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 #include <linux/sched/mm.h>
  53
  54 #include <linux/sunrpc/clnt.h>
  55
  56 #include "nfs4_fs.h"
  57 #include "callback.h"
  58 #include "delegation.h"
  59 #include "internal.h"
  60 #include "nfs4idmap.h"
  61 #include "nfs4session.h"
  62 #include "pnfs.h"
  63 #include "netns.h"
  64 #include "nfs4trace.h"
  65
  66 #define NFSDBG_FACILITY         NFSDBG_STATE
  67
  68 #define OPENOWNER_POOL_SIZE     8
  69
  70 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp);
  71
  72 const nfs4_stateid zero_stateid = {
  73         { .data = { 0 } },
  74         .type = NFS4_SPECIAL_STATEID_TYPE,
  75 };
  76 const nfs4_stateid invalid_stateid = {
  77         {
  78                 /* Funky initialiser keeps older gcc versions happy */
  79                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
  80         },
  81         .type = NFS4_INVALID_STATEID_TYPE,
  82 };
  83
  84 const nfs4_stateid current_stateid = {
  85         {
  86                 /* Funky initialiser keeps older gcc versions happy */
  87                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
  88         },
  89         .type = NFS4_SPECIAL_STATEID_TYPE,
  90 };
  91
  92 static DEFINE_MUTEX(nfs_clid_init_mutex);
  93
  94 static int nfs4_setup_state_renewal(struct nfs_client *clp)
  95 {
  96         int status;
  97         struct nfs_fsinfo fsinfo;
  98
  99         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 100                 nfs4_schedule_state_renewal(clp);
 101                 return 0;
 102         }
 103
 104         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 105         if (status == 0) {
 106                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
 107                 nfs4_schedule_state_renewal(clp);
 108         }
 109
 110         return status;
 111 }
 112
 113 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
 114 {
 115         struct nfs4_setclientid_res clid = {
 116                 .clientid = clp->cl_clientid,
 117                 .confirm = clp->cl_confirm,
 118         };
 119         unsigned short port;
 120         int status;
 121         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 122
 123         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 124                 goto do_confirm;
 125         port = nn->nfs_callback_tcpport;
 126         if (clp->cl_addr.ss_family == AF_INET6)
 127                 port = nn->nfs_callback_tcpport6;
 128
 129         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 130         if (status != 0)
 131                 goto out;
 132         clp->cl_clientid = clid.clientid;
 133         clp->cl_confirm = clid.confirm;
 134         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 135 do_confirm:
 136         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 137         if (status != 0)
 138                 goto out;
 139         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 140         nfs4_setup_state_renewal(clp);
 141 out:
 142         return status;
 143 }
 144
 145 /**
 146  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 147  *
 148  * @clp: nfs_client under test
 149  * @result: OUT: found nfs_client, or clp
 150  * @cred: credential to use for trunking test
 151  *
 152  * Returns zero, a negative errno, or a negative NFS4ERR status.
 153  * If zero is returned, an nfs_client pointer is planted in
 154  * "result".
 155  *
 156  * Note: The returned client may not yet be marked ready.
 157  */
 158 int nfs40_discover_server_trunking(struct nfs_client *clp,
 159                                    struct nfs_client **result,
 160                                    const struct cred *cred)
 161 {
 162         struct nfs4_setclientid_res clid = {
 163                 .clientid = clp->cl_clientid,
 164                 .confirm = clp->cl_confirm,
 165         };
 166         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 167         unsigned short port;
 168         int status;
 169
 170         port = nn->nfs_callback_tcpport;
 171         if (clp->cl_addr.ss_family == AF_INET6)
 172                 port = nn->nfs_callback_tcpport6;
 173
 174         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 175         if (status != 0)
 176                 goto out;
 177         clp->cl_clientid = clid.clientid;
 178         clp->cl_confirm = clid.confirm;
 179
 180         status = nfs40_walk_client_list(clp, result, cred);
 181         if (status == 0) {
 182                 /* Sustain the lease, even if it's empty.  If the clientid4
 183                  * goes stale it's of no use for trunking discovery. */
 184                 nfs4_schedule_state_renewal(*result);
 185
 186                 /* If the client state need to recover, do it. */
 187                 if (clp->cl_state)
 188                         nfs4_schedule_state_manager(clp);
 189         }
 190 out:
 191         return status;
 192 }
 193
 194 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
 195 {
 196         return get_cred(rpc_machine_cred());
 197 }
 198
 199 static void nfs4_root_machine_cred(struct nfs_client *clp)
 200 {
 201
 202         /* Force root creds instead of machine */
 203         clp->cl_principal = NULL;
 204         clp->cl_rpcclient->cl_principal = NULL;
 205 }
 206
 207 static const struct cred *
 208 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 209 {
 210         const struct cred *cred = NULL;
 211         struct nfs4_state_owner *sp;
 212         struct rb_node *pos;
 213
 214         for (pos = rb_first(&server->state_owners);
 215              pos != NULL;
 216              pos = rb_next(pos)) {
 217                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 218                 if (list_empty(&sp->so_states))
 219                         continue;
 220                 cred = get_cred(sp->so_cred);
 221                 break;
 222         }
 223         return cred;
 224 }
 225
 226 /**
 227  * nfs4_get_renew_cred - Acquire credential for a renew operation
 228  * @clp: client state handle
 229  *
 230  * Returns an rpc_cred with reference count bumped, or NULL.
 231  * Caller must hold clp->cl_lock.
 232  */
 233 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
 234 {
 235         const struct cred *cred = NULL;
 236         struct nfs_server *server;
 237
 238         /* Use machine credentials if available */
 239         cred = nfs4_get_machine_cred(clp);
 240         if (cred != NULL)
 241                 goto out;
 242
 243         spin_lock(&clp->cl_lock);
 244         rcu_read_lock();
 245         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 246                 cred = nfs4_get_renew_cred_server_locked(server);
 247                 if (cred != NULL)
 248                         break;
 249         }
 250         rcu_read_unlock();
 251         spin_unlock(&clp->cl_lock);
 252
 253 out:
 254         return cred;
 255 }
 256
 257 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 258 {
 259         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 260                 spin_lock(&tbl->slot_tbl_lock);
 261                 nfs41_wake_slot_table(tbl);
 262                 spin_unlock(&tbl->slot_tbl_lock);
 263         }
 264 }
 265
 266 static void nfs4_end_drain_session(struct nfs_client *clp)
 267 {
 268         struct nfs4_session *ses = clp->cl_session;
 269
 270         if (clp->cl_slot_tbl) {
 271                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 272                 return;
 273         }
 274
 275         if (ses != NULL) {
 276                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 277                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 278         }
 279 }
 280
 281 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 282 {
 283         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 284         spin_lock(&tbl->slot_tbl_lock);
 285         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 286                 reinit_completion(&tbl->complete);
 287                 spin_unlock(&tbl->slot_tbl_lock);
 288                 return wait_for_completion_interruptible(&tbl->complete);
 289         }
 290         spin_unlock(&tbl->slot_tbl_lock);
 291         return 0;
 292 }
 293
 294 static int nfs4_begin_drain_session(struct nfs_client *clp)
 295 {
 296         struct nfs4_session *ses = clp->cl_session;
 297         int ret;
 298
 299         if (clp->cl_slot_tbl)
 300                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 301
 302         /* back channel */
 303         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 304         if (ret)
 305                 return ret;
 306         /* fore channel */
 307         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 308 }
 309
 310 #if defined(CONFIG_NFS_V4_1)
 311
 312 static void nfs41_finish_session_reset(struct nfs_client *clp)
 313 {
 314         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 315         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 316         /* create_session negotiated new slot table */
 317         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 318         nfs4_setup_state_renewal(clp);
 319 }
 320
 321 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
 322 {
 323         int status;
 324
 325         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 326                 goto do_confirm;
 327         status = nfs4_proc_exchange_id(clp, cred);
 328         if (status != 0)
 329                 goto out;
 330         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 331 do_confirm:
 332         status = nfs4_proc_create_session(clp, cred);
 333         if (status != 0)
 334                 goto out;
 335         if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R))
 336                 nfs4_state_start_reclaim_reboot(clp);
 337         nfs41_finish_session_reset(clp);
 338         nfs_mark_client_ready(clp, NFS_CS_READY);
 339 out:
 340         return status;
 341 }
 342
 343 /**
 344  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 345  *
 346  * @clp: nfs_client under test
 347  * @result: OUT: found nfs_client, or clp
 348  * @cred: credential to use for trunking test
 349  *
 350  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 351  * If NFS4_OK is returned, an nfs_client pointer is planted in
 352  * "result".
 353  *
 354  * Note: The returned client may not yet be marked ready.
 355  */
 356 int nfs41_discover_server_trunking(struct nfs_client *clp,
 357                                    struct nfs_client **result,
 358                                    const struct cred *cred)
 359 {
 360         int status;
 361
 362         status = nfs4_proc_exchange_id(clp, cred);
 363         if (status != NFS4_OK)
 364                 return status;
 365
 366         status = nfs41_walk_client_list(clp, result, cred);
 367         if (status < 0)
 368                 return status;
 369         if (clp != *result)
 370                 return 0;
 371
 372         /*
 373          * Purge state if the client id was established in a prior
 374          * instance and the client id could not have arrived on the
 375          * server via Transparent State Migration.
 376          */
 377         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
 378                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
 379                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 380                 else
 381                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 382         }
 383         nfs4_schedule_state_manager(clp);
 384         status = nfs_wait_client_init_complete(clp);
 385         if (status < 0)
 386                 nfs_put_client(clp);
 387         return status;
 388 }
 389
 390 #endif /* CONFIG_NFS_V4_1 */
 391
 392 /**
 393  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 394  * @clp: client state handle
 395  *
 396  * Returns a cred with reference count bumped, or NULL.
