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 <[email protected]>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid = {
  69         { .data = { 0 } },
  70         .type = NFS4_SPECIAL_STATEID_TYPE,
  71 };
  72 const nfs4_stateid invalid_stateid = {
  73         {
  74                 .seqid = cpu_to_be32(0xffffffffU),
  75                 .other = { 0 },
  76         },
  77         .type = NFS4_INVALID_STATEID_TYPE,
  78 };
  79
  80 static DEFINE_MUTEX(nfs_clid_init_mutex);
  81
  82 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  83 {
  84         struct nfs4_setclientid_res clid = {
  85                 .clientid = clp->cl_clientid,
  86                 .confirm = clp->cl_confirm,
  87         };
  88         unsigned short port;
  89         int status;
  90         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  91
  92         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  93                 goto do_confirm;
  94         port = nn->nfs_callback_tcpport;
  95         if (clp->cl_addr.ss_family == AF_INET6)
  96                 port = nn->nfs_callback_tcpport6;
  97
  98         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  99         if (status != 0)
 100                 goto out;
 101         clp->cl_clientid = clid.clientid;
 102         clp->cl_confirm = clid.confirm;
 103         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 104 do_confirm:
 105         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 106         if (status != 0)
 107                 goto out;
 108         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 109         nfs4_schedule_state_renewal(clp);
 110 out:
 111         return status;
 112 }
 113
 114 /**
 115  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 116  *
 117  * @clp: nfs_client under test
 118  * @result: OUT: found nfs_client, or clp
 119  * @cred: credential to use for trunking test
 120  *
 121  * Returns zero, a negative errno, or a negative NFS4ERR status.
 122  * If zero is returned, an nfs_client pointer is planted in
 123  * "result".
 124  *
 125  * Note: The returned client may not yet be marked ready.
 126  */
 127 int nfs40_discover_server_trunking(struct nfs_client *clp,
 128                                    struct nfs_client **result,
 129                                    struct rpc_cred *cred)
 130 {
 131         struct nfs4_setclientid_res clid = {
 132                 .clientid = clp->cl_clientid,
 133                 .confirm = clp->cl_confirm,
 134         };
 135         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 136         unsigned short port;
 137         int status;
 138
 139         port = nn->nfs_callback_tcpport;
 140         if (clp->cl_addr.ss_family == AF_INET6)
 141                 port = nn->nfs_callback_tcpport6;
 142
 143         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 144         if (status != 0)
 145                 goto out;
 146         clp->cl_clientid = clid.clientid;
 147         clp->cl_confirm = clid.confirm;
 148
 149         status = nfs40_walk_client_list(clp, result, cred);
 150         if (status == 0) {
 151                 /* Sustain the lease, even if it's empty.  If the clientid4
 152                  * goes stale it's of no use for trunking discovery. */
 153                 nfs4_schedule_state_renewal(*result);
 154         }
 155 out:
 156         return status;
 157 }
 158
 159 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 160 {
 161         struct rpc_cred *cred = NULL;
 162
 163         if (clp->cl_machine_cred != NULL)
 164                 cred = get_rpccred(clp->cl_machine_cred);
 165         return cred;
 166 }
 167
 168 static void nfs4_root_machine_cred(struct nfs_client *clp)
 169 {
 170         struct rpc_cred *cred, *new;
 171
 172         new = rpc_lookup_machine_cred(NULL);
 173         spin_lock(&clp->cl_lock);
 174         cred = clp->cl_machine_cred;
 175         clp->cl_machine_cred = new;
 176         spin_unlock(&clp->cl_lock);
 177         if (cred != NULL)
 178                 put_rpccred(cred);
 179 }
 180
 181 static struct rpc_cred *
 182 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 183 {
 184         struct rpc_cred *cred = NULL;
 185         struct nfs4_state_owner *sp;
 186         struct rb_node *pos;
 187
 188         for (pos = rb_first(&server->state_owners);
 189              pos != NULL;
 190              pos = rb_next(pos)) {
 191                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 192                 if (list_empty(&sp->so_states))
 193                         continue;
 194                 cred = get_rpccred(sp->so_cred);
 195                 break;
 196         }
 197         return cred;
 198 }
 199
 200 /**
 201  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 202  * @clp: client state handle
 203  *
 204  * Returns an rpc_cred with reference count bumped, or NULL.
 205  * Caller must hold clp->cl_lock.
 206  */
 207 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 208 {
 209         struct rpc_cred *cred = NULL;
 210         struct nfs_server *server;
 211
 212         /* Use machine credentials if available */
 213         cred = nfs4_get_machine_cred_locked(clp);
 214         if (cred != NULL)
 215                 goto out;
 216
 217         rcu_read_lock();
 218         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 219                 cred = nfs4_get_renew_cred_server_locked(server);
 220                 if (cred != NULL)
 221                         break;
 222         }
 223         rcu_read_unlock();
 224
 225 out:
 226         return cred;
 227 }
 228
 229 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 230 {
 231         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 232                 spin_lock(&tbl->slot_tbl_lock);
 233                 nfs41_wake_slot_table(tbl);
 234                 spin_unlock(&tbl->slot_tbl_lock);
 235         }
 236 }
 237
 238 static void nfs4_end_drain_session(struct nfs_client *clp)
 239 {
 240         struct nfs4_session *ses = clp->cl_session;
 241
 242         if (clp->cl_slot_tbl) {
 243                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 244                 return;
 245         }
 246
 247         if (ses != NULL) {
 248                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 249                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 250         }
 251 }
 252
 253 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 254 {
 255         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 256         spin_lock(&tbl->slot_tbl_lock);
 257         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 258                 reinit_completion(&tbl->complete);
 259                 spin_unlock(&tbl->slot_tbl_lock);
 260                 return wait_for_completion_interruptible(&tbl->complete);
 261         }
 262         spin_unlock(&tbl->slot_tbl_lock);
 263         return 0;
 264 }
 265
 266 static int nfs4_begin_drain_session(struct nfs_client *clp)
 267 {
 268         struct nfs4_session *ses = clp->cl_session;
 269         int ret = 0;
 270
 271         if (clp->cl_slot_tbl)
 272                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 273
 274         /* back channel */
 275         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 276         if (ret)
 277                 return ret;
 278         /* fore channel */
 279         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 280 }
 281
 282 #if defined(CONFIG_NFS_V4_1)
 283
 284 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 285 {
 286         int status;
 287         struct nfs_fsinfo fsinfo;
 288         unsigned long now;
 289
 290         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 291                 nfs4_schedule_state_renewal(clp);
 292                 return 0;
 293         }
 294
 295         now = jiffies;
 296         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 297         if (status == 0) {
 298                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 299                 nfs4_schedule_state_renewal(clp);
 300         }
 301
 302         return status;
 303 }
 304
 305 static void nfs41_finish_session_reset(struct nfs_client *clp)
 306 {
 307         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 308         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 309         /* create_session negotiated new slot table */
 310         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 311         nfs41_setup_state_renewal(clp);
 312 }
 313
 314 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 315 {
 316         int status;
 317
 318         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 319                 goto do_confirm;
 320         status = nfs4_proc_exchange_id(clp, cred);
 321         if (status != 0)
 322                 goto out;
 323         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 324 do_confirm:
 325         status = nfs4_proc_create_session(clp, cred);
 326         if (status != 0)
 327                 goto out;
 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  *
 622  * Called at umount time.  Remaining state owners will be on
 623  * the LRU with ref count of zero.
