fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid = {
  69         { .data = { 0 } },
  70         .type = NFS4_SPECIAL_STATEID_TYPE,
  71 };
  72 const nfs4_stateid invalid_stateid = {
  73         {
  74                 /* Funky initialiser keeps older gcc versions happy */
  75                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
  76         },
  77         .type = NFS4_INVALID_STATEID_TYPE,
  78 };
  79
  80 const nfs4_stateid current_stateid = {
  81         {
  82                 /* Funky initialiser keeps older gcc versions happy */
  83                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
  84         },
  85         .type = NFS4_SPECIAL_STATEID_TYPE,
  86 };
  87
  88 static DEFINE_MUTEX(nfs_clid_init_mutex);
  89
  90 static int nfs4_setup_state_renewal(struct nfs_client *clp)
  91 {
  92         int status;
  93         struct nfs_fsinfo fsinfo;
  94         unsigned long now;
  95
  96         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
  97                 nfs4_schedule_state_renewal(clp);
  98                 return 0;
  99         }
 100
 101         now = jiffies;
 102         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 103         if (status == 0) {
 104                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 105                 nfs4_schedule_state_renewal(clp);
 106         }
 107
 108         return status;
 109 }
 110
 111 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
 112 {
 113         struct nfs4_setclientid_res clid = {
 114                 .clientid = clp->cl_clientid,
 115                 .confirm = clp->cl_confirm,
 116         };
 117         unsigned short port;
 118         int status;
 119         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 120
 121         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 122                 goto do_confirm;
 123         port = nn->nfs_callback_tcpport;
 124         if (clp->cl_addr.ss_family == AF_INET6)
 125                 port = nn->nfs_callback_tcpport6;
 126
 127         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 128         if (status != 0)
 129                 goto out;
 130         clp->cl_clientid = clid.clientid;
 131         clp->cl_confirm = clid.confirm;
 132         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 133 do_confirm:
 134         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 135         if (status != 0)
 136                 goto out;
 137         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 138         nfs4_setup_state_renewal(clp);
 139 out:
 140         return status;
 141 }
 142
 143 /**
 144  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 145  *
 146  * @clp: nfs_client under test
 147  * @result: OUT: found nfs_client, or clp
 148  * @cred: credential to use for trunking test
 149  *
 150  * Returns zero, a negative errno, or a negative NFS4ERR status.
 151  * If zero is returned, an nfs_client pointer is planted in
 152  * "result".
 153  *
 154  * Note: The returned client may not yet be marked ready.
 155  */
 156 int nfs40_discover_server_trunking(struct nfs_client *clp,
 157                                    struct nfs_client **result,
 158                                    const struct cred *cred)
 159 {
 160         struct nfs4_setclientid_res clid = {
 161                 .clientid = clp->cl_clientid,
 162                 .confirm = clp->cl_confirm,
 163         };
 164         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 165         unsigned short port;
 166         int status;
 167
 168         port = nn->nfs_callback_tcpport;
 169         if (clp->cl_addr.ss_family == AF_INET6)
 170                 port = nn->nfs_callback_tcpport6;
 171
 172         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 173         if (status != 0)
 174                 goto out;
 175         clp->cl_clientid = clid.clientid;
 176         clp->cl_confirm = clid.confirm;
 177
 178         status = nfs40_walk_client_list(clp, result, cred);
 179         if (status == 0) {
 180                 /* Sustain the lease, even if it's empty.  If the clientid4
 181                  * goes stale it's of no use for trunking discovery. */
 182                 nfs4_schedule_state_renewal(*result);
 183
 184                 /* If the client state need to recover, do it. */
 185                 if (clp->cl_state)
 186                         nfs4_schedule_state_manager(clp);
 187         }
 188 out:
 189         return status;
 190 }
 191
 192 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
 193 {
 194         return get_cred(rpc_machine_cred());
 195 }
 196
 197 static void nfs4_root_machine_cred(struct nfs_client *clp)
 198 {
 199
 200         /* Force root creds instead of machine */
 201         clp->cl_principal = NULL;
 202         clp->cl_rpcclient->cl_principal = NULL;
 203 }
 204
 205 static const struct cred *
 206 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 207 {
 208         const struct cred *cred = NULL;
 209         struct nfs4_state_owner *sp;
 210         struct rb_node *pos;
 211
 212         for (pos = rb_first(&server->state_owners);
 213              pos != NULL;
 214              pos = rb_next(pos)) {
 215                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 216                 if (list_empty(&sp->so_states))
 217                         continue;
 218                 cred = get_cred(sp->so_cred);
 219                 break;
 220         }
 221         return cred;
 222 }
 223
 224 /**
 225  * nfs4_get_renew_cred - Acquire credential for a renew operation
 226  * @clp: client state handle
 227  *
 228  * Returns an rpc_cred with reference count bumped, or NULL.
 229  * Caller must hold clp->cl_lock.
 230  */
 231 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
 232 {
 233         const struct cred *cred = NULL;
 234         struct nfs_server *server;
 235
 236         /* Use machine credentials if available */
 237         cred = nfs4_get_machine_cred(clp);
 238         if (cred != NULL)
 239                 goto out;
 240
 241         spin_lock(&clp->cl_lock);
 242         rcu_read_lock();
 243         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 244                 cred = nfs4_get_renew_cred_server_locked(server);
 245                 if (cred != NULL)
 246                         break;
 247         }
 248         rcu_read_unlock();
 249         spin_unlock(&clp->cl_lock);
 250
 251 out:
 252         return cred;
 253 }
 254
 255 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 256 {
 257         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 258                 spin_lock(&tbl->slot_tbl_lock);
 259                 nfs41_wake_slot_table(tbl);
 260                 spin_unlock(&tbl->slot_tbl_lock);
 261         }
 262 }
 263
 264 static void nfs4_end_drain_session(struct nfs_client *clp)
 265 {
 266         struct nfs4_session *ses = clp->cl_session;
 267
 268         if (clp->cl_slot_tbl) {
 269                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 270                 return;
 271         }
 272
 273         if (ses != NULL) {
 274                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 275                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 276         }
 277 }
 278
 279 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 280 {
 281         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 282         spin_lock(&tbl->slot_tbl_lock);
 283         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 284                 reinit_completion(&tbl->complete);
 285                 spin_unlock(&tbl->slot_tbl_lock);
 286                 return wait_for_completion_interruptible(&tbl->complete);
 287         }
 288         spin_unlock(&tbl->slot_tbl_lock);
 289         return 0;
 290 }
 291
 292 static int nfs4_begin_drain_session(struct nfs_client *clp)
 293 {
 294         struct nfs4_session *ses = clp->cl_session;
 295         int ret;
 296
 297         if (clp->cl_slot_tbl)
 298                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 299
 300         /* back channel */
 301         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 302         if (ret)
 303                 return ret;
 304         /* fore channel */
 305         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 306 }
 307
 308 #if defined(CONFIG_NFS_V4_1)
 309
 310 static void nfs41_finish_session_reset(struct nfs_client *clp)
 311 {
 312         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 313         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 314         /* create_session negotiated new slot table */
 315         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 316         nfs4_setup_state_renewal(clp);
 317 }
 318
 319 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
 320 {
 321         int status;
 322
 323         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 324                 goto do_confirm;
 325         status = nfs4_proc_exchange_id(clp, cred);
 326         if (status != 0)
 327                 goto out;
 328         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 329 do_confirm:
 330         status = nfs4_proc_create_session(clp, cred);
 331         if (status != 0)
 332                 goto out;
 333         nfs41_finish_session_reset(clp);
 334         nfs_mark_client_ready(clp, NFS_CS_READY);
 335 out:
 336         return status;
 337 }
 338
 339 /**
 340  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 341  *
 342  * @clp: nfs_client under test
 343  * @result: OUT: found nfs_client, or clp
 344  * @cred: credential to use for trunking test
 345  *
 346  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 347  * If NFS4_OK is returned, an nfs_client pointer is planted in
 348  * "result".
 349  *
 350  * Note: The returned client may not yet be marked ready.
 351  */
 352 int nfs41_discover_server_trunking(struct nfs_client *clp,
 353                                    struct nfs_client **result,
 354                                    const struct cred *cred)
 355 {
 356         int status;
 357
 358         status = nfs4_proc_exchange_id(clp, cred);
 359         if (status != NFS4_OK)
 360                 return status;
 361
 362         status = nfs41_walk_client_list(clp, result, cred);
 363         if (status < 0)
 364                 return status;
 365         if (clp != *result)
 366                 return 0;
 367
 368         /*
 369          * Purge state if the client id was established in a prior
 370          * instance and the client id could not have arrived on the
 371          * server via Transparent State Migration.
 372          */
 373         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
 374                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
 375                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 376                 else
 377                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 378         }
 379         nfs4_schedule_state_manager(clp);
 380         status = nfs_wait_client_init_complete(clp);
 381         if (status < 0)
 382                 nfs_put_client(clp);
 383         return status;
 384 }
 385
 386 #endif /* CONFIG_NFS_V4_1 */
 387
 388 /**
 389  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 390  * @clp: client state handle
 391  *
 392  * Returns a cred with reference count bumped, or NULL.
