fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52 #include <linux/jiffies.h>
  53
  54 #include <linux/sunrpc/clnt.h>
  55
  56 #include "nfs4_fs.h"
  57 #include "callback.h"
  58 #include "delegation.h"
  59 #include "internal.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid;
  69 static DEFINE_MUTEX(nfs_clid_init_mutex);
  70
  71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  72 {
  73         struct nfs4_setclientid_res clid = {
  74                 .clientid = clp->cl_clientid,
  75                 .confirm = clp->cl_confirm,
  76         };
  77         unsigned short port;
  78         int status;
  79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  80
  81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  82                 goto do_confirm;
  83         port = nn->nfs_callback_tcpport;
  84         if (clp->cl_addr.ss_family == AF_INET6)
  85                 port = nn->nfs_callback_tcpport6;
  86
  87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  88         if (status != 0)
  89                 goto out;
  90         clp->cl_clientid = clid.clientid;
  91         clp->cl_confirm = clid.confirm;
  92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  93 do_confirm:
  94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  95         if (status != 0)
  96                 goto out;
  97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98         nfs4_schedule_state_renewal(clp);
  99 out:
 100         return status;
 101 }
 102
 103 /**
 104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 105  *
 106  * @clp: nfs_client under test
 107  * @result: OUT: found nfs_client, or clp
 108  * @cred: credential to use for trunking test
 109  *
 110  * Returns zero, a negative errno, or a negative NFS4ERR status.
 111  * If zero is returned, an nfs_client pointer is planted in
 112  * "result".
 113  *
 114  * Note: The returned client may not yet be marked ready.
 115  */
 116 int nfs40_discover_server_trunking(struct nfs_client *clp,
 117                                    struct nfs_client **result,
 118                                    struct rpc_cred *cred)
 119 {
 120         struct nfs4_setclientid_res clid = {
 121                 .clientid = clp->cl_clientid,
 122                 .confirm = clp->cl_confirm,
 123         };
 124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 125         unsigned short port;
 126         int status;
 127
 128         port = nn->nfs_callback_tcpport;
 129         if (clp->cl_addr.ss_family == AF_INET6)
 130                 port = nn->nfs_callback_tcpport6;
 131
 132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 133         if (status != 0)
 134                 goto out;
 135         clp->cl_clientid = clid.clientid;
 136         clp->cl_confirm = clid.confirm;
 137
 138         status = nfs40_walk_client_list(clp, result, cred);
 139         if (status == 0) {
 140                 /* Sustain the lease, even if it's empty.  If the clientid4
 141                  * goes stale it's of no use for trunking discovery. */
 142                 nfs4_schedule_state_renewal(*result);
 143         }
 144 out:
 145         return status;
 146 }
 147
 148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 149 {
 150         struct rpc_cred *cred = NULL;
 151
 152         if (clp->cl_machine_cred != NULL)
 153                 cred = get_rpccred(clp->cl_machine_cred);
 154         return cred;
 155 }
 156
 157 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 158 {
 159         struct rpc_cred *cred;
 160
 161         spin_lock(&clp->cl_lock);
 162         cred = clp->cl_machine_cred;
 163         clp->cl_machine_cred = NULL;
 164         spin_unlock(&clp->cl_lock);
 165         if (cred != NULL)
 166                 put_rpccred(cred);
 167 }
 168
 169 static struct rpc_cred *
 170 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 171 {
 172         struct rpc_cred *cred = NULL;
 173         struct nfs4_state_owner *sp;
 174         struct rb_node *pos;
 175
 176         for (pos = rb_first(&server->state_owners);
 177              pos != NULL;
 178              pos = rb_next(pos)) {
 179                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 180                 if (list_empty(&sp->so_states))
 181                         continue;
 182                 cred = get_rpccred(sp->so_cred);
 183                 break;
 184         }
 185         return cred;
 186 }
 187
 188 /**
 189  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 190  * @clp: client state handle
 191  *
 192  * Returns an rpc_cred with reference count bumped, or NULL.
 193  * Caller must hold clp->cl_lock.
 194  */
 195 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 196 {
 197         struct rpc_cred *cred = NULL;
 198         struct nfs_server *server;
 199
 200         /* Use machine credentials if available */
 201         cred = nfs4_get_machine_cred_locked(clp);
 202         if (cred != NULL)
 203                 goto out;
 204
 205         rcu_read_lock();
 206         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 207                 cred = nfs4_get_renew_cred_server_locked(server);
 208                 if (cred != NULL)
 209                         break;
 210         }
 211         rcu_read_unlock();
 212
 213 out:
 214         return cred;
 215 }
 216
 217 #if defined(CONFIG_NFS_V4_1)
 218
 219 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 220 {
 221         int status;
 222         struct nfs_fsinfo fsinfo;
 223
 224         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 225                 nfs4_schedule_state_renewal(clp);
 226                 return 0;
 227         }
 228
 229         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 230         if (status == 0) {
 231                 /* Update lease time and schedule renewal */
 232                 spin_lock(&clp->cl_lock);
 233                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 234                 clp->cl_last_renewal = jiffies;
 235                 spin_unlock(&clp->cl_lock);
 236
 237                 nfs4_schedule_state_renewal(clp);
 238         }
 239
 240         return status;
 241 }
 242
 243 /*
 244  * Back channel returns NFS4ERR_DELAY for new requests when
 245  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 246  * is ended.
