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                 const struct nfs_lockowner *lockowner)
 915 {
 916         struct nfs4_lock_state *lsp;
 917         fl_owner_t fl_owner;
 918         pid_t fl_pid;
 919         bool ret = false;
 920
 921
 922         if (lockowner == NULL)
 923                 goto out;
 924
 925         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 926                 goto out;
 927
 928         fl_owner = lockowner->l_owner;
 929         fl_pid = lockowner->l_pid;
 930         spin_lock(&state->state_lock);
 931         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 932         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
 933                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 934                 ret = true;
 935         }
 936         spin_unlock(&state->state_lock);
 937         nfs4_put_lock_state(lsp);
 938 out:
 939         return ret;
 940 }
 941
 942 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 943 {
 944         int seq;
 945
 946         do {
 947                 seq = read_seqbegin(&state->seqlock);
 948                 nfs4_stateid_copy(dst, &state->stateid);
 949         } while (read_seqretry(&state->seqlock, seq));
 950 }
 951
 952 /*
 953  * Byte-range lock aware utility to initialize the stateid of read/write
 954  * requests.
 955  */
 956 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 957                 fmode_t fmode, const struct nfs_lockowner *lockowner)
 958 {
 959         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
 960                 return;
 961         if (nfs4_copy_lock_stateid(dst, state, lockowner))
 962                 return;
 963         nfs4_copy_open_stateid(dst, state);
 964 }
 965
 966 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 967 {
 968         struct nfs_seqid *new;
 969
 970         new = kmalloc(sizeof(*new), gfp_mask);
 971         if (new != NULL) {
 972                 new->sequence = counter;
 973                 INIT_LIST_HEAD(&new->list);
 974                 new->task = NULL;
 975         }
 976         return new;
 977 }
 978
 979 void nfs_release_seqid(struct nfs_seqid *seqid)
 980 {
 981         struct nfs_seqid_counter *sequence;
 982
 983         if (list_empty(&seqid->list))
 984                 return;
 985         sequence = seqid->sequence;
 986         spin_lock(&sequence->lock);
 987         list_del_init(&seqid->list);
 988         if (!list_empty(&sequence->list)) {
 989                 struct nfs_seqid *next;
 990
 991                 next = list_first_entry(&sequence->list,
 992                                 struct nfs_seqid, list);
 993                 rpc_wake_up_queued_task(&sequence->wait, next->task);
 994         }
 995         spin_unlock(&sequence->lock);
 996 }
 997
 998 void nfs_free_seqid(struct nfs_seqid *seqid)
 999 {
1000         nfs_release_seqid(seqid);
1001         kfree(seqid);
1002 }
1003
1004 /*
1005  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1006  * failed with a seqid incrementing error -
1007  * see comments nfs_fs.h:seqid_mutating_error()
1008  */
1009 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1010 {
1011         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
1012         switch (status) {
1013                 case 0:
1014                         break;
1015                 case -NFS4ERR_BAD_SEQID:
1016                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1017                                 return;
1018                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1019                                         " sequence-id error on an"
1020                                         " unconfirmed sequence %p!\n",
1021                                         seqid->sequence);
1022                 case -NFS4ERR_STALE_CLIENTID:
1023                 case -NFS4ERR_STALE_STATEID:
1024                 case -NFS4ERR_BAD_STATEID:
1025                 case -NFS4ERR_BADXDR:
1026                 case -NFS4ERR_RESOURCE:
1027                 case -NFS4ERR_NOFILEHANDLE:
1028                         /* Non-seqid mutating errors */
1029                         return;
1030         };
1031         /*
1032          * Note: no locking needed as we are guaranteed to be first
1033          * on the sequence list
1034          */
1035         seqid->sequence->counter++;
1036 }
1037
1038 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1039 {
1040         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1041                                         struct nfs4_state_owner, so_seqid);
1042         struct nfs_server *server = sp->so_server;
1043
1044         if (status == -NFS4ERR_BAD_SEQID)
1045                 nfs4_drop_state_owner(sp);
1046         if (!nfs4_has_session(server->nfs_client))
1047                 nfs_increment_seqid(status, seqid);
1048 }
1049
1050 /*
1051  * Increment the seqid if the LOCK/LOCKU succeeded, or
1052  * failed with a seqid incrementing error -
1053  * see comments nfs_fs.h:seqid_mutating_error()
1054  */
1055 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1056 {
1057         nfs_increment_seqid(status, seqid);
1058 }
1059
1060 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1061 {
1062         struct nfs_seqid_counter *sequence = seqid->sequence;
1063         int status = 0;
1064
1065         spin_lock(&sequence->lock);
1066         seqid->task = task;
1067         if (list_empty(&seqid->list))
1068                 list_add_tail(&seqid->list, &sequence->list);
1069         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1070                 goto unlock;
