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 "nfs4_fs.h"
  55 #include "callback.h"
  56 #include "delegation.h"
  57 #include "internal.h"
  58 #include "pnfs.h"
  59
  60 #define NFSDBG_FACILITY         NFSDBG_STATE
  61
  62 #define OPENOWNER_POOL_SIZE     8
  63
  64 const nfs4_stateid zero_stateid;
  65
  66 static LIST_HEAD(nfs4_clientid_list);
  67
  68 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  69 {
  70         struct nfs4_setclientid_res clid = {
  71                 .clientid = clp->cl_clientid,
  72                 .confirm = clp->cl_confirm,
  73         };
  74         unsigned short port;
  75         int status;
  76
  77         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  78                 goto do_confirm;
  79         port = nfs_callback_tcpport;
  80         if (clp->cl_addr.ss_family == AF_INET6)
  81                 port = nfs_callback_tcpport6;
  82
  83         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  84         if (status != 0)
  85                 goto out;
  86         clp->cl_clientid = clid.clientid;
  87         clp->cl_confirm = clid.confirm;
  88         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  89 do_confirm:
  90         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  91         if (status != 0)
  92                 goto out;
  93         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  94         nfs4_schedule_state_renewal(clp);
  95 out:
  96         return status;
  97 }
  98
  99 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 100 {
 101         struct rpc_cred *cred = NULL;
 102
 103         if (clp->cl_machine_cred != NULL)
 104                 cred = get_rpccred(clp->cl_machine_cred);
 105         return cred;
 106 }
 107
 108 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 109 {
 110         struct rpc_cred *cred;
 111
 112         spin_lock(&clp->cl_lock);
 113         cred = clp->cl_machine_cred;
 114         clp->cl_machine_cred = NULL;
 115         spin_unlock(&clp->cl_lock);
 116         if (cred != NULL)
 117                 put_rpccred(cred);
 118 }
 119
 120 static struct rpc_cred *
 121 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 122 {
 123         struct rpc_cred *cred = NULL;
 124         struct nfs4_state_owner *sp;
 125         struct rb_node *pos;
 126
 127         for (pos = rb_first(&server->state_owners);
 128              pos != NULL;
 129              pos = rb_next(pos)) {
 130                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 131                 if (list_empty(&sp->so_states))
 132                         continue;
 133                 cred = get_rpccred(sp->so_cred);
 134                 break;
 135         }
 136         return cred;
 137 }
 138
 139 /**
 140  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 141  * @clp: client state handle
 142  *
 143  * Returns an rpc_cred with reference count bumped, or NULL.
 144  * Caller must hold clp->cl_lock.
 145  */
 146 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 147 {
 148         struct rpc_cred *cred = NULL;
 149         struct nfs_server *server;
 150
 151         /* Use machine credentials if available */
 152         cred = nfs4_get_machine_cred_locked(clp);
 153         if (cred != NULL)
 154                 goto out;
 155
 156         rcu_read_lock();
 157         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 158                 cred = nfs4_get_renew_cred_server_locked(server);
 159                 if (cred != NULL)
 160                         break;
 161         }
 162         rcu_read_unlock();
 163
 164 out:
 165         return cred;
 166 }
 167
 168 #if defined(CONFIG_NFS_V4_1)
 169
 170 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 171 {
 172         int status;
 173         struct nfs_fsinfo fsinfo;
 174
 175         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 176                 nfs4_schedule_state_renewal(clp);
 177                 return 0;
 178         }
 179
 180         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 181         if (status == 0) {
 182                 /* Update lease time and schedule renewal */
 183                 spin_lock(&clp->cl_lock);
 184                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 185                 clp->cl_last_renewal = jiffies;
 186                 spin_unlock(&clp->cl_lock);
 187
 188                 nfs4_schedule_state_renewal(clp);
 189         }
 190
 191         return status;
 192 }
 193
 194 /*
 195  * Back channel returns NFS4ERR_DELAY for new requests when
 196  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 197  * is ended.
 198  */
 199 static void nfs4_end_drain_session(struct nfs_client *clp)
 200 {
 201         struct nfs4_session *ses = clp->cl_session;
 202         struct nfs4_slot_table *tbl;
 203         int max_slots;
 204
 205         if (ses == NULL)
 206                 return;
 207         tbl = &ses->fc_slot_table;
 208         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 209                 spin_lock(&tbl->slot_tbl_lock);
 210                 max_slots = tbl->max_slots;
 211                 while (max_slots--) {
 212                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 213                                                 nfs4_set_task_privileged,
 214                                                 NULL) == NULL)
 215                                 break;
 216                 }
 217                 spin_unlock(&tbl->slot_tbl_lock);
 218         }
 219 }
 220
 221 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 222 {
 223         spin_lock(&tbl->slot_tbl_lock);
 224         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 225                 INIT_COMPLETION(tbl->complete);
 226                 spin_unlock(&tbl->slot_tbl_lock);
 227                 return wait_for_completion_interruptible(&tbl->complete);
 228         }
 229         spin_unlock(&tbl->slot_tbl_lock);
 230         return 0;
 231 }
 232
 233 static int nfs4_begin_drain_session(struct nfs_client *clp)
 234 {
 235         struct nfs4_session *ses = clp->cl_session;
 236         int ret = 0;
 237
 238         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 239         /* back channel */
 240         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 241         if (ret)
 242                 return ret;
 243         /* fore channel */
 244         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 245 }
 246
 247 static void nfs41_finish_session_reset(struct nfs_client *clp)
 248 {
 249         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 250         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 251         /* create_session negotiated new slot table */
 252         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
 253         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 254         nfs41_setup_state_renewal(clp);
 255 }
 256
 257 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 258 {
 259         int status;
 260
 261         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 262                 goto do_confirm;
 263         nfs4_begin_drain_session(clp);
 264         status = nfs4_proc_exchange_id(clp, cred);
 265         if (status != 0)
 266                 goto out;
 267         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 268 do_confirm:
 269         status = nfs4_proc_create_session(clp, cred);
 270         if (status != 0)
 271                 goto out;
 272         nfs41_finish_session_reset(clp);
 273         nfs_mark_client_ready(clp, NFS_CS_READY);
 274 out:
 275         return status;
 276 }
 277
 278 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 279 {
 280         struct rpc_cred *cred;
 281
 282         spin_lock(&clp->cl_lock);
 283         cred = nfs4_get_machine_cred_locked(clp);
 284         spin_unlock(&clp->cl_lock);
 285         return cred;
 286 }
 287
 288 #endif /* CONFIG_NFS_V4_1 */
 289
 290 static struct rpc_cred *
 291 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 292 {
 293         struct nfs_client *clp = server->nfs_client;
 294         struct rpc_cred *cred = NULL;
 295         struct nfs4_state_owner *sp;
 296         struct rb_node *pos;
 297
 298         spin_lock(&clp->cl_lock);
 299         pos = rb_first(&server->state_owners);
 300         if (pos != NULL) {
 301                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 302                 cred = get_rpccred(sp->so_cred);
 303         }
 304         spin_unlock(&clp->cl_lock);
 305         return cred;
 306 }
 307
 308 /**
 309  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 310  * @clp: client state handle
 311  *
 312  * Returns an rpc_cred with reference count bumped, or NULL.