 247  */
 248 static void nfs4_end_drain_session(struct nfs_client *clp)
 249 {
 250         struct nfs4_session *ses = clp->cl_session;
 251         struct nfs4_slot_table *tbl;
 252
 253         if (ses == NULL)
 254                 return;
 255         tbl = &ses->fc_slot_table;
 256         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 257                 spin_lock(&tbl->slot_tbl_lock);
 258                 nfs41_wake_slot_table(tbl);
 259                 spin_unlock(&tbl->slot_tbl_lock);
 260         }
 261 }
 262
 263 /*
 264  * Signal state manager thread if session fore channel is drained
 265  */
 266 void nfs4_session_drain_complete(struct nfs4_session *session,
 267                 struct nfs4_slot_table *tbl)
 268 {
 269         if (nfs4_session_draining(session))
 270                 complete(&tbl->complete);
 271 }
 272
 273 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 274 {
 275         spin_lock(&tbl->slot_tbl_lock);
 276         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 277                 INIT_COMPLETION(tbl->complete);
 278                 spin_unlock(&tbl->slot_tbl_lock);
 279                 return wait_for_completion_interruptible(&tbl->complete);
 280         }
 281         spin_unlock(&tbl->slot_tbl_lock);
 282         return 0;
 283 }
 284
 285 static int nfs4_begin_drain_session(struct nfs_client *clp)
 286 {
 287         struct nfs4_session *ses = clp->cl_session;
 288         int ret = 0;
 289
 290         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 291         /* back channel */
 292         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 293         if (ret)
 294                 return ret;
 295         /* fore channel */
 296         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 297 }
 298
 299 static void nfs41_finish_session_reset(struct nfs_client *clp)
 300 {
 301         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 302         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 303         /* create_session negotiated new slot table */
 304         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 305         nfs41_setup_state_renewal(clp);
 306 }
 307
 308 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 309 {
 310         int status;
 311
 312         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 313                 goto do_confirm;
 314         nfs4_begin_drain_session(clp);
 315         status = nfs4_proc_exchange_id(clp, cred);
 316         if (status != 0)
 317                 goto out;
 318         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 319 do_confirm:
 320         status = nfs4_proc_create_session(clp, cred);
 321         if (status != 0)
 322                 goto out;
 323         nfs41_finish_session_reset(clp);
 324         nfs_mark_client_ready(clp, NFS_CS_READY);
 325 out:
 326         return status;
 327 }
 328
 329 /**
 330  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 331  *
 332  * @clp: nfs_client under test
 333  * @result: OUT: found nfs_client, or clp
 334  * @cred: credential to use for trunking test
 335  *
 336  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 337  * If NFS4_OK is returned, an nfs_client pointer is planted in
 338  * "result".
 339  *
 340  * Note: The returned client may not yet be marked ready.
 341  */
 342 int nfs41_discover_server_trunking(struct nfs_client *clp,
 343                                    struct nfs_client **result,
 344                                    struct rpc_cred *cred)
 345 {
 346         int status;
 347
 348         status = nfs4_proc_exchange_id(clp, cred);
 349         if (status != NFS4_OK)
 350                 return status;
 351         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 352
 353         return nfs41_walk_client_list(clp, result, cred);
 354 }
 355
 356 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 357 {
 358         struct rpc_cred *cred;
 359
 360         spin_lock(&clp->cl_lock);
 361         cred = nfs4_get_machine_cred_locked(clp);
 362         spin_unlock(&clp->cl_lock);
 363         return cred;
 364 }
 365
 366 #endif /* CONFIG_NFS_V4_1 */
 367
 368 static struct rpc_cred *
 369 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 370 {
 371         struct nfs_client *clp = server->nfs_client;
 372         struct rpc_cred *cred = NULL;
 373         struct nfs4_state_owner *sp;
 374         struct rb_node *pos;
 375
 376         spin_lock(&clp->cl_lock);
 377         pos = rb_first(&server->state_owners);
 378         if (pos != NULL) {
 379                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 380                 cred = get_rpccred(sp->so_cred);
 381         }
 382         spin_unlock(&clp->cl_lock);
 383         return cred;
 384 }
 385
 386 /**
 387  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 388  * @clp: client state handle
 389  *
 390  * Returns an rpc_cred with reference count bumped, or NULL.
 391  */
 392 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 393 {
 394         struct nfs_server *server;
 395         struct rpc_cred *cred;
 396
 397         spin_lock(&clp->cl_lock);
 398         cred = nfs4_get_machine_cred_locked(clp);
 399         spin_unlock(&clp->cl_lock);
 400         if (cred != NULL)
 401                 goto out;
 402
 403         rcu_read_lock();
 404         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 405                 cred = nfs4_get_setclientid_cred_server(server);
 406                 if (cred != NULL)
 407                         break;
 408         }
 409         rcu_read_unlock();
 410
 411 out:
 412         return cred;
 413 }
 414
 415 static struct nfs4_state_owner *
 416 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 417 {
 418         struct rb_node **p = &server->state_owners.rb_node,
 419                        *parent = NULL;
 420         struct nfs4_state_owner *sp;
 421
 422         while (*p != NULL) {
 423                 parent = *p;
 424                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 425
 426                 if (cred < sp->so_cred)
 427                         p = &parent->rb_left;
 428                 else if (cred > sp->so_cred)
 429                         p = &parent->rb_right;
 430                 else {
 431                         if (!list_empty(&sp->so_lru))
 432                                 list_del_init(&sp->so_lru);
 433                         atomic_inc(&sp->so_count);
 434                         return sp;
 435                 }
 436         }
 437         return NULL;
 438 }
 439
 440 static struct nfs4_state_owner *
 441 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 442 {
 443         struct nfs_server *server = new->so_server;
 444         struct rb_node **p = &server->state_owners.rb_node,
 445                        *parent = NULL;
 446         struct nfs4_state_owner *sp;
 447         int err;
 448
 449         while (*p != NULL) {
 450                 parent = *p;
 451                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 452
 453                 if (new->so_cred < sp->so_cred)
 454                         p = &parent->rb_left;
 455                 else if (new->so_cred > sp->so_cred)
 456                         p = &parent->rb_right;
 457                 else {
 458                         if (!list_empty(&sp->so_lru))
 459                                 list_del_init(&sp->so_lru);
 460                         atomic_inc(&sp->so_count);
 461                         return sp;
 462                 }
 463         }
 464         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 465         if (err)
 466                 return ERR_PTR(err);
 467         rb_link_node(&new->so_server_node, parent, p);
 468         rb_insert_color(&new->so_server_node, &server->state_owners);
 469         return new;
 470 }
 471
 472 static void
 473 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 474 {
 475         struct nfs_server *server = sp->so_server;
 476
 477         if (!RB_EMPTY_NODE(&sp->so_server_node))
 478                 rb_erase(&sp->so_server_node, &server->state_owners);
 479         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 480 }
 481
 482 static void
 483 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 484 {
 485         sc->create_time = ktime_get();
 486         sc->flags = 0;
 487         sc->counter = 0;
 488         spin_lock_init(&sc->lock);
 489         INIT_LIST_HEAD(&sc->list);
 490         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 491 }
 492
 493 static void
 494 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 495 {
 496         rpc_destroy_wait_queue(&sc->wait);
 497 }
 498
 499 /*
 500  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 501  * create a new state_owner.