 393  */
 394 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
 395 {
 396         const struct cred *cred;
 397
 398         cred = nfs4_get_machine_cred(clp);
 399         return cred;
 400 }
 401
 402 static struct nfs4_state_owner *
 403 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
 404 {
 405         struct rb_node **p = &server->state_owners.rb_node,
 406                        *parent = NULL;
 407         struct nfs4_state_owner *sp;
 408         int cmp;
 409
 410         while (*p != NULL) {
 411                 parent = *p;
 412                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 413                 cmp = cred_fscmp(cred, sp->so_cred);
 414
 415                 if (cmp < 0)
 416                         p = &parent->rb_left;
 417                 else if (cmp > 0)
 418                         p = &parent->rb_right;
 419                 else {
 420                         if (!list_empty(&sp->so_lru))
 421                                 list_del_init(&sp->so_lru);
 422                         atomic_inc(&sp->so_count);
 423                         return sp;
 424                 }
 425         }
 426         return NULL;
 427 }
 428
 429 static struct nfs4_state_owner *
 430 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 431 {
 432         struct nfs_server *server = new->so_server;
 433         struct rb_node **p = &server->state_owners.rb_node,
 434                        *parent = NULL;
 435         struct nfs4_state_owner *sp;
 436         int cmp;
 437
 438         while (*p != NULL) {
 439                 parent = *p;
 440                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 441                 cmp = cred_fscmp(new->so_cred, sp->so_cred);
 442
 443                 if (cmp < 0)
 444                         p = &parent->rb_left;
 445                 else if (cmp > 0)
 446                         p = &parent->rb_right;
 447                 else {
 448                         if (!list_empty(&sp->so_lru))
 449                                 list_del_init(&sp->so_lru);
 450                         atomic_inc(&sp->so_count);
 451                         return sp;
 452                 }
 453         }
 454         rb_link_node(&new->so_server_node, parent, p);
 455         rb_insert_color(&new->so_server_node, &server->state_owners);
 456         return new;
 457 }
 458
 459 static void
 460 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 461 {
 462         struct nfs_server *server = sp->so_server;
 463
 464         if (!RB_EMPTY_NODE(&sp->so_server_node))
 465                 rb_erase(&sp->so_server_node, &server->state_owners);
 466 }
 467
 468 static void
 469 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 470 {
 471         sc->create_time = ktime_get();
 472         sc->flags = 0;
 473         sc->counter = 0;
 474         spin_lock_init(&sc->lock);
 475         INIT_LIST_HEAD(&sc->list);
 476         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 477 }
 478
 479 static void
 480 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 481 {
 482         rpc_destroy_wait_queue(&sc->wait);
 483 }
 484
 485 /*
 486  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 487  * create a new state_owner.
 488  *
 489  */
 490 static struct nfs4_state_owner *
 491 nfs4_alloc_state_owner(struct nfs_server *server,
 492                 const struct cred *cred,
 493                 gfp_t gfp_flags)
 494 {
 495         struct nfs4_state_owner *sp;
 496
 497         sp = kzalloc(sizeof(*sp), gfp_flags);
 498         if (!sp)
 499                 return NULL;
 500         sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
 501                                                 gfp_flags);
 502         if (sp->so_seqid.owner_id < 0) {
 503                 kfree(sp);
 504                 return NULL;
 505         }
 506         sp->so_server = server;
 507         sp->so_cred = get_cred(cred);
 508         spin_lock_init(&sp->so_lock);
 509         INIT_LIST_HEAD(&sp->so_states);
 510         nfs4_init_seqid_counter(&sp->so_seqid);
 511         atomic_set(&sp->so_count, 1);
 512         INIT_LIST_HEAD(&sp->so_lru);
 513         seqcount_init(&sp->so_reclaim_seqcount);
 514         mutex_init(&sp->so_delegreturn_mutex);
 515         return sp;
 516 }
 517
 518 static void
 519 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 520 {
 521         /* This state_owner is no longer usable, but must
 522          * remain in place so that state recovery can find it
 523          * and the opens associated with it.
 524          * It may also be used for new 'open' request to
 525          * return a delegation to the server.
 526          * So update the 'create_time' so that it looks like
 527          * a new state_owner.  This will cause the server to
 528          * request an OPEN_CONFIRM to start a new sequence.
 529          */
 530         sp->so_seqid.create_time = ktime_get();
 531 }
 532
 533 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 534 {
 535         nfs4_destroy_seqid_counter(&sp->so_seqid);
 536         put_cred(sp->so_cred);
 537         ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
 538         kfree(sp);
 539 }
 540
 541 static void nfs4_gc_state_owners(struct nfs_server *server)
 542 {
 543         struct nfs_client *clp = server->nfs_client;
 544         struct nfs4_state_owner *sp, *tmp;
 545         unsigned long time_min, time_max;
 546         LIST_HEAD(doomed);
 547
 548         spin_lock(&clp->cl_lock);
 549         time_max = jiffies;
 550         time_min = (long)time_max - (long)clp->cl_lease_time;
 551         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 552                 /* NB: LRU is sorted so that oldest is at the head */
 553                 if (time_in_range(sp->so_expires, time_min, time_max))
 554                         break;
 555                 list_move(&sp->so_lru, &doomed);
 556                 nfs4_remove_state_owner_locked(sp);
 557         }
 558         spin_unlock(&clp->cl_lock);
 559
 560         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 561                 list_del(&sp->so_lru);
 562                 nfs4_free_state_owner(sp);
 563         }
 564 }
 565
 566 /**
 567  * nfs4_get_state_owner - Look up a state owner given a credential
 568  * @server: nfs_server to search
 569  * @cred: RPC credential to match
 570  * @gfp_flags: allocation mode
 571  *
 572  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 573  */
 574 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 575                                               const struct cred *cred,
 576                                               gfp_t gfp_flags)
 577 {
 578         struct nfs_client *clp = server->nfs_client;
 579         struct nfs4_state_owner *sp, *new;
 580
 581         spin_lock(&clp->cl_lock);
 582         sp = nfs4_find_state_owner_locked(server, cred);
 583         spin_unlock(&clp->cl_lock);
 584         if (sp != NULL)
 585                 goto out;
 586         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 587         if (new == NULL)
 588                 goto out;
 589         spin_lock(&clp->cl_lock);
 590         sp = nfs4_insert_state_owner_locked(new);
 591         spin_unlock(&clp->cl_lock);
 592         if (sp != new)
 593                 nfs4_free_state_owner(new);
 594 out:
 595         nfs4_gc_state_owners(server);
 596         return sp;
 597 }
 598
 599 /**
 600  * nfs4_put_state_owner - Release a nfs4_state_owner
 601  * @sp: state owner data to release
 602  *
 603  * Note that we keep released state owners on an LRU
 604  * list.
 605  * This caches valid state owners so that they can be
 606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 607  * It also pins the uniquifier of dropped state owners for
 608  * a while, to ensure that those state owner names are
 609  * never reused.
 610  */
 611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 612 {
 613         struct nfs_server *server = sp->so_server;
 614         struct nfs_client *clp = server->nfs_client;
 615
 616         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 617                 return;
 618
 619         sp->so_expires = jiffies;
 620         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 621         spin_unlock(&clp->cl_lock);
 622 }
 623
 624 /**
 625  * nfs4_purge_state_owners - Release all cached state owners
 626  * @server: nfs_server with cached state owners to release
 627  * @head: resulting list of state owners
 628  *
 629  * Called at umount time.  Remaining state owners will be on
 630  * the LRU with ref count of zero.
 631  * Note that the state owners are not freed, but are added
 632  * to the list @head, which can later be used as an argument
 633  * to nfs4_free_state_owners.
 634  */
 635 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 636 {
 637         struct nfs_client *clp = server->nfs_client;
 638         struct nfs4_state_owner *sp, *tmp;
 639
 640         spin_lock(&clp->cl_lock);
 641         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 642                 list_move(&sp->so_lru, head);
 643                 nfs4_remove_state_owner_locked(sp);
 644         }
 645         spin_unlock(&clp->cl_lock);
 646 }
 647
 648 /**
 649  * nfs4_purge_state_owners - Release all cached state owners
 650  * @head: resulting list of state owners
 651  *
 652  * Frees a list of state owners that was generated by
 653  * nfs4_purge_state_owners
 654  */
 655 void nfs4_free_state_owners(struct list_head *head)
 656 {
 657         struct nfs4_state_owner *sp, *tmp;
 658
 659         list_for_each_entry_safe(sp, tmp, head, so_lru) {
 660                 list_del(&sp->so_lru);
 661                 nfs4_free_state_owner(sp);
 662         }
 663 }
 664
 665 static struct nfs4_state *
 666 nfs4_alloc_open_state(void)
 667 {
 668         struct nfs4_state *state;
 669
 670         state = kzalloc(sizeof(*state), GFP_NOFS);
 671         if (!state)
 672                 return NULL;
 673         refcount_set(&state->count, 1);
 674         INIT_LIST_HEAD(&state->lock_states);
 675         spin_lock_init(&state->state_lock);
 676         seqlock_init(&state->seqlock);
 677         init_waitqueue_head(&state->waitq);
 678         return state;
 679 }
 680
 681 void
 682 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 683 {
 684         if (state->state == fmode)
 685                 return;
 686         /* NB! List reordering - see the reclaim code for why.  */
 687         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 688                 if (fmode & FMODE_WRITE)
 689                         list_move(&state->open_states, &state->owner->so_states);
 690                 else
 691                         list_move_tail(&state->open_states, &state->owner->so_states);
 692         }
 693         state->state = fmode;
 694 }
 695
 696 static struct nfs4_state *
 697 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 698 {
 699         struct nfs_inode *nfsi = NFS_I(inode);
 700         struct nfs4_state *state;
 701
 702         list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
 703                 if (state->owner != owner)
 704                         continue;
 705                 if (!nfs4_valid_open_stateid(state))
 706                         continue;
 707                 if (refcount_inc_not_zero(&state->count))
 708                         return state;
 709         }
 710         return NULL;
 711 }
 712
 713 static void
 714 nfs4_free_open_state(struct nfs4_state *state)
 715 {
 716         kfree_rcu(state, rcu_head);
 717 }
 718
 719 struct nfs4_state *
 720 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 721 {
 722         struct nfs4_state *state, *new;
 723         struct nfs_inode *nfsi = NFS_I(inode);
 724
 725         rcu_read_lock();
 726         state = __nfs4_find_state_byowner(inode, owner);
 727         rcu_read_unlock();
 728         if (state)
 729                 goto out;
 730         new = nfs4_alloc_open_state();
 731         spin_lock(&owner->so_lock);
 732         spin_lock(&inode->i_lock);
 733         state = __nfs4_find_state_byowner(inode, owner);
 734         if (state == NULL && new != NULL) {
 735                 state = new;
 736                 state->owner = owner;
 737                 atomic_inc(&owner->so_count);
 738                 list_add_rcu(&state->inode_states, &nfsi->open_states);
 739                 ihold(inode);
 740                 state->inode = inode;
 741                 spin_unlock(&inode->i_lock);
 742                 /* Note: The reclaim code dictates that we add stateless
 743                  * and read-only stateids to the end of the list */
 744                 list_add_tail(&state->open_states, &owner->so_states);
 745                 spin_unlock(&owner->so_lock);
 746         } else {
 747                 spin_unlock(&inode->i_lock);
 748                 spin_unlock(&owner->so_lock);
 749                 if (new)
 750                         nfs4_free_open_state(new);
 751         }
 752 out:
 753         return state;
 754 }
 755
 756 void nfs4_put_open_state(struct nfs4_state *state)
 757 {
 758         struct inode *inode = state->inode;
 759         struct nfs4_state_owner *owner = state->owner;
 760
 761         if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
 762                 return;
 763         spin_lock(&inode->i_lock);
 764         list_del_rcu(&state->inode_states);
 765         list_del(&state->open_states);
 766         spin_unlock(&inode->i_lock);
 767         spin_unlock(&owner->so_lock);
 768         iput(inode);
 769         nfs4_free_open_state(state);
 770         nfs4_put_state_owner(owner);
 771 }
 772
 773 /*
 774  * Close the current file.
