fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63 #include "nfs4trace.h"
  64
  65 #define NFSDBG_FACILITY         NFSDBG_STATE
  66
  67 #define OPENOWNER_POOL_SIZE     8
  68
  69 const nfs4_stateid zero_stateid = {
  70         { .data = { 0 } },
  71         .type = NFS4_SPECIAL_STATEID_TYPE,
  72 };
  73 const nfs4_stateid invalid_stateid = {
  74         {
  75                 /* Funky initialiser keeps older gcc versions happy */
  76                 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
  77         },
  78         .type = NFS4_INVALID_STATEID_TYPE,
  79 };
  80
  81 const nfs4_stateid current_stateid = {
  82         {
  83                 /* Funky initialiser keeps older gcc versions happy */
  84                 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
  85         },
  86         .type = NFS4_SPECIAL_STATEID_TYPE,
  87 };
  88
  89 static DEFINE_MUTEX(nfs_clid_init_mutex);
  90
  91 static int nfs4_setup_state_renewal(struct nfs_client *clp)
  92 {
  93         int status;
  94         struct nfs_fsinfo fsinfo;
  95
  96         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
  97                 nfs4_schedule_state_renewal(clp);
  98                 return 0;
  99         }
 100
 101         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 102         if (status == 0) {
 103                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
 104                 nfs4_schedule_state_renewal(clp);
 105         }
 106
 107         return status;
 108 }
 109
 110 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
 111 {
 112         struct nfs4_setclientid_res clid = {
 113                 .clientid = clp->cl_clientid,
 114                 .confirm = clp->cl_confirm,
 115         };
 116         unsigned short port;
 117         int status;
 118         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 119
 120         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 121                 goto do_confirm;
 122         port = nn->nfs_callback_tcpport;
 123         if (clp->cl_addr.ss_family == AF_INET6)
 124                 port = nn->nfs_callback_tcpport6;
 125
 126         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 127         if (status != 0)
 128                 goto out;
 129         clp->cl_clientid = clid.clientid;
 130         clp->cl_confirm = clid.confirm;
 131         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 132 do_confirm:
 133         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
 134         if (status != 0)
 135                 goto out;
 136         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 137         nfs4_setup_state_renewal(clp);
 138 out:
 139         return status;
 140 }
 141
 142 /**
 143  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 144  *
 145  * @clp: nfs_client under test
 146  * @result: OUT: found nfs_client, or clp
 147  * @cred: credential to use for trunking test
 148  *
 149  * Returns zero, a negative errno, or a negative NFS4ERR status.
 150  * If zero is returned, an nfs_client pointer is planted in
 151  * "result".
 152  *
 153  * Note: The returned client may not yet be marked ready.
 154  */
 155 int nfs40_discover_server_trunking(struct nfs_client *clp,
 156                                    struct nfs_client **result,
 157                                    const struct cred *cred)
 158 {
 159         struct nfs4_setclientid_res clid = {
 160                 .clientid = clp->cl_clientid,
 161                 .confirm = clp->cl_confirm,
 162         };
 163         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 164         unsigned short port;
 165         int status;
 166
 167         port = nn->nfs_callback_tcpport;
 168         if (clp->cl_addr.ss_family == AF_INET6)
 169                 port = nn->nfs_callback_tcpport6;
 170
 171         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 172         if (status != 0)
 173                 goto out;
 174         clp->cl_clientid = clid.clientid;
 175         clp->cl_confirm = clid.confirm;
 176
 177         status = nfs40_walk_client_list(clp, result, cred);
 178         if (status == 0) {
 179                 /* Sustain the lease, even if it's empty.  If the clientid4
 180                  * goes stale it's of no use for trunking discovery. */
 181                 nfs4_schedule_state_renewal(*result);
 182
 183                 /* If the client state need to recover, do it. */
 184                 if (clp->cl_state)
 185                         nfs4_schedule_state_manager(clp);
 186         }
 187 out:
 188         return status;
 189 }
 190
 191 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
 192 {
 193         return get_cred(rpc_machine_cred());
 194 }
 195
 196 static void nfs4_root_machine_cred(struct nfs_client *clp)
 197 {
 198
 199         /* Force root creds instead of machine */
 200         clp->cl_principal = NULL;
 201         clp->cl_rpcclient->cl_principal = NULL;
 202 }
 203
 204 static const struct cred *
 205 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 206 {
 207         const struct cred *cred = NULL;
 208         struct nfs4_state_owner *sp;
 209         struct rb_node *pos;
 210
 211         for (pos = rb_first(&server->state_owners);
 212              pos != NULL;
 213              pos = rb_next(pos)) {
 214                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 215                 if (list_empty(&sp->so_states))
 216                         continue;
 217                 cred = get_cred(sp->so_cred);
 218                 break;
 219         }
 220         return cred;
 221 }
 222
 223 /**
 224  * nfs4_get_renew_cred - Acquire credential for a renew operation
 225  * @clp: client state handle
 226  *
 227  * Returns an rpc_cred with reference count bumped, or NULL.
 228  * Caller must hold clp->cl_lock.
 229  */
 230 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
 231 {
 232         const struct cred *cred = NULL;
 233         struct nfs_server *server;
 234
 235         /* Use machine credentials if available */
 236         cred = nfs4_get_machine_cred(clp);
 237         if (cred != NULL)
 238                 goto out;
 239
 240         spin_lock(&clp->cl_lock);
 241         rcu_read_lock();
 242         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 243                 cred = nfs4_get_renew_cred_server_locked(server);
 244                 if (cred != NULL)
 245                         break;
 246         }
 247         rcu_read_unlock();
 248         spin_unlock(&clp->cl_lock);
 249
 250 out:
 251         return cred;
 252 }
 253
 254 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 255 {
 256         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 257                 spin_lock(&tbl->slot_tbl_lock);
 258                 nfs41_wake_slot_table(tbl);
 259                 spin_unlock(&tbl->slot_tbl_lock);
 260         }
 261 }
 262
 263 static void nfs4_end_drain_session(struct nfs_client *clp)
 264 {
 265         struct nfs4_session *ses = clp->cl_session;
 266
 267         if (clp->cl_slot_tbl) {
 268                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 269                 return;
 270         }
 271
 272         if (ses != NULL) {
 273                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 274                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 275         }
 276 }
 277
 278 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 279 {
 280         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 281         spin_lock(&tbl->slot_tbl_lock);
 282         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 283                 reinit_completion(&tbl->complete);
 284                 spin_unlock(&tbl->slot_tbl_lock);
 285                 return wait_for_completion_interruptible(&tbl->complete);
 286         }
 287         spin_unlock(&tbl->slot_tbl_lock);
 288         return 0;
 289 }
 290
 291 static int nfs4_begin_drain_session(struct nfs_client *clp)
 292 {
 293         struct nfs4_session *ses = clp->cl_session;
 294         int ret;
 295
 296         if (clp->cl_slot_tbl)
 297                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 298
 299         /* back channel */
 300         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 301         if (ret)
 302                 return ret;
 303         /* fore channel */
 304         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 305 }
 306
 307 #if defined(CONFIG_NFS_V4_1)
 308
 309 static void nfs41_finish_session_reset(struct nfs_client *clp)
 310 {
 311         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 312         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 313         /* create_session negotiated new slot table */
 314         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 315         nfs4_setup_state_renewal(clp);
 316 }
 317
 318 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
 319 {
 320         int status;
 321
 322         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 323                 goto do_confirm;
 324         status = nfs4_proc_exchange_id(clp, cred);
 325         if (status != 0)
 326                 goto out;
 327         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 328 do_confirm:
 329         status = nfs4_proc_create_session(clp, cred);
 330         if (status != 0)
 331                 goto out;
 332         nfs41_finish_session_reset(clp);
 333         nfs_mark_client_ready(clp, NFS_CS_READY);
 334 out:
 335         return status;
 336 }
 337
 338 /**
 339  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 340  *
 341  * @clp: nfs_client under test
 342  * @result: OUT: found nfs_client, or clp
 343  * @cred: credential to use for trunking test
 344  *
 345  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 346  * If NFS4_OK is returned, an nfs_client pointer is planted in
 347  * "result".
 348  *
 349  * Note: The returned client may not yet be marked ready.
 350  */
 351 int nfs41_discover_server_trunking(struct nfs_client *clp,
 352                                    struct nfs_client **result,
 353                                    const struct cred *cred)
 354 {
 355         int status;
 356
 357         status = nfs4_proc_exchange_id(clp, cred);
 358         if (status != NFS4_OK)
 359                 return status;
 360
 361         status = nfs41_walk_client_list(clp, result, cred);
 362         if (status < 0)
 363                 return status;
 364         if (clp != *result)
 365                 return 0;
 366
 367         /*
 368          * Purge state if the client id was established in a prior
 369          * instance and the client id could not have arrived on the
 370          * server via Transparent State Migration.
 371          */
 372         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
 373                 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
 374                         set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 375                 else
 376                         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 377         }
 378         nfs4_schedule_state_manager(clp);
 379         status = nfs_wait_client_init_complete(clp);
 380         if (status < 0)
 381                 nfs_put_client(clp);
 382         return status;
 383 }
 384
 385 #endif /* CONFIG_NFS_V4_1 */
 386
 387 /**
 388  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 389  * @clp: client state handle
 390  *
 391  * Returns a cred with reference count bumped, or NULL.
