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 #if defined(CONFIG_NFS_V4_1)
 243
 244 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 245 {
 246         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 247         spin_lock(&tbl->slot_tbl_lock);
 248         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 249                 INIT_COMPLETION(tbl->complete);
 250                 spin_unlock(&tbl->slot_tbl_lock);
 251                 return wait_for_completion_interruptible(&tbl->complete);
 252         }
 253         spin_unlock(&tbl->slot_tbl_lock);
 254         return 0;
 255 }
 256
 257 static int nfs4_begin_drain_session(struct nfs_client *clp)
 258 {
 259         struct nfs4_session *ses = clp->cl_session;
 260         int ret = 0;
 261
 262         if (clp->cl_slot_tbl)
 263                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 264
 265         /* back channel */
 266         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 267         if (ret)
 268                 return ret;
 269         /* fore channel */
 270         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 271 }
 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, pid_t fl_pid, unsigned int type)
 791 {
 792         struct nfs4_lock_state *pos;
 793         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 794                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 795                         continue;
 796                 switch (pos->ls_owner.lo_type) {
 797                 case NFS4_POSIX_LOCK_TYPE:
 798                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 799                                 continue;
 800                         break;
 801                 case NFS4_FLOCK_LOCK_TYPE:
 802                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 803                                 continue;
 804                 }
 805                 atomic_inc(&pos->ls_count);
 806                 return pos;
 807         }
 808         return NULL;
 809 }
 810
 811 /*
 812  * Return a compatible lock_state. If no initialized lock_state structure
 813  * exists, return an uninitialized one.
 814  *
 815  */
 816 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 817 {
 818         struct nfs4_lock_state *lsp;
 819         struct nfs_server *server = state->owner->so_server;
 820
 821         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 822         if (lsp == NULL)
 823                 return NULL;
 824         nfs4_init_seqid_counter(&lsp->ls_seqid);
 825         atomic_set(&lsp->ls_count, 1);
 826         lsp->ls_state = state;
 827         lsp->ls_owner.lo_type = type;
 828         switch (lsp->ls_owner.lo_type) {
 829         case NFS4_FLOCK_LOCK_TYPE:
 830                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 831                 break;
 832         case NFS4_POSIX_LOCK_TYPE:
 833                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 834                 break;
 835         default:
 836                 goto out_free;
 837         }
 838         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 839         if (lsp->ls_seqid.owner_id < 0)
 840                 goto out_free;
 841         INIT_LIST_HEAD(&lsp->ls_locks);
 842         return lsp;
 843 out_free:
 844         kfree(lsp);
 845         return NULL;
 846 }
 847
 848 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 849 {
 850         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 851         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 852         kfree(lsp);
 853 }
 854
 855 /*
 856  * Return a compatible lock_state. If no initialized lock_state structure
 857  * exists, return an uninitialized one.
 858  *
 859  */
 860 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 861 {
 862         struct nfs4_lock_state *lsp, *new = NULL;
 863
 864         for(;;) {
 865                 spin_lock(&state->state_lock);
 866                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 867                 if (lsp != NULL)
 868                         break;
 869                 if (new != NULL) {
 870                         list_add(&new->ls_locks, &state->lock_states);
 871                         set_bit(LK_STATE_IN_USE, &state->flags);
 872                         lsp = new;
 873                         new = NULL;
 874                         break;
 875                 }
 876                 spin_unlock(&state->state_lock);
 877                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 878                 if (new == NULL)
 879                         return NULL;
 880         }
 881         spin_unlock(&state->state_lock);
 882         if (new != NULL)
 883                 nfs4_free_lock_state(state->owner->so_server, new);
 884         return lsp;
 885 }
 886
 887 /*
 888  * Release reference to lock_state, and free it if we see that
 889  * it is no longer in use
 890  */
 891 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 892 {
 893         struct nfs_server *server;
 894         struct nfs4_state *state;
 895
 896         if (lsp == NULL)
 897                 return;
 898         state = lsp->ls_state;
 899         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 900                 return;
 901         list_del(&lsp->ls_locks);
 902         if (list_empty(&state->lock_states))
 903                 clear_bit(LK_STATE_IN_USE, &state->flags);
 904         spin_unlock(&state->state_lock);
 905         server = state->owner->so_server;
 906         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 907                 struct nfs_client *clp = server->nfs_client;
 908
 909                 clp->cl_mvops->free_lock_state(server, lsp);
 910         } else
 911                 nfs4_free_lock_state(server, lsp);
 912 }
 913
 914 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 915 {
 916         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 917
 918         dst->fl_u.nfs4_fl.owner = lsp;
 919         atomic_inc(&lsp->ls_count);
 920 }
 921
 922 static void nfs4_fl_release_lock(struct file_lock *fl)
 923 {
 924         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 925 }
 926
 927 static const struct file_lock_operations nfs4_fl_lock_ops = {
 928         .fl_copy_lock = nfs4_fl_copy_lock,
 929         .fl_release_private = nfs4_fl_release_lock,
 930 };
 931
 932 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 933 {
 934         struct nfs4_lock_state *lsp;
 935
 936         if (fl->fl_ops != NULL)
 937                 return 0;
 938         if (fl->fl_flags & FL_POSIX)
 939                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 940         else if (fl->fl_flags & FL_FLOCK)
 941                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 942                                 NFS4_FLOCK_LOCK_TYPE);
 943         else
 944                 return -EINVAL;
 945         if (lsp == NULL)
 946                 return -ENOMEM;
 947         fl->fl_u.nfs4_fl.owner = lsp;
 948         fl->fl_ops = &nfs4_fl_lock_ops;
 949         return 0;
 950 }
 951