 397  */
 398 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
 399 {
 400         const struct cred *cred;
 401
 402         cred = nfs4_get_machine_cred(clp);
 403         return cred;
 404 }
 405
 406 static struct nfs4_state_owner *
 407 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
 408 {
 409         struct rb_node **p = &server->state_owners.rb_node,
 410                        *parent = NULL;
 411         struct nfs4_state_owner *sp;
 412         int cmp;
 413
 414         while (*p != NULL) {
 415                 parent = *p;
 416                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 417                 cmp = cred_fscmp(cred, sp->so_cred);
 418
 419                 if (cmp < 0)
 420                         p = &parent->rb_left;
 421                 else if (cmp > 0)
 422                         p = &parent->rb_right;
 423                 else {
 424                         if (!list_empty(&sp->so_lru))
 425                                 list_del_init(&sp->so_lru);
 426                         atomic_inc(&sp->so_count);
 427                         return sp;
 428                 }
 429         }
 430         return NULL;
 431 }
 432
 433 static struct nfs4_state_owner *
 434 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 435 {
 436         struct nfs_server *server = new->so_server;
 437         struct rb_node **p = &server->state_owners.rb_node,
 438                        *parent = NULL;
 439         struct nfs4_state_owner *sp;
 440         int cmp;
 441
 442         while (*p != NULL) {
 443                 parent = *p;
 444                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 445                 cmp = cred_fscmp(new->so_cred, sp->so_cred);
 446
 447                 if (cmp < 0)
 448                         p = &parent->rb_left;
 449                 else if (cmp > 0)
 450                         p = &parent->rb_right;
 451                 else {
 452                         if (!list_empty(&sp->so_lru))
 453                                 list_del_init(&sp->so_lru);
 454                         atomic_inc(&sp->so_count);
 455                         return sp;
 456                 }
 457         }
 458         rb_link_node(&new->so_server_node, parent, p);
 459         rb_insert_color(&new->so_server_node, &server->state_owners);
 460         return new;
 461 }
 462
 463 static void
 464 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 465 {
 466         struct nfs_server *server = sp->so_server;
 467
 468         if (!RB_EMPTY_NODE(&sp->so_server_node))
 469                 rb_erase(&sp->so_server_node, &server->state_owners);
 470 }
 471
 472 static void
 473 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 474 {
 475         sc->create_time = ktime_get();
 476         sc->flags = 0;
 477         sc->counter = 0;
 478         spin_lock_init(&sc->lock);
 479         INIT_LIST_HEAD(&sc->list);
 480         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 481 }
 482
 483 static void
 484 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 485 {
 486         rpc_destroy_wait_queue(&sc->wait);
 487 }
 488
 489 /*
 490  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 491  * create a new state_owner.
 492  *
 493  */
 494 static struct nfs4_state_owner *
 495 nfs4_alloc_state_owner(struct nfs_server *server,
 496                 const struct cred *cred,
 497                 gfp_t gfp_flags)
 498 {
 499         struct nfs4_state_owner *sp;
 500
 501         sp = kzalloc(sizeof(*sp), gfp_flags);
 502         if (!sp)
 503                 return NULL;
 504         sp->so_seqid.owner_id = ida_alloc(&server->openowner_id, gfp_flags);
 505         if (sp->so_seqid.owner_id < 0) {
 506                 kfree(sp);
 507                 return NULL;
 508         }
 509         sp->so_server = server;
 510         sp->so_cred = get_cred(cred);
 511         spin_lock_init(&sp->so_lock);
 512         INIT_LIST_HEAD(&sp->so_states);
 513         nfs4_init_seqid_counter(&sp->so_seqid);
 514         atomic_set(&sp->so_count, 1);
 515         INIT_LIST_HEAD(&sp->so_lru);
 516         mutex_init(&sp->so_delegreturn_mutex);
 517         return sp;
 518 }
 519
 520 static void
 521 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 522 {
 523         /* This state_owner is no longer usable, but must
 524          * remain in place so that state recovery can find it
 525          * and the opens associated with it.
 526          * It may also be used for new 'open' request to
 527          * return a delegation to the server.
 528          * So update the 'create_time' so that it looks like
 529          * a new state_owner.  This will cause the server to
 530          * request an OPEN_CONFIRM to start a new sequence.
 531          */
 532         sp->so_seqid.create_time = ktime_get();
 533 }
 534
 535 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 536 {
 537         nfs4_destroy_seqid_counter(&sp->so_seqid);
 538         put_cred(sp->so_cred);
 539         ida_free(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
 540         kfree(sp);
 541 }
 542
 543 static void nfs4_gc_state_owners(struct nfs_server *server)
 544 {
 545         struct nfs_client *clp = server->nfs_client;
 546         struct nfs4_state_owner *sp, *tmp;
 547         unsigned long time_min, time_max;
 548         LIST_HEAD(doomed);
 549
 550         spin_lock(&clp->cl_lock);
 551         time_max = jiffies;
 552         time_min = (long)time_max - (long)clp->cl_lease_time;
 553         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 554                 /* NB: LRU is sorted so that oldest is at the head */
 555                 if (time_in_range(sp->so_expires, time_min, time_max))
 556                         break;
 557                 list_move(&sp->so_lru, &doomed);
 558                 nfs4_remove_state_owner_locked(sp);
 559         }
 560         spin_unlock(&clp->cl_lock);
 561
 562         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 563                 list_del(&sp->so_lru);
 564                 nfs4_free_state_owner(sp);
 565         }
 566 }
 567
 568 /**
 569  * nfs4_get_state_owner - Look up a state owner given a credential
 570  * @server: nfs_server to search
 571  * @cred: RPC credential to match
 572  * @gfp_flags: allocation mode
 573  *
 574  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 575  */
 576 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 577                                               const struct cred *cred,
 578                                               gfp_t gfp_flags)
 579 {
 580         struct nfs_client *clp = server->nfs_client;
 581         struct nfs4_state_owner *sp, *new;
 582
 583         spin_lock(&clp->cl_lock);
 584         sp = nfs4_find_state_owner_locked(server, cred);
 585         spin_unlock(&clp->cl_lock);
 586         if (sp != NULL)
 587                 goto out;
 588         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 589         if (new == NULL)
 590                 goto out;
 591         spin_lock(&clp->cl_lock);
 592         sp = nfs4_insert_state_owner_locked(new);
 593         spin_unlock(&clp->cl_lock);
 594         if (sp != new)
 595                 nfs4_free_state_owner(new);
 596 out:
 597         nfs4_gc_state_owners(server);
 598         return sp;
 599 }
 600
 601 /**
 602  * nfs4_put_state_owner - Release a nfs4_state_owner
 603  * @sp: state owner data to release
 604  *
 605  * Note that we keep released state owners on an LRU
 606  * list.
 607  * This caches valid state owners so that they can be
 608  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 609  * It also pins the uniquifier of dropped state owners for
 610  * a while, to ensure that those state owner names are
 611  * never reused.
 612  */
 613 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 614 {
 615         struct nfs_server *server = sp->so_server;
 616         struct nfs_client *clp = server->nfs_client;
 617
 618         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 619                 return;
 620
 621         sp->so_expires = jiffies;
 622         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 623         spin_unlock(&clp->cl_lock);
 624 }
 625
 626 /**
 627  * nfs4_purge_state_owners - Release all cached state owners
 628  * @server: nfs_server with cached state owners to release
 629  * @head: resulting list of state owners
 630  *
 631  * Called at umount time.  Remaining state owners will be on
 632  * the LRU with ref count of zero.
 633  * Note that the state owners are not freed, but are added
 634  * to the list @head, which can later be used as an argument
 635  * to nfs4_free_state_owners.
 636  */
 637 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 638 {
 639         struct nfs_client *clp = server->nfs_client;
 640         struct nfs4_state_owner *sp, *tmp;
 641
 642         spin_lock(&clp->cl_lock);
 643         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 644                 list_move(&sp->so_lru, head);
 645                 nfs4_remove_state_owner_locked(sp);
 646         }
 647         spin_unlock(&clp->cl_lock);
 648 }
 649
 650 /**
 651  * nfs4_free_state_owners - Release all cached state owners
 652  * @head: resulting list of state owners
 653  *
 654  * Frees a list of state owners that was generated by
 655  * nfs4_purge_state_owners
 656  */
 657 void nfs4_free_state_owners(struct list_head *head)
 658 {
 659         struct nfs4_state_owner *sp, *tmp;
 660
 661         list_for_each_entry_safe(sp, tmp, head, so_lru) {
 662                 list_del(&sp->so_lru);
 663                 nfs4_free_state_owner(sp);
 664         }
 665 }
 666
 667 static struct nfs4_state *
 668 nfs4_alloc_open_state(void)
 669 {
 670         struct nfs4_state *state;
 671
 672         state = kzalloc(sizeof(*state), GFP_KERNEL_ACCOUNT);
 673         if (!state)
 674                 return NULL;
 675         refcount_set(&state->count, 1);
 676         INIT_LIST_HEAD(&state->lock_states);
 677         spin_lock_init(&state->state_lock);
 678         seqlock_init(&state->seqlock);
 679         init_waitqueue_head(&state->waitq);
 680         return state;
 681 }
 682
 683 void
 684 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 685 {
 686         if (state->state == fmode)
 687                 return;
 688         /* NB! List reordering - see the reclaim code for why.  */
 689         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 690                 if (fmode & FMODE_WRITE)
 691                         list_move(&state->open_states, &state->owner->so_states);
 692                 else
 693                         list_move_tail(&state->open_states, &state->owner->so_states);
 694         }
 695         state->state = fmode;
 696 }
 697
 698 static struct nfs4_state *
 699 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 700 {
 701         struct nfs_inode *nfsi = NFS_I(inode);
 702         struct nfs4_state *state;
 703
 704         list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
 705                 if (state->owner != owner)
 706                         continue;
 707                 if (!nfs4_valid_open_stateid(state))
 708                         continue;
 709                 if (refcount_inc_not_zero(&state->count))
 710                         return state;
 711         }
 712         return NULL;
 713 }
 714
 715 static void
 716 nfs4_free_open_state(struct nfs4_state *state)
 717 {
 718         kfree_rcu(state, rcu_head);
 719 }
 720
 721 struct nfs4_state *
 722 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 723 {
 724         struct nfs4_state *state, *new;
 725         struct nfs_inode *nfsi = NFS_I(inode);
 726
 727         rcu_read_lock();
 728         state = __nfs4_find_state_byowner(inode, owner);
 729         rcu_read_unlock();
 730         if (state)
 731                 goto out;
 732         new = nfs4_alloc_open_state();
 733         spin_lock(&owner->so_lock);
 734         spin_lock(&inode->i_lock);
 735         state = __nfs4_find_state_byowner(inode, owner);
 736         if (state == NULL && new != NULL) {
 737                 state = new;
 738                 state->owner = owner;
 739                 atomic_inc(&owner->so_count);
 740                 ihold(inode);
 741                 state->inode = inode;
 742                 list_add_rcu(&state->inode_states, &nfsi->open_states);
 743                 spin_unlock(&inode->i_lock);
 744                 /* Note: The reclaim code dictates that we add stateless
 745                  * and read-only stateids to the end of the list */
 746                 list_add_tail(&state->open_states, &owner->so_states);
 747                 spin_unlock(&owner->so_lock);
 748         } else {
 749                 spin_unlock(&inode->i_lock);
 750                 spin_unlock(&owner->so_lock);
 751                 if (new)
 752                         nfs4_free_open_state(new);
 753         }
 754 out:
 755         return state;
 756 }
 757
 758 void nfs4_put_open_state(struct nfs4_state *state)
 759 {
 760         struct inode *inode = state->inode;
 761         struct nfs4_state_owner *owner = state->owner;
 762
 763         if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
 764                 return;
 765         spin_lock(&inode->i_lock);
 766         list_del_rcu(&state->inode_states);
 767         list_del(&state->open_states);
 768         spin_unlock(&inode->i_lock);
 769         spin_unlock(&owner->so_lock);
 770         nfs4_inode_return_delegation_on_close(inode);
 771         iput(inode);
 772         nfs4_free_open_state(state);
 773         nfs4_put_state_owner(owner);
 774 }
 775
 776 /*
 777  * Close the current file.
