fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid;
  69 static DEFINE_MUTEX(nfs_clid_init_mutex);
  70
  71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  72 {
  73         struct nfs4_setclientid_res clid = {
  74                 .clientid = clp->cl_clientid,
  75                 .confirm = clp->cl_confirm,
  76         };
  77         unsigned short port;
  78         int status;
  79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  80
  81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  82                 goto do_confirm;
  83         port = nn->nfs_callback_tcpport;
  84         if (clp->cl_addr.ss_family == AF_INET6)
  85                 port = nn->nfs_callback_tcpport6;
  86
  87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  88         if (status != 0)
  89                 goto out;
  90         clp->cl_clientid = clid.clientid;
  91         clp->cl_confirm = clid.confirm;
  92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  93 do_confirm:
  94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  95         if (status != 0)
  96                 goto out;
  97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98         nfs4_schedule_state_renewal(clp);
  99 out:
 100         return status;
 101 }
 102
 103 /**
 104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 105  *
 106  * @clp: nfs_client under test
 107  * @result: OUT: found nfs_client, or clp
 108  * @cred: credential to use for trunking test
 109  *
 110  * Returns zero, a negative errno, or a negative NFS4ERR status.
 111  * If zero is returned, an nfs_client pointer is planted in
 112  * "result".
 113  *
 114  * Note: The returned client may not yet be marked ready.
 115  */
 116 int nfs40_discover_server_trunking(struct nfs_client *clp,
 117                                    struct nfs_client **result,
 118                                    struct rpc_cred *cred)
 119 {
 120         struct nfs4_setclientid_res clid = {
 121                 .clientid = clp->cl_clientid,
 122                 .confirm = clp->cl_confirm,
 123         };
 124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 125         unsigned short port;
 126         int status;
 127
 128         port = nn->nfs_callback_tcpport;
 129         if (clp->cl_addr.ss_family == AF_INET6)
 130                 port = nn->nfs_callback_tcpport6;
 131
 132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 133         if (status != 0)
 134                 goto out;
 135         clp->cl_clientid = clid.clientid;
 136         clp->cl_confirm = clid.confirm;
 137
 138         status = nfs40_walk_client_list(clp, result, cred);
 139         if (status == 0) {
 140                 /* Sustain the lease, even if it's empty.  If the clientid4
 141                  * goes stale it's of no use for trunking discovery. */
 142                 nfs4_schedule_state_renewal(*result);
 143
 144                 /* If the client state need to recover, do it. */
 145                 if (clp->cl_state)
 146                         nfs4_schedule_state_manager(clp);
 147         }
 148 out:
 149         return status;
 150 }
 151
 152 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 153 {
 154         struct rpc_cred *cred = NULL;
 155
 156         if (clp->cl_machine_cred != NULL)
 157                 cred = get_rpccred(clp->cl_machine_cred);
 158         return cred;
 159 }
 160
 161 static void nfs4_root_machine_cred(struct nfs_client *clp)
 162 {
 163         struct rpc_cred *cred, *new;
 164
 165         new = rpc_lookup_machine_cred(NULL);
 166         spin_lock(&clp->cl_lock);
 167         cred = clp->cl_machine_cred;
 168         clp->cl_machine_cred = new;
 169         spin_unlock(&clp->cl_lock);
 170         if (cred != NULL)
 171                 put_rpccred(cred);
 172 }
 173
 174 static struct rpc_cred *
 175 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 176 {
 177         struct rpc_cred *cred = NULL;
 178         struct nfs4_state_owner *sp;
 179         struct rb_node *pos;
 180
 181         for (pos = rb_first(&server->state_owners);
 182              pos != NULL;
 183              pos = rb_next(pos)) {
 184                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 185                 if (list_empty(&sp->so_states))
 186                         continue;
 187                 cred = get_rpccred(sp->so_cred);
 188                 break;
 189         }
 190         return cred;
 191 }
 192
 193 /**
 194  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 195  * @clp: client state handle
 196  *
 197  * Returns an rpc_cred with reference count bumped, or NULL.
 198  * Caller must hold clp->cl_lock.
 199  */
 200 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 201 {
 202         struct rpc_cred *cred = NULL;
 203         struct nfs_server *server;
 204
 205         /* Use machine credentials if available */
 206         cred = nfs4_get_machine_cred_locked(clp);
 207         if (cred != NULL)
 208                 goto out;
 209
 210         rcu_read_lock();
 211         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 212                 cred = nfs4_get_renew_cred_server_locked(server);
 213                 if (cred != NULL)
 214                         break;
 215         }
 216         rcu_read_unlock();
 217
 218 out:
 219         return cred;
 220 }
 221
 222 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 223 {
 224         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 225                 spin_lock(&tbl->slot_tbl_lock);
 226                 nfs41_wake_slot_table(tbl);
 227                 spin_unlock(&tbl->slot_tbl_lock);
 228         }
 229 }
 230
 231 static void nfs4_end_drain_session(struct nfs_client *clp)
 232 {
 233         struct nfs4_session *ses = clp->cl_session;
 234
 235         if (clp->cl_slot_tbl) {
 236                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 237                 return;
 238         }
 239
 240         if (ses != NULL) {
 241                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 242                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 243         }
 244 }
 245
 246 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 247 {
 248         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 249         spin_lock(&tbl->slot_tbl_lock);
 250         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 251                 reinit_completion(&tbl->complete);
 252                 spin_unlock(&tbl->slot_tbl_lock);
 253                 return wait_for_completion_interruptible(&tbl->complete);
 254         }
 255         spin_unlock(&tbl->slot_tbl_lock);
 256         return 0;
 257 }
 258
 259 static int nfs4_begin_drain_session(struct nfs_client *clp)
 260 {
 261         struct nfs4_session *ses = clp->cl_session;
 262         int ret = 0;
 263
 264         if (clp->cl_slot_tbl)
 265                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 266
 267         /* back channel */
 268         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 269         if (ret)
 270                 return ret;
 271         /* fore channel */
 272         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 273 }
 274
 275 #if defined(CONFIG_NFS_V4_1)
 276
 277 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 278 {
 279         int status;
 280         struct nfs_fsinfo fsinfo;
 281
 282         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 283                 nfs4_schedule_state_renewal(clp);
 284                 return 0;
 285         }
 286
 287         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 288         if (status == 0) {
 289                 /* Update lease time and schedule renewal */
 290                 spin_lock(&clp->cl_lock);
 291                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 292                 clp->cl_last_renewal = jiffies;
 293                 spin_unlock(&clp->cl_lock);
 294
 295                 nfs4_schedule_state_renewal(clp);
 296         }
 297
 298         return status;
 299 }
 300
 301 static void nfs41_finish_session_reset(struct nfs_client *clp)
 302 {
 303         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 304         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 305         /* create_session negotiated new slot table */
 306         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 307         nfs41_setup_state_renewal(clp);
 308 }
 309
 310 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 311 {
 312         int status;
 313
 314         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 315                 goto do_confirm;
 316         status = nfs4_proc_exchange_id(clp, cred);
 317         if (status != 0)
 318                 goto out;
 319         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 320 do_confirm:
 321         status = nfs4_proc_create_session(clp, cred);
 322         if (status != 0)
 323                 goto out;
 324         nfs41_finish_session_reset(clp);
 325         nfs_mark_client_ready(clp, NFS_CS_READY);
 326 out:
 327         return status;
 328 }
 329
 330 /**
 331  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 332  *
 333  * @clp: nfs_client under test
 334  * @result: OUT: found nfs_client, or clp
 335  * @cred: credential to use for trunking test
 336  *
 337  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 338  * If NFS4_OK is returned, an nfs_client pointer is planted in
 339  * "result".
 340  *
 341  * Note: The returned client may not yet be marked ready.
 342  */
 343 int nfs41_discover_server_trunking(struct nfs_client *clp,
 344                                    struct nfs_client **result,
 345                                    struct rpc_cred *cred)
 346 {
 347         int status;
 348
 349         status = nfs4_proc_exchange_id(clp, cred);
 350         if (status != NFS4_OK)
 351                 return status;
 352
 353         status = nfs41_walk_client_list(clp, result, cred);
 354         if (status < 0)
 355                 return status;
 356         if (clp != *result)
 357                 return 0;
 358
 359         /* Purge state if the client id was established in a prior instance */
 360         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
 361                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 362         else
 363                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 364         nfs4_schedule_state_manager(clp);
 365         status = nfs_wait_client_init_complete(clp);
 366         if (status < 0)
 367                 nfs_put_client(clp);
 368         return status;
 369 }
 370
 371 #endif /* CONFIG_NFS_V4_1 */
 372
 373 /**
 374  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 375  * @clp: client state handle
 376  *
 377  * Returns an rpc_cred with reference count bumped, or NULL.