 313  */
 314 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 315 {
 316         struct nfs_server *server;
 317         struct rpc_cred *cred;
 318
 319         spin_lock(&clp->cl_lock);
 320         cred = nfs4_get_machine_cred_locked(clp);
 321         spin_unlock(&clp->cl_lock);
 322         if (cred != NULL)
 323                 goto out;
 324
 325         rcu_read_lock();
 326         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 327                 cred = nfs4_get_setclientid_cred_server(server);
 328                 if (cred != NULL)
 329                         break;
 330         }
 331         rcu_read_unlock();
 332
 333 out:
 334         return cred;
 335 }
 336
 337 static struct nfs4_state_owner *
 338 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 339 {
 340         struct rb_node **p = &server->state_owners.rb_node,
 341                        *parent = NULL;
 342         struct nfs4_state_owner *sp;
 343
 344         while (*p != NULL) {
 345                 parent = *p;
 346                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 347
 348                 if (cred < sp->so_cred)
 349                         p = &parent->rb_left;
 350                 else if (cred > sp->so_cred)
 351                         p = &parent->rb_right;
 352                 else {
 353                         if (!list_empty(&sp->so_lru))
 354                                 list_del_init(&sp->so_lru);
 355                         atomic_inc(&sp->so_count);
 356                         return sp;
 357                 }
 358         }
 359         return NULL;
 360 }
 361
 362 static struct nfs4_state_owner *
 363 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 364 {
 365         struct nfs_server *server = new->so_server;
 366         struct rb_node **p = &server->state_owners.rb_node,
 367                        *parent = NULL;
 368         struct nfs4_state_owner *sp;
 369         int err;
 370
 371         while (*p != NULL) {
 372                 parent = *p;
 373                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 374
 375                 if (new->so_cred < sp->so_cred)
 376                         p = &parent->rb_left;
 377                 else if (new->so_cred > sp->so_cred)
 378                         p = &parent->rb_right;
 379                 else {
 380                         if (!list_empty(&sp->so_lru))
 381                                 list_del_init(&sp->so_lru);
 382                         atomic_inc(&sp->so_count);
 383                         return sp;
 384                 }
 385         }
 386         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 387         if (err)
 388                 return ERR_PTR(err);
 389         rb_link_node(&new->so_server_node, parent, p);
 390         rb_insert_color(&new->so_server_node, &server->state_owners);
 391         return new;
 392 }
 393
 394 static void
 395 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 396 {
 397         struct nfs_server *server = sp->so_server;
 398
 399         if (!RB_EMPTY_NODE(&sp->so_server_node))
 400                 rb_erase(&sp->so_server_node, &server->state_owners);
 401         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 402 }
 403
 404 static void
 405 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 406 {
 407         sc->create_time = ktime_get();
 408         sc->flags = 0;
 409         sc->counter = 0;
 410         spin_lock_init(&sc->lock);
 411         INIT_LIST_HEAD(&sc->list);
 412         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 413 }
 414
 415 static void
 416 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 417 {
 418         rpc_destroy_wait_queue(&sc->wait);
 419 }
 420
 421 /*
 422  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 423  * create a new state_owner.
 424  *
 425  */
 426 static struct nfs4_state_owner *
 427 nfs4_alloc_state_owner(struct nfs_server *server,
 428                 struct rpc_cred *cred,
 429                 gfp_t gfp_flags)
 430 {
 431         struct nfs4_state_owner *sp;
 432
 433         sp = kzalloc(sizeof(*sp), gfp_flags);
 434         if (!sp)
 435                 return NULL;
 436         sp->so_server = server;
 437         sp->so_cred = get_rpccred(cred);
 438         spin_lock_init(&sp->so_lock);
 439         INIT_LIST_HEAD(&sp->so_states);
 440         nfs4_init_seqid_counter(&sp->so_seqid);
 441         atomic_set(&sp->so_count, 1);
 442         INIT_LIST_HEAD(&sp->so_lru);
 443         return sp;
 444 }
 445
 446 static void
 447 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 448 {
 449         struct rb_node *rb_node = &sp->so_server_node;
 450
 451         if (!RB_EMPTY_NODE(rb_node)) {
 452                 struct nfs_server *server = sp->so_server;
 453                 struct nfs_client *clp = server->nfs_client;
 454
 455                 spin_lock(&clp->cl_lock);
 456                 if (!RB_EMPTY_NODE(rb_node)) {
 457                         rb_erase(rb_node, &server->state_owners);
 458                         RB_CLEAR_NODE(rb_node);
 459                 }
 460                 spin_unlock(&clp->cl_lock);
 461         }
 462 }
 463
 464 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 465 {
 466         nfs4_destroy_seqid_counter(&sp->so_seqid);
 467         put_rpccred(sp->so_cred);
 468         kfree(sp);
 469 }
 470
 471 static void nfs4_gc_state_owners(struct nfs_server *server)
 472 {
 473         struct nfs_client *clp = server->nfs_client;
 474         struct nfs4_state_owner *sp, *tmp;
 475         unsigned long time_min, time_max;
 476         LIST_HEAD(doomed);
 477
 478         spin_lock(&clp->cl_lock);
 479         time_max = jiffies;
 480         time_min = (long)time_max - (long)clp->cl_lease_time;
 481         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 482                 /* NB: LRU is sorted so that oldest is at the head */
 483                 if (time_in_range(sp->so_expires, time_min, time_max))
 484                         break;
 485                 list_move(&sp->so_lru, &doomed);
 486                 nfs4_remove_state_owner_locked(sp);
 487         }
 488         spin_unlock(&clp->cl_lock);
 489
 490         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 491                 list_del(&sp->so_lru);
 492                 nfs4_free_state_owner(sp);
 493         }
 494 }
 495
 496 /**
 497  * nfs4_get_state_owner - Look up a state owner given a credential
 498  * @server: nfs_server to search
 499  * @cred: RPC credential to match
 500  *
 501  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 502  */
 503 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 504                                               struct rpc_cred *cred,
 505                                               gfp_t gfp_flags)
 506 {
 507         struct nfs_client *clp = server->nfs_client;
 508         struct nfs4_state_owner *sp, *new;
 509
 510         spin_lock(&clp->cl_lock);
 511         sp = nfs4_find_state_owner_locked(server, cred);
 512         spin_unlock(&clp->cl_lock);
 513         if (sp != NULL)
 514                 goto out;
 515         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 516         if (new == NULL)
 517                 goto out;
 518         do {
 519                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 520                         break;
 521                 spin_lock(&clp->cl_lock);
 522                 sp = nfs4_insert_state_owner_locked(new);
 523                 spin_unlock(&clp->cl_lock);
 524         } while (sp == ERR_PTR(-EAGAIN));
 525         if (sp != new)
 526                 nfs4_free_state_owner(new);
 527 out:
 528         nfs4_gc_state_owners(server);
 529         return sp;
 530 }
 531
 532 /**
 533  * nfs4_put_state_owner - Release a nfs4_state_owner
 534  * @sp: state owner data to release
 535  *
 536  * Note that we keep released state owners on an LRU
 537  * list.
 538  * This caches valid state owners so that they can be
 539  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 540  * It also pins the uniquifier of dropped state owners for
 541  * a while, to ensure that those state owner names are
 542  * never reused.