 502  *
 503  */
 504 static struct nfs4_state_owner *
 505 nfs4_alloc_state_owner(struct nfs_server *server,
 506                 struct rpc_cred *cred,
 507                 gfp_t gfp_flags)
 508 {
 509         struct nfs4_state_owner *sp;
 510
 511         sp = kzalloc(sizeof(*sp), gfp_flags);
 512         if (!sp)
 513                 return NULL;
 514         sp->so_server = server;
 515         sp->so_cred = get_rpccred(cred);
 516         spin_lock_init(&sp->so_lock);
 517         INIT_LIST_HEAD(&sp->so_states);
 518         nfs4_init_seqid_counter(&sp->so_seqid);
 519         atomic_set(&sp->so_count, 1);
 520         INIT_LIST_HEAD(&sp->so_lru);
 521         seqcount_init(&sp->so_reclaim_seqcount);
 522         mutex_init(&sp->so_delegreturn_mutex);
 523         return sp;
 524 }
 525
 526 static void
 527 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 528 {
 529         struct rb_node *rb_node = &sp->so_server_node;
 530
 531         if (!RB_EMPTY_NODE(rb_node)) {
 532                 struct nfs_server *server = sp->so_server;
 533                 struct nfs_client *clp = server->nfs_client;
 534
 535                 spin_lock(&clp->cl_lock);
 536                 if (!RB_EMPTY_NODE(rb_node)) {
 537                         rb_erase(rb_node, &server->state_owners);
 538                         RB_CLEAR_NODE(rb_node);
 539                 }
 540                 spin_unlock(&clp->cl_lock);
 541         }
 542 }
 543
 544 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 545 {
 546         nfs4_destroy_seqid_counter(&sp->so_seqid);
 547         put_rpccred(sp->so_cred);
 548         kfree(sp);
 549 }
 550
 551 static void nfs4_gc_state_owners(struct nfs_server *server)
 552 {
 553         struct nfs_client *clp = server->nfs_client;
 554         struct nfs4_state_owner *sp, *tmp;
 555         unsigned long time_min, time_max;
 556         LIST_HEAD(doomed);
 557
 558         spin_lock(&clp->cl_lock);
 559         time_max = jiffies;
 560         time_min = (long)time_max - (long)clp->cl_lease_time;
 561         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 562                 /* NB: LRU is sorted so that oldest is at the head */
 563                 if (time_in_range(sp->so_expires, time_min, time_max))
 564                         break;
 565                 list_move(&sp->so_lru, &doomed);
 566                 nfs4_remove_state_owner_locked(sp);
 567         }
 568         spin_unlock(&clp->cl_lock);
 569
 570         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 571                 list_del(&sp->so_lru);
 572                 nfs4_free_state_owner(sp);
 573         }
 574 }
 575
 576 /**
 577  * nfs4_get_state_owner - Look up a state owner given a credential
 578  * @server: nfs_server to search
 579  * @cred: RPC credential to match
 580  *
 581  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 582  */
 583 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 584                                               struct rpc_cred *cred,
 585                                               gfp_t gfp_flags)
 586 {
 587         struct nfs_client *clp = server->nfs_client;
 588         struct nfs4_state_owner *sp, *new;
 589
 590         spin_lock(&clp->cl_lock);
 591         sp = nfs4_find_state_owner_locked(server, cred);
 592         spin_unlock(&clp->cl_lock);
 593         if (sp != NULL)
 594                 goto out;
 595         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 596         if (new == NULL)
 597                 goto out;
 598         do {
 599                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 600                         break;
 601                 spin_lock(&clp->cl_lock);
 602                 sp = nfs4_insert_state_owner_locked(new);
 603                 spin_unlock(&clp->cl_lock);
 604         } while (sp == ERR_PTR(-EAGAIN));
 605         if (sp != new)
 606                 nfs4_free_state_owner(new);
 607 out:
 608         nfs4_gc_state_owners(server);
 609         return sp;
 610 }
 611
 612 /**
 613  * nfs4_put_state_owner - Release a nfs4_state_owner
 614  * @sp: state owner data to release
 615  *
 616  * Note that we keep released state owners on an LRU
 617  * list.
 618  * This caches valid state owners so that they can be
 619  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 620  * It also pins the uniquifier of dropped state owners for
 621  * a while, to ensure that those state owner names are
 622  * never reused.
 623  */
 624 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 625 {
 626         struct nfs_server *server = sp->so_server;
 627         struct nfs_client *clp = server->nfs_client;
 628
 629         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 630                 return;
 631
 632         sp->so_expires = jiffies;
 633         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 634         spin_unlock(&clp->cl_lock);
 635 }
 636
 637 /**
 638  * nfs4_purge_state_owners - Release all cached state owners
 639  * @server: nfs_server with cached state owners to release
 640  *
 641  * Called at umount time.  Remaining state owners will be on
 642  * the LRU with ref count of zero.