 952 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 953                 struct nfs4_state *state,
 954                 const struct nfs_lockowner *lockowner)
 955 {
 956         struct nfs4_lock_state *lsp;
 957         fl_owner_t fl_owner;
 958         pid_t fl_pid;
 959         int ret = -ENOENT;
 960
 961
 962         if (lockowner == NULL)
 963                 goto out;
 964
 965         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 966                 goto out;
 967
 968         fl_owner = lockowner->l_owner;
 969         fl_pid = lockowner->l_pid;
 970         spin_lock(&state->state_lock);
 971         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 972         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 973                 ret = -EIO;
 974         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 975                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 976                 ret = 0;
 977                 smp_rmb();
 978                 if (!list_empty(&lsp->ls_seqid.list))
 979                         ret = -EWOULDBLOCK;
 980         }
 981         spin_unlock(&state->state_lock);
 982         nfs4_put_lock_state(lsp);
 983 out:
 984         return ret;
 985 }
 986
 987 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 988 {
 989         const nfs4_stateid *src;
 990         int ret;
 991         int seq;
 992
 993         do {
 994                 src = &zero_stateid;
 995                 seq = read_seqbegin(&state->seqlock);
 996                 if (test_bit(NFS_OPEN_STATE, &state->flags))
 997                         src = &state->open_stateid;
 998                 nfs4_stateid_copy(dst, src);
 999                 ret = 0;
1000                 smp_rmb();
1001                 if (!list_empty(&state->owner->so_seqid.list))
1002                         ret = -EWOULDBLOCK;
1003         } while (read_seqretry(&state->seqlock, seq));
1004         return ret;
1005 }
1006
1007 /*
1008  * Byte-range lock aware utility to initialize the stateid of read/write
1009  * requests.
1010  */
1011 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1012                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1013 {
1014         int ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1015         if (ret == -EIO)
1016                 /* A lost lock - don't even consider delegations */
1017                 goto out;
1018         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1019                 goto out;
1020         if (ret != -ENOENT)
1021                 /* nfs4_copy_delegation_stateid() didn't over-write
1022                  * dst, so it still has the lock stateid which we now
1023                  * choose to use.
1024                  */
1025                 goto out;
1026         ret = nfs4_copy_open_stateid(dst, state);
1027 out:
1028         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1029                 dst->seqid = 0;
1030         return ret;
1031 }
1032
1033 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1034 {
1035         struct nfs_seqid *new;
1036
1037         new = kmalloc(sizeof(*new), gfp_mask);
1038         if (new != NULL) {
1039                 new->sequence = counter;
1040                 INIT_LIST_HEAD(&new->list);
1041                 new->task = NULL;
1042         }
1043         return new;
1044 }
1045
1046 void nfs_release_seqid(struct nfs_seqid *seqid)
1047 {
1048         struct nfs_seqid_counter *sequence;
1049
1050         if (list_empty(&seqid->list))
1051                 return;
1052         sequence = seqid->sequence;
1053         spin_lock(&sequence->lock);
1054         list_del_init(&seqid->list);
1055         if (!list_empty(&sequence->list)) {
1056                 struct nfs_seqid *next;
1057
1058                 next = list_first_entry(&sequence->list,
1059                                 struct nfs_seqid, list);
1060                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1061         }
1062         spin_unlock(&sequence->lock);
1063 }
1064
1065 void nfs_free_seqid(struct nfs_seqid *seqid)
1066 {
1067         nfs_release_seqid(seqid);
1068         kfree(seqid);
1069 }
1070
1071 /*
1072  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1073  * failed with a seqid incrementing error -
1074  * see comments nfs_fs.h:seqid_mutating_error()
1075  */
1076 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1077 {
1078         switch (status) {
1079                 case 0:
1080                         break;
1081                 case -NFS4ERR_BAD_SEQID:
1082                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1083                                 return;
1084                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1085                                         " sequence-id error on an"
1086                                         " unconfirmed sequence %p!\n",
1087                                         seqid->sequence);
1088                 case -NFS4ERR_STALE_CLIENTID:
1089                 case -NFS4ERR_STALE_STATEID:
1090                 case -NFS4ERR_BAD_STATEID:
1091                 case -NFS4ERR_BADXDR:
1092                 case -NFS4ERR_RESOURCE:
1093                 case -NFS4ERR_NOFILEHANDLE:
1094                         /* Non-seqid mutating errors */
1095                         return;
1096         };
1097         /*
1098          * Note: no locking needed as we are guaranteed to be first
1099          * on the sequence list
1100          */
1101         seqid->sequence->counter++;
1102 }
1103
1104 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1105 {
1106         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1107                                         struct nfs4_state_owner, so_seqid);
1108         struct nfs_server *server = sp->so_server;
1109
1110         if (status == -NFS4ERR_BAD_SEQID)
1111                 nfs4_drop_state_owner(sp);
1112         if (!nfs4_has_session(server->nfs_client))
1113                 nfs_increment_seqid(status, seqid);
1114 }
1115
1116 /*
1117  * Increment the seqid if the LOCK/LOCKU succeeded, or
1118  * failed with a seqid incrementing error -
1119  * see comments nfs_fs.h:seqid_mutating_error()
1120  */
1121 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1122 {
1123         nfs_increment_seqid(status, seqid);
1124 }
1125
1126 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1127 {
1128         struct nfs_seqid_counter *sequence = seqid->sequence;
1129         int status = 0;
1130
1131         spin_lock(&sequence->lock);
1132         seqid->task = task;
1133         if (list_empty(&seqid->list))
1134                 list_add_tail(&seqid->list, &sequence->list);
1135         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1136                 goto unlock;
1137         rpc_sleep_on(&sequence->wait, task, NULL);
1138         status = -EAGAIN;
1139 unlock:
1140         spin_unlock(&sequence->lock);
1141         return status;
1142 }
1143
1144 static int nfs4_run_state_manager(void *);
1145
1146 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1147 {
1148         smp_mb__before_clear_bit();