 392  */
 393 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
 394 {
 395         const struct cred *cred;
 396
 397         cred = nfs4_get_machine_cred(clp);
 398         return cred;
 399 }
 400
 401 static struct nfs4_state_owner *
 402 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
 403 {
 404         struct rb_node **p = &server->state_owners.rb_node,
 405                        *parent = NULL;
 406         struct nfs4_state_owner *sp;
 407         int cmp;
 408
 409         while (*p != NULL) {
 410                 parent = *p;
 411                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 412                 cmp = cred_fscmp(cred, sp->so_cred);
 413
 414                 if (cmp < 0)
 415                         p = &parent->rb_left;
 416                 else if (cmp > 0)
 417                         p = &parent->rb_right;
 418                 else {
 419                         if (!list_empty(&sp->so_lru))
 420                                 list_del_init(&sp->so_lru);
 421                         atomic_inc(&sp->so_count);
 422                         return sp;
 423                 }
 424         }
 425         return NULL;
 426 }
 427
 428 static struct nfs4_state_owner *
 429 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 430 {
 431         struct nfs_server *server = new->so_server;
 432         struct rb_node **p = &server->state_owners.rb_node,
 433                        *parent = NULL;
 434         struct nfs4_state_owner *sp;
 435         int cmp;
 436
 437         while (*p != NULL) {
 438                 parent = *p;
 439                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 440                 cmp = cred_fscmp(new->so_cred, sp->so_cred);
 441
 442                 if (cmp < 0)
 443                         p = &parent->rb_left;
 444                 else if (cmp > 0)
 445                         p = &parent->rb_right;
 446                 else {
 447                         if (!list_empty(&sp->so_lru))
 448                                 list_del_init(&sp->so_lru);
 449                         atomic_inc(&sp->so_count);
 450                         return sp;
 451                 }
 452         }
 453         rb_link_node(&new->so_server_node, parent, p);
 454         rb_insert_color(&new->so_server_node, &server->state_owners);
 455         return new;
 456 }
 457
 458 static void
 459 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 460 {
 461         struct nfs_server *server = sp->so_server;
 462
 463         if (!RB_EMPTY_NODE(&sp->so_server_node))
 464                 rb_erase(&sp->so_server_node, &server->state_owners);
 465 }
 466
 467 static void
 468 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 469 {
 470         sc->create_time = ktime_get();
 471         sc->flags = 0;
 472         sc->counter = 0;
 473         spin_lock_init(&sc->lock);
 474         INIT_LIST_HEAD(&sc->list);
 475         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 476 }
 477
 478 static void
 479 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 480 {
 481         rpc_destroy_wait_queue(&sc->wait);
 482 }
 483
 484 /*
 485  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 486  * create a new state_owner.
 487  *
 488  */
 489 static struct nfs4_state_owner *
 490 nfs4_alloc_state_owner(struct nfs_server *server,
 491                 const struct cred *cred,
 492                 gfp_t gfp_flags)
 493 {
 494         struct nfs4_state_owner *sp;
 495
 496         sp = kzalloc(sizeof(*sp), gfp_flags);
 497         if (!sp)
 498                 return NULL;
 499         sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
 500                                                 gfp_flags);
 501         if (sp->so_seqid.owner_id < 0) {
 502                 kfree(sp);
 503                 return NULL;
 504         }
 505         sp->so_server = server;
 506         sp->so_cred = get_cred(cred);
 507         spin_lock_init(&sp->so_lock);
 508         INIT_LIST_HEAD(&sp->so_states);
 509         nfs4_init_seqid_counter(&sp->so_seqid);
 510         atomic_set(&sp->so_count, 1);
 511         INIT_LIST_HEAD(&sp->so_lru);
 512         seqcount_spinlock_init(&sp->so_reclaim_seqcount, &sp->so_lock);
 513         mutex_init(&sp->so_delegreturn_mutex);
 514         return sp;
 515 }
 516
 517 static void
 518 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
 519 {
 520         /* This state_owner is no longer usable, but must
 521          * remain in place so that state recovery can find it
 522          * and the opens associated with it.
 523          * It may also be used for new 'open' request to
 524          * return a delegation to the server.
 525          * So update the 'create_time' so that it looks like
 526          * a new state_owner.  This will cause the server to
 527          * request an OPEN_CONFIRM to start a new sequence.
 528          */
 529         sp->so_seqid.create_time = ktime_get();
 530 }
 531
 532 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 533 {
 534         nfs4_destroy_seqid_counter(&sp->so_seqid);
 535         put_cred(sp->so_cred);
 536         ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
 537         kfree(sp);
 538 }
 539
 540 static void nfs4_gc_state_owners(struct nfs_server *server)
 541 {
 542         struct nfs_client *clp = server->nfs_client;
 543         struct nfs4_state_owner *sp, *tmp;
 544         unsigned long time_min, time_max;
 545         LIST_HEAD(doomed);
 546
 547         spin_lock(&clp->cl_lock);
 548         time_max = jiffies;
 549         time_min = (long)time_max - (long)clp->cl_lease_time;
 550         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 551                 /* NB: LRU is sorted so that oldest is at the head */
 552                 if (time_in_range(sp->so_expires, time_min, time_max))
 553                         break;
 554                 list_move(&sp->so_lru, &doomed);
 555                 nfs4_remove_state_owner_locked(sp);
 556         }
 557         spin_unlock(&clp->cl_lock);
 558
 559         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 560                 list_del(&sp->so_lru);
 561                 nfs4_free_state_owner(sp);
 562         }
 563 }
 564
 565 /**
 566  * nfs4_get_state_owner - Look up a state owner given a credential
 567  * @server: nfs_server to search
 568  * @cred: RPC credential to match
 569  * @gfp_flags: allocation mode
 570  *
 571  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 572  */
 573 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 574                                               const struct cred *cred,
 575                                               gfp_t gfp_flags)
 576 {
 577         struct nfs_client *clp = server->nfs_client;
 578         struct nfs4_state_owner *sp, *new;
 579
 580         spin_lock(&clp->cl_lock);
 581         sp = nfs4_find_state_owner_locked(server, cred);
 582         spin_unlock(&clp->cl_lock);
 583         if (sp != NULL)
 584                 goto out;
 585         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 586         if (new == NULL)
 587                 goto out;
 588         spin_lock(&clp->cl_lock);
 589         sp = nfs4_insert_state_owner_locked(new);
 590         spin_unlock(&clp->cl_lock);
 591         if (sp != new)
 592                 nfs4_free_state_owner(new);
 593 out:
 594         nfs4_gc_state_owners(server);
 595         return sp;
 596 }
 597
 598 /**
 599  * nfs4_put_state_owner - Release a nfs4_state_owner
 600  * @sp: state owner data to release
 601  *
 602  * Note that we keep released state owners on an LRU
 603  * list.
 604  * This caches valid state owners so that they can be
 605  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 606  * It also pins the uniquifier of dropped state owners for
 607  * a while, to ensure that those state owner names are
 608  * never reused.
 609  */
 610 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 611 {
 612         struct nfs_server *server = sp->so_server;
 613         struct nfs_client *clp = server->nfs_client;
 614
 615         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 616                 return;
 617
 618         sp->so_expires = jiffies;
 619         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 620         spin_unlock(&clp->cl_lock);
 621 }
 622
 623 /**
 624  * nfs4_purge_state_owners - Release all cached state owners
 625  * @server: nfs_server with cached state owners to release
 626  * @head: resulting list of state owners
 627  *
 628  * Called at umount time.  Remaining state owners will be on
 629  * the LRU with ref count of zero.
 630  * Note that the state owners are not freed, but are added
 631  * to the list @head, which can later be used as an argument
 632  * to nfs4_free_state_owners.
 633  */
 634 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
 635 {
 636         struct nfs_client *clp = server->nfs_client;
 637         struct nfs4_state_owner *sp, *tmp;
 638
 639         spin_lock(&clp->cl_lock);
 640         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 641                 list_move(&sp->so_lru, head);
 642                 nfs4_remove_state_owner_locked(sp);
 643         }
 644         spin_unlock(&clp->cl_lock);
 645 }
 646
 647 /**
 648  * nfs4_purge_state_owners - Release all cached state owners
 649  * @head: resulting list of state owners
 650  *
 651  * Frees a list of state owners that was generated by
 652  * nfs4_purge_state_owners
 653  */
 654 void nfs4_free_state_owners(struct list_head *head)
 655 {
 656         struct nfs4_state_owner *sp, *tmp;
 657
 658         list_for_each_entry_safe(sp, tmp, head, so_lru) {
 659                 list_del(&sp->so_lru);
 660                 nfs4_free_state_owner(sp);
 661         }
 662 }
 663
 664 static struct nfs4_state *
 665 nfs4_alloc_open_state(void)
 666 {
 667         struct nfs4_state *state;
 668
 669         state = kzalloc(sizeof(*state), GFP_NOFS);
 670         if (!state)
 671                 return NULL;
 672         refcount_set(&state->count, 1);
 673         INIT_LIST_HEAD(&state->lock_states);
 674         spin_lock_init(&state->state_lock);
 675         seqlock_init(&state->seqlock);
 676         init_waitqueue_head(&state->waitq);
 677         return state;
 678 }
 679
 680 void
 681 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 682 {
 683         if (state->state == fmode)
 684                 return;
 685         /* NB! List reordering - see the reclaim code for why.  */
 686         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 687                 if (fmode & FMODE_WRITE)
 688                         list_move(&state->open_states, &state->owner->so_states);
 689                 else
 690                         list_move_tail(&state->open_states, &state->owner->so_states);
 691         }
 692         state->state = fmode;
 693 }
 694
 695 static struct nfs4_state *
 696 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 697 {
 698         struct nfs_inode *nfsi = NFS_I(inode);
 699         struct nfs4_state *state;
 700
 701         list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
 702                 if (state->owner != owner)
 703                         continue;
 704                 if (!nfs4_valid_open_stateid(state))
 705                         continue;
 706                 if (refcount_inc_not_zero(&state->count))
 707                         return state;
 708         }
 709         return NULL;
 710 }
 711
 712 static void
 713 nfs4_free_open_state(struct nfs4_state *state)
 714 {
 715         kfree_rcu(state, rcu_head);
 716 }
 717
 718 struct nfs4_state *
 719 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 720 {
 721         struct nfs4_state *state, *new;
 722         struct nfs_inode *nfsi = NFS_I(inode);
 723
 724         rcu_read_lock();
 725         state = __nfs4_find_state_byowner(inode, owner);
 726         rcu_read_unlock();
 727         if (state)
 728                 goto out;
 729         new = nfs4_alloc_open_state();
 730         spin_lock(&owner->so_lock);
 731         spin_lock(&inode->i_lock);
 732         state = __nfs4_find_state_byowner(inode, owner);
 733         if (state == NULL && new != NULL) {
 734                 state = new;
 735                 state->owner = owner;
 736                 atomic_inc(&owner->so_count);
 737                 ihold(inode);
 738                 state->inode = inode;
 739                 list_add_rcu(&state->inode_states, &nfsi->open_states);
 740                 spin_unlock(&inode->i_lock);
 741                 /* Note: The reclaim code dictates that we add stateless
 742                  * and read-only stateids to the end of the list */
 743                 list_add_tail(&state->open_states, &owner->so_states);
 744                 spin_unlock(&owner->so_lock);
 745         } else {
 746                 spin_unlock(&inode->i_lock);
 747                 spin_unlock(&owner->so_lock);
 748                 if (new)
 749                         nfs4_free_open_state(new);
 750         }
 751 out:
 752         return state;
 753 }
 754
 755 void nfs4_put_open_state(struct nfs4_state *state)
 756 {
 757         struct inode *inode = state->inode;
 758         struct nfs4_state_owner *owner = state->owner;
 759
 760         if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
 761                 return;
 762         spin_lock(&inode->i_lock);
 763         list_del_rcu(&state->inode_states);
 764         list_del(&state->open_states);
 765         spin_unlock(&inode->i_lock);
 766         spin_unlock(&owner->so_lock);
 767         nfs4_inode_return_delegation_on_close(inode);
 768         iput(inode);
 769         nfs4_free_open_state(state);
 770         nfs4_put_state_owner(owner);
 771 }
 772
 773 /*
 774  * Close the current file.
