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/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52 #include <linux/jiffies.h>
  53
  54 #include <linux/sunrpc/clnt.h>
  55
  56 #include "nfs4_fs.h"
  57 #include "callback.h"
  58 #include "delegation.h"
  59 #include "internal.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid;
  69 static DEFINE_MUTEX(nfs_clid_init_mutex);
  70
  71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  72 {
  73         struct nfs4_setclientid_res clid = {
  74                 .clientid = clp->cl_clientid,
  75                 .confirm = clp->cl_confirm,
  76         };
  77         unsigned short port;
  78         int status;
  79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  80
  81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  82                 goto do_confirm;
  83         port = nn->nfs_callback_tcpport;
  84         if (clp->cl_addr.ss_family == AF_INET6)
  85                 port = nn->nfs_callback_tcpport6;
  86
  87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  88         if (status != 0)
  89                 goto out;
  90         clp->cl_clientid = clid.clientid;
  91         clp->cl_confirm = clid.confirm;
  92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  93 do_confirm:
  94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  95         if (status != 0)
  96                 goto out;
  97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98         nfs4_schedule_state_renewal(clp);
  99 out:
 100         return status;
 101 }
 102
 103 /**
 104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 105  *
 106  * @clp: nfs_client under test
 107  * @result: OUT: found nfs_client, or clp
 108  * @cred: credential to use for trunking test
 109  *
 110  * Returns zero, a negative errno, or a negative NFS4ERR status.
 111  * If zero is returned, an nfs_client pointer is planted in
 112  * "result".
 113  *
 114  * Note: The returned client may not yet be marked ready.
 115  */
 116 int nfs40_discover_server_trunking(struct nfs_client *clp,
 117                                    struct nfs_client **result,
 118                                    struct rpc_cred *cred)
 119 {
 120         struct nfs4_setclientid_res clid = {
 121                 .clientid = clp->cl_clientid,
 122                 .confirm = clp->cl_confirm,
 123         };
 124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 125         unsigned short port;
 126         int status;
 127
 128         port = nn->nfs_callback_tcpport;
 129         if (clp->cl_addr.ss_family == AF_INET6)
 130                 port = nn->nfs_callback_tcpport6;
 131
 132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 133         if (status != 0)
 134                 goto out;
 135         clp->cl_clientid = clid.clientid;
 136         clp->cl_confirm = clid.confirm;
 137
 138         status = nfs40_walk_client_list(clp, result, cred);
 139         if (status == 0) {
 140                 /* Sustain the lease, even if it's empty.  If the clientid4
 141                  * goes stale it's of no use for trunking discovery. */
 142                 nfs4_schedule_state_renewal(*result);
 143         }
 144 out:
 145         return status;
 146 }
 147
 148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 149 {
 150         struct rpc_cred *cred = NULL;
 151
 152         if (clp->cl_machine_cred != NULL)
 153                 cred = get_rpccred(clp->cl_machine_cred);
 154         return cred;
 155 }
 156
 157 static void nfs4_root_machine_cred(struct nfs_client *clp)
 158 {
 159         struct rpc_cred *cred, *new;
 160
 161         new = rpc_lookup_machine_cred(NULL);
 162         spin_lock(&clp->cl_lock);
 163         cred = clp->cl_machine_cred;
 164         clp->cl_machine_cred = new;
 165         spin_unlock(&clp->cl_lock);
 166         if (cred != NULL)
 167                 put_rpccred(cred);
 168 }
 169
 170 static struct rpc_cred *
 171 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 172 {
 173         struct rpc_cred *cred = NULL;
 174         struct nfs4_state_owner *sp;
 175         struct rb_node *pos;
 176
 177         for (pos = rb_first(&server->state_owners);
 178              pos != NULL;
 179              pos = rb_next(pos)) {
 180                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 181                 if (list_empty(&sp->so_states))
 182                         continue;
 183                 cred = get_rpccred(sp->so_cred);
 184                 break;
 185         }
 186         return cred;
 187 }
 188
 189 /**
 190  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 191  * @clp: client state handle
 192  *
 193  * Returns an rpc_cred with reference count bumped, or NULL.
 194  * Caller must hold clp->cl_lock.
 195  */
 196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 197 {
 198         struct rpc_cred *cred = NULL;
 199         struct nfs_server *server;
 200
 201         /* Use machine credentials if available */
 202         cred = nfs4_get_machine_cred_locked(clp);
 203         if (cred != NULL)
 204                 goto out;
 205
 206         rcu_read_lock();
 207         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 208                 cred = nfs4_get_renew_cred_server_locked(server);
 209                 if (cred != NULL)
 210                         break;
 211         }
 212         rcu_read_unlock();
 213
 214 out:
 215         return cred;
 216 }
 217
 218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 219 {
 220         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 221                 spin_lock(&tbl->slot_tbl_lock);
 222                 nfs41_wake_slot_table(tbl);
 223                 spin_unlock(&tbl->slot_tbl_lock);
 224         }
 225 }
 226
 227 static void nfs4_end_drain_session(struct nfs_client *clp)
 228 {
 229         struct nfs4_session *ses = clp->cl_session;
 230
 231         if (clp->cl_slot_tbl) {
 232                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 233                 return;
 234         }
 235
 236         if (ses != NULL) {
 237                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 238                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 239         }
 240 }
 241
 242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 243 {
 244         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 245         spin_lock(&tbl->slot_tbl_lock);
 246         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 247                 reinit_completion(&tbl->complete);
 248                 spin_unlock(&tbl->slot_tbl_lock);
 249                 return wait_for_completion_interruptible(&tbl->complete);
 250         }
 251         spin_unlock(&tbl->slot_tbl_lock);
 252         return 0;
 253 }
 254
 255 static int nfs4_begin_drain_session(struct nfs_client *clp)
 256 {
 257         struct nfs4_session *ses = clp->cl_session;
 258         int ret = 0;
 259
 260         if (clp->cl_slot_tbl)
 261                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 262
 263         /* back channel */
 264         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 265         if (ret)
 266                 return ret;
 267         /* fore channel */
 268         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 269 }
 270
 271 #if defined(CONFIG_NFS_V4_1)
 272
 273 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 274 {
 275         int status;
 276         struct nfs_fsinfo fsinfo;
 277
 278         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 279                 nfs4_schedule_state_renewal(clp);
 280                 return 0;
 281         }
 282
 283         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 284         if (status == 0) {
 285                 /* Update lease time and schedule renewal */
 286                 spin_lock(&clp->cl_lock);
 287                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 288                 clp->cl_last_renewal = jiffies;
 289                 spin_unlock(&clp->cl_lock);
 290
 291                 nfs4_schedule_state_renewal(clp);
 292         }
 293
 294         return status;
 295 }
 296
 297 static void nfs41_finish_session_reset(struct nfs_client *clp)
 298 {
 299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 301         /* create_session negotiated new slot table */
 302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 303         nfs41_setup_state_renewal(clp);
 304 }
 305
 306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 307 {
 308         int status;
 309
 310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 311                 goto do_confirm;
 312         nfs4_begin_drain_session(clp);
 313         status = nfs4_proc_exchange_id(clp, cred);
 314         if (status != 0)
 315                 goto out;
 316         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 317 do_confirm:
 318         status = nfs4_proc_create_session(clp, cred);
 319         if (status != 0)
 320                 goto out;
 321         nfs41_finish_session_reset(clp);
 322         nfs_mark_client_ready(clp, NFS_CS_READY);
 323 out:
 324         return status;
 325 }
 326
 327 /**
 328  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 329  *
 330  * @clp: nfs_client under test
 331  * @result: OUT: found nfs_client, or clp
 332  * @cred: credential to use for trunking test
 333  *
 334  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 335  * If NFS4_OK is returned, an nfs_client pointer is planted in
 336  * "result".
 337  *
 338  * Note: The returned client may not yet be marked ready.
 339  */
 340 int nfs41_discover_server_trunking(struct nfs_client *clp,
 341                                    struct nfs_client **result,
 342                                    struct rpc_cred *cred)
 343 {
 344         int status;
 345
 346         status = nfs4_proc_exchange_id(clp, cred);
 347         if (status != NFS4_OK)
 348                 return status;
 349         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 350
 351         return nfs41_walk_client_list(clp, result, cred);
 352 }
 353
 354 #endif /* CONFIG_NFS_V4_1 */
 355
 356 /**
 357  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 358  * @clp: client state handle
 359  *
 360  * Returns an rpc_cred with reference count bumped, or NULL.