1149         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1150         smp_mb__after_clear_bit();
1151         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1152         rpc_wake_up(&clp->cl_rpcwaitq);
1153 }
1154
1155 /*
1156  * Schedule the nfs_client asynchronous state management routine
1157  */
1158 void nfs4_schedule_state_manager(struct nfs_client *clp)
1159 {
1160         struct task_struct *task;
1161         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1162
1163         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1164                 return;
1165         __module_get(THIS_MODULE);
1166         atomic_inc(&clp->cl_count);
1167
1168         /* The rcu_read_lock() is not strictly necessary, as the state
1169          * manager is the only thread that ever changes the rpc_xprt
1170          * after it's initialized.  At this point, we're single threaded. */
1171         rcu_read_lock();
1172         snprintf(buf, sizeof(buf), "%s-manager",
1173                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1174         rcu_read_unlock();
1175         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1176         if (IS_ERR(task)) {
1177                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1178                         __func__, PTR_ERR(task));
1179                 nfs4_clear_state_manager_bit(clp);
1180                 nfs_put_client(clp);
1181                 module_put(THIS_MODULE);
1182         }
1183 }
1184
1185 /*
1186  * Schedule a lease recovery attempt
1187  */
1188 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1189 {
1190         if (!clp)
1191                 return;
1192         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1193                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1194         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1195                         clp->cl_hostname);
1196         nfs4_schedule_state_manager(clp);
1197 }
1198 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1199
1200 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1201 {
1202         int res;
1203
1204         might_sleep();
1205
1206         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1207                         nfs_wait_bit_killable, TASK_KILLABLE);
1208         if (res)
1209                 return res;
1210
1211         if (clp->cl_cons_state < 0)
1212                 return clp->cl_cons_state;
1213         return 0;
1214 }
1215
1216 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1217 {
1218         unsigned int loop;
1219         int ret;
1220
1221         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1222                 ret = nfs4_wait_clnt_recover(clp);
1223                 if (ret != 0)
1224                         break;
1225                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1226                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1227                         break;
1228                 nfs4_schedule_state_manager(clp);
1229                 ret = -EIO;
1230         }
1231         return ret;
1232 }
1233
1234 /*
1235  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1236  * @clp: client to process
1237  *
1238  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1239  * resend of the SETCLIENTID and hence re-establish the
1240  * callback channel. Then return all existing delegations.
1241  */
1242 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1243 {
1244         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1245         nfs_expire_all_delegations(clp);
1246         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1247                         clp->cl_hostname);
1248 }
1249
1250 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1251 {
1252         nfs40_handle_cb_pathdown(clp);
1253         nfs4_schedule_state_manager(clp);
1254 }
1255
1256 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1257 {
1258
1259         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1260         /* Don't recover state that expired before the reboot */
1261         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1262                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1263                 return 0;
1264         }
1265         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1266         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1267         return 1;
1268 }
1269
1270 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1271 {
1272         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1273         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1274         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1275         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1276         return 1;
1277 }
1278
1279 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1280 {
1281         struct nfs_client *clp = server->nfs_client;
1282
1283         if (!nfs4_valid_open_stateid(state))
1284                 return -EBADF;
1285         nfs4_state_mark_reclaim_nograce(clp, state);
1286         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1287                         clp->cl_hostname);
1288         nfs4_schedule_state_manager(clp);
1289         return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1292
1293 void nfs_inode_find_state_and_recover(struct inode *inode,
1294                 const nfs4_stateid *stateid)
1295 {
1296         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1297         struct nfs_inode *nfsi = NFS_I(inode);
1298         struct nfs_open_context *ctx;
1299         struct nfs4_state *state;
1300         bool found = false;
1301
1302         spin_lock(&inode->i_lock);
1303         list_for_each_entry(ctx, &nfsi->open_files, list) {
1304                 state = ctx->state;
1305                 if (state == NULL)
1306                         continue;
1307                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1308                         continue;
1309                 if (!nfs4_stateid_match(&state->stateid, stateid))
1310                         continue;
1311                 nfs4_state_mark_reclaim_nograce(clp, state);
1312                 found = true;
1313         }
1314         spin_unlock(&inode->i_lock);
1315         if (found)
1316                 nfs4_schedule_state_manager(clp);
1317 }
1318
1319 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1320 {
1321         struct inode *inode = state->inode;
1322         struct nfs_inode *nfsi = NFS_I(inode);
1323         struct nfs_open_context *ctx;
1324
1325         spin_lock(&inode->i_lock);
1326         list_for_each_entry(ctx, &nfsi->open_files, list) {
1327                 if (ctx->state != state)
1328                         continue;
1329                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1330         }
1331         spin_unlock(&inode->i_lock);
1332 }
1333
1334 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1335 {
1336         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1337         nfs4_state_mark_open_context_bad(state);
1338 }
1339
1340
1341 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1342 {