 775  */
 776 static void __nfs4_close(struct nfs4_state *state,
 777                 fmode_t fmode, gfp_t gfp_mask, int wait)
 778 {
 779         struct nfs4_state_owner *owner = state->owner;
 780         int call_close = 0;
 781         fmode_t newstate;
 782
 783         atomic_inc(&owner->so_count);
 784         /* Protect against nfs4_find_state() */
 785         spin_lock(&owner->so_lock);
 786         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 787                 case FMODE_READ:
 788                         state->n_rdonly--;
 789                         break;
 790                 case FMODE_WRITE:
 791                         state->n_wronly--;
 792                         break;
 793                 case FMODE_READ|FMODE_WRITE:
 794                         state->n_rdwr--;
 795         }
 796         newstate = FMODE_READ|FMODE_WRITE;
 797         if (state->n_rdwr == 0) {
 798                 if (state->n_rdonly == 0) {
 799                         newstate &= ~FMODE_READ;
 800                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 801                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 802                 }
 803                 if (state->n_wronly == 0) {
 804                         newstate &= ~FMODE_WRITE;
 805                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 806                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 807                 }
 808                 if (newstate == 0)
 809                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 810         }
 811         nfs4_state_set_mode_locked(state, newstate);
 812         spin_unlock(&owner->so_lock);
 813
 814         if (!call_close) {
 815                 nfs4_put_open_state(state);
 816                 nfs4_put_state_owner(owner);
 817         } else
 818                 nfs4_do_close(state, gfp_mask, wait);
 819 }
 820
 821 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 822 {
 823         __nfs4_close(state, fmode, GFP_NOFS, 0);
 824 }
 825
 826 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 827 {
 828         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 829 }
 830
 831 /*
 832  * Search the state->lock_states for an existing lock_owner
 833  * that is compatible with either of the given owners.
 834  * If the second is non-zero, then the first refers to a Posix-lock
 835  * owner (current->files) and the second refers to a flock/OFD
 836  * owner (struct file*).  In that case, prefer a match for the first
 837  * owner.
 838  * If both sorts of locks are held on the one file we cannot know
 839  * which stateid was intended to be used, so a "correct" choice cannot
 840  * be made.  Failing that, a "consistent" choice is preferable.  The
 841  * consistent choice we make is to prefer the first owner, that of a
 842  * Posix lock.
 843  */
 844 static struct nfs4_lock_state *
 845 __nfs4_find_lock_state(struct nfs4_state *state,
 846                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 847 {
 848         struct nfs4_lock_state *pos, *ret = NULL;
 849         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 850                 if (pos->ls_owner == fl_owner) {
 851                         ret = pos;
 852                         break;
 853                 }
 854                 if (pos->ls_owner == fl_owner2)
 855                         ret = pos;
 856         }
 857         if (ret)
 858                 refcount_inc(&ret->ls_count);
 859         return ret;
 860 }
 861
 862 /*
 863  * Return a compatible lock_state. If no initialized lock_state structure
 864  * exists, return an uninitialized one.
 865  *
 866  */
 867 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 868 {
 869         struct nfs4_lock_state *lsp;
 870         struct nfs_server *server = state->owner->so_server;
 871
 872         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 873         if (lsp == NULL)
 874                 return NULL;
 875         nfs4_init_seqid_counter(&lsp->ls_seqid);
 876         refcount_set(&lsp->ls_count, 1);
 877         lsp->ls_state = state;
 878         lsp->ls_owner = fl_owner;
 879         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 880         if (lsp->ls_seqid.owner_id < 0)
 881                 goto out_free;
 882         INIT_LIST_HEAD(&lsp->ls_locks);
 883         return lsp;
 884 out_free:
 885         kfree(lsp);
 886         return NULL;
 887 }
 888
 889 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 890 {
 891         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 892         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 893         kfree(lsp);
 894 }
 895
 896 /*
 897  * Return a compatible lock_state. If no initialized lock_state structure
 898  * exists, return an uninitialized one.
 899  *
 900  */
 901 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 902 {
 903         struct nfs4_lock_state *lsp, *new = NULL;
 904
 905         for(;;) {
 906                 spin_lock(&state->state_lock);
 907                 lsp = __nfs4_find_lock_state(state, owner, NULL);
 908                 if (lsp != NULL)
 909                         break;
 910                 if (new != NULL) {
 911                         list_add(&new->ls_locks, &state->lock_states);
 912                         set_bit(LK_STATE_IN_USE, &state->flags);
 913                         lsp = new;
 914                         new = NULL;
 915                         break;
 916                 }
 917                 spin_unlock(&state->state_lock);
 918                 new = nfs4_alloc_lock_state(state, owner);
 919                 if (new == NULL)
 920                         return NULL;
 921         }
 922         spin_unlock(&state->state_lock);
 923         if (new != NULL)
 924                 nfs4_free_lock_state(state->owner->so_server, new);
 925         return lsp;
 926 }
 927
 928 /*
 929  * Release reference to lock_state, and free it if we see that
 930  * it is no longer in use
 931  */
 932 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 933 {
 934         struct nfs_server *server;
 935         struct nfs4_state *state;
 936
 937         if (lsp == NULL)
 938                 return;
 939         state = lsp->ls_state;
 940         if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
 941                 return;
 942         list_del(&lsp->ls_locks);
 943         if (list_empty(&state->lock_states))
 944                 clear_bit(LK_STATE_IN_USE, &state->flags);
 945         spin_unlock(&state->state_lock);
 946         server = state->owner->so_server;
 947         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 948                 struct nfs_client *clp = server->nfs_client;
 949
 950                 clp->cl_mvops->free_lock_state(server, lsp);
 951         } else
 952                 nfs4_free_lock_state(server, lsp);
 953 }
 954
 955 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 956 {
 957         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 958
 959         dst->fl_u.nfs4_fl.owner = lsp;
 960         refcount_inc(&lsp->ls_count);
 961 }
 962
 963 static void nfs4_fl_release_lock(struct file_lock *fl)
 964 {
 965         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 966 }
 967
 968 static const struct file_lock_operations nfs4_fl_lock_ops = {
 969         .fl_copy_lock = nfs4_fl_copy_lock,
 970         .fl_release_private = nfs4_fl_release_lock,
 971 };
 972
 973 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 974 {
 975         struct nfs4_lock_state *lsp;
 976
 977         if (fl->fl_ops != NULL)
 978                 return 0;
 979         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 980         if (lsp == NULL)
 981                 return -ENOMEM;
 982         fl->fl_u.nfs4_fl.owner = lsp;
 983         fl->fl_ops = &nfs4_fl_lock_ops;
 984         return 0;
 985 }
 986
 987 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 988                 struct nfs4_state *state,
 989                 const struct nfs_lock_context *l_ctx)
 990 {
 991         struct nfs4_lock_state *lsp;
 992         fl_owner_t fl_owner, fl_flock_owner;
 993         int ret = -ENOENT;
 994
 995         if (l_ctx == NULL)
 996                 goto out;
 997
 998         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 999                 goto out;
1000
1001         fl_owner = l_ctx->lockowner;
1002         fl_flock_owner = l_ctx->open_context->flock_owner;
1003
1004         spin_lock(&state->state_lock);
1005         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1006         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1007                 ret = -EIO;
1008         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1009                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1010                 ret = 0;
1011         }
1012         spin_unlock(&state->state_lock);
1013         nfs4_put_lock_state(lsp);
1014 out:
1015         return ret;
1016 }
1017
1018 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1019 {
1020         bool ret;
1021         const nfs4_stateid *src;
1022         int seq;
1023
1024         do {
1025                 ret = false;
1026                 src = &zero_stateid;
1027                 seq = read_seqbegin(&state->seqlock);
1028                 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1029                         src = &state->open_stateid;
1030                         ret = true;
1031                 }
1032                 nfs4_stateid_copy(dst, src);
1033         } while (read_seqretry(&state->seqlock, seq));
1034         return ret;
1035 }
1036
1037 /*
1038  * Byte-range lock aware utility to initialize the stateid of read/write
1039  * requests.