 775  */
 776 static void __nfs4_close(struct nfs4_state *state,
 777                 fmode_t fmode, gfp_t gfp_mask, int wait)
 778 {
 779         struct nfs4_state_owner *owner = state->owner;
 780         int call_close = 0;
 781         fmode_t newstate;
 782
 783         atomic_inc(&owner->so_count);
 784         /* Protect against nfs4_find_state() */
 785         spin_lock(&owner->so_lock);
 786         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 787                 case FMODE_READ:
 788                         state->n_rdonly--;
 789                         break;
 790                 case FMODE_WRITE:
 791                         state->n_wronly--;
 792                         break;
 793                 case FMODE_READ|FMODE_WRITE:
 794                         state->n_rdwr--;
 795         }
 796         newstate = FMODE_READ|FMODE_WRITE;
 797         if (state->n_rdwr == 0) {
 798                 if (state->n_rdonly == 0) {
 799                         newstate &= ~FMODE_READ;
 800                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 801                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 802                 }
 803                 if (state->n_wronly == 0) {
 804                         newstate &= ~FMODE_WRITE;
 805                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 806                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 807                 }
 808                 if (newstate == 0)
 809                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 810         }
 811         nfs4_state_set_mode_locked(state, newstate);
 812         spin_unlock(&owner->so_lock);
 813
 814         if (!call_close) {
 815                 nfs4_put_open_state(state);
 816                 nfs4_put_state_owner(owner);
 817         } else
 818                 nfs4_do_close(state, gfp_mask, wait);
 819 }
 820
 821 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 822 {
 823         __nfs4_close(state, fmode, GFP_NOFS, 0);
 824 }
 825
 826 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 827 {
 828         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 829 }
 830
 831 /*
 832  * Search the state->lock_states for an existing lock_owner
 833  * that is compatible with either of the given owners.
 834  * If the second is non-zero, then the first refers to a Posix-lock
 835  * owner (current->files) and the second refers to a flock/OFD
 836  * owner (struct file*).  In that case, prefer a match for the first
 837  * owner.
 838  * If both sorts of locks are held on the one file we cannot know
 839  * which stateid was intended to be used, so a "correct" choice cannot
 840  * be made.  Failing that, a "consistent" choice is preferable.  The
 841  * consistent choice we make is to prefer the first owner, that of a
 842  * Posix lock.
 843  */
 844 static struct nfs4_lock_state *
 845 __nfs4_find_lock_state(struct nfs4_state *state,
 846                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 847 {
 848         struct nfs4_lock_state *pos, *ret = NULL;
 849         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 850                 if (pos->ls_owner == fl_owner) {
 851                         ret = pos;
 852                         break;
 853                 }
 854                 if (pos->ls_owner == fl_owner2)
 855                         ret = pos;
 856         }
 857         if (ret)
 858                 refcount_inc(&ret->ls_count);
 859         return ret;
 860 }
 861
 862 /*
 863  * Return a compatible lock_state. If no initialized lock_state structure
 864  * exists, return an uninitialized one.
 865  *
 866  */
 867 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 868 {
 869         struct nfs4_lock_state *lsp;
 870         struct nfs_server *server = state->owner->so_server;
 871
 872         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 873         if (lsp == NULL)
 874                 return NULL;
 875         nfs4_init_seqid_counter(&lsp->ls_seqid);
 876         refcount_set(&lsp->ls_count, 1);
 877         lsp->ls_state = state;
 878         lsp->ls_owner = fl_owner;
 879         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 880         if (lsp->ls_seqid.owner_id < 0)
 881                 goto out_free;
 882         INIT_LIST_HEAD(&lsp->ls_locks);
 883         return lsp;
 884 out_free:
 885         kfree(lsp);
 886         return NULL;
 887 }
 888
 889 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 890 {
 891         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 892         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 893         kfree(lsp);
 894 }
 895
 896 /*
 897  * Return a compatible lock_state. If no initialized lock_state structure
 898  * exists, return an uninitialized one.
 899  *
 900  */
 901 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 902 {
 903         struct nfs4_lock_state *lsp, *new = NULL;
 904
 905         for(;;) {
 906                 spin_lock(&state->state_lock);
 907                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 908                 if (lsp != NULL)
 909                         break;
 910                 if (new != NULL) {
 911                         list_add(&new->ls_locks, &state->lock_states);
 912                         set_bit(LK_STATE_IN_USE, &state->flags);
 913                         lsp = new;
 914                         new = NULL;
 915                         break;
 916                 }
 917                 spin_unlock(&state->state_lock);
 918                 new = nfs4_alloc_lock_state(state, owner);
 919                 if (new == NULL)
 920                         return NULL;
 921         }
 922         spin_unlock(&state->state_lock);
 923         if (new != NULL)
 924                 nfs4_free_lock_state(state->owner->so_server, new);
 925         return lsp;
 926 }
 927
 928 /*
 929  * Release reference to lock_state, and free it if we see that
 930  * it is no longer in use
 931  */
 932 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 933 {
 934         struct nfs_server *server;
 935         struct nfs4_state *state;
 936
 937         if (lsp == NULL)
 938                 return;
 939         state = lsp->ls_state;
 940         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 941                 return;
 942         list_del(&lsp->ls_locks);
 943         if (list_empty(&state->lock_states))
 944                 clear_bit(LK_STATE_IN_USE, &state->flags);
 945         spin_unlock(&state->state_lock);
 946         server = state->owner->so_server;
 947         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 948                 struct nfs_client *clp = server->nfs_client;
 949
 950                 clp->cl_mvops->free_lock_state(server, lsp);
 951         } else
 952                 nfs4_free_lock_state(server, lsp);
 953 }
 954
 955 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 956 {
 957         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 958
 959         dst->fl_u.nfs4_fl.owner = lsp;
 960         refcount_inc(&lsp->ls_count);
 961 }
 962
 963 static void nfs4_fl_release_lock(struct file_lock *fl)
 964 {
 965         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 966 }
 967
 968 static const struct file_lock_operations nfs4_fl_lock_ops = {
 969         .fl_copy_lock = nfs4_fl_copy_lock,
 970         .fl_release_private = nfs4_fl_release_lock,
 971 };
 972
 973 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 974 {
 975         struct nfs4_lock_state *lsp;
 976
 977         if (fl->fl_ops != NULL)
 978                 return 0;
 979         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 980         if (lsp == NULL)
 981                 return -ENOMEM;
 982         fl->fl_u.nfs4_fl.owner = lsp;
 983         fl->fl_ops = &nfs4_fl_lock_ops;
 984         return 0;
 985 }
 986
 987 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 988                 struct nfs4_state *state,
 989                 const struct nfs_lock_context *l_ctx)
 990 {
 991         struct nfs4_lock_state *lsp;
 992         fl_owner_t fl_owner, fl_flock_owner;
 993         int ret = -ENOENT;
 994
 995         if (l_ctx == NULL)
 996                 goto out;
 997
 998         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 999                 goto out;
1000
1001         fl_owner = l_ctx->lockowner;
1002         fl_flock_owner = l_ctx->open_context->flock_owner;
1003
1004         spin_lock(&state->state_lock);
1005         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1006         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1007                 ret = -EIO;
1008         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1009                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1010                 ret = 0;
1011         }
1012         spin_unlock(&state->state_lock);
1013         nfs4_put_lock_state(lsp);
1014 out:
1015         return ret;
1016 }
1017
1018 bool nfs4_refresh_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1019 {
1020         bool ret;
1021         int seq;
1022
1023         do {
1024                 ret = false;
1025                 seq = read_seqbegin(&state->seqlock);
1026                 if (nfs4_state_match_open_stateid_other(state, dst)) {
1027                         dst->seqid = state->open_stateid.seqid;
1028                         ret = true;
1029                 }
1030         } while (read_seqretry(&state->seqlock, seq));
1031         return ret;
1032 }
1033
1034 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1035 {
1036         bool ret;
1037         const nfs4_stateid *src;
1038         int seq;
1039
1040         do {
1041                 ret = false;
1042                 src = &zero_stateid;
1043                 seq = read_seqbegin(&state->seqlock);
1044                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1045                         src = &state->open_stateid;
1046                         ret = true;
1047                 }
1048                 nfs4_stateid_copy(dst, src);
1049         } while (read_seqretry(&state->seqlock, seq));
1050         return ret;
1051 }
1052
1053 /*
1054  * Byte-range lock aware utility to initialize the stateid of read/write
1055  * requests.