 778  */
 779 static void __nfs4_close(struct nfs4_state *state,
 780                 fmode_t fmode, gfp_t gfp_mask, int wait)
 781 {
 782         struct nfs4_state_owner *owner = state->owner;
 783         int call_close = 0;
 784         fmode_t newstate;
 785
 786         atomic_inc(&owner->so_count);
 787         /* Protect against nfs4_find_state() */
 788         spin_lock(&owner->so_lock);
 789         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 790                 case FMODE_READ:
 791                         state->n_rdonly--;
 792                         break;
 793                 case FMODE_WRITE:
 794                         state->n_wronly--;
 795                         break;
 796                 case FMODE_READ|FMODE_WRITE:
 797                         state->n_rdwr--;
 798         }
 799         newstate = FMODE_READ|FMODE_WRITE;
 800         if (state->n_rdwr == 0) {
 801                 if (state->n_rdonly == 0) {
 802                         newstate &= ~FMODE_READ;
 803                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 804                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 805                 }
 806                 if (state->n_wronly == 0) {
 807                         newstate &= ~FMODE_WRITE;
 808                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 809                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 810                 }
 811                 if (newstate == 0)
 812                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 813         }
 814         nfs4_state_set_mode_locked(state, newstate);
 815         spin_unlock(&owner->so_lock);
 816
 817         if (!call_close) {
 818                 nfs4_put_open_state(state);
 819                 nfs4_put_state_owner(owner);
 820         } else
 821                 nfs4_do_close(state, gfp_mask, wait);
 822 }
 823
 824 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 825 {
 826         __nfs4_close(state, fmode, GFP_KERNEL, 0);
 827 }
 828
 829 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 830 {
 831         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 832 }
 833
 834 /*
 835  * Search the state->lock_states for an existing lock_owner
 836  * that is compatible with either of the given owners.
 837  * If the second is non-zero, then the first refers to a Posix-lock
 838  * owner (current->files) and the second refers to a flock/OFD
 839  * owner (struct file*).  In that case, prefer a match for the first
 840  * owner.
 841  * If both sorts of locks are held on the one file we cannot know
 842  * which stateid was intended to be used, so a "correct" choice cannot
 843  * be made.  Failing that, a "consistent" choice is preferable.  The
 844  * consistent choice we make is to prefer the first owner, that of a
 845  * Posix lock.
 846  */
 847 static struct nfs4_lock_state *
 848 __nfs4_find_lock_state(struct nfs4_state *state,
 849                        fl_owner_t owner, fl_owner_t owner2)
 850 {
 851         struct nfs4_lock_state *pos, *ret = NULL;
 852         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 853                 if (pos->ls_owner == owner) {
 854                         ret = pos;
 855                         break;
 856                 }
 857                 if (pos->ls_owner == owner2)
 858                         ret = pos;
 859         }
 860         if (ret)
 861                 refcount_inc(&ret->ls_count);
 862         return ret;
 863 }
 864
 865 /*
 866  * Return a compatible lock_state. If no initialized lock_state structure
 867  * exists, return an uninitialized one.
 868  *
 869  */
 870 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t owner)
 871 {
 872         struct nfs4_lock_state *lsp;
 873         struct nfs_server *server = state->owner->so_server;
 874
 875         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL_ACCOUNT);
 876         if (lsp == NULL)
 877                 return NULL;
 878         nfs4_init_seqid_counter(&lsp->ls_seqid);
 879         refcount_set(&lsp->ls_count, 1);
 880         lsp->ls_state = state;
 881         lsp->ls_owner = owner;
 882         lsp->ls_seqid.owner_id = ida_alloc(&server->lockowner_id, GFP_KERNEL_ACCOUNT);
 883         if (lsp->ls_seqid.owner_id < 0)
 884                 goto out_free;
 885         INIT_LIST_HEAD(&lsp->ls_locks);
 886         return lsp;
 887 out_free:
 888         kfree(lsp);
 889         return NULL;
 890 }
 891
 892 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 893 {
 894         ida_free(&server->lockowner_id, lsp->ls_seqid.owner_id);
 895         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 896         kfree(lsp);
 897 }
 898
 899 /*
 900  * Return a compatible lock_state. If no initialized lock_state structure
 901  * exists, return an uninitialized one.
 902  *
 903  */
 904 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 905 {
 906         struct nfs4_lock_state *lsp, *new = NULL;
 907
 908         for(;;) {
 909                 spin_lock(&state->state_lock);
 910                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 911                 if (lsp != NULL)
 912                         break;
 913                 if (new != NULL) {
 914                         list_add(&new->ls_locks, &state->lock_states);
 915                         set_bit(LK_STATE_IN_USE, &state->flags);
 916                         lsp = new;
 917                         new = NULL;
 918                         break;
 919                 }
 920                 spin_unlock(&state->state_lock);
 921                 new = nfs4_alloc_lock_state(state, owner);
 922                 if (new == NULL)
 923                         return NULL;
 924         }
 925         spin_unlock(&state->state_lock);
 926         if (new != NULL)
 927                 nfs4_free_lock_state(state->owner->so_server, new);
 928         return lsp;
 929 }
 930
 931 /*
 932  * Release reference to lock_state, and free it if we see that
 933  * it is no longer in use
 934  */
 935 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 936 {
 937         struct nfs_server *server;
 938         struct nfs4_state *state;
 939
 940         if (lsp == NULL)
 941                 return;
 942         state = lsp->ls_state;
 943         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 944                 return;
 945         list_del(&lsp->ls_locks);
 946         if (list_empty(&state->lock_states))
 947                 clear_bit(LK_STATE_IN_USE, &state->flags);
 948         spin_unlock(&state->state_lock);
 949         server = state->owner->so_server;
 950         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 951                 struct nfs_client *clp = server->nfs_client;
 952
 953                 clp->cl_mvops->free_lock_state(server, lsp);
 954         } else
 955                 nfs4_free_lock_state(server, lsp);
 956 }
 957
 958 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 959 {
 960         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 961
 962         dst->fl_u.nfs4_fl.owner = lsp;
 963         refcount_inc(&lsp->ls_count);
 964 }
 965
 966 static void nfs4_fl_release_lock(struct file_lock *fl)
 967 {
 968         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 969 }
 970
 971 static const struct file_lock_operations nfs4_fl_lock_ops = {
 972         .fl_copy_lock = nfs4_fl_copy_lock,
 973         .fl_release_private = nfs4_fl_release_lock,
 974 };
 975
 976 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 977 {
 978         struct nfs4_lock_state *lsp;
 979
 980         if (fl->fl_ops != NULL)
 981                 return 0;
 982         lsp = nfs4_get_lock_state(state, fl->c.flc_owner);
 983         if (lsp == NULL)
 984                 return -ENOMEM;
 985         fl->fl_u.nfs4_fl.owner = lsp;
 986         fl->fl_ops = &nfs4_fl_lock_ops;
 987         return 0;
 988 }
 989
 990 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 991                 struct nfs4_state *state,
 992                 const struct nfs_lock_context *l_ctx)
 993 {
 994         struct nfs4_lock_state *lsp;
 995         fl_owner_t owner, fl_flock_owner;
 996         int ret = -ENOENT;
 997
 998         if (l_ctx == NULL)
 999                 goto out;
1000
1001         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1002                 goto out;
1003
1004         owner = l_ctx->lockowner;
1005         fl_flock_owner = l_ctx->open_context->flock_owner;
1006
1007         spin_lock(&state->state_lock);
1008         lsp = __nfs4_find_lock_state(state, owner, fl_flock_owner);
1009         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1010                 ret = -EIO;
1011         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1012                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1013                 ret = 0;
1014         }
1015         spin_unlock(&state->state_lock);
1016         nfs4_put_lock_state(lsp);
1017 out:
1018         return ret;
1019 }
1020
1021 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1022 {
1023         bool ret;
1024         const nfs4_stateid *src;
1025         int seq;
1026
1027         do {
1028                 ret = false;
1029                 src = &zero_stateid;
1030                 seq = read_seqbegin(&state->seqlock);
1031                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1032                         src = &state->open_stateid;
1033                         ret = true;
1034                 }
1035                 nfs4_stateid_copy(dst, src);
1036         } while (read_seqretry(&state->seqlock, seq));
1037         return ret;
1038 }
1039
1040 /*
1041  * Byte-range lock aware utility to initialize the stateid of read/write
1042  * requests.