1071         rpc_sleep_on(&sequence->wait, task, NULL);
1072         status = -EAGAIN;
1073 unlock:
1074         spin_unlock(&sequence->lock);
1075         return status;
1076 }
1077
1078 static int nfs4_run_state_manager(void *);
1079
1080 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1081 {
1082         smp_mb__before_clear_bit();
1083         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1084         smp_mb__after_clear_bit();
1085         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1086         rpc_wake_up(&clp->cl_rpcwaitq);
1087 }
1088
1089 /*
1090  * Schedule the nfs_client asynchronous state management routine
1091  */
1092 void nfs4_schedule_state_manager(struct nfs_client *clp)
1093 {
1094         struct task_struct *task;
1095         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1096
1097         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1098                 return;
1099         __module_get(THIS_MODULE);
1100         atomic_inc(&clp->cl_count);
1101
1102         /* The rcu_read_lock() is not strictly necessary, as the state
1103          * manager is the only thread that ever changes the rpc_xprt
1104          * after it's initialized.  At this point, we're single threaded. */
1105         rcu_read_lock();
1106         snprintf(buf, sizeof(buf), "%s-manager",
1107                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1108         rcu_read_unlock();
1109         task = kthread_run(nfs4_run_state_manager, clp, buf);
1110         if (IS_ERR(task)) {
1111                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1112                         __func__, PTR_ERR(task));
1113                 nfs4_clear_state_manager_bit(clp);
1114                 nfs_put_client(clp);
1115                 module_put(THIS_MODULE);
1116         }
1117 }
1118
1119 /*
1120  * Schedule a lease recovery attempt
1121  */
1122 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1123 {
1124         if (!clp)
1125                 return;
1126         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1127                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1128         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1129                         clp->cl_hostname);
1130         nfs4_schedule_state_manager(clp);
1131 }
1132 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1133
1134 /*
1135  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1136  * @clp: client to process
1137  *
1138  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1139  * resend of the SETCLIENTID and hence re-establish the
1140  * callback channel. Then return all existing delegations.
1141  */
1142 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1143 {
1144         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1145         nfs_expire_all_delegations(clp);
1146         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1147                         clp->cl_hostname);
1148 }
1149
1150 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1151 {
1152         nfs40_handle_cb_pathdown(clp);
1153         nfs4_schedule_state_manager(clp);
1154 }
1155
1156 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1157 {
1158
1159         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1160         /* Don't recover state that expired before the reboot */
1161         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1162                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1163                 return 0;
1164         }
1165         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1166         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1167         return 1;
1168 }
1169
1170 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1171 {
1172         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1173         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1174         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1175         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1176         return 1;
1177 }
1178
1179 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1180 {
1181         struct nfs_client *clp = server->nfs_client;
1182
1183         nfs4_state_mark_reclaim_nograce(clp, state);
1184         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1185                         clp->cl_hostname);
1186         nfs4_schedule_state_manager(clp);
1187 }
1188 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1189
1190 void nfs_inode_find_state_and_recover(struct inode *inode,
1191                 const nfs4_stateid *stateid)
1192 {
1193         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1194         struct nfs_inode *nfsi = NFS_I(inode);
1195         struct nfs_open_context *ctx;
1196         struct nfs4_state *state;
1197         bool found = false;
1198
1199         spin_lock(&inode->i_lock);
1200         list_for_each_entry(ctx, &nfsi->open_files, list) {
1201                 state = ctx->state;
1202                 if (state == NULL)
1203                         continue;
1204                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1205                         continue;
1206                 if (!nfs4_stateid_match(&state->stateid, stateid))
1207                         continue;
1208                 nfs4_state_mark_reclaim_nograce(clp, state);
1209                 found = true;
1210         }
1211         spin_unlock(&inode->i_lock);
1212         if (found)
1213                 nfs4_schedule_state_manager(clp);
1214 }
1215
1216
1217 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1218 {