 624  */
 625 void nfs4_purge_state_owners(struct nfs_server *server)
 626 {
 627         struct nfs_client *clp = server->nfs_client;
 628         struct nfs4_state_owner *sp, *tmp;
 629         LIST_HEAD(doomed);
 630
 631         spin_lock(&clp->cl_lock);
 632         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 633                 list_move(&sp->so_lru, &doomed);
 634                 nfs4_remove_state_owner_locked(sp);
 635         }
 636         spin_unlock(&clp->cl_lock);
 637
 638         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 639                 list_del(&sp->so_lru);
 640                 nfs4_free_state_owner(sp);
 641         }
 642 }
 643
 644 static struct nfs4_state *
 645 nfs4_alloc_open_state(void)
 646 {
 647         struct nfs4_state *state;
 648
 649         state = kzalloc(sizeof(*state), GFP_NOFS);
 650         if (!state)
 651                 return NULL;
 652         atomic_set(&state->count, 1);
 653         INIT_LIST_HEAD(&state->lock_states);
 654         spin_lock_init(&state->state_lock);
 655         seqlock_init(&state->seqlock);
 656         init_waitqueue_head(&state->waitq);
 657         return state;
 658 }
 659
 660 void
 661 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 662 {
 663         if (state->state == fmode)
 664                 return;
 665         /* NB! List reordering - see the reclaim code for why.  */
 666         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 667                 if (fmode & FMODE_WRITE)
 668                         list_move(&state->open_states, &state->owner->so_states);
 669                 else
 670                         list_move_tail(&state->open_states, &state->owner->so_states);
 671         }
 672         state->state = fmode;
 673 }
 674
 675 static struct nfs4_state *
 676 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 677 {
 678         struct nfs_inode *nfsi = NFS_I(inode);
 679         struct nfs4_state *state;
 680
 681         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 682                 if (state->owner != owner)
 683                         continue;
 684                 if (!nfs4_valid_open_stateid(state))
 685                         continue;
 686                 if (atomic_inc_not_zero(&state->count))
 687                         return state;
 688         }
 689         return NULL;
 690 }
 691
 692 static void
 693 nfs4_free_open_state(struct nfs4_state *state)
 694 {
 695         kfree(state);
 696 }
 697
 698 struct nfs4_state *
 699 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 700 {
 701         struct nfs4_state *state, *new;
 702         struct nfs_inode *nfsi = NFS_I(inode);
 703
 704         spin_lock(&inode->i_lock);
 705         state = __nfs4_find_state_byowner(inode, owner);
 706         spin_unlock(&inode->i_lock);
 707         if (state)
 708                 goto out;
 709         new = nfs4_alloc_open_state();
 710         spin_lock(&owner->so_lock);
 711         spin_lock(&inode->i_lock);
 712         state = __nfs4_find_state_byowner(inode, owner);
 713         if (state == NULL && new != NULL) {
 714                 state = new;
 715                 state->owner = owner;
 716                 atomic_inc(&owner->so_count);
 717                 list_add(&state->inode_states, &nfsi->open_states);
 718                 ihold(inode);
 719                 state->inode = inode;
 720                 spin_unlock(&inode->i_lock);
 721                 /* Note: The reclaim code dictates that we add stateless
 722                  * and read-only stateids to the end of the list */
 723                 list_add_tail(&state->open_states, &owner->so_states);
 724                 spin_unlock(&owner->so_lock);
 725         } else {
 726                 spin_unlock(&inode->i_lock);
 727                 spin_unlock(&owner->so_lock);
 728                 if (new)
 729                         nfs4_free_open_state(new);
 730         }
 731 out:
 732         return state;
 733 }
 734
 735 void nfs4_put_open_state(struct nfs4_state *state)
 736 {
 737         struct inode *inode = state->inode;
 738         struct nfs4_state_owner *owner = state->owner;
 739
 740         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 741                 return;
 742         spin_lock(&inode->i_lock);
 743         list_del(&state->inode_states);
 744         list_del(&state->open_states);
 745         spin_unlock(&inode->i_lock);
 746         spin_unlock(&owner->so_lock);
 747         iput(inode);
 748         nfs4_free_open_state(state);
 749         nfs4_put_state_owner(owner);
 750 }
 751
 752 /*
 753  * Close the current file.
 754  */
 755 static void __nfs4_close(struct nfs4_state *state,
 756                 fmode_t fmode, gfp_t gfp_mask, int wait)
 757 {
 758         struct nfs4_state_owner *owner = state->owner;
 759         int call_close = 0;
 760         fmode_t newstate;
 761
 762         atomic_inc(&owner->so_count);
 763         /* Protect against nfs4_find_state() */
 764         spin_lock(&owner->so_lock);
 765         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 766                 case FMODE_READ:
 767                         state->n_rdonly--;
 768                         break;
 769                 case FMODE_WRITE:
 770                         state->n_wronly--;
 771                         break;
 772                 case FMODE_READ|FMODE_WRITE:
 773                         state->n_rdwr--;
 774         }
 775         newstate = FMODE_READ|FMODE_WRITE;
 776         if (state->n_rdwr == 0) {
 777                 if (state->n_rdonly == 0) {
 778                         newstate &= ~FMODE_READ;
 779                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 780                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 781                 }
 782                 if (state->n_wronly == 0) {
 783                         newstate &= ~FMODE_WRITE;
 784                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 785                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 786                 }
 787                 if (newstate == 0)
 788                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 789         }
 790         nfs4_state_set_mode_locked(state, newstate);
 791         spin_unlock(&owner->so_lock);
 792
 793         if (!call_close) {
 794                 nfs4_put_open_state(state);
 795                 nfs4_put_state_owner(owner);
 796         } else
 797                 nfs4_do_close(state, gfp_mask, wait);
 798 }
 799
 800 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 801 {
 802         __nfs4_close(state, fmode, GFP_NOFS, 0);
 803 }
 804
 805 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 806 {
 807         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 808 }
 809
 810 /*
 811  * Search the state->lock_states for an existing lock_owner
 812  * that is compatible with either of the given owners.
 813  * If the second is non-zero, then the first refers to a Posix-lock
 814  * owner (current->files) and the second refers to a flock/OFD
 815  * owner (struct file*).  In that case, prefer a match for the first
 816  * owner.
 817  * If both sorts of locks are held on the one file we cannot know
 818  * which stateid was intended to be used, so a "correct" choice cannot
 819  * be made.  Failing that, a "consistent" choice is preferable.  The
 820  * consistent choice we make is to prefer the first owner, that of a
 821  * Posix lock.
 822  */
 823 static struct nfs4_lock_state *
 824 __nfs4_find_lock_state(struct nfs4_state *state,
 825                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 826 {
 827         struct nfs4_lock_state *pos, *ret = NULL;
 828         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 829                 if (pos->ls_owner == fl_owner) {
 830                         ret = pos;
 831                         break;
 832                 }
 833                 if (pos->ls_owner == fl_owner2)
 834                         ret = pos;
 835         }
 836         if (ret)
 837                 refcount_inc(&ret->ls_count);
 838         return ret;
 839 }
 840
 841 /*
 842  * Return a compatible lock_state. If no initialized lock_state structure
 843  * exists, return an uninitialized one.
 844  *
 845  */
 846 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 847 {
 848         struct nfs4_lock_state *lsp;
 849         struct nfs_server *server = state->owner->so_server;
 850
 851         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 852         if (lsp == NULL)
 853                 return NULL;
 854         nfs4_init_seqid_counter(&lsp->ls_seqid);
 855         refcount_set(&lsp->ls_count, 1);
 856         lsp->ls_state = state;
 857         lsp->ls_owner = fl_owner;
 858         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 859         if (lsp->ls_seqid.owner_id < 0)
 860                 goto out_free;
 861         INIT_LIST_HEAD(&lsp->ls_locks);
 862         return lsp;
 863 out_free:
 864         kfree(lsp);
 865         return NULL;
 866 }
 867
 868 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 869 {
 870         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 871         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 872         kfree(lsp);
 873 }
 874
 875 /*
 876  * Return a compatible lock_state. If no initialized lock_state structure
 877  * exists, return an uninitialized one.