 361  */
 362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 363 {
 364         struct rpc_cred *cred;
 365
 366         spin_lock(&clp->cl_lock);
 367         cred = nfs4_get_machine_cred_locked(clp);
 368         spin_unlock(&clp->cl_lock);
 369         return cred;
 370 }
 371
 372 static struct nfs4_state_owner *
 373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 374 {
 375         struct rb_node **p = &server->state_owners.rb_node,
 376                        *parent = NULL;
 377         struct nfs4_state_owner *sp;
 378
 379         while (*p != NULL) {
 380                 parent = *p;
 381                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 382
 383                 if (cred < sp->so_cred)
 384                         p = &parent->rb_left;
 385                 else if (cred > sp->so_cred)
 386                         p = &parent->rb_right;
 387                 else {
 388                         if (!list_empty(&sp->so_lru))
 389                                 list_del_init(&sp->so_lru);
 390                         atomic_inc(&sp->so_count);
 391                         return sp;
 392                 }
 393         }
 394         return NULL;
 395 }
 396
 397 static struct nfs4_state_owner *
 398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 399 {
 400         struct nfs_server *server = new->so_server;
 401         struct rb_node **p = &server->state_owners.rb_node,
 402                        *parent = NULL;
 403         struct nfs4_state_owner *sp;
 404         int err;
 405
 406         while (*p != NULL) {
 407                 parent = *p;
 408                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 409
 410                 if (new->so_cred < sp->so_cred)
 411                         p = &parent->rb_left;
 412                 else if (new->so_cred > sp->so_cred)
 413                         p = &parent->rb_right;
 414                 else {
 415                         if (!list_empty(&sp->so_lru))
 416                                 list_del_init(&sp->so_lru);
 417                         atomic_inc(&sp->so_count);
 418                         return sp;
 419                 }
 420         }
 421         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 422         if (err)
 423                 return ERR_PTR(err);
 424         rb_link_node(&new->so_server_node, parent, p);
 425         rb_insert_color(&new->so_server_node, &server->state_owners);
 426         return new;
 427 }
 428
 429 static void
 430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 431 {
 432         struct nfs_server *server = sp->so_server;
 433
 434         if (!RB_EMPTY_NODE(&sp->so_server_node))
 435                 rb_erase(&sp->so_server_node, &server->state_owners);
 436         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 437 }
 438
 439 static void
 440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 441 {
 442         sc->create_time = ktime_get();
 443         sc->flags = 0;
 444         sc->counter = 0;
 445         spin_lock_init(&sc->lock);
 446         INIT_LIST_HEAD(&sc->list);
 447         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 448 }
 449
 450 static void
 451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 452 {
 453         rpc_destroy_wait_queue(&sc->wait);
 454 }
 455
 456 /*
 457  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 458  * create a new state_owner.
 459  *
 460  */
 461 static struct nfs4_state_owner *
 462 nfs4_alloc_state_owner(struct nfs_server *server,
 463                 struct rpc_cred *cred,
 464                 gfp_t gfp_flags)
 465 {
 466         struct nfs4_state_owner *sp;
 467
 468         sp = kzalloc(sizeof(*sp), gfp_flags);
 469         if (!sp)
 470                 return NULL;
 471         sp->so_server = server;
 472         sp->so_cred = get_rpccred(cred);
 473         spin_lock_init(&sp->so_lock);
 474         INIT_LIST_HEAD(&sp->so_states);
 475         nfs4_init_seqid_counter(&sp->so_seqid);
 476         atomic_set(&sp->so_count, 1);
 477         INIT_LIST_HEAD(&sp->so_lru);
 478         seqcount_init(&sp->so_reclaim_seqcount);
 479         mutex_init(&sp->so_delegreturn_mutex);
 480         return sp;
 481 }
 482
 483 static void
 484 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 485 {
 486         struct rb_node *rb_node = &sp->so_server_node;
 487
 488         if (!RB_EMPTY_NODE(rb_node)) {
 489                 struct nfs_server *server = sp->so_server;
 490                 struct nfs_client *clp = server->nfs_client;
 491
 492                 spin_lock(&clp->cl_lock);
 493                 if (!RB_EMPTY_NODE(rb_node)) {
 494                         rb_erase(rb_node, &server->state_owners);
 495                         RB_CLEAR_NODE(rb_node);
 496                 }
 497                 spin_unlock(&clp->cl_lock);
 498         }
 499 }
 500
 501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 502 {
 503         nfs4_destroy_seqid_counter(&sp->so_seqid);
 504         put_rpccred(sp->so_cred);
 505         kfree(sp);
 506 }
 507
 508 static void nfs4_gc_state_owners(struct nfs_server *server)
 509 {
 510         struct nfs_client *clp = server->nfs_client;
 511         struct nfs4_state_owner *sp, *tmp;
 512         unsigned long time_min, time_max;
 513         LIST_HEAD(doomed);
 514
 515         spin_lock(&clp->cl_lock);
 516         time_max = jiffies;
 517         time_min = (long)time_max - (long)clp->cl_lease_time;
 518         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 519                 /* NB: LRU is sorted so that oldest is at the head */
 520                 if (time_in_range(sp->so_expires, time_min, time_max))
 521                         break;
 522                 list_move(&sp->so_lru, &doomed);
 523                 nfs4_remove_state_owner_locked(sp);
 524         }
 525         spin_unlock(&clp->cl_lock);
 526
 527         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 528                 list_del(&sp->so_lru);
 529                 nfs4_free_state_owner(sp);
 530         }
 531 }
 532
 533 /**
 534  * nfs4_get_state_owner - Look up a state owner given a credential
 535  * @server: nfs_server to search
 536  * @cred: RPC credential to match
 537  *
 538  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 539  */
 540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 541                                               struct rpc_cred *cred,
 542                                               gfp_t gfp_flags)
 543 {
 544         struct nfs_client *clp = server->nfs_client;
 545         struct nfs4_state_owner *sp, *new;
 546
 547         spin_lock(&clp->cl_lock);
 548         sp = nfs4_find_state_owner_locked(server, cred);
 549         spin_unlock(&clp->cl_lock);
 550         if (sp != NULL)
 551                 goto out;
 552         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 553         if (new == NULL)
 554                 goto out;
 555         do {
 556                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 557                         break;
 558                 spin_lock(&clp->cl_lock);
 559                 sp = nfs4_insert_state_owner_locked(new);
 560                 spin_unlock(&clp->cl_lock);
 561         } while (sp == ERR_PTR(-EAGAIN));
 562         if (sp != new)
 563                 nfs4_free_state_owner(new);
 564 out:
 565         nfs4_gc_state_owners(server);
 566         return sp;
 567 }
 568
 569 /**
 570  * nfs4_put_state_owner - Release a nfs4_state_owner
 571  * @sp: state owner data to release
 572  *
 573  * Note that we keep released state owners on an LRU
 574  * list.
 575  * This caches valid state owners so that they can be
 576  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 577  * It also pins the uniquifier of dropped state owners for
 578  * a while, to ensure that those state owner names are
 579  * never reused.
 580  */
 581 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 582 {
 583         struct nfs_server *server = sp->so_server;
 584         struct nfs_client *clp = server->nfs_client;
 585
 586         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 587                 return;
 588
 589         sp->so_expires = jiffies;
 590         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 591         spin_unlock(&clp->cl_lock);
 592 }
 593
 594 /**
 595  * nfs4_purge_state_owners - Release all cached state owners
 596  * @server: nfs_server with cached state owners to release
 597  *
 598  * Called at umount time.  Remaining state owners will be on
 599  * the LRU with ref count of zero.