1219         struct inode *inode = state->inode;
1220         struct nfs_inode *nfsi = NFS_I(inode);
1221         struct file_lock *fl;
1222         int status = 0;
1223
1224         if (inode->i_flock == NULL)
1225                 return 0;
1226
1227         /* Guard against delegation returns and new lock/unlock calls */
1228         down_write(&nfsi->rwsem);
1229         /* Protect inode->i_flock using the BKL */
1230         lock_flocks();
1231         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1232                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1233                         continue;
1234                 if (nfs_file_open_context(fl->fl_file)->state != state)
1235                         continue;
1236                 unlock_flocks();
1237                 status = ops->recover_lock(state, fl);
1238                 switch (status) {
1239                         case 0:
1240                                 break;
1241                         case -ESTALE:
1242                         case -NFS4ERR_ADMIN_REVOKED:
1243                         case -NFS4ERR_STALE_STATEID:
1244                         case -NFS4ERR_BAD_STATEID:
1245                         case -NFS4ERR_EXPIRED:
1246                         case -NFS4ERR_NO_GRACE:
1247                         case -NFS4ERR_STALE_CLIENTID:
1248                         case -NFS4ERR_BADSESSION:
1249                         case -NFS4ERR_BADSLOT:
1250                         case -NFS4ERR_BAD_HIGH_SLOT:
1251                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1252                                 goto out;
1253                         default:
1254                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1255                                         "Zeroing state\n", __func__, status);
1256                         case -ENOMEM:
1257                         case -NFS4ERR_DENIED:
1258                         case -NFS4ERR_RECLAIM_BAD:
1259                         case -NFS4ERR_RECLAIM_CONFLICT:
1260                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1261                                 status = 0;
1262                 }
1263                 lock_flocks();
1264         }
1265         unlock_flocks();
1266 out:
1267         up_write(&nfsi->rwsem);
1268         return status;
1269 }
1270
1271 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1272 {
1273         struct nfs4_state *state;
1274         struct nfs4_lock_state *lock;
1275         int status = 0;
1276
1277         /* Note: we rely on the sp->so_states list being ordered
1278          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1279          * states first.
1280          * This is needed to ensure that the server won't give us any
1281          * read delegations that we have to return if, say, we are
1282          * recovering after a network partition or a reboot from a
1283          * server that doesn't support a grace period.
1284          */
1285 restart:
1286         spin_lock(&sp->so_lock);
1287         list_for_each_entry(state, &sp->so_states, open_states) {
1288                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1289                         continue;
1290                 if (state->state == 0)
1291                         continue;
1292                 atomic_inc(&state->count);
1293                 spin_unlock(&sp->so_lock);
1294                 status = ops->recover_open(sp, state);
1295                 if (status >= 0) {
1296                         status = nfs4_reclaim_locks(state, ops);
1297                         if (status >= 0) {
1298                                 spin_lock(&state->state_lock);
1299                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1300                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1301                                                 pr_warn_ratelimited("NFS: "
1302                                                         "%s: Lock reclaim "
1303                                                         "failed!\n", __func__);
1304                                 }
1305                                 spin_unlock(&state->state_lock);
1306                                 nfs4_put_open_state(state);
1307                                 goto restart;
1308                         }
1309                 }
1310                 switch (status) {
1311                         default:
1312                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1313                                         "Zeroing state\n", __func__, status);
1314                         case -ENOENT:
1315                         case -ENOMEM:
1316                         case -ESTALE:
1317                                 /*
1318                                  * Open state on this file cannot be recovered
1319                                  * All we can do is revert to using the zero stateid.
1320                                  */
1321                                 memset(&state->stateid, 0,
1322                                         sizeof(state->stateid));
1323                                 /* Mark the file as being 'closed' */
1324                                 state->state = 0;
1325                                 break;
1326                         case -EKEYEXPIRED:
1327                                 /*
1328                                  * User RPCSEC_GSS context has expired.
1329                                  * We cannot recover this stateid now, so
1330                                  * skip it and allow recovery thread to
1331                                  * proceed.