1343         struct inode *inode = state->inode;
1344         struct nfs_inode *nfsi = NFS_I(inode);
1345         struct file_lock *fl;
1346         int status = 0;
1347
1348         if (inode->i_flock == NULL)
1349                 return 0;
1350
1351         /* Guard against delegation returns and new lock/unlock calls */
1352         down_write(&nfsi->rwsem);
1353         /* Protect inode->i_flock using the BKL */
1354         spin_lock(&inode->i_lock);
1355         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1356                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1357                         continue;
1358                 if (nfs_file_open_context(fl->fl_file)->state != state)
1359                         continue;
1360                 spin_unlock(&inode->i_lock);
1361                 status = ops->recover_lock(state, fl);
1362                 switch (status) {
1363                         case 0:
1364                                 break;
1365                         case -ESTALE:
1366                         case -NFS4ERR_ADMIN_REVOKED:
1367                         case -NFS4ERR_STALE_STATEID:
1368                         case -NFS4ERR_BAD_STATEID:
1369                         case -NFS4ERR_EXPIRED:
1370                         case -NFS4ERR_NO_GRACE:
1371                         case -NFS4ERR_STALE_CLIENTID:
1372                         case -NFS4ERR_BADSESSION:
1373                         case -NFS4ERR_BADSLOT:
1374                         case -NFS4ERR_BAD_HIGH_SLOT:
1375                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1376                                 goto out;
1377                         default:
1378                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1379                                         "Zeroing state\n", __func__, status);
1380                         case -ENOMEM:
1381                         case -NFS4ERR_DENIED:
1382                         case -NFS4ERR_RECLAIM_BAD:
1383                         case -NFS4ERR_RECLAIM_CONFLICT:
1384                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1385                                 status = 0;
1386                 }
1387                 spin_lock(&inode->i_lock);
1388         }
1389         spin_unlock(&inode->i_lock);
1390 out:
1391         up_write(&nfsi->rwsem);
1392         return status;
1393 }
1394
1395 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1396 {
1397         struct nfs4_state *state;
1398         struct nfs4_lock_state *lock;
1399         int status = 0;
1400
1401         /* Note: we rely on the sp->so_states list being ordered
1402          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1403          * states first.
1404          * This is needed to ensure that the server won't give us any
1405          * read delegations that we have to return if, say, we are
1406          * recovering after a network partition or a reboot from a
1407          * server that doesn't support a grace period.
1408          */
1409         spin_lock(&sp->so_lock);
1410         write_seqcount_begin(&sp->so_reclaim_seqcount);
1411 restart:
1412         list_for_each_entry(state, &sp->so_states, open_states) {
1413                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1414                         continue;
1415                 if (!nfs4_valid_open_stateid(state))
1416                         continue;
1417                 if (state->state == 0)
1418                         continue;
1419                 atomic_inc(&state->count);
1420                 spin_unlock(&sp->so_lock);
1421                 status = ops->recover_open(sp, state);
1422                 if (status >= 0) {
1423                         status = nfs4_reclaim_locks(state, ops);
1424                         if (status >= 0) {
1425                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1426                                         spin_lock(&state->state_lock);
1427                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1428                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1429                                                         pr_warn_ratelimited("NFS: "
1430                                                                             "%s: Lock reclaim "
1431                                                                             "failed!\n", __func__);
1432                                         }
1433                                         spin_unlock(&state->state_lock);
1434                                 }
1435                                 nfs4_put_open_state(state);
1436                                 spin_lock(&sp->so_lock);
1437                                 goto restart;
1438                         }
1439                 }
1440                 switch (status) {
1441                         default:
1442                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1443                                         "Zeroing state\n", __func__, status);
1444                         case -ENOENT:
1445                         case -ENOMEM:
1446                         case -ESTALE:
1447                                 /*
1448                                  * Open state on this file cannot be recovered
1449                                  * All we can do is revert to using the zero stateid.
1450                                  */
1451                                 nfs4_state_mark_recovery_failed(state, status);
1452                                 break;
1453                         case -EAGAIN:
1454                                 ssleep(1);
1455                         case -NFS4ERR_ADMIN_REVOKED:
1456                         case -NFS4ERR_STALE_STATEID:
1457                         case -NFS4ERR_BAD_STATEID:
1458                         case -NFS4ERR_RECLAIM_BAD:
1459                         case -NFS4ERR_RECLAIM_CONFLICT:
1460                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1461                                 break;
1462                         case -NFS4ERR_EXPIRED:
1463                         case -NFS4ERR_NO_GRACE:
1464                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1465                         case -NFS4ERR_STALE_CLIENTID:
1466                         case -NFS4ERR_BADSESSION:
1467                         case -NFS4ERR_BADSLOT:
1468                         case -NFS4ERR_BAD_HIGH_SLOT:
1469                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1470                                 goto out_err;
1471                 }
1472                 nfs4_put_open_state(state);
1473                 spin_lock(&sp->so_lock);
1474                 goto restart;
1475         }
1476         write_seqcount_end(&sp->so_reclaim_seqcount);
1477         spin_unlock(&sp->so_lock);
1478         return 0;
1479 out_err:
1480         nfs4_put_open_state(state);
1481         spin_lock(&sp->so_lock);
1482         write_seqcount_end(&sp->so_reclaim_seqcount);
1483         spin_unlock(&sp->so_lock);
1484         return status;
1485 }
1486
1487 static void nfs4_clear_open_state(struct nfs4_state *state)