 600  */
 601 void nfs4_purge_state_owners(struct nfs_server *server)
 602 {
 603         struct nfs_client *clp = server->nfs_client;
 604         struct nfs4_state_owner *sp, *tmp;
 605         LIST_HEAD(doomed);
 606
 607         spin_lock(&clp->cl_lock);
 608         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 609                 list_move(&sp->so_lru, &doomed);
 610                 nfs4_remove_state_owner_locked(sp);
 611         }
 612         spin_unlock(&clp->cl_lock);
 613
 614         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 615                 list_del(&sp->so_lru);
 616                 nfs4_free_state_owner(sp);
 617         }
 618 }
 619
 620 static struct nfs4_state *
 621 nfs4_alloc_open_state(void)
 622 {
 623         struct nfs4_state *state;
 624
 625         state = kzalloc(sizeof(*state), GFP_NOFS);
 626         if (!state)
 627                 return NULL;
 628         atomic_set(&state->count, 1);
 629         INIT_LIST_HEAD(&state->lock_states);
 630         spin_lock_init(&state->state_lock);
 631         seqlock_init(&state->seqlock);
 632         return state;
 633 }
 634
 635 void
 636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 637 {
 638         if (state->state == fmode)
 639                 return;
 640         /* NB! List reordering - see the reclaim code for why.  */
 641         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 642                 if (fmode & FMODE_WRITE)
 643                         list_move(&state->open_states, &state->owner->so_states);
 644                 else
 645                         list_move_tail(&state->open_states, &state->owner->so_states);
 646         }
 647         state->state = fmode;
 648 }
 649
 650 static struct nfs4_state *
 651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 652 {
 653         struct nfs_inode *nfsi = NFS_I(inode);
 654         struct nfs4_state *state;
 655
 656         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 657                 if (state->owner != owner)
 658                         continue;
 659                 if (!nfs4_valid_open_stateid(state))
 660                         continue;
 661                 if (atomic_inc_not_zero(&state->count))
 662                         return state;
 663         }
 664         return NULL;
 665 }
 666
 667 static void
 668 nfs4_free_open_state(struct nfs4_state *state)
 669 {
 670         kfree(state);
 671 }
 672
 673 struct nfs4_state *
 674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 675 {
 676         struct nfs4_state *state, *new;
 677         struct nfs_inode *nfsi = NFS_I(inode);
 678
 679         spin_lock(&inode->i_lock);
 680         state = __nfs4_find_state_byowner(inode, owner);
 681         spin_unlock(&inode->i_lock);
 682         if (state)
 683                 goto out;
 684         new = nfs4_alloc_open_state();
 685         spin_lock(&owner->so_lock);
 686         spin_lock(&inode->i_lock);
 687         state = __nfs4_find_state_byowner(inode, owner);
 688         if (state == NULL && new != NULL) {
 689                 state = new;
 690                 state->owner = owner;
 691                 atomic_inc(&owner->so_count);
 692                 list_add(&state->inode_states, &nfsi->open_states);
 693                 ihold(inode);
 694                 state->inode = inode;
 695                 spin_unlock(&inode->i_lock);
 696                 /* Note: The reclaim code dictates that we add stateless
 697                  * and read-only stateids to the end of the list */
 698                 list_add_tail(&state->open_states, &owner->so_states);
 699                 spin_unlock(&owner->so_lock);
 700         } else {
 701                 spin_unlock(&inode->i_lock);
 702                 spin_unlock(&owner->so_lock);
 703                 if (new)
 704                         nfs4_free_open_state(new);
 705         }
 706 out:
 707         return state;
 708 }
 709
 710 void nfs4_put_open_state(struct nfs4_state *state)
 711 {
 712         struct inode *inode = state->inode;
 713         struct nfs4_state_owner *owner = state->owner;
 714
 715         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 716                 return;
 717         spin_lock(&inode->i_lock);
 718         list_del(&state->inode_states);
 719         list_del(&state->open_states);
 720         spin_unlock(&inode->i_lock);
 721         spin_unlock(&owner->so_lock);
 722         iput(inode);
 723         nfs4_free_open_state(state);
 724         nfs4_put_state_owner(owner);
 725 }
 726
 727 /*
 728  * Close the current file.
 729  */
 730 static void __nfs4_close(struct nfs4_state *state,
 731                 fmode_t fmode, gfp_t gfp_mask, int wait)
 732 {
 733         struct nfs4_state_owner *owner = state->owner;
 734         int call_close = 0;
 735         fmode_t newstate;
 736
 737         atomic_inc(&owner->so_count);
 738         /* Protect against nfs4_find_state() */
 739         spin_lock(&owner->so_lock);
 740         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 741                 case FMODE_READ:
 742                         state->n_rdonly--;
 743                         break;
 744                 case FMODE_WRITE:
 745                         state->n_wronly--;
 746                         break;
 747                 case FMODE_READ|FMODE_WRITE:
 748                         state->n_rdwr--;
 749         }
 750         newstate = FMODE_READ|FMODE_WRITE;
 751         if (state->n_rdwr == 0) {
 752                 if (state->n_rdonly == 0) {
 753                         newstate &= ~FMODE_READ;
 754                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 755                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 756                 }
 757                 if (state->n_wronly == 0) {
 758                         newstate &= ~FMODE_WRITE;
 759                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 760                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 761                 }
 762                 if (newstate == 0)
 763                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 764         }
 765         nfs4_state_set_mode_locked(state, newstate);
 766         spin_unlock(&owner->so_lock);
 767
 768         if (!call_close) {
 769                 nfs4_put_open_state(state);
 770                 nfs4_put_state_owner(owner);
 771         } else
 772                 nfs4_do_close(state, gfp_mask, wait);
 773 }
 774
 775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 776 {
 777         __nfs4_close(state, fmode, GFP_NOFS, 0);
 778 }
 779
 780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 781 {
 782         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 783 }
 784
 785 /*
 786  * Search the state->lock_states for an existing lock_owner
 787  * that is compatible with current->files
 788  */
 789 static struct nfs4_lock_state *
 790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 791 {
 792         struct nfs4_lock_state *pos;
 793         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 794                 if (pos->ls_owner != fl_owner)
 795                         continue;
 796                 atomic_inc(&pos->ls_count);
 797                 return pos;
 798         }
 799         return NULL;
 800 }
 801
 802 /*
 803  * Return a compatible lock_state. If no initialized lock_state structure
 804  * exists, return an uninitialized one.
 805  *
 806  */
 807 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 808 {
 809         struct nfs4_lock_state *lsp;
 810         struct nfs_server *server = state->owner->so_server;
 811
 812         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 813         if (lsp == NULL)
 814                 return NULL;
 815         nfs4_init_seqid_counter(&lsp->ls_seqid);
 816         atomic_set(&lsp->ls_count, 1);
 817         lsp->ls_state = state;
 818         lsp->ls_owner = fl_owner;
 819         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 820         if (lsp->ls_seqid.owner_id < 0)
 821                 goto out_free;
 822         INIT_LIST_HEAD(&lsp->ls_locks);
 823         return lsp;
 824 out_free:
 825         kfree(lsp);
 826         return NULL;
 827 }
 828
 829 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 830 {
 831         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 832         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 833         kfree(lsp);
 834 }
 835
 836 /*
 837  * Return a compatible lock_state. If no initialized lock_state structure
 838  * exists, return an uninitialized one.
 839  *
 840  */
 841 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 842 {
 843         struct nfs4_lock_state *lsp, *new = NULL;
 844
 845         for(;;) {
 846                 spin_lock(&state->state_lock);
 847                 lsp = __nfs4_find_lock_state(state, owner);
 848                 if (lsp != NULL)
 849                         break;
 850                 if (new != NULL) {
 851                         list_add(&new->ls_locks, &state->lock_states);
 852                         set_bit(LK_STATE_IN_USE, &state->flags);
 853                         lsp = new;
 854                         new = NULL;
 855                         break;
 856                 }
 857                 spin_unlock(&state->state_lock);
 858                 new = nfs4_alloc_lock_state(state, owner);
 859                 if (new == NULL)
 860                         return NULL;
 861         }
 862         spin_unlock(&state->state_lock);
 863         if (new != NULL)
 864                 nfs4_free_lock_state(state->owner->so_server, new);
 865         return lsp;
 866 }
 867
 868 /*
 869  * Release reference to lock_state, and free it if we see that
 870  * it is no longer in use
 871  */
 872 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 873 {
 874         struct nfs_server *server;
 875         struct nfs4_state *state;
 876
 877         if (lsp == NULL)
 878                 return;
 879         state = lsp->ls_state;
 880         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 881                 return;
 882         list_del(&lsp->ls_locks);
 883         if (list_empty(&state->lock_states))
 884                 clear_bit(LK_STATE_IN_USE, &state->flags);
 885         spin_unlock(&state->state_lock);
 886         server = state->owner->so_server;
 887         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 888                 struct nfs_client *clp = server->nfs_client;
 889
 890                 clp->cl_mvops->free_lock_state(server, lsp);
 891         } else
 892                 nfs4_free_lock_state(server, lsp);
 893 }
 894
 895 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 896 {
 897         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 898
 899         dst->fl_u.nfs4_fl.owner = lsp;
 900         atomic_inc(&lsp->ls_count);
 901 }
 902
 903 static void nfs4_fl_release_lock(struct file_lock *fl)
 904 {
 905         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 906 }
 907
 908 static const struct file_lock_operations nfs4_fl_lock_ops = {
 909         .fl_copy_lock = nfs4_fl_copy_lock,
 910         .fl_release_private = nfs4_fl_release_lock,
 911 };
 912
 913 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 914 {
 915         struct nfs4_lock_state *lsp;
 916
 917         if (fl->fl_ops != NULL)
 918                 return 0;
 919         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 920         if (lsp == NULL)
 921                 return -ENOMEM;
 922         fl->fl_u.nfs4_fl.owner = lsp;
 923         fl->fl_ops = &nfs4_fl_lock_ops;
 924         return 0;
 925 }
 926
 927 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 928                 struct nfs4_state *state,
 929                 const struct nfs_lockowner *lockowner)
 930 {
 931         struct nfs4_lock_state *lsp;
 932         fl_owner_t fl_owner;
 933         int ret = -ENOENT;
 934
 935
 936         if (lockowner == NULL)
 937                 goto out;
 938
 939         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 940                 goto out;
 941
 942         fl_owner = lockowner->l_owner;
 943         spin_lock(&state->state_lock);
 944         lsp = __nfs4_find_lock_state(state, fl_owner);
 945         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 946                 ret = -EIO;
 947         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 948                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 949                 ret = 0;
 950         }
 951         spin_unlock(&state->state_lock);
 952         nfs4_put_lock_state(lsp);
 953 out:
 954         return ret;
 955 }
 956
 957 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 958 {
 959         const nfs4_stateid *src;
 960         int seq;
 961
 962         do {
 963                 src = &zero_stateid;
 964                 seq = read_seqbegin(&state->seqlock);
 965                 if (test_bit(NFS_OPEN_STATE, &state->flags))
 966                         src = &state->open_stateid;
 967                 nfs4_stateid_copy(dst, src);
 968         } while (read_seqretry(&state->seqlock, seq));
 969 }
 970
 971 /*
 972  * Byte-range lock aware utility to initialize the stateid of read/write
 973  * requests.
