cl_mvops->reboot_recovery_ops;
1929         int status;
1930
1931         nfs4_begin_drain_session(clp);
1932         cred = nfs4_get_clid_cred(clp);
1933         if (cred == NULL)
1934                 return -ENOENT;
1935         status = ops->establish_clid(clp, cred);
1936         put_rpccred(cred);
1937         if (status != 0)
1938                 return status;
1939         pnfs_destroy_all_layouts(clp);
1940         return 0;
1941 }
1942
1943 /*
1944  * Returns zero or a negative errno.  NFS4ERR values are converted
1945  * to local errno values.
1946  */
1947 static int nfs4_reclaim_lease(struct nfs_client *clp)
1948 {
1949         int status;
1950
1951         status = nfs4_establish_lease(clp);
1952         if (status < 0)
1953                 return nfs4_handle_reclaim_lease_error(clp, status);
1954         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1955                 nfs4_state_start_reclaim_nograce(clp);
1956         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1957                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1958         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1959         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1960         return 0;
1961 }
1962
1963 static int nfs4_purge_lease(struct nfs_client *clp)
1964 {
1965         int status;
1966
1967         status = nfs4_establish_lease(clp);
1968         if (status < 0)
1969                 return nfs4_handle_reclaim_lease_error(clp, status);
1970         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1971         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1972         nfs4_state_start_reclaim_nograce(clp);
1973         return 0;
1974 }
1975
1976 /*
1977  * Try remote migration of one FSID from a source server to a
1978  * destination server.  The source server provides a list of
1979  * potential destinations.
1980  *
1981  * Returns zero or a negative NFS4ERR status code.
1982  */
1983 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1984 {
1985         struct nfs_client *clp = server->nfs_client;
1986         struct nfs4_fs_locations *locations = NULL;
1987         struct inode *inode;
1988         struct page *page;
1989         int status, result;
1990
1991         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1992                         (unsigned long long)server->fsid.major,
1993                         (unsigned long long)server->fsid.minor,
1994                         clp->cl_hostname);
1995
1996         result = 0;
1997         page = alloc_page(GFP_KERNEL);
1998         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1999         if (page == NULL || locations == NULL) {
2000                 dprintk("<-- %s: no memory\n", __func__);
2001                 goto out;
2002         }
2003
2004         inode = d_inode(server->super->s_root);
2005         result = nfs4_proc_get_locations(inode, locations, page, cred);
2006         if (result) {
2007                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2008                         __func__, result);
2009                 goto out;
2010         }
2011
2012         result = -NFS4ERR_NXIO;
2013         if (!locations->nlocations)
2014                 goto out;
2015
2016         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2017                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2018                         __func__);
2019                 goto out;
2020         }
2021
2022         nfs4_begin_drain_session(clp);
2023
2024         status = nfs4_replace_transport(server, locations);
2025         if (status != 0) {
2026                 dprintk("<-- %s: failed to replace transport: %d\n",
2027                         __func__, status);
2028                 goto out;
2029         }
2030
2031         result = 0;
2032         dprintk("<-- %s: migration succeeded\n", __func__);
2033
2034 out:
2035         if (page != NULL)
2036                 __free_page(page);
2037         kfree(locations);
2038         if (result) {
2039                 pr_err("NFS: migration recovery failed (server %s)\n",
2040                                 clp->cl_hostname);
2041                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2042         }
2043         return result;
2044 }
2045
2046 /*
2047  * Returns zero or a negative NFS4ERR status code.