1332                                  */
1333                                 break;
1334                         case -NFS4ERR_ADMIN_REVOKED:
1335                         case -NFS4ERR_STALE_STATEID:
1336                         case -NFS4ERR_BAD_STATEID:
1337                         case -NFS4ERR_RECLAIM_BAD:
1338                         case -NFS4ERR_RECLAIM_CONFLICT:
1339                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1340                                 break;
1341                         case -NFS4ERR_EXPIRED:
1342                         case -NFS4ERR_NO_GRACE:
1343                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1344                         case -NFS4ERR_STALE_CLIENTID:
1345                         case -NFS4ERR_BADSESSION:
1346                         case -NFS4ERR_BADSLOT:
1347                         case -NFS4ERR_BAD_HIGH_SLOT:
1348                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1349                                 goto out_err;
1350                 }
1351                 nfs4_put_open_state(state);
1352                 goto restart;
1353         }
1354         spin_unlock(&sp->so_lock);
1355         return 0;
1356 out_err:
1357         nfs4_put_open_state(state);
1358         return status;
1359 }
1360
1361 static void nfs4_clear_open_state(struct nfs4_state *state)
1362 {
1363         struct nfs4_lock_state *lock;
1364
1365         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1366         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1367         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1368         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1369         spin_lock(&state->state_lock);
1370         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1371                 lock->ls_seqid.flags = 0;
1372                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1373         }
1374         spin_unlock(&state->state_lock);
1375 }
1376
1377 static void nfs4_reset_seqids(struct nfs_server *server,
1378         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1379 {
1380         struct nfs_client *clp = server->nfs_client;
1381         struct nfs4_state_owner *sp;
1382         struct rb_node *pos;
1383         struct nfs4_state *state;
1384
1385         spin_lock(&clp->cl_lock);
1386         for (pos = rb_first(&server->state_owners);
1387              pos != NULL;
1388              pos = rb_next(pos)) {
1389                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1390                 sp->so_seqid.flags = 0;
1391                 spin_lock(&sp->so_lock);
1392                 list_for_each_entry(state, &sp->so_states, open_states) {
1393                         if (mark_reclaim(clp, state))
1394                                 nfs4_clear_open_state(state);
1395                 }
1396                 spin_unlock(&sp->so_lock);
1397         }
1398         spin_unlock(&clp->cl_lock);
1399 }
1400
1401 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1402         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1403 {
1404         struct nfs_server *server;
1405
1406         rcu_read_lock();
1407         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1408                 nfs4_reset_seqids(server, mark_reclaim);
1409         rcu_read_unlock();
1410 }
1411
1412 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1413 {
1414         /* Mark all delegations for reclaim */
1415         nfs_delegation_mark_reclaim(clp);
1416         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1417 }
1418
1419 static void nfs4_reclaim_complete(struct nfs_client *clp,
1420                                  const struct nfs4_state_recovery_ops *ops)
1421 {
1422         /* Notify the server we're done reclaiming our state */
1423         if (ops->reclaim_complete)
1424                 (void)ops->reclaim_complete(clp);
1425 }
1426
1427 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1428 {
1429         struct nfs_client *clp = server->nfs_client;
1430         struct nfs4_state_owner *sp;
1431         struct rb_node *pos;
1432         struct nfs4_state *state;
1433
1434         spin_lock(&clp->cl_lock);
1435         for (pos = rb_first(&server->state_owners);
1436              pos != NULL;
1437              pos = rb_next(pos)) {
1438                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1439                 spin_lock(&sp->so_lock);
1440                 list_for_each_entry(state, &sp->so_states, open_states) {
1441                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1442                                                 &state->flags))
1443                                 continue;
1444                         nfs4_state_mark_reclaim_nograce(clp, state);
1445                 }
1446                 spin_unlock(&sp->so_lock);
1447         }
1448         spin_unlock(&clp->cl_lock);
1449 }
1450
1451 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1452 {
1453         struct nfs_server *server;
1454
1455         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1456                 return 0;
1457
1458         rcu_read_lock();
1459         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1460                 nfs4_clear_reclaim_server(server);
1461         rcu_read_unlock();
1462
1463         nfs_delegation_reap_unclaimed(clp);
1464         return 1;
1465 }
1466
1467 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1468 {
1469         if (!nfs4_state_clear_reclaim_reboot(clp))
1470                 return;
1471         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1472 }
1473
1474 static void nfs_delegation_clear_all(struct nfs_client *clp)
1475 {
1476         nfs_delegation_mark_reclaim(clp);
1477         nfs_delegation_reap_unclaimed(clp);
1478 }
1479