 974  */
 975 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 976                 fmode_t fmode, const struct nfs_lockowner *lockowner)
 977 {
 978         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
 979         if (ret == -EIO)
 980                 /* A lost lock - don't even consider delegations */
 981                 goto out;
 982         /* returns true if delegation stateid found and copied */
 983         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) {
 984                 ret = 0;
 985                 goto out;
 986         }
 987         if (ret != -ENOENT)
 988                 /* nfs4_copy_delegation_stateid() didn't over-write
 989                  * dst, so it still has the lock stateid which we now
 990                  * choose to use.
 991                  */
 992                 goto out;
 993         nfs4_copy_open_stateid(dst, state);
 994         ret = 0;
 995 out:
 996         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
 997                 dst->seqid = 0;
 998         return ret;
 999 }
1000
1001 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1002 {
1003         struct nfs_seqid *new;
1004
1005         new = kmalloc(sizeof(*new), gfp_mask);
1006         if (new == NULL)
1007                 return ERR_PTR(-ENOMEM);
1008         new->sequence = counter;
1009         INIT_LIST_HEAD(&new->list);
1010         new->task = NULL;
1011         return new;
1012 }
1013
1014 void nfs_release_seqid(struct nfs_seqid *seqid)
1015 {
1016         struct nfs_seqid_counter *sequence;
1017
1018         if (seqid == NULL || list_empty(&seqid->list))
1019                 return;
1020         sequence = seqid->sequence;
1021         spin_lock(&sequence->lock);
1022         list_del_init(&seqid->list);
1023         if (!list_empty(&sequence->list)) {
1024                 struct nfs_seqid *next;
1025
1026                 next = list_first_entry(&sequence->list,
1027                                 struct nfs_seqid, list);
1028                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1029         }
1030         spin_unlock(&sequence->lock);
1031 }
1032
1033 void nfs_free_seqid(struct nfs_seqid *seqid)
1034 {
1035         nfs_release_seqid(seqid);
1036         kfree(seqid);
1037 }
1038
1039 /*
1040  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1041  * failed with a seqid incrementing error -
1042  * see comments nfs4.h:seqid_mutating_error()
1043  */
1044 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1045 {
1046         switch (status) {
1047                 case 0:
1048                         break;
1049                 case -NFS4ERR_BAD_SEQID:
1050                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1051                                 return;
1052                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1053                                         " sequence-id error on an"
1054                                         " unconfirmed sequence %p!\n",
1055                                         seqid->sequence);
1056                 case -NFS4ERR_STALE_CLIENTID:
1057                 case -NFS4ERR_STALE_STATEID:
1058                 case -NFS4ERR_BAD_STATEID:
1059                 case -NFS4ERR_BADXDR:
1060                 case -NFS4ERR_RESOURCE:
1061                 case -NFS4ERR_NOFILEHANDLE:
1062                         /* Non-seqid mutating errors */
1063                         return;
1064         };
1065         /*
1066          * Note: no locking needed as we are guaranteed to be first
1067          * on the sequence list
1068          */
1069         seqid->sequence->counter++;
1070 }
1071
1072 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1073 {
1074         struct nfs4_state_owner *sp;
1075
1076         if (seqid == NULL)
1077                 return;
1078
1079         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1080         if (status == -NFS4ERR_BAD_SEQID)
1081                 nfs4_drop_state_owner(sp);
1082         if (!nfs4_has_session(sp->so_server->nfs_client))
1083                 nfs_increment_seqid(status, seqid);
1084 }
1085
1086 /*
1087  * Increment the seqid if the LOCK/LOCKU succeeded, or
1088  * failed with a seqid incrementing error -
1089  * see comments nfs4.h:seqid_mutating_error()
1090  */
1091 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1092 {
1093         if (seqid != NULL)
1094                 nfs_increment_seqid(status, seqid);
1095 }
1096
1097 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1098 {
1099         struct nfs_seqid_counter *sequence;
1100         int status = 0;
1101
1102         if (seqid == NULL)
1103                 goto out;
1104         sequence = seqid->sequence;
1105         spin_lock(&sequence->lock);
1106         seqid->task = task;
1107         if (list_empty(&seqid->list))
1108                 list_add_tail(&seqid->list, &sequence->list);
1109         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1110                 goto unlock;
1111         rpc_sleep_on(&sequence->wait, task, NULL);
1112         status = -EAGAIN;
1113 unlock:
1114         spin_unlock(&sequence->lock);
1115 out:
1116         return status;
1117 }
1118
1119 static int nfs4_run_state_manager(void *);
1120
1121 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1122 {
1123         smp_mb__before_atomic();
1124         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1125         smp_mb__after_atomic();
1126         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1127         rpc_wake_up(&clp->cl_rpcwaitq);
1128 }
1129
1130 /*
1131  * Schedule the nfs_client asynchronous state management routine
1132  */
1133 void nfs4_schedule_state_manager(struct nfs_client *clp)
1134 {
1135         struct task_struct *task;
1136         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1137
1138         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1139                 return;
1140         __module_get(THIS_MODULE);
1141         atomic_inc(&clp->cl_count);
1142
1143         /* The rcu_read_lock() is not strictly necessary, as the state
1144          * manager is the only thread that ever changes the rpc_xprt
1145          * after it's initialized.  At this point, we're single threaded. */
1146         rcu_read_lock();
1147         snprintf(buf, sizeof(buf), "%s-manager",
1148                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1149         rcu_read_unlock();
1150         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1151         if (IS_ERR(task)) {
1152                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1153                         __func__, PTR_ERR(task));
1154                 nfs4_clear_state_manager_bit(clp);
1155                 nfs_put_client(clp);
1156                 module_put(THIS_MODULE);
1157         }
1158 }
1159
1160 /*
1161  * Schedule a lease recovery attempt
1162  */
1163 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1164 {
1165         if (!clp)
1166                 return;
1167         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1168                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1169         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1170                         clp->cl_hostname);
1171         nfs4_schedule_state_manager(clp);
1172 }
1173 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1174
1175 /**
1176  * nfs4_schedule_migration_recovery - trigger migration recovery
1177  *
1178  * @server: FSID that is migrating
1179  *
1180  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1181  * value is returned.
1182  */
1183 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1184 {
1185         struct nfs_client *clp = server->nfs_client;
1186
1187         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1188                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1189                                 clp->cl_hostname);
1190                 return -NFS4ERR_IO;
1191         }
1192
1193         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1194                 return -NFS4ERR_IO;
1195
1196         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1197                         __func__,
1198                         (unsigned long long)server->fsid.major,
1199                         (unsigned long long)server->fsid.minor,
1200                         clp->cl_hostname);
1201
1202         set_bit(NFS_MIG_IN_TRANSITION,
1203                         &((struct nfs_server *)server)->mig_status);
1204         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1205
1206         nfs4_schedule_state_manager(clp);
1207         return 0;
1208 }
1209 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1210
1211 /**
1212  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1213  *
1214  * @clp: server to check for moved leases
1215  *
1216  */
1217 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1218 {
1219         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1220                 __func__, clp->cl_clientid, clp->cl_hostname);
1221
1222         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1223         nfs4_schedule_state_manager(clp);
1224 }
1225 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1226
1227 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1228 {
1229         int res;
1230
1231         might_sleep();
1232
1233         atomic_inc(&clp->cl_count);
1234         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1235                                  nfs_wait_bit_killable, TASK_KILLABLE);
1236         if (res)
1237                 goto out;
1238         if (clp->cl_cons_state < 0)
1239                 res = clp->cl_cons_state;
1240 out:
1241         nfs_put_client(clp);
1242         return res;
1243 }
1244
1245 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1246 {
1247         unsigned int loop;
1248         int ret;
1249
1250         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1251                 ret = nfs4_wait_clnt_recover(clp);
1252                 if (ret != 0)
1253                         break;
1254                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1255                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1256                         break;
1257                 nfs4_schedule_state_manager(clp);
1258                 ret = -EIO;
1259         }
1260         return ret;
1261 }
1262
1263 /*
1264  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1265  * @clp: client to process
1266  *
1267  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1268  * resend of the SETCLIENTID and hence re-establish the
1269  * callback channel. Then return all existing delegations.