1056  */
1057 int nfs4_select_rw_stateid(struct nfs4_state *state,
1058                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1059                 nfs4_stateid *dst, const struct cred **cred)
1060 {
1061         int ret;
1062
1063         if (!nfs4_valid_open_stateid(state))
1064                 return -EIO;
1065         if (cred != NULL)
1066                 *cred = NULL;
1067         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1068         if (ret == -EIO)
1069                 /* A lost lock - don't even consider delegations */
1070                 goto out;
1071         /* returns true if delegation stateid found and copied */
1072         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1073                 ret = 0;
1074                 goto out;
1075         }
1076         if (ret != -ENOENT)
1077                 /* nfs4_copy_delegation_stateid() didn't over-write
1078                  * dst, so it still has the lock stateid which we now
1079                  * choose to use.
1080                  */
1081                 goto out;
1082         ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1083 out:
1084         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1085                 dst->seqid = 0;
1086         return ret;
1087 }
1088
1089 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1090 {
1091         struct nfs_seqid *new;
1092
1093         new = kmalloc(sizeof(*new), gfp_mask);
1094         if (new == NULL)
1095                 return ERR_PTR(-ENOMEM);
1096         new->sequence = counter;
1097         INIT_LIST_HEAD(&new->list);
1098         new->task = NULL;
1099         return new;
1100 }
1101
1102 void nfs_release_seqid(struct nfs_seqid *seqid)
1103 {
1104         struct nfs_seqid_counter *sequence;
1105
1106         if (seqid == NULL || list_empty(&seqid->list))
1107                 return;
1108         sequence = seqid->sequence;
1109         spin_lock(&sequence->lock);
1110         list_del_init(&seqid->list);
1111         if (!list_empty(&sequence->list)) {
1112                 struct nfs_seqid *next;
1113
1114                 next = list_first_entry(&sequence->list,
1115                                 struct nfs_seqid, list);
1116                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1117         }
1118         spin_unlock(&sequence->lock);
1119 }
1120
1121 void nfs_free_seqid(struct nfs_seqid *seqid)
1122 {
1123         nfs_release_seqid(seqid);
1124         kfree(seqid);
1125 }
1126
1127 /*
1128  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1129  * failed with a seqid incrementing error -
1130  * see comments nfs4.h:seqid_mutating_error()
1131  */
1132 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1133 {
1134         switch (status) {
1135                 case 0:
1136                         break;
1137                 case -NFS4ERR_BAD_SEQID:
1138                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1139                                 return;
1140                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1141                                         " sequence-id error on an"
1142                                         " unconfirmed sequence %p!\n",
1143                                         seqid->sequence);
1144                 case -NFS4ERR_STALE_CLIENTID:
1145                 case -NFS4ERR_STALE_STATEID:
1146                 case -NFS4ERR_BAD_STATEID:
1147                 case -NFS4ERR_BADXDR:
1148                 case -NFS4ERR_RESOURCE:
1149                 case -NFS4ERR_NOFILEHANDLE:
1150                 case -NFS4ERR_MOVED:
1151                         /* Non-seqid mutating errors */
1152                         return;
1153         };
1154         /*
1155          * Note: no locking needed as we are guaranteed to be first
1156          * on the sequence list
1157          */
1158         seqid->sequence->counter++;
1159 }
1160
1161 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1162 {
1163         struct nfs4_state_owner *sp;
1164
1165         if (seqid == NULL)
1166                 return;
1167
1168         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1169         if (status == -NFS4ERR_BAD_SEQID)
1170                 nfs4_reset_state_owner(sp);
1171         if (!nfs4_has_session(sp->so_server->nfs_client))
1172                 nfs_increment_seqid(status, seqid);
1173 }
1174
1175 /*
1176  * Increment the seqid if the LOCK/LOCKU succeeded, or
1177  * failed with a seqid incrementing error -
1178  * see comments nfs4.h:seqid_mutating_error()
1179  */
1180 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1181 {
1182         if (seqid != NULL)
1183                 nfs_increment_seqid(status, seqid);
1184 }
1185
1186 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1187 {
1188         struct nfs_seqid_counter *sequence;
1189         int status = 0;
1190
1191         if (seqid == NULL)
1192                 goto out;
1193         sequence = seqid->sequence;
1194         spin_lock(&sequence->lock);
1195         seqid->task = task;
1196         if (list_empty(&seqid->list))
1197                 list_add_tail(&seqid->list, &sequence->list);
1198         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1199                 goto unlock;
1200         rpc_sleep_on(&sequence->wait, task, NULL);
1201         status = -EAGAIN;
1202 unlock:
1203         spin_unlock(&sequence->lock);
1204 out:
1205         return status;
1206 }
1207
1208 static int nfs4_run_state_manager(void *);
1209
1210 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1211 {
1212         smp_mb__before_atomic();
1213         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1214         smp_mb__after_atomic();
1215         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1216         rpc_wake_up(&clp->cl_rpcwaitq);
1217 }
1218
1219 /*
1220  * Schedule the nfs_client asynchronous state management routine
1221  */
1222 void nfs4_schedule_state_manager(struct nfs_client *clp)
1223 {
1224         struct task_struct *task;
1225         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1226
1227         set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1228         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1229                 return;
1230         __module_get(THIS_MODULE);
1231         refcount_inc(&clp->cl_count);
1232
1233         /* The rcu_read_lock() is not strictly necessary, as the state
1234          * manager is the only thread that ever changes the rpc_xprt
1235          * after it's initialized.  At this point, we're single threaded. */
1236         rcu_read_lock();
1237         snprintf(buf, sizeof(buf), "%s-manager",
1238                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1239         rcu_read_unlock();
1240         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1241         if (IS_ERR(task)) {
1242                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1243                         __func__, PTR_ERR(task));
1244                 nfs4_clear_state_manager_bit(clp);
1245                 nfs_put_client(clp);
1246                 module_put(THIS_MODULE);
1247         }
1248 }
1249
1250 /*
1251  * Schedule a lease recovery attempt
1252  */
1253 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1254 {
1255         if (!clp)
1256                 return;
1257         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1258                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1259         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1260                         clp->cl_hostname);
1261         nfs4_schedule_state_manager(clp);
1262 }
1263 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1264
1265 /**
1266  * nfs4_schedule_migration_recovery - trigger migration recovery
1267  *
1268  * @server: FSID that is migrating
1269  *
1270  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1271  * value is returned.
1272  */
1273 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1274 {
1275         struct nfs_client *clp = server->nfs_client;
1276
1277         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1278                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1279                                 clp->cl_hostname);
1280                 return -NFS4ERR_IO;
1281         }
1282
1283         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1284                 return -NFS4ERR_IO;
1285
1286         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1287                         __func__,
1288                         (unsigned long long)server->fsid.major,
1289                         (unsigned long long)server->fsid.minor,
1290                         clp->cl_hostname);
1291
1292         set_bit(NFS_MIG_IN_TRANSITION,
1293                         &((struct nfs_server *)server)->mig_status);
1294         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1295
1296         nfs4_schedule_state_manager(clp);
1297         return 0;
1298 }
1299 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1300
1301 /**
1302  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1303  *
1304  * @clp: server to check for moved leases
1305  *
1306  */
1307 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1308 {
1309         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1310                 __func__, clp->cl_clientid, clp->cl_hostname);
1311
1312         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1313         nfs4_schedule_state_manager(clp);
1314 }
1315 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1316
1317 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1318 {
1319         int res;
1320
1321         might_sleep();
1322
1323         refcount_inc(&clp->cl_count);
1324         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1325                                  nfs_wait_bit_killable, TASK_KILLABLE);
1326         if (res)
1327                 goto out;
1328         if (clp->cl_cons_state < 0)
1329                 res = clp->cl_cons_state;
1330 out:
1331         nfs_put_client(clp);
1332         return res;
1333 }
1334
1335 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1336 {
1337         unsigned int loop;
1338         int ret;
1339
1340         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1341                 ret = nfs4_wait_clnt_recover(clp);
1342                 if (ret != 0)
1343                         break;
1344                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1345                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1346                         break;
1347                 nfs4_schedule_state_manager(clp);
1348                 ret = -EIO;
1349         }
1350         return ret;
1351 }
1352
1353 /*
1354  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1355  * @clp: client to process
1356  *
1357  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1358  * resend of the SETCLIENTID and hence re-establish the
1359  * callback channel. Then return all existing delegations.