2048  */
2049 static int nfs4_handle_migration(struct nfs_client *clp)
2050 {
2051         const struct nfs4_state_maintenance_ops *ops =
2052                                 clp->cl_mvops->state_renewal_ops;
2053         struct nfs_server *server;
2054         struct rpc_cred *cred;
2055
2056         dprintk("%s: migration reported on \"%s\"\n", __func__,
2057                         clp->cl_hostname);
2058
2059         spin_lock(&clp->cl_lock);
2060         cred = ops->get_state_renewal_cred_locked(clp);
2061         spin_unlock(&clp->cl_lock);
2062         if (cred == NULL)
2063                 return -NFS4ERR_NOENT;
2064
2065         clp->cl_mig_gen++;
2066 restart:
2067         rcu_read_lock();
2068         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2069                 int status;
2070
2071                 if (server->mig_gen == clp->cl_mig_gen)
2072                         continue;
2073                 server->mig_gen = clp->cl_mig_gen;
2074
2075                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2076                                                 &server->mig_status))
2077                         continue;
2078
2079                 rcu_read_unlock();
2080                 status = nfs4_try_migration(server, cred);
2081                 if (status < 0) {
2082                         put_rpccred(cred);
2083                         return status;
2084                 }
2085                 goto restart;
2086         }
2087         rcu_read_unlock();
2088         put_rpccred(cred);
2089         return 0;
2090 }
2091
2092 /*
2093  * Test each nfs_server on the clp's cl_superblocks list to see
2094  * if it's moved to another server.  Stop when the server no longer
2095  * returns NFS4ERR_LEASE_MOVED.
2096  */
2097 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2098 {
2099         const struct nfs4_state_maintenance_ops *ops =
2100                                 clp->cl_mvops->state_renewal_ops;
2101         struct nfs_server *server;
2102         struct rpc_cred *cred;
2103
2104         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2105                         clp->cl_hostname);
2106
2107         spin_lock(&clp->cl_lock);
2108         cred = ops->get_state_renewal_cred_locked(clp);
2109         spin_unlock(&clp->cl_lock);
2110         if (cred == NULL)
2111                 return -NFS4ERR_NOENT;
2112
2113         clp->cl_mig_gen++;
2114 restart:
2115         rcu_read_lock();
2116         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2117                 struct inode *inode;
2118                 int status;
2119
2120                 if (server->mig_gen == clp->cl_mig_gen)
2121                         continue;
2122                 server->mig_gen = clp->cl_mig_gen;
2123
2124                 rcu_read_unlock();
2125
2126                 inode = d_inode(server->super->s_root);
2127                 status = nfs4_proc_fsid_present(inode, cred);
2128                 if (status != -NFS4ERR_MOVED)
2129                         goto restart;   /* wasn't this one */
2130                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2131                         goto restart;   /* there are more */
2132                 goto out;
2133         }
2134         rcu_read_unlock();
2135
2136 out:
2137         put_rpccred(cred);
2138         return 0;
2139 }
2140
2141 /**
2142  * nfs4_discover_server_trunking - Detect server IP address trunking
2143  *
2144  * @clp: nfs_client under test
2145  * @result: OUT: found nfs_client, or clp
2146  *
2147  * Returns zero or a negative errno.  If zero is returned,
2148  * an nfs_client pointer is planted in "result".
2149  *
2150  * Note: since we are invoked in process context, and
2151  * not from inside the state manager, we cannot use
2152  * nfs4_handle_reclaim_lease_error().