1043  */
1044 int nfs4_select_rw_stateid(struct nfs4_state *state,
1045                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1046                 nfs4_stateid *dst, const struct cred **cred)
1047 {
1048         int ret;
1049
1050         if (!nfs4_valid_open_stateid(state))
1051                 return -EIO;
1052         if (cred != NULL)
1053                 *cred = NULL;
1054         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1055         if (ret == -EIO)
1056                 /* A lost lock - don't even consider delegations */
1057                 goto out;
1058         /* returns true if delegation stateid found and copied */
1059         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1060                 ret = 0;
1061                 goto out;
1062         }
1063         if (ret != -ENOENT)
1064                 /* nfs4_copy_delegation_stateid() didn't over-write
1065                  * dst, so it still has the lock stateid which we now
1066                  * choose to use.
1067                  */
1068                 goto out;
1069         ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1070 out:
1071         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1072                 dst->seqid = 0;
1073         return ret;
1074 }
1075
1076 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1077 {
1078         struct nfs_seqid *new;
1079
1080         new = kmalloc(sizeof(*new), gfp_mask);
1081         if (new == NULL)
1082                 return ERR_PTR(-ENOMEM);
1083         new->sequence = counter;
1084         INIT_LIST_HEAD(&new->list);
1085         new->task = NULL;
1086         return new;
1087 }
1088
1089 void nfs_release_seqid(struct nfs_seqid *seqid)
1090 {
1091         struct nfs_seqid_counter *sequence;
1092
1093         if (seqid == NULL || list_empty(&seqid->list))
1094                 return;
1095         sequence = seqid->sequence;
1096         spin_lock(&sequence->lock);
1097         list_del_init(&seqid->list);
1098         if (!list_empty(&sequence->list)) {
1099                 struct nfs_seqid *next;
1100
1101                 next = list_first_entry(&sequence->list,
1102                                 struct nfs_seqid, list);
1103                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1104         }
1105         spin_unlock(&sequence->lock);
1106 }
1107
1108 void nfs_free_seqid(struct nfs_seqid *seqid)
1109 {
1110         nfs_release_seqid(seqid);
1111         kfree(seqid);
1112 }
1113
1114 /*
1115  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1116  * failed with a seqid incrementing error -
1117  * see comments nfs4.h:seqid_mutating_error()
1118  */
1119 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1120 {
1121         switch (status) {
1122                 case 0:
1123                         break;
1124                 case -NFS4ERR_BAD_SEQID:
1125                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1126                                 return;
1127                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1128                                         " sequence-id error on an"
1129                                         " unconfirmed sequence %p!\n",
1130                                         seqid->sequence);
1131                         return;
1132                 case -NFS4ERR_STALE_CLIENTID:
1133                 case -NFS4ERR_STALE_STATEID:
1134                 case -NFS4ERR_BAD_STATEID:
1135                 case -NFS4ERR_BADXDR:
1136                 case -NFS4ERR_RESOURCE:
1137                 case -NFS4ERR_NOFILEHANDLE:
1138                 case -NFS4ERR_MOVED:
1139                         /* Non-seqid mutating errors */
1140                         return;
1141         }
1142         /*
1143          * Note: no locking needed as we are guaranteed to be first
1144          * on the sequence list
1145          */
1146         seqid->sequence->counter++;
1147 }
1148
1149 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1150 {
1151         struct nfs4_state_owner *sp;
1152
1153         if (seqid == NULL)
1154                 return;
1155
1156         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1157         if (status == -NFS4ERR_BAD_SEQID)
1158                 nfs4_reset_state_owner(sp);
1159         if (!nfs4_has_session(sp->so_server->nfs_client))
1160                 nfs_increment_seqid(status, seqid);
1161 }
1162
1163 /*
1164  * Increment the seqid if the LOCK/LOCKU succeeded, or
1165  * failed with a seqid incrementing error -
1166  * see comments nfs4.h:seqid_mutating_error()
1167  */
1168 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1169 {
1170         if (seqid != NULL)
1171                 nfs_increment_seqid(status, seqid);
1172 }
1173
1174 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1175 {
1176         struct nfs_seqid_counter *sequence;
1177         int status = 0;
1178
1179         if (seqid == NULL)
1180                 goto out;
1181         sequence = seqid->sequence;
1182         spin_lock(&sequence->lock);
1183         seqid->task = task;
1184         if (list_empty(&seqid->list))
1185                 list_add_tail(&seqid->list, &sequence->list);
1186         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1187                 goto unlock;
1188         rpc_sleep_on(&sequence->wait, task, NULL);
1189         status = -EAGAIN;
1190 unlock:
1191         spin_unlock(&sequence->lock);
1192 out:
1193         return status;
1194 }
1195
1196 static int nfs4_run_state_manager(void *);
1197
1198 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1199 {
1200         clear_and_wake_up_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1201         rpc_wake_up(&clp->cl_rpcwaitq);
1202 }
1203
1204 /*
1205  * Schedule the nfs_client asynchronous state management routine
1206  */
1207 void nfs4_schedule_state_manager(struct nfs_client *clp)
1208 {
1209         struct task_struct *task;
1210         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1211         struct rpc_clnt *clnt = clp->cl_rpcclient;
1212         bool swapon = false;
1213
1214         if (clnt->cl_shutdown)
1215                 return;
1216
1217         set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1218
1219         if (atomic_read(&clnt->cl_swapper)) {
1220                 swapon = !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE,
1221                                            &clp->cl_state);
1222                 if (!swapon) {
1223                         wake_up_var(&clp->cl_state);
1224                         return;
1225                 }
1226         }
1227
1228         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1229                 return;
1230
1231         __module_get(THIS_MODULE);
1232         refcount_inc(&clp->cl_count);
1233
1234         /* The rcu_read_lock() is not strictly necessary, as the state
1235          * manager is the only thread that ever changes the rpc_xprt
1236          * after it's initialized.  At this point, we're single threaded. */
1237         rcu_read_lock();
1238         snprintf(buf, sizeof(buf), "%s-manager",
1239                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1240         rcu_read_unlock();
1241         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1242         if (IS_ERR(task)) {
1243                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1244                         __func__, PTR_ERR(task));
1245                 if (!nfs_client_init_is_complete(clp))
1246                         nfs_mark_client_ready(clp, PTR_ERR(task));
1247                 if (swapon)
1248                         clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1249                 nfs4_clear_state_manager_bit(clp);
1250                 nfs_put_client(clp);
1251                 module_put(THIS_MODULE);
1252         }
1253 }
1254
1255 /*
1256  * Schedule a lease recovery attempt
1257  */
1258 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1259 {
1260         if (!clp)
1261                 return;
1262         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1263                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1264         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1265                         clp->cl_hostname);
1266         nfs4_schedule_state_manager(clp);
1267 }
1268 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1269
1270 /**
1271  * nfs4_schedule_migration_recovery - trigger migration recovery
1272  *
1273  * @server: FSID that is migrating
1274  *
1275  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1276  * value is returned.
1277  */
1278 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1279 {
1280         struct nfs_client *clp = server->nfs_client;
1281
1282         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1283                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1284                                 clp->cl_hostname);
1285                 return -NFS4ERR_IO;
1286         }
1287
1288         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1289                 return -NFS4ERR_IO;
1290
1291         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1292                         __func__,
1293                         (unsigned long long)server->fsid.major,
1294                         (unsigned long long)server->fsid.minor,
1295                         clp->cl_hostname);
1296
1297         set_bit(NFS_MIG_IN_TRANSITION,
1298                         &((struct nfs_server *)server)->mig_status);
1299         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1300
1301         nfs4_schedule_state_manager(clp);
1302         return 0;
1303 }
1304 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1305
1306 /**
1307  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1308  *
1309  * @clp: server to check for moved leases
1310  *
1311  */
1312 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1313 {
1314         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1315                 __func__, clp->cl_clientid, clp->cl_hostname);
1316
1317         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1318         nfs4_schedule_state_manager(clp);
1319 }
1320 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1321
1322 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1323 {
1324         int res;
1325
1326         might_sleep();
1327
1328         refcount_inc(&clp->cl_count);
1329         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1330                                  nfs_wait_bit_killable,
1331                                  TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1332         if (res)
1333                 goto out;
1334         if (clp->cl_cons_state < 0)
1335                 res = clp->cl_cons_state;
1336 out:
1337         nfs_put_client(clp);
1338         return res;
1339 }
1340
1341 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1342 {
1343         unsigned int loop;
1344         int ret;
1345
1346         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1347                 ret = nfs4_wait_clnt_recover(clp);
1348                 if (ret != 0)
1349                         break;
1350                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1351                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1352                         break;
1353                 nfs4_schedule_state_manager(clp);
1354                 ret = -EIO;
1355         }
1356         return ret;
1357 }
1358
1359 /*
1360  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1361  * @clp: client to process
1362  *
1363  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1364  * resend of the SETCLIENTID and hence re-establish the
1365  * callback channel. Then return all existing delegations.