1270  */
1271 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1272 {
1273         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1274         nfs_expire_all_delegations(clp);
1275         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1276                         clp->cl_hostname);
1277 }
1278
1279 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1280 {
1281         nfs40_handle_cb_pathdown(clp);
1282         nfs4_schedule_state_manager(clp);
1283 }
1284
1285 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1286 {
1287
1288         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1289         /* Don't recover state that expired before the reboot */
1290         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1291                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1292                 return 0;
1293         }
1294         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1295         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1296         return 1;
1297 }
1298
1299 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1300 {
1301         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1302         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1303         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1304         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1305         return 1;
1306 }
1307
1308 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1309 {
1310         struct nfs_client *clp = server->nfs_client;
1311
1312         if (!nfs4_valid_open_stateid(state))
1313                 return -EBADF;
1314         nfs4_state_mark_reclaim_nograce(clp, state);
1315         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1316                         clp->cl_hostname);
1317         nfs4_schedule_state_manager(clp);
1318         return 0;
1319 }
1320 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1321
1322 void nfs_inode_find_state_and_recover(struct inode *inode,
1323                 const nfs4_stateid *stateid)
1324 {
1325         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1326         struct nfs_inode *nfsi = NFS_I(inode);
1327         struct nfs_open_context *ctx;
1328         struct nfs4_state *state;
1329         bool found = false;
1330
1331         spin_lock(&inode->i_lock);
1332         list_for_each_entry(ctx, &nfsi->open_files, list) {
1333                 state = ctx->state;
1334                 if (state == NULL)
1335                         continue;
1336                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1337                         continue;
1338                 if (!nfs4_stateid_match(&state->stateid, stateid))
1339                         continue;
1340                 nfs4_state_mark_reclaim_nograce(clp, state);
1341                 found = true;
1342         }
1343         spin_unlock(&inode->i_lock);
1344         if (found)
1345                 nfs4_schedule_state_manager(clp);
1346 }
1347
1348 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1349 {
1350         struct inode *inode = state->inode;
1351         struct nfs_inode *nfsi = NFS_I(inode);
1352         struct nfs_open_context *ctx;
1353
1354         spin_lock(&inode->i_lock);
1355         list_for_each_entry(ctx, &nfsi->open_files, list) {
1356                 if (ctx->state != state)
1357                         continue;
1358                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1359         }
1360         spin_unlock(&inode->i_lock);
1361 }
1362
1363 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1364 {
1365         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1366         nfs4_state_mark_open_context_bad(state);
1367 }
1368
1369
1370 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1371 {
1372         struct inode *inode = state->inode;
1373         struct nfs_inode *nfsi = NFS_I(inode);
1374         struct file_lock *fl;
1375         int status = 0;
1376
1377         if (inode->i_flock == NULL)
1378                 return 0;
1379
1380         /* Guard against delegation returns and new lock/unlock calls */
1381         down_write(&nfsi->rwsem);
1382         /* Protect inode->i_flock using the BKL */
1383         spin_lock(&inode->i_lock);
1384         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1385                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1386                         continue;
1387                 if (nfs_file_open_context(fl->fl_file)->state != state)
1388                         continue;
1389                 spin_unlock(&inode->i_lock);
1390                 status = ops->recover_lock(state, fl);
1391                 switch (status) {
1392                         case 0:
1393                                 break;
1394                         case -ESTALE:
1395                         case -NFS4ERR_ADMIN_REVOKED:
1396                         case -NFS4ERR_STALE_STATEID:
1397                         case -NFS4ERR_BAD_STATEID:
1398                         case -NFS4ERR_EXPIRED:
1399                         case -NFS4ERR_NO_GRACE:
1400                         case -NFS4ERR_STALE_CLIENTID:
1401                         case -NFS4ERR_BADSESSION:
1402                         case -NFS4ERR_BADSLOT:
1403                         case -NFS4ERR_BAD_HIGH_SLOT:
1404                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1405                                 goto out;
1406                         default:
1407                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1408                                          __func__, status);
1409                         case -ENOMEM:
1410                         case -NFS4ERR_DENIED:
1411                         case -NFS4ERR_RECLAIM_BAD:
1412                         case -NFS4ERR_RECLAIM_CONFLICT:
1413                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1414                                 status = 0;
1415                 }
1416                 spin_lock(&inode->i_lock);
1417         }
1418         spin_unlock(&inode->i_lock);
1419 out:
1420         up_write(&nfsi->rwsem);
1421         return status;
1422 }
1423
1424 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1425 {
1426         struct nfs4_state *state;
1427         struct nfs4_lock_state *lock;
1428         int status = 0;
1429
1430         /* Note: we rely on the sp->so_states list being ordered
1431          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1432          * states first.
1433          * This is needed to ensure that the server won't give us any
1434          * read delegations that we have to return if, say, we are
1435          * recovering after a network partition or a reboot from a
1436          * server that doesn't support a grace period.
1437          */
1438         spin_lock(&sp->so_lock);
1439         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1440 restart:
1441         list_for_each_entry(state, &sp->so_states, open_states) {
1442                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1443                         continue;
1444                 if (!nfs4_valid_open_stateid(state))
1445                         continue;
1446                 if (state->state == 0)
1447                         continue;
1448                 atomic_inc(&state->count);
1449                 spin_unlock(&sp->so_lock);
1450                 status = ops->recover_open(sp, state);
1451                 if (status >= 0) {
1452                         status = nfs4_reclaim_locks(state, ops);
1453                         if (status >= 0) {
1454                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1455                                         spin_lock(&state->state_lock);
1456                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1457                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1458                                                         pr_warn_ratelimited("NFS: "
1459                                                                             "%s: Lock reclaim "
1460                                                                             "failed!\n", __func__);
1461                                         }
1462                                         spin_unlock(&state->state_lock);
1463                                 }
1464                                 nfs4_put_open_state(state);
1465                                 spin_lock(&sp->so_lock);
1466                                 goto restart;
1467                         }
1468                 }
1469                 switch (status) {
1470                         default:
1471                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1472                                         __func__, status);
1473                         case -ENOENT:
1474                         case -ENOMEM:
1475                         case -ESTALE:
1476                                 /* Open state on this file cannot be recovered */
1477                                 nfs4_state_mark_recovery_failed(state, status);
1478                                 break;
1479                         case -EAGAIN:
1480                                 ssleep(1);
1481                         case -NFS4ERR_ADMIN_REVOKED:
1482                         case -NFS4ERR_STALE_STATEID:
1483                         case -NFS4ERR_BAD_STATEID:
1484                         case -NFS4ERR_RECLAIM_BAD:
1485                         case -NFS4ERR_RECLAIM_CONFLICT:
1486                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1487                                 break;
1488                         case -NFS4ERR_EXPIRED:
1489                         case -NFS4ERR_NO_GRACE:
1490                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1491                         case -NFS4ERR_STALE_CLIENTID:
1492                         case -NFS4ERR_BADSESSION:
1493                         case -NFS4ERR_BADSLOT:
1494                         case -NFS4ERR_BAD_HIGH_SLOT:
1495                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1496                                 goto out_err;
1497                 }
1498                 nfs4_put_open_state(state);
1499                 spin_lock(&sp->so_lock);
1500                 goto restart;
1501         }
1502         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1503         spin_unlock(&sp->so_lock);
1504         return 0;
1505 out_err:
1506         nfs4_put_open_state(state);
1507         spin_lock(&sp->so_lock);
1508         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1509         spin_unlock(&sp->so_lock);
1510         return status;
1511 }
1512
1513 static void nfs4_clear_open_state(struct nfs4_state *state)
1514 {
1515         struct nfs4_lock_state *lock;
1516
1517         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1518         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1519         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1520         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1521         spin_lock(&state->state_lock);
1522         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1523                 lock->ls_seqid.flags = 0;
1524                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1525         }
1526         spin_unlock(&state->state_lock);
1527 }
1528
1529 static void nfs4_reset_seqids(struct nfs_server *server,
1530         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1531 {
1532         struct nfs_client *clp = server->nfs_client;
1533         struct nfs4_state_owner *sp;
1534         struct rb_node *pos;
1535         struct nfs4_state *state;
1536
1537         spin_lock(&clp->cl_lock);
1538         for (pos = rb_first(&server->state_owners);