 378  */
 379 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 380 {
 381         struct rpc_cred *cred;
 382
 383         spin_lock(&clp->cl_lock);
 384         cred = nfs4_get_machine_cred_locked(clp);
 385         spin_unlock(&clp->cl_lock);
 386         return cred;
 387 }
 388
 389 static struct nfs4_state_owner *
 390 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 391 {
 392         struct rb_node **p = &server->state_owners.rb_node,
 393                        *parent = NULL;
 394         struct nfs4_state_owner *sp;
 395
 396         while (*p != NULL) {
 397                 parent = *p;
 398                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 399
 400                 if (cred < sp->so_cred)
 401                         p = &parent->rb_left;
 402                 else if (cred > sp->so_cred)
 403                         p = &parent->rb_right;
 404                 else {
 405                         if (!list_empty(&sp->so_lru))
 406                                 list_del_init(&sp->so_lru);
 407                         atomic_inc(&sp->so_count);
 408                         return sp;
 409                 }
 410         }
 411         return NULL;
 412 }
 413
 414 static struct nfs4_state_owner *
 415 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 416 {
 417         struct nfs_server *server = new->so_server;
 418         struct rb_node **p = &server->state_owners.rb_node,
 419                        *parent = NULL;
 420         struct nfs4_state_owner *sp;
 421         int err;
 422
 423         while (*p != NULL) {
 424                 parent = *p;
 425                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 426
 427                 if (new->so_cred < sp->so_cred)
 428                         p = &parent->rb_left;
 429                 else if (new->so_cred > sp->so_cred)
 430                         p = &parent->rb_right;
 431                 else {
 432                         if (!list_empty(&sp->so_lru))
 433                                 list_del_init(&sp->so_lru);
 434                         atomic_inc(&sp->so_count);
 435                         return sp;
 436                 }
 437         }
 438         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 439         if (err)
 440                 return ERR_PTR(err);
 441         rb_link_node(&new->so_server_node, parent, p);
 442         rb_insert_color(&new->so_server_node, &server->state_owners);
 443         return new;
 444 }
 445
 446 static void
 447 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 448 {
 449         struct nfs_server *server = sp->so_server;
 450
 451         if (!RB_EMPTY_NODE(&sp->so_server_node))
 452                 rb_erase(&sp->so_server_node, &server->state_owners);
 453         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 454 }
 455
 456 static void
 457 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 458 {
 459         sc->create_time = ktime_get();
 460         sc->flags = 0;
 461         sc->counter = 0;
 462         spin_lock_init(&sc->lock);
 463         INIT_LIST_HEAD(&sc->list);
 464         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 465 }
 466
 467 static void
 468 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 469 {
 470         rpc_destroy_wait_queue(&sc->wait);
 471 }
 472
 473 /*
 474  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 475  * create a new state_owner.
 476  *
 477  */
 478 static struct nfs4_state_owner *
 479 nfs4_alloc_state_owner(struct nfs_server *server,
 480                 struct rpc_cred *cred,
 481                 gfp_t gfp_flags)
 482 {
 483         struct nfs4_state_owner *sp;
 484
 485         sp = kzalloc(sizeof(*sp), gfp_flags);
 486         if (!sp)
 487                 return NULL;
 488         sp->so_server = server;
 489         sp->so_cred = get_rpccred(cred);
 490         spin_lock_init(&sp->so_lock);
 491         INIT_LIST_HEAD(&sp->so_states);
 492         nfs4_init_seqid_counter(&sp->so_seqid);
 493         atomic_set(&sp->so_count, 1);
 494         INIT_LIST_HEAD(&sp->so_lru);
 495         seqcount_init(&sp->so_reclaim_seqcount);
 496         mutex_init(&sp->so_delegreturn_mutex);
 497         return sp;
 498 }
 499
 500 static void
 501 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 502 {
 503         struct rb_node *rb_node = &sp->so_server_node;
 504
 505         if (!RB_EMPTY_NODE(rb_node)) {
 506                 struct nfs_server *server = sp->so_server;
 507                 struct nfs_client *clp = server->nfs_client;
 508
 509                 spin_lock(&clp->cl_lock);
 510                 if (!RB_EMPTY_NODE(rb_node)) {
 511                         rb_erase(rb_node, &server->state_owners);
 512                         RB_CLEAR_NODE(rb_node);
 513                 }
 514                 spin_unlock(&clp->cl_lock);
 515         }
 516 }
 517
 518 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 519 {
 520         nfs4_destroy_seqid_counter(&sp->so_seqid);
 521         put_rpccred(sp->so_cred);
 522         kfree(sp);
 523 }
 524
 525 static void nfs4_gc_state_owners(struct nfs_server *server)
 526 {
 527         struct nfs_client *clp = server->nfs_client;
 528         struct nfs4_state_owner *sp, *tmp;
 529         unsigned long time_min, time_max;
 530         LIST_HEAD(doomed);
 531
 532         spin_lock(&clp->cl_lock);
 533         time_max = jiffies;
 534         time_min = (long)time_max - (long)clp->cl_lease_time;
 535         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 536                 /* NB: LRU is sorted so that oldest is at the head */
 537                 if (time_in_range(sp->so_expires, time_min, time_max))
 538                         break;
 539                 list_move(&sp->so_lru, &doomed);
 540                 nfs4_remove_state_owner_locked(sp);
 541         }
 542         spin_unlock(&clp->cl_lock);
 543
 544         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 545                 list_del(&sp->so_lru);
 546                 nfs4_free_state_owner(sp);
 547         }
 548 }
 549
 550 /**
 551  * nfs4_get_state_owner - Look up a state owner given a credential
 552  * @server: nfs_server to search
 553  * @cred: RPC credential to match
 554  *
 555  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 556  */
 557 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 558                                               struct rpc_cred *cred,
 559                                               gfp_t gfp_flags)
 560 {
 561         struct nfs_client *clp = server->nfs_client;
 562         struct nfs4_state_owner *sp, *new;
 563
 564         spin_lock(&clp->cl_lock);
 565         sp = nfs4_find_state_owner_locked(server, cred);
 566         spin_unlock(&clp->cl_lock);
 567         if (sp != NULL)
 568                 goto out;
 569         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 570         if (new == NULL)
 571                 goto out;
 572         do {
 573                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 574                         break;
 575                 spin_lock(&clp->cl_lock);
 576                 sp = nfs4_insert_state_owner_locked(new);
 577                 spin_unlock(&clp->cl_lock);
 578         } while (sp == ERR_PTR(-EAGAIN));
 579         if (sp != new)
 580                 nfs4_free_state_owner(new);
 581 out:
 582         nfs4_gc_state_owners(server);
 583         return sp;
 584 }
 585
 586 /**
 587  * nfs4_put_state_owner - Release a nfs4_state_owner
 588  * @sp: state owner data to release
 589  *
 590  * Note that we keep released state owners on an LRU
 591  * list.
 592  * This caches valid state owners so that they can be
 593  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 594  * It also pins the uniquifier of dropped state owners for
 595  * a while, to ensure that those state owner names are
 596  * never reused.
 597  */
 598 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 599 {
 600         struct nfs_server *server = sp->so_server;
 601         struct nfs_client *clp = server->nfs_client;
 602
 603         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 604                 return;
 605
 606         sp->so_expires = jiffies;
 607         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 608         spin_unlock(&clp->cl_lock);
 609 }
 610
 611 /**
 612  * nfs4_purge_state_owners - Release all cached state owners
 613  * @server: nfs_server with cached state owners to release
 614  * @head: resulting list of state owners
 615  *
 616  * Called at umount time.  Remaining state owners will be on
 617  * the LRU with ref count of zero.
 618  * Note that the state owners are not freed, but are added
 619  * to the list @head, which can later be used as an argument
 620  * to nfs4_free_state_owners.