1480 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1481 {
1482         nfs_delegation_clear_all(clp);
1483         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1484 }
1485
1486 static void nfs4_warn_keyexpired(const char *s)
1487 {
1488         printk_ratelimited(KERN_WARNING "Error: state manager"
1489                         " encountered RPCSEC_GSS session"
1490                         " expired against NFSv4 server %s.\n",
1491                         s);
1492 }
1493
1494 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1495 {
1496         switch (error) {
1497                 case 0:
1498                         break;
1499                 case -NFS4ERR_CB_PATH_DOWN:
1500                         nfs40_handle_cb_pathdown(clp);
1501                         break;
1502                 case -NFS4ERR_NO_GRACE:
1503                         nfs4_state_end_reclaim_reboot(clp);
1504                         break;
1505                 case -NFS4ERR_STALE_CLIENTID:
1506                 case -NFS4ERR_LEASE_MOVED:
1507                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1508                         nfs4_state_clear_reclaim_reboot(clp);
1509                         nfs4_state_start_reclaim_reboot(clp);
1510                         break;
1511                 case -NFS4ERR_EXPIRED:
1512                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1513                         nfs4_state_start_reclaim_nograce(clp);
1514                         break;
1515                 case -NFS4ERR_BADSESSION:
1516                 case -NFS4ERR_BADSLOT:
1517                 case -NFS4ERR_BAD_HIGH_SLOT:
1518                 case -NFS4ERR_DEADSESSION:
1519                 case -NFS4ERR_SEQ_FALSE_RETRY:
1520                 case -NFS4ERR_SEQ_MISORDERED:
1521                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1522                         /* Zero session reset errors */
1523                         break;
1524                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1525                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1526                         break;
1527                 case -EKEYEXPIRED:
1528                         /* Nothing we can do */
1529                         nfs4_warn_keyexpired(clp->cl_hostname);
1530                         break;
1531                 default:
1532                         dprintk("%s: failed to handle error %d for server %s\n",
1533                                         __func__, error, clp->cl_hostname);
1534                         return error;
1535         }
1536         dprintk("%s: handled error %d for server %s\n", __func__, error,
1537                         clp->cl_hostname);
1538         return 0;
1539 }
1540
1541 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1542 {
1543         struct nfs4_state_owner *sp;
1544         struct nfs_server *server;
1545         struct rb_node *pos;
1546         int status = 0;
1547
1548 restart:
1549         rcu_read_lock();
1550         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1551                 nfs4_purge_state_owners(server);
1552                 spin_lock(&clp->cl_lock);
1553                 for (pos = rb_first(&server->state_owners);
1554                      pos != NULL;
1555                      pos = rb_next(pos)) {
1556                         sp = rb_entry(pos,
1557                                 struct nfs4_state_owner, so_server_node);
1558                         if (!test_and_clear_bit(ops->owner_flag_bit,
1559                                                         &sp->so_flags))
1560                                 continue;
1561                         atomic_inc(&sp->so_count);
1562                         spin_unlock(&clp->cl_lock);
1563                         rcu_read_unlock();
1564
1565                         status = nfs4_reclaim_open_state(sp, ops);
1566                         if (status < 0) {
1567                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1568                                 nfs4_put_state_owner(sp);
1569                                 return nfs4_recovery_handle_error(clp, status);
1570                         }
1571
1572                         nfs4_put_state_owner(sp);
1573                         goto restart;
1574                 }
1575                 spin_unlock(&clp->cl_lock);
1576         }
1577         rcu_read_unlock();
1578         return status;
1579 }
1580
1581 static int nfs4_check_lease(struct nfs_client *clp)
1582 {
1583         struct rpc_cred *cred;
1584         const struct nfs4_state_maintenance_ops *ops =
1585                 clp->cl_mvops->state_renewal_ops;
1586         int status;
1587
1588         /* Is the client already known to have an expired lease? */
1589         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1590                 return 0;
1591         spin_lock(&clp->cl_lock);
1592         cred = ops->get_state_renewal_cred_locked(clp);
1593         spin_unlock(&clp->cl_lock);
1594         if (cred == NULL) {
1595                 cred = nfs4_get_setclientid_cred(clp);
1596                 status = -ENOKEY;
1597                 if (cred == NULL)
1598                         goto out;
1599         }
1600         status = ops->renew_lease(clp, cred);
1601         put_rpccred(cred);
1602 out:
1603         return nfs4_recovery_handle_error(clp, status);
1604 }
1605
1606 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1607  * on EXCHANGE_ID for v4.1
1608  */
1609 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1610 {
1611         switch (status) {
1612         case -NFS4ERR_SEQ_MISORDERED:
1613                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1614                         return -ESERVERFAULT;
1615                 /* Lease confirmation error: retry after purging the lease */