1539              pos != NULL;
1540              pos = rb_next(pos)) {
1541                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1542                 sp->so_seqid.flags = 0;
1543                 spin_lock(&sp->so_lock);
1544                 list_for_each_entry(state, &sp->so_states, open_states) {
1545                         if (mark_reclaim(clp, state))
1546                                 nfs4_clear_open_state(state);
1547                 }
1548                 spin_unlock(&sp->so_lock);
1549         }
1550         spin_unlock(&clp->cl_lock);
1551 }
1552
1553 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1554         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1555 {
1556         struct nfs_server *server;
1557
1558         rcu_read_lock();
1559         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1560                 nfs4_reset_seqids(server, mark_reclaim);
1561         rcu_read_unlock();
1562 }
1563
1564 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1565 {
1566         /* Mark all delegations for reclaim */
1567         nfs_delegation_mark_reclaim(clp);
1568         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1569 }
1570
1571 static void nfs4_reclaim_complete(struct nfs_client *clp,
1572                                  const struct nfs4_state_recovery_ops *ops,
1573                                  struct rpc_cred *cred)
1574 {
1575         /* Notify the server we're done reclaiming our state */
1576         if (ops->reclaim_complete)
1577                 (void)ops->reclaim_complete(clp, cred);
1578 }
1579
1580 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1581 {
1582         struct nfs_client *clp = server->nfs_client;
1583         struct nfs4_state_owner *sp;
1584         struct rb_node *pos;
1585         struct nfs4_state *state;
1586
1587         spin_lock(&clp->cl_lock);
1588         for (pos = rb_first(&server->state_owners);
1589              pos != NULL;
1590              pos = rb_next(pos)) {
1591                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1592                 spin_lock(&sp->so_lock);
1593                 list_for_each_entry(state, &sp->so_states, open_states) {
1594                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1595                                                 &state->flags))
1596                                 continue;
1597                         nfs4_state_mark_reclaim_nograce(clp, state);
1598                 }
1599                 spin_unlock(&sp->so_lock);
1600         }
1601         spin_unlock(&clp->cl_lock);
1602 }
1603
1604 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1605 {
1606         struct nfs_server *server;
1607
1608         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1609                 return 0;
1610
1611         rcu_read_lock();
1612         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1613                 nfs4_clear_reclaim_server(server);
1614         rcu_read_unlock();
1615
1616         nfs_delegation_reap_unclaimed(clp);
1617         return 1;
1618 }
1619
1620 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1621 {
1622         const struct nfs4_state_recovery_ops *ops;
1623         struct rpc_cred *cred;
1624
1625         if (!nfs4_state_clear_reclaim_reboot(clp))
1626                 return;
1627         ops = clp->cl_mvops->reboot_recovery_ops;
1628         cred = nfs4_get_clid_cred(clp);
1629         nfs4_reclaim_complete(clp, ops, cred);
1630         put_rpccred(cred);
1631 }
1632
1633 static void nfs_delegation_clear_all(struct nfs_client *clp)
1634 {
1635         nfs_delegation_mark_reclaim(clp);
1636         nfs_delegation_reap_unclaimed(clp);
1637 }
1638
1639 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1640 {
1641         nfs_delegation_clear_all(clp);
1642         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1643 }
1644
1645 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1646 {
1647         switch (error) {
1648                 case 0:
1649                         break;
1650                 case -NFS4ERR_CB_PATH_DOWN:
1651                         nfs40_handle_cb_pathdown(clp);
1652                         break;
1653                 case -NFS4ERR_NO_GRACE:
1654                         nfs4_state_end_reclaim_reboot(clp);
1655                         break;
1656                 case -NFS4ERR_STALE_CLIENTID:
1657                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1658                         nfs4_state_clear_reclaim_reboot(clp);
1659                         nfs4_state_start_reclaim_reboot(clp);
1660                         break;
1661                 case -NFS4ERR_EXPIRED:
1662                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1663                         nfs4_state_start_reclaim_nograce(clp);
1664                         break;
1665                 case -NFS4ERR_BADSESSION:
1666                 case -NFS4ERR_BADSLOT:
1667                 case -NFS4ERR_BAD_HIGH_SLOT:
1668                 case -NFS4ERR_DEADSESSION:
1669                 case -NFS4ERR_SEQ_FALSE_RETRY:
1670                 case -NFS4ERR_SEQ_MISORDERED:
1671                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1672                         /* Zero session reset errors */
1673                         break;
1674                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1675                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1676                         break;
1677                 default:
1678                         dprintk("%s: failed to handle error %d for server %s\n",
1679                                         __func__, error, clp->cl_hostname);
1680                         return error;
1681         }
1682         dprintk("%s: handled error %d for server %s\n", __func__, error,
1683                         clp->cl_hostname);
1684         return 0;
1685 }
1686
1687 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1688 {
1689         struct nfs4_state_owner *sp;
1690         struct nfs_server *server;
1691         struct rb_node *pos;
1692         int status = 0;
1693
1694 restart:
1695         rcu_read_lock();
1696         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1697                 nfs4_purge_state_owners(server);
1698                 spin_lock(&clp->cl_lock);
1699                 for (pos = rb_first(&server->state_owners);
1700                      pos != NULL;
1701                      pos = rb_next(pos)) {
1702                         sp = rb_entry(pos,
1703                                 struct nfs4_state_owner, so_server_node);
1704                         if (!test_and_clear_bit(ops->owner_flag_bit,
1705                                                         &sp->so_flags))
1706                                 continue;
1707                         atomic_inc(&sp->so_count);
1708                         spin_unlock(&clp->cl_lock);
1709                         rcu_read_unlock();
1710
1711                         status = nfs4_reclaim_open_state(sp, ops);
1712                         if (status < 0) {
1713                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1714                                 nfs4_put_state_owner(sp);
1715                                 status = nfs4_recovery_handle_error(clp, status);
1716                                 return (status != 0) ? status : -EAGAIN;
1717                         }
1718
1719                         nfs4_put_state_owner(sp);
1720                         goto restart;
1721                 }
1722                 spin_unlock(&clp->cl_lock);
1723         }
1724         rcu_read_unlock();
1725         return 0;
1726 }
1727
1728 static int nfs4_check_lease(struct nfs_client *clp)
1729 {
1730         struct rpc_cred *cred;
1731         const struct nfs4_state_maintenance_ops *ops =
1732                 clp->cl_mvops->state_renewal_ops;
1733         int status;
1734
1735         /* Is the client already known to have an expired lease? */
1736         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1737                 return 0;
1738         spin_lock(&clp->cl_lock);
1739         cred = ops->get_state_renewal_cred_locked(clp);
1740         spin_unlock(&clp->cl_lock);
1741         if (cred == NULL) {
1742                 cred = nfs4_get_clid_cred(clp);
1743                 status = -ENOKEY;
1744                 if (cred == NULL)
1745                         goto out;
1746         }
1747         status = ops->renew_lease(clp, cred);
1748         put_rpccred(cred);
1749         if (status == -ETIMEDOUT) {
1750                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1751                 return 0;
1752         }
1753 out:
1754         return nfs4_recovery_handle_error(clp, status);
1755 }
1756
1757 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1758  * and for recoverable errors on EXCHANGE_ID for v4.1
1759  */
1760 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1761 {
1762         switch (status) {
1763         case -NFS4ERR_SEQ_MISORDERED:
1764                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1765                         return -ESERVERFAULT;
1766                 /* Lease confirmation error: retry after purging the lease */
1767                 ssleep(1);
1768                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1769                 break;
1770         case -NFS4ERR_STALE_CLIENTID:
1771                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1772                 nfs4_state_start_reclaim_reboot(clp);
1773                 break;
1774         case -NFS4ERR_CLID_INUSE:
1775                 pr_err("NFS: Server %s reports our clientid is in use\n",
1776                         clp->cl_hostname);
1777                 nfs_mark_client_ready(clp, -EPERM);
1778                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1779                 return -EPERM;
1780         case -EACCES:
1781         case -NFS4ERR_DELAY:
1782         case -ETIMEDOUT:
1783         case -EAGAIN:
1784                 ssleep(1);
1785                 break;
1786
1787         case -NFS4ERR_MINOR_VERS_MISMATCH:
1788                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1789                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1790                 dprintk("%s: exit with error %d for server %s\n",
1791                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1792                 return -EPROTONOSUPPORT;
1793         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1794                                  * in nfs4_exchange_id */
1795         default:
1796                 dprintk("%s: exit with error %d for server %s\n", __func__,
1797                                 status, clp->cl_hostname);
1798                 return status;
1799         }
1800         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1801         dprintk("%s: handled error %d for server %s\n", __func__, status,
1802                         clp->cl_hostname);
1803         return 0;
1804 }
1805
1806 static int nfs4_establish_lease(struct nfs_client *clp)
1807 {
1808         struct rpc_cred *cred;
1809         const struct nfs4_state_recovery_ops *ops =
1810                 clp->cl_mvops->reboot_recovery_ops;
1811         int status;
1812
1813         cred = nfs4_get_clid_cred(clp);
1814         if (cred == NULL)
1815                 return -ENOENT;
1816         status = ops->establish_clid(clp, cred);
1817         put_rpccred(cred);
1818         if (status != 0)
1819                 return status;
1820         pnfs_destroy_all_layouts(clp);
1821         return 0;
1822 }
1823
1824 /*
1825  * Returns zero or a negative errno.  NFS4ERR values are converted
1826  * to local errno values.