1360  */
1361 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1362 {
1363         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1364         nfs_expire_all_delegations(clp);
1365         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1366                         clp->cl_hostname);
1367 }
1368
1369 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1370 {
1371         nfs40_handle_cb_pathdown(clp);
1372         nfs4_schedule_state_manager(clp);
1373 }
1374
1375 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1376 {
1377
1378         if (!nfs4_valid_open_stateid(state))
1379                 return 0;
1380         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1381         /* Don't recover state that expired before the reboot */
1382         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1383                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1384                 return 0;
1385         }
1386         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1387         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1388         return 1;
1389 }
1390
1391 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1392 {
1393         if (!nfs4_valid_open_stateid(state))
1394                 return 0;
1395         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1396         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1397         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1398         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1399         return 1;
1400 }
1401
1402 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1403 {
1404         struct nfs_client *clp = server->nfs_client;
1405
1406         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1407                 return -EBADF;
1408         nfs_inode_find_delegation_state_and_recover(state->inode,
1409                         &state->stateid);
1410         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1411                         clp->cl_hostname);
1412         nfs4_schedule_state_manager(clp);
1413         return 0;
1414 }
1415 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1416
1417 static struct nfs4_lock_state *
1418 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1419                 const nfs4_stateid *stateid)
1420 {
1421         struct nfs4_lock_state *pos;
1422
1423         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1424                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1425                         continue;
1426                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1427                         return pos;
1428         }
1429         return NULL;
1430 }
1431
1432 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1433                 const nfs4_stateid *stateid)
1434 {
1435         bool found = false;
1436
1437         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1438                 spin_lock(&state->state_lock);
1439                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1440                         found = true;
1441                 spin_unlock(&state->state_lock);
1442         }
1443         return found;
1444 }
1445
1446 void nfs_inode_find_state_and_recover(struct inode *inode,
1447                 const nfs4_stateid *stateid)
1448 {
1449         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1450         struct nfs_inode *nfsi = NFS_I(inode);
1451         struct nfs_open_context *ctx;
1452         struct nfs4_state *state;
1453         bool found = false;
1454
1455         rcu_read_lock();
1456         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1457                 state = ctx->state;
1458                 if (state == NULL)
1459                         continue;
1460                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1461                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1462                         found = true;
1463                         continue;
1464                 }
1465                 if (nfs4_stateid_match_other(&state->open_stateid, stateid) &&
1466                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1467                         found = true;
1468                         continue;
1469                 }
1470                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1471                     nfs4_state_mark_reclaim_nograce(clp, state))
1472                         found = true;
1473         }
1474         rcu_read_unlock();
1475
1476         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1477         if (found)
1478                 nfs4_schedule_state_manager(clp);
1479 }
1480
1481 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1482 {
1483         struct inode *inode = state->inode;
1484         struct nfs_inode *nfsi = NFS_I(inode);
1485         struct nfs_open_context *ctx;
1486
1487         rcu_read_lock();
1488         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1489                 if (ctx->state != state)
1490                         continue;
1491                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1492                 pr_warn("NFSv4: state recovery failed for open file %pd2, "
1493                                 "error = %d\n", ctx->dentry, err);
1494         }
1495         rcu_read_unlock();
1496 }
1497
1498 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1499 {
1500         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1501         nfs4_state_mark_open_context_bad(state, error);
1502 }
1503
1504
1505 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1506 {
1507         struct inode *inode = state->inode;
1508         struct nfs_inode *nfsi = NFS_I(inode);
1509         struct file_lock *fl;
1510         struct nfs4_lock_state *lsp;
1511         int status = 0;
1512         struct file_lock_context *flctx = inode->i_flctx;
1513         struct list_head *list;
1514
1515         if (flctx == NULL)
1516                 return 0;
1517
1518         list = &flctx->flc_posix;
1519
1520         /* Guard against delegation returns and new lock/unlock calls */
1521         down_write(&nfsi->rwsem);
1522         spin_lock(&flctx->flc_lock);
1523 restart:
1524         list_for_each_entry(fl, list, fl_list) {
1525                 if (nfs_file_open_context(fl->fl_file)->state != state)
1526                         continue;
1527                 spin_unlock(&flctx->flc_lock);
1528                 status = ops->recover_lock(state, fl);
1529                 switch (status) {
1530                 case 0:
1531                         break;
1532                 case -ETIMEDOUT:
1533                 case -ESTALE:
1534                 case -NFS4ERR_ADMIN_REVOKED:
1535                 case -NFS4ERR_STALE_STATEID:
1536                 case -NFS4ERR_BAD_STATEID:
1537                 case -NFS4ERR_EXPIRED:
1538                 case -NFS4ERR_NO_GRACE:
1539                 case -NFS4ERR_STALE_CLIENTID:
1540                 case -NFS4ERR_BADSESSION:
1541                 case -NFS4ERR_BADSLOT:
1542                 case -NFS4ERR_BAD_HIGH_SLOT:
1543                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1544                         goto out;
1545                 default:
1546                         pr_err("NFS: %s: unhandled error %d\n",
1547                                         __func__, status);
1548                         /* Fall through */
1549                 case -ENOMEM:
1550                 case -NFS4ERR_DENIED:
1551                 case -NFS4ERR_RECLAIM_BAD:
1552                 case -NFS4ERR_RECLAIM_CONFLICT:
1553                         lsp = fl->fl_u.nfs4_fl.owner;
1554                         if (lsp)
1555                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1556                         status = 0;
1557                 }
1558                 spin_lock(&flctx->flc_lock);
1559         }
1560         if (list == &flctx->flc_posix) {
1561                 list = &flctx->flc_flock;
1562                 goto restart;
1563         }
1564         spin_unlock(&flctx->flc_lock);
1565 out:
1566         up_write(&nfsi->rwsem);
1567         return status;
1568 }
1569
1570 #ifdef CONFIG_NFS_V4_2
1571 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1572 {
1573         struct nfs4_copy_state *copy;
1574
1575         if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags))
1576                 return;
1577
1578         spin_lock(&sp->so_server->nfs_client->cl_lock);
1579         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1580                 if (!nfs4_stateid_match_other(&state->stateid, &copy->parent_state->stateid))
1581                         continue;
1582                 copy->flags = 1;
1583                 complete(&copy->completion);
1584                 break;
1585         }
1586         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1587 }
1588 #else /* !CONFIG_NFS_V4_2 */
1589 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1590                                          struct nfs4_state *state)
1591 {
1592 }
1593 #endif /* CONFIG_NFS_V4_2 */
1594
1595 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1596                                      const struct nfs4_state_recovery_ops *ops)
1597 {
1598         struct nfs4_lock_state *lock;
1599         int status;
1600
1601         status = ops->recover_open(sp, state);
1602         if (status < 0)
1603                 return status;
1604
1605         status = nfs4_reclaim_locks(state, ops);
1606         if (status < 0)
1607                 return status;
1608
1609         if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1610                 spin_lock(&state->state_lock);
1611                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1612                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1613                                 pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1614                 }
1615                 spin_unlock(&state->state_lock);
1616         }
1617
1618         nfs42_complete_copies(sp, state);
1619         clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1620         return status;
1621 }
1622
1623 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1624 {
1625         struct nfs4_state *state;
1626         unsigned int loop = 0;
1627         int status = 0;
1628
1629         /* Note: we rely on the sp->so_states list being ordered
1630          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1631          * states first.
1632          * This is needed to ensure that the server won't give us any
1633          * read delegations that we have to return if, say, we are
1634          * recovering after a network partition or a reboot from a
1635          * server that doesn't support a grace period.