 621  */
 622 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 623 {
 624         struct nfs_client *clp = server->nfs_client;
 625         struct nfs4_state_owner *sp, *tmp;
 626
 627         spin_lock(&clp->cl_lock);
 628         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 629                 list_move(&sp->so_lru, head);
 630                 nfs4_remove_state_owner_locked(sp);
 631         }
 632         spin_unlock(&clp->cl_lock);
 633 }
 634
 635 /**
 636  * nfs4_purge_state_owners - Release all cached state owners
 637  * @head: resulting list of state owners
 638  *
 639  * Frees a list of state owners that was generated by
 640  * nfs4_purge_state_owners
 641  */
 642 void nfs4_free_state_owners(struct list_head *head)
 643 {
 644         struct nfs4_state_owner *sp, *tmp;
 645
 646         list_for_each_entry_safe(sp, tmp, head, so_lru) {
 647                 list_del(&sp->so_lru);
 648                 nfs4_free_state_owner(sp);
 649         }
 650 }
 651
 652 static struct nfs4_state *
 653 nfs4_alloc_open_state(void)
 654 {
 655         struct nfs4_state *state;
 656
 657         state = kzalloc(sizeof(*state), GFP_NOFS);
 658         if (!state)
 659                 return NULL;
 660         atomic_set(&state->count, 1);
 661         INIT_LIST_HEAD(&state->lock_states);
 662         spin_lock_init(&state->state_lock);
 663         seqlock_init(&state->seqlock);
 664         return state;
 665 }
 666
 667 void
 668 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 669 {
 670         if (state->state == fmode)
 671                 return;
 672         /* NB! List reordering - see the reclaim code for why.  */
 673         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 674                 if (fmode & FMODE_WRITE)
 675                         list_move(&state->open_states, &state->owner->so_states);
 676                 else
 677                         list_move_tail(&state->open_states, &state->owner->so_states);
 678         }
 679         state->state = fmode;
 680 }
 681
 682 static struct nfs4_state *
 683 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 684 {
 685         struct nfs_inode *nfsi = NFS_I(inode);
 686         struct nfs4_state *state;
 687
 688         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 689                 if (state->owner != owner)
 690                         continue;
 691                 if (!nfs4_valid_open_stateid(state))
 692                         continue;
 693                 if (atomic_inc_not_zero(&state->count))
 694                         return state;
 695         }
 696         return NULL;
 697 }
 698
 699 static void
 700 nfs4_free_open_state(struct nfs4_state *state)
 701 {
 702         kfree(state);
 703 }
 704
 705 struct nfs4_state *
 706 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 707 {
 708         struct nfs4_state *state, *new;
 709         struct nfs_inode *nfsi = NFS_I(inode);
 710
 711         spin_lock(&inode->i_lock);
 712         state = __nfs4_find_state_byowner(inode, owner);
 713         spin_unlock(&inode->i_lock);
 714         if (state)
 715                 goto out;
 716         new = nfs4_alloc_open_state();
 717         spin_lock(&owner->so_lock);
 718         spin_lock(&inode->i_lock);
 719         state = __nfs4_find_state_byowner(inode, owner);
 720         if (state == NULL && new != NULL) {
 721                 state = new;
 722                 state->owner = owner;
 723                 atomic_inc(&owner->so_count);
 724                 list_add(&state->inode_states, &nfsi->open_states);
 725                 ihold(inode);
 726                 state->inode = inode;
 727                 spin_unlock(&inode->i_lock);
 728                 /* Note: The reclaim code dictates that we add stateless
 729                  * and read-only stateids to the end of the list */
 730                 list_add_tail(&state->open_states, &owner->so_states);
 731                 spin_unlock(&owner->so_lock);
 732         } else {
 733                 spin_unlock(&inode->i_lock);
 734                 spin_unlock(&owner->so_lock);
 735                 if (new)
 736                         nfs4_free_open_state(new);
 737         }
 738 out:
 739         return state;
 740 }
 741
 742 void nfs4_put_open_state(struct nfs4_state *state)
 743 {
 744         struct inode *inode = state->inode;
 745         struct nfs4_state_owner *owner = state->owner;
 746
 747         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 748                 return;
 749         spin_lock(&inode->i_lock);
 750         list_del(&state->inode_states);
 751         list_del(&state->open_states);
 752         spin_unlock(&inode->i_lock);
 753         spin_unlock(&owner->so_lock);
 754         iput(inode);
 755         nfs4_free_open_state(state);
 756         nfs4_put_state_owner(owner);
 757 }
 758
 759 /*
 760  * Close the current file.
 761  */
 762 static void __nfs4_close(struct nfs4_state *state,
 763                 fmode_t fmode, gfp_t gfp_mask, int wait)
 764 {
 765         struct nfs4_state_owner *owner = state->owner;
 766         int call_close = 0;
 767         fmode_t newstate;
 768
 769         atomic_inc(&owner->so_count);
 770         /* Protect against nfs4_find_state() */
 771         spin_lock(&owner->so_lock);
 772         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 773                 case FMODE_READ:
 774                         state->n_rdonly--;
 775                         break;
 776                 case FMODE_WRITE:
 777                         state->n_wronly--;
 778                         break;
 779                 case FMODE_READ|FMODE_WRITE:
 780                         state->n_rdwr--;
 781         }
 782         newstate = FMODE_READ|FMODE_WRITE;
 783         if (state->n_rdwr == 0) {
 784                 if (state->n_rdonly == 0) {
 785                         newstate &= ~FMODE_READ;
 786                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 787                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 788                 }
 789                 if (state->n_wronly == 0) {
 790                         newstate &= ~FMODE_WRITE;
 791                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 792                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 793                 }
 794                 if (newstate == 0)
 795                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 796         }
 797         nfs4_state_set_mode_locked(state, newstate);
 798         spin_unlock(&owner->so_lock);
 799
 800         if (!call_close) {
 801                 nfs4_put_open_state(state);
 802                 nfs4_put_state_owner(owner);
 803         } else
 804                 nfs4_do_close(state, gfp_mask, wait);
 805 }
 806
 807 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 808 {
 809         __nfs4_close(state, fmode, GFP_NOFS, 0);
 810 }
 811
 812 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 813 {
 814         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 815 }
 816
 817 /*
 818  * Search the state->lock_states for an existing lock_owner
 819  * that is compatible with current->files
 820  */
 821 static struct nfs4_lock_state *
 822 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 823 {
 824         struct nfs4_lock_state *pos;
 825         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 826                 if (pos->ls_owner != fl_owner)
 827                         continue;
 828                 atomic_inc(&pos->ls_count);
 829                 return pos;
 830         }
 831         return NULL;
 832 }
 833
 834 /*
 835  * Return a compatible lock_state. If no initialized lock_state structure
 836  * exists, return an uninitialized one.
 837  *
 838  */
 839 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 840 {
 841         struct nfs4_lock_state *lsp;
 842         struct nfs_server *server = state->owner->so_server;
 843
 844         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 845         if (lsp == NULL)
 846                 return NULL;
 847         nfs4_init_seqid_counter(&lsp->ls_seqid);
 848         atomic_set(&lsp->ls_count, 1);
 849         lsp->ls_state = state;
 850         lsp->ls_owner = fl_owner;
 851         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 852         if (lsp->ls_seqid.owner_id < 0)
 853                 goto out_free;
 854         INIT_LIST_HEAD(&lsp->ls_locks);
 855         return lsp;
 856 out_free:
 857         kfree(lsp);
 858         return NULL;
 859 }
 860
 861 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 862 {
 863         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 864         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 865         kfree(lsp);
 866 }
 867
 868 /*
 869  * Return a compatible lock_state. If no initialized lock_state structure
 870  * exists, return an uninitialized one.
 871  *
 872  */
 873 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 874 {
 875         struct nfs4_lock_state *lsp, *new = NULL;
 876
 877         for(;;) {
 878                 spin_lock(&state->state_lock);
 879                 lsp = __nfs4_find_lock_state(state, owner);
 880                 if (lsp != NULL)
 881                         break;
 882                 if (new != NULL) {
 883                         list_add(&new->ls_locks, &state->lock_states);
 884                         set_bit(LK_STATE_IN_USE, &state->flags);
 885                         lsp = new;
 886                         new = NULL;
 887                         break;
 888                 }
 889                 spin_unlock(&state->state_lock);
 890                 new = nfs4_alloc_lock_state(state, owner);
 891                 if (new == NULL)
 892                         return NULL;
 893         }
 894         spin_unlock(&state->state_lock);
 895         if (new != NULL)
 896                 nfs4_free_lock_state(state->owner->so_server, new);
 897         return lsp;
 898 }
 899
 900 /*
 901  * Release reference to lock_state, and free it if we see that
 902  * it is no longer in use
 903  */
 904 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 905 {
 906         struct nfs_server *server;
 907         struct nfs4_state *state;
 908
 909         if (lsp == NULL)
 910                 return;
 911         state = lsp->ls_state;
 912         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 913                 return;
 914         list_del(&lsp->ls_locks);
 915         if (list_empty(&state->lock_states))
 916                 clear_bit(LK_STATE_IN_USE, &state->flags);
 917         spin_unlock(&state->state_lock);
 918         server = state->owner->so_server;
 919         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 920                 struct nfs_client *clp = server->nfs_client;
 921
 922                 clp->cl_mvops->free_lock_state(server, lsp);
 923         } else
 924                 nfs4_free_lock_state(server, lsp);
 925 }
 926
 927 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 928 {
 929         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 930
 931         dst->fl_u.nfs4_fl.owner = lsp;
 932         atomic_inc(&lsp->ls_count);
 933 }
 934
 935 static void nfs4_fl_release_lock(struct file_lock *fl)
 936 {
 937         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 938 }
 939
 940 static const struct file_lock_operations nfs4_fl_lock_ops = {
 941         .fl_copy_lock = nfs4_fl_copy_lock,
 942         .fl_release_private = nfs4_fl_release_lock,
 943 };
 944
 945 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 946 {
 947         struct nfs4_lock_state *lsp;
 948
 949         if (fl->fl_ops != NULL)
 950                 return 0;
 951         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 952         if (lsp == NULL)
 953                 return -ENOMEM;
 954         fl->fl_u.nfs4_fl.owner = lsp;
 955         fl->fl_ops = &nfs4_fl_lock_ops;
 956         return 0;
 957 }
 958
 959 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 960                 struct nfs4_state *state,
 961                 const struct nfs_lockowner *lockowner)
 962 {
 963         struct nfs4_lock_state *lsp;
 964         fl_owner_t fl_owner;
 965         int ret = -ENOENT;
 966
 967
 968         if (lockowner == NULL)
 969                 goto out;
 970
 971         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 972                 goto out;
 973
 974         fl_owner = lockowner->l_owner;
 975         spin_lock(&state->state_lock);
 976         lsp = __nfs4_find_lock_state(state, fl_owner);
 977         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 978                 ret = -EIO;
 979         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 980                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 981                 ret = 0;
 982         }
 983         spin_unlock(&state->state_lock);
 984         nfs4_put_lock_state(lsp);
 985 out:
 986         return ret;
 987 }
 988
 989 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 990 {
 991         const nfs4_stateid *src;
 992         int seq;
 993
 994         do {
 995                 src = &zero_stateid;
 996                 seq = read_seqbegin(&state->seqlock);
 997                 if (test_bit(NFS_OPEN_STATE, &state->flags))
 998                         src = &state->open_stateid;
 999                 nfs4_stateid_copy(dst, src);
1000         } while (read_seqretry(&state->seqlock, seq));
1001 }
1002
1003 /*
1004  * Byte-range lock aware utility to initialize the stateid of read/write
1005  * requests.