1040  */
1041 int nfs4_select_rw_stateid(struct nfs4_state *state,
1042                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1043                 nfs4_stateid *dst, const struct cred **cred)
1044 {
1045         int ret;
1046
1047         if (!nfs4_valid_open_stateid(state))
1048                 return -EIO;
1049         if (cred != NULL)
1050                 *cred = NULL;
1051         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1052         if (ret == -EIO)
1053                 /* A lost lock - don't even consider delegations */
1054                 goto out;
1055         /* returns true if delegation stateid found and copied */
1056         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1057                 ret = 0;
1058                 goto out;
1059         }
1060         if (ret != -ENOENT)
1061                 /* nfs4_copy_delegation_stateid() didn't over-write
1062                  * dst, so it still has the lock stateid which we now
1063                  * choose to use.
1064                  */
1065                 goto out;
1066         ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1067 out:
1068         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1069                 dst->seqid = 0;
1070         return ret;
1071 }
1072
1073 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1074 {
1075         struct nfs_seqid *new;
1076
1077         new = kmalloc(sizeof(*new), gfp_mask);
1078         if (new == NULL)
1079                 return ERR_PTR(-ENOMEM);
1080         new->sequence = counter;
1081         INIT_LIST_HEAD(&new->list);
1082         new->task = NULL;
1083         return new;
1084 }
1085
1086 void nfs_release_seqid(struct nfs_seqid *seqid)
1087 {
1088         struct nfs_seqid_counter *sequence;
1089
1090         if (seqid == NULL || list_empty(&seqid->list))
1091                 return;
1092         sequence = seqid->sequence;
1093         spin_lock(&sequence->lock);
1094         list_del_init(&seqid->list);
1095         if (!list_empty(&sequence->list)) {
1096                 struct nfs_seqid *next;
1097
1098                 next = list_first_entry(&sequence->list,
1099                                 struct nfs_seqid, list);
1100                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1101         }
1102         spin_unlock(&sequence->lock);
1103 }
1104
1105 void nfs_free_seqid(struct nfs_seqid *seqid)
1106 {
1107         nfs_release_seqid(seqid);
1108         kfree(seqid);
1109 }
1110
1111 /*
1112  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1113  * failed with a seqid incrementing error -
1114  * see comments nfs4.h:seqid_mutating_error()
1115  */
1116 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1117 {
1118         switch (status) {
1119                 case 0:
1120                         break;
1121                 case -NFS4ERR_BAD_SEQID:
1122                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1123                                 return;
1124                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1125                                         " sequence-id error on an"
1126                                         " unconfirmed sequence %p!\n",
1127                                         seqid->sequence);
1128                         return;
1129                 case -NFS4ERR_STALE_CLIENTID:
1130                 case -NFS4ERR_STALE_STATEID:
1131                 case -NFS4ERR_BAD_STATEID:
1132                 case -NFS4ERR_BADXDR:
1133                 case -NFS4ERR_RESOURCE:
1134                 case -NFS4ERR_NOFILEHANDLE:
1135                 case -NFS4ERR_MOVED:
1136                         /* Non-seqid mutating errors */
1137                         return;
1138         }
1139         /*
1140          * Note: no locking needed as we are guaranteed to be first
1141          * on the sequence list
1142          */
1143         seqid->sequence->counter++;
1144 }
1145
1146 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1147 {
1148         struct nfs4_state_owner *sp;
1149
1150         if (seqid == NULL)
1151                 return;
1152
1153         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1154         if (status == -NFS4ERR_BAD_SEQID)
1155                 nfs4_reset_state_owner(sp);
1156         if (!nfs4_has_session(sp->so_server->nfs_client))
1157                 nfs_increment_seqid(status, seqid);
1158 }
1159
1160 /*
1161  * Increment the seqid if the LOCK/LOCKU succeeded, or
1162  * failed with a seqid incrementing error -
1163  * see comments nfs4.h:seqid_mutating_error()
1164  */
1165 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1166 {
1167         if (seqid != NULL)
1168                 nfs_increment_seqid(status, seqid);
1169 }
1170
1171 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1172 {
1173         struct nfs_seqid_counter *sequence;
1174         int status = 0;
1175
1176         if (seqid == NULL)
1177                 goto out;
1178         sequence = seqid->sequence;
1179         spin_lock(&sequence->lock);
1180         seqid->task = task;
1181         if (list_empty(&seqid->list))
1182                 list_add_tail(&seqid->list, &sequence->list);
1183         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1184                 goto unlock;
1185         rpc_sleep_on(&sequence->wait, task, NULL);
1186         status = -EAGAIN;
1187 unlock:
1188         spin_unlock(&sequence->lock);
1189 out:
1190         return status;
1191 }
1192
1193 static int nfs4_run_state_manager(void *);
1194
1195 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1196 {
1197         smp_mb__before_atomic();
1198         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1199         smp_mb__after_atomic();
1200         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
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
1212         set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1213         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1214                 return;
1215         __module_get(THIS_MODULE);
1216         refcount_inc(&clp->cl_count);
1217
1218         /* The rcu_read_lock() is not strictly necessary, as the state
1219          * manager is the only thread that ever changes the rpc_xprt
1220          * after it's initialized.  At this point, we're single threaded. */
1221         rcu_read_lock();
1222         snprintf(buf, sizeof(buf), "%s-manager",
1223                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1224         rcu_read_unlock();
1225         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1226         if (IS_ERR(task)) {
1227                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1228                         __func__, PTR_ERR(task));
1229                 nfs4_clear_state_manager_bit(clp);
1230                 nfs_put_client(clp);
1231                 module_put(THIS_MODULE);
1232         }
1233 }
1234
1235 /*
1236  * Schedule a lease recovery attempt
1237  */
1238 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1239 {
1240         if (!clp)
1241                 return;
1242         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1243                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1244         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1245                         clp->cl_hostname);
1246         nfs4_schedule_state_manager(clp);
1247 }
1248 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1249
1250 /**
1251  * nfs4_schedule_migration_recovery - trigger migration recovery
1252  *
1253  * @server: FSID that is migrating
1254  *
1255  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1256  * value is returned.
1257  */
1258 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1259 {
1260         struct nfs_client *clp = server->nfs_client;
1261
1262         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1263                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1264                                 clp->cl_hostname);
1265                 return -NFS4ERR_IO;
1266         }
1267
1268         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1269                 return -NFS4ERR_IO;
1270
1271         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1272                         __func__,
1273                         (unsigned long long)server->fsid.major,
1274                         (unsigned long long)server->fsid.minor,
1275                         clp->cl_hostname);
1276
1277         set_bit(NFS_MIG_IN_TRANSITION,
1278                         &((struct nfs_server *)server)->mig_status);
1279         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1280
1281         nfs4_schedule_state_manager(clp);
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1285
1286 /**
1287  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1288  *
1289  * @clp: server to check for moved leases
1290  *
1291  */
1292 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1293 {
1294         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1295                 __func__, clp->cl_clientid, clp->cl_hostname);
1296
1297         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1298         nfs4_schedule_state_manager(clp);
1299 }
1300 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1301
1302 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1303 {
1304         int res;
1305
1306         might_sleep();
1307
1308         refcount_inc(&clp->cl_count);
1309         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1310                                  nfs_wait_bit_killable, TASK_KILLABLE);
1311         if (res)
1312                 goto out;
1313         if (clp->cl_cons_state < 0)
1314                 res = clp->cl_cons_state;
1315 out:
1316         nfs_put_client(clp);
1317         return res;
1318 }
1319
1320 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1321 {
1322         unsigned int loop;
1323         int ret;
1324
1325         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1326                 ret = nfs4_wait_clnt_recover(clp);
1327                 if (ret != 0)
1328                         break;
1329                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1330                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1331                         break;
1332                 nfs4_schedule_state_manager(clp);
1333                 ret = -EIO;
1334         }
1335         return ret;
1336 }
1337
1338 /*
1339  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1340  * @clp: client to process
1341  *
1342  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1343  * resend of the SETCLIENTID and hence re-establish the
1344  * callback channel. Then return all existing delegations.