1636          */
1637         spin_lock(&sp->so_lock);
1638         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1639 restart:
1640         list_for_each_entry(state, &sp->so_states, open_states) {
1641                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1642                         continue;
1643                 if (!nfs4_valid_open_stateid(state))
1644                         continue;
1645                 if (state->state == 0)
1646                         continue;
1647                 refcount_inc(&state->count);
1648                 spin_unlock(&sp->so_lock);
1649                 status = __nfs4_reclaim_open_state(sp, state, ops);
1650
1651                 switch (status) {
1652                 default:
1653                         if (status >= 0) {
1654                                 loop = 0;
1655                                 break;
1656                         }
1657                         printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1658                         /* Fall through */
1659                 case -ENOENT:
1660                 case -ENOMEM:
1661                 case -EACCES:
1662                 case -EROFS:
1663                 case -EIO:
1664                 case -ESTALE:
1665                         /* Open state on this file cannot be recovered */
1666                         nfs4_state_mark_recovery_failed(state, status);
1667                         break;
1668                 case -EAGAIN:
1669                         ssleep(1);
1670                         if (loop++ < 10) {
1671                                 set_bit(ops->state_flag_bit, &state->flags);
1672                                 break;
1673                         }
1674                         /* Fall through */
1675                 case -NFS4ERR_ADMIN_REVOKED:
1676                 case -NFS4ERR_STALE_STATEID:
1677                 case -NFS4ERR_OLD_STATEID:
1678                 case -NFS4ERR_BAD_STATEID:
1679                 case -NFS4ERR_RECLAIM_BAD:
1680                 case -NFS4ERR_RECLAIM_CONFLICT:
1681                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1682                         break;
1683                 case -NFS4ERR_EXPIRED:
1684                 case -NFS4ERR_NO_GRACE:
1685                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1686                         /* Fall through */
1687                 case -NFS4ERR_STALE_CLIENTID:
1688                 case -NFS4ERR_BADSESSION:
1689                 case -NFS4ERR_BADSLOT:
1690                 case -NFS4ERR_BAD_HIGH_SLOT:
1691                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1692                 case -ETIMEDOUT:
1693                         goto out_err;
1694                 }
1695                 nfs4_put_open_state(state);
1696                 spin_lock(&sp->so_lock);
1697                 goto restart;
1698         }
1699         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1700         spin_unlock(&sp->so_lock);
1701         return 0;
1702 out_err:
1703         nfs4_put_open_state(state);
1704         spin_lock(&sp->so_lock);
1705         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1706         spin_unlock(&sp->so_lock);
1707         return status;
1708 }
1709
1710 static void nfs4_clear_open_state(struct nfs4_state *state)
1711 {
1712         struct nfs4_lock_state *lock;
1713
1714         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1715         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1716         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1717         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1718         spin_lock(&state->state_lock);
1719         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1720                 lock->ls_seqid.flags = 0;
1721                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1722         }
1723         spin_unlock(&state->state_lock);
1724 }
1725
1726 static void nfs4_reset_seqids(struct nfs_server *server,
1727         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1728 {
1729         struct nfs_client *clp = server->nfs_client;
1730         struct nfs4_state_owner *sp;
1731         struct rb_node *pos;
1732         struct nfs4_state *state;
1733
1734         spin_lock(&clp->cl_lock);
1735         for (pos = rb_first(&server->state_owners);
1736              pos != NULL;
1737              pos = rb_next(pos)) {
1738                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1739                 sp->so_seqid.flags = 0;
1740                 spin_lock(&sp->so_lock);
1741                 list_for_each_entry(state, &sp->so_states, open_states) {
1742                         if (mark_reclaim(clp, state))
1743                                 nfs4_clear_open_state(state);
1744                 }
1745                 spin_unlock(&sp->so_lock);
1746         }
1747         spin_unlock(&clp->cl_lock);
1748 }
1749
1750 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1751         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1752 {
1753         struct nfs_server *server;
1754
1755         rcu_read_lock();
1756         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1757                 nfs4_reset_seqids(server, mark_reclaim);
1758         rcu_read_unlock();
1759 }
1760
1761 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1762 {
1763         /* Mark all delegations for reclaim */
1764         nfs_delegation_mark_reclaim(clp);
1765         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1766 }
1767
1768 static int nfs4_reclaim_complete(struct nfs_client *clp,
1769                                  const struct nfs4_state_recovery_ops *ops,
1770                                  const struct cred *cred)
1771 {
1772         /* Notify the server we're done reclaiming our state */
1773         if (ops->reclaim_complete)
1774                 return ops->reclaim_complete(clp, cred);
1775         return 0;
1776 }
1777
1778 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1779 {
1780         struct nfs_client *clp = server->nfs_client;
1781         struct nfs4_state_owner *sp;
1782         struct rb_node *pos;
1783         struct nfs4_state *state;
1784
1785         spin_lock(&clp->cl_lock);
1786         for (pos = rb_first(&server->state_owners);
1787              pos != NULL;
1788              pos = rb_next(pos)) {
1789                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1790                 spin_lock(&sp->so_lock);
1791                 list_for_each_entry(state, &sp->so_states, open_states) {
1792                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1793                                                 &state->flags))
1794                                 continue;
1795                         nfs4_state_mark_reclaim_nograce(clp, state);
1796                 }
1797                 spin_unlock(&sp->so_lock);
1798         }
1799         spin_unlock(&clp->cl_lock);
1800 }
1801
1802 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1803 {
1804         struct nfs_server *server;
1805
1806         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1807                 return 0;
1808
1809         rcu_read_lock();
1810         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1811                 nfs4_clear_reclaim_server(server);
1812         rcu_read_unlock();
1813
1814         nfs_delegation_reap_unclaimed(clp);
1815         return 1;
1816 }
1817
1818 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1819 {
1820         const struct nfs4_state_recovery_ops *ops;
1821         const struct cred *cred;
1822         int err;
1823
1824         if (!nfs4_state_clear_reclaim_reboot(clp))
1825                 return;
1826         ops = clp->cl_mvops->reboot_recovery_ops;
1827         cred = nfs4_get_clid_cred(clp);
1828         err = nfs4_reclaim_complete(clp, ops, cred);
1829         put_cred(cred);
1830         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1831                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1832 }
1833
1834 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1835 {
1836         nfs_mark_test_expired_all_delegations(clp);
1837         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1838 }
1839
1840 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1841 {
1842         switch (error) {
1843         case 0:
1844                 break;
1845         case -NFS4ERR_CB_PATH_DOWN:
1846                 nfs40_handle_cb_pathdown(clp);
1847                 break;
1848         case -NFS4ERR_NO_GRACE:
1849                 nfs4_state_end_reclaim_reboot(clp);
1850                 break;
1851         case -NFS4ERR_STALE_CLIENTID:
1852                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1853                 nfs4_state_start_reclaim_reboot(clp);
1854                 break;
1855         case -NFS4ERR_EXPIRED:
1856                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1857                 nfs4_state_start_reclaim_nograce(clp);
1858                 break;
1859         case -NFS4ERR_BADSESSION:
1860         case -NFS4ERR_BADSLOT:
1861         case -NFS4ERR_BAD_HIGH_SLOT:
1862         case -NFS4ERR_DEADSESSION:
1863         case -NFS4ERR_SEQ_FALSE_RETRY:
1864         case -NFS4ERR_SEQ_MISORDERED:
1865                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1866                 /* Zero session reset errors */
1867                 break;
1868         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1869                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1870                 break;
1871         default:
1872                 dprintk("%s: failed to handle error %d for server %s\n",
1873                                 __func__, error, clp->cl_hostname);
1874                 return error;
1875         }
1876         dprintk("%s: handled error %d for server %s\n", __func__, error,
1877                         clp->cl_hostname);
1878         return 0;
1879 }
1880
1881 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1882 {
1883         struct nfs4_state_owner *sp;
1884         struct nfs_server *server;
1885         struct rb_node *pos;
1886         LIST_HEAD(freeme);
1887         int status = 0;
1888
1889 restart:
1890         rcu_read_lock();
1891         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1892                 nfs4_purge_state_owners(server, &freeme);
1893                 spin_lock(&clp->cl_lock);
1894                 for (pos = rb_first(&server->state_owners);
1895                      pos != NULL;
1896                      pos = rb_next(pos)) {
1897                         sp = rb_entry(pos,
1898                                 struct nfs4_state_owner, so_server_node);
1899                         if (!test_and_clear_bit(ops->owner_flag_bit,
1900                                                         &sp->so_flags))
1901                                 continue;
1902                         if (!atomic_inc_not_zero(&sp->so_count))
1903                                 continue;
1904                         spin_unlock(&clp->cl_lock);
1905                         rcu_read_unlock();
1906
1907                         status = nfs4_reclaim_open_state(sp, ops);
1908                         if (status < 0) {
1909                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1910                                 nfs4_put_state_owner(sp);
1911                                 status = nfs4_recovery_handle_error(clp, status);
1912                                 return (status != 0) ? status : -EAGAIN;
1913                         }
1914
1915                         nfs4_put_state_owner(sp);
1916                         goto restart;
1917                 }
1918                 spin_unlock(&clp->cl_lock);
1919         }
1920         rcu_read_unlock();
1921         nfs4_free_state_owners(&freeme);
1922         return 0;
1923 }
1924
1925 static int nfs4_check_lease(struct nfs_client *clp)
1926 {
1927         const struct cred *cred;
1928         const struct nfs4_state_maintenance_ops *ops =
1929                 clp->cl_mvops->state_renewal_ops;
1930         int status;
1931
1932         /* Is the client already known to have an expired lease? */
1933         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1934                 return 0;
1935         cred = ops->get_state_renewal_cred(clp);
1936         if (cred == NULL) {
1937                 cred = nfs4_get_clid_cred(clp);
1938                 status = -ENOKEY;
1939                 if (cred == NULL)
1940                         goto out;
1941         }
1942         status = ops->renew_lease(clp, cred);
1943         put_cred(cred);
1944         if (status == -ETIMEDOUT) {
1945                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1946                 return 0;
1947         }
1948 out:
1949         return nfs4_recovery_handle_error(clp, status);
1950 }
1951
1952 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1953  * and for recoverable errors on EXCHANGE_ID for v4.1
1954  */
1955 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1956 {
1957         switch (status) {
1958         case -NFS4ERR_SEQ_MISORDERED:
1959                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1960                         return -ESERVERFAULT;
1961                 /* Lease confirmation error: retry after purging the lease */
1962                 ssleep(1);
1963                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1964                 break;
1965         case -NFS4ERR_STALE_CLIENTID:
1966                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1967                 nfs4_state_start_reclaim_reboot(clp);
1968                 break;
1969         case -NFS4ERR_CLID_INUSE:
1970                 pr_err("NFS: Server %s reports our clientid is in use\n",
1971                         clp->cl_hostname);
1972                 nfs_mark_client_ready(clp, -EPERM);
1973                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1974                 return -EPERM;
1975         case -EACCES:
1976         case -NFS4ERR_DELAY:
1977         case -EAGAIN:
1978                 ssleep(1);
1979                 break;
1980
1981         case -NFS4ERR_MINOR_VERS_MISMATCH:
1982                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1983                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1984                 dprintk("%s: exit with error %d for server %s\n",
1985                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1986                 return -EPROTONOSUPPORT;
1987         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1988                                  * in nfs4_exchange_id */
1989         default:
1990                 dprintk("%s: exit with error %d for server %s\n", __func__,
1991                                 status, clp->cl_hostname);
1992                 return status;
1993         }
1994         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1995         dprintk("%s: handled error %d for server %s\n", __func__, status,
1996                         clp->cl_hostname);
1997         return 0;
1998 }
1999
2000 static int nfs4_establish_lease(struct nfs_client *clp)
2001 {
2002         const struct cred *cred;
2003         const struct nfs4_state_recovery_ops *ops =
2004                 clp->cl_mvops->reboot_recovery_ops;
2005         int status;
2006
2007         status = nfs4_begin_drain_session(clp);
2008         if (status != 0)
2009                 return status;
2010         cred = nfs4_get_clid_cred(clp);
2011         if (cred == NULL)
2012                 return -ENOENT;
2013         status = ops->establish_clid(clp, cred);
2014         put_cred(cred);
2015         if (status != 0)
2016                 return status;
2017         pnfs_destroy_all_layouts(clp);
2018         return 0;
2019 }
2020
2021 /*
2022  * Returns zero or a negative errno.  NFS4ERR values are converted
2023  * to local errno values.