1827  */
1828 static int nfs4_reclaim_lease(struct nfs_client *clp)
1829 {
1830         int status;
1831
1832         status = nfs4_establish_lease(clp);
1833         if (status < 0)
1834                 return nfs4_handle_reclaim_lease_error(clp, status);
1835         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1836                 nfs4_state_start_reclaim_nograce(clp);
1837         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1838                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1839         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1840         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1841         return 0;
1842 }
1843
1844 static int nfs4_purge_lease(struct nfs_client *clp)
1845 {
1846         int status;
1847
1848         status = nfs4_establish_lease(clp);
1849         if (status < 0)
1850                 return nfs4_handle_reclaim_lease_error(clp, status);
1851         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1852         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1853         nfs4_state_start_reclaim_nograce(clp);
1854         return 0;
1855 }
1856
1857 /*
1858  * Try remote migration of one FSID from a source server to a
1859  * destination server.  The source server provides a list of
1860  * potential destinations.
1861  *
1862  * Returns zero or a negative NFS4ERR status code.
1863  */
1864 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1865 {
1866         struct nfs_client *clp = server->nfs_client;
1867         struct nfs4_fs_locations *locations = NULL;
1868         struct inode *inode;
1869         struct page *page;
1870         int status, result;
1871
1872         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1873                         (unsigned long long)server->fsid.major,
1874                         (unsigned long long)server->fsid.minor,
1875                         clp->cl_hostname);
1876
1877         result = 0;
1878         page = alloc_page(GFP_KERNEL);
1879         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1880         if (page == NULL || locations == NULL) {
1881                 dprintk("<-- %s: no memory\n", __func__);
1882                 goto out;
1883         }
1884
1885         inode = server->super->s_root->d_inode;
1886         result = nfs4_proc_get_locations(inode, locations, page, cred);
1887         if (result) {
1888                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1889                         __func__, result);
1890                 goto out;
1891         }
1892
1893         result = -NFS4ERR_NXIO;
1894         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1895                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1896                         __func__);
1897                 goto out;
1898         }
1899
1900         nfs4_begin_drain_session(clp);
1901
1902         status = nfs4_replace_transport(server, locations);
1903         if (status != 0) {
1904                 dprintk("<-- %s: failed to replace transport: %d\n",
1905                         __func__, status);
1906                 goto out;
1907         }
1908
1909         result = 0;
1910         dprintk("<-- %s: migration succeeded\n", __func__);
1911
1912 out:
1913         if (page != NULL)
1914                 __free_page(page);
1915         kfree(locations);
1916         if (result) {
1917                 pr_err("NFS: migration recovery failed (server %s)\n",
1918                                 clp->cl_hostname);
1919                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1920         }
1921         return result;
1922 }
1923
1924 /*
1925  * Returns zero or a negative NFS4ERR status code.
1926  */
1927 static int nfs4_handle_migration(struct nfs_client *clp)
1928 {
1929         const struct nfs4_state_maintenance_ops *ops =
1930                                 clp->cl_mvops->state_renewal_ops;
1931         struct nfs_server *server;
1932         struct rpc_cred *cred;
1933
1934         dprintk("%s: migration reported on \"%s\"\n", __func__,
1935                         clp->cl_hostname);
1936
1937         spin_lock(&clp->cl_lock);
1938         cred = ops->get_state_renewal_cred_locked(clp);
1939         spin_unlock(&clp->cl_lock);
1940         if (cred == NULL)
1941                 return -NFS4ERR_NOENT;
1942
1943         clp->cl_mig_gen++;
1944 restart:
1945         rcu_read_lock();
1946         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1947                 int status;
1948
1949                 if (server->mig_gen == clp->cl_mig_gen)
1950                         continue;
1951                 server->mig_gen = clp->cl_mig_gen;
1952
1953                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
1954                                                 &server->mig_status))
1955                         continue;
1956
1957                 rcu_read_unlock();
1958                 status = nfs4_try_migration(server, cred);
1959                 if (status < 0) {
1960                         put_rpccred(cred);
1961                         return status;
1962                 }
1963                 goto restart;
1964         }
1965         rcu_read_unlock();
1966         put_rpccred(cred);
1967         return 0;
1968 }
1969
1970 /*
1971  * Test each nfs_server on the clp's cl_superblocks list to see
1972  * if it's moved to another server.  Stop when the server no longer
1973  * returns NFS4ERR_LEASE_MOVED.
1974  */
1975 static int nfs4_handle_lease_moved(struct nfs_client *clp)
1976 {
1977         const struct nfs4_state_maintenance_ops *ops =
1978                                 clp->cl_mvops->state_renewal_ops;
1979         struct nfs_server *server;
1980         struct rpc_cred *cred;
1981
1982         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
1983                         clp->cl_hostname);
1984
1985         spin_lock(&clp->cl_lock);
1986         cred = ops->get_state_renewal_cred_locked(clp);
1987         spin_unlock(&clp->cl_lock);
1988         if (cred == NULL)
1989                 return -NFS4ERR_NOENT;
1990
1991         clp->cl_mig_gen++;
1992 restart:
1993         rcu_read_lock();
1994         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1995                 struct inode *inode;
1996                 int status;
1997
1998                 if (server->mig_gen == clp->cl_mig_gen)
1999                         continue;
2000                 server->mig_gen = clp->cl_mig_gen;
2001
2002                 rcu_read_unlock();
2003
2004                 inode = server->super->s_root->d_inode;
2005                 status = nfs4_proc_fsid_present(inode, cred);
2006                 if (status != -NFS4ERR_MOVED)
2007                         goto restart;   /* wasn't this one */
2008                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2009                         goto restart;   /* there are more */
2010                 goto out;
2011         }
2012         rcu_read_unlock();
2013
2014 out:
2015         put_rpccred(cred);
2016         return 0;
2017 }
2018
2019 /**
2020  * nfs4_discover_server_trunking - Detect server IP address trunking
2021  *
2022  * @clp: nfs_client under test
2023  * @result: OUT: found nfs_client, or clp
2024  *
2025  * Returns zero or a negative errno.  If zero is returned,
2026  * an nfs_client pointer is planted in "result".
2027  *
2028  * Note: since we are invoked in process context, and
2029  * not from inside the state manager, we cannot use
2030  * nfs4_handle_reclaim_lease_error().