1345  */
1346 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1347 {
1348         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1349         nfs_expire_all_delegations(clp);
1350         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1351                         clp->cl_hostname);
1352 }
1353
1354 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1355 {
1356         nfs40_handle_cb_pathdown(clp);
1357         nfs4_schedule_state_manager(clp);
1358 }
1359
1360 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1361 {
1362
1363         if (!nfs4_valid_open_stateid(state))
1364                 return 0;
1365         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1366         /* Don't recover state that expired before the reboot */
1367         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1368                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1369                 return 0;
1370         }
1371         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1372         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1373         return 1;
1374 }
1375
1376 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1377 {
1378         if (!nfs4_valid_open_stateid(state))
1379                 return 0;
1380         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1381         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1382         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1383         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1384         return 1;
1385 }
1386
1387 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1388 {
1389         struct nfs_client *clp = server->nfs_client;
1390
1391         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1392                 return -EBADF;
1393         nfs_inode_find_delegation_state_and_recover(state->inode,
1394                         &state->stateid);
1395         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1396                         clp->cl_hostname);
1397         nfs4_schedule_state_manager(clp);
1398         return 0;
1399 }
1400 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1401
1402 static struct nfs4_lock_state *
1403 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1404                 const nfs4_stateid *stateid)
1405 {
1406         struct nfs4_lock_state *pos;
1407
1408         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1409                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1410                         continue;
1411                 if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1412                         return pos;
1413         }
1414         return NULL;
1415 }
1416
1417 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1418                 const nfs4_stateid *stateid)
1419 {
1420         bool found = false;
1421
1422         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1423                 spin_lock(&state->state_lock);
1424                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1425                         found = true;
1426                 spin_unlock(&state->state_lock);
1427         }
1428         return found;
1429 }
1430
1431 void nfs_inode_find_state_and_recover(struct inode *inode,
1432                 const nfs4_stateid *stateid)
1433 {
1434         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1435         struct nfs_inode *nfsi = NFS_I(inode);
1436         struct nfs_open_context *ctx;
1437         struct nfs4_state *state;
1438         bool found = false;
1439
1440         rcu_read_lock();
1441         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1442                 state = ctx->state;
1443                 if (state == NULL)
1444                         continue;
1445                 if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1446                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1447                         found = true;
1448                         continue;
1449                 }
1450                 if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1451                     nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1452                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1453                         found = true;
1454                         continue;
1455                 }
1456                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1457                     nfs4_state_mark_reclaim_nograce(clp, state))
1458                         found = true;
1459         }
1460         rcu_read_unlock();
1461
1462         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1463         if (found)
1464                 nfs4_schedule_state_manager(clp);
1465 }
1466
1467 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1468 {
1469         struct inode *inode = state->inode;
1470         struct nfs_inode *nfsi = NFS_I(inode);
1471         struct nfs_open_context *ctx;
1472
1473         rcu_read_lock();
1474         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1475                 if (ctx->state != state)
1476                         continue;
1477                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1478                 pr_warn("NFSv4: state recovery failed for open file %pd2, "
1479                                 "error = %d\n", ctx->dentry, err);
1480         }
1481         rcu_read_unlock();
1482 }
1483
1484 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1485 {
1486         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1487         nfs4_state_mark_open_context_bad(state, error);
1488 }
1489
1490
1491 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1492 {
1493         struct inode *inode = state->inode;
1494         struct nfs_inode *nfsi = NFS_I(inode);
1495         struct file_lock *fl;
1496         struct nfs4_lock_state *lsp;
1497         int status = 0;
1498         struct file_lock_context *flctx = inode->i_flctx;
1499         struct list_head *list;
1500
1501         if (flctx == NULL)
1502                 return 0;
1503
1504         list = &flctx->flc_posix;
1505
1506         /* Guard against delegation returns and new lock/unlock calls */
1507         down_write(&nfsi->rwsem);
1508         spin_lock(&flctx->flc_lock);
1509 restart:
1510         list_for_each_entry(fl, list, fl_list) {
1511                 if (nfs_file_open_context(fl->fl_file)->state != state)
1512                         continue;
1513                 spin_unlock(&flctx->flc_lock);
1514                 status = ops->recover_lock(state, fl);
1515                 switch (status) {
1516                 case 0:
1517                         break;
1518                 case -ETIMEDOUT:
1519                 case -ESTALE:
1520                 case -NFS4ERR_ADMIN_REVOKED:
1521                 case -NFS4ERR_STALE_STATEID:
1522                 case -NFS4ERR_BAD_STATEID:
1523                 case -NFS4ERR_EXPIRED:
1524                 case -NFS4ERR_NO_GRACE:
1525                 case -NFS4ERR_STALE_CLIENTID:
1526                 case -NFS4ERR_BADSESSION:
1527                 case -NFS4ERR_BADSLOT:
1528                 case -NFS4ERR_BAD_HIGH_SLOT:
1529                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1530                         goto out;
1531                 default:
1532                         pr_err("NFS: %s: unhandled error %d\n",
1533                                         __func__, status);
1534                         fallthrough;
1535                 case -ENOMEM:
1536                 case -NFS4ERR_DENIED:
1537                 case -NFS4ERR_RECLAIM_BAD:
1538                 case -NFS4ERR_RECLAIM_CONFLICT:
1539                         lsp = fl->fl_u.nfs4_fl.owner;
1540                         if (lsp)
1541                                 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1542                         status = 0;
1543                 }
1544                 spin_lock(&flctx->flc_lock);
1545         }
1546         if (list == &flctx->flc_posix) {
1547                 list = &flctx->flc_flock;
1548                 goto restart;
1549         }
1550         spin_unlock(&flctx->flc_lock);
1551 out:
1552         up_write(&nfsi->rwsem);
1553         return status;
1554 }
1555
1556 #ifdef CONFIG_NFS_V4_2
1557 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1558 {
1559         struct nfs4_copy_state *copy;
1560
1561         if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1562                 !test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1563                 return;
1564
1565         spin_lock(&sp->so_server->nfs_client->cl_lock);
1566         list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1567                 if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1568                                 !nfs4_stateid_match_other(&state->stateid,
1569                                 &copy->parent_dst_state->stateid)))
1570                                 continue;
1571                 copy->flags = 1;
1572                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1573                                 &state->flags)) {
1574                         clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1575                         complete(&copy->completion);
1576                 }
1577         }
1578         list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1579                 if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1580                                 !nfs4_stateid_match_other(&state->stateid,
1581                                 &copy->parent_src_state->stateid)))
1582                                 continue;
1583                 copy->flags = 1;
1584                 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1585                                 &state->flags))
1586                         complete(&copy->completion);
1587         }
1588         spin_unlock(&sp->so_server->nfs_client->cl_lock);
1589 }
1590 #else /* !CONFIG_NFS_V4_2 */
1591 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1592                                          struct nfs4_state *state)
1593 {
1594 }
1595 #endif /* CONFIG_NFS_V4_2 */
1596
1597 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1598                                      const struct nfs4_state_recovery_ops *ops)
1599 {
1600         struct nfs4_lock_state *lock;
1601         int status;
1602
1603         status = ops->recover_open(sp, state);
1604         if (status < 0)
1605                 return status;
1606
1607         status = nfs4_reclaim_locks(state, ops);
1608         if (status < 0)
1609                 return status;
1610
1611         if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1612                 spin_lock(&state->state_lock);
1613                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1614                         trace_nfs4_state_lock_reclaim(state, lock);
1615                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1616                                 pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1617                 }
1618                 spin_unlock(&state->state_lock);
1619         }
1620
1621         nfs42_complete_copies(sp, state);
1622         clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1623         return status;
1624 }
1625
1626 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1627 {
1628         struct nfs4_state *state;
1629         unsigned int loop = 0;
1630         int status = 0;
1631 #ifdef CONFIG_NFS_V4_2
1632         bool found_ssc_copy_state = false;
1633 #endif /* CONFIG_NFS_V4_2 */
1634
1635         /* Note: we rely on the sp->so_states list being ordered
1636          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1637          * states first.
1638          * This is needed to ensure that the server won't give us any
1639          * read delegations that we have to return if, say, we are
1640          * recovering after a network partition or a reboot from a
1641          * server that doesn't support a grace period.