1006  */
1007 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1008                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1009 {
1010         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1011         if (ret == -EIO)
1012                 /* A lost lock - don't even consider delegations */
1013                 goto out;
1014         /* returns true if delegation stateid found and copied */
1015         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
1016                 ret = 0;
1017                 goto out;
1018         }
1019         if (ret != -ENOENT)
1020                 /* nfs4_copy_delegation_stateid() didn't over-write
1021                  * dst, so it still has the lock stateid which we now
1022                  * choose to use.
1023                  */
1024                 goto out;
1025         nfs4_copy_open_stateid(dst, state);
1026         ret = 0;
1027 out:
1028         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1029                 dst->seqid = 0;
1030         return ret;
1031 }
1032
1033 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1034 {
1035         struct nfs_seqid *new;
1036
1037         new = kmalloc(sizeof(*new), gfp_mask);
1038         if (new == NULL)
1039                 return ERR_PTR(-ENOMEM);
1040         new->sequence = counter;
1041         INIT_LIST_HEAD(&new->list);
1042         new->task = NULL;
1043         return new;
1044 }
1045
1046 void nfs_release_seqid(struct nfs_seqid *seqid)
1047 {
1048         struct nfs_seqid_counter *sequence;
1049
1050         if (seqid == NULL || list_empty(&seqid->list))
1051                 return;
1052         sequence = seqid->sequence;
1053         spin_lock(&sequence->lock);
1054         list_del_init(&seqid->list);
1055         if (!list_empty(&sequence->list)) {
1056                 struct nfs_seqid *next;
1057
1058                 next = list_first_entry(&sequence->list,
1059                                 struct nfs_seqid, list);
1060                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1061         }
1062         spin_unlock(&sequence->lock);
1063 }
1064
1065 void nfs_free_seqid(struct nfs_seqid *seqid)
1066 {
1067         nfs_release_seqid(seqid);
1068         kfree(seqid);
1069 }
1070
1071 /*
1072  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1073  * failed with a seqid incrementing error -
1074  * see comments nfs4.h:seqid_mutating_error()
1075  */
1076 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1077 {
1078         switch (status) {
1079                 case 0:
1080                         break;
1081                 case -NFS4ERR_BAD_SEQID:
1082                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1083                                 return;
1084                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1085                                         " sequence-id error on an"
1086                                         " unconfirmed sequence %p!\n",
1087                                         seqid->sequence);
1088                 case -NFS4ERR_STALE_CLIENTID:
1089                 case -NFS4ERR_STALE_STATEID:
1090                 case -NFS4ERR_BAD_STATEID:
1091                 case -NFS4ERR_BADXDR:
1092                 case -NFS4ERR_RESOURCE:
1093                 case -NFS4ERR_NOFILEHANDLE:
1094                 case -NFS4ERR_MOVED:
1095                         /* Non-seqid mutating errors */
1096                         return;
1097         };
1098         /*
1099          * Note: no locking needed as we are guaranteed to be first
1100          * on the sequence list
1101          */
1102         seqid->sequence->counter++;
1103 }
1104
1105 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1106 {
1107         struct nfs4_state_owner *sp;
1108
1109         if (seqid == NULL)
1110                 return;
1111
1112         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1113         if (status == -NFS4ERR_BAD_SEQID)
1114                 nfs4_drop_state_owner(sp);
1115         if (!nfs4_has_session(sp->so_server->nfs_client))
1116                 nfs_increment_seqid(status, seqid);
1117 }
1118
1119 /*
1120  * Increment the seqid if the LOCK/LOCKU succeeded, or
1121  * failed with a seqid incrementing error -
1122  * see comments nfs4.h:seqid_mutating_error()
1123  */
1124 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1125 {
1126         if (seqid != NULL)
1127                 nfs_increment_seqid(status, seqid);
1128 }
1129
1130 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1131 {
1132         struct nfs_seqid_counter *sequence;
1133         int status = 0;
1134
1135         if (seqid == NULL)
1136                 goto out;
1137         sequence = seqid->sequence;
1138         spin_lock(&sequence->lock);
1139         seqid->task = task;
1140         if (list_empty(&seqid->list))
1141                 list_add_tail(&seqid->list, &sequence->list);
1142         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1143                 goto unlock;
1144         rpc_sleep_on(&sequence->wait, task, NULL);
1145         status = -EAGAIN;
1146 unlock:
1147         spin_unlock(&sequence->lock);
1148 out:
1149         return status;
1150 }
1151
1152 static int nfs4_run_state_manager(void *);
1153
1154 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1155 {
1156         smp_mb__before_atomic();
1157         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1158         smp_mb__after_atomic();
1159         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1160         rpc_wake_up(&clp->cl_rpcwaitq);
1161 }
1162
1163 /*
1164  * Schedule the nfs_client asynchronous state management routine
1165  */
1166 void nfs4_schedule_state_manager(struct nfs_client *clp)
1167 {
1168         struct task_struct *task;
1169         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1170
1171         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1172                 return;
1173         __module_get(THIS_MODULE);
1174         atomic_inc(&clp->cl_count);
1175
1176         /* The rcu_read_lock() is not strictly necessary, as the state
1177          * manager is the only thread that ever changes the rpc_xprt
1178          * after it's initialized.  At this point, we're single threaded. */
1179         rcu_read_lock();
1180         snprintf(buf, sizeof(buf), "%s-manager",
1181                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1182         rcu_read_unlock();
1183         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1184         if (IS_ERR(task)) {
1185                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1186                         __func__, PTR_ERR(task));
1187                 nfs4_clear_state_manager_bit(clp);
1188                 nfs_put_client(clp);
1189                 module_put(THIS_MODULE);
1190         }
1191 }
1192
1193 /*
1194  * Schedule a lease recovery attempt
1195  */
1196 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1197 {
1198         if (!clp)
1199                 return;
1200         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1201                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1202         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1203                         clp->cl_hostname);
1204         nfs4_schedule_state_manager(clp);
1205 }
1206 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1207
1208 /**
1209  * nfs4_schedule_migration_recovery - trigger migration recovery
1210  *
1211  * @server: FSID that is migrating
1212  *
1213  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1214  * value is returned.
1215  */
1216 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1217 {
1218         struct nfs_client *clp = server->nfs_client;
1219
1220         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1221                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1222                                 clp->cl_hostname);
1223                 return -NFS4ERR_IO;
1224         }
1225
1226         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1227                 return -NFS4ERR_IO;
1228
1229         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1230                         __func__,
1231                         (unsigned long long)server->fsid.major,
1232                         (unsigned long long)server->fsid.minor,
1233                         clp->cl_hostname);
1234
1235         set_bit(NFS_MIG_IN_TRANSITION,
1236                         &((struct nfs_server *)server)->mig_status);
1237         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1238
1239         nfs4_schedule_state_manager(clp);
1240         return 0;
1241 }
1242 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1243
1244 /**
1245  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1246  *
1247  * @clp: server to check for moved leases
1248  *
1249  */
1250 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1251 {
1252         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1253                 __func__, clp->cl_clientid, clp->cl_hostname);
1254
1255         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1256         nfs4_schedule_state_manager(clp);
1257 }
1258 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1259
1260 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1261 {
1262         int res;
1263
1264         might_sleep();
1265
1266         atomic_inc(&clp->cl_count);
1267         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1268                                  nfs_wait_bit_killable, TASK_KILLABLE);
1269         if (res)
1270                 goto out;
1271         if (clp->cl_cons_state < 0)
1272                 res = clp->cl_cons_state;
1273 out:
1274         nfs_put_client(clp);
1275         return res;
1276 }
1277
1278 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1279 {
1280         unsigned int loop;
1281         int ret;
1282
1283         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1284                 ret = nfs4_wait_clnt_recover(clp);
1285                 if (ret != 0)
1286                         break;
1287                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1288                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1289                         break;
1290                 nfs4_schedule_state_manager(clp);
1291                 ret = -EIO;
1292         }
1293         return ret;
1294 }
1295
1296 /*
1297  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1298  * @clp: client to process
1299  *
1300  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1301  * resend of the SETCLIENTID and hence re-establish the
1302  * callback channel. Then return all existing delegations.