 878  *
 879  */
 880 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 881 {
 882         struct nfs4_lock_state *lsp, *new = NULL;
 883
 884         for(;;) {
 885                 spin_lock(&state->state_lock);
 886                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 887                 if (lsp != NULL)
 888                         break;
 889                 if (new != NULL) {
 890                         list_add(&new->ls_locks, &state->lock_states);
 891                         set_bit(LK_STATE_IN_USE, &state->flags);
 892                         lsp = new;
 893                         new = NULL;
 894                         break;
 895                 }
 896                 spin_unlock(&state->state_lock);
 897                 new = nfs4_alloc_lock_state(state, owner);
 898                 if (new == NULL)
 899                         return NULL;
 900         }
 901         spin_unlock(&state->state_lock);
 902         if (new != NULL)
 903                 nfs4_free_lock_state(state->owner->so_server, new);
 904         return lsp;
 905 }
 906
 907 /*
 908  * Release reference to lock_state, and free it if we see that
 909  * it is no longer in use
 910  */
 911 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 912 {
 913         struct nfs_server *server;
 914         struct nfs4_state *state;
 915
 916         if (lsp == NULL)
 917                 return;
 918         state = lsp->ls_state;
 919         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 920                 return;
 921         list_del(&lsp->ls_locks);
 922         if (list_empty(&state->lock_states))
 923                 clear_bit(LK_STATE_IN_USE, &state->flags);
 924         spin_unlock(&state->state_lock);
 925         server = state->owner->so_server;
 926         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 927                 struct nfs_client *clp = server->nfs_client;
 928
 929                 clp->cl_mvops->free_lock_state(server, lsp);
 930         } else
 931                 nfs4_free_lock_state(server, lsp);
 932 }
 933
 934 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 935 {
 936         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 937
 938         dst->fl_u.nfs4_fl.owner = lsp;
 939         refcount_inc(&lsp->ls_count);
 940 }
 941
 942 static void nfs4_fl_release_lock(struct file_lock *fl)
 943 {
 944         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 945 }
 946
 947 static const struct file_lock_operations nfs4_fl_lock_ops = {
 948         .fl_copy_lock = nfs4_fl_copy_lock,
 949         .fl_release_private = nfs4_fl_release_lock,
 950 };
 951
 952 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 953 {
 954         struct nfs4_lock_state *lsp;
 955
 956         if (fl->fl_ops != NULL)
 957                 return 0;
 958         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 959         if (lsp == NULL)
 960                 return -ENOMEM;
 961         fl->fl_u.nfs4_fl.owner = lsp;
 962         fl->fl_ops = &nfs4_fl_lock_ops;
 963         return 0;
 964 }
 965
 966 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 967                 struct nfs4_state *state,
 968                 const struct nfs_lock_context *l_ctx)
 969 {
 970         struct nfs4_lock_state *lsp;
 971         fl_owner_t fl_owner, fl_flock_owner;
 972         int ret = -ENOENT;
 973
 974         if (l_ctx == NULL)
 975                 goto out;
 976
 977         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 978                 goto out;
 979
 980         fl_owner = l_ctx->lockowner;
 981         fl_flock_owner = l_ctx->open_context->flock_owner;
 982
 983         spin_lock(&state->state_lock);
 984         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
 985         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 986                 ret = -EIO;
 987         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 988                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 989                 ret = 0;
 990         }
 991         spin_unlock(&state->state_lock);
 992         nfs4_put_lock_state(lsp);
 993 out:
 994         return ret;
 995 }
 996
 997 bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 998 {
 999         bool ret;
1000         int seq;
1001
1002         do {
1003                 ret = false;
1004                 seq = read_seqbegin(&state->seqlock);
1005                 if (nfs4_state_match_open_stateid_other(state, dst)) {
1006                         dst->seqid = state->open_stateid.seqid;
1007                         ret = true;
1008                 }
1009         } while (read_seqretry(&state->seqlock, seq));
1010         return ret;
1011 }
1012
1013 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1014 {
1015         bool ret;
1016         const nfs4_stateid *src;
1017         int seq;
1018
1019         do {
1020                 ret = false;
1021                 src = &zero_stateid;
1022                 seq = read_seqbegin(&state->seqlock);
1023                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1024                         src = &state->open_stateid;
1025                         ret = true;
1026                 }
1027                 nfs4_stateid_copy(dst, src);
1028         } while (read_seqretry(&state->seqlock, seq));
1029         return ret;
1030 }
1031
1032 /*
1033  * Byte-range lock aware utility to initialize the stateid of read/write
1034  * requests.
1035  */
1036 int nfs4_select_rw_stateid(struct nfs4_state *state,
1037                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1038                 nfs4_stateid *dst, struct rpc_cred **cred)
1039 {
1040         int ret;
1041
1042         if (!nfs4_valid_open_stateid(state))
1043                 return -EIO;
1044         if (cred != NULL)
1045                 *cred = NULL;
1046         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1047         if (ret == -EIO)
1048                 /* A lost lock - don't even consider delegations */
1049                 goto out;
1050         /* returns true if delegation stateid found and copied */
1051         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1052                 ret = 0;
1053                 goto out;
1054         }
1055         if (ret != -ENOENT)
1056                 /* nfs4_copy_delegation_stateid() didn't over-write
1057                  * dst, so it still has the lock stateid which we now
1058                  * choose to use.
1059                  */
1060                 goto out;
1061         nfs4_copy_open_stateid(dst, state);
1062         ret = 0;
1063 out:
1064         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1065                 dst->seqid = 0;
1066         return ret;
1067 }
1068
1069 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1070 {
1071         struct nfs_seqid *new;
1072
1073         new = kmalloc(sizeof(*new), gfp_mask);
1074         if (new == NULL)
1075                 return ERR_PTR(-ENOMEM);
1076         new->sequence = counter;
1077         INIT_LIST_HEAD(&new->list);
1078         new->task = NULL;
1079         return new;
1080 }
1081
1082 void nfs_release_seqid(struct nfs_seqid *seqid)
1083 {
1084         struct nfs_seqid_counter *sequence;
1085
1086         if (seqid == NULL || list_empty(&seqid->list))
1087                 return;
1088         sequence = seqid->sequence;
1089         spin_lock(&sequence->lock);
1090         list_del_init(&seqid->list);
1091         if (!list_empty(&sequence->list)) {
1092                 struct nfs_seqid *next;
1093
1094                 next = list_first_entry(&sequence->list,
1095                                 struct nfs_seqid, list);
1096                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1097         }
1098         spin_unlock(&sequence->lock);
1099 }
1100
1101 void nfs_free_seqid(struct nfs_seqid *seqid)
1102 {
1103         nfs_release_seqid(seqid);
1104         kfree(seqid);
1105 }
1106
1107 /*
1108  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1109  * failed with a seqid incrementing error -
1110  * see comments nfs4.h:seqid_mutating_error()
1111  */
1112 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1113 {
1114         switch (status) {
1115                 case 0:
1116                         break;
1117                 case -NFS4ERR_BAD_SEQID:
1118                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1119                                 return;
1120                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1121                                         " sequence-id error on an"
1122                                         " unconfirmed sequence %p!\n",
1123                                         seqid->sequence);
1124                 case -NFS4ERR_STALE_CLIENTID:
1125                 case -NFS4ERR_STALE_STATEID:
1126                 case -NFS4ERR_BAD_STATEID:
1127                 case -NFS4ERR_BADXDR:
1128                 case -NFS4ERR_RESOURCE:
1129                 case -NFS4ERR_NOFILEHANDLE:
1130                 case -NFS4ERR_MOVED:
1131                         /* Non-seqid mutating errors */
1132                         return;
1133         };
1134         /*
1135          * Note: no locking needed as we are guaranteed to be first
1136          * on the sequence list
1137          */
1138         seqid->sequence->counter++;
1139 }
1140
1141 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1142 {
1143         struct nfs4_state_owner *sp;
1144
1145         if (seqid == NULL)
1146                 return;
1147
1148         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1149         if (status == -NFS4ERR_BAD_SEQID)
1150                 nfs4_reset_state_owner(sp);
1151         if (!nfs4_has_session(sp->so_server->nfs_client))
1152                 nfs_increment_seqid(status, seqid);
1153 }
1154
1155 /*
1156  * Increment the seqid if the LOCK/LOCKU succeeded, or
1157  * failed with a seqid incrementing error -
1158  * see comments nfs4.h:seqid_mutating_error()
1159  */
1160 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1161 {
1162         if (seqid != NULL)
1163                 nfs_increment_seqid(status, seqid);
1164 }
1165
1166 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1167 {
1168         struct nfs_seqid_counter *sequence;
1169         int status = 0;
1170
1171         if (seqid == NULL)
1172                 goto out;
1173         sequence = seqid->sequence;
1174         spin_lock(&sequence->lock);
1175         seqid->task = task;
1176         if (list_empty(&seqid->list))
1177                 list_add_tail(&seqid->list, &sequence->list);
1178         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1179                 goto unlock;
1180         rpc_sleep_on(&sequence->wait, task, NULL);
1181         status = -EAGAIN;
1182 unlock:
1183         spin_unlock(&sequence->lock);
1184 out:
1185         return status;
1186 }
1187
1188 static int nfs4_run_state_manager(void *);
1189
1190 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1191 {
1192         smp_mb__before_atomic();
1193         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1194         smp_mb__after_atomic();
1195         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1196         rpc_wake_up(&clp->cl_rpcwaitq);
1197 }
1198
1199 /*
1200  * Schedule the nfs_client asynchronous state management routine
1201  */
1202 void nfs4_schedule_state_manager(struct nfs_client *clp)
1203 {
1204         struct task_struct *task;
1205         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1206
1207         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1208                 return;
1209         __module_get(THIS_MODULE);
1210         refcount_inc(&clp->cl_count);
1211
1212         /* The rcu_read_lock() is not strictly necessary, as the state
1213          * manager is the only thread that ever changes the rpc_xprt
1214          * after it's initialized.  At this point, we're single threaded. */
1215         rcu_read_lock();
1216         snprintf(buf, sizeof(buf), "%s-manager",
1217                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1218         rcu_read_unlock();
1219         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1220         if (IS_ERR(task)) {
1221                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1222                         __func__, PTR_ERR(task));
1223                 nfs4_clear_state_manager_bit(clp);
1224                 nfs_put_client(clp);
1225                 module_put(THIS_MODULE);
1226         }
1227 }
1228
1229 /*
1230  * Schedule a lease recovery attempt
1231  */
1232 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1233 {
1234         if (!clp)
1235                 return;
1236         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1237                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1238         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1239                         clp->cl_hostname);
1240         nfs4_schedule_state_manager(clp);
1241 }
1242 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1243
1244 /**
1245  * nfs4_schedule_migration_recovery - trigger migration recovery
1246  *
1247  * @server: FSID that is migrating
1248  *
1249  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1250  * value is returned.