1366  */
1367 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1368 {
1369         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1370         nfs_expire_all_delegations(clp);
1371         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1372                         clp->cl_hostname);
1373 }
1374
1375 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1376 {
1377         nfs40_handle_cb_pathdown(clp);
1378         nfs4_schedule_state_manager(clp);
1379 }
1380
1381 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1382 {
1383
1384         if (!nfs4_valid_open_stateid(state))
1385                 return 0;
1386         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1387         /* Don't recover state that expired before the reboot */
1388         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1389                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1390                 return 0;
1391         }
1392         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1393         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1394         return 1;
1395 }
1396
1397 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1398 {
1399         if (!nfs4_valid_open_stateid(state))
1400                 return 0;
1401         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1402         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1403         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1404         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1405         return 1;
1406 }
1407
1408 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1409 {
1410         struct nfs_client *clp = server->nfs_client;
1411
1412         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1413                 return -EBADF;
1414         nfs_inode_find_delegation_state_and_recover(state->inode,
1415                         &state->stateid);
1416         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1417                         clp->cl_hostname);
1418         nfs4_schedule_state_manager(clp);
1419         return 0;
1420 }
1421 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1422
1423 static struct nfs4_lock_state *
1424 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1425                 const nfs4_stateid *stateid)
1426 {
1427         struct nfs4_lock_state *pos;
1428
1429         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1430                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1431                         continue;
1432                 if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1433                         return pos;
1434         }
1435         return NULL;
1436 }
1437
1438 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1439                 const nfs4_stateid *stateid)
1440 {
1441         bool found = false;
1442
1443         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1444                 spin_lock(&state->state_lock);
1445                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1446                         found = true;
1447                 spin_unlock(&state->state_lock);
1448         }
1449         return found;
1450 }
1451
1452 void nfs_inode_find_state_and_recover(struct inode *inode,
1453                 const nfs4_stateid *stateid)
1454 {
1455         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1456         struct nfs_inode *nfsi = NFS_I(inode);
1457         struct nfs_open_context *ctx;
1458         struct nfs4_state *state;
1459         bool found = false;
1460
1461         rcu_read_lock();
1462         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1463                 state = ctx->state;
1464                 if (state == NULL)
1465                         continue;
1466                 if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1467                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1468                         found = true;
1469                         continue;
1470                 }
1471                 if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1472                     nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1473                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1474                         found = true;
1475                         continue;
1476                 }
1477                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1478                     nfs4_state_mark_reclaim_nograce(clp, state))
1479                         found = true;
1480         }
1481         rcu_read_unlock();
1482
1483         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1484         if (found)
1485                 nfs4_schedule_state_manager(clp);
1486 }
1487
1488 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1489 {
1490         struct inode *inode = state->inode;
1491         struct nfs_inode *nfsi = NFS_I(inode);
1492         struct nfs_open_context *ctx;
1493
1494         rcu_read_lock();
1495         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1496                 if (ctx->state != state)
1497                         continue;
1498                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1499                 pr_warn("NFSv4: state recovery failed for open file %pd2, "
1500                                 "error = %d\n", ctx->dentry, err);
1501         }
1502         rcu_read_unlock();
1503 }
1504
1505 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1506 {
1507         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1508         nfs4_state_mark_open_context_bad(state, error);
1509 }
1510
1511
1512 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1513 {
1514         struct inode *inode = state->inode;
1515         struct nfs_inode *nfsi = NFS_I(inode);
1516         struct file_lock *fl;
1517         struct nfs4_lock_state *lsp;
1518         int status = 0;
1519         struct file_lock_context *flctx = locks_inode_context(inode);
1520         struct list_head *list;
1521
1522         if (flctx == NULL)
1523                 return 0;
1524
1525         list = &flctx->flc_posix;
1526
1527         /* Guard against delegation returns and new lock/unlock calls */
1528         down_write(&nfsi->rwsem);
1529         spin_lock(&flctx->flc_lock);
1530 restart:
1531         for_each_file_lock(fl, list) {
1532                 if (nfs_file_open_context(fl->c.flc_file)->state != state)
1533                         continue;
1534                 spin_unlock(&flctx->flc_lock);
1535                 status = ops->recover_lock(state, fl);
1536                 switch (status) {
1537                 case 0:
1538                         break;
1539                 case -ETIMEDOUT:
1540                 case -ESTALE:
1541                 case -NFS4ERR_ADMIN_REVOKED:
1542                 case -NFS4ERR_STALE_STATEID:
1543                 case -NFS4ERR_BAD_STATEID:
1544                 case -NFS4ERR_EXPIRED:
1545                 case -NFS4ERR_NO_GRACE:
1546                 case -NFS4ERR_STALE_CLIENTID:
1547                 case -NFS4ERR_BADSESSION:
1548                 case -NFS4ERR_BADSLOT:
1549                 case -NFS4ERR_BAD_HIGH_SLOT:
1550                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1551                         goto out;
1552                 default:
1553                         pr_err("NFS: %s: unhandled error %d\n",
1554                                         __func__, status);
1555                         fallthrough;
1556                 case -ENOMEM:
1557                 case -NFS4ERR_DENIED:
1558                 case -NFS4ERR_RECLAIM_BAD:
1559                 case -NFS4ERR_RECLAIM_CONFLICT:
1560                         lsp = fl->fl_u.nfs4_fl.owner;
1561                         if (lsp)
1562                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1563                         status = 0;
1564                 }
1565                 spin_lock(&flctx->flc_lock);
1566         }
1567         if (list == &flctx->flc_posix) {
1568                 list = &flctx->flc_flock;
1569                 goto restart;
1570         }
1571         spin_unlock(&flctx->flc_lock);
1572 out:
1573         up_write(&nfsi->rwsem);
1574         return status;
1575 }
1576
1577 #ifdef CONFIG_NFS_V4_2
1578 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1579 {
1580         struct nfs4_copy_state *copy;
1581
1582         if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1583                 !test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1584                 return;
1585
1586         spin_lock(&sp->so_server->nfs_client->cl_lock);
1587         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1588                 if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1589                                 !nfs4_stateid_match_other(&state->stateid,
1590                                 &copy->parent_dst_state->stateid)))
1591                                 continue;
1592                 copy->flags = 1;
1593                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1594                                 &state->flags)) {
1595                         clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1596                         complete(&copy->completion);
1597                 }
1598         }
1599         list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1600                 if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1601                                 !nfs4_stateid_match_other(&state->stateid,
1602                                 &copy->parent_src_state->stateid)))
1603                                 continue;
1604                 copy->flags = 1;
1605                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1606                                 &state->flags))
1607                         complete(&copy->completion);
1608         }
1609         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1610 }
1611 #else /* !CONFIG_NFS_V4_2 */
1612 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1613                                          struct nfs4_state *state)
1614 {
1615 }
1616 #endif /* CONFIG_NFS_V4_2 */
1617
1618 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1619                                      const struct nfs4_state_recovery_ops *ops,
1620                                      int *lost_locks)
1621 {
1622         struct nfs4_lock_state *lock;
1623         int status;
1624
1625         status = ops->recover_open(sp, state);
1626         if (status < 0)
1627                 return status;
1628
1629         status = nfs4_reclaim_locks(state, ops);
1630         if (status < 0)
1631                 return status;
1632
1633         if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1634                 spin_lock(&state->state_lock);
1635                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1636                         trace_nfs4_state_lock_reclaim(state, lock);
1637                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags) &&
1638                             !test_bit(NFS_LOCK_UNLOCKING, &lock->ls_flags))
1639                                 *lost_locks += 1;
1640                 }
1641                 spin_unlock(&state->state_lock);
1642         }
1643
1644         nfs42_complete_copies(sp, state);
1645         clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1646         return status;
1647 }
1648
1649 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp,
1650                                    const struct nfs4_state_recovery_ops *ops,
1651                                    int *lost_locks)
1652 {
1653         struct nfs4_state *state;
1654         unsigned int loop = 0;
1655         int status = 0;
1656 #ifdef CONFIG_NFS_V4_2
1657         bool found_ssc_copy_state = false;
1658 #endif /* CONFIG_NFS_V4_2 */
1659
1660         /* Note: we rely on the sp->so_states list being ordered
1661          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1662          * states first.
1663          * This is needed to ensure that the server won't give us any
1664          * read delegations that we have to return if, say, we are
1665          * recovering after a network partition or a reboot from a
1666          * server that doesn't support a grace period.
1667          */
1668         spin_lock(&sp->so_lock);
1669 restart:
1670         list_for_each_entry(state, &sp->so_states, open_states) {
1671                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1672                         continue;
1673                 if (!nfs4_valid_open_stateid(state))
1674                         continue;
1675                 if (state->state == 0)
1676                         continue;
1677 #ifdef CONFIG_NFS_V4_2
1678                 if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1679                         nfs4_state_mark_recovery_failed(state, -EIO);
1680                         found_ssc_copy_state = true;
1681                         continue;
1682                 }
1683 #endif /* CONFIG_NFS_V4_2 */
1684                 refcount_inc(&state->count);
1685                 spin_unlock(&sp->so_lock);
1686                 status = __nfs4_reclaim_open_state(sp, state, ops, lost_locks);
1687
1688                 switch (status) {
1689                 default:
1690                         if (status >= 0) {
1691                                 loop = 0;
1692                                 break;
1693                         }
1694                         printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1695                         fallthrough;
1696                 case -ENOENT:
1697                 case -ENOMEM:
1698                 case -EACCES:
1699                 case -EROFS:
1700                 case -EIO:
1701                 case -ESTALE:
1702                         /* Open state on this file cannot be recovered */
1703                         nfs4_state_mark_recovery_failed(state, status);
1704                         break;
1705                 case -EAGAIN:
1706                         ssleep(1);
1707                         if (loop++ < 10) {
1708                                 set_bit(ops->state_flag_bit, &state->flags);
1709                                 break;
1710                         }
1711                         fallthrough;
1712                 case -NFS4ERR_ADMIN_REVOKED:
1713                 case -NFS4ERR_STALE_STATEID:
1714                 case -NFS4ERR_OLD_STATEID:
1715                 case -NFS4ERR_BAD_STATEID:
1716                 case -NFS4ERR_RECLAIM_BAD:
1717                 case -NFS4ERR_RECLAIM_CONFLICT:
1718                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1719                         break;
1720                 case -NFS4ERR_EXPIRED:
1721                 case -NFS4ERR_NO_GRACE:
1722                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1723                         fallthrough;
1724                 case -NFS4ERR_STALE_CLIENTID:
1725                 case -NFS4ERR_BADSESSION:
1726                 case -NFS4ERR_BADSLOT:
1727                 case -NFS4ERR_BAD_HIGH_SLOT:
1728                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1729                 case -ETIMEDOUT:
1730                         goto out_err;
1731                 }
1732                 nfs4_put_open_state(state);
1733                 spin_lock(&sp->so_lock);
1734                 goto restart;
1735         }
1736         spin_unlock(&sp->so_lock);
1737 #ifdef CONFIG_NFS_V4_2
1738         if (found_ssc_copy_state)
1739                 return -EIO;
1740 #endif /* CONFIG_NFS_V4_2 */
1741         return 0;
1742 out_err:
1743         nfs4_put_open_state(state);
1744         spin_lock(&sp->so_lock);
1745         spin_unlock(&sp->so_lock);
1746         return status;
1747 }
1748
1749 static void nfs4_clear_open_state(struct nfs4_state *state)
1750 {
1751         struct nfs4_lock_state *lock;
1752
1753         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1754         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1755         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1756         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1757         spin_lock(&state->state_lock);
1758         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1759                 lock->ls_seqid.flags = 0;
1760                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1761         }
1762         spin_unlock(&state->state_lock);
1763 }
1764
1765 static void nfs4_reset_seqids(struct nfs_server *server,
1766         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1767 {
1768         struct nfs_client *clp = server->nfs_client;
1769         struct nfs4_state_owner *sp;
1770         struct rb_node *pos;
1771         struct nfs4_state *state;
1772
1773         spin_lock(&clp->cl_lock);
1774         for (pos = rb_first(&server->state_owners);
1775              pos != NULL;
1776              pos = rb_next(pos)) {
1777                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1778                 sp->so_seqid.flags = 0;
1779                 spin_lock(&sp->so_lock);
1780                 list_for_each_entry(state, &sp->so_states, open_states) {
1781                         if (mark_reclaim(clp, state))
1782                                 nfs4_clear_open_state(state);
1783                 }
1784                 spin_unlock(&sp->so_lock);
1785         }
1786         spin_unlock(&clp->cl_lock);
1787 }
1788
1789 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1790         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1791 {
1792         struct nfs_server *server;
1793
1794         rcu_read_lock();
1795         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1796                 nfs4_reset_seqids(server, mark_reclaim);
1797         rcu_read_unlock();
1798 }
1799
1800 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1801 {
1802         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1803         /* Mark all delegations for reclaim */
1804         nfs_delegation_mark_reclaim(clp);
1805         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1806 }
1807
1808 static int nfs4_reclaim_complete(struct nfs_client *clp,
1809                                  const struct nfs4_state_recovery_ops *ops,
1810                                  const struct cred *cred)
1811 {
1812         /* Notify the server we're done reclaiming our state */
1813         if (ops->reclaim_complete)
1814                 return ops->reclaim_complete(clp, cred);
1815         return 0;
1816 }
1817
1818 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1819 {
1820         struct nfs_client *clp = server->nfs_client;
1821         struct nfs4_state_owner *sp;
1822         struct rb_node *pos;
1823         struct nfs4_state *state;
1824
1825         spin_lock(&clp->cl_lock);
1826         for (pos = rb_first(&server->state_owners);
1827              pos != NULL;
1828              pos = rb_next(pos)) {
1829                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1830                 spin_lock(&sp->so_lock);
1831                 list_for_each_entry(state, &sp->so_states, open_states) {
1832                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1833                                                 &state->flags))
1834                                 continue;
1835                         nfs4_state_mark_reclaim_nograce(clp, state);
1836                 }
1837                 spin_unlock(&sp->so_lock);
1838         }
1839         spin_unlock(&clp->cl_lock);
1840 }
1841
1842 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1843 {
1844         struct nfs_server *server;
1845
1846         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1847                 return 0;
1848
1849         rcu_read_lock();
1850         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1851                 nfs4_clear_reclaim_server(server);
1852         rcu_read_unlock();
1853
1854         nfs_delegation_reap_unclaimed(clp);
1855         return 1;
1856 }
1857
1858 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1859 {
1860         const struct nfs4_state_recovery_ops *ops;
1861         const struct cred *cred;
1862         int err;
1863
1864         if (!nfs4_state_clear_reclaim_reboot(clp))
1865                 return;
1866         ops = clp->cl_mvops->reboot_recovery_ops;
1867         cred = nfs4_get_clid_cred(clp);
1868         err = nfs4_reclaim_complete(clp, ops, cred);
1869         put_cred(cred);
1870         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1871                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1872 }
1873
1874 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1875 {
1876         nfs_mark_test_expired_all_delegations(clp);
1877         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1878 }
1879
1880 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1881 {
1882         switch (error) {
1883         case 0:
1884                 break;
1885         case -NFS4ERR_CB_PATH_DOWN:
1886                 nfs40_handle_cb_pathdown(clp);
1887                 break;
1888         case -NFS4ERR_NO_GRACE:
1889                 nfs4_state_end_reclaim_reboot(clp);
1890                 break;
1891         case -NFS4ERR_STALE_CLIENTID:
1892                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1893                 nfs4_state_start_reclaim_reboot(clp);
1894                 break;
1895         case -NFS4ERR_EXPIRED:
1896                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1897                 nfs4_state_start_reclaim_nograce(clp);
1898                 break;
1899         case -NFS4ERR_BADSESSION:
1900         case -NFS4ERR_BADSLOT:
1901         case -NFS4ERR_BAD_HIGH_SLOT:
1902         case -NFS4ERR_DEADSESSION:
1903         case -NFS4ERR_SEQ_FALSE_RETRY:
1904         case -NFS4ERR_SEQ_MISORDERED:
1905                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1906                 /* Zero session reset errors */
1907                 break;
1908         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1909                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1910                 break;
1911         default:
1912                 dprintk("%s: failed to handle error %d for server %s\n",
1913                                 __func__, error, clp->cl_hostname);
1914                 return error;
1915         }
1916         dprintk("%s: handled error %d for server %s\n", __func__, error,
1917                         clp->cl_hostname);
1918         return 0;
1919 }
1920
1921 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1922 {
1923         struct nfs4_state_owner *sp;
1924         struct nfs_server *server;
1925         struct rb_node *pos;
1926         LIST_HEAD(freeme);
1927         int lost_locks = 0;
1928         int status;
1929
1930         status = nfs4_begin_drain_session(clp);
1931         if (status < 0)
1932                 return status;
1933 restart:
1934         rcu_read_lock();
1935         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1936                 nfs4_purge_state_owners(server, &freeme);
1937                 spin_lock(&clp->cl_lock);
1938                 for (pos = rb_first(&server->state_owners);
1939                      pos != NULL;
1940                      pos = rb_next(pos)) {
1941                         sp = rb_entry(pos,
1942                                 struct nfs4_state_owner, so_server_node);
1943                         if (!test_and_clear_bit(ops->owner_flag_bit,
1944                                                         &sp->so_flags))
1945                                 continue;
1946                         if (!atomic_inc_not_zero(&sp->so_count))
1947                                 continue;
1948                         spin_unlock(&clp->cl_lock);
1949                         rcu_read_unlock();
1950
1951                         status = nfs4_reclaim_open_state(sp, ops, &lost_locks);
1952                         if (status < 0) {
1953                                 if (lost_locks)
1954                                         pr_warn("NFS: %s: lost %d locks\n",
1955                                                 clp->cl_hostname, lost_locks);
1956                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1957                                 nfs4_put_state_owner(sp);
1958                                 status = nfs4_recovery_handle_error(clp, status);
1959                                 return (status != 0) ? status : -EAGAIN;
1960                         }
1961
1962                         nfs4_put_state_owner(sp);
1963                         goto restart;
1964                 }
1965                 spin_unlock(&clp->cl_lock);
1966         }
1967         rcu_read_unlock();
1968         nfs4_free_state_owners(&freeme);
1969         if (lost_locks)
1970                 pr_warn("NFS: %s: lost %d locks\n",
1971                         clp->cl_hostname, lost_locks);
1972         return 0;
1973 }
1974
1975 static int nfs4_check_lease(struct nfs_client *clp)
1976 {
1977         const struct cred *cred;
1978         const struct nfs4_state_maintenance_ops *ops =
1979                 clp->cl_mvops->state_renewal_ops;
1980         int status;
1981
1982         /* Is the client already known to have an expired lease? */
1983         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1984                 return 0;
1985         cred = ops->get_state_renewal_cred(clp);
1986         if (cred == NULL) {
1987                 cred = nfs4_get_clid_cred(clp);
1988                 status = -ENOKEY;
1989                 if (cred == NULL)
1990                         goto out;
1991         }
1992         status = ops->renew_lease(clp, cred);
1993         put_cred(cred);
1994         if (status == -ETIMEDOUT) {
1995                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1996                 return 0;
1997         }
1998 out:
1999         return nfs4_recovery_handle_error(clp, status);
2000 }
2001
2002 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
2003  * and for recoverable errors on EXCHANGE_ID for v4.1
2004  */
2005 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
2006 {
2007         switch (status) {
2008         case -NFS4ERR_SEQ_MISORDERED:
2009                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
2010                         return -ESERVERFAULT;
2011                 /* Lease confirmation error: retry after purging the lease */
2012                 ssleep(1);
2013                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2014                 break;
2015         case -NFS4ERR_STALE_CLIENTID:
2016                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2017                 nfs4_state_start_reclaim_reboot(clp);
2018                 break;
2019         case -NFS4ERR_CLID_INUSE:
2020                 pr_err("NFS: Server %s reports our clientid is in use\n",
2021                         clp->cl_hostname);
2022                 nfs_mark_client_ready(clp, -EPERM);
2023                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2024                 return -EPERM;
2025         case -EACCES:
2026         case -NFS4ERR_DELAY:
2027         case -EAGAIN:
2028                 ssleep(1);
2029                 break;
2030
2031         case -NFS4ERR_MINOR_VERS_MISMATCH:
2032                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2033                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2034                 dprintk("%s: exit with error %d for server %s\n",
2035                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
2036                 return -EPROTONOSUPPORT;
2037         case -ENOSPC:
2038                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2039                         nfs_mark_client_ready(clp, -EIO);
2040                 return -EIO;
2041         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2042                                  * in nfs4_exchange_id */
2043         default:
2044                 dprintk("%s: exit with error %d for server %s\n", __func__,
2045                                 status, clp->cl_hostname);
2046                 return status;
2047         }
2048         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2049         dprintk("%s: handled error %d for server %s\n", __func__, status,
2050                         clp->cl_hostname);
2051         return 0;
2052 }
2053
2054 static int nfs4_establish_lease(struct nfs_client *clp)
2055 {
2056         const struct cred *cred;
2057         const struct nfs4_state_recovery_ops *ops =
2058                 clp->cl_mvops->reboot_recovery_ops;
2059         int status;
2060
2061         status = nfs4_begin_drain_session(clp);
2062         if (status != 0)
2063                 return status;
2064         cred = nfs4_get_clid_cred(clp);
2065         if (cred == NULL)
2066                 return -ENOENT;
2067         status = ops->establish_clid(clp, cred);
2068         put_cred(cred);
2069         if (status != 0)
2070                 return status;
2071         pnfs_destroy_all_layouts(clp);
2072         return 0;
2073 }
2074
2075 /*
2076  * Returns zero or a negative errno.  NFS4ERR values are converted
2077  * to local errno values.