1303  */
1304 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1305 {
1306         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1307         nfs_expire_all_delegations(clp);
1308         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1309                         clp->cl_hostname);
1310 }
1311
1312 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1313 {
1314         nfs40_handle_cb_pathdown(clp);
1315         nfs4_schedule_state_manager(clp);
1316 }
1317
1318 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1319 {
1320
1321         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1322         /* Don't recover state that expired before the reboot */
1323         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1324                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1325                 return 0;
1326         }
1327         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1328         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1329         return 1;
1330 }
1331
1332 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1333 {
1334         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1335         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1336         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1337         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1338         return 1;
1339 }
1340
1341 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1342 {
1343         struct nfs_client *clp = server->nfs_client;
1344
1345         if (!nfs4_valid_open_stateid(state))
1346                 return -EBADF;
1347         nfs4_state_mark_reclaim_nograce(clp, state);
1348         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1349                         clp->cl_hostname);
1350         nfs4_schedule_state_manager(clp);
1351         return 0;
1352 }
1353 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1354
1355 void nfs_inode_find_state_and_recover(struct inode *inode,
1356                 const nfs4_stateid *stateid)
1357 {
1358         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1359         struct nfs_inode *nfsi = NFS_I(inode);
1360         struct nfs_open_context *ctx;
1361         struct nfs4_state *state;
1362         bool found = false;
1363
1364         spin_lock(&inode->i_lock);
1365         list_for_each_entry(ctx, &nfsi->open_files, list) {
1366                 state = ctx->state;
1367                 if (state == NULL)
1368                         continue;
1369                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1370                         continue;
1371                 if (!nfs4_stateid_match(&state->stateid, stateid))
1372                         continue;
1373                 nfs4_state_mark_reclaim_nograce(clp, state);
1374                 found = true;
1375         }
1376         spin_unlock(&inode->i_lock);
1377         if (found)
1378                 nfs4_schedule_state_manager(clp);
1379 }
1380
1381 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1382 {
1383         struct inode *inode = state->inode;
1384         struct nfs_inode *nfsi = NFS_I(inode);
1385         struct nfs_open_context *ctx;
1386
1387         spin_lock(&inode->i_lock);
1388         list_for_each_entry(ctx, &nfsi->open_files, list) {
1389                 if (ctx->state != state)
1390                         continue;
1391                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1392         }
1393         spin_unlock(&inode->i_lock);
1394 }
1395
1396 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1397 {
1398         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1399         nfs4_state_mark_open_context_bad(state);
1400 }
1401
1402
1403 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1404 {
1405         struct inode *inode = state->inode;
1406         struct nfs_inode *nfsi = NFS_I(inode);
1407         struct file_lock *fl;
1408         struct nfs4_lock_state *lsp;
1409         int status = 0;
1410         struct file_lock_context *flctx = inode->i_flctx;
1411         struct list_head *list;
1412
1413         if (flctx == NULL)
1414                 return 0;
1415
1416         list = &flctx->flc_posix;
1417
1418         /* Guard against delegation returns and new lock/unlock calls */
1419         down_write(&nfsi->rwsem);
1420         spin_lock(&flctx->flc_lock);
1421 restart:
1422         list_for_each_entry(fl, list, fl_list) {
1423                 if (nfs_file_open_context(fl->fl_file)->state != state)
1424                         continue;
1425                 spin_unlock(&flctx->flc_lock);
1426                 status = ops->recover_lock(state, fl);
1427                 switch (status) {
1428                 case 0:
1429                         break;
1430                 case -ESTALE:
1431                 case -NFS4ERR_ADMIN_REVOKED:
1432                 case -NFS4ERR_STALE_STATEID:
1433                 case -NFS4ERR_BAD_STATEID:
1434                 case -NFS4ERR_EXPIRED:
1435                 case -NFS4ERR_NO_GRACE:
1436                 case -NFS4ERR_STALE_CLIENTID:
1437                 case -NFS4ERR_BADSESSION:
1438                 case -NFS4ERR_BADSLOT:
1439                 case -NFS4ERR_BAD_HIGH_SLOT:
1440                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1441                         goto out;
1442                 default:
1443                         pr_err("NFS: %s: unhandled error %d\n",
1444                                         __func__, status);
1445                 case -ENOMEM:
1446                 case -NFS4ERR_DENIED:
1447                 case -NFS4ERR_RECLAIM_BAD:
1448                 case -NFS4ERR_RECLAIM_CONFLICT:
1449                         lsp = fl->fl_u.nfs4_fl.owner;
1450                         if (lsp)
1451                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1452                         status = 0;
1453                 }
1454                 spin_lock(&flctx->flc_lock);
1455         }
1456         if (list == &flctx->flc_posix) {
1457                 list = &flctx->flc_flock;
1458                 goto restart;
1459         }
1460         spin_unlock(&flctx->flc_lock);
1461 out:
1462         up_write(&nfsi->rwsem);
1463         return status;
1464 }
1465
1466 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1467 {
1468         struct nfs4_state *state;
1469         struct nfs4_lock_state *lock;
1470         int status = 0;
1471
1472         /* Note: we rely on the sp->so_states list being ordered
1473          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1474          * states first.
1475          * This is needed to ensure that the server won't give us any
1476          * read delegations that we have to return if, say, we are
1477          * recovering after a network partition or a reboot from a
1478          * server that doesn't support a grace period.
1479          */
1480         spin_lock(&sp->so_lock);
1481         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1482 restart:
1483         list_for_each_entry(state, &sp->so_states, open_states) {
1484                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1485                         continue;
1486                 if (!nfs4_valid_open_stateid(state))
1487                         continue;
1488                 if (state->state == 0)
1489                         continue;
1490                 atomic_inc(&state->count);
1491                 spin_unlock(&sp->so_lock);
1492                 status = ops->recover_open(sp, state);
1493                 if (status >= 0) {
1494                         status = nfs4_reclaim_locks(state, ops);
1495                         if (status >= 0) {
1496                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1497                                         spin_lock(&state->state_lock);
1498                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1499                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1500                                                         pr_warn_ratelimited("NFS: "
1501                                                                             "%s: Lock reclaim "
1502                                                                             "failed!\n", __func__);
1503                                         }
1504                                         spin_unlock(&state->state_lock);
1505                                 }
1506                                 nfs4_put_open_state(state);
1507                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1508                                         &state->flags);
1509                                 spin_lock(&sp->so_lock);
1510                                 goto restart;
1511                         }
1512                 }
1513                 switch (status) {
1514                         default:
1515                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1516                                         __func__, status);
1517                         case -ENOENT:
1518                         case -ENOMEM:
1519                         case -EACCES:
1520                         case -EROFS:
1521                         case -EIO:
1522                         case -ESTALE:
1523                                 /* Open state on this file cannot be recovered */
1524                                 nfs4_state_mark_recovery_failed(state, status);
1525                                 break;
1526                         case -EAGAIN:
1527                                 ssleep(1);
1528                         case -NFS4ERR_ADMIN_REVOKED:
1529                         case -NFS4ERR_STALE_STATEID:
1530                         case -NFS4ERR_BAD_STATEID:
1531                         case -NFS4ERR_RECLAIM_BAD:
1532                         case -NFS4ERR_RECLAIM_CONFLICT:
1533                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1534                                 break;
1535                         case -NFS4ERR_EXPIRED:
1536                         case -NFS4ERR_NO_GRACE:
1537                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1538                         case -NFS4ERR_STALE_CLIENTID:
1539                         case -NFS4ERR_BADSESSION:
1540                         case -NFS4ERR_BADSLOT:
1541                         case -NFS4ERR_BAD_HIGH_SLOT:
1542                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1543                                 goto out_err;
1544                 }
1545                 nfs4_put_open_state(state);
1546                 spin_lock(&sp->so_lock);
1547                 goto restart;
1548         }
1549         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1550         spin_unlock(&sp->so_lock);
1551         return 0;
1552 out_err:
1553         nfs4_put_open_state(state);
1554         spin_lock(&sp->so_lock);
1555         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1556         spin_unlock(&sp->so_lock);
1557         return status;
1558 }
1559
1560 static void nfs4_clear_open_state(struct nfs4_state *state)
1561 {
1562         struct nfs4_lock_state *lock;
1563
1564         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1565         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1566         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1567         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1568         spin_lock(&state->state_lock);
1569         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1570                 lock->ls_seqid.flags = 0;
1571                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1572         }
1573         spin_unlock(&state->state_lock);
1574 }
1575
1576 static void nfs4_reset_seqids(struct nfs_server *server,
1577         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1578 {
1579         struct nfs_client *clp = server->nfs_client;
1580         struct nfs4_state_owner *sp;
1581         struct rb_node *pos;
1582         struct nfs4_state *state;
1583
1584         spin_lock(&clp->cl_lock);