 543  */
 544 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 545 {
 546         struct nfs_server *server = sp->so_server;
 547         struct nfs_client *clp = server->nfs_client;
 548
 549         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 550                 return;
 551
 552         sp->so_expires = jiffies;
 553         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 554         spin_unlock(&clp->cl_lock);
 555 }
 556
 557 /**
 558  * nfs4_purge_state_owners - Release all cached state owners
 559  * @server: nfs_server with cached state owners to release
 560  *
 561  * Called at umount time.  Remaining state owners will be on
 562  * the LRU with ref count of zero.
 563  */
 564 void nfs4_purge_state_owners(struct nfs_server *server)
 565 {
 566         struct nfs_client *clp = server->nfs_client;
 567         struct nfs4_state_owner *sp, *tmp;
 568         LIST_HEAD(doomed);
 569
 570         spin_lock(&clp->cl_lock);
 571         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 572                 list_move(&sp->so_lru, &doomed);
 573                 nfs4_remove_state_owner_locked(sp);
 574         }
 575         spin_unlock(&clp->cl_lock);
 576
 577         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 578                 list_del(&sp->so_lru);
 579                 nfs4_free_state_owner(sp);
 580         }
 581 }
 582
 583 static struct nfs4_state *
 584 nfs4_alloc_open_state(void)
 585 {
 586         struct nfs4_state *state;
 587
 588         state = kzalloc(sizeof(*state), GFP_NOFS);
 589         if (!state)
 590                 return NULL;
 591         atomic_set(&state->count, 1);
 592         INIT_LIST_HEAD(&state->lock_states);
 593         spin_lock_init(&state->state_lock);
 594         seqlock_init(&state->seqlock);
 595         return state;
 596 }
 597
 598 void
 599 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 600 {
 601         if (state->state == fmode)
 602                 return;
 603         /* NB! List reordering - see the reclaim code for why.  */
 604         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 605                 if (fmode & FMODE_WRITE)
 606                         list_move(&state->open_states, &state->owner->so_states);
 607                 else
 608                         list_move_tail(&state->open_states, &state->owner->so_states);
 609         }
 610         state->state = fmode;
 611 }
 612
 613 static struct nfs4_state *
 614 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 615 {
 616         struct nfs_inode *nfsi = NFS_I(inode);
 617         struct nfs4_state *state;
 618
 619         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 620                 if (state->owner != owner)
 621                         continue;
 622                 if (atomic_inc_not_zero(&state->count))
 623                         return state;
 624         }
 625         return NULL;
 626 }
 627
 628 static void
 629 nfs4_free_open_state(struct nfs4_state *state)
 630 {
 631         kfree(state);
 632 }
 633
 634 struct nfs4_state *
 635 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 636 {
 637         struct nfs4_state *state, *new;
 638         struct nfs_inode *nfsi = NFS_I(inode);
 639
 640         spin_lock(&inode->i_lock);
 641         state = __nfs4_find_state_byowner(inode, owner);
 642         spin_unlock(&inode->i_lock);
 643         if (state)
 644                 goto out;
 645         new = nfs4_alloc_open_state();
 646         spin_lock(&owner->so_lock);
 647         spin_lock(&inode->i_lock);
 648         state = __nfs4_find_state_byowner(inode, owner);
 649         if (state == NULL && new != NULL) {
 650                 state = new;
 651                 state->owner = owner;
 652                 atomic_inc(&owner->so_count);
 653                 list_add(&state->inode_states, &nfsi->open_states);
 654                 ihold(inode);
 655                 state->inode = inode;
 656                 spin_unlock(&inode->i_lock);
 657                 /* Note: The reclaim code dictates that we add stateless
 658                  * and read-only stateids to the end of the list */
 659                 list_add_tail(&state->open_states, &owner->so_states);
 660                 spin_unlock(&owner->so_lock);
 661         } else {
 662                 spin_unlock(&inode->i_lock);
 663                 spin_unlock(&owner->so_lock);
 664                 if (new)
 665                         nfs4_free_open_state(new);
 666         }
 667 out:
 668         return state;
 669 }
 670
 671 void nfs4_put_open_state(struct nfs4_state *state)
 672 {
 673         struct inode *inode = state->inode;
 674         struct nfs4_state_owner *owner = state->owner;
 675
 676         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 677                 return;
 678         spin_lock(&inode->i_lock);
 679         list_del(&state->inode_states);
 680         list_del(&state->open_states);
 681         spin_unlock(&inode->i_lock);
 682         spin_unlock(&owner->so_lock);
 683         iput(inode);
 684         nfs4_free_open_state(state);
 685         nfs4_put_state_owner(owner);
 686 }
 687
 688 /*
 689  * Close the current file.
 690  */
 691 static void __nfs4_close(struct nfs4_state *state,
 692                 fmode_t fmode, gfp_t gfp_mask, int wait)
 693 {
 694         struct nfs4_state_owner *owner = state->owner;
 695         int call_close = 0;
 696         fmode_t newstate;
 697
 698         atomic_inc(&owner->so_count);
 699         /* Protect against nfs4_find_state() */
 700         spin_lock(&owner->so_lock);
 701         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 702                 case FMODE_READ:
 703                         state->n_rdonly--;
 704                         break;
 705                 case FMODE_WRITE:
 706                         state->n_wronly--;
 707                         break;
 708                 case FMODE_READ|FMODE_WRITE:
 709                         state->n_rdwr--;
 710         }
 711         newstate = FMODE_READ|FMODE_WRITE;
 712         if (state->n_rdwr == 0) {
 713                 if (state->n_rdonly == 0) {
 714                         newstate &= ~FMODE_READ;
 715                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 716                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 717                 }
 718                 if (state->n_wronly == 0) {
 719                         newstate &= ~FMODE_WRITE;
 720                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 721                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 722                 }
 723                 if (newstate == 0)
 724                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 725         }
 726         nfs4_state_set_mode_locked(state, newstate);
 727         spin_unlock(&owner->so_lock);
 728
 729         if (!call_close) {
 730                 nfs4_put_open_state(state);
 731                 nfs4_put_state_owner(owner);
 732         } else {
 733                 bool roc = pnfs_roc(state->inode);
 734
 735                 nfs4_do_close(state, gfp_mask, wait, roc);
 736         }
 737 }
 738
 739 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 740 {
 741         __nfs4_close(state, fmode, GFP_NOFS, 0);
 742 }
 743
 744 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 745 {
 746         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 747 }
 748
 749 /*
 750  * Search the state->lock_states for an existing lock_owner
 751  * that is compatible with current->files
 752  */
 753 static struct nfs4_lock_state *
 754 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 755 {
 756         struct nfs4_lock_state *pos;
 757         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 758                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 759                         continue;
 760                 switch (pos->ls_owner.lo_type) {
 761                 case NFS4_POSIX_LOCK_TYPE:
 762                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 763                                 continue;
 764                         break;
 765                 case NFS4_FLOCK_LOCK_TYPE:
 766                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 767                                 continue;
 768                 }
 769                 atomic_inc(&pos->ls_count);
 770                 return pos;
 771         }
 772         return NULL;
 773 }
 774
 775 /*
 776  * Return a compatible lock_state. If no initialized lock_state structure
 777  * exists, return an uninitialized one.