2153  */
2154 int nfs4_discover_server_trunking(struct nfs_client *clp,
2155                                   struct nfs_client **result)
2156 {
2157         const struct nfs4_state_recovery_ops *ops =
2158                                 clp->cl_mvops->reboot_recovery_ops;
2159         struct rpc_clnt *clnt;
2160         struct rpc_cred *cred;
2161         int i, status;
2162
2163         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2164
2165         clnt = clp->cl_rpcclient;
2166         i = 0;
2167
2168         mutex_lock(&nfs_clid_init_mutex);
2169 again:
2170         status  = -ENOENT;
2171         cred = nfs4_get_clid_cred(clp);
2172         if (cred == NULL)
2173                 goto out_unlock;
2174
2175         status = ops->detect_trunking(clp, result, cred);
2176         put_rpccred(cred);
2177         switch (status) {
2178         case 0:
2179         case -EINTR:
2180         case -ERESTARTSYS:
2181                 break;
2182         case -ETIMEDOUT:
2183                 if (clnt->cl_softrtry)
2184                         break;
2185         case -NFS4ERR_DELAY:
2186         case -EAGAIN:
2187                 ssleep(1);
2188         case -NFS4ERR_STALE_CLIENTID:
2189                 dprintk("NFS: %s after status %d, retrying\n",
2190                         __func__, status);
2191                 goto again;
2192         case -EACCES:
2193                 if (i++ == 0) {
2194                         nfs4_root_machine_cred(clp);
2195                         goto again;
2196                 }
2197                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2198                         break;
2199         case -NFS4ERR_CLID_INUSE:
2200         case -NFS4ERR_WRONGSEC:
2201                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2202                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2203                         status = -EPERM;
2204                         break;
2205                 }
2206                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2207                 if (IS_ERR(clnt)) {
2208                         status = PTR_ERR(clnt);
2209                         break;
2210                 }
2211                 /* Note: this is safe because we haven't yet marked the
2212                  * client as ready, so we are the only user of
2213                  * clp->cl_rpcclient
2214                  */
2215                 clnt = xchg(&clp->cl_rpcclient, clnt);
2216                 rpc_shutdown_client(clnt);
2217                 clnt = clp->cl_rpcclient;
2218                 goto again;
2219
2220         case -NFS4ERR_MINOR_VERS_MISMATCH:
2221                 status = -EPROTONOSUPPORT;
2222                 break;
2223
2224         case -EKEYEXPIRED:
2225         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2226                                  * in nfs4_exchange_id */
2227                 status = -EKEYEXPIRED;
2228                 break;
2229         default:
2230                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2231                                 __func__, status);
2232                 status = -EIO;
2233         }
2234
2235 out_unlock:
2236         mutex_unlock(&nfs_clid_init_mutex);
2237         dprintk("NFS: %s: status = %d\n", __func__, status);
2238         return status;
2239 }
2240
2241 #ifdef CONFIG_NFS_V4_1
2242 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2243 {
2244         struct nfs_client *clp = session->clp;
2245
2246         switch (err) {
2247         default:
2248                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2249                 break;
2250         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2251                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2252         }
2253         nfs4_schedule_state_manager(clp);
2254 }
2255 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2256
2257 void nfs41_notify_server(struct nfs_client *clp)
2258 {
2259         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2260         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2261         nfs4_schedule_state_manager(clp);
2262 }
2263
2264 static void nfs4_reset_all_state(struct nfs_client *clp)
2265 {
2266         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2267                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2268                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2269                 nfs4_state_start_reclaim_nograce(clp);
2270                 dprintk("%s: scheduling reset of all state for server %s!\n",
2271                                 __func__, clp->cl_hostname);
2272                 nfs4_schedule_state_manager(clp);
2273         }
2274 }
2275
2276 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2277 {
2278         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2279                 nfs4_state_start_reclaim_reboot(clp);
2280                 dprintk("%s: server %s rebooted!\n", __func__,
2281                                 clp->cl_hostname);
2282                 nfs4_schedule_state_manager(clp);
2283         }
2284 }
2285
2286 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2287 {
2288         nfs4_reset_all_state(clp);
2289         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2290 }
2291
2292 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2293 {
2294         nfs4_state_start_reclaim_nograce(clp);
2295         nfs4_schedule_state_manager(clp);
2296
2297         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2298 }
2299
2300 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2301 {
2302         /* FIXME: For now, we destroy all layouts. */
2303         pnfs_destroy_all_layouts(clp);
2304         /* FIXME: For now, we test all delegations+open state+locks. */
2305         nfs41_handle_some_state_revoked(clp);
2306         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2307                         clp->cl_hostname);
2308 }
2309
2310 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2311 {
2312         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2313         nfs4_schedule_state_manager(clp);
2314
2315         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2316                         clp->cl_hostname);
2317 }
2318
2319 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2320 {
2321         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2322                 &clp->cl_state) == 0)
2323                 nfs4_schedule_state_manager(clp);
2324 }
2325
2326 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2327                 bool recovery)
2328 {
2329         if (!flags)
2330                 return;
2331
2332         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2333                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2334         /*
2335          * If we're called from the state manager thread, then assume we're
2336          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2337          * Those flags are expected to remain set until we're done
2338          * recovering (see RFC5661, section 18.46.3).