1251  */
1252 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1253 {
1254         struct nfs_client *clp = server->nfs_client;
1255
1256         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1257                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1258                                 clp->cl_hostname);
1259                 return -NFS4ERR_IO;
1260         }
1261
1262         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1263                 return -NFS4ERR_IO;
1264
1265         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1266                         __func__,
1267                         (unsigned long long)server->fsid.major,
1268                         (unsigned long long)server->fsid.minor,
1269                         clp->cl_hostname);
1270
1271         set_bit(NFS_MIG_IN_TRANSITION,
1272                         &((struct nfs_server *)server)->mig_status);
1273         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1274
1275         nfs4_schedule_state_manager(clp);
1276         return 0;
1277 }
1278 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1279
1280 /**
1281  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1282  *
1283  * @clp: server to check for moved leases
1284  *
1285  */
1286 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1287 {
1288         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1289                 __func__, clp->cl_clientid, clp->cl_hostname);
1290
1291         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1292         nfs4_schedule_state_manager(clp);
1293 }
1294 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1295
1296 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1297 {
1298         int res;
1299
1300         might_sleep();
1301
1302         refcount_inc(&clp->cl_count);
1303         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1304                                  nfs_wait_bit_killable, TASK_KILLABLE);
1305         if (res)
1306                 goto out;
1307         if (clp->cl_cons_state < 0)
1308                 res = clp->cl_cons_state;
1309 out:
1310         nfs_put_client(clp);
1311         return res;
1312 }
1313
1314 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1315 {
1316         unsigned int loop;
1317         int ret;
1318
1319         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1320                 ret = nfs4_wait_clnt_recover(clp);
1321                 if (ret != 0)
1322                         break;
1323                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1324                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1325                         break;
1326                 nfs4_schedule_state_manager(clp);
1327                 ret = -EIO;
1328         }
1329         return ret;
1330 }
1331
1332 /*
1333  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1334  * @clp: client to process
1335  *
1336  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1337  * resend of the SETCLIENTID and hence re-establish the
1338  * callback channel. Then return all existing delegations.
1339  */
1340 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1341 {
1342         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1343         nfs_expire_all_delegations(clp);
1344         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1345                         clp->cl_hostname);
1346 }
1347
1348 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1349 {
1350         nfs40_handle_cb_pathdown(clp);
1351         nfs4_schedule_state_manager(clp);
1352 }
1353
1354 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1355 {
1356
1357         if (!nfs4_valid_open_stateid(state))
1358                 return 0;
1359         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1360         /* Don't recover state that expired before the reboot */
1361         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1362                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1363                 return 0;
1364         }
1365         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1366         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1367         return 1;
1368 }
1369
1370 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1371 {
1372         if (!nfs4_valid_open_stateid(state))
1373                 return 0;
1374         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1375         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1376         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1377         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1378         return 1;
1379 }
1380
1381 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1382 {
1383         struct nfs_client *clp = server->nfs_client;
1384
1385         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1386                 return -EBADF;
1387         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1388                         clp->cl_hostname);
1389         nfs4_schedule_state_manager(clp);
1390         return 0;
1391 }
1392 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1393
1394 static struct nfs4_lock_state *
1395 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1396                 const nfs4_stateid *stateid)
1397 {
1398         struct nfs4_lock_state *pos;
1399
1400         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1401                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1402                         continue;
1403                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1404                         return pos;
1405         }
1406         return NULL;
1407 }
1408
1409 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1410                 const nfs4_stateid *stateid)
1411 {
1412         bool found = false;
1413
1414         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1415                 spin_lock(&state->state_lock);
1416                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1417                         found = true;
1418                 spin_unlock(&state->state_lock);
1419         }
1420         return found;
1421 }
1422
1423 void nfs_inode_find_state_and_recover(struct inode *inode,
1424                 const nfs4_stateid *stateid)
1425 {
1426         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1427         struct nfs_inode *nfsi = NFS_I(inode);
1428         struct nfs_open_context *ctx;
1429         struct nfs4_state *state;
1430         bool found = false;
1431
1432         spin_lock(&inode->i_lock);
1433         list_for_each_entry(ctx, &nfsi->open_files, list) {
1434                 state = ctx->state;
1435                 if (state == NULL)
1436                         continue;
1437                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1438                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1439                         found = true;
1440                         continue;
1441                 }
1442                 if (nfs4_stateid_match_other(&state->open_stateid, stateid) &&
1443                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1444                         found = true;
1445                         continue;
1446                 }
1447                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1448                     nfs4_state_mark_reclaim_nograce(clp, state))
1449                         found = true;
1450         }
1451         spin_unlock(&inode->i_lock);
1452
1453         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1454         if (found)
1455                 nfs4_schedule_state_manager(clp);
1456 }
1457
1458 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1459 {
1460         struct inode *inode = state->inode;
1461         struct nfs_inode *nfsi = NFS_I(inode);
1462         struct nfs_open_context *ctx;
1463
1464         spin_lock(&inode->i_lock);
1465         list_for_each_entry(ctx, &nfsi->open_files, list) {
1466                 if (ctx->state != state)
1467                         continue;
1468                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1469         }
1470         spin_unlock(&inode->i_lock);
1471 }
1472
1473 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1474 {
1475         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1476         nfs4_state_mark_open_context_bad(state);
1477 }
1478
1479
1480 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1481 {
1482         struct inode *inode = state->inode;
1483         struct nfs_inode *nfsi = NFS_I(inode);
1484         struct file_lock *fl;
1485         int status = 0;
1486         struct file_lock_context *flctx = inode->i_flctx;
1487         struct list_head *list;
1488
1489         if (flctx == NULL)
1490                 return 0;
1491
1492         list = &flctx->flc_posix;
1493
1494         /* Guard against delegation returns and new lock/unlock calls */
1495         down_write(&nfsi->rwsem);
1496         spin_lock(&flctx->flc_lock);
1497 restart:
1498         list_for_each_entry(fl, list, fl_list) {
1499                 if (nfs_file_open_context(fl->fl_file)->state != state)
1500                         continue;
1501                 spin_unlock(&flctx->flc_lock);
1502                 status = ops->recover_lock(state, fl);
1503                 switch (status) {
1504                 case 0:
1505                         break;
1506                 case -ESTALE:
1507                 case -NFS4ERR_ADMIN_REVOKED:
1508                 case -NFS4ERR_STALE_STATEID:
1509                 case -NFS4ERR_BAD_STATEID:
1510                 case -NFS4ERR_EXPIRED:
1511                 case -NFS4ERR_NO_GRACE:
1512                 case -NFS4ERR_STALE_CLIENTID:
1513                 case -NFS4ERR_BADSESSION:
1514                 case -NFS4ERR_BADSLOT:
1515                 case -NFS4ERR_BAD_HIGH_SLOT:
1516                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1517                         goto out;
1518                 default:
1519                         pr_err("NFS: %s: unhandled error %d\n",
1520                                         __func__, status);
1521                 case -ENOMEM:
1522                 case -NFS4ERR_DENIED:
1523                 case -NFS4ERR_RECLAIM_BAD:
1524                 case -NFS4ERR_RECLAIM_CONFLICT:
1525                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1526                         status = 0;
1527                 }
1528                 spin_lock(&flctx->flc_lock);
1529         }
1530         if (list == &flctx->flc_posix) {
1531                 list = &flctx->flc_flock;
1532                 goto restart;
1533         }
1534         spin_unlock(&flctx->flc_lock);
1535 out:
1536         up_write(&nfsi->rwsem);
1537         return status;
1538 }
1539
1540 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1541 {
1542         struct nfs4_state *state;
1543         struct nfs4_lock_state *lock;
1544         int status = 0;
1545
1546         /* Note: we rely on the sp->so_states list being ordered
1547          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1548          * states first.