1488 {
1489         struct nfs4_lock_state *lock;
1490
1491         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1492         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1493         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1494         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1495         spin_lock(&state->state_lock);
1496         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1497                 lock->ls_seqid.flags = 0;
1498                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1499         }
1500         spin_unlock(&state->state_lock);
1501 }
1502
1503 static void nfs4_reset_seqids(struct nfs_server *server,
1504         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1505 {
1506         struct nfs_client *clp = server->nfs_client;
1507         struct nfs4_state_owner *sp;
1508         struct rb_node *pos;
1509         struct nfs4_state *state;
1510
1511         spin_lock(&clp->cl_lock);
1512         for (pos = rb_first(&server->state_owners);
1513              pos != NULL;
1514              pos = rb_next(pos)) {
1515                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1516                 sp->so_seqid.flags = 0;
1517                 spin_lock(&sp->so_lock);
1518                 list_for_each_entry(state, &sp->so_states, open_states) {
1519                         if (mark_reclaim(clp, state))
1520                                 nfs4_clear_open_state(state);
1521                 }
1522                 spin_unlock(&sp->so_lock);
1523         }
1524         spin_unlock(&clp->cl_lock);
1525 }
1526
1527 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1528         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1529 {
1530         struct nfs_server *server;
1531
1532         rcu_read_lock();
1533         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1534                 nfs4_reset_seqids(server, mark_reclaim);
1535         rcu_read_unlock();
1536 }
1537
1538 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1539 {
1540         /* Mark all delegations for reclaim */
1541         nfs_delegation_mark_reclaim(clp);
1542         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1543 }
1544
1545 static void nfs4_reclaim_complete(struct nfs_client *clp,
1546                                  const struct nfs4_state_recovery_ops *ops,
1547                                  struct rpc_cred *cred)
1548 {
1549         /* Notify the server we're done reclaiming our state */
1550         if (ops->reclaim_complete)
1551                 (void)ops->reclaim_complete(clp, cred);
1552 }
1553
1554 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1555 {
1556         struct nfs_client *clp = server->nfs_client;
1557         struct nfs4_state_owner *sp;
1558         struct rb_node *pos;
1559         struct nfs4_state *state;
1560
1561         spin_lock(&clp->cl_lock);
1562         for (pos = rb_first(&server->state_owners);
1563              pos != NULL;
1564              pos = rb_next(pos)) {
1565                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1566                 spin_lock(&sp->so_lock);
1567                 list_for_each_entry(state, &sp->so_states, open_states) {
1568                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1569                                                 &state->flags))
1570                                 continue;
1571                         nfs4_state_mark_reclaim_nograce(clp, state);
1572                 }
1573                 spin_unlock(&sp->so_lock);
1574         }
1575         spin_unlock(&clp->cl_lock);
1576 }
1577
1578 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1579 {
1580         struct nfs_server *server;
1581
1582         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1583                 return 0;
1584
1585         rcu_read_lock();
1586         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1587                 nfs4_clear_reclaim_server(server);
1588         rcu_read_unlock();
1589
1590         nfs_delegation_reap_unclaimed(clp);
1591         return 1;
1592 }
1593
1594 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1595 {
1596         const struct nfs4_state_recovery_ops *ops;
1597         struct rpc_cred *cred;
1598
1599         if (!nfs4_state_clear_reclaim_reboot(clp))
1600                 return;
1601         ops = clp->cl_mvops->reboot_recovery_ops;
1602         cred = nfs4_get_clid_cred(clp);
1603         nfs4_reclaim_complete(clp, ops, cred);
1604         put_rpccred(cred);
1605 }
1606
1607 static void nfs_delegation_clear_all(struct nfs_client *clp)
1608 {
1609         nfs_delegation_mark_reclaim(clp);
1610         nfs_delegation_reap_unclaimed(clp);
1611 }
1612
1613 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1614 {
1615         nfs_delegation_clear_all(clp);
1616         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1617 }
1618
1619 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1620 {
1621         switch (error) {
1622                 case 0:
1623                         break;
1624                 case -NFS4ERR_CB_PATH_DOWN:
1625                         nfs40_handle_cb_pathdown(clp);
1626                         break;
1627                 case -NFS4ERR_NO_GRACE:
1628                         nfs4_state_end_reclaim_reboot(clp);
1629                         break;
1630                 case -NFS4ERR_STALE_CLIENTID:
1631                 case -NFS4ERR_LEASE_MOVED:
1632                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1633                         nfs4_state_clear_reclaim_reboot(clp);
1634                         nfs4_state_start_reclaim_reboot(clp);
1635                         break;
1636                 case -NFS4ERR_EXPIRED:
1637                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1638                         nfs4_state_start_reclaim_nograce(clp);
1639                         break;
1640                 case -NFS4ERR_BADSESSION:
1641                 case -NFS4ERR_BADSLOT:
1642                 case -NFS4ERR_BAD_HIGH_SLOT:
1643                 case -NFS4ERR_DEADSESSION:
1644                 case -NFS4ERR_SEQ_FALSE_RETRY:
1645                 case -NFS4ERR_SEQ_MISORDERED:
1646                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1647                         /* Zero session reset errors */
1648                         break;
1649                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1650                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1651                         break;
1652                 default:
1653                         dprintk("%s: failed to handle error %d for server %s\n",
1654                                         __func__, error, clp->cl_hostname);
1655                         return error;
1656         }
1657         dprintk("%s: handled error %d for server %s\n", __func__, error,
1658                         clp->cl_hostname);
1659         return 0;
1660 }
1661
1662 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1663 {
1664         struct nfs4_state_owner *sp;
1665         struct nfs_server *server;