2339          */
2340         if (recovery)
2341                 goto out_recovery;
2342
2343         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2344                 nfs41_handle_server_reboot(clp);
2345         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2346                 nfs41_handle_all_state_revoked(clp);
2347         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2348                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2349                 nfs41_handle_some_state_revoked(clp);
2350         if (flags & SEQ4_STATUS_LEASE_MOVED)
2351                 nfs4_schedule_lease_moved_recovery(clp);
2352         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2353                 nfs41_handle_recallable_state_revoked(clp);
2354 out_recovery:
2355         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2356                 nfs41_handle_backchannel_fault(clp);
2357         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2358                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2359                 nfs41_handle_cb_path_down(clp);
2360 }
2361
2362 static int nfs4_reset_session(struct nfs_client *clp)
2363 {
2364         struct rpc_cred *cred;
2365         int status;
2366
2367         if (!nfs4_has_session(clp))
2368                 return 0;
2369         nfs4_begin_drain_session(clp);
2370         cred = nfs4_get_clid_cred(clp);
2371         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2372         switch (status) {
2373         case 0:
2374         case -NFS4ERR_BADSESSION:
2375         case -NFS4ERR_DEADSESSION:
2376                 break;
2377         case -NFS4ERR_BACK_CHAN_BUSY:
2378         case -NFS4ERR_DELAY:
2379                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2380                 status = 0;
2381                 ssleep(1);
2382                 goto out;
2383         default:
2384                 status = nfs4_recovery_handle_error(clp, status);
2385                 goto out;
2386         }
2387
2388         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2389         status = nfs4_proc_create_session(clp, cred);
2390         if (status) {
2391                 dprintk("%s: session reset failed with status %d for server %s!\n",
2392                         __func__, status, clp->cl_hostname);
2393                 status = nfs4_handle_reclaim_lease_error(clp, status);
2394                 goto out;
2395         }
2396         nfs41_finish_session_reset(clp);
2397         dprintk("%s: session reset was successful for server %s!\n",
2398                         __func__, clp->cl_hostname);
2399 out:
2400         if (cred)
2401                 put_rpccred(cred);
2402         return status;
2403 }
2404
2405 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2406 {
2407         struct rpc_cred *cred;
2408         int ret;
2409
2410         if (!nfs4_has_session(clp))
2411                 return 0;
2412         nfs4_begin_drain_session(clp);
2413         cred = nfs4_get_clid_cred(clp);
2414         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2415         if (cred)
2416                 put_rpccred(cred);
2417         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2418         switch (ret) {
2419         case 0:
2420                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2421                         __func__, clp->cl_hostname);
2422                 break;
2423         case -NFS4ERR_DELAY:
2424                 ssleep(1);
2425                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2426                 break;
2427         default:
2428                 return nfs4_recovery_handle_error(clp, ret);
2429         }
2430         return 0;
2431 }
2432 #else /* CONFIG_NFS_V4_1 */
2433 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2434
2435 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2436 {
2437         return 0;
2438 }
2439 #endif /* CONFIG_NFS_V4_1 */
2440
2441 static void nfs4_state_manager(struct nfs_client *clp)
2442 {
2443         int status = 0;
2444         const char *section = "", *section_sep = "";
2445
2446         /* Ensure exclusive access to NFSv4 state */
2447         do {
2448                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2449                         section = "purge state";
2450                         status = nfs4_purge_lease(clp);
2451                         if (status < 0)
2452                                 goto out_error;
2453                         continue;
2454                 }
2455
2456                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2457                         section = "lease expired";
2458                         /* We're going to have to re-establish a clientid */
2459                         status = nfs4_reclaim_lease(clp);
2460                         if (status < 0)
2461                                 goto out_error;
2462                         continue;