2078  */
2079 static int nfs4_reclaim_lease(struct nfs_client *clp)
2080 {
2081         int status;
2082
2083         status = nfs4_establish_lease(clp);
2084         if (status < 0)
2085                 return nfs4_handle_reclaim_lease_error(clp, status);
2086         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2087                 nfs4_state_start_reclaim_nograce(clp);
2088         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2089                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2090         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2091         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2092         return 0;
2093 }
2094
2095 static int nfs4_purge_lease(struct nfs_client *clp)
2096 {
2097         int status;
2098
2099         status = nfs4_establish_lease(clp);
2100         if (status < 0)
2101                 return nfs4_handle_reclaim_lease_error(clp, status);
2102         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2103         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2104         nfs4_state_start_reclaim_nograce(clp);
2105         return 0;
2106 }
2107
2108 /*
2109  * Try remote migration of one FSID from a source server to a
2110  * destination server.  The source server provides a list of
2111  * potential destinations.
2112  *
2113  * Returns zero or a negative NFS4ERR status code.
2114  */
2115 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2116 {
2117         struct nfs_client *clp = server->nfs_client;
2118         struct nfs4_fs_locations *locations = NULL;
2119         struct inode *inode;
2120         struct page *page;
2121         int status, result;
2122
2123         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2124                         (unsigned long long)server->fsid.major,
2125                         (unsigned long long)server->fsid.minor,
2126                         clp->cl_hostname);
2127
2128         result = 0;
2129         page = alloc_page(GFP_KERNEL);
2130         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2131         if (page == NULL || locations == NULL) {
2132                 dprintk("<-- %s: no memory\n", __func__);
2133                 goto out;
2134         }
2135         locations->fattr = nfs_alloc_fattr();
2136         if (locations->fattr == NULL) {
2137                 dprintk("<-- %s: no memory\n", __func__);
2138                 goto out;
2139         }
2140
2141         inode = d_inode(server->super->s_root);
2142         result = nfs4_proc_get_locations(server, NFS_FH(inode), locations,
2143                                          page, cred);
2144         if (result) {
2145                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2146                         __func__, result);
2147                 goto out;
2148         }
2149
2150         result = -NFS4ERR_NXIO;
2151         if (!locations->nlocations)
2152                 goto out;
2153
2154         if (!(locations->fattr->valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2155                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2156                         __func__);
2157                 goto out;
2158         }
2159
2160         status = nfs4_begin_drain_session(clp);
2161         if (status != 0) {
2162                 result = status;
2163                 goto out;
2164         }
2165
2166         status = nfs4_replace_transport(server, locations);
2167         if (status != 0) {
2168                 dprintk("<-- %s: failed to replace transport: %d\n",
2169                         __func__, status);
2170                 goto out;
2171         }
2172
2173         result = 0;
2174         dprintk("<-- %s: migration succeeded\n", __func__);
2175
2176 out:
2177         if (page != NULL)
2178                 __free_page(page);
2179         if (locations != NULL)
2180                 kfree(locations->fattr);
2181         kfree(locations);
2182         if (result) {
2183                 pr_err("NFS: migration recovery failed (server %s)\n",
2184                                 clp->cl_hostname);
2185                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2186         }
2187         return result;
2188 }
2189
2190 /*
2191  * Returns zero or a negative NFS4ERR status code.
2192  */
2193 static int nfs4_handle_migration(struct nfs_client *clp)
2194 {
2195         const struct nfs4_state_maintenance_ops *ops =
2196                                 clp->cl_mvops->state_renewal_ops;
2197         struct nfs_server *server;
2198         const struct cred *cred;
2199
2200         dprintk("%s: migration reported on \"%s\"\n", __func__,
2201                         clp->cl_hostname);
2202
2203         cred = ops->get_state_renewal_cred(clp);
2204         if (cred == NULL)
2205                 return -NFS4ERR_NOENT;
2206
2207         clp->cl_mig_gen++;
2208 restart:
2209         rcu_read_lock();
2210         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2211                 int status;
2212
2213                 if (server->mig_gen == clp->cl_mig_gen)
2214                         continue;
2215                 server->mig_gen = clp->cl_mig_gen;
2216
2217                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2218                                                 &server->mig_status))
2219                         continue;
2220
2221                 rcu_read_unlock();
2222                 status = nfs4_try_migration(server, cred);
2223                 if (status < 0) {
2224                         put_cred(cred);
2225                         return status;
2226                 }
2227                 goto restart;
2228         }
2229         rcu_read_unlock();
2230         put_cred(cred);
2231         return 0;
2232 }
2233
2234 /*
2235  * Test each nfs_server on the clp's cl_superblocks list to see
2236  * if it's moved to another server.  Stop when the server no longer
2237  * returns NFS4ERR_LEASE_MOVED.
2238  */
2239 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2240 {
2241         const struct nfs4_state_maintenance_ops *ops =
2242                                 clp->cl_mvops->state_renewal_ops;
2243         struct nfs_server *server;
2244         const struct cred *cred;
2245
2246         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2247                         clp->cl_hostname);
2248
2249         cred = ops->get_state_renewal_cred(clp);
2250         if (cred == NULL)
2251                 return -NFS4ERR_NOENT;
2252
2253         clp->cl_mig_gen++;
2254 restart:
2255         rcu_read_lock();
2256         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2257                 struct inode *inode;
2258                 int status;
2259
2260                 if (server->mig_gen == clp->cl_mig_gen)
2261                         continue;
2262                 server->mig_gen = clp->cl_mig_gen;
2263
2264                 rcu_read_unlock();
2265
2266                 inode = d_inode(server->super->s_root);
2267                 status = nfs4_proc_fsid_present(inode, cred);
2268                 if (status != -NFS4ERR_MOVED)
2269                         goto restart;   /* wasn't this one */
2270                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2271                         goto restart;   /* there are more */
2272                 goto out;
2273         }
2274         rcu_read_unlock();
2275
2276 out:
2277         put_cred(cred);
2278         return 0;
2279 }
2280
2281 /**
2282  * nfs4_discover_server_trunking - Detect server IP address trunking
2283  *
2284  * @clp: nfs_client under test
2285  * @result: OUT: found nfs_client, or clp
2286  *
2287  * Returns zero or a negative errno.  If zero is returned,
2288  * an nfs_client pointer is planted in "result".
2289  *
2290  * Note: since we are invoked in process context, and
2291  * not from inside the state manager, we cannot use
2292  * nfs4_handle_reclaim_lease_error().
2293  */
2294 int nfs4_discover_server_trunking(struct nfs_client *clp,
2295                                   struct nfs_client **result)
2296 {
2297         const struct nfs4_state_recovery_ops *ops =
2298                                 clp->cl_mvops->reboot_recovery_ops;
2299         struct rpc_clnt *clnt;
2300         const struct cred *cred;
2301         int i, status;
2302
2303         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2304
2305         clnt = clp->cl_rpcclient;
2306         i = 0;
2307
2308         mutex_lock(&nfs_clid_init_mutex);
2309 again:
2310         status  = -ENOENT;
2311         cred = nfs4_get_clid_cred(clp);
2312         if (cred == NULL)
2313                 goto out_unlock;
2314
2315         status = ops->detect_trunking(clp, result, cred);
2316         put_cred(cred);
2317         switch (status) {
2318         case 0:
2319         case -EINTR:
2320         case -ERESTARTSYS:
2321                 break;
2322         case -ETIMEDOUT:
2323                 if (clnt->cl_softrtry)
2324                         break;
2325                 fallthrough;
2326         case -NFS4ERR_DELAY:
2327         case -EAGAIN:
2328                 ssleep(1);
2329                 fallthrough;
2330         case -NFS4ERR_STALE_CLIENTID:
2331                 dprintk("NFS: %s after status %d, retrying\n",
2332                         __func__, status);
2333                 goto again;
2334         case -EACCES:
2335                 if (i++ == 0) {
2336                         nfs4_root_machine_cred(clp);
2337                         goto again;
2338                 }
2339                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2340                         break;
2341                 fallthrough;
2342         case -NFS4ERR_CLID_INUSE:
2343         case -NFS4ERR_WRONGSEC:
2344                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2345                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2346                         status = -EPERM;
2347                         break;
2348                 }
2349                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2350                 if (IS_ERR(clnt)) {
2351                         status = PTR_ERR(clnt);
2352                         break;
2353                 }
2354                 /* Note: this is safe because we haven't yet marked the
2355                  * client as ready, so we are the only user of
2356                  * clp->cl_rpcclient
2357                  */
2358                 clnt = xchg(&clp->cl_rpcclient, clnt);
2359                 rpc_shutdown_client(clnt);
2360                 clnt = clp->cl_rpcclient;
2361                 goto again;
2362
2363         case -NFS4ERR_MINOR_VERS_MISMATCH:
2364                 status = -EPROTONOSUPPORT;
2365                 break;
2366
2367         case -EKEYEXPIRED:
2368         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2369                                  * in nfs4_exchange_id */
2370                 status = -EKEYEXPIRED;
2371                 break;
2372         default:
2373                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2374                                 __func__, status);
2375                 status = -EIO;
2376         }
2377
2378 out_unlock:
2379         mutex_unlock(&nfs_clid_init_mutex);
2380         dprintk("NFS: %s: status = %d\n", __func__, status);
2381         return status;
2382 }
2383
2384 #ifdef CONFIG_NFS_V4_1
2385 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2386 {
2387         struct nfs_client *clp = session->clp;
2388
2389         switch (err) {
2390         default:
2391                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2392                 break;
2393         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2394                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2395         }
2396         nfs4_schedule_state_manager(clp);
2397 }
2398 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2399
2400 void nfs41_notify_server(struct nfs_client *clp)
2401 {
2402         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2403         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2404         nfs4_schedule_state_manager(clp);
2405 }
2406
2407 static void nfs4_reset_all_state(struct nfs_client *clp)
2408 {
2409         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2410                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2411                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2412                 nfs4_state_start_reclaim_nograce(clp);
2413                 dprintk("%s: scheduling reset of all state for server %s!\n",
2414                                 __func__, clp->cl_hostname);
2415                 nfs4_schedule_state_manager(clp);
2416         }
2417 }
2418
2419 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2420 {
2421         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2422                 nfs4_state_start_reclaim_reboot(clp);
2423                 dprintk("%s: server %s rebooted!\n", __func__,
2424                                 clp->cl_hostname);
2425                 nfs4_schedule_state_manager(clp);
2426         }
2427 }
2428
2429 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2430 {
2431         nfs4_reset_all_state(clp);
2432         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2433 }
2434
2435 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2436 {
2437         nfs4_state_start_reclaim_nograce(clp);
2438         nfs4_schedule_state_manager(clp);
2439
2440         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2441 }
2442
2443 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2444 {
2445         /* FIXME: For now, we destroy all layouts. */
2446         pnfs_destroy_all_layouts(clp);
2447         nfs_test_expired_all_delegations(clp);
2448         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2449                         clp->cl_hostname);
2450 }
2451
2452 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2453 {
2454         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2455         nfs4_schedule_state_manager(clp);
2456
2457         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2458                         clp->cl_hostname);
2459 }
2460
2461 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2462 {
2463         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2464                 &clp->cl_state) == 0)
2465                 nfs4_schedule_state_manager(clp);
2466 }
2467
2468 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2469                 bool recovery)
2470 {
2471         if (!flags)
2472                 return;
2473
2474         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2475                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2476         /*
2477          * If we're called from the state manager thread, then assume we're
2478          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2479          * Those flags are expected to remain set until we're done
2480          * recovering (see RFC5661, section 18.46.3).