2024  */
2025 static int nfs4_reclaim_lease(struct nfs_client *clp)
2026 {
2027         int status;
2028
2029         status = nfs4_establish_lease(clp);
2030         if (status < 0)
2031                 return nfs4_handle_reclaim_lease_error(clp, status);
2032         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2033                 nfs4_state_start_reclaim_nograce(clp);
2034         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2035                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2036         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2037         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2038         return 0;
2039 }
2040
2041 static int nfs4_purge_lease(struct nfs_client *clp)
2042 {
2043         int status;
2044
2045         status = nfs4_establish_lease(clp);
2046         if (status < 0)
2047                 return nfs4_handle_reclaim_lease_error(clp, status);
2048         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2049         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2050         nfs4_state_start_reclaim_nograce(clp);
2051         return 0;
2052 }
2053
2054 /*
2055  * Try remote migration of one FSID from a source server to a
2056  * destination server.  The source server provides a list of
2057  * potential destinations.
2058  *
2059  * Returns zero or a negative NFS4ERR status code.
2060  */
2061 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2062 {
2063         struct nfs_client *clp = server->nfs_client;
2064         struct nfs4_fs_locations *locations = NULL;
2065         struct inode *inode;
2066         struct page *page;
2067         int status, result;
2068
2069         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2070                         (unsigned long long)server->fsid.major,
2071                         (unsigned long long)server->fsid.minor,
2072                         clp->cl_hostname);
2073
2074         result = 0;
2075         page = alloc_page(GFP_KERNEL);
2076         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2077         if (page == NULL || locations == NULL) {
2078                 dprintk("<-- %s: no memory\n", __func__);
2079                 goto out;
2080         }
2081
2082         inode = d_inode(server->super->s_root);
2083         result = nfs4_proc_get_locations(inode, locations, page, cred);
2084         if (result) {
2085                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2086                         __func__, result);
2087                 goto out;
2088         }
2089
2090         result = -NFS4ERR_NXIO;
2091         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2092                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2093                         __func__);
2094                 goto out;
2095         }
2096
2097         status = nfs4_begin_drain_session(clp);
2098         if (status != 0)
2099                 return status;
2100
2101         status = nfs4_replace_transport(server, locations);
2102         if (status != 0) {
2103                 dprintk("<-- %s: failed to replace transport: %d\n",
2104                         __func__, status);
2105                 goto out;
2106         }
2107
2108         result = 0;
2109         dprintk("<-- %s: migration succeeded\n", __func__);
2110
2111 out:
2112         if (page != NULL)
2113                 __free_page(page);
2114         kfree(locations);
2115         if (result) {
2116                 pr_err("NFS: migration recovery failed (server %s)\n",
2117                                 clp->cl_hostname);
2118                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2119         }
2120         return result;
2121 }
2122
2123 /*
2124  * Returns zero or a negative NFS4ERR status code.
2125  */
2126 static int nfs4_handle_migration(struct nfs_client *clp)
2127 {
2128         const struct nfs4_state_maintenance_ops *ops =
2129                                 clp->cl_mvops->state_renewal_ops;
2130         struct nfs_server *server;
2131         const struct cred *cred;
2132
2133         dprintk("%s: migration reported on \"%s\"\n", __func__,
2134                         clp->cl_hostname);
2135
2136         cred = ops->get_state_renewal_cred(clp);
2137         if (cred == NULL)
2138                 return -NFS4ERR_NOENT;
2139
2140         clp->cl_mig_gen++;
2141 restart:
2142         rcu_read_lock();
2143         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2144                 int status;
2145
2146                 if (server->mig_gen == clp->cl_mig_gen)
2147                         continue;
2148                 server->mig_gen = clp->cl_mig_gen;
2149
2150                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2151                                                 &server->mig_status))
2152                         continue;
2153
2154                 rcu_read_unlock();
2155                 status = nfs4_try_migration(server, cred);
2156                 if (status < 0) {
2157                         put_cred(cred);
2158                         return status;
2159                 }
2160                 goto restart;
2161         }
2162         rcu_read_unlock();
2163         put_cred(cred);
2164         return 0;
2165 }
2166
2167 /*
2168  * Test each nfs_server on the clp's cl_superblocks list to see
2169  * if it's moved to another server.  Stop when the server no longer
2170  * returns NFS4ERR_LEASE_MOVED.
2171  */
2172 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2173 {
2174         const struct nfs4_state_maintenance_ops *ops =
2175                                 clp->cl_mvops->state_renewal_ops;
2176         struct nfs_server *server;
2177         const struct cred *cred;
2178
2179         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2180                         clp->cl_hostname);
2181
2182         cred = ops->get_state_renewal_cred(clp);
2183         if (cred == NULL)
2184                 return -NFS4ERR_NOENT;
2185
2186         clp->cl_mig_gen++;
2187 restart:
2188         rcu_read_lock();
2189         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2190                 struct inode *inode;
2191                 int status;
2192
2193                 if (server->mig_gen == clp->cl_mig_gen)
2194                         continue;
2195                 server->mig_gen = clp->cl_mig_gen;
2196
2197                 rcu_read_unlock();
2198
2199                 inode = d_inode(server->super->s_root);
2200                 status = nfs4_proc_fsid_present(inode, cred);
2201                 if (status != -NFS4ERR_MOVED)
2202                         goto restart;   /* wasn't this one */
2203                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2204                         goto restart;   /* there are more */
2205                 goto out;
2206         }
2207         rcu_read_unlock();
2208
2209 out:
2210         put_cred(cred);
2211         return 0;
2212 }
2213
2214 /**
2215  * nfs4_discover_server_trunking - Detect server IP address trunking
2216  *
2217  * @clp: nfs_client under test
2218  * @result: OUT: found nfs_client, or clp
2219  *
2220  * Returns zero or a negative errno.  If zero is returned,
2221  * an nfs_client pointer is planted in "result".
2222  *
2223  * Note: since we are invoked in process context, and
2224  * not from inside the state manager, we cannot use
2225  * nfs4_handle_reclaim_lease_error().
2226  */
2227 int nfs4_discover_server_trunking(struct nfs_client *clp,
2228                                   struct nfs_client **result)
2229 {
2230         const struct nfs4_state_recovery_ops *ops =
2231                                 clp->cl_mvops->reboot_recovery_ops;
2232         struct rpc_clnt *clnt;
2233         const struct cred *cred;
2234         int i, status;
2235
2236         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2237
2238         clnt = clp->cl_rpcclient;
2239         i = 0;
2240
2241         mutex_lock(&nfs_clid_init_mutex);
2242 again:
2243         status  = -ENOENT;
2244         cred = nfs4_get_clid_cred(clp);
2245         if (cred == NULL)
2246                 goto out_unlock;
2247
2248         status = ops->detect_trunking(clp, result, cred);
2249         put_cred(cred);
2250         switch (status) {
2251         case 0:
2252         case -EINTR:
2253         case -ERESTARTSYS:
2254                 break;
2255         case -ETIMEDOUT:
2256                 if (clnt->cl_softrtry)
2257                         break;
2258                 /* Fall through */
2259         case -NFS4ERR_DELAY:
2260         case -EAGAIN:
2261                 ssleep(1);
2262                 /* Fall through */
2263         case -NFS4ERR_STALE_CLIENTID:
2264                 dprintk("NFS: %s after status %d, retrying\n",
2265                         __func__, status);
2266                 goto again;
2267         case -EACCES:
2268                 if (i++ == 0) {
2269                         nfs4_root_machine_cred(clp);
2270                         goto again;
2271                 }
2272                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2273                         break;
2274                 /* Fall through */
2275         case -NFS4ERR_CLID_INUSE:
2276         case -NFS4ERR_WRONGSEC:
2277                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2278                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2279                         status = -EPERM;
2280                         break;
2281                 }
2282                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2283                 if (IS_ERR(clnt)) {
2284                         status = PTR_ERR(clnt);
2285                         break;
2286                 }
2287                 /* Note: this is safe because we haven't yet marked the
2288                  * client as ready, so we are the only user of
2289                  * clp->cl_rpcclient
2290                  */
2291                 clnt = xchg(&clp->cl_rpcclient, clnt);
2292                 rpc_shutdown_client(clnt);
2293                 clnt = clp->cl_rpcclient;
2294                 goto again;
2295
2296         case -NFS4ERR_MINOR_VERS_MISMATCH:
2297                 status = -EPROTONOSUPPORT;
2298                 break;
2299
2300         case -EKEYEXPIRED:
2301         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2302                                  * in nfs4_exchange_id */
2303                 status = -EKEYEXPIRED;
2304                 break;
2305         default:
2306                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2307                                 __func__, status);
2308                 status = -EIO;
2309         }
2310
2311 out_unlock:
2312         mutex_unlock(&nfs_clid_init_mutex);
2313         dprintk("NFS: %s: status = %d\n", __func__, status);
2314         return status;
2315 }
2316
2317 #ifdef CONFIG_NFS_V4_1
2318 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2319 {
2320         struct nfs_client *clp = session->clp;
2321
2322         switch (err) {
2323         default:
2324                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2325                 break;
2326         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2327                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2328         }
2329         nfs4_schedule_state_manager(clp);
2330 }
2331 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2332
2333 void nfs41_notify_server(struct nfs_client *clp)
2334 {
2335         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2336         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2337         nfs4_schedule_state_manager(clp);
2338 }
2339
2340 static void nfs4_reset_all_state(struct nfs_client *clp)
2341 {
2342         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2343                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2344                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2345                 nfs4_state_start_reclaim_nograce(clp);
2346                 dprintk("%s: scheduling reset of all state for server %s!\n",
2347                                 __func__, clp->cl_hostname);
2348                 nfs4_schedule_state_manager(clp);
2349         }
2350 }
2351
2352 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2353 {
2354         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2355                 nfs4_state_start_reclaim_reboot(clp);
2356                 dprintk("%s: server %s rebooted!\n", __func__,
2357                                 clp->cl_hostname);
2358                 nfs4_schedule_state_manager(clp);
2359         }
2360 }
2361
2362 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2363 {
2364         nfs4_reset_all_state(clp);
2365         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2366 }
2367
2368 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2369 {
2370         nfs4_state_start_reclaim_nograce(clp);
2371         nfs4_schedule_state_manager(clp);
2372
2373         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2374 }
2375
2376 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2377 {
2378         /* FIXME: For now, we destroy all layouts. */
2379         pnfs_destroy_all_layouts(clp);
2380         nfs_test_expired_all_delegations(clp);
2381         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2382                         clp->cl_hostname);
2383 }
2384
2385 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2386 {
2387         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2388         nfs4_schedule_state_manager(clp);
2389
2390         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2391                         clp->cl_hostname);
2392 }
2393
2394 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2395 {
2396         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2397                 &clp->cl_state) == 0)
2398                 nfs4_schedule_state_manager(clp);
2399 }
2400
2401 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2402                 bool recovery)
2403 {
2404         if (!flags)
2405                 return;
2406
2407         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2408                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2409         /*
2410          * If we're called from the state manager thread, then assume we're
2411          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2412          * Those flags are expected to remain set until we're done
2413          * recovering (see RFC5661, section 18.46.3).