1585         for (pos = rb_first(&server->state_owners);
1586              pos != NULL;
1587              pos = rb_next(pos)) {
1588                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1589                 sp->so_seqid.flags = 0;
1590                 spin_lock(&sp->so_lock);
1591                 list_for_each_entry(state, &sp->so_states, open_states) {
1592                         if (mark_reclaim(clp, state))
1593                                 nfs4_clear_open_state(state);
1594                 }
1595                 spin_unlock(&sp->so_lock);
1596         }
1597         spin_unlock(&clp->cl_lock);
1598 }
1599
1600 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1601         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1602 {
1603         struct nfs_server *server;
1604
1605         rcu_read_lock();
1606         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1607                 nfs4_reset_seqids(server, mark_reclaim);
1608         rcu_read_unlock();
1609 }
1610
1611 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1612 {
1613         /* Mark all delegations for reclaim */
1614         nfs_delegation_mark_reclaim(clp);
1615         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1616 }
1617
1618 static int nfs4_reclaim_complete(struct nfs_client *clp,
1619                                  const struct nfs4_state_recovery_ops *ops,
1620                                  struct rpc_cred *cred)
1621 {
1622         /* Notify the server we're done reclaiming our state */
1623         if (ops->reclaim_complete)
1624                 return ops->reclaim_complete(clp, cred);
1625         return 0;
1626 }
1627
1628 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1629 {
1630         struct nfs_client *clp = server->nfs_client;
1631         struct nfs4_state_owner *sp;
1632         struct rb_node *pos;
1633         struct nfs4_state *state;
1634
1635         spin_lock(&clp->cl_lock);
1636         for (pos = rb_first(&server->state_owners);
1637              pos != NULL;
1638              pos = rb_next(pos)) {
1639                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1640                 spin_lock(&sp->so_lock);
1641                 list_for_each_entry(state, &sp->so_states, open_states) {
1642                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1643                                                 &state->flags))
1644                                 continue;
1645                         nfs4_state_mark_reclaim_nograce(clp, state);
1646                 }
1647                 spin_unlock(&sp->so_lock);
1648         }
1649         spin_unlock(&clp->cl_lock);
1650 }
1651
1652 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1653 {
1654         struct nfs_server *server;
1655
1656         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1657                 return 0;
1658
1659         rcu_read_lock();
1660         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1661                 nfs4_clear_reclaim_server(server);
1662         rcu_read_unlock();
1663
1664         nfs_delegation_reap_unclaimed(clp);
1665         return 1;
1666 }
1667
1668 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1669 {
1670         const struct nfs4_state_recovery_ops *ops;
1671         struct rpc_cred *cred;
1672         int err;
1673
1674         if (!nfs4_state_clear_reclaim_reboot(clp))
1675                 return;
1676         ops = clp->cl_mvops->reboot_recovery_ops;
1677         cred = nfs4_get_clid_cred(clp);
1678         err = nfs4_reclaim_complete(clp, ops, cred);
1679         put_rpccred(cred);
1680         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1681                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1682 }
1683
1684 static void nfs_delegation_clear_all(struct nfs_client *clp)
1685 {
1686         nfs_delegation_mark_reclaim(clp);
1687         nfs_delegation_reap_unclaimed(clp);
1688 }
1689
1690 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1691 {
1692         nfs_delegation_clear_all(clp);
1693         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1694 }
1695
1696 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1697 {
1698         switch (error) {
1699                 case 0:
1700                         break;
1701                 case -NFS4ERR_CB_PATH_DOWN:
1702                         nfs40_handle_cb_pathdown(clp);
1703                         break;
1704                 case -NFS4ERR_NO_GRACE:
1705                         nfs4_state_end_reclaim_reboot(clp);
1706                         break;
1707                 case -NFS4ERR_STALE_CLIENTID:
1708                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1709                         nfs4_state_start_reclaim_reboot(clp);
1710                         break;
1711                 case -NFS4ERR_EXPIRED:
1712                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1713                         nfs4_state_start_reclaim_nograce(clp);
1714                         break;
1715                 case -NFS4ERR_BADSESSION:
1716                 case -NFS4ERR_BADSLOT:
1717                 case -NFS4ERR_BAD_HIGH_SLOT:
1718                 case -NFS4ERR_DEADSESSION:
1719                 case -NFS4ERR_SEQ_FALSE_RETRY:
1720                 case -NFS4ERR_SEQ_MISORDERED:
1721                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1722                         /* Zero session reset errors */
1723                         break;
1724                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1725                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1726                         break;
1727                 default:
1728                         dprintk("%s: failed to handle error %d for server %s\n",
1729                                         __func__, error, clp->cl_hostname);
1730                         return error;
1731         }
1732         dprintk("%s: handled error %d for server %s\n", __func__, error,
1733                         clp->cl_hostname);
1734         return 0;
1735 }
1736
1737 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1738 {
1739         struct nfs4_state_owner *sp;
1740         struct nfs_server *server;
1741         struct rb_node *pos;
1742         LIST_HEAD(freeme);
1743         int status = 0;
1744
1745 restart:
1746         rcu_read_lock();
1747         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1748                 nfs4_purge_state_owners(server, &freeme);
1749                 spin_lock(&clp->cl_lock);
1750                 for (pos = rb_first(&server->state_owners);
1751                      pos != NULL;
1752                      pos = rb_next(pos)) {
1753                         sp = rb_entry(pos,
1754                                 struct nfs4_state_owner, so_server_node);
1755                         if (!test_and_clear_bit(ops->owner_flag_bit,
1756                                                         &sp->so_flags))
1757                                 continue;
1758                         if (!atomic_inc_not_zero(&sp->so_count))
1759                                 continue;
1760                         spin_unlock(&clp->cl_lock);
1761                         rcu_read_unlock();
1762
1763                         status = nfs4_reclaim_open_state(sp, ops);
1764                         if (status < 0) {
1765                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1766                                 nfs4_put_state_owner(sp);
1767                                 status = nfs4_recovery_handle_error(clp, status);
1768                                 return (status != 0) ? status : -EAGAIN;
1769                         }
1770
1771                         nfs4_put_state_owner(sp);
1772                         goto restart;
1773                 }
1774                 spin_unlock(&clp->cl_lock);
1775         }
1776         rcu_read_unlock();
1777         nfs4_free_state_owners(&freeme);
1778         return 0;
1779 }
1780
1781 static int nfs4_check_lease(struct nfs_client *clp)
1782 {
1783         struct rpc_cred *cred;
1784         const struct nfs4_state_maintenance_ops *ops =
1785                 clp->cl_mvops->state_renewal_ops;
1786         int status;
1787
1788         /* Is the client already known to have an expired lease? */
1789         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1790                 return 0;
1791         spin_lock(&clp->cl_lock);
1792         cred = ops->get_state_renewal_cred_locked(clp);
1793         spin_unlock(&clp->cl_lock);
1794         if (cred == NULL) {
1795                 cred = nfs4_get_clid_cred(clp);
1796                 status = -ENOKEY;
1797                 if (cred == NULL)
1798                         goto out;
1799         }
1800         status = ops->renew_lease(clp, cred);
1801         put_rpccred(cred);
1802         if (status == -ETIMEDOUT) {
1803                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1804                 return 0;
1805         }
1806 out:
1807         return nfs4_recovery_handle_error(clp, status);
1808 }
1809
1810 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1811  * and for recoverable errors on EXCHANGE_ID for v4.1
1812  */
1813 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1814 {
1815         switch (status) {
1816         case -NFS4ERR_SEQ_MISORDERED:
1817                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1818                         return -ESERVERFAULT;
1819                 /* Lease confirmation error: retry after purging the lease */
1820                 ssleep(1);
1821                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1822                 break;
1823         case -NFS4ERR_STALE_CLIENTID:
1824                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1825                 nfs4_state_start_reclaim_reboot(clp);
1826                 break;
1827         case -NFS4ERR_CLID_INUSE:
1828                 pr_err("NFS: Server %s reports our clientid is in use\n",
1829                         clp->cl_hostname);
1830                 nfs_mark_client_ready(clp, -EPERM);
1831                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1832                 return -EPERM;
1833         case -EACCES:
1834         case -NFS4ERR_DELAY:
1835         case -ETIMEDOUT:
1836         case -EAGAIN:
1837                 ssleep(1);
1838                 break;
1839
1840         case -NFS4ERR_MINOR_VERS_MISMATCH:
1841                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1842                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1843                 dprintk("%s: exit with error %d for server %s\n",
1844                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1845                 return -EPROTONOSUPPORT;
1846         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1847                                  * in nfs4_exchange_id */
1848         default:
1849                 dprintk("%s: exit with error %d for server %s\n", __func__,
1850                                 status, clp->cl_hostname);
1851                 return status;
1852         }
1853         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1854         dprintk("%s: handled error %d for server %s\n", __func__, status,
1855                         clp->cl_hostname);
1856         return 0;
1857 }
1858
1859 static int nfs4_establish_lease(struct nfs_client *clp)
1860 {
1861         struct rpc_cred *cred;
1862         const struct nfs4_state_recovery_ops *ops =
1863                 clp->cl_mvops->reboot_recovery_ops;
1864         int status;
1865
1866         nfs4_begin_drain_session(clp);
1867         cred = nfs4_get_clid_cred(clp);
1868         if (cred == NULL)
1869                 return -ENOENT;
1870         status = ops->establish_clid(clp, cred);
1871         put_rpccred(cred);
1872         if (status != 0)
1873                 return status;
1874         pnfs_destroy_all_layouts(clp);
1875         return 0;
1876 }
1877
1878 /*
1879  * Returns zero or a negative errno.  NFS4ERR values are converted
1880  * to local errno values.