1616                 ssleep(1);
1617         case -NFS4ERR_STALE_CLIENTID:
1618                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1619                 break;
1620         case -NFS4ERR_CLID_INUSE:
1621                 pr_err("NFS: Server %s reports our clientid is in use\n",
1622                         clp->cl_hostname);
1623                 nfs_mark_client_ready(clp, -EPERM);
1624                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1625                 return -EPERM;
1626         case -EACCES:
1627                 if (clp->cl_machine_cred == NULL)
1628                         return -EACCES;
1629                 /* Handle case where the user hasn't set up machine creds */
1630                 nfs4_clear_machine_cred(clp);
1631         case -NFS4ERR_DELAY:
1632         case -ETIMEDOUT:
1633         case -EAGAIN:
1634                 ssleep(1);
1635                 break;
1636
1637         case -NFS4ERR_MINOR_VERS_MISMATCH:
1638                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1639                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1640                 dprintk("%s: exit with error %d for server %s\n",
1641                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1642                 return -EPROTONOSUPPORT;
1643         case -EKEYEXPIRED:
1644                 nfs4_warn_keyexpired(clp->cl_hostname);
1645         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1646                                  * in nfs4_exchange_id */
1647         default:
1648                 dprintk("%s: exit with error %d for server %s\n", __func__,
1649                                 status, clp->cl_hostname);
1650                 return status;
1651         }
1652         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1653         dprintk("%s: handled error %d for server %s\n", __func__, status,
1654                         clp->cl_hostname);
1655         return 0;
1656 }
1657
1658 static int nfs4_establish_lease(struct nfs_client *clp)
1659 {
1660         struct rpc_cred *cred;
1661         const struct nfs4_state_recovery_ops *ops =
1662                 clp->cl_mvops->reboot_recovery_ops;
1663         int status;
1664
1665         cred = ops->get_clid_cred(clp);
1666         if (cred == NULL)
1667                 return -ENOENT;
1668         status = ops->establish_clid(clp, cred);
1669         put_rpccred(cred);
1670         if (status != 0)
1671                 return status;
1672         pnfs_destroy_all_layouts(clp);
1673         return 0;
1674 }
1675
1676 /*
1677  * Returns zero or a negative errno.  NFS4ERR values are converted
1678  * to local errno values.
1679  */
1680 static int nfs4_reclaim_lease(struct nfs_client *clp)
1681 {
1682         int status;
1683
1684         status = nfs4_establish_lease(clp);
1685         if (status < 0)
1686                 return nfs4_handle_reclaim_lease_error(clp, status);
1687         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1688                 nfs4_state_start_reclaim_nograce(clp);
1689         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1690                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1691         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1692         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1693         return 0;
1694 }
1695
1696 static int nfs4_purge_lease(struct nfs_client *clp)
1697 {
1698         int status;
1699
1700         status = nfs4_establish_lease(clp);
1701         if (status < 0)
1702                 return nfs4_handle_reclaim_lease_error(clp, status);
1703         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1704         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1705         nfs4_state_start_reclaim_nograce(clp);
1706         return 0;
1707 }
1708
1709 #ifdef CONFIG_NFS_V4_1
1710 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1711 {
1712         struct nfs_client *clp = session->clp;
1713
1714         switch (err) {
1715         default:
1716                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1717                 break;
1718         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1719                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1720         }
1721         nfs4_schedule_lease_recovery(clp);
1722 }
1723 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1724
1725 void nfs41_handle_recall_slot(struct nfs_client *clp)
1726 {
1727         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1728         dprintk("%s: scheduling slot recall for server %s\n", __func__,
1729                         clp->cl_hostname);
1730         nfs4_schedule_state_manager(clp);
1731 }
1732
1733 static void nfs4_reset_all_state(struct nfs_client *clp)
1734 {
1735         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1736                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1737                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1738                 nfs4_state_start_reclaim_nograce(clp);
1739                 dprintk("%s: scheduling reset of all state for server %s!\n",
1740                                 __func__, clp->cl_hostname);
1741                 nfs4_schedule_state_manager(clp);
1742         }
1743 }
1744
1745 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1746 {
1747         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1748                 nfs4_state_start_reclaim_reboot(clp);
1749                 dprintk("%s: server %s rebooted!\n", __func__,
1750                                 clp->cl_hostname);
1751                 nfs4_schedule_state_manager(clp);
1752         }
1753 }
1754
1755 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1756 {
1757         nfs4_reset_all_state(clp);