2463                 }
2464
2465                 /* Initialize or reset the session */
2466                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2467                         section = "reset session";
2468                         status = nfs4_reset_session(clp);
2469                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2470                                 continue;
2471                         if (status < 0)
2472                                 goto out_error;
2473                 }
2474
2475                 /* Send BIND_CONN_TO_SESSION */
2476                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2477                                 &clp->cl_state)) {
2478                         section = "bind conn to session";
2479                         status = nfs4_bind_conn_to_session(clp);
2480                         if (status < 0)
2481                                 goto out_error;
2482                         continue;
2483                 }
2484
2485                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2486                         section = "check lease";
2487                         status = nfs4_check_lease(clp);
2488                         if (status < 0)
2489                                 goto out_error;
2490                         continue;
2491                 }
2492
2493                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2494                         section = "migration";
2495                         status = nfs4_handle_migration(clp);
2496                         if (status < 0)
2497                                 goto out_error;
2498                 }
2499
2500                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2501                         section = "lease moved";
2502                         status = nfs4_handle_lease_moved(clp);
2503                         if (status < 0)
2504                                 goto out_error;
2505                 }
2506
2507                 /* First recover reboot state... */
2508                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2509                         section = "reclaim reboot";
2510                         status = nfs4_do_reclaim(clp,
2511                                 clp->cl_mvops->reboot_recovery_ops);
2512                         if (status == -EAGAIN)
2513                                 continue;
2514                         if (status < 0)
2515                                 goto out_error;
2516                         nfs4_state_end_reclaim_reboot(clp);
2517                         continue;
2518                 }
2519
2520                 /* Detect expired delegations... */
2521                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2522                         section = "detect expired delegations";
2523                         nfs_reap_expired_delegations(clp);
2524                         continue;
2525                 }
2526
2527                 /* Now recover expired state... */
2528                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2529                         section = "reclaim nograce";
2530                         status = nfs4_do_reclaim(clp,
2531                                 clp->cl_mvops->nograce_recovery_ops);
2532                         if (status == -EAGAIN)
2533                                 continue;
2534                         if (status < 0)
2535                                 goto out_error;
2536                 }
2537
2538                 nfs4_end_drain_session(clp);
2539                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2540                         nfs_client_return_marked_delegations(clp);
2541                         continue;
2542                 }
2543
2544                 nfs4_clear_state_manager_bit(clp);
2545                 /* Did we race with an attempt to give us more work? */
2546                 if (clp->cl_state == 0)
2547                         break;
2548                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2549                         break;
2550         } while (atomic_read(&clp->cl_count) > 1);
2551         return;
2552 out_error:
2553         if (strlen(section))
2554                 section_sep = ": ";
2555         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2556                         " with error %d\n", section_sep, section,
2557                         clp->cl_hostname, -status);
2558         ssleep(1);
2559         nfs4_end_drain_session(clp);
2560         nfs4_clear_state_manager_bit(clp);
2561 }
2562
2563 static int nfs4_run_state_manager(void *ptr)
2564 {
2565         struct nfs_client *clp = ptr;
2566
2567         allow_signal(SIGKILL);
2568         nfs4_state_manager(clp);
2569         nfs_put_client(clp);
2570         module_put_and_exit(0);
2571         return 0;
2572 }
2573
2574 /*
2575  * Local variables:
2576  *  c-basic-offset: 8
2577  * End:
2578  */