1666         struct rb_node *pos;
1667         int status = 0;
1668
1669 restart:
1670         rcu_read_lock();
1671         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1672                 nfs4_purge_state_owners(server);
1673                 spin_lock(&clp->cl_lock);
1674                 for (pos = rb_first(&server->state_owners);
1675                      pos != NULL;
1676                      pos = rb_next(pos)) {
1677                         sp = rb_entry(pos,
1678                                 struct nfs4_state_owner, so_server_node);
1679                         if (!test_and_clear_bit(ops->owner_flag_bit,
1680                                                         &sp->so_flags))
1681                                 continue;
1682                         atomic_inc(&sp->so_count);
1683                         spin_unlock(&clp->cl_lock);
1684                         rcu_read_unlock();
1685
1686                         status = nfs4_reclaim_open_state(sp, ops);
1687                         if (status < 0) {
1688                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1689                                 nfs4_put_state_owner(sp);
1690                                 return nfs4_recovery_handle_error(clp, status);
1691                         }
1692
1693                         nfs4_put_state_owner(sp);
1694                         goto restart;
1695                 }
1696                 spin_unlock(&clp->cl_lock);
1697         }
1698         rcu_read_unlock();
1699         return status;
1700 }
1701
1702 static int nfs4_check_lease(struct nfs_client *clp)
1703 {
1704         struct rpc_cred *cred;
1705         const struct nfs4_state_maintenance_ops *ops =
1706                 clp->cl_mvops->state_renewal_ops;
1707         int status;
1708
1709         /* Is the client already known to have an expired lease? */
1710         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1711                 return 0;
1712         spin_lock(&clp->cl_lock);
1713         cred = ops->get_state_renewal_cred_locked(clp);
1714         spin_unlock(&clp->cl_lock);
1715         if (cred == NULL) {
1716                 cred = nfs4_get_clid_cred(clp);
1717                 status = -ENOKEY;
1718                 if (cred == NULL)
1719                         goto out;
1720         }
1721         status = ops->renew_lease(clp, cred);
1722         put_rpccred(cred);
1723         if (status == -ETIMEDOUT) {
1724                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1725                 return 0;
1726         }
1727 out:
1728         return nfs4_recovery_handle_error(clp, status);
1729 }
1730
1731 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1732  * and for recoverable errors on EXCHANGE_ID for v4.1
1733  */
1734 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1735 {
1736         switch (status) {
1737         case -NFS4ERR_SEQ_MISORDERED:
1738                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1739                         return -ESERVERFAULT;
1740                 /* Lease confirmation error: retry after purging the lease */
1741                 ssleep(1);
1742                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1743                 break;
1744         case -NFS4ERR_STALE_CLIENTID:
1745                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1746                 nfs4_state_clear_reclaim_reboot(clp);
1747                 nfs4_state_start_reclaim_reboot(clp);
1748                 break;
1749         case -NFS4ERR_CLID_INUSE:
1750                 pr_err("NFS: Server %s reports our clientid is in use\n",
1751                         clp->cl_hostname);
1752                 nfs_mark_client_ready(clp, -EPERM);
1753                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1754                 return -EPERM;
1755         case -EACCES:
1756         case -NFS4ERR_DELAY:
1757         case -ETIMEDOUT:
1758         case -EAGAIN:
1759                 ssleep(1);
1760                 break;
1761
1762         case -NFS4ERR_MINOR_VERS_MISMATCH:
1763                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1764                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1765                 dprintk("%s: exit with error %d for server %s\n",
1766                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1767                 return -EPROTONOSUPPORT;
1768         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1769                                  * in nfs4_exchange_id */
1770         default:
1771                 dprintk("%s: exit with error %d for server %s\n", __func__,
1772                                 status, clp->cl_hostname);
1773                 return status;
1774         }
1775         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1776         dprintk("%s: handled error %d for server %s\n", __func__, status,
1777                         clp->cl_hostname);
1778         return 0;
1779 }
1780
1781 static int nfs4_establish_lease(struct nfs_client *clp)
1782 {
1783         struct rpc_cred *cred;
1784         const struct nfs4_state_recovery_ops *ops =
1785                 clp->cl_mvops->reboot_recovery_ops;
1786         int status;
1787
1788         cred = nfs4_get_clid_cred(clp);
1789         if (cred == NULL)
1790                 return -ENOENT;
1791         status = ops->establish_clid(clp, cred);
1792         put_rpccred(cred);
1793         if (status != 0)
1794                 return status;
1795         pnfs_destroy_all_layouts(clp);
1796         return 0;
1797 }
1798
1799 /*
1800  * Returns zero or a negative errno.  NFS4ERR values are converted
1801  * to local errno values.
1802  */
1803 static int nfs4_reclaim_lease(struct nfs_client *clp)
1804 {
1805         int status;
1806
1807         status = nfs4_establish_lease(clp);
1808         if (status < 0)
1809                 return nfs4_handle_reclaim_lease_error(clp, status);
1810         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1811                 nfs4_state_start_reclaim_nograce(clp);
1812         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1813                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1814         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1815         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1816         return 0;
1817 }
1818
1819 static int nfs4_purge_lease(struct nfs_client *clp)
1820 {
1821         int status;
1822
1823         status = nfs4_establish_lease(clp);
1824         if (status < 0)
1825                 return nfs4_handle_reclaim_lease_error(clp, status);
1826         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1827         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1828         nfs4_state_start_reclaim_nograce(clp);
1829         return 0;
1830 }
1831
1832 /**
1833  * nfs4_discover_server_trunking - Detect server IP address trunking
1834  *
1835  * @clp: nfs_client under test
1836  * @result: OUT: found nfs_client, or clp
1837  *
1838  * Returns zero or a negative errno.  If zero is returned,
1839  * an nfs_client pointer is planted in "result".