1549          * This is needed to ensure that the server won't give us any
1550          * read delegations that we have to return if, say, we are
1551          * recovering after a network partition or a reboot from a
1552          * server that doesn't support a grace period.
1553          */
1554         spin_lock(&sp->so_lock);
1555         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1556 restart:
1557         list_for_each_entry(state, &sp->so_states, open_states) {
1558                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1559                         continue;
1560                 if (!nfs4_valid_open_stateid(state))
1561                         continue;
1562                 if (state->state == 0)
1563                         continue;
1564                 atomic_inc(&state->count);
1565                 spin_unlock(&sp->so_lock);
1566                 status = ops->recover_open(sp, state);
1567                 if (status >= 0) {
1568                         status = nfs4_reclaim_locks(state, ops);
1569                         if (status >= 0) {
1570                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1571                                         spin_lock(&state->state_lock);
1572                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1573                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1574                                                         pr_warn_ratelimited("NFS: "
1575                                                                             "%s: Lock reclaim "
1576                                                                             "failed!\n", __func__);
1577                                         }
1578                                         spin_unlock(&state->state_lock);
1579                                 }
1580                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1581                                         &state->flags);
1582                                 nfs4_put_open_state(state);
1583                                 spin_lock(&sp->so_lock);
1584                                 goto restart;
1585                         }
1586                 }
1587                 switch (status) {
1588                         default:
1589                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1590                                         __func__, status);
1591                         case -ENOENT:
1592                         case -ENOMEM:
1593                         case -EACCES:
1594                         case -EROFS:
1595                         case -EIO:
1596                         case -ESTALE:
1597                                 /* Open state on this file cannot be recovered */
1598                                 nfs4_state_mark_recovery_failed(state, status);
1599                                 break;
1600                         case -EAGAIN:
1601                                 ssleep(1);
1602                         case -NFS4ERR_ADMIN_REVOKED:
1603                         case -NFS4ERR_STALE_STATEID:
1604                         case -NFS4ERR_OLD_STATEID:
1605                         case -NFS4ERR_BAD_STATEID:
1606                         case -NFS4ERR_RECLAIM_BAD:
1607                         case -NFS4ERR_RECLAIM_CONFLICT:
1608                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1609                                 break;
1610                         case -NFS4ERR_EXPIRED:
1611                         case -NFS4ERR_NO_GRACE:
1612                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1613                         case -NFS4ERR_STALE_CLIENTID:
1614                         case -NFS4ERR_BADSESSION:
1615                         case -NFS4ERR_BADSLOT:
1616                         case -NFS4ERR_BAD_HIGH_SLOT:
1617                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1618                                 goto out_err;
1619                 }
1620                 nfs4_put_open_state(state);
1621                 spin_lock(&sp->so_lock);
1622                 goto restart;
1623         }
1624         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1625         spin_unlock(&sp->so_lock);
1626         return 0;
1627 out_err:
1628         nfs4_put_open_state(state);
1629         spin_lock(&sp->so_lock);
1630         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1631         spin_unlock(&sp->so_lock);
1632         return status;
1633 }
1634
1635 static void nfs4_clear_open_state(struct nfs4_state *state)
1636 {
1637         struct nfs4_lock_state *lock;
1638
1639         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1640         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1641         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1642         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1643         spin_lock(&state->state_lock);
1644         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1645                 lock->ls_seqid.flags = 0;
1646                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1647         }
1648         spin_unlock(&state->state_lock);
1649 }
1650
1651 static void nfs4_reset_seqids(struct nfs_server *server,
1652         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1653 {
1654         struct nfs_client *clp = server->nfs_client;
1655         struct nfs4_state_owner *sp;
1656         struct rb_node *pos;
1657         struct nfs4_state *state;
1658
1659         spin_lock(&clp->cl_lock);
1660         for (pos = rb_first(&server->state_owners);
1661              pos != NULL;
1662              pos = rb_next(pos)) {
1663                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1664                 sp->so_seqid.flags = 0;
1665                 spin_lock(&sp->so_lock);
1666                 list_for_each_entry(state, &sp->so_states, open_states) {
1667                         if (mark_reclaim(clp, state))
1668                                 nfs4_clear_open_state(state);
1669                 }
1670                 spin_unlock(&sp->so_lock);
1671         }
1672         spin_unlock(&clp->cl_lock);
1673 }
1674
1675 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1676         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1677 {
1678         struct nfs_server *server;
1679
1680         rcu_read_lock();
1681         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1682                 nfs4_reset_seqids(server, mark_reclaim);
1683         rcu_read_unlock();
1684 }
1685
1686 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1687 {
1688         /* Mark all delegations for reclaim */
1689         nfs_delegation_mark_reclaim(clp);
1690         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1691 }
1692
1693 static int nfs4_reclaim_complete(struct nfs_client *clp,
1694                                  const struct nfs4_state_recovery_ops *ops,
1695                                  struct rpc_cred *cred)
1696 {
1697         /* Notify the server we're done reclaiming our state */
1698         if (ops->reclaim_complete)
1699                 return ops->reclaim_complete(clp, cred);
1700         return 0;
1701 }
1702
1703 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1704 {
1705         struct nfs_client *clp = server->nfs_client;
1706         struct nfs4_state_owner *sp;
1707         struct rb_node *pos;
1708         struct nfs4_state *state;
1709
1710         spin_lock(&clp->cl_lock);
1711         for (pos = rb_first(&server->state_owners);
1712              pos != NULL;
1713              pos = rb_next(pos)) {
1714                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1715                 spin_lock(&sp->so_lock);
1716                 list_for_each_entry(state, &sp->so_states, open_states) {
1717                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1718                                                 &state->flags))
1719                                 continue;
1720                         nfs4_state_mark_reclaim_nograce(clp, state);
1721                 }
1722                 spin_unlock(&sp->so_lock);
1723         }
1724         spin_unlock(&clp->cl_lock);
1725 }
1726
1727 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1728 {
1729         struct nfs_server *server;
1730
1731         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1732                 return 0;
1733
1734         rcu_read_lock();
1735         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1736                 nfs4_clear_reclaim_server(server);
1737         rcu_read_unlock();
1738
1739         nfs_delegation_reap_unclaimed(clp);
1740         return 1;
1741 }
1742
1743 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1744 {
1745         const struct nfs4_state_recovery_ops *ops;
1746         struct rpc_cred *cred;
1747         int err;
1748
1749         if (!nfs4_state_clear_reclaim_reboot(clp))
1750                 return;
1751         ops = clp->cl_mvops->reboot_recovery_ops;
1752         cred = nfs4_get_clid_cred(clp);
1753         err = nfs4_reclaim_complete(clp, ops, cred);
1754         put_rpccred(cred);
1755         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1756                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1757 }
1758
1759 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1760 {
1761         nfs_mark_test_expired_all_delegations(clp);
1762         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1763 }
1764
1765 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1766 {
1767         switch (error) {
1768                 case 0:
1769                         break;
1770                 case -NFS4ERR_CB_PATH_DOWN:
1771                         nfs40_handle_cb_pathdown(clp);
1772                         break;
1773                 case -NFS4ERR_NO_GRACE:
1774                         nfs4_state_end_reclaim_reboot(clp);
1775                         break;
1776                 case -NFS4ERR_STALE_CLIENTID:
1777                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1778                         nfs4_state_start_reclaim_reboot(clp);
1779                         break;
1780                 case -NFS4ERR_EXPIRED:
1781                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1782                         nfs4_state_start_reclaim_nograce(clp);
1783                         break;
1784                 case -NFS4ERR_BADSESSION:
1785                 case -NFS4ERR_BADSLOT:
1786                 case -NFS4ERR_BAD_HIGH_SLOT:
1787                 case -NFS4ERR_DEADSESSION:
1788                 case -NFS4ERR_SEQ_FALSE_RETRY:
1789                 case -NFS4ERR_SEQ_MISORDERED:
1790                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1791                         /* Zero session reset errors */
1792                         break;
1793                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1794                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1795                         break;
1796                 default:
1797                         dprintk("%s: failed to handle error %d for server %s\n",
1798                                         __func__, error, clp->cl_hostname);
1799                         return error;
1800         }
1801         dprintk("%s: handled error %d for server %s\n", __func__, error,
1802                         clp->cl_hostname);
1803         return 0;
1804 }
1805
1806 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1807 {
1808         struct nfs4_state_owner *sp;
1809         struct nfs_server *server;
1810         struct rb_node *pos;
1811         int status = 0;
1812
1813 restart:
1814         rcu_read_lock();
1815         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1816                 nfs4_purge_state_owners(server);
1817                 spin_lock(&clp->cl_lock);
1818                 for (pos = rb_first(&server->state_owners);
1819                      pos != NULL;
1820                      pos = rb_next(pos)) {
1821                         sp = rb_entry(pos,
1822                                 struct nfs4_state_owner, so_server_node);
1823                         if (!test_and_clear_bit(ops->owner_flag_bit,
1824                                                         &sp->so_flags))
1825                                 continue;
1826                         if (!atomic_inc_not_zero(&sp->so_count))
1827                                 continue;
1828                         spin_unlock(&clp->cl_lock);
1829                         rcu_read_unlock();
1830
1831                         status = nfs4_reclaim_open_state(sp, ops);
1832                         if (status < 0) {
1833                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1834                                 nfs4_put_state_owner(sp);
1835                                 status = nfs4_recovery_handle_error(clp, status);
1836                                 return (status != 0) ? status : -EAGAIN;
1837                         }
1838
1839                         nfs4_put_state_owner(sp);
1840                         goto restart;
1841                 }
1842                 spin_unlock(&clp->cl_lock);
1843         }
1844         rcu_read_unlock();
1845         return 0;
1846 }
1847
1848 static int nfs4_check_lease(struct nfs_client *clp)
1849 {
1850         struct rpc_cred *cred;
1851         const struct nfs4_state_maintenance_ops *ops =
1852                 clp->cl_mvops->state_renewal_ops;
1853         int status;
1854
1855         /* Is the client already known to have an expired lease? */
1856         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1857                 return 0;
1858         spin_lock(&clp->cl_lock);
1859         cred = ops->get_state_renewal_cred_locked(clp);
1860         spin_unlock(&clp->cl_lock);
1861         if (cred == NULL) {
1862                 cred = nfs4_get_clid_cred(clp);
1863                 status = -ENOKEY;
1864                 if (cred == NULL)
1865                         goto out;
1866         }
1867         status = ops->renew_lease(clp, cred);
1868         put_rpccred(cred);
1869         if (status == -ETIMEDOUT) {
1870                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1871                 return 0;
1872         }
1873 out:
1874         return nfs4_recovery_handle_error(clp, status);
1875 }
1876
1877 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1878  * and for recoverable errors on EXCHANGE_ID for v4.1
1879  */
1880 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1881 {
1882         switch (status) {
1883         case -NFS4ERR_SEQ_MISORDERED:
1884                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1885                         return -ESERVERFAULT;
1886                 /* Lease confirmation error: retry after purging the lease */
1887                 ssleep(1);
1888                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1889                 break;
1890         case -NFS4ERR_STALE_CLIENTID:
1891                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1892                 nfs4_state_start_reclaim_reboot(clp);
1893                 break;
1894         case -NFS4ERR_CLID_INUSE:
1895                 pr_err("NFS: Server %s reports our clientid is in use\n",
1896                         clp->cl_hostname);
1897                 nfs_mark_client_ready(clp, -EPERM);
1898                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1899                 return -EPERM;
1900         case -EACCES:
1901         case -NFS4ERR_DELAY:
1902         case -ETIMEDOUT:
1903         case -EAGAIN:
1904                 ssleep(1);
1905                 break;
1906
1907         case -NFS4ERR_MINOR_VERS_MISMATCH:
1908                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1909                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1910                 dprintk("%s: exit with error %d for server %s\n",
1911                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1912                 return -EPROTONOSUPPORT;
1913         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1914                                  * in nfs4_exchange_id */
1915         default:
1916                 dprintk("%s: exit with error %d for server %s\n", __func__,
1917                                 status, clp->cl_hostname);
1918                 return status;
1919         }
1920         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1921         dprintk("%s: handled error %d for server %s\n", __func__, status,
1922                         clp->cl_hostname);
1923         return 0;
1924 }
1925
1926 static int nfs4_establish_lease(struct nfs_client *clp)
1927 {
1928         struct rpc_cred *cred;
1929         const struct nfs4_state_recovery_ops *ops =
1930                 clp->cl_mvops->reboot_recovery_ops;
1931         int status;
1932
1933         nfs4_begin_drain_session(clp);
1934         cred = nfs4_get_clid_cred(clp);
1935         if (cred == NULL)
1936                 return -ENOENT;
1937         status = ops->establish_clid(clp, cred);
1938         put_rpccred(cred);
1939         if (status != 0)
1940                 return status;
1941         pnfs_destroy_all_layouts(clp);
1942         return 0;
1943 }
1944
1945 /*
1946  * Returns zero or a negative errno.  NFS4ERR values are converted
1947  * to local errno values.
1948  */
1949 static int nfs4_reclaim_lease(struct nfs_client *clp)
1950 {
1951         int status;
1952
1953         status = nfs4_establish_lease(clp);
1954         if (status < 0)
1955                 return nfs4_handle_reclaim_lease_error(clp, status);
1956         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1957                 nfs4_state_start_reclaim_nograce(clp);
1958         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1959                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1960         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1961         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1962         return 0;
1963 }
1964
1965 static int nfs4_purge_lease(struct nfs_client *clp)
1966 {
1967         int status;
1968
1969         status = nfs4_establish_lease(clp);
1970         if (status < 0)
1971                 return nfs4_handle_reclaim_lease_error(clp, status);
1972         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1973         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1974         nfs4_state_start_reclaim_nograce(clp);
1975         return 0;
1976 }
1977
1978 /*
1979  * Try remote migration of one FSID from a source server to a
1980  * destination server.  The source server provides a list of
1981  * potential destinations.
1982  *
1983  * Returns zero or a negative NFS4ERR status code.
1984  */
1985 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1986 {
1987         struct nfs_client *clp = server->nfs_client;
1988         struct nfs4_fs_locations *locations = NULL;
1989         struct inode *inode;
1990         struct page *page;
1991         int status, result;
1992
1993         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1994                         (unsigned long long)server->fsid.major,
1995                         (unsigned long long)server->fsid.minor,
1996                         clp->cl_hostname);
1997
1998         result = 0;
1999         page = alloc_page(GFP_KERNEL);
2000         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2001         if (page == NULL || locations == NULL) {
2002                 dprintk("<-- %s: no memory\n", __func__);
2003                 goto out;
2004         }
2005
2006         inode = d_inode(server->super->s_root);
2007         result = nfs4_proc_get_locations(inode, locations, page, cred);
2008         if (result) {
2009                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2010                         __func__, result);
2011                 goto out;
2012         }
2013
2014         result = -NFS4ERR_NXIO;
2015         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2016                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2017                         __func__);
2018                 goto out;
2019         }
2020
2021         nfs4_begin_drain_session(clp);
2022
2023         status = nfs4_replace_transport(server, locations);
2024         if (status != 0) {
2025                 dprintk("<-- %s: failed to replace transport: %d\n",
2026                         __func__, status);
2027                 goto out;
2028         }
2029
2030         result = 0;
2031         dprintk("<-- %s: migration succeeded\n", __func__);
2032
2033 out:
2034         if (page != NULL)
2035                 __free_page(page);
2036         kfree(locations);
2037         if (result) {
2038                 pr_err("NFS: migration recovery failed (server %s)\n",
2039                                 clp->cl_hostname);
2040                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2041         }
2042         return result;
2043 }
2044
2045 /*
2046  * Returns zero or a negative NFS4ERR status code.