1758         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1759 }
1760
1761 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1762 {
1763         /* This will need to handle layouts too */
1764         nfs_expire_all_delegations(clp);
1765         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1766                         clp->cl_hostname);
1767 }
1768
1769 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1770 {
1771         nfs_expire_all_delegations(clp);
1772         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1773                 nfs4_schedule_state_manager(clp);
1774         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1775                         clp->cl_hostname);
1776 }
1777
1778 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1779 {
1780         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1781                 &clp->cl_state) == 0)
1782                 nfs4_schedule_state_manager(clp);
1783 }
1784
1785 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1786 {
1787         if (!flags)
1788                 return;
1789
1790         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1791                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
1792
1793         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1794                 nfs41_handle_server_reboot(clp);
1795         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1796                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1797                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1798                             SEQ4_STATUS_LEASE_MOVED))
1799                 nfs41_handle_state_revoked(clp);
1800         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1801                 nfs41_handle_recallable_state_revoked(clp);
1802         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1803                 nfs41_handle_backchannel_fault(clp);
1804         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1805                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1806                 nfs41_handle_cb_path_down(clp);
1807 }
1808
1809 static int nfs4_reset_session(struct nfs_client *clp)
1810 {
1811         struct rpc_cred *cred;
1812         int status;
1813
1814         if (!nfs4_has_session(clp))
1815                 return 0;
1816         nfs4_begin_drain_session(clp);
1817         cred = nfs4_get_exchange_id_cred(clp);
1818         status = nfs4_proc_destroy_session(clp->cl_session, cred);
1819         if (status && status != -NFS4ERR_BADSESSION &&
1820             status != -NFS4ERR_DEADSESSION) {
1821                 status = nfs4_recovery_handle_error(clp, status);
1822                 goto out;
1823         }
1824
1825         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1826         status = nfs4_proc_create_session(clp, cred);
1827         if (status) {
1828                 dprintk("%s: session reset failed with status %d for server %s!\n",
1829                         __func__, status, clp->cl_hostname);
1830                 status = nfs4_handle_reclaim_lease_error(clp, status);
1831                 goto out;
1832         }
1833         nfs41_finish_session_reset(clp);
1834         dprintk("%s: session reset was successful for server %s!\n",
1835                         __func__, clp->cl_hostname);
1836 out:
1837         if (cred)
1838                 put_rpccred(cred);
1839         return status;
1840 }
1841
1842 static int nfs4_recall_slot(struct nfs_client *clp)
1843 {
1844         struct nfs4_slot_table *fc_tbl;
1845         struct nfs4_slot *new, *old;
1846         int i;
1847
1848         if (!nfs4_has_session(clp))
1849                 return 0;
1850         nfs4_begin_drain_session(clp);
1851         fc_tbl = &clp->cl_session->fc_slot_table;
1852         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1853                       GFP_NOFS);
1854         if (!new)
1855                 return -ENOMEM;
1856
1857         spin_lock(&fc_tbl->slot_tbl_lock);
1858         for (i = 0; i < fc_tbl->target_max_slots; i++)
1859                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1860         old = fc_tbl->slots;
1861         fc_tbl->slots = new;
1862         fc_tbl->max_slots = fc_tbl->target_max_slots;
1863         fc_tbl->target_max_slots = 0;
1864         clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
1865         spin_unlock(&fc_tbl->slot_tbl_lock);
1866
1867         kfree(old);
1868         return 0;
1869 }
1870
1871 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1872 {
1873         struct rpc_cred *cred;
1874         int ret;
1875
1876         if (!nfs4_has_session(clp))
1877                 return 0;
1878         nfs4_begin_drain_session(clp);
1879         cred = nfs4_get_exchange_id_cred(clp);
1880         ret = nfs4_proc_bind_conn_to_session(clp, cred);
1881         if (cred)
1882                 put_rpccred(cred);
1883         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1884         switch (ret) {
1885         case 0:
1886                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
1887                         __func__, clp->cl_hostname);
1888                 break;
1889         case -NFS4ERR_DELAY:
1890                 ssleep(1);
1891                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1892                 break;
1893         default:
1894                 return nfs4_recovery_handle_error(clp, ret);
1895         }
1896         return 0;
1897 }
1898 #else /* CONFIG_NFS_V4_1 */
1899 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1900 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1901 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1902
1903 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
1904 {
1905         return 0;
1906 }