1881  */
1882 static int nfs4_reclaim_lease(struct nfs_client *clp)
1883 {
1884         int status;
1885
1886         status = nfs4_establish_lease(clp);
1887         if (status < 0)
1888                 return nfs4_handle_reclaim_lease_error(clp, status);
1889         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1890                 nfs4_state_start_reclaim_nograce(clp);
1891         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1892                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1893         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1894         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1895         return 0;
1896 }
1897
1898 static int nfs4_purge_lease(struct nfs_client *clp)
1899 {
1900         int status;
1901
1902         status = nfs4_establish_lease(clp);
1903         if (status < 0)
1904                 return nfs4_handle_reclaim_lease_error(clp, status);
1905         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1906         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1907         nfs4_state_start_reclaim_nograce(clp);
1908         return 0;
1909 }
1910
1911 /*
1912  * Try remote migration of one FSID from a source server to a
1913  * destination server.  The source server provides a list of
1914  * potential destinations.
1915  *
1916  * Returns zero or a negative NFS4ERR status code.
1917  */
1918 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1919 {
1920         struct nfs_client *clp = server->nfs_client;
1921         struct nfs4_fs_locations *locations = NULL;
1922         struct inode *inode;
1923         struct page *page;
1924         int status, result;
1925
1926         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1927                         (unsigned long long)server->fsid.major,
1928                         (unsigned long long)server->fsid.minor,
1929                         clp->cl_hostname);
1930
1931         result = 0;
1932         page = alloc_page(GFP_KERNEL);
1933         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1934         if (page == NULL || locations == NULL) {
1935                 dprintk("<-- %s: no memory\n", __func__);
1936                 goto out;
1937         }
1938
1939         inode = d_inode(server->super->s_root);
1940         result = nfs4_proc_get_locations(inode, locations, page, cred);
1941         if (result) {
1942                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1943                         __func__, result);
1944                 goto out;
1945         }
1946
1947         result = -NFS4ERR_NXIO;
1948         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1949                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1950                         __func__);
1951                 goto out;
1952         }
1953
1954         nfs4_begin_drain_session(clp);
1955
1956         status = nfs4_replace_transport(server, locations);
1957         if (status != 0) {
1958                 dprintk("<-- %s: failed to replace transport: %d\n",
1959                         __func__, status);
1960                 goto out;
1961         }
1962
1963         result = 0;
1964         dprintk("<-- %s: migration succeeded\n", __func__);
1965
1966 out:
1967         if (page != NULL)
1968                 __free_page(page);
1969         kfree(locations);
1970         if (result) {
1971                 pr_err("NFS: migration recovery failed (server %s)\n",
1972                                 clp->cl_hostname);
1973                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1974         }
1975         return result;
1976 }
1977
1978 /*
1979  * Returns zero or a negative NFS4ERR status code.
1980  */
1981 static int nfs4_handle_migration(struct nfs_client *clp)
1982 {
1983         const struct nfs4_state_maintenance_ops *ops =
1984                                 clp->cl_mvops->state_renewal_ops;
1985         struct nfs_server *server;
1986         struct rpc_cred *cred;
1987
1988         dprintk("%s: migration reported on \"%s\"\n", __func__,
1989                         clp->cl_hostname);
1990
1991         spin_lock(&clp->cl_lock);
1992         cred = ops->get_state_renewal_cred_locked(clp);
1993         spin_unlock(&clp->cl_lock);
1994         if (cred == NULL)
1995                 return -NFS4ERR_NOENT;
1996
1997         clp->cl_mig_gen++;
1998 restart:
1999         rcu_read_lock();
2000         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2001                 int status;
2002
2003                 if (server->mig_gen == clp->cl_mig_gen)
2004                         continue;
2005                 server->mig_gen = clp->cl_mig_gen;
2006
2007                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2008                                                 &server->mig_status))
2009                         continue;
2010
2011                 rcu_read_unlock();
2012                 status = nfs4_try_migration(server, cred);
2013                 if (status < 0) {
2014                         put_rpccred(cred);
2015                         return status;
2016                 }
2017                 goto restart;
2018         }
2019         rcu_read_unlock();
2020         put_rpccred(cred);
2021         return 0;
2022 }
2023
2024 /*
2025  * Test each nfs_server on the clp's cl_superblocks list to see
2026  * if it's moved to another server.  Stop when the server no longer
2027  * returns NFS4ERR_LEASE_MOVED.
2028  */
2029 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2030 {
2031         const struct nfs4_state_maintenance_ops *ops =
2032                                 clp->cl_mvops->state_renewal_ops;
2033         struct nfs_server *server;
2034         struct rpc_cred *cred;
2035
2036         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2037                         clp->cl_hostname);
2038
2039         spin_lock(&clp->cl_lock);
2040         cred = ops->get_state_renewal_cred_locked(clp);
2041         spin_unlock(&clp->cl_lock);
2042         if (cred == NULL)
2043                 return -NFS4ERR_NOENT;
2044
2045         clp->cl_mig_gen++;
2046 restart:
2047         rcu_read_lock();
2048         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2049                 struct inode *inode;
2050                 int status;
2051
2052                 if (server->mig_gen == clp->cl_mig_gen)
2053                         continue;
2054                 server->mig_gen = clp->cl_mig_gen;
2055
2056                 rcu_read_unlock();
2057
2058                 inode = d_inode(server->super->s_root);
2059                 status = nfs4_proc_fsid_present(inode, cred);
2060                 if (status != -NFS4ERR_MOVED)
2061                         goto restart;   /* wasn't this one */
2062                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2063                         goto restart;   /* there are more */
2064                 goto out;
2065         }
2066         rcu_read_unlock();
2067
2068 out:
2069         put_rpccred(cred);
2070         return 0;
2071 }
2072
2073 /**
2074  * nfs4_discover_server_trunking - Detect server IP address trunking
2075  *
2076  * @clp: nfs_client under test
2077  * @result: OUT: found nfs_client, or clp
2078  *
2079  * Returns zero or a negative errno.  If zero is returned,
2080  * an nfs_client pointer is planted in "result".
2081  *
2082  * Note: since we are invoked in process context, and
2083  * not from inside the state manager, we cannot use
2084  * nfs4_handle_reclaim_lease_error().