 778  *
 779  */
 780 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)
 781 {
 782         struct nfs4_lock_state *lsp;
 783         struct nfs_server *server = state->owner->so_server;
 784
 785         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 786         if (lsp == NULL)
 787                 return NULL;
 788         nfs4_init_seqid_counter(&lsp->ls_seqid);
 789         atomic_set(&lsp->ls_count, 1);
 790         lsp->ls_state = state;
 791         lsp->ls_owner.lo_type = type;
 792         switch (lsp->ls_owner.lo_type) {
 793         case NFS4_FLOCK_LOCK_TYPE:
 794                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 795                 break;
 796         case NFS4_POSIX_LOCK_TYPE:
 797                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 798                 break;
 799         default:
 800                 goto out_free;
 801         }
 802         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 803         if (lsp->ls_seqid.owner_id < 0)
 804                 goto out_free;
 805         INIT_LIST_HEAD(&lsp->ls_locks);
 806         return lsp;
 807 out_free:
 808         kfree(lsp);
 809         return NULL;
 810 }
 811
 812 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 813 {
 814         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 815         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 816         kfree(lsp);
 817 }
 818
 819 /*
 820  * Return a compatible lock_state. If no initialized lock_state structure
 821  * exists, return an uninitialized one.
 822  *
 823  */
 824 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 825 {
 826         struct nfs4_lock_state *lsp, *new = NULL;
 827
 828         for(;;) {
 829                 spin_lock(&state->state_lock);
 830                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 831                 if (lsp != NULL)
 832                         break;
 833                 if (new != NULL) {
 834                         list_add(&new->ls_locks, &state->lock_states);
 835                         set_bit(LK_STATE_IN_USE, &state->flags);
 836                         lsp = new;
 837                         new = NULL;
 838                         break;
 839                 }
 840                 spin_unlock(&state->state_lock);
 841                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 842                 if (new == NULL)
 843                         return NULL;
 844         }
 845         spin_unlock(&state->state_lock);
 846         if (new != NULL)
 847                 nfs4_free_lock_state(state->owner->so_server, new);
 848         return lsp;
 849 }
 850
 851 /*
 852  * Release reference to lock_state, and free it if we see that
 853  * it is no longer in use
 854  */
 855 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 856 {
 857         struct nfs4_state *state;
 858
 859         if (lsp == NULL)
 860                 return;
 861         state = lsp->ls_state;
 862         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 863                 return;
 864         list_del(&lsp->ls_locks);
 865         if (list_empty(&state->lock_states))
 866                 clear_bit(LK_STATE_IN_USE, &state->flags);
 867         spin_unlock(&state->state_lock);
 868         if (lsp->ls_flags & NFS_LOCK_INITIALIZED) {
 869                 if (nfs4_release_lockowner(lsp) == 0)
 870                         return;
 871         }
 872         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 873 }
 874
 875 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 876 {
 877         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 878
 879         dst->fl_u.nfs4_fl.owner = lsp;
 880         atomic_inc(&lsp->ls_count);
 881 }
 882
 883 static void nfs4_fl_release_lock(struct file_lock *fl)
 884 {
 885         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 886 }
 887
 888 static const struct file_lock_operations nfs4_fl_lock_ops = {
 889         .fl_copy_lock = nfs4_fl_copy_lock,
 890         .fl_release_private = nfs4_fl_release_lock,
 891 };
 892
 893 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 894 {
 895         struct nfs4_lock_state *lsp;
 896
 897         if (fl->fl_ops != NULL)
 898                 return 0;
 899         if (fl->fl_flags & FL_POSIX)
 900                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 901         else if (fl->fl_flags & FL_FLOCK)
 902                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 903                                 NFS4_FLOCK_LOCK_TYPE);
 904         else
 905                 return -EINVAL;
 906         if (lsp == NULL)
 907                 return -ENOMEM;
 908         fl->fl_u.nfs4_fl.owner = lsp;
 909         fl->fl_ops = &nfs4_fl_lock_ops;
 910         return 0;
 911 }
 912
 913 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 914                 fl_owner_t fl_owner, pid_t fl_pid)
 915 {
 916         struct nfs4_lock_state *lsp;
 917         bool ret = false;
 918
 919         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 920                 goto out;
 921
 922         spin_lock(&state->state_lock);
 923         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 924         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
 925                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 926                 ret = true;
 927         }
 928         spin_unlock(&state->state_lock);
 929         nfs4_put_lock_state(lsp);
 930 out:
 931         return ret;
 932 }
 933
 934 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 935 {
 936         int seq;
 937
 938         do {
 939                 seq = read_seqbegin(&state->seqlock);
 940                 nfs4_stateid_copy(dst, &state->stateid);
 941         } while (read_seqretry(&state->seqlock, seq));
 942 }
 943
 944 /*
 945  * Byte-range lock aware utility to initialize the stateid of read/write
 946  * requests.
 947  */
 948 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 949                 fmode_t fmode, fl_owner_t fl_owner, pid_t fl_pid)
 950 {
 951         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
 952                 return;
 953         if (nfs4_copy_lock_stateid(dst, state, fl_owner, fl_pid))
 954                 return;
 955         nfs4_copy_open_stateid(dst, state);
 956 }
 957
 958 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 959 {
 960         struct nfs_seqid *new;
 961
 962         new = kmalloc(sizeof(*new), gfp_mask);
 963         if (new != NULL) {
 964                 new->sequence = counter;
 965                 INIT_LIST_HEAD(&new->list);
 966                 new->task = NULL;
 967         }
 968         return new;
 969 }
 970
 971 void nfs_release_seqid(struct nfs_seqid *seqid)
 972 {
 973         struct nfs_seqid_counter *sequence;
 974
 975         if (list_empty(&seqid->list))
 976                 return;
 977         sequence = seqid->sequence;
 978         spin_lock(&sequence->lock);
 979         list_del_init(&seqid->list);
 980         if (!list_empty(&sequence->list)) {
 981                 struct nfs_seqid *next;
 982
 983                 next = list_first_entry(&sequence->list,
 984                                 struct nfs_seqid, list);
 985                 rpc_wake_up_queued_task(&sequence->wait, next->task);
 986         }
 987         spin_unlock(&sequence->lock);
 988 }
 989
 990 void nfs_free_seqid(struct nfs_seqid *seqid)
 991 {
 992         nfs_release_seqid(seqid);
 993         kfree(seqid);
 994 }
 995
 996 /*
 997  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 998  * failed with a seqid incrementing error -
 999  * see comments nfs_fs.h:seqid_mutating_error()
1000  */
1001 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1002 {
1003         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
1004         switch (status) {
1005                 case 0:
1006                         break;
1007                 case -NFS4ERR_BAD_SEQID:
1008                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1009                                 return;
1010                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1011                                         " sequence-id error on an"
1012                                         " unconfirmed sequence %p!\n",
1013                                         seqid->sequence);
1014                 case -NFS4ERR_STALE_CLIENTID:
1015                 case -NFS4ERR_STALE_STATEID:
1016                 case -NFS4ERR_BAD_STATEID:
1017                 case -NFS4ERR_BADXDR:
1018                 case -NFS4ERR_RESOURCE:
1019                 case -NFS4ERR_NOFILEHANDLE:
1020                         /* Non-seqid mutating errors */
1021                         return;
1022         };
1023         /*
1024          * Note: no locking needed as we are guaranteed to be first
1025          * on the sequence list
1026          */
1027         seqid->sequence->counter++;
1028 }
1029
1030 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1031 {
1032         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1033                                         struct nfs4_state_owner, so_seqid);
1034         struct nfs_server *server = sp->so_server;
1035
1036         if (status == -NFS4ERR_BAD_SEQID)
1037                 nfs4_drop_state_owner(sp);
1038         if (!nfs4_has_session(server->nfs_client))