 643  */
 644 void nfs4_purge_state_owners(struct nfs_server *server)
 645 {
 646         struct nfs_client *clp = server->nfs_client;
 647         struct nfs4_state_owner *sp, *tmp;
 648         LIST_HEAD(doomed);
 649
 650         spin_lock(&clp->cl_lock);
 651         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 652                 list_move(&sp->so_lru, &doomed);
 653                 nfs4_remove_state_owner_locked(sp);
 654         }
 655         spin_unlock(&clp->cl_lock);
 656
 657         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 658                 list_del(&sp->so_lru);
 659                 nfs4_free_state_owner(sp);
 660         }
 661 }
 662
 663 static struct nfs4_state *
 664 nfs4_alloc_open_state(void)
 665 {
 666         struct nfs4_state *state;
 667
 668         state = kzalloc(sizeof(*state), GFP_NOFS);
 669         if (!state)
 670                 return NULL;
 671         atomic_set(&state->count, 1);
 672         INIT_LIST_HEAD(&state->lock_states);
 673         spin_lock_init(&state->state_lock);
 674         seqlock_init(&state->seqlock);
 675         return state;
 676 }
 677
 678 void
 679 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 680 {
 681         if (state->state == fmode)
 682                 return;
 683         /* NB! List reordering - see the reclaim code for why.  */
 684         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 685                 if (fmode & FMODE_WRITE)
 686                         list_move(&state->open_states, &state->owner->so_states);
 687                 else
 688                         list_move_tail(&state->open_states, &state->owner->so_states);
 689         }
 690         state->state = fmode;
 691 }
 692
 693 static struct nfs4_state *
 694 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 695 {
 696         struct nfs_inode *nfsi = NFS_I(inode);
 697         struct nfs4_state *state;
 698
 699         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 700                 if (state->owner != owner)
 701                         continue;
 702                 if (!nfs4_valid_open_stateid(state))
 703                         continue;
 704                 if (atomic_inc_not_zero(&state->count))
 705                         return state;
 706         }
 707         return NULL;
 708 }
 709
 710 static void
 711 nfs4_free_open_state(struct nfs4_state *state)
 712 {
 713         kfree(state);
 714 }
 715
 716 struct nfs4_state *
 717 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 718 {
 719         struct nfs4_state *state, *new;
 720         struct nfs_inode *nfsi = NFS_I(inode);
 721
 722         spin_lock(&inode->i_lock);
 723         state = __nfs4_find_state_byowner(inode, owner);
 724         spin_unlock(&inode->i_lock);
 725         if (state)
 726                 goto out;
 727         new = nfs4_alloc_open_state();
 728         spin_lock(&owner->so_lock);
 729         spin_lock(&inode->i_lock);
 730         state = __nfs4_find_state_byowner(inode, owner);
 731         if (state == NULL && new != NULL) {
 732                 state = new;
 733                 state->owner = owner;
 734                 atomic_inc(&owner->so_count);
 735                 list_add(&state->inode_states, &nfsi->open_states);
 736                 ihold(inode);
 737                 state->inode = inode;
 738                 spin_unlock(&inode->i_lock);
 739                 /* Note: The reclaim code dictates that we add stateless
 740                  * and read-only stateids to the end of the list */
 741                 list_add_tail(&state->open_states, &owner->so_states);
 742                 spin_unlock(&owner->so_lock);
 743         } else {
 744                 spin_unlock(&inode->i_lock);
 745                 spin_unlock(&owner->so_lock);
 746                 if (new)
 747                         nfs4_free_open_state(new);
 748         }
 749 out:
 750         return state;
 751 }
 752
 753 void nfs4_put_open_state(struct nfs4_state *state)
 754 {
 755         struct inode *inode = state->inode;
 756         struct nfs4_state_owner *owner = state->owner;
 757
 758         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 759                 return;
 760         spin_lock(&inode->i_lock);
 761         list_del(&state->inode_states);
 762         list_del(&state->open_states);
 763         spin_unlock(&inode->i_lock);
 764         spin_unlock(&owner->so_lock);
 765         iput(inode);
 766         nfs4_free_open_state(state);
 767         nfs4_put_state_owner(owner);
 768 }
 769
 770 /*
 771  * Close the current file.
 772  */
 773 static void __nfs4_close(struct nfs4_state *state,
 774                 fmode_t fmode, gfp_t gfp_mask, int wait)
 775 {
 776         struct nfs4_state_owner *owner = state->owner;
 777         int call_close = 0;
 778         fmode_t newstate;
 779
 780         atomic_inc(&owner->so_count);
 781         /* Protect against nfs4_find_state() */
 782         spin_lock(&owner->so_lock);
 783         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 784                 case FMODE_READ:
 785                         state->n_rdonly--;
 786                         break;
 787                 case FMODE_WRITE:
 788                         state->n_wronly--;
 789                         break;
 790                 case FMODE_READ|FMODE_WRITE:
 791                         state->n_rdwr--;
 792         }
 793         newstate = FMODE_READ|FMODE_WRITE;
 794         if (state->n_rdwr == 0) {
 795                 if (state->n_rdonly == 0) {
 796                         newstate &= ~FMODE_READ;
 797                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 798                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 799                 }
 800                 if (state->n_wronly == 0) {
 801                         newstate &= ~FMODE_WRITE;
 802                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 803                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 804                 }
 805                 if (newstate == 0)
 806                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 807         }
 808         nfs4_state_set_mode_locked(state, newstate);
 809         spin_unlock(&owner->so_lock);
 810
 811         if (!call_close) {
 812                 nfs4_put_open_state(state);
 813                 nfs4_put_state_owner(owner);
 814         } else
 815                 nfs4_do_close(state, gfp_mask, wait);
 816 }
 817
 818 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 819 {
 820         __nfs4_close(state, fmode, GFP_NOFS, 0);
 821 }
 822
 823 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 824 {
 825         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 826 }
 827
 828 /*
 829  * Search the state->lock_states for an existing lock_owner
 830  * that is compatible with current->files
 831  */
 832 static struct nfs4_lock_state *
 833 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 834 {
 835         struct nfs4_lock_state *pos;
 836         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 837                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 838                         continue;
 839                 switch (pos->ls_owner.lo_type) {
 840                 case NFS4_POSIX_LOCK_TYPE:
 841                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 842                                 continue;
 843                         break;
 844                 case NFS4_FLOCK_LOCK_TYPE:
 845                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 846                                 continue;
 847                 }
 848                 atomic_inc(&pos->ls_count);
 849                 return pos;
 850         }
 851         return NULL;
 852 }
 853
 854 /*
 855  * Return a compatible lock_state. If no initialized lock_state structure
 856  * exists, return an uninitialized one.