1642          */
1643         spin_lock(&sp->so_lock);
1644         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1645 restart:
1646         list_for_each_entry(state, &sp->so_states, open_states) {
1647                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1648                         continue;
1649                 if (!nfs4_valid_open_stateid(state))
1650                         continue;
1651                 if (state->state == 0)
1652                         continue;
1653 #ifdef CONFIG_NFS_V4_2
1654                 if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1655                         nfs4_state_mark_recovery_failed(state, -EIO);
1656                         found_ssc_copy_state = true;
1657                         continue;
1658                 }
1659 #endif /* CONFIG_NFS_V4_2 */
1660                 refcount_inc(&state->count);
1661                 spin_unlock(&sp->so_lock);
1662                 status = __nfs4_reclaim_open_state(sp, state, ops);
1663
1664                 switch (status) {
1665                 default:
1666                         if (status >= 0) {
1667                                 loop = 0;
1668                                 break;
1669                         }
1670                         printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1671                         fallthrough;
1672                 case -ENOENT:
1673                 case -ENOMEM:
1674                 case -EACCES:
1675                 case -EROFS:
1676                 case -EIO:
1677                 case -ESTALE:
1678                         /* Open state on this file cannot be recovered */
1679                         nfs4_state_mark_recovery_failed(state, status);
1680                         break;
1681                 case -EAGAIN:
1682                         ssleep(1);
1683                         if (loop++ < 10) {
1684                                 set_bit(ops->state_flag_bit, &state->flags);
1685                                 break;
1686                         }
1687                         fallthrough;
1688                 case -NFS4ERR_ADMIN_REVOKED:
1689                 case -NFS4ERR_STALE_STATEID:
1690                 case -NFS4ERR_OLD_STATEID:
1691                 case -NFS4ERR_BAD_STATEID:
1692                 case -NFS4ERR_RECLAIM_BAD:
1693                 case -NFS4ERR_RECLAIM_CONFLICT:
1694                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1695                         break;
1696                 case -NFS4ERR_EXPIRED:
1697                 case -NFS4ERR_NO_GRACE:
1698                         nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1699                         fallthrough;
1700                 case -NFS4ERR_STALE_CLIENTID:
1701                 case -NFS4ERR_BADSESSION:
1702                 case -NFS4ERR_BADSLOT:
1703                 case -NFS4ERR_BAD_HIGH_SLOT:
1704                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1705                 case -ETIMEDOUT:
1706                         goto out_err;
1707                 }
1708                 nfs4_put_open_state(state);
1709                 spin_lock(&sp->so_lock);
1710                 goto restart;
1711         }
1712         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1713         spin_unlock(&sp->so_lock);
1714 #ifdef CONFIG_NFS_V4_2
1715         if (found_ssc_copy_state)
1716                 return -EIO;
1717 #endif /* CONFIG_NFS_V4_2 */
1718         return 0;
1719 out_err:
1720         nfs4_put_open_state(state);
1721         spin_lock(&sp->so_lock);
1722         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1723         spin_unlock(&sp->so_lock);
1724         return status;
1725 }
1726
1727 static void nfs4_clear_open_state(struct nfs4_state *state)
1728 {
1729         struct nfs4_lock_state *lock;
1730
1731         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1732         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1733         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1734         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1735         spin_lock(&state->state_lock);
1736         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1737                 lock->ls_seqid.flags = 0;
1738                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1739         }
1740         spin_unlock(&state->state_lock);
1741 }
1742
1743 static void nfs4_reset_seqids(struct nfs_server *server,
1744         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1745 {
1746         struct nfs_client *clp = server->nfs_client;
1747         struct nfs4_state_owner *sp;
1748         struct rb_node *pos;
1749         struct nfs4_state *state;
1750
1751         spin_lock(&clp->cl_lock);
1752         for (pos = rb_first(&server->state_owners);
1753              pos != NULL;
1754              pos = rb_next(pos)) {
1755                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1756                 sp->so_seqid.flags = 0;
1757                 spin_lock(&sp->so_lock);
1758                 list_for_each_entry(state, &sp->so_states, open_states) {
1759                         if (mark_reclaim(clp, state))
1760                                 nfs4_clear_open_state(state);
1761                 }
1762                 spin_unlock(&sp->so_lock);
1763         }
1764         spin_unlock(&clp->cl_lock);
1765 }
1766
1767 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1768         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1769 {
1770         struct nfs_server *server;
1771
1772         rcu_read_lock();
1773         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1774                 nfs4_reset_seqids(server, mark_reclaim);
1775         rcu_read_unlock();
1776 }
1777
1778 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1779 {
1780         /* Mark all delegations for reclaim */
1781         nfs_delegation_mark_reclaim(clp);
1782         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1783 }
1784
1785 static int nfs4_reclaim_complete(struct nfs_client *clp,
1786                                  const struct nfs4_state_recovery_ops *ops,
1787                                  const struct cred *cred)
1788 {
1789         /* Notify the server we're done reclaiming our state */
1790         if (ops->reclaim_complete)
1791                 return ops->reclaim_complete(clp, cred);
1792         return 0;
1793 }
1794
1795 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1796 {
1797         struct nfs_client *clp = server->nfs_client;
1798         struct nfs4_state_owner *sp;
1799         struct rb_node *pos;
1800         struct nfs4_state *state;
1801
1802         spin_lock(&clp->cl_lock);
1803         for (pos = rb_first(&server->state_owners);
1804              pos != NULL;
1805              pos = rb_next(pos)) {
1806                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1807                 spin_lock(&sp->so_lock);
1808                 list_for_each_entry(state, &sp->so_states, open_states) {
1809                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1810                                                 &state->flags))
1811                                 continue;
1812                         nfs4_state_mark_reclaim_nograce(clp, state);
1813                 }
1814                 spin_unlock(&sp->so_lock);
1815         }
1816         spin_unlock(&clp->cl_lock);
1817 }
1818
1819 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1820 {
1821         struct nfs_server *server;
1822
1823         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1824                 return 0;
1825
1826         rcu_read_lock();
1827         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1828                 nfs4_clear_reclaim_server(server);
1829         rcu_read_unlock();
1830
1831         nfs_delegation_reap_unclaimed(clp);
1832         return 1;
1833 }
1834
1835 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1836 {
1837         const struct nfs4_state_recovery_ops *ops;
1838         const struct cred *cred;
1839         int err;
1840
1841         if (!nfs4_state_clear_reclaim_reboot(clp))
1842                 return;
1843         ops = clp->cl_mvops->reboot_recovery_ops;
1844         cred = nfs4_get_clid_cred(clp);
1845         err = nfs4_reclaim_complete(clp, ops, cred);
1846         put_cred(cred);
1847         if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1848                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1849 }
1850
1851 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1852 {
1853         nfs_mark_test_expired_all_delegations(clp);
1854         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1855 }
1856
1857 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1858 {
1859         switch (error) {
1860         case 0:
1861                 break;
1862         case -NFS4ERR_CB_PATH_DOWN:
1863                 nfs40_handle_cb_pathdown(clp);
1864                 break;
1865         case -NFS4ERR_NO_GRACE:
1866                 nfs4_state_end_reclaim_reboot(clp);
1867                 break;
1868         case -NFS4ERR_STALE_CLIENTID:
1869                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1870                 nfs4_state_start_reclaim_reboot(clp);
1871                 break;
1872         case -NFS4ERR_EXPIRED:
1873                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1874                 nfs4_state_start_reclaim_nograce(clp);
1875                 break;
1876         case -NFS4ERR_BADSESSION:
1877         case -NFS4ERR_BADSLOT:
1878         case -NFS4ERR_BAD_HIGH_SLOT:
1879         case -NFS4ERR_DEADSESSION:
1880         case -NFS4ERR_SEQ_FALSE_RETRY:
1881         case -NFS4ERR_SEQ_MISORDERED:
1882                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1883                 /* Zero session reset errors */
1884                 break;
1885         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1886                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1887                 break;
1888         default:
1889                 dprintk("%s: failed to handle error %d for server %s\n",
1890                                 __func__, error, clp->cl_hostname);
1891                 return error;
1892         }
1893         dprintk("%s: handled error %d for server %s\n", __func__, error,
1894                         clp->cl_hostname);
1895         return 0;
1896 }
1897
1898 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1899 {
1900         struct nfs4_state_owner *sp;
1901         struct nfs_server *server;
1902         struct rb_node *pos;
1903         LIST_HEAD(freeme);
1904         int status = 0;
1905
1906 restart:
1907         rcu_read_lock();
1908         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1909                 nfs4_purge_state_owners(server, &freeme);
1910                 spin_lock(&clp->cl_lock);
1911                 for (pos = rb_first(&server->state_owners);
1912                      pos != NULL;
1913                      pos = rb_next(pos)) {
1914                         sp = rb_entry(pos,
1915                                 struct nfs4_state_owner, so_server_node);
1916                         if (!test_and_clear_bit(ops->owner_flag_bit,
1917                                                         &sp->so_flags))
1918                                 continue;
1919                         if (!atomic_inc_not_zero(&sp->so_count))
1920                                 continue;
1921                         spin_unlock(&clp->cl_lock);
1922                         rcu_read_unlock();
1923
1924                         status = nfs4_reclaim_open_state(sp, ops);
1925                         if (status < 0) {
1926                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1927                                 nfs4_put_state_owner(sp);
1928                                 status = nfs4_recovery_handle_error(clp, status);
1929                                 return (status != 0) ? status : -EAGAIN;
1930                         }
1931
1932                         nfs4_put_state_owner(sp);
1933                         goto restart;
1934                 }
1935                 spin_unlock(&clp->cl_lock);
1936         }
1937         rcu_read_unlock();
1938         nfs4_free_state_owners(&freeme);
1939         return 0;
1940 }
1941
1942 static int nfs4_check_lease(struct nfs_client *clp)
1943 {
1944         const struct cred *cred;
1945         const struct nfs4_state_maintenance_ops *ops =
1946                 clp->cl_mvops->state_renewal_ops;
1947         int status;
1948
1949         /* Is the client already known to have an expired lease? */
1950         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1951                 return 0;
1952         cred = ops->get_state_renewal_cred(clp);
1953         if (cred == NULL) {
1954                 cred = nfs4_get_clid_cred(clp);
1955                 status = -ENOKEY;
1956                 if (cred == NULL)
1957                         goto out;
1958         }
1959         status = ops->renew_lease(clp, cred);
1960         put_cred(cred);
1961         if (status == -ETIMEDOUT) {
1962                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1963                 return 0;
1964         }
1965 out:
1966         return nfs4_recovery_handle_error(clp, status);
1967 }
1968
1969 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1970  * and for recoverable errors on EXCHANGE_ID for v4.1
1971  */
1972 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1973 {
1974         switch (status) {
1975         case -NFS4ERR_SEQ_MISORDERED:
1976                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1977                         return -ESERVERFAULT;
1978                 /* Lease confirmation error: retry after purging the lease */
1979                 ssleep(1);
1980                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1981                 break;
1982         case -NFS4ERR_STALE_CLIENTID:
1983                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1984                 nfs4_state_start_reclaim_reboot(clp);
1985                 break;
1986         case -NFS4ERR_CLID_INUSE:
1987                 pr_err("NFS: Server %s reports our clientid is in use\n",
1988                         clp->cl_hostname);
1989                 nfs_mark_client_ready(clp, -EPERM);
1990                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1991                 return -EPERM;
1992         case -EACCES:
1993         case -NFS4ERR_DELAY:
1994         case -EAGAIN:
1995                 ssleep(1);
1996                 break;
1997
1998         case -NFS4ERR_MINOR_VERS_MISMATCH:
1999                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2000                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2001                 dprintk("%s: exit with error %d for server %s\n",
2002                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
2003                 return -EPROTONOSUPPORT;
2004         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2005                                  * in nfs4_exchange_id */
2006         default:
2007                 dprintk("%s: exit with error %d for server %s\n", __func__,
2008                                 status, clp->cl_hostname);
2009                 return status;
2010         }
2011         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2012         dprintk("%s: handled error %d for server %s\n", __func__, status,
2013                         clp->cl_hostname);
2014         return 0;
2015 }
2016
2017 static int nfs4_establish_lease(struct nfs_client *clp)
2018 {
2019         const struct cred *cred;
2020         const struct nfs4_state_recovery_ops *ops =
2021                 clp->cl_mvops->reboot_recovery_ops;
2022         int status;
2023
2024         status = nfs4_begin_drain_session(clp);
2025         if (status != 0)
2026                 return status;
2027         cred = nfs4_get_clid_cred(clp);
2028         if (cred == NULL)
2029                 return -ENOENT;
2030         status = ops->establish_clid(clp, cred);
2031         put_cred(cred);
2032         if (status != 0)
2033                 return status;
2034         pnfs_destroy_all_layouts(clp);
2035         return 0;
2036 }
2037
2038 /*
2039  * Returns zero or a negative errno.  NFS4ERR values are converted
2040  * to local errno values.