2414          */
2415         if (recovery)
2416                 goto out_recovery;
2417
2418         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2419                 nfs41_handle_server_reboot(clp);
2420         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2421                 nfs41_handle_all_state_revoked(clp);
2422         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2423                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2424                 nfs41_handle_some_state_revoked(clp);
2425         if (flags & SEQ4_STATUS_LEASE_MOVED)
2426                 nfs4_schedule_lease_moved_recovery(clp);
2427         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2428                 nfs41_handle_recallable_state_revoked(clp);
2429 out_recovery:
2430         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2431                 nfs41_handle_backchannel_fault(clp);
2432         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2433                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2434                 nfs41_handle_cb_path_down(clp);
2435 }
2436
2437 static int nfs4_reset_session(struct nfs_client *clp)
2438 {
2439         const struct cred *cred;
2440         int status;
2441
2442         if (!nfs4_has_session(clp))
2443                 return 0;
2444         status = nfs4_begin_drain_session(clp);
2445         if (status != 0)
2446                 return status;
2447         cred = nfs4_get_clid_cred(clp);
2448         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2449         switch (status) {
2450         case 0:
2451         case -NFS4ERR_BADSESSION:
2452         case -NFS4ERR_DEADSESSION:
2453                 break;
2454         case -NFS4ERR_BACK_CHAN_BUSY:
2455         case -NFS4ERR_DELAY:
2456                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2457                 status = 0;
2458                 ssleep(1);
2459                 goto out;
2460         default:
2461                 status = nfs4_recovery_handle_error(clp, status);
2462                 goto out;
2463         }
2464
2465         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2466         status = nfs4_proc_create_session(clp, cred);
2467         if (status) {
2468                 dprintk("%s: session reset failed with status %d for server %s!\n",
2469                         __func__, status, clp->cl_hostname);
2470                 status = nfs4_handle_reclaim_lease_error(clp, status);
2471                 goto out;
2472         }
2473         nfs41_finish_session_reset(clp);
2474         dprintk("%s: session reset was successful for server %s!\n",
2475                         __func__, clp->cl_hostname);
2476 out:
2477         put_cred(cred);
2478         return status;
2479 }
2480
2481 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2482 {
2483         const struct cred *cred;
2484         int ret;
2485
2486         if (!nfs4_has_session(clp))
2487                 return 0;
2488         ret = nfs4_begin_drain_session(clp);
2489         if (ret != 0)
2490                 return ret;
2491         cred = nfs4_get_clid_cred(clp);
2492         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2493         put_cred(cred);
2494         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2495         switch (ret) {
2496         case 0:
2497                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2498                         __func__, clp->cl_hostname);
2499                 break;
2500         case -NFS4ERR_DELAY:
2501                 ssleep(1);
2502                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2503                 break;
2504         default:
2505                 return nfs4_recovery_handle_error(clp, ret);
2506         }
2507         return 0;
2508 }
2509 #else /* CONFIG_NFS_V4_1 */
2510 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2511
2512 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2513 {
2514         return 0;
2515 }
2516 #endif /* CONFIG_NFS_V4_1 */
2517
2518 static void nfs4_state_manager(struct nfs_client *clp)
2519 {
2520         int status = 0;
2521         const char *section = "", *section_sep = "";
2522
2523         /* Ensure exclusive access to NFSv4 state */
2524         do {
2525                 clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2526                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2527                         section = "purge state";
2528                         status = nfs4_purge_lease(clp);
2529                         if (status < 0)
2530                                 goto out_error;
2531                         continue;
2532                 }
2533
2534                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2535                         section = "lease expired";
2536                         /* We're going to have to re-establish a clientid */
2537                         status = nfs4_reclaim_lease(clp);
2538                         if (status < 0)
2539                                 goto out_error;
2540                         continue;
2541                 }
2542
2543                 /* Initialize or reset the session */
2544                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2545                         section = "reset session";
2546                         status = nfs4_reset_session(clp);
2547                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2548                                 continue;
2549                         if (status < 0)
2550                                 goto out_error;
2551                 }
2552
2553                 /* Send BIND_CONN_TO_SESSION */
2554                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2555                                 &clp->cl_state)) {
2556                         section = "bind conn to session";
2557                         status = nfs4_bind_conn_to_session(clp);
2558                         if (status < 0)
2559                                 goto out_error;
2560                         continue;
2561                 }
2562
2563                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2564                         section = "check lease";
2565                         status = nfs4_check_lease(clp);
2566                         if (status < 0)
2567                                 goto out_error;
2568                         continue;
2569                 }
2570
2571                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2572                         section = "migration";
2573                         status = nfs4_handle_migration(clp);
2574                         if (status < 0)
2575                                 goto out_error;
2576                 }
2577
2578                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2579                         section = "lease moved";
2580                         status = nfs4_handle_lease_moved(clp);
2581                         if (status < 0)
2582                                 goto out_error;
2583                 }
2584
2585                 /* First recover reboot state... */
2586                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2587                         section = "reclaim reboot";
2588                         status = nfs4_do_reclaim(clp,
2589                                 clp->cl_mvops->reboot_recovery_ops);
2590                         if (status == -EAGAIN)
2591                                 continue;
2592                         if (status < 0)
2593                                 goto out_error;
2594                         nfs4_state_end_reclaim_reboot(clp);
2595                 }
2596
2597                 /* Detect expired delegations... */
2598                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2599                         section = "detect expired delegations";
2600                         nfs_reap_expired_delegations(clp);
2601                         continue;
2602                 }
2603
2604                 /* Now recover expired state... */
2605                 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2606                         section = "reclaim nograce";
2607                         status = nfs4_do_reclaim(clp,
2608                                 clp->cl_mvops->nograce_recovery_ops);
2609                         if (status == -EAGAIN)
2610                                 continue;
2611                         if (status < 0)
2612                                 goto out_error;
2613                         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2614                 }
2615
2616                 nfs4_end_drain_session(clp);
2617                 nfs4_clear_state_manager_bit(clp);
2618
2619                 if (!test_and_set_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state)) {
2620                         if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2621                                 nfs_client_return_marked_delegations(clp);
2622                                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2623                         }
2624                         clear_bit(NFS4CLNT_DELEGRETURN_RUNNING, &clp->cl_state);
2625                 }
2626
2627                 /* Did we race with an attempt to give us more work? */
2628                 if (!test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state))
2629                         return;
2630                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2631                         return;
2632         } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2633         goto out_drain;
2634
2635 out_error:
2636         if (strlen(section))
2637                 section_sep = ": ";
2638         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2639                         " with error %d\n", section_sep, section,
2640                         clp->cl_hostname, -status);
2641         ssleep(1);
2642 out_drain:
2643         nfs4_end_drain_session(clp);
2644         nfs4_clear_state_manager_bit(clp);
2645 }
2646
2647 static int nfs4_run_state_manager(void *ptr)
2648 {
2649         struct nfs_client *clp = ptr;
2650
2651         allow_signal(SIGKILL);
2652         nfs4_state_manager(clp);
2653         nfs_put_client(clp);
2654         module_put_and_exit(0);
2655         return 0;
2656 }
2657
2658 /*
2659  * Local variables:
2660  *  c-basic-offset: 8
2661  * End:
2662  */