1039                 nfs_increment_seqid(status, seqid);
1040 }
1041
1042 /*
1043  * Increment the seqid if the LOCK/LOCKU succeeded, or
1044  * failed with a seqid incrementing error -
1045  * see comments nfs_fs.h:seqid_mutating_error()
1046  */
1047 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1048 {
1049         nfs_increment_seqid(status, seqid);
1050 }
1051
1052 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1053 {
1054         struct nfs_seqid_counter *sequence = seqid->sequence;
1055         int status = 0;
1056
1057         spin_lock(&sequence->lock);
1058         seqid->task = task;
1059         if (list_empty(&seqid->list))
1060                 list_add_tail(&seqid->list, &sequence->list);
1061         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1062                 goto unlock;
1063         rpc_sleep_on(&sequence->wait, task, NULL);
1064         status = -EAGAIN;
1065 unlock:
1066         spin_unlock(&sequence->lock);
1067         return status;
1068 }
1069
1070 static int nfs4_run_state_manager(void *);
1071
1072 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1073 {
1074         smp_mb__before_clear_bit();
1075         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1076         smp_mb__after_clear_bit();
1077         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1078         rpc_wake_up(&clp->cl_rpcwaitq);
1079 }
1080
1081 /*
1082  * Schedule the nfs_client asynchronous state management routine
1083  */
1084 void nfs4_schedule_state_manager(struct nfs_client *clp)
1085 {
1086         struct task_struct *task;
1087         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1088
1089         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1090                 return;
1091         __module_get(THIS_MODULE);
1092         atomic_inc(&clp->cl_count);
1093
1094         /* The rcu_read_lock() is not strictly necessary, as the state
1095          * manager is the only thread that ever changes the rpc_xprt
1096          * after it's initialized.  At this point, we're single threaded. */
1097         rcu_read_lock();
1098         snprintf(buf, sizeof(buf), "%s-manager",
1099                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1100         rcu_read_unlock();
1101         task = kthread_run(nfs4_run_state_manager, clp, buf);
1102         if (IS_ERR(task)) {
1103                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1104                         __func__, PTR_ERR(task));
1105                 nfs4_clear_state_manager_bit(clp);
1106                 nfs_put_client(clp);
1107                 module_put(THIS_MODULE);
1108         }
1109 }
1110
1111 /*
1112  * Schedule a lease recovery attempt
1113  */
1114 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1115 {
1116         if (!clp)
1117                 return;
1118         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1119                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1120         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1121                         clp->cl_hostname);
1122         nfs4_schedule_state_manager(clp);
1123 }
1124 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1125
1126 /*
1127  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1128  * @clp: client to process
1129  *
1130  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1131  * resend of the SETCLIENTID and hence re-establish the
1132  * callback channel. Then return all existing delegations.
1133  */
1134 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1135 {
1136         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1137         nfs_expire_all_delegations(clp);
1138         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1139                         clp->cl_hostname);
1140 }
1141
1142 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1143 {
1144         nfs40_handle_cb_pathdown(clp);
1145         nfs4_schedule_state_manager(clp);
1146 }
1147
1148 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1149 {
1150
1151         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1152         /* Don't recover state that expired before the reboot */
1153         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1154                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1155                 return 0;
1156         }
1157         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1158         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1159         return 1;
1160 }
1161
1162 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1163 {
1164         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1165         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1166         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1167         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1168         return 1;
1169 }
1170
1171 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1172 {
1173         struct nfs_client *clp = server->nfs_client;
1174
1175         nfs4_state_mark_reclaim_nograce(clp, state);
1176         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1177                         clp->cl_hostname);
1178         nfs4_schedule_state_manager(clp);
1179 }
1180 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1181
1182 void nfs_inode_find_state_and_recover(struct inode *inode,
1183                 const nfs4_stateid *stateid)
1184 {
1185         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1186         struct nfs_inode *nfsi = NFS_I(inode);
1187         struct nfs_open_context *ctx;
1188         struct nfs4_state *state;
1189         bool found = false;
1190
1191         spin_lock(&inode->i_lock);
1192         list_for_each_entry(ctx, &nfsi->open_files, list) {
1193                 state = ctx->state;
1194                 if (state == NULL)
1195                         continue;
1196                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1197                         continue;
1198                 if (!nfs4_stateid_match(&state->stateid, stateid))
1199                         continue;
1200                 nfs4_state_mark_reclaim_nograce(clp, state);
1201                 found = true;
1202         }
1203         spin_unlock(&inode->i_lock);
1204         if (found)
1205                 nfs4_schedule_state_manager(clp);
1206 }
1207
1208
1209 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1210 {
1211         struct inode *inode = state->inode;
1212         struct nfs_inode *nfsi = NFS_I(inode);
1213         struct file_lock *fl;
1214         int status = 0;
1215
1216         if (inode->i_flock == NULL)
1217                 return 0;
1218
1219         /* Guard against delegation returns and new lock/unlock calls */
1220         down_write(&nfsi->rwsem);
1221         /* Protect inode->i_flock using the BKL */
1222         lock_flocks();
1223         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1224                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1225                         continue;
1226                 if (nfs_file_open_context(fl->fl_file)->state != state)
1227                         continue;
1228                 unlock_flocks();
1229                 status = ops->recover_lock(state, fl);
1230                 switch (status) {
1231                         case 0:
1232                                 break;
1233                         case -ESTALE:
1234                         case -NFS4ERR_ADMIN_REVOKED:
1235                         case -NFS4ERR_STALE_STATEID:
1236                         case -NFS4ERR_BAD_STATEID:
1237                         case -NFS4ERR_EXPIRED:
1238                         case -NFS4ERR_NO_GRACE:
1239                         case -NFS4ERR_STALE_CLIENTID:
1240                         case -NFS4ERR_BADSESSION:
1241                         case -NFS4ERR_BADSLOT:
1242                         case -NFS4ERR_BAD_HIGH_SLOT:
1243                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1244                                 goto out;
1245                         default:
1246                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1247                                         "Zeroing state\n", __func__, status);
1248                         case -ENOMEM:
1249                         case -NFS4ERR_DENIED:
1250                         case -NFS4ERR_RECLAIM_BAD:
1251                         case -NFS4ERR_RECLAIM_CONFLICT:
1252                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1253                                 status = 0;
1254                 }
1255                 lock_flocks();
1256         }
1257         unlock_flocks();
1258 out:
1259         up_write(&nfsi->rwsem);
1260         return status;
1261 }
1262
1263 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1264 {
1265         struct nfs4_state *state;
1266         struct nfs4_lock_state *lock;
1267         int status = 0;
1268
1269         /* Note: we rely on the sp->so_states list being ordered
1270          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1271          * states first.
1272          * This is needed to ensure that the server won't give us any
1273          * read delegations that we have to return if, say, we are
1274          * recovering after a network partition or a reboot from a
1275          * server that doesn't support a grace period.