2031  */
2032 int nfs4_discover_server_trunking(struct nfs_client *clp,
2033                                   struct nfs_client **result)
2034 {
2035         const struct nfs4_state_recovery_ops *ops =
2036                                 clp->cl_mvops->reboot_recovery_ops;
2037         struct rpc_clnt *clnt;
2038         struct rpc_cred *cred;
2039         int i, status;
2040
2041         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2042
2043         clnt = clp->cl_rpcclient;
2044         i = 0;
2045
2046         mutex_lock(&nfs_clid_init_mutex);
2047 again:
2048         status  = -ENOENT;
2049         cred = nfs4_get_clid_cred(clp);
2050         if (cred == NULL)
2051                 goto out_unlock;
2052
2053         status = ops->detect_trunking(clp, result, cred);
2054         put_rpccred(cred);
2055         switch (status) {
2056         case 0:
2057                 break;
2058         case -ETIMEDOUT:
2059                 if (clnt->cl_softrtry)
2060                         break;
2061         case -NFS4ERR_DELAY:
2062         case -EAGAIN:
2063                 ssleep(1);
2064         case -NFS4ERR_STALE_CLIENTID:
2065                 dprintk("NFS: %s after status %d, retrying\n",
2066                         __func__, status);
2067                 goto again;
2068         case -EACCES:
2069                 if (i++ == 0) {
2070                         nfs4_root_machine_cred(clp);
2071                         goto again;
2072                 }
2073                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2074                         break;
2075         case -NFS4ERR_CLID_INUSE:
2076         case -NFS4ERR_WRONGSEC:
2077                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2078                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2079                         status = -EPERM;
2080                         break;
2081                 }
2082                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2083                 if (IS_ERR(clnt)) {
2084                         status = PTR_ERR(clnt);
2085                         break;
2086                 }
2087                 /* Note: this is safe because we haven't yet marked the
2088                  * client as ready, so we are the only user of
2089                  * clp->cl_rpcclient
2090                  */
2091                 clnt = xchg(&clp->cl_rpcclient, clnt);
2092                 rpc_shutdown_client(clnt);
2093                 clnt = clp->cl_rpcclient;
2094                 goto again;
2095
2096         case -NFS4ERR_MINOR_VERS_MISMATCH:
2097                 status = -EPROTONOSUPPORT;
2098                 break;
2099
2100         case -EKEYEXPIRED:
2101         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2102                                  * in nfs4_exchange_id */
2103                 status = -EKEYEXPIRED;
2104                 break;
2105         default:
2106                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2107                                 __func__, status);
2108                 status = -EIO;
2109         }
2110
2111 out_unlock:
2112         mutex_unlock(&nfs_clid_init_mutex);
2113         dprintk("NFS: %s: status = %d\n", __func__, status);
2114         return status;
2115 }
2116
2117 #ifdef CONFIG_NFS_V4_1
2118 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2119 {
2120         struct nfs_client *clp = session->clp;
2121
2122         switch (err) {
2123         default:
2124                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2125                 break;
2126         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2127                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2128         }
2129         nfs4_schedule_lease_recovery(clp);
2130 }
2131 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2132
2133 static void nfs41_ping_server(struct nfs_client *clp)
2134 {
2135         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2136         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2137         nfs4_schedule_state_manager(clp);
2138 }
2139
2140 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
2141 {
2142         nfs41_ping_server(clp);
2143 }
2144
2145 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
2146 {
2147         nfs41_ping_server(clp);
2148 }
2149
2150 static void nfs4_reset_all_state(struct nfs_client *clp)
2151 {
2152         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2153                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2154                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2155                 nfs4_state_start_reclaim_nograce(clp);
2156                 dprintk("%s: scheduling reset of all state for server %s!\n",
2157                                 __func__, clp->cl_hostname);
2158                 nfs4_schedule_state_manager(clp);
2159         }
2160 }
2161
2162 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2163 {
2164         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2165                 nfs4_state_start_reclaim_reboot(clp);
2166                 dprintk("%s: server %s rebooted!\n", __func__,
2167                                 clp->cl_hostname);
2168                 nfs4_schedule_state_manager(clp);
2169         }
2170 }
2171
2172 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2173 {
2174         nfs4_reset_all_state(clp);
2175         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2176 }
2177
2178 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2179 {
2180         /* This will need to handle layouts too */
2181         nfs_expire_all_delegations(clp);
2182         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2183                         clp->cl_hostname);
2184 }
2185
2186 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2187 {
2188         nfs_expire_all_delegations(clp);
2189         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2190                 nfs4_schedule_state_manager(clp);
2191         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2192                         clp->cl_hostname);
2193 }
2194
2195 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2196 {
2197         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2198                 &clp->cl_state) == 0)
2199                 nfs4_schedule_state_manager(clp);
2200 }
2201
2202 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2203 {
2204         if (!flags)
2205                 return;
2206
2207         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2208                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2209
2210         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2211                 nfs41_handle_server_reboot(clp);
2212         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2213                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2214                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2215                 nfs41_handle_state_revoked(clp);
2216         if (flags & SEQ4_STATUS_LEASE_MOVED)
2217                 nfs4_schedule_lease_moved_recovery(clp);
2218         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2219                 nfs41_handle_recallable_state_revoked(clp);
2220         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2221                 nfs41_handle_backchannel_fault(clp);
2222         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2223                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2224                 nfs41_handle_cb_path_down(clp);
2225 }
2226
2227 static int nfs4_reset_session(struct nfs_client *clp)
2228 {
2229         struct rpc_cred *cred;
2230         int status;
2231
2232         if (!nfs4_has_session(clp))
2233                 return 0;
2234         nfs4_begin_drain_session(clp);
2235         cred = nfs4_get_clid_cred(clp);
2236         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2237         switch (status) {
2238         case 0:
2239         case -NFS4ERR_BADSESSION:
2240         case -NFS4ERR_DEADSESSION:
2241                 break;
2242         case -NFS4ERR_BACK_CHAN_BUSY:
2243         case -NFS4ERR_DELAY:
2244                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2245                 status = 0;
2246                 ssleep(1);
2247                 goto out;
2248         default:
2249                 status = nfs4_recovery_handle_error(clp, status);
2250                 goto out;
2251         }
2252
2253         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2254         status = nfs4_proc_create_session(clp, cred);
2255         if (status) {
2256                 dprintk("%s: session reset failed with status %d for server %s!\n",
2257                         __func__, status, clp->cl_hostname);
2258                 status = nfs4_handle_reclaim_lease_error(clp, status);
2259                 goto out;
2260         }
2261         nfs41_finish_session_reset(clp);
2262         dprintk("%s: session reset was successful for server %s!\n",
2263                         __func__, clp->cl_hostname);
2264 out:
2265         if (cred)
2266                 put_rpccred(cred);
2267         return status;
2268 }
2269
2270 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2271 {
2272         struct rpc_cred *cred;
2273         int ret;
2274
2275         if (!nfs4_has_session(clp))
2276                 return 0;
2277         nfs4_begin_drain_session(clp);
2278         cred = nfs4_get_clid_cred(clp);
2279         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2280         if (cred)
2281                 put_rpccred(cred);
2282         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2283         switch (ret) {
2284         case 0:
2285                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2286                         __func__, clp->cl_hostname);
2287                 break;
2288         case -NFS4ERR_DELAY:
2289                 ssleep(1);
2290                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2291                 break;
2292         default:
2293                 return nfs4_recovery_handle_error(clp, ret);
2294         }
2295         return 0;
2296 }
2297 #else /* CONFIG_NFS_V4_1 */
2298 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2299
2300 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2301 {
2302         return 0;
2303 }
2304 #endif /* CONFIG_NFS_V4_1 */
2305
2306 static void nfs4_state_manager(struct nfs_client *clp)
2307 {
2308         int status = 0;
2309         const char *section = "", *section_sep = "";
2310
2311         /* Ensure exclusive access to NFSv4 state */
2312         do {
2313                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2314                         section = "purge state";
2315                         status = nfs4_purge_lease(clp);
2316                         if (status < 0)
2317                                 goto out_error;
2318                         continue;
2319                 }
2320
2321                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2322                         section = "lease expired";
2323                         /* We're going to have to re-establish a clientid */
2324                         status = nfs4_reclaim_lease(clp);
2325                         if (status < 0)
2326                                 goto out_error;
2327                         continue;
2328                 }
2329
2330                 /* Initialize or reset the session */
2331                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2332                         section = "reset session";
2333                         status = nfs4_reset_session(clp);
2334                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2335                                 continue;
2336                         if (status < 0)
2337                                 goto out_error;
2338                 }
2339
2340                 /* Send BIND_CONN_TO_SESSION */
2341                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2342                                 &clp->cl_state)) {
2343                         section = "bind conn to session";
2344                         status = nfs4_bind_conn_to_session(clp);
2345                         if (status < 0)
2346                                 goto out_error;
2347                         continue;
2348                 }
2349
2350                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2351                         section = "check lease";
2352                         status = nfs4_check_lease(clp);
2353                         if (status < 0)
2354                                 goto out_error;
2355                         continue;
2356                 }
2357
2358                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2359                         section = "migration";
2360                         status = nfs4_handle_migration(clp);
2361                         if (status < 0)
2362                                 goto out_error;
2363                 }
2364
2365                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2366                         section = "lease moved";
2367                         status = nfs4_handle_lease_moved(clp);
2368                         if (status < 0)
2369                                 goto out_error;
2370                 }
2371
2372                 /* First recover reboot state... */
2373                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2374                         section = "reclaim reboot";
2375                         status = nfs4_do_reclaim(clp,
2376                                 clp->cl_mvops->reboot_recovery_ops);
2377                         if (status == -EAGAIN)
2378                                 continue;
2379                         if (status < 0)
2380                                 goto out_error;
2381                         nfs4_state_end_reclaim_reboot(clp);
2382                 }
2383
2384                 /* Now recover expired state... */
2385                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2386                         section = "reclaim nograce";
2387                         status = nfs4_do_reclaim(clp,
2388                                 clp->cl_mvops->nograce_recovery_ops);
2389                         if (status == -EAGAIN)
2390                                 continue;
2391                         if (status < 0)
2392                                 goto out_error;
2393                 }
2394
2395                 nfs4_end_drain_session(clp);
2396                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2397                         nfs_client_return_marked_delegations(clp);
2398                         continue;
2399                 }
2400
2401                 nfs4_clear_state_manager_bit(clp);
2402                 /* Did we race with an attempt to give us more work? */
2403                 if (clp->cl_state == 0)
2404                         break;
2405                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2406                         break;
2407         } while (atomic_read(&clp->cl_count) > 1);
2408         return;
2409 out_error:
2410         if (strlen(section))
2411                 section_sep = ": ";
2412         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2413                         " with error %d\n", section_sep, section,
2414                         clp->cl_hostname, -status);
2415         ssleep(1);
2416         nfs4_end_drain_session(clp);
2417         nfs4_clear_state_manager_bit(clp);
2418 }
2419
2420 static int nfs4_run_state_manager(void *ptr)
2421 {
2422         struct nfs_client *clp = ptr;
2423
2424         allow_signal(SIGKILL);
2425         nfs4_state_manager(clp);
2426         nfs_put_client(clp);
2427         module_put_and_exit(0);
2428         return 0;
2429 }
2430
2431 /*
2432  * Local variables:
2433  *  c-basic-offset: 8
2434  * End:
2435  */