 857  *
 858  */
 859 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)
 860 {
 861         struct nfs4_lock_state *lsp;
 862         struct nfs_server *server = state->owner->so_server;
 863
 864         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 865         if (lsp == NULL)
 866                 return NULL;
 867         nfs4_init_seqid_counter(&lsp->ls_seqid);
 868         atomic_set(&lsp->ls_count, 1);
 869         lsp->ls_state = state;
 870         lsp->ls_owner.lo_type = type;
 871         switch (lsp->ls_owner.lo_type) {
 872         case NFS4_FLOCK_LOCK_TYPE:
 873                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 874                 break;
 875         case NFS4_POSIX_LOCK_TYPE:
 876                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 877                 break;
 878         default:
 879                 goto out_free;
 880         }
 881         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 882         if (lsp->ls_seqid.owner_id < 0)
 883                 goto out_free;
 884         INIT_LIST_HEAD(&lsp->ls_locks);
 885         return lsp;
 886 out_free:
 887         kfree(lsp);
 888         return NULL;
 889 }
 890
 891 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 892 {
 893         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 894         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 895         kfree(lsp);
 896 }
 897
 898 /*
 899  * Return a compatible lock_state. If no initialized lock_state structure
 900  * exists, return an uninitialized one.
 901  *
 902  */
 903 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 904 {
 905         struct nfs4_lock_state *lsp, *new = NULL;
 906
 907         for(;;) {
 908                 spin_lock(&state->state_lock);
 909                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 910                 if (lsp != NULL)
 911                         break;
 912                 if (new != NULL) {
 913                         list_add(&new->ls_locks, &state->lock_states);
 914                         set_bit(LK_STATE_IN_USE, &state->flags);
 915                         lsp = new;
 916                         new = NULL;
 917                         break;
 918                 }
 919                 spin_unlock(&state->state_lock);
 920                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 921                 if (new == NULL)
 922                         return NULL;
 923         }
 924         spin_unlock(&state->state_lock);
 925         if (new != NULL)
 926                 nfs4_free_lock_state(state->owner->so_server, new);
 927         return lsp;
 928 }
 929
 930 /*
 931  * Release reference to lock_state, and free it if we see that
 932  * it is no longer in use
 933  */
 934 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 935 {
 936         struct nfs4_state *state;
 937
 938         if (lsp == NULL)
 939                 return;
 940         state = lsp->ls_state;
 941         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 942                 return;
 943         list_del(&lsp->ls_locks);
 944         if (list_empty(&state->lock_states))
 945                 clear_bit(LK_STATE_IN_USE, &state->flags);
 946         spin_unlock(&state->state_lock);
 947         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 948                 if (nfs4_release_lockowner(lsp) == 0)
 949                         return;
 950         }
 951         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 952 }
 953
 954 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 955 {
 956         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 957
 958         dst->fl_u.nfs4_fl.owner = lsp;
 959         atomic_inc(&lsp->ls_count);
 960 }
 961
 962 static void nfs4_fl_release_lock(struct file_lock *fl)
 963 {
 964         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 965 }
 966
 967 static const struct file_lock_operations nfs4_fl_lock_ops = {
 968         .fl_copy_lock = nfs4_fl_copy_lock,
 969         .fl_release_private = nfs4_fl_release_lock,
 970 };
 971
 972 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 973 {
 974         struct nfs4_lock_state *lsp;
 975
 976         if (fl->fl_ops != NULL)
 977                 return 0;
 978         if (fl->fl_flags & FL_POSIX)
 979                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 980         else if (fl->fl_flags & FL_FLOCK)
 981                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 982                                 NFS4_FLOCK_LOCK_TYPE);
 983         else
 984                 return -EINVAL;
 985         if (lsp == NULL)
 986                 return -ENOMEM;
 987         fl->fl_u.nfs4_fl.owner = lsp;
 988         fl->fl_ops = &nfs4_fl_lock_ops;
 989         return 0;
 990 }
 991
 992 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 993                 struct nfs4_state *state,
 994                 const struct nfs_lockowner *lockowner)
 995 {
 996         struct nfs4_lock_state *lsp;
 997         fl_owner_t fl_owner;
 998         pid_t fl_pid;
 999         int ret = -ENOENT;
1000
1001
1002         if (lockowner == NULL)
1003                 goto out;
1004
1005         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1006                 goto out;
1007
1008         fl_owner = lockowner->l_owner;
1009         fl_pid = lockowner->l_pid;
1010         spin_lock(&state->state_lock);
1011         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1012         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1013                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1014                 ret = 0;
1015                 smp_rmb();
1016                 if (!list_empty(&lsp->ls_seqid.list))
1017                         ret = -EWOULDBLOCK;
1018         }
1019         spin_unlock(&state->state_lock);
1020         nfs4_put_lock_state(lsp);
1021 out:
1022         return ret;
1023 }
1024
1025 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1026 {
1027         int ret;
1028         int seq;
1029
1030         do {
1031                 seq = read_seqbegin(&state->seqlock);
1032                 nfs4_stateid_copy(dst, &state->stateid);
1033                 ret = 0;
1034                 smp_rmb();
1035                 if (!list_empty(&state->owner->so_seqid.list))
1036                         ret = -EWOULDBLOCK;
1037         } while (read_seqretry(&state->seqlock, seq));
1038         return ret;
1039 }
1040
1041 /*
1042  * Byte-range lock aware utility to initialize the stateid of read/write
1043  * requests.