1276          */
1277 restart:
1278         spin_lock(&sp->so_lock);
1279         list_for_each_entry(state, &sp->so_states, open_states) {
1280                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1281                         continue;
1282                 if (state->state == 0)
1283                         continue;
1284                 atomic_inc(&state->count);
1285                 spin_unlock(&sp->so_lock);
1286                 status = ops->recover_open(sp, state);
1287                 if (status >= 0) {
1288                         status = nfs4_reclaim_locks(state, ops);
1289                         if (status >= 0) {
1290                                 spin_lock(&state->state_lock);
1291                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1292                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1293                                                 pr_warn_ratelimited("NFS: "
1294                                                         "%s: Lock reclaim "
1295                                                         "failed!\n", __func__);
1296                                 }
1297                                 spin_unlock(&state->state_lock);
1298                                 nfs4_put_open_state(state);
1299                                 goto restart;
1300                         }
1301                 }
1302                 switch (status) {
1303                         default:
1304                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1305                                         "Zeroing state\n", __func__, status);
1306                         case -ENOENT:
1307                         case -ENOMEM:
1308                         case -ESTALE:
1309                                 /*
1310                                  * Open state on this file cannot be recovered
1311                                  * All we can do is revert to using the zero stateid.
1312                                  */
1313                                 memset(&state->stateid, 0,
1314                                         sizeof(state->stateid));
1315                                 /* Mark the file as being 'closed' */
1316                                 state->state = 0;
1317                                 break;
1318                         case -EKEYEXPIRED:
1319                                 /*
1320                                  * User RPCSEC_GSS context has expired.
1321                                  * We cannot recover this stateid now, so
1322                                  * skip it and allow recovery thread to
1323                                  * proceed.
1324                                  */
1325                                 break;
1326                         case -NFS4ERR_ADMIN_REVOKED:
1327                         case -NFS4ERR_STALE_STATEID:
1328                         case -NFS4ERR_BAD_STATEID:
1329                         case -NFS4ERR_RECLAIM_BAD:
1330                         case -NFS4ERR_RECLAIM_CONFLICT:
1331                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1332                                 break;
1333                         case -NFS4ERR_EXPIRED:
1334                         case -NFS4ERR_NO_GRACE:
1335                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1336                         case -NFS4ERR_STALE_CLIENTID:
1337                         case -NFS4ERR_BADSESSION:
1338                         case -NFS4ERR_BADSLOT:
1339                         case -NFS4ERR_BAD_HIGH_SLOT:
1340                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1341                                 goto out_err;
1342                 }
1343                 nfs4_put_open_state(state);
1344                 goto restart;
1345         }
1346         spin_unlock(&sp->so_lock);
1347         return 0;
1348 out_err:
1349         nfs4_put_open_state(state);
1350         return status;
1351 }
1352
1353 static void nfs4_clear_open_state(struct nfs4_state *state)
1354 {
1355         struct nfs4_lock_state *lock;
1356
1357         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1358         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1359         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1360         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1361         spin_lock(&state->state_lock);
1362         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1363                 lock->ls_seqid.flags = 0;
1364                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1365         }
1366         spin_unlock(&state->state_lock);
1367 }
1368
1369 static void nfs4_reset_seqids(struct nfs_server *server,
1370         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1371 {
1372         struct nfs_client *clp = server->nfs_client;
1373         struct nfs4_state_owner *sp;
1374         struct rb_node *pos;
1375         struct nfs4_state *state;
1376
1377         spin_lock(&clp->cl_lock);
1378         for (pos = rb_first(&server->state_owners);
1379              pos != NULL;
1380              pos = rb_next(pos)) {
1381                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1382                 sp->so_seqid.flags = 0;
1383                 spin_lock(&sp->so_lock);
1384                 list_for_each_entry(state, &sp->so_states, open_states) {
1385                         if (mark_reclaim(clp, state))
1386                                 nfs4_clear_open_state(state);
1387                 }
1388                 spin_unlock(&sp->so_lock);
1389         }
1390         spin_unlock(&clp->cl_lock);
1391 }
1392
1393 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1394         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1395 {
1396         struct nfs_server *server;
1397
1398         rcu_read_lock();
1399         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1400                 nfs4_reset_seqids(server, mark_reclaim);
1401         rcu_read_unlock();
1402 }
1403
1404 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1405 {
1406         /* Mark all delegations for reclaim */
1407         nfs_delegation_mark_reclaim(clp);
1408         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1409 }
1410
1411 static void nfs4_reclaim_complete(struct nfs_client *clp,
1412                                  const struct nfs4_state_recovery_ops *ops)
1413 {
1414         /* Notify the server we're done reclaiming our state */
1415         if (ops->reclaim_complete)
1416                 (void)ops->reclaim_complete(clp);
1417 }
1418
1419 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1420 {
1421         struct nfs_client *clp = server->nfs_client;
1422         struct nfs4_state_owner *sp;
1423         struct rb_node *pos;
1424         struct nfs4_state *state;
1425
1426         spin_lock(&clp->cl_lock);
1427         for (pos = rb_first(&server->state_owners);
1428              pos != NULL;
1429              pos = rb_next(pos)) {
1430                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1431                 spin_lock(&sp->so_lock);
1432                 list_for_each_entry(state, &sp->so_states, open_states) {
1433                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1434                                                 &state->flags))
1435                                 continue;
1436                         nfs4_state_mark_reclaim_nograce(clp, state);
1437                 }
1438                 spin_unlock(&sp->so_lock);
1439         }
1440         spin_unlock(&clp->cl_lock);
1441 }
1442
1443 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1444 {
1445         struct nfs_server *server;
1446
1447         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1448                 return 0;
1449
1450         rcu_read_lock();
1451         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1452                 nfs4_clear_reclaim_server(server);
1453         rcu_read_unlock();
1454
1455         nfs_delegation_reap_unclaimed(clp);
1456         return 1;
1457 }
1458
1459 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1460 {
1461         if (!nfs4_state_clear_reclaim_reboot(clp))
1462                 return;
1463         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1464 }
1465
1466 static void nfs_delegation_clear_all(struct nfs_client *clp)
1467 {
1468         nfs_delegation_mark_reclaim(clp);
1469         nfs_delegation_reap_unclaimed(clp);
1470 }
1471
1472 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1473 {
1474         nfs_delegation_clear_all(clp);
1475         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1476 }
1477
1478 static void nfs4_warn_keyexpired(const char *s)
1479 {
1480         printk_ratelimited(KERN_WARNING "Error: state manager"
1481                         " encountered RPCSEC_GSS session"
1482                         " expired against NFSv4 server %s.\n",
1483                         s);
1484 }
1485
1486 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1487 {
1488         switch (error) {
1489                 case 0:
1490                         break;
1491                 case -NFS4ERR_CB_PATH_DOWN:
1492                         nfs40_handle_cb_pathdown(clp);
1493                         break;
1494                 case -NFS4ERR_NO_GRACE:
1495                         nfs4_state_end_reclaim_reboot(clp);
1496                         break;
1497                 case -NFS4ERR_STALE_CLIENTID:
1498                 case -NFS4ERR_LEASE_MOVED:
1499                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1500                         nfs4_state_clear_reclaim_reboot(clp);
1501                         nfs4_state_start_reclaim_reboot(clp);
1502                         break;
1503                 case -NFS4ERR_EXPIRED:
1504                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1505                         nfs4_state_start_reclaim_nograce(clp);
1506                         break;
1507                 case -NFS4ERR_BADSESSION:
1508                 case -NFS4ERR_BADSLOT:
1509                 case -NFS4ERR_BAD_HIGH_SLOT:
1510                 case -NFS4ERR_DEADSESSION:
1511                 case -NFS4ERR_SEQ_FALSE_RETRY:
1512                 case -NFS4ERR_SEQ_MISORDERED:
1513                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1514                         /* Zero session reset errors */
1515                         break;
1516                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1517                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1518                         break;
1519                 case -EKEYEXPIRED:
1520                         /* Nothing we can do */
1521                         nfs4_warn_keyexpired(clp->cl_hostname);
1522                         break;
1523                 default:
1524                         dprintk("%s: failed to handle error %d for server %s\n",
1525                                         __func__, error, clp->cl_hostname);
1526                         return error;
1527         }
1528         dprintk("%s: handled error %d for server %s\n", __func__, error,
1529                         clp->cl_hostname);
1530         return 0;
1531 }
1532
1533 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1534 {
1535         struct nfs4_state_owner *sp;
1536         struct nfs_server *server;
1537         struct rb_node *pos;
1538         int status = 0;
1539
1540 restart:
1541         rcu_read_lock();
1542         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1543                 nfs4_purge_state_owners(server);
1544                 spin_lock(&clp->cl_lock);
1545                 for (pos = rb_first(&server->state_owners);
1546                      pos != NULL;
1547                      pos = rb_next(pos)) {
1548                         sp = rb_entry(pos,
1549                                 struct nfs4_state_owner, so_server_node);
1550                         if (!test_and_clear_bit(ops->owner_flag_bit,
1551                                                         &sp->so_flags))
1552                                 continue;
1553                         atomic_inc(&sp->so_count);
1554                         spin_unlock(&clp->cl_lock);
1555                         rcu_read_unlock();
1556
1557                         status = nfs4_reclaim_open_state(sp, ops);
1558                         if (status < 0) {
1559                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1560                                 nfs4_put_state_owner(sp);
1561                                 return nfs4_recovery_handle_error(clp, status);
1562                         }
1563
1564                         nfs4_put_state_owner(sp);
1565                         goto restart;
1566                 }
1567                 spin_unlock(&clp->cl_lock);
1568         }
1569         rcu_read_unlock();
1570         return status;
1571 }
1572
1573 static int nfs4_check_lease(struct nfs_client *clp)
1574 {
1575         struct rpc_cred *cred;
1576         const struct nfs4_state_maintenance_ops *ops =
1577                 clp->cl_mvops->state_renewal_ops;
1578         int status;
1579
1580         /* Is the client already known to have an expired lease? */
1581         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1582                 return 0;
1583         spin_lock(&clp->cl_lock);
1584         cred = ops->get_state_renewal_cred_locked(clp);
1585         spin_unlock(&clp->cl_lock);
1586         if (cred == NULL) {
1587                 cred = nfs4_get_setclientid_cred(clp);
1588                 status = -ENOKEY;
1589                 if (cred == NULL)
1590                         goto out;
1591         }
1592         status = ops->renew_lease(clp, cred);
1593         put_rpccred(cred);
1594 out:
1595         return nfs4_recovery_handle_error(clp, status);
1596 }
1597
1598 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1599  * on EXCHANGE_ID for v4.1
1600  */
1601 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1602 {
1603         switch (status) {
1604         case -NFS4ERR_SEQ_MISORDERED:
1605                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1606                         return -ESERVERFAULT;
1607                 /* Lease confirmation error: retry after purging the lease */
1608                 ssleep(1);
1609         case -NFS4ERR_CLID_INUSE:
1610         case -NFS4ERR_STALE_CLIENTID:
1611                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1612                 break;
1613         case -EACCES:
1614                 if (clp->cl_machine_cred == NULL)
1615                         return -EACCES;
1616                 /* Handle case where the user hasn't set up machine creds */
1617                 nfs4_clear_machine_cred(clp);
1618         case -NFS4ERR_DELAY:
1619         case -ETIMEDOUT:
1620         case -EAGAIN:
1621                 ssleep(1);
1622                 break;
1623
1624         case -NFS4ERR_MINOR_VERS_MISMATCH:
1625                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1626                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1627                 dprintk("%s: exit with error %d for server %s\n",
1628                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1629                 return -EPROTONOSUPPORT;
1630         case -EKEYEXPIRED:
1631                 nfs4_warn_keyexpired(clp->cl_hostname);
1632         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1633                                  * in nfs4_exchange_id */
1634         default:
1635                 dprintk("%s: exit with error %d for server %s\n", __func__,
1636                                 status, clp->cl_hostname);
1637                 return status;
1638         }
1639         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1640         dprintk("%s: handled error %d for server %s\n", __func__, status,
1641                         clp->cl_hostname);
1642         return 0;
1643 }
1644
1645 static int nfs4_reclaim_lease(struct nfs_client *clp)
1646 {
1647         struct rpc_cred *cred;
1648         const struct nfs4_state_recovery_ops *ops =
1649                 clp->cl_mvops->reboot_recovery_ops;
1650         int status;
1651
1652         cred = ops->get_clid_cred(clp);
1653         if (cred == NULL)
1654                 return -ENOENT;
1655         status = ops->establish_clid(clp, cred);
1656         put_rpccred(cred);
1657         if (status != 0)
1658                 return nfs4_handle_reclaim_lease_error(clp, status);
1659         return 0;
1660 }
1661
1662 #ifdef CONFIG_NFS_V4_1
1663 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1664 {
1665         struct nfs_client *clp = session->clp;
1666
1667         switch (err) {
1668         default:
1669                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1670                 break;
1671         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1672                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1673         }
1674         nfs4_schedule_lease_recovery(clp);
1675 }
1676 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1677
1678 void nfs41_handle_recall_slot(struct nfs_client *clp)
1679 {
1680         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1681         dprintk("%s: scheduling slot recall for server %s\n", __func__,
1682                         clp->cl_hostname);
1683         nfs4_schedule_state_manager(clp);
1684 }
1685
1686 static void nfs4_reset_all_state(struct nfs_client *clp)
1687 {
1688         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1689                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1690                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1691                 nfs4_state_start_reclaim_nograce(clp);
1692                 dprintk("%s: scheduling reset of all state for server %s!\n",
1693                                 __func__, clp->cl_hostname);
1694                 nfs4_schedule_state_manager(clp);
1695         }
1696 }
1697
1698 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1699 {
1700         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1701                 nfs4_state_start_reclaim_reboot(clp);
1702                 dprintk("%s: server %s rebooted!\n", __func__,
1703                                 clp->cl_hostname);
1704                 nfs4_schedule_state_manager(clp);
1705         }
1706 }
1707
1708 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1709 {
1710         nfs4_reset_all_state(clp);
1711         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1712 }
1713
1714 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1715 {
1716         /* This will need to handle layouts too */
1717         nfs_expire_all_delegations(clp);
1718         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1719                         clp->cl_hostname);
1720 }
1721
1722 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1723 {
1724         nfs_expire_all_delegations(clp);
1725         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1726                 nfs4_schedule_state_manager(clp);
1727         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1728                         clp->cl_hostname);
1729 }
1730
1731 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1732 {
1733         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1734                 &clp->cl_state) == 0)
1735                 nfs4_schedule_state_manager(clp);
1736 }
1737
1738 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1739 {
1740         if (!flags)
1741                 return;
1742
1743         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1744                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
1745
1746         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1747                 nfs41_handle_server_reboot(clp);
1748         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1749                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1750                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1751                             SEQ4_STATUS_LEASE_MOVED))
1752                 nfs41_handle_state_revoked(clp);
1753         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1754                 nfs41_handle_recallable_state_revoked(clp);
1755         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1756                 nfs41_handle_backchannel_fault(clp);
1757         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1758                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1759                 nfs41_handle_cb_path_down(clp);