2047  */
2048 static int nfs4_handle_migration(struct nfs_client *clp)
2049 {
2050         const struct nfs4_state_maintenance_ops *ops =
2051                                 clp->cl_mvops->state_renewal_ops;
2052         struct nfs_server *server;
2053         struct rpc_cred *cred;
2054
2055         dprintk("%s: migration reported on \"%s\"\n", __func__,
2056                         clp->cl_hostname);
2057
2058         spin_lock(&clp->cl_lock);
2059         cred = ops->get_state_renewal_cred_locked(clp);
2060         spin_unlock(&clp->cl_lock);
2061         if (cred == NULL)
2062                 return -NFS4ERR_NOENT;
2063
2064         clp->cl_mig_gen++;
2065 restart:
2066         rcu_read_lock();
2067         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2068                 int status;
2069
2070                 if (server->mig_gen == clp->cl_mig_gen)
2071                         continue;
2072                 server->mig_gen = clp->cl_mig_gen;
2073
2074                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2075                                                 &server->mig_status))
2076                         continue;
2077
2078                 rcu_read_unlock();
2079                 status = nfs4_try_migration(server, cred);
2080                 if (status < 0) {
2081                         put_rpccred(cred);
2082                         return status;
2083                 }
2084                 goto restart;
2085         }
2086         rcu_read_unlock();
2087         put_rpccred(cred);
2088         return 0;
2089 }
2090
2091 /*
2092  * Test each nfs_server on the clp's cl_superblocks list to see
2093  * if it's moved to another server.  Stop when the server no longer
2094  * returns NFS4ERR_LEASE_MOVED.
2095  */
2096 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2097 {
2098         const struct nfs4_state_maintenance_ops *ops =
2099                                 clp->cl_mvops->state_renewal_ops;
2100         struct nfs_server *server;
2101         struct rpc_cred *cred;
2102
2103         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2104                         clp->cl_hostname);
2105
2106         spin_lock(&clp->cl_lock);
2107         cred = ops->get_state_renewal_cred_locked(clp);
2108         spin_unlock(&clp->cl_lock);
2109         if (cred == NULL)
2110                 return -NFS4ERR_NOENT;
2111
2112         clp->cl_mig_gen++;
2113 restart:
2114         rcu_read_lock();
2115         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2116                 struct inode *inode;
2117                 int status;
2118
2119                 if (server->mig_gen == clp->cl_mig_gen)
2120                         continue;
2121                 server->mig_gen = clp->cl_mig_gen;
2122
2123                 rcu_read_unlock();
2124
2125                 inode = d_inode(server->super->s_root);
2126                 status = nfs4_proc_fsid_present(inode, cred);
2127                 if (status != -NFS4ERR_MOVED)
2128                         goto restart;   /* wasn't this one */
2129                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2130                         goto restart;   /* there are more */
2131                 goto out;
2132         }
2133         rcu_read_unlock();
2134
2135 out:
2136         put_rpccred(cred);
2137         return 0;
2138 }
2139
2140 /**
2141  * nfs4_discover_server_trunking - Detect server IP address trunking
2142  *
2143  * @clp: nfs_client under test
2144  * @result: OUT: found nfs_client, or clp
2145  *
2146  * Returns zero or a negative errno.  If zero is returned,
2147  * an nfs_client pointer is planted in "result".
2148  *
2149  * Note: since we are invoked in process context, and
2150  * not from inside the state manager, we cannot use
2151  * nfs4_handle_reclaim_lease_error().
2152  */
2153 int nfs4_discover_server_trunking(struct nfs_client *clp,
2154                                   struct nfs_client **result)
2155 {
2156         const struct nfs4_state_recovery_ops *ops =
2157                                 clp->cl_mvops->reboot_recovery_ops;
2158         struct rpc_clnt *clnt;
2159         struct rpc_cred *cred;
2160         int i, status;
2161
2162         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2163
2164         clnt = clp->cl_rpcclient;
2165         i = 0;
2166
2167         mutex_lock(&nfs_clid_init_mutex);
2168 again:
2169         status  = -ENOENT;
2170         cred = nfs4_get_clid_cred(clp);
2171         if (cred == NULL)
2172                 goto out_unlock;
2173
2174         status = ops->detect_trunking(clp, result, cred);
2175         put_rpccred(cred);
2176         switch (status) {
2177         case 0:
2178         case -EINTR:
2179         case -ERESTARTSYS:
2180                 break;
2181         case -ETIMEDOUT:
2182                 if (clnt->cl_softrtry)
2183                         break;
2184         case -NFS4ERR_DELAY:
2185         case -EAGAIN:
2186                 ssleep(1);
2187         case -NFS4ERR_STALE_CLIENTID:
2188                 dprintk("NFS: %s after status %d, retrying\n",
2189                         __func__, status);
2190                 goto again;
2191         case -EACCES:
2192                 if (i++ == 0) {
2193                         nfs4_root_machine_cred(clp);
2194                         goto again;
2195                 }
2196                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2197                         break;
2198         case -NFS4ERR_CLID_INUSE:
2199         case -NFS4ERR_WRONGSEC:
2200                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2201                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2202                         status = -EPERM;
2203                         break;
2204                 }
2205                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2206                 if (IS_ERR(clnt)) {
2207                         status = PTR_ERR(clnt);
2208                         break;
2209                 }
2210                 /* Note: this is safe because we haven't yet marked the
2211                  * client as ready, so we are the only user of
2212                  * clp->cl_rpcclient
2213                  */
2214                 clnt = xchg(&clp->cl_rpcclient, clnt);
2215                 rpc_shutdown_client(clnt);
2216                 clnt = clp->cl_rpcclient;
2217                 goto again;
2218
2219         case -NFS4ERR_MINOR_VERS_MISMATCH:
2220                 status = -EPROTONOSUPPORT;
2221                 break;
2222
2223         case -EKEYEXPIRED:
2224         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2225                                  * in nfs4_exchange_id */
2226                 status = -EKEYEXPIRED;
2227                 break;
2228         default:
2229                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2230                                 __func__, status);
2231                 status = -EIO;
2232         }
2233
2234 out_unlock:
2235         mutex_unlock(&nfs_clid_init_mutex);
2236         dprintk("NFS: %s: status = %d\n", __func__, status);
2237         return status;
2238 }
2239
2240 #ifdef CONFIG_NFS_V4_1
2241 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2242 {
2243         struct nfs_client *clp = session->clp;
2244
2245         switch (err) {
2246         default:
2247                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2248                 break;
2249         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2250                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2251         }
2252         nfs4_schedule_state_manager(clp);
2253 }
2254 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2255
2256 void nfs41_notify_server(struct nfs_client *clp)
2257 {
2258         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2259         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2260         nfs4_schedule_state_manager(clp);
2261 }
2262
2263 static void nfs4_reset_all_state(struct nfs_client *clp)
2264 {
2265         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2266                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2267                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2268                 nfs4_state_start_reclaim_nograce(clp);
2269                 dprintk("%s: scheduling reset of all state for server %s!\n",
2270                                 __func__, clp->cl_hostname);
2271                 nfs4_schedule_state_manager(clp);
2272         }
2273 }
2274
2275 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2276 {
2277         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2278                 nfs4_state_start_reclaim_reboot(clp);
2279                 dprintk("%s: server %s rebooted!\n", __func__,
2280                                 clp->cl_hostname);
2281                 nfs4_schedule_state_manager(clp);
2282         }
2283 }
2284
2285 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2286 {
2287         nfs4_reset_all_state(clp);
2288         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2289 }
2290
2291 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2292 {
2293         nfs4_state_start_reclaim_nograce(clp);
2294         nfs4_schedule_state_manager(clp);
2295
2296         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2297 }
2298
2299 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2300 {
2301         /* FIXME: For now, we destroy all layouts. */
2302         pnfs_destroy_all_layouts(clp);
2303         /* FIXME: For now, we test all delegations+open state+locks. */
2304         nfs41_handle_some_state_revoked(clp);
2305         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2306                         clp->cl_hostname);
2307 }
2308
2309 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2310 {
2311         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2312         nfs4_schedule_state_manager(clp);
2313
2314         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2315                         clp->cl_hostname);
2316 }
2317
2318 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2319 {
2320         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2321                 &clp->cl_state) == 0)
2322                 nfs4_schedule_state_manager(clp);
2323 }
2324
2325 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2326                 bool recovery)
2327 {
2328         if (!flags)
2329                 return;
2330
2331         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2332                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2333         /*
2334          * If we're called from the state manager thread, then assume we're
2335          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2336          * Those flags are expected to remain set until we're done
2337          * recovering (see RFC5661, section 18.46.3).