1044  */
1045 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1046                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1047 {
1048         int ret = 0;
1049         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1050                 goto out;
1051         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1052         if (ret != -ENOENT)
1053                 goto out;
1054         ret = nfs4_copy_open_stateid(dst, state);
1055 out:
1056         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1057                 dst->seqid = 0;
1058         return ret;
1059 }
1060
1061 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1062 {
1063         struct nfs_seqid *new;
1064
1065         new = kmalloc(sizeof(*new), gfp_mask);
1066         if (new != NULL) {
1067                 new->sequence = counter;
1068                 INIT_LIST_HEAD(&new->list);
1069                 new->task = NULL;
1070         }
1071         return new;
1072 }
1073
1074 void nfs_release_seqid(struct nfs_seqid *seqid)
1075 {
1076         struct nfs_seqid_counter *sequence;
1077
1078         if (list_empty(&seqid->list))
1079                 return;
1080         sequence = seqid->sequence;
1081         spin_lock(&sequence->lock);
1082         list_del_init(&seqid->list);
1083         if (!list_empty(&sequence->list)) {
1084                 struct nfs_seqid *next;
1085
1086                 next = list_first_entry(&sequence->list,
1087                                 struct nfs_seqid, list);
1088                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1089         }
1090         spin_unlock(&sequence->lock);
1091 }
1092
1093 void nfs_free_seqid(struct nfs_seqid *seqid)
1094 {
1095         nfs_release_seqid(seqid);
1096         kfree(seqid);
1097 }
1098
1099 /*
1100  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1101  * failed with a seqid incrementing error -
1102  * see comments nfs_fs.h:seqid_mutating_error()
1103  */
1104 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1105 {
1106         switch (status) {
1107                 case 0:
1108                         break;
1109                 case -NFS4ERR_BAD_SEQID:
1110                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1111                                 return;
1112                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1113                                         " sequence-id error on an"
1114                                         " unconfirmed sequence %p!\n",
1115                                         seqid->sequence);
1116                 case -NFS4ERR_STALE_CLIENTID:
1117                 case -NFS4ERR_STALE_STATEID:
1118                 case -NFS4ERR_BAD_STATEID:
1119                 case -NFS4ERR_BADXDR:
1120                 case -NFS4ERR_RESOURCE:
1121                 case -NFS4ERR_NOFILEHANDLE:
1122                         /* Non-seqid mutating errors */
1123                         return;
1124         };
1125         /*
1126          * Note: no locking needed as we are guaranteed to be first
1127          * on the sequence list
1128          */
1129         seqid->sequence->counter++;
1130 }
1131
1132 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1133 {
1134         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1135                                         struct nfs4_state_owner, so_seqid);
1136         struct nfs_server *server = sp->so_server;
1137
1138         if (status == -NFS4ERR_BAD_SEQID)
1139                 nfs4_drop_state_owner(sp);
1140         if (!nfs4_has_session(server->nfs_client))
1141                 nfs_increment_seqid(status, seqid);
1142 }
1143
1144 /*
1145  * Increment the seqid if the LOCK/LOCKU succeeded, or
1146  * failed with a seqid incrementing error -
1147  * see comments nfs_fs.h:seqid_mutating_error()
1148  */
1149 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1150 {
1151         nfs_increment_seqid(status, seqid);
1152 }
1153
1154 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1155 {
1156         struct nfs_seqid_counter *sequence = seqid->sequence;
1157         int status = 0;
1158
1159         spin_lock(&sequence->lock);
1160         seqid->task = task;
1161         if (list_empty(&seqid->list))
1162                 list_add_tail(&seqid->list, &sequence->list);
1163         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1164                 goto unlock;
1165         rpc_sleep_on(&sequence->wait, task, NULL);
1166         status = -EAGAIN;
1167 unlock:
1168         spin_unlock(&sequence->lock);
1169         return status;
1170 }
1171
1172 static int nfs4_run_state_manager(void *);
1173
1174 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1175 {
1176         smp_mb__before_clear_bit();
1177         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1178         smp_mb__after_clear_bit();
1179         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1180         rpc_wake_up(&clp->cl_rpcwaitq);
1181 }
1182
1183 /*
1184  * Schedule the nfs_client asynchronous state management routine
1185  */
1186 void nfs4_schedule_state_manager(struct nfs_client *clp)
1187 {
1188         struct task_struct *task;
1189         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1190
1191         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1192                 return;
1193         __module_get(THIS_MODULE);
1194         atomic_inc(&clp->cl_count);
1195
1196         /* The rcu_read_lock() is not strictly necessary, as the state
1197          * manager is the only thread that ever changes the rpc_xprt
1198          * after it's initialized.  At this point, we're single threaded. */
1199         rcu_read_lock();
1200         snprintf(buf, sizeof(buf), "%s-manager",
1201                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1202         rcu_read_unlock();
1203         task = kthread_run(nfs4_run_state_manager, clp, buf);
1204         if (IS_ERR(task)) {
1205                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1206                         __func__, PTR_ERR(task));
1207                 nfs4_clear_state_manager_bit(clp);
1208                 nfs_put_client(clp);
1209                 module_put(THIS_MODULE);
1210         }
1211 }
1212
1213 /*
1214  * Schedule a lease recovery attempt
1215  */
1216 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1217 {
1218         if (!clp)
1219                 return;
1220         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1221                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1222         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1223                         clp->cl_hostname);
1224         nfs4_schedule_state_manager(clp);
1225 }
1226 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1227
1228 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1229 {
1230         int res;
1231
1232         might_sleep();
1233
1234         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1235                         nfs_wait_bit_killable, TASK_KILLABLE);
1236         if (res)
1237                 return res;
1238
1239         if (clp->cl_cons_state < 0)
1240                 return clp->cl_cons_state;
1241         return 0;
1242 }
1243
1244 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1245 {
1246         unsigned int loop;
1247         int ret;
1248
1249         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1250                 ret = nfs4_wait_clnt_recover(clp);
1251                 if (ret != 0)
1252                         break;
1253                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1254                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1255                         break;
1256                 nfs4_schedule_state_manager(clp);
1257                 ret = -EIO;
1258         }
1259         return ret;
1260 }
1261
1262 /*
1263  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1264  * @clp: client to process
1265  *
1266  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1267  * resend of the SETCLIENTID and hence re-establish the
1268  * callback channel. Then return all existing delegations.