2041  */
2042 static int nfs4_reclaim_lease(struct nfs_client *clp)
2043 {
2044         int status;
2045
2046         status = nfs4_establish_lease(clp);
2047         if (status < 0)
2048                 return nfs4_handle_reclaim_lease_error(clp, status);
2049         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2050                 nfs4_state_start_reclaim_nograce(clp);
2051         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2052                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2053         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2054         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2055         return 0;
2056 }
2057
2058 static int nfs4_purge_lease(struct nfs_client *clp)
2059 {
2060         int status;
2061
2062         status = nfs4_establish_lease(clp);
2063         if (status < 0)
2064                 return nfs4_handle_reclaim_lease_error(clp, status);
2065         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2066         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2067         nfs4_state_start_reclaim_nograce(clp);
2068         return 0;
2069 }
2070
2071 /*
2072  * Try remote migration of one FSID from a source server to a
2073  * destination server.  The source server provides a list of
2074  * potential destinations.
2075  *
2076  * Returns zero or a negative NFS4ERR status code.
2077  */
2078 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2079 {
2080         struct nfs_client *clp = server->nfs_client;
2081         struct nfs4_fs_locations *locations = NULL;
2082         struct inode *inode;
2083         struct page *page;
2084         int status, result;
2085
2086         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2087                         (unsigned long long)server->fsid.major,
2088                         (unsigned long long)server->fsid.minor,
2089                         clp->cl_hostname);
2090
2091         result = 0;
2092         page = alloc_page(GFP_KERNEL);
2093         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2094         if (page == NULL || locations == NULL) {
2095                 dprintk("<-- %s: no memory\n", __func__);
2096                 goto out;
2097         }
2098
2099         inode = d_inode(server->super->s_root);
2100         result = nfs4_proc_get_locations(inode, locations, page, cred);
2101         if (result) {
2102                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2103                         __func__, result);
2104                 goto out;
2105         }
2106
2107         result = -NFS4ERR_NXIO;
2108         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2109                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2110                         __func__);
2111                 goto out;
2112         }
2113
2114         status = nfs4_begin_drain_session(clp);
2115         if (status != 0) {
2116                 result = status;
2117                 goto out;
2118         }
2119
2120         status = nfs4_replace_transport(server, locations);
2121         if (status != 0) {
2122                 dprintk("<-- %s: failed to replace transport: %d\n",
2123                         __func__, status);
2124                 goto out;
2125         }
2126
2127         result = 0;
2128         dprintk("<-- %s: migration succeeded\n", __func__);
2129
2130 out:
2131         if (page != NULL)
2132                 __free_page(page);
2133         kfree(locations);
2134         if (result) {
2135                 pr_err("NFS: migration recovery failed (server %s)\n",
2136                                 clp->cl_hostname);
2137                 set_bit(NFS_MIG_FAILED, &server->mig_status);
2138         }
2139         return result;
2140 }
2141
2142 /*
2143  * Returns zero or a negative NFS4ERR status code.
2144  */
2145 static int nfs4_handle_migration(struct nfs_client *clp)
2146 {
2147         const struct nfs4_state_maintenance_ops *ops =
2148                                 clp->cl_mvops->state_renewal_ops;
2149         struct nfs_server *server;
2150         const struct cred *cred;
2151
2152         dprintk("%s: migration reported on \"%s\"\n", __func__,
2153                         clp->cl_hostname);
2154
2155         cred = ops->get_state_renewal_cred(clp);
2156         if (cred == NULL)
2157                 return -NFS4ERR_NOENT;
2158
2159         clp->cl_mig_gen++;
2160 restart:
2161         rcu_read_lock();
2162         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2163                 int status;
2164
2165                 if (server->mig_gen == clp->cl_mig_gen)
2166                         continue;
2167                 server->mig_gen = clp->cl_mig_gen;
2168
2169                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2170                                                 &server->mig_status))
2171                         continue;
2172
2173                 rcu_read_unlock();
2174                 status = nfs4_try_migration(server, cred);
2175                 if (status < 0) {
2176                         put_cred(cred);
2177                         return status;
2178                 }
2179                 goto restart;
2180         }
2181         rcu_read_unlock();
2182         put_cred(cred);
2183         return 0;
2184 }
2185
2186 /*
2187  * Test each nfs_server on the clp's cl_superblocks list to see
2188  * if it's moved to another server.  Stop when the server no longer
2189  * returns NFS4ERR_LEASE_MOVED.
2190  */
2191 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2192 {
2193         const struct nfs4_state_maintenance_ops *ops =
2194                                 clp->cl_mvops->state_renewal_ops;
2195         struct nfs_server *server;
2196         const struct cred *cred;
2197
2198         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2199                         clp->cl_hostname);
2200
2201         cred = ops->get_state_renewal_cred(clp);
2202         if (cred == NULL)
2203                 return -NFS4ERR_NOENT;
2204
2205         clp->cl_mig_gen++;
2206 restart:
2207         rcu_read_lock();
2208         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2209                 struct inode *inode;
2210                 int status;
2211
2212                 if (server->mig_gen == clp->cl_mig_gen)
2213                         continue;
2214                 server->mig_gen = clp->cl_mig_gen;
2215
2216                 rcu_read_unlock();
2217
2218                 inode = d_inode(server->super->s_root);
2219                 status = nfs4_proc_fsid_present(inode, cred);
2220                 if (status != -NFS4ERR_MOVED)
2221                         goto restart;   /* wasn't this one */
2222                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2223                         goto restart;   /* there are more */
2224                 goto out;
2225         }
2226         rcu_read_unlock();
2227
2228 out:
2229         put_cred(cred);
2230         return 0;
2231 }
2232
2233 /**
2234  * nfs4_discover_server_trunking - Detect server IP address trunking
2235  *
2236  * @clp: nfs_client under test
2237  * @result: OUT: found nfs_client, or clp
2238  *
2239  * Returns zero or a negative errno.  If zero is returned,
2240  * an nfs_client pointer is planted in "result".
2241  *
2242  * Note: since we are invoked in process context, and
2243  * not from inside the state manager, we cannot use
2244  * nfs4_handle_reclaim_lease_error().
2245  */
2246 int nfs4_discover_server_trunking(struct nfs_client *clp,
2247                                   struct nfs_client **result)
2248 {
2249         const struct nfs4_state_recovery_ops *ops =
2250                                 clp->cl_mvops->reboot_recovery_ops;
2251         struct rpc_clnt *clnt;
2252         const struct cred *cred;
2253         int i, status;
2254
2255         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2256
2257         clnt = clp->cl_rpcclient;
2258         i = 0;
2259
2260         mutex_lock(&nfs_clid_init_mutex);
2261 again:
2262         status  = -ENOENT;
2263         cred = nfs4_get_clid_cred(clp);
2264         if (cred == NULL)
2265                 goto out_unlock;
2266
2267         status = ops->detect_trunking(clp, result, cred);
2268         put_cred(cred);
2269         switch (status) {
2270         case 0:
2271         case -EINTR:
2272         case -ERESTARTSYS:
2273                 break;
2274         case -ETIMEDOUT:
2275                 if (clnt->cl_softrtry)
2276                         break;
2277                 fallthrough;
2278         case -NFS4ERR_DELAY:
2279         case -EAGAIN:
2280                 ssleep(1);
2281                 fallthrough;
2282         case -NFS4ERR_STALE_CLIENTID:
2283                 dprintk("NFS: %s after status %d, retrying\n",
2284                         __func__, status);
2285                 goto again;
2286         case -EACCES:
2287                 if (i++ == 0) {
2288                         nfs4_root_machine_cred(clp);
2289                         goto again;
2290                 }
2291                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2292                         break;
2293                 fallthrough;
2294         case -NFS4ERR_CLID_INUSE:
2295         case -NFS4ERR_WRONGSEC:
2296                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2297                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2298                         status = -EPERM;
2299                         break;
2300                 }
2301                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2302                 if (IS_ERR(clnt)) {
2303                         status = PTR_ERR(clnt);
2304                         break;
2305                 }
2306                 /* Note: this is safe because we haven't yet marked the
2307                  * client as ready, so we are the only user of
2308                  * clp->cl_rpcclient
2309                  */
2310                 clnt = xchg(&clp->cl_rpcclient, clnt);
2311                 rpc_shutdown_client(clnt);
2312                 clnt = clp->cl_rpcclient;
2313                 goto again;
2314
2315         case -NFS4ERR_MINOR_VERS_MISMATCH:
2316                 status = -EPROTONOSUPPORT;
2317                 break;
2318
2319         case -EKEYEXPIRED:
2320         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2321                                  * in nfs4_exchange_id */
2322                 status = -EKEYEXPIRED;
2323                 break;
2324         default:
2325                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2326                                 __func__, status);
2327                 status = -EIO;
2328         }
2329
2330 out_unlock:
2331         mutex_unlock(&nfs_clid_init_mutex);
2332         dprintk("NFS: %s: status = %d\n", __func__, status);
2333         return status;
2334 }
2335
2336 #ifdef CONFIG_NFS_V4_1
2337 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2338 {
2339         struct nfs_client *clp = session->clp;
2340
2341         switch (err) {
2342         default:
2343                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2344                 break;
2345         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2346                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2347         }
2348         nfs4_schedule_state_manager(clp);
2349 }
2350 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2351
2352 void nfs41_notify_server(struct nfs_client *clp)
2353 {
2354         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2355         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2356         nfs4_schedule_state_manager(clp);
2357 }
2358
2359 static void nfs4_reset_all_state(struct nfs_client *clp)
2360 {
2361         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2362                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2363                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2364                 nfs4_state_start_reclaim_nograce(clp);
2365                 dprintk("%s: scheduling reset of all state for server %s!\n",
2366                                 __func__, clp->cl_hostname);
2367                 nfs4_schedule_state_manager(clp);
2368         }
2369 }
2370
2371 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2372 {
2373         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2374                 nfs4_state_start_reclaim_reboot(clp);
2375                 dprintk("%s: server %s rebooted!\n", __func__,
2376                                 clp->cl_hostname);
2377                 nfs4_schedule_state_manager(clp);
2378         }
2379 }
2380
2381 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2382 {
2383         nfs4_reset_all_state(clp);
2384         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2385 }
2386
2387 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2388 {
2389         nfs4_state_start_reclaim_nograce(clp);
2390         nfs4_schedule_state_manager(clp);
2391
2392         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2393 }
2394
2395 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2396 {
2397         /* FIXME: For now, we destroy all layouts. */
2398         pnfs_destroy_all_layouts(clp);
2399         nfs_test_expired_all_delegations(clp);
2400         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2401                         clp->cl_hostname);
2402 }
2403
2404 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2405 {
2406         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2407         nfs4_schedule_state_manager(clp);
2408
2409         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2410                         clp->cl_hostname);
2411 }
2412
2413 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2414 {
2415         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2416                 &clp->cl_state) == 0)
2417                 nfs4_schedule_state_manager(clp);
2418 }
2419
2420 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2421                 bool recovery)
2422 {
2423         if (!flags)
2424                 return;
2425
2426         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2427                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2428         /*
2429          * If we're called from the state manager thread, then assume we're
2430          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2431          * Those flags are expected to remain set until we're done
2432          * recovering (see RFC5661, section 18.46.3).