1760 }
1761
1762 static int nfs4_reset_session(struct nfs_client *clp)
1763 {
1764         struct rpc_cred *cred;
1765         int status;
1766
1767         nfs4_begin_drain_session(clp);
1768         cred = nfs4_get_exchange_id_cred(clp);
1769         status = nfs4_proc_destroy_session(clp->cl_session, cred);
1770         if (status && status != -NFS4ERR_BADSESSION &&
1771             status != -NFS4ERR_DEADSESSION) {
1772                 status = nfs4_recovery_handle_error(clp, status);
1773                 goto out;
1774         }
1775
1776         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1777         status = nfs4_proc_create_session(clp, cred);
1778         if (status) {
1779                 dprintk("%s: session reset failed with status %d for server %s!\n",
1780                         __func__, status, clp->cl_hostname);
1781                 status = nfs4_handle_reclaim_lease_error(clp, status);
1782                 goto out;
1783         }
1784         nfs41_finish_session_reset(clp);
1785         dprintk("%s: session reset was successful for server %s!\n",
1786                         __func__, clp->cl_hostname);
1787 out:
1788         if (cred)
1789                 put_rpccred(cred);
1790         return status;
1791 }
1792
1793 static int nfs4_recall_slot(struct nfs_client *clp)
1794 {
1795         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1796         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1797         struct nfs4_slot *new, *old;
1798         int i;
1799
1800         nfs4_begin_drain_session(clp);
1801         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1802                       GFP_NOFS);
1803         if (!new)
1804                 return -ENOMEM;
1805
1806         spin_lock(&fc_tbl->slot_tbl_lock);
1807         for (i = 0; i < fc_tbl->target_max_slots; i++)
1808                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1809         old = fc_tbl->slots;
1810         fc_tbl->slots = new;
1811         fc_tbl->max_slots = fc_tbl->target_max_slots;
1812         fc_tbl->target_max_slots = 0;
1813         fc_attrs->max_reqs = fc_tbl->max_slots;
1814         spin_unlock(&fc_tbl->slot_tbl_lock);
1815
1816         kfree(old);
1817         nfs4_end_drain_session(clp);
1818         return 0;
1819 }
1820
1821 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1822 {
1823         struct rpc_cred *cred;
1824         int ret;
1825
1826         nfs4_begin_drain_session(clp);
1827         cred = nfs4_get_exchange_id_cred(clp);
1828         ret = nfs4_proc_bind_conn_to_session(clp, cred);
1829         if (cred)
1830                 put_rpccred(cred);
1831         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1832         switch (ret) {
1833         case 0:
1834                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
1835                         __func__, clp->cl_hostname);
1836                 break;
1837         case -NFS4ERR_DELAY:
1838                 ssleep(1);
1839                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1840                 break;
1841         default:
1842                 return nfs4_recovery_handle_error(clp, ret);
1843         }
1844         return 0;
1845 }
1846 #else /* CONFIG_NFS_V4_1 */
1847 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1848 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1849 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1850
1851 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1852 {
1853         return 0;
1854 }
1855 #endif /* CONFIG_NFS_V4_1 */
1856
1857 static void nfs4_state_manager(struct nfs_client *clp)
1858 {
1859         int status = 0;
1860         const char *section = "", *section_sep = "";
1861
1862         /* Ensure exclusive access to NFSv4 state */
1863         do {
1864                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
1865                         section = "purge state";
1866                         status = nfs4_reclaim_lease(clp);
1867                         if (status < 0)
1868                                 goto out_error;
1869                         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1870                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1871                 }
1872
1873                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1874                         section = "lease expired";
1875                         /* We're going to have to re-establish a clientid */
1876                         status = nfs4_reclaim_lease(clp);
1877                         if (status < 0)
1878                                 goto out_error;
1879                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1880                                 continue;
1881                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1882
1883                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1884                                                &clp->cl_state))
1885                                 nfs4_state_start_reclaim_nograce(clp);
1886                         else
1887                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1888                                         &clp->cl_state);
1889
1890                         pnfs_destroy_all_layouts(clp);
1891                 }
1892
1893                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1894                         section = "check lease";
1895                         status = nfs4_check_lease(clp);
1896                         if (status < 0)
1897                                 goto out_error;
1898                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1899                                 continue;
1900                 }
1901
1902                 /* Initialize or reset the session */
1903                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1904                    && nfs4_has_session(clp)) {
1905                         section = "reset session";
1906                         status = nfs4_reset_session(clp);
1907                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1908                                 continue;
1909                         if (status < 0)
1910                                 goto out_error;
1911                 }
1912
1913                 /* Send BIND_CONN_TO_SESSION */
1914                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1915                                 &clp->cl_state) && nfs4_has_session(clp)) {
1916                         section = "bind conn to session";
1917                         status = nfs4_bind_conn_to_session(clp);
1918                         if (status < 0)
1919                                 goto out_error;
1920                         continue;
1921                 }
1922
1923                 /* First recover reboot state... */
1924                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1925                         section = "reclaim reboot";
1926                         status = nfs4_do_reclaim(clp,
1927                                 clp->cl_mvops->reboot_recovery_ops);
1928                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1929                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1930                                 continue;
1931                         nfs4_state_end_reclaim_reboot(clp);
1932                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1933                                 continue;
1934                         if (status < 0)
1935                                 goto out_error;
1936                 }
1937
1938                 /* Now recover expired state... */
1939                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1940                         section = "reclaim nograce";
1941                         status = nfs4_do_reclaim(clp,
1942                                 clp->cl_mvops->nograce_recovery_ops);
1943                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1944                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1945                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1946                                 continue;
1947                         if (status < 0)
1948                                 goto out_error;
1949                 }
1950
1951                 nfs4_end_drain_session(clp);
1952                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1953                         nfs_client_return_marked_delegations(clp);
1954                         continue;
1955                 }
1956                 /* Recall session slots */
1957                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1958                    && nfs4_has_session(clp)) {
1959                         section = "recall slot";
1960                         status = nfs4_recall_slot(clp);
1961                         if (status < 0)
1962                                 goto out_error;
1963                         continue;
1964                 }
1965
1966
1967                 nfs4_clear_state_manager_bit(clp);
1968                 /* Did we race with an attempt to give us more work? */
1969                 if (clp->cl_state == 0)
1970                         break;
1971                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1972                         break;
1973         } while (atomic_read(&clp->cl_count) > 1);
1974         return;
1975 out_error:
1976         if (strlen(section))
1977                 section_sep = ": ";
1978         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
1979                         " with error %d\n", section_sep, section,
1980                         clp->cl_hostname, -status);
1981         nfs4_end_drain_session(clp);
1982         nfs4_clear_state_manager_bit(clp);
1983 }
1984
1985 static int nfs4_run_state_manager(void *ptr)
1986 {
1987         struct nfs_client *clp = ptr;
1988
1989         allow_signal(SIGKILL);
1990         nfs4_state_manager(clp);
1991         nfs_put_client(clp);
1992         module_put_and_exit(0);
1993         return 0;
1994 }
1995
1996 /*
1997  * Local variables:
1998  *  c-basic-offset: 8
1999  * End:
2000  */