1840  *
1841  * Note: since we are invoked in process context, and
1842  * not from inside the state manager, we cannot use
1843  * nfs4_handle_reclaim_lease_error().
1844  */
1845 int nfs4_discover_server_trunking(struct nfs_client *clp,
1846                                   struct nfs_client **result)
1847 {
1848         const struct nfs4_state_recovery_ops *ops =
1849                                 clp->cl_mvops->reboot_recovery_ops;
1850         struct rpc_clnt *clnt;
1851         struct rpc_cred *cred;
1852         int i, status;
1853
1854         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1855
1856         clnt = clp->cl_rpcclient;
1857         i = 0;
1858
1859         mutex_lock(&nfs_clid_init_mutex);
1860 again:
1861         status  = -ENOENT;
1862         cred = nfs4_get_clid_cred(clp);
1863         if (cred == NULL)
1864                 goto out_unlock;
1865
1866         status = ops->detect_trunking(clp, result, cred);
1867         put_rpccred(cred);
1868         switch (status) {
1869         case 0:
1870                 break;
1871         case -NFS4ERR_DELAY:
1872         case -ETIMEDOUT:
1873         case -EAGAIN:
1874                 ssleep(1);
1875         case -NFS4ERR_STALE_CLIENTID:
1876                 dprintk("NFS: %s after status %d, retrying\n",
1877                         __func__, status);
1878                 goto again;
1879         case -EACCES:
1880                 if (i++ == 0) {
1881                         nfs4_root_machine_cred(clp);
1882                         goto again;
1883                 }
1884                 if (i > 2)
1885                         break;
1886         case -NFS4ERR_CLID_INUSE:
1887         case -NFS4ERR_WRONGSEC:
1888                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
1889                 if (IS_ERR(clnt)) {
1890                         status = PTR_ERR(clnt);
1891                         break;
1892                 }
1893                 /* Note: this is safe because we haven't yet marked the
1894                  * client as ready, so we are the only user of
1895                  * clp->cl_rpcclient
1896                  */
1897                 clnt = xchg(&clp->cl_rpcclient, clnt);
1898                 rpc_shutdown_client(clnt);
1899                 clnt = clp->cl_rpcclient;
1900                 goto again;
1901
1902         case -NFS4ERR_MINOR_VERS_MISMATCH:
1903                 status = -EPROTONOSUPPORT;
1904                 break;
1905
1906         case -EKEYEXPIRED:
1907         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1908                                  * in nfs4_exchange_id */
1909                 status = -EKEYEXPIRED;
1910                 break;
1911         default:
1912                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
1913                                 __func__, status);
1914                 status = -EIO;
1915         }
1916
1917 out_unlock:
1918         mutex_unlock(&nfs_clid_init_mutex);
1919         dprintk("NFS: %s: status = %d\n", __func__, status);
1920         return status;
1921 }
1922
1923 #ifdef CONFIG_NFS_V4_1
1924 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1925 {
1926         struct nfs_client *clp = session->clp;
1927
1928         switch (err) {
1929         default:
1930                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1931                 break;
1932         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1933                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1934         }
1935         nfs4_schedule_lease_recovery(clp);
1936 }
1937 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1938
1939 static void nfs41_ping_server(struct nfs_client *clp)
1940 {
1941         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1942         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1943         nfs4_schedule_state_manager(clp);
1944 }
1945
1946 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1947 {
1948         nfs41_ping_server(clp);
1949 }
1950
1951 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1952 {
1953         nfs41_ping_server(clp);
1954 }
1955
1956 static void nfs4_reset_all_state(struct nfs_client *clp)
1957 {
1958         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1959                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1960                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1961                 nfs4_state_start_reclaim_nograce(clp);
1962                 dprintk("%s: scheduling reset of all state for server %s!\n",
1963                                 __func__, clp->cl_hostname);
1964                 nfs4_schedule_state_manager(clp);
1965         }
1966 }
1967
1968 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1969 {
1970         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1971                 nfs4_state_start_reclaim_reboot(clp);
1972                 dprintk("%s: server %s rebooted!\n", __func__,
1973                                 clp->cl_hostname);
1974                 nfs4_schedule_state_manager(clp);
1975         }
1976 }
1977
1978 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1979 {
1980         nfs4_reset_all_state(clp);
1981         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1982 }
1983
1984 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1985 {
1986         /* This will need to handle layouts too */
1987         nfs_expire_all_delegations(clp);
1988         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1989                         clp->cl_hostname);
1990 }
1991
1992 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1993 {
1994         nfs_expire_all_delegations(clp);
1995         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1996                 nfs4_schedule_state_manager(clp);
1997         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1998                         clp->cl_hostname);
1999 }
2000
2001 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2002 {
2003         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2004                 &clp->cl_state) == 0)
2005                 nfs4_schedule_state_manager(clp);
2006 }
2007
2008 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2009 {
2010         if (!flags)
2011                 return;
2012
2013         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2014                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2015
2016         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2017                 nfs41_handle_server_reboot(clp);
2018         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2019                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2020                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2021                             SEQ4_STATUS_LEASE_MOVED))
2022                 nfs41_handle_state_revoked(clp);
2023         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2024                 nfs41_handle_recallable_state_revoked(clp);
2025         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2026                 nfs41_handle_backchannel_fault(clp);
2027         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2028                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2029                 nfs41_handle_cb_path_down(clp);
2030 }
2031
2032 static int nfs4_reset_session(struct nfs_client *clp)
2033 {
2034         struct rpc_cred *cred;
2035         int status;
2036
2037         if (!nfs4_has_session(clp))
2038                 return 0;
2039         nfs4_begin_drain_session(clp);
2040         cred = nfs4_get_clid_cred(clp);
2041         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2042         switch (status) {
2043         case 0:
2044         case -NFS4ERR_BADSESSION:
2045         case -NFS4ERR_DEADSESSION:
2046                 break;
2047         case -NFS4ERR_BACK_CHAN_BUSY:
2048         case -NFS4ERR_DELAY:
2049                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2050                 status = 0;
2051                 ssleep(1);
2052                 goto out;
2053         default:
2054                 status = nfs4_recovery_handle_error(clp, status);
2055                 goto out;
2056         }
2057
2058         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2059         status = nfs4_proc_create_session(clp, cred);
2060         if (status) {
2061                 dprintk("%s: session reset failed with status %d for server %s!\n",
2062                         __func__, status, clp->cl_hostname);
2063                 status = nfs4_handle_reclaim_lease_error(clp, status);
2064                 goto out;
2065         }
2066         nfs41_finish_session_reset(clp);
2067         dprintk("%s: session reset was successful for server %s!\n",
2068                         __func__, clp->cl_hostname);
2069 out:
2070         if (cred)
2071                 put_rpccred(cred);
2072         return status;
2073 }
2074
2075 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2076 {
2077         struct rpc_cred *cred;
2078         int ret;
2079
2080         if (!nfs4_has_session(clp))
2081                 return 0;
2082         nfs4_begin_drain_session(clp);
2083         cred = nfs4_get_clid_cred(clp);
2084         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2085         if (cred)
2086                 put_rpccred(cred);
2087         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2088         switch (ret) {
2089         case 0:
2090                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2091                         __func__, clp->cl_hostname);
2092                 break;
2093         case -NFS4ERR_DELAY:
2094                 ssleep(1);
2095                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2096                 break;
2097         default:
2098                 return nfs4_recovery_handle_error(clp, ret);
2099         }
2100         return 0;
2101 }
2102 #else /* CONFIG_NFS_V4_1 */
2103 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2104
2105 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2106 {
2107         return 0;
2108 }
2109 #endif /* CONFIG_NFS_V4_1 */
2110
2111 static void nfs4_state_manager(struct nfs_client *clp)
2112 {
2113         int status = 0;
2114         const char *section = "", *section_sep = "";
2115
2116         /* Ensure exclusive access to NFSv4 state */
2117         do {
2118                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2119                         section = "purge state";
2120                         status = nfs4_purge_lease(clp);
2121                         if (status < 0)
2122                                 goto out_error;
2123                         continue;
2124                 }
2125
2126                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2127                         section = "lease expired";
2128                         /* We're going to have to re-establish a clientid */
2129                         status = nfs4_reclaim_lease(clp);
2130                         if (status < 0)
2131                                 goto out_error;
2132                         continue;
2133                 }
2134
2135                 /* Initialize or reset the session */
2136                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2137                         section = "reset session";
2138                         status = nfs4_reset_session(clp);
2139                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2140                                 continue;
2141                         if (status < 0)
2142                                 goto out_error;
2143                 }
2144
2145                 /* Send BIND_CONN_TO_SESSION */
2146                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2147                                 &clp->cl_state)) {
2148                         section = "bind conn to session";
2149                         status = nfs4_bind_conn_to_session(clp);
2150                         if (status < 0)
2151                                 goto out_error;
2152                         continue;
2153                 }
2154
2155                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2156                         section = "check lease";
2157                         status = nfs4_check_lease(clp);
2158                         if (status < 0)
2159                                 goto out_error;
2160                         continue;
2161                 }
2162
2163                 /* First recover reboot state... */
2164                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2165                         section = "reclaim reboot";
2166                         status = nfs4_do_reclaim(clp,
2167                                 clp->cl_mvops->reboot_recovery_ops);
2168                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2169                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2170                                 continue;
2171                         nfs4_state_end_reclaim_reboot(clp);
2172                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2173                                 continue;
2174                         if (status < 0)
2175                                 goto out_error;
2176                 }
2177
2178                 /* Now recover expired state... */
2179                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2180                         section = "reclaim nograce";
2181                         status = nfs4_do_reclaim(clp,
2182                                 clp->cl_mvops->nograce_recovery_ops);
2183                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2184                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2185                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2186                                 continue;
2187                         if (status < 0)
2188                                 goto out_error;
2189                 }
2190
2191                 nfs4_end_drain_session(clp);
2192                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2193                         nfs_client_return_marked_delegations(clp);
2194                         continue;
2195                 }
2196
2197                 nfs4_clear_state_manager_bit(clp);
2198                 /* Did we race with an attempt to give us more work? */
2199                 if (clp->cl_state == 0)
2200                         break;
2201                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2202                         break;
2203         } while (atomic_read(&clp->cl_count) > 1);
2204         return;
2205 out_error:
2206         if (strlen(section))
2207                 section_sep = ": ";
2208         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2209                         " with error %d\n", section_sep, section,
2210                         clp->cl_hostname, -status);
2211         ssleep(1);
2212         nfs4_end_drain_session(clp);
2213         nfs4_clear_state_manager_bit(clp);
2214 }
2215
2216 static int nfs4_run_state_manager(void *ptr)
2217 {
2218         struct nfs_client *clp = ptr;
2219
2220         allow_signal(SIGKILL);
2221         nfs4_state_manager(clp);
2222         nfs_put_client(clp);
2223         module_put_and_exit(0);
2224         return 0;
2225 }
2226
2227 /*
2228  * Local variables:
2229  *  c-basic-offset: 8
2230  * End:
2231  */