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