2338          */
2339         if (recovery)
2340                 goto out_recovery;
2341
2342         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2343                 nfs41_handle_server_reboot(clp);
2344         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2345                 nfs41_handle_all_state_revoked(clp);
2346         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2347                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2348                 nfs41_handle_some_state_revoked(clp);
2349         if (flags & SEQ4_STATUS_LEASE_MOVED)
2350                 nfs4_schedule_lease_moved_recovery(clp);
2351         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2352                 nfs41_handle_recallable_state_revoked(clp);
2353 out_recovery:
2354         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2355                 nfs41_handle_backchannel_fault(clp);
2356         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2357                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2358                 nfs41_handle_cb_path_down(clp);
2359 }
2360
2361 static int nfs4_reset_session(struct nfs_client *clp)
2362 {
2363         struct rpc_cred *cred;
2364         int status;
2365
2366         if (!nfs4_has_session(clp))
2367                 return 0;
2368         nfs4_begin_drain_session(clp);
2369         cred = nfs4_get_clid_cred(clp);
2370         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2371         switch (status) {
2372         case 0:
2373         case -NFS4ERR_BADSESSION:
2374         case -NFS4ERR_DEADSESSION:
2375                 break;
2376         case -NFS4ERR_BACK_CHAN_BUSY:
2377         case -NFS4ERR_DELAY:
2378                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2379                 status = 0;
2380                 ssleep(1);
2381                 goto out;
2382         default:
2383                 status = nfs4_recovery_handle_error(clp, status);
2384                 goto out;
2385         }
2386
2387         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2388         status = nfs4_proc_create_session(clp, cred);
2389         if (status) {
2390                 dprintk("%s: session reset failed with status %d for server %s!\n",
2391                         __func__, status, clp->cl_hostname);
2392                 status = nfs4_handle_reclaim_lease_error(clp, status);
2393                 goto out;
2394         }
2395         nfs41_finish_session_reset(clp);
2396         dprintk("%s: session reset was successful for server %s!\n",
2397                         __func__, clp->cl_hostname);
2398 out:
2399         if (cred)
2400                 put_rpccred(cred);
2401         return status;
2402 }
2403
2404 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2405 {
2406         struct rpc_cred *cred;
2407         int ret;
2408
2409         if (!nfs4_has_session(clp))
2410                 return 0;
2411         nfs4_begin_drain_session(clp);
2412         cred = nfs4_get_clid_cred(clp);
2413         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2414         if (cred)
2415                 put_rpccred(cred);
2416         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2417         switch (ret) {
2418         case 0:
2419                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2420                         __func__, clp->cl_hostname);
2421                 break;
2422         case -NFS4ERR_DELAY:
2423                 ssleep(1);
2424                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2425                 break;
2426         default:
2427                 return nfs4_recovery_handle_error(clp, ret);
2428         }
2429         return 0;
2430 }
2431 #else /* CONFIG_NFS_V4_1 */
2432 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2433
2434 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2435 {
2436         return 0;
2437 }
2438 #endif /* CONFIG_NFS_V4_1 */
2439
2440 static void nfs4_state_manager(struct nfs_client *clp)
2441 {
2442         int status = 0;
2443         const char *section = "", *section_sep = "";
2444
2445         /* Ensure exclusive access to NFSv4 state */
2446         do {
2447                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2448                         section = "purge state";
2449                         status = nfs4_purge_lease(clp);
2450                         if (status < 0)
2451                                 goto out_error;
2452                         continue;
2453                 }
2454
2455                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2456                         section = "lease expired";
2457                         /* We're going to have to re-establish a clientid */
2458                         status = nfs4_reclaim_lease(clp);
2459                         if (status < 0)
2460                                 goto out_error;
2461                         continue;
2462                 }
2463
2464                 /* Initialize or reset the session */
2465                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2466                         section = "reset session";
2467                         status = nfs4_reset_session(clp);
2468                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2469                                 continue;
2470                         if (status < 0)
2471                                 goto out_error;
2472                 }
2473
2474                 /* Send BIND_CONN_TO_SESSION */
2475                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2476                                 &clp->cl_state)) {
2477                         section = "bind conn to session";
2478                         status = nfs4_bind_conn_to_session(clp);
2479                         if (status < 0)
2480                                 goto out_error;
2481                         continue;
2482                 }
2483
2484                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2485                         section = "check lease";
2486                         status = nfs4_check_lease(clp);
2487                         if (status < 0)
2488                                 goto out_error;
2489                         continue;
2490                 }
2491
2492                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2493                         section = "migration";
2494                         status = nfs4_handle_migration(clp);
2495                         if (status < 0)
2496                                 goto out_error;
2497                 }
2498
2499                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2500                         section = "lease moved";
2501                         status = nfs4_handle_lease_moved(clp);
2502                         if (status < 0)
2503                                 goto out_error;
2504                 }
2505
2506                 /* First recover reboot state... */
2507                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2508                         section = "reclaim reboot";
2509                         status = nfs4_do_reclaim(clp,
2510                                 clp->cl_mvops->reboot_recovery_ops);
2511                         if (status == -EAGAIN)
2512                                 continue;
2513                         if (status < 0)
2514                                 goto out_error;
2515                         nfs4_state_end_reclaim_reboot(clp);
2516                 }
2517
2518                 /* Detect expired delegations... */
2519                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2520                         section = "detect expired delegations";
2521                         nfs_reap_expired_delegations(clp);
2522                         continue;
2523                 }
2524
2525                 /* Now recover expired state... */
2526                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2527                         section = "reclaim nograce";
2528                         status = nfs4_do_reclaim(clp,
2529                                 clp->cl_mvops->nograce_recovery_ops);
2530                         if (status == -EAGAIN)
2531                                 continue;
2532                         if (status < 0)
2533                                 goto out_error;
2534                 }
2535
2536                 nfs4_end_drain_session(clp);
2537                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2538                         nfs_client_return_marked_delegations(clp);
2539                         continue;
2540                 }
2541
2542                 nfs4_clear_state_manager_bit(clp);
2543                 /* Did we race with an attempt to give us more work? */
2544                 if (clp->cl_state == 0)
2545                         break;
2546                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2547                         break;
2548         } while (refcount_read(&clp->cl_count) > 1);
2549         return;
2550 out_error:
2551         if (strlen(section))
2552                 section_sep = ": ";
2553         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2554                         " with error %d\n", section_sep, section,
2555                         clp->cl_hostname, -status);
2556         ssleep(1);
2557         nfs4_end_drain_session(clp);
2558         nfs4_clear_state_manager_bit(clp);
2559 }
2560
2561 static int nfs4_run_state_manager(void *ptr)
2562 {
2563         struct nfs_client *clp = ptr;
2564
2565         allow_signal(SIGKILL);
2566         nfs4_state_manager(clp);
2567         nfs_put_client(clp);
2568         module_put_and_exit(0);
2569         return 0;
2570 }
2571
2572 /*
2573  * Local variables:
2574  *  c-basic-offset: 8
2575  * End:
2576  */