1269  */
1270 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1271 {
1272         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1273         nfs_expire_all_delegations(clp);
1274         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1275                         clp->cl_hostname);
1276 }
1277
1278 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1279 {
1280         nfs40_handle_cb_pathdown(clp);
1281         nfs4_schedule_state_manager(clp);
1282 }
1283
1284 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1285 {
1286
1287         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1288         /* Don't recover state that expired before the reboot */
1289         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1290                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1291                 return 0;
1292         }
1293         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1294         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1295         return 1;
1296 }
1297
1298 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1299 {
1300         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1301         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1302         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1303         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1304         return 1;
1305 }
1306
1307 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1308 {
1309         struct nfs_client *clp = server->nfs_client;
1310
1311         if (!nfs4_valid_open_stateid(state))
1312                 return -EBADF;
1313         nfs4_state_mark_reclaim_nograce(clp, state);
1314         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1315                         clp->cl_hostname);
1316         nfs4_schedule_state_manager(clp);
1317         return 0;
1318 }
1319 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1320
1321 void nfs_inode_find_state_and_recover(struct inode *inode,
1322                 const nfs4_stateid *stateid)
1323 {
1324         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1325         struct nfs_inode *nfsi = NFS_I(inode);
1326         struct nfs_open_context *ctx;
1327         struct nfs4_state *state;
1328         bool found = false;
1329
1330         spin_lock(&inode->i_lock);
1331         list_for_each_entry(ctx, &nfsi->open_files, list) {
1332                 state = ctx->state;
1333                 if (state == NULL)
1334                         continue;
1335                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1336                         continue;
1337                 if (!nfs4_stateid_match(&state->stateid, stateid))
1338                         continue;
1339                 nfs4_state_mark_reclaim_nograce(clp, state);
1340                 found = true;
1341         }
1342         spin_unlock(&inode->i_lock);
1343         if (found)
1344                 nfs4_schedule_state_manager(clp);
1345 }
1346
1347 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1348 {
1349         struct inode *inode = state->inode;
1350         struct nfs_inode *nfsi = NFS_I(inode);
1351         struct nfs_open_context *ctx;
1352
1353         spin_lock(&inode->i_lock);
1354         list_for_each_entry(ctx, &nfsi->open_files, list) {
1355                 if (ctx->state != state)
1356                         continue;
1357                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1358         }
1359         spin_unlock(&inode->i_lock);
1360 }
1361
1362 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1363 {
1364         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1365         nfs4_state_mark_open_context_bad(state);
1366 }
1367
1368
1369 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1370 {
1371         struct inode *inode = state->inode;
1372         struct nfs_inode *nfsi = NFS_I(inode);
1373         struct file_lock *fl;
1374         int status = 0;
1375
1376         if (inode->i_flock == NULL)
1377                 return 0;
1378
1379         /* Guard against delegation returns and new lock/unlock calls */
1380         down_write(&nfsi->rwsem);
1381         /* Protect inode->i_flock using the BKL */
1382         lock_flocks();
1383         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1384                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1385                         continue;
1386                 if (nfs_file_open_context(fl->fl_file)->state != state)
1387                         continue;
1388                 unlock_flocks();
1389                 status = ops->recover_lock(state, fl);
1390                 switch (status) {
1391                         case 0:
1392                                 break;
1393                         case -ESTALE:
1394                         case -NFS4ERR_ADMIN_REVOKED:
1395                         case -NFS4ERR_STALE_STATEID:
1396                         case -NFS4ERR_BAD_STATEID:
1397                         case -NFS4ERR_EXPIRED:
1398                         case -NFS4ERR_NO_GRACE:
1399                         case -NFS4ERR_STALE_CLIENTID:
1400                         case -NFS4ERR_BADSESSION:
1401                         case -NFS4ERR_BADSLOT:
1402                         case -NFS4ERR_BAD_HIGH_SLOT:
1403                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1404                                 goto out;
1405                         default:
1406                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1407                                         "Zeroing state\n", __func__, status);
1408                         case -ENOMEM:
1409                         case -NFS4ERR_DENIED:
1410                         case -NFS4ERR_RECLAIM_BAD:
1411                         case -NFS4ERR_RECLAIM_CONFLICT:
1412                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1413                                 status = 0;
1414                 }
1415                 lock_flocks();
1416         }
1417         unlock_flocks();
1418 out:
1419         up_write(&nfsi->rwsem);
1420         return status;
1421 }
1422
1423 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1424 {
1425         struct nfs4_state *state;
1426         struct nfs4_lock_state *lock;
1427         int status = 0;
1428
1429         /* Note: we rely on the sp->so_states list being ordered
1430          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1431          * states first.
1432          * This is needed to ensure that the server won't give us any
1433          * read delegations that we have to return if, say, we are
1434          * recovering after a network partition or a reboot from a
1435          * server that doesn't support a grace period.