1907 #endif /* CONFIG_NFS_V4_1 */
1908
1909 static void nfs4_state_manager(struct nfs_client *clp)
1910 {
1911         int status = 0;
1912         const char *section = "", *section_sep = "";
1913
1914         /* Ensure exclusive access to NFSv4 state */
1915         do {
1916                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
1917                         section = "purge state";
1918                         status = nfs4_purge_lease(clp);
1919                         if (status < 0)
1920                                 goto out_error;
1921                         continue;
1922                 }
1923
1924                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1925                         section = "lease expired";
1926                         /* We're going to have to re-establish a clientid */
1927                         status = nfs4_reclaim_lease(clp);
1928                         if (status < 0)
1929                                 goto out_error;
1930                         continue;
1931                 }
1932
1933                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1934                         section = "check lease";
1935                         status = nfs4_check_lease(clp);
1936                         if (status < 0)
1937                                 goto out_error;
1938                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1939                                 continue;
1940                 }
1941
1942                 /* Initialize or reset the session */
1943                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
1944                         section = "reset session";
1945                         status = nfs4_reset_session(clp);
1946                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1947                                 continue;
1948                         if (status < 0)
1949                                 goto out_error;
1950                 }
1951
1952                 /* Send BIND_CONN_TO_SESSION */
1953                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1954                                 &clp->cl_state)) {
1955                         section = "bind conn to session";
1956                         status = nfs4_bind_conn_to_session(clp);
1957                         if (status < 0)
1958                                 goto out_error;
1959                         continue;
1960                 }
1961
1962                 /* Recall session slots */
1963                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)) {
1964                         section = "recall slot";
1965                         status = nfs4_recall_slot(clp);
1966                         if (status < 0)
1967                                 goto out_error;
1968                         continue;
1969                 }
1970
1971                 /* First recover reboot state... */
1972                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1973                         section = "reclaim reboot";
1974                         status = nfs4_do_reclaim(clp,
1975                                 clp->cl_mvops->reboot_recovery_ops);
1976                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1977                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1978                                 continue;
1979                         nfs4_state_end_reclaim_reboot(clp);
1980                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1981                                 continue;
1982                         if (status < 0)
1983                                 goto out_error;
1984                 }
1985
1986                 /* Now recover expired state... */
1987                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1988                         section = "reclaim nograce";
1989                         status = nfs4_do_reclaim(clp,
1990                                 clp->cl_mvops->nograce_recovery_ops);
1991                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1992                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1993                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1994                                 continue;
1995                         if (status < 0)
1996                                 goto out_error;
1997                 }
1998
1999                 nfs4_end_drain_session(clp);
2000                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2001                         nfs_client_return_marked_delegations(clp);
2002                         continue;
2003                 }
2004
2005                 nfs4_clear_state_manager_bit(clp);
2006                 /* Did we race with an attempt to give us more work? */
2007                 if (clp->cl_state == 0)
2008                         break;
2009                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2010                         break;
2011         } while (atomic_read(&clp->cl_count) > 1);
2012         return;
2013 out_error:
2014         if (strlen(section))
2015                 section_sep = ": ";
2016         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2017                         " with error %d\n", section_sep, section,
2018                         clp->cl_hostname, -status);
2019         nfs4_end_drain_session(clp);
2020         nfs4_clear_state_manager_bit(clp);
2021 }
2022
2023 static int nfs4_run_state_manager(void *ptr)
2024 {
2025         struct nfs_client *clp = ptr;
2026
2027         allow_signal(SIGKILL);
2028         nfs4_state_manager(clp);
2029         nfs_put_client(clp);
2030         module_put_and_exit(0);
2031         return 0;
2032 }
2033
2034 /*
2035  * Local variables:
2036  *  c-basic-offset: 8
2037  * End:
2038  */