2433          */
2434         if (recovery)
2435                 goto out_recovery;
2436
2437         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2438                 nfs41_handle_server_reboot(clp);
2439         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2440                 nfs41_handle_all_state_revoked(clp);
2441         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2442                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2443                 nfs41_handle_some_state_revoked(clp);
2444         if (flags & SEQ4_STATUS_LEASE_MOVED)
2445                 nfs4_schedule_lease_moved_recovery(clp);
2446         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2447                 nfs41_handle_recallable_state_revoked(clp);
2448 out_recovery:
2449         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2450                 nfs41_handle_backchannel_fault(clp);
2451         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2452                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2453                 nfs41_handle_cb_path_down(clp);
2454 }
2455
2456 static int nfs4_reset_session(struct nfs_client *clp)
2457 {
2458         const struct cred *cred;
2459         int status;
2460
2461         if (!nfs4_has_session(clp))
2462                 return 0;
2463         status = nfs4_begin_drain_session(clp);
2464         if (status != 0)
2465                 return status;
2466         cred = nfs4_get_clid_cred(clp);
2467         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2468         switch (status) {
2469         case 0:
2470         case -NFS4ERR_BADSESSION:
2471         case -NFS4ERR_DEADSESSION:
2472                 break;
2473         case -NFS4ERR_BACK_CHAN_BUSY:
2474         case -NFS4ERR_DELAY:
2475                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2476                 status = 0;
2477                 ssleep(1);
2478                 goto out;
2479         default:
2480                 status = nfs4_recovery_handle_error(clp, status);
2481                 goto out;
2482         }
2483
2484         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2485         status = nfs4_proc_create_session(clp, cred);
2486         if (status) {
2487                 dprintk("%s: session reset failed with status %d for server %s!\n",
2488                         __func__, status, clp->cl_hostname);
2489                 status = nfs4_handle_reclaim_lease_error(clp, status);
2490                 goto out;
2491         }
2492         nfs41_finish_session_reset(clp);
2493         dprintk("%s: session reset was successful for server %s!\n",
2494                         __func__, clp->cl_hostname);
2495 out:
2496         put_cred(cred);
2497         return status;
2498 }
2499
2500 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2501 {
2502         const struct cred *cred;
2503         int ret;
2504
2505         if (!nfs4_has_session(clp))
2506                 return 0;
2507         ret = nfs4_begin_drain_session(clp);
2508         if (ret != 0)
2509                 return ret;
2510         cred = nfs4_get_clid_cred(clp);
2511         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2512         put_cred(cred);
2513         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2514         switch (ret) {
2515         case 0:
2516                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2517                         __func__, clp->cl_hostname);
2518                 break;
2519         case -NFS4ERR_DELAY:
2520                 ssleep(1);
2521                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2522                 break;
2523         default:
2524                 return nfs4_recovery_handle_error(clp, ret);
2525         }
2526         return 0;
2527 }
2528
2529 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2530 {
2531         int iomode = 0;
2532
2533         if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2534                 iomode += IOMODE_READ;
2535         if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2536                 iomode += IOMODE_RW;
2537         /* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2538         if (iomode) {
2539                 pnfs_layout_return_unused_byclid(clp, iomode);
2540                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2541         }
2542 }
2543 #else /* CONFIG_NFS_V4_1 */
2544 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2545
2546 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2547 {
2548         return 0;
2549 }
2550
2551 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2552 {
2553 }
2554 #endif /* CONFIG_NFS_V4_1 */
2555
2556 static void nfs4_state_manager(struct nfs_client *clp)
2557 {
2558         int status = 0;
2559         const char *section = "", *section_sep = "";
2560
2561         /* Ensure exclusive access to NFSv4 state */
2562         do {
2563                 trace_nfs4_state_mgr(clp);
2564                 clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2565                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2566                         section = "purge state";
2567                         status = nfs4_purge_lease(clp);
2568                         if (status < 0)
2569                                 goto out_error;
2570                         continue;
2571                 }
2572
2573                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2574                         section = "lease expired";
2575                         /* We're going to have to re-establish a clientid */
2576                         status = nfs4_reclaim_lease(clp);
2577                         if (status < 0)
2578                                 goto out_error;
2579                         continue;
2580                 }
2581
2582                 /* Initialize or reset the session */
2583                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2584                         section = "reset session";
2585                         status = nfs4_reset_session(clp);
2586                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2587                                 continue;
2588                         if (status < 0)
2589                                 goto out_error;
2590                 }
2591
2592                 /* Send BIND_CONN_TO_SESSION */
2593                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2594                                 &clp->cl_state)) {
2595                         section = "bind conn to session";
2596                         status = nfs4_bind_conn_to_session(clp);
2597                         if (status < 0)
2598                                 goto out_error;
2599                         continue;
2600                 }
2601
2602                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2603                         section = "check lease";
2604                         status = nfs4_check_lease(clp);
2605                         if (status < 0)
2606                                 goto out_error;
2607                         continue;
2608                 }
2609
2610                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2611                         section = "migration";
2612                         status = nfs4_handle_migration(clp);
2613                         if (status < 0)
2614                                 goto out_error;
2615                 }
2616
2617                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2618                         section = "lease moved";
2619                         status = nfs4_handle_lease_moved(clp);
2620                         if (status < 0)
2621                                 goto out_error;
2622                 }
2623
2624                 /* First recover reboot state... */
2625                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2626                         section = "reclaim reboot";
2627                         status = nfs4_do_reclaim(clp,
2628                                 clp->cl_mvops->reboot_recovery_ops);
2629                         if (status == -EAGAIN)
2630                                 continue;
2631                         if (status < 0)
2632                                 goto out_error;
2633                         nfs4_state_end_reclaim_reboot(clp);
2634                 }
2635
2636                 /* Detect expired delegations... */
2637                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2638                         section = "detect expired delegations";
2639                         nfs_reap_expired_delegations(clp);
2640                         continue;
2641                 }
2642
2643                 /* Now recover expired state... */
2644                 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2645                         section = "reclaim nograce";
2646                         status = nfs4_do_reclaim(clp,
2647                                 clp->cl_mvops->nograce_recovery_ops);
2648                         if (status == -EAGAIN)
2649                                 continue;
2650                         if (status < 0)
2651                                 goto out_error;
2652                         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2653                 }
2654
2655                 nfs4_end_drain_session(clp);
2656                 nfs4_clear_state_manager_bit(clp);
2657
2658                 if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2659                         if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2660                                 nfs_client_return_marked_delegations(clp);
2661                                 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2662                         }
2663                         nfs4_layoutreturn_any_run(clp);
2664                         clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2665                 }
2666
2667                 /* Did we race with an attempt to give us more work? */
2668                 if (!test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state))
2669                         return;
2670                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2671                         return;
2672         } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2673         goto out_drain;
2674
2675 out_error:
2676         if (strlen(section))
2677                 section_sep = ": ";
2678         trace_nfs4_state_mgr_failed(clp, section, status);
2679         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2680                         " with error %d\n", section_sep, section,
2681                         clp->cl_hostname, -status);
2682         ssleep(1);
2683 out_drain:
2684         nfs4_end_drain_session(clp);
2685         nfs4_clear_state_manager_bit(clp);
2686 }
2687
2688 static int nfs4_run_state_manager(void *ptr)
2689 {
2690         struct nfs_client *clp = ptr;
2691
2692         allow_signal(SIGKILL);
2693         nfs4_state_manager(clp);
2694         nfs_put_client(clp);
2695         module_put_and_exit(0);
2696         return 0;
2697 }
2698
2699 /*
2700  * Local variables:
2701  *  c-basic-offset: 8
2702  * End:
2703  */