1436          */
1437         spin_lock(&sp->so_lock);
1438         write_seqcount_begin(&sp->so_reclaim_seqcount);
1439 restart:
1440         list_for_each_entry(state, &sp->so_states, open_states) {
1441                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1442                         continue;
1443                 if (!nfs4_valid_open_stateid(state))
1444                         continue;
1445                 if (state->state == 0)
1446                         continue;
1447                 atomic_inc(&state->count);
1448                 spin_unlock(&sp->so_lock);
1449                 status = ops->recover_open(sp, state);
1450                 if (status >= 0) {
1451                         status = nfs4_reclaim_locks(state, ops);
1452                         if (status >= 0) {
1453                                 spin_lock(&state->state_lock);
1454                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1455                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1456                                                 pr_warn_ratelimited("NFS: "
1457                                                         "%s: Lock reclaim "
1458                                                         "failed!\n", __func__);
1459                                 }
1460                                 spin_unlock(&state->state_lock);
1461                                 nfs4_put_open_state(state);
1462                                 spin_lock(&sp->so_lock);
1463                                 goto restart;
1464                         }
1465                 }
1466                 switch (status) {
1467                         default:
1468                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1469                                         "Zeroing state\n", __func__, status);
1470                         case -ENOENT:
1471                         case -ENOMEM:
1472                         case -ESTALE:
1473                                 /*
1474                                  * Open state on this file cannot be recovered
1475                                  * All we can do is revert to using the zero stateid.
1476                                  */
1477                                 nfs4_state_mark_recovery_failed(state, status);
1478                                 break;
1479                         case -EAGAIN:
1480                                 ssleep(1);
1481                         case -NFS4ERR_ADMIN_REVOKED:
1482                         case -NFS4ERR_STALE_STATEID:
1483                         case -NFS4ERR_BAD_STATEID:
1484                         case -NFS4ERR_RECLAIM_BAD:
1485                         case -NFS4ERR_RECLAIM_CONFLICT:
1486                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1487                                 break;
1488                         case -NFS4ERR_EXPIRED:
1489                         case -NFS4ERR_NO_GRACE:
1490                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1491                         case -NFS4ERR_STALE_CLIENTID:
1492                         case -NFS4ERR_BADSESSION:
1493                         case -NFS4ERR_BADSLOT:
1494                         case -NFS4ERR_BAD_HIGH_SLOT:
1495                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1496                                 goto out_err;
1497                 }
1498                 nfs4_put_open_state(state);
1499                 spin_lock(&sp->so_lock);
1500                 goto restart;
1501         }
1502         write_seqcount_end(&sp->so_reclaim_seqcount);
1503         spin_unlock(&sp->so_lock);
1504         return 0;
1505 out_err:
1506         nfs4_put_open_state(state);
1507         spin_lock(&sp->so_lock);
1508         write_seqcount_end(&sp->so_reclaim_seqcount);
1509         spin_unlock(&sp->so_lock);
1510         return status;
1511 }
1512
1513 static void nfs4_clear_open_state(struct nfs4_state *state)
1514 {
1515         struct nfs4_lock_state *lock;
1516
1517         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1518         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1519         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1520         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1521         spin_lock(&state->state_lock);
1522         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1523                 lock->ls_seqid.flags = 0;
1524                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1525         }
1526         spin_unlock(&state->state_lock);
1527 }
1528
1529 static void nfs4_reset_seqids(struct nfs_server *server,
1530         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1531 {
1532         struct nfs_client *clp = server->nfs_client;
1533         struct nfs4_state_owner *sp;
1534         struct rb_node *pos;
1535         struct nfs4_state *state;
1536
1537         spin_lock(&clp->cl_lock);
1538         for (pos = rb_first(&server->state_owners);
1539              pos != NULL;
1540              pos = rb_next(pos)) {
1541                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1542                 sp->so_seqid.flags = 0;
1543                 spin_lock(&sp->so_lock);
1544                 list_for_each_entry(state, &sp->so_states, open_states) {
1545                         if (mark_reclaim(clp, state))
1546                                 nfs4_clear_open_state(state);
1547                 }
1548                 spin_unlock(&sp->so_lock);
1549         }
1550         spin_unlock(&clp->cl_lock);
1551 }
1552
1553 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1554         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1555 {
1556         struct nfs_server *server;
1557
1558         rcu_read_lock();
1559         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1560                 nfs4_reset_seqids(server, mark_reclaim);
1561         rcu_read_unlock();
1562 }
1563
1564 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1565 {
1566         /* Mark all delegations for reclaim */
1567         nfs_delegation_mark_reclaim(clp);
1568         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1569 }
1570
1571 static void nfs4_reclaim_complete(struct nfs_client *clp,
1572                                  const struct nfs4_state_recovery_ops *ops)
1573 {
1574         /* Notify the server we're done reclaiming our state */
1575         if (ops->reclaim_complete)
1576                 (void)ops->reclaim_complete(clp);
1577 }
1578
1579 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1580 {
1581         struct nfs_client *clp = server->nfs_client;
1582         struct nfs4_state_owner *sp;
1583         struct rb_node *pos;
1584         struct nfs4_state *state;
1585
1586         spin_lock(&clp->cl_lock);
1587         for (pos = rb_first(&server->state_owners);
1588              pos != NULL;
1589              pos = rb_next(pos)) {
1590                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1591                 spin_lock(&sp->so_lock);
1592                 list_for_each_entry(state, &sp->so_states, open_states) {
1593                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1594                                                 &state->flags))
1595                                 continue;
1596                         nfs4_state_mark_reclaim_nograce(clp, state);
1597                 }
1598                 spin_unlock(&sp->so_lock);
1599         }
1600         spin_unlock(&clp->cl_lock);
1601 }
1602
1603 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1604 {
1605         struct nfs_server *server;
1606
1607         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1608                 return 0;
1609
1610         rcu_read_lock();
1611         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1612                 nfs4_clear_reclaim_server(server);
1613         rcu_read_unlock();
1614
1615         nfs_delegation_reap_unclaimed(clp);
1616         return 1;
1617 }
1618
1619 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1620 {
1621         if (!nfs4_state_clear_reclaim_reboot(clp))
1622                 return;
1623         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1624 }
1625
1626 static void nfs_delegation_clear_all(struct nfs_client *clp)
1627 {
1628         nfs_delegation_mark_reclaim(clp);
1629         nfs_delegation_reap_unclaimed(clp);
1630