2481          */
2482         if (recovery)
2483                 goto out_recovery;
2484
2485         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2486                 nfs41_handle_server_reboot(clp);
2487         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2488                 nfs41_handle_all_state_revoked(clp);
2489         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2490                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2491                 nfs41_handle_some_state_revoked(clp);
2492         if (flags & SEQ4_STATUS_LEASE_MOVED)
2493                 nfs4_schedule_lease_moved_recovery(clp);
2494         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2495                 nfs41_handle_recallable_state_revoked(clp);
2496 out_recovery:
2497         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2498                 nfs41_handle_backchannel_fault(clp);
2499         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2500                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2501                 nfs41_handle_cb_path_down(clp);
2502 }
2503
2504 static int nfs4_reset_session(struct nfs_client *clp)
2505 {
2506         const struct cred *cred;
2507         int status;
2508
2509         if (!nfs4_has_session(clp))
2510                 return 0;
2511         status = nfs4_begin_drain_session(clp);
2512         if (status != 0)
2513                 return status;
2514         cred = nfs4_get_clid_cred(clp);
2515         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2516         switch (status) {
2517         case 0:
2518         case -NFS4ERR_BADSESSION:
2519         case -NFS4ERR_DEADSESSION:
2520                 break;
2521         case -NFS4ERR_BACK_CHAN_BUSY:
2522         case -NFS4ERR_DELAY:
2523                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2524                 status = 0;
2525                 ssleep(1);
2526                 goto out;
2527         default:
2528                 status = nfs4_recovery_handle_error(clp, status);
2529                 goto out;
2530         }
2531
2532         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2533         status = nfs4_proc_create_session(clp, cred);
2534         if (status) {
2535                 dprintk("%s: session reset failed with status %d for server %s!\n",
2536                         __func__, status, clp->cl_hostname);
2537                 status = nfs4_handle_reclaim_lease_error(clp, status);
2538                 goto out;
2539         }
2540         nfs41_finish_session_reset(clp);
2541         dprintk("%s: session reset was successful for server %s!\n",
2542                         __func__, clp->cl_hostname);
2543 out:
2544         put_cred(cred);
2545         return status;
2546 }
2547
2548 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2549 {
2550         const struct cred *cred;
2551         int ret;
2552
2553         if (!nfs4_has_session(clp))
2554                 return 0;
2555         ret = nfs4_begin_drain_session(clp);
2556         if (ret != 0)
2557                 return ret;
2558         cred = nfs4_get_clid_cred(clp);
2559         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2560         put_cred(cred);
2561         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2562         switch (ret) {
2563         case 0:
2564                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2565                         __func__, clp->cl_hostname);
2566                 break;
2567         case -NFS4ERR_DELAY:
2568                 ssleep(1);
2569                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2570                 break;
2571         default:
2572                 return nfs4_recovery_handle_error(clp, ret);
2573         }
2574         return 0;
2575 }
2576
2577 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2578 {
2579         int iomode = 0;
2580
2581         if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2582                 iomode += IOMODE_READ;
2583         if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2584                 iomode += IOMODE_RW;
2585         /* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2586         if (iomode) {
2587                 pnfs_layout_return_unused_byclid(clp, iomode);
2588                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2589         }
2590 }
2591 #else /* CONFIG_NFS_V4_1 */
2592 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2593
2594 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2595 {
2596         return 0;
2597 }
2598
2599 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2600 {
2601 }
2602 #endif /* CONFIG_NFS_V4_1 */
2603
2604 static void nfs4_state_manager(struct nfs_client *clp)
2605 {
2606         unsigned int memflags;
2607         int status = 0;
2608         const char *section = "", *section_sep = "";
2609
2610         /*
2611          * State recovery can deadlock if the direct reclaim code tries
2612          * start NFS writeback. So ensure memory allocations are all
2613          * GFP_NOFS.
2614          */
2615         memflags = memalloc_nofs_save();
2616
2617         /* Ensure exclusive access to NFSv4 state */
2618         do {
2619                 trace_nfs4_state_mgr(clp);
2620                 clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2621                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2622                         section = "purge state";
2623                         status = nfs4_purge_lease(clp);
2624                         if (status < 0)
2625                                 goto out_error;
2626                         continue;
2627                 }
2628
2629                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2630                         section = "lease expired";
2631                         /* We're going to have to re-establish a clientid */
2632                         status = nfs4_reclaim_lease(clp);
2633                         if (status < 0)
2634                                 goto out_error;
2635                         continue;
2636                 }
2637
2638                 /* Initialize or reset the session */
2639                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2640                         section = "reset session";
2641                         status = nfs4_reset_session(clp);
2642                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2643                                 continue;
2644                         if (status < 0)
2645                                 goto out_error;
2646                 }
2647
2648                 /* Send BIND_CONN_TO_SESSION */
2649                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2650                                 &clp->cl_state)) {
2651                         section = "bind conn to session";
2652                         status = nfs4_bind_conn_to_session(clp);
2653                         if (status < 0)
2654                                 goto out_error;
2655                         continue;
2656                 }
2657
2658                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2659                         section = "check lease";
2660                         status = nfs4_check_lease(clp);
2661                         if (status < 0)
2662                                 goto out_error;
2663                         continue;
2664                 }
2665
2666                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2667                         section = "migration";
2668                         status = nfs4_handle_migration(clp);
2669                         if (status < 0)
2670                                 goto out_error;
2671                 }
2672
2673                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2674                         section = "lease moved";
2675                         status = nfs4_handle_lease_moved(clp);
2676                         if (status < 0)
2677                                 goto out_error;
2678                 }
2679
2680                 /* First recover reboot state... */
2681                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2682                         section = "reclaim reboot";
2683                         status = nfs4_do_reclaim(clp,
2684                                 clp->cl_mvops->reboot_recovery_ops);
2685                         if (status == -EAGAIN)
2686                                 continue;
2687                         if (status < 0)
2688                                 goto out_error;
2689                         nfs4_state_end_reclaim_reboot(clp);
2690                         continue;
2691                 }
2692
2693                 /* Detect expired delegations... */
2694                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2695                         section = "detect expired delegations";
2696                         status = nfs4_begin_drain_session(clp);
2697                         if (status < 0)
2698                                 goto out_error;
2699                         nfs_reap_expired_delegations(clp);
2700                         continue;
2701                 }
2702
2703                 /* Now recover expired state... */
2704                 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2705                         section = "reclaim nograce";
2706                         status = nfs4_do_reclaim(clp,
2707                                 clp->cl_mvops->nograce_recovery_ops);
2708                         if (status == -EAGAIN)
2709                                 continue;
2710                         if (status < 0)
2711                                 goto out_error;
2712                         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2713                 }
2714
2715                 memalloc_nofs_restore(memflags);
2716                 nfs4_end_drain_session(clp);
2717                 nfs4_clear_state_manager_bit(clp);
2718
2719                 if (test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2720                     !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING,
2721                                       &clp->cl_state)) {
2722                         memflags = memalloc_nofs_save();
2723                         continue;
2724                 }
2725
2726                 if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2727                         if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2728                                 nfs_client_return_marked_delegations(clp);
2729                                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2730                         }
2731                         nfs4_layoutreturn_any_run(clp);
2732                         clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2733                 }
2734
2735                 return;
2736
2737         } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2738         goto out_drain;
2739
2740 out_error:
2741         if (strlen(section))
2742                 section_sep = ": ";
2743         trace_nfs4_state_mgr_failed(clp, section, status);
2744         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2745                         " with error %d\n", section_sep, section,
2746                         clp->cl_hostname, -status);
2747         ssleep(1);
2748 out_drain:
2749         memalloc_nofs_restore(memflags);
2750         nfs4_end_drain_session(clp);
2751         nfs4_clear_state_manager_bit(clp);
2752 }
2753
2754 static int nfs4_run_state_manager(void *ptr)
2755 {
2756         struct nfs_client *clp = ptr;
2757         struct rpc_clnt *cl = clp->cl_rpcclient;
2758
2759         while (cl != cl->cl_parent)
2760                 cl = cl->cl_parent;
2761
2762         allow_signal(SIGKILL);
2763 again:
2764         nfs4_state_manager(clp);
2765
2766         if (test_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) &&
2767             !test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state)) {
2768                 wait_var_event_interruptible(&clp->cl_state,
2769                                              test_bit(NFS4CLNT_RUN_MANAGER,
2770                                                       &clp->cl_state));
2771                 if (!atomic_read(&cl->cl_swapper))
2772                         clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2773                 if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2774                     !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2775                         goto again;
2776                 /* Either no longer a swapper, or were signalled */
2777                 clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2778         }
2779
2780         if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2781             test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2782             !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2783                 goto again;
2784
2785         nfs_put_client(clp);
2786         module_put_and_kthread_exit(0);
2787         return 0;
2788 }