2085  */
2086 int nfs4_discover_server_trunking(struct nfs_client *clp,
2087                                   struct nfs_client **result)
2088 {
2089         const struct nfs4_state_recovery_ops *ops =
2090                                 clp->cl_mvops->reboot_recovery_ops;
2091         struct rpc_clnt *clnt;
2092         struct rpc_cred *cred;
2093         int i, status;
2094
2095         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2096
2097         clnt = clp->cl_rpcclient;
2098         i = 0;
2099
2100         mutex_lock(&nfs_clid_init_mutex);
2101 again:
2102         status  = -ENOENT;
2103         cred = nfs4_get_clid_cred(clp);
2104         if (cred == NULL)
2105                 goto out_unlock;
2106
2107         status = ops->detect_trunking(clp, result, cred);
2108         put_rpccred(cred);
2109         switch (status) {
2110         case 0:
2111                 break;
2112         case -ETIMEDOUT:
2113                 if (clnt->cl_softrtry)
2114                         break;
2115         case -NFS4ERR_DELAY:
2116         case -EAGAIN:
2117                 ssleep(1);
2118         case -NFS4ERR_STALE_CLIENTID:
2119                 dprintk("NFS: %s after status %d, retrying\n",
2120                         __func__, status);
2121                 goto again;
2122         case -EACCES:
2123                 if (i++ == 0) {
2124                         nfs4_root_machine_cred(clp);
2125                         goto again;
2126                 }
2127                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2128                         break;
2129         case -NFS4ERR_CLID_INUSE:
2130         case -NFS4ERR_WRONGSEC:
2131                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2132                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2133                         status = -EPERM;
2134                         break;
2135                 }
2136                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2137                 if (IS_ERR(clnt)) {
2138                         status = PTR_ERR(clnt);
2139                         break;
2140                 }
2141                 /* Note: this is safe because we haven't yet marked the
2142                  * client as ready, so we are the only user of
2143                  * clp->cl_rpcclient
2144                  */
2145                 clnt = xchg(&clp->cl_rpcclient, clnt);
2146                 rpc_shutdown_client(clnt);
2147                 clnt = clp->cl_rpcclient;
2148                 goto again;
2149
2150         case -NFS4ERR_MINOR_VERS_MISMATCH:
2151                 status = -EPROTONOSUPPORT;
2152                 break;
2153
2154         case -EKEYEXPIRED:
2155         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2156                                  * in nfs4_exchange_id */
2157                 status = -EKEYEXPIRED;
2158                 break;
2159         default:
2160                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2161                                 __func__, status);
2162                 status = -EIO;
2163         }
2164
2165 out_unlock:
2166         mutex_unlock(&nfs_clid_init_mutex);
2167         dprintk("NFS: %s: status = %d\n", __func__, status);
2168         return status;
2169 }
2170
2171 #ifdef CONFIG_NFS_V4_1
2172 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2173 {
2174         struct nfs_client *clp = session->clp;
2175
2176         switch (err) {
2177         default:
2178                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2179                 break;
2180         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2181                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2182         }
2183         nfs4_schedule_lease_recovery(clp);
2184 }
2185 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2186
2187 void nfs41_notify_server(struct nfs_client *clp)
2188 {
2189         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2190         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2191         nfs4_schedule_state_manager(clp);
2192 }
2193
2194 static void nfs4_reset_all_state(struct nfs_client *clp)
2195 {
2196         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2197                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2198                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2199                 nfs4_state_start_reclaim_nograce(clp);
2200                 dprintk("%s: scheduling reset of all state for server %s!\n",
2201                                 __func__, clp->cl_hostname);
2202                 nfs4_schedule_state_manager(clp);
2203         }
2204 }
2205
2206 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2207 {
2208         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2209                 nfs4_state_start_reclaim_reboot(clp);
2210                 dprintk("%s: server %s rebooted!\n", __func__,
2211                                 clp->cl_hostname);
2212                 nfs4_schedule_state_manager(clp);
2213         }
2214 }
2215
2216 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2217 {
2218         nfs4_reset_all_state(clp);
2219         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2220 }
2221
2222 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2223 {
2224         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
2225         nfs4_schedule_state_manager(clp);
2226
2227         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2228 }
2229
2230 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2231 {
2232         /* FIXME: For now, we destroy all layouts. */
2233         pnfs_destroy_all_layouts(clp);
2234         /* FIXME: For now, we test all delegations+open state+locks. */
2235         nfs41_handle_some_state_revoked(clp);
2236         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2237                         clp->cl_hostname);
2238 }
2239
2240 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2241 {
2242         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2243         nfs4_schedule_state_manager(clp);
2244
2245         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2246                         clp->cl_hostname);
2247 }
2248
2249 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2250 {
2251         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2252                 &clp->cl_state) == 0)
2253                 nfs4_schedule_state_manager(clp);
2254 }
2255
2256 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2257 {
2258         if (!flags)
2259                 return;
2260
2261         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2262                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2263
2264         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2265                 nfs41_handle_server_reboot(clp);
2266         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2267                 nfs41_handle_all_state_revoked(clp);
2268         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2269                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2270                 nfs41_handle_some_state_revoked(clp);
2271         if (flags & SEQ4_STATUS_LEASE_MOVED)
2272                 nfs4_schedule_lease_moved_recovery(clp);
2273         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2274                 nfs41_handle_recallable_state_revoked(clp);
2275         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2276                 nfs41_handle_backchannel_fault(clp);
2277         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2278                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2279                 nfs41_handle_cb_path_down(clp);
2280 }
2281
2282 static int nfs4_reset_session(struct nfs_client *clp)
2283 {
2284         struct rpc_cred *cred;
2285         int status;
2286
2287         if (!nfs4_has_session(clp))
2288                 return 0;
2289         nfs4_begin_drain_session(clp);
2290         cred = nfs4_get_clid_cred(clp);
2291         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2292         switch (status) {
2293         case 0:
2294         case -NFS4ERR_BADSESSION:
2295         case -NFS4ERR_DEADSESSION:
2296                 break;
2297         case -NFS4ERR_BACK_CHAN_BUSY:
2298         case -NFS4ERR_DELAY:
2299                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2300                 status = 0;
2301                 ssleep(1);
2302                 goto out;
2303         default:
2304                 status = nfs4_recovery_handle_error(clp, status);
2305                 goto out;
2306         }
2307
2308         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2309         status = nfs4_proc_create_session(clp, cred);
2310         if (status) {
2311                 dprintk("%s: session reset failed with status %d for server %s!\n",
2312                         __func__, status, clp->cl_hostname);
2313                 status = nfs4_handle_reclaim_lease_error(clp, status);
2314                 goto out;
2315         }
2316         nfs41_finish_session_reset(clp);
2317         dprintk("%s: session reset was successful for server %s!\n",
2318                         __func__, clp->cl_hostname);
2319 out:
2320         if (cred)
2321                 put_rpccred(cred);
2322         return status;
2323 }
2324
2325 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2326 {
2327         struct rpc_cred *cred;
2328         int ret;
2329
2330         if (!nfs4_has_session(clp))
2331                 return 0;
2332         nfs4_begin_drain_session(clp);
2333         cred = nfs4_get_clid_cred(clp);
2334         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2335         if (cred)
2336                 put_rpccred(cred);
2337         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2338         switch (ret) {
2339         case 0:
2340                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2341                         __func__, clp->cl_hostname);
2342                 break;
2343         case -NFS4ERR_DELAY:
2344                 ssleep(1);
2345                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2346                 break;
2347         default:
2348                 return nfs4_recovery_handle_error(clp, ret);
2349         }
2350         return 0;
2351 }
2352 #else /* CONFIG_NFS_V4_1 */
2353 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2354
2355 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2356 {
2357         return 0;
2358 }
2359 #endif /* CONFIG_NFS_V4_1 */
2360
2361 static void nfs4_state_manager(struct nfs_client *clp)
2362 {
2363         int status = 0;
2364         const char *section = "", *section_sep = "";
2365
2366         /* Ensure exclusive access to NFSv4 state */
2367         do {
2368                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2369                         section = "purge state";
2370                         status = nfs4_purge_lease(clp);
2371                         if (status < 0)
2372                                 goto out_error;
2373                         continue;
2374                 }
2375
2376                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2377                         section = "lease expired";
2378                         /* We're going to have to re-establish a clientid */
2379                         status = nfs4_reclaim_lease(clp);
2380                         if (status < 0)
2381                                 goto out_error;
2382                         continue;
2383                 }
2384
2385                 /* Initialize or reset the session */
2386                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2387                         section = "reset session";
2388                         status = nfs4_reset_session(clp);
2389                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2390                                 continue;
2391                         if (status < 0)
2392                                 goto out_error;
2393                 }
2394
2395                 /* Send BIND_CONN_TO_SESSION */
2396                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2397                                 &clp->cl_state)) {
2398                         section = "bind conn to session";
2399                         status = nfs4_bind_conn_to_session(clp);
2400                         if (status < 0)
2401                                 goto out_error;
2402                         continue;
2403                 }
2404
2405                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2406                         section = "check lease";
2407                         status = nfs4_check_lease(clp);
2408                         if (status < 0)
2409                                 goto out_error;
2410                         continue;
2411                 }
2412
2413                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2414                         section = "migration";
2415                         status = nfs4_handle_migration(clp);
2416                         if (status < 0)
2417                                 goto out_error;
2418                 }
2419
2420                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2421                         section = "lease moved";
2422                         status = nfs4_handle_lease_moved(clp);
2423                         if (status < 0)
2424                                 goto out_error;
2425                 }
2426
2427                 /* First recover reboot state... */
2428                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2429                         section = "reclaim reboot";
2430                         status = nfs4_do_reclaim(clp,
2431                                 clp->cl_mvops->reboot_recovery_ops);
2432                         if (status == -EAGAIN)
2433                                 continue;
2434                         if (status < 0)
2435                                 goto out_error;
2436                         nfs4_state_end_reclaim_reboot(clp);
2437                 }
2438
2439                 /* Now recover expired state... */
2440                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2441                         section = "reclaim nograce";
2442                         status = nfs4_do_reclaim(clp,
2443                                 clp->cl_mvops->nograce_recovery_ops);
2444                         if (status == -EAGAIN)
2445                                 continue;
2446                         if (status < 0)
2447                                 goto out_error;
2448                 }
2449
2450                 nfs4_end_drain_session(clp);
2451                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2452                         nfs_client_return_marked_delegations(clp);
2453                         continue;
2454                 }
2455
2456                 nfs4_clear_state_manager_bit(clp);
2457                 /* Did we race with an attempt to give us more work? */
2458                 if (clp->cl_state == 0)
2459                         break;
2460                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2461                         break;
2462         } while (atomic_read(&clp->cl_count) > 1);
2463         return;
2464 out_error:
2465         if (strlen(section))
2466                 section_sep = ": ";
2467         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2468                         " with error %d\n", section_sep, section,
2469                         clp->cl_hostname, -status);
2470         ssleep(1);
2471         nfs4_end_drain_session(clp);
2472         nfs4_clear_state_manager_bit(clp);
2473 }
2474
2475 static int nfs4_run_state_manager(void *ptr)
2476 {
2477         struct nfs_client *clp = ptr;
2478
2479         allow_signal(SIGKILL);
2480         nfs4_state_manager(clp);
2481         nfs_put_client(clp);
2482         module_put_and_exit(0);
2483         return 0;
2484 }
2485
2486 /*
2487  * Local variables:
2488  *  c-basic-offset: 8
2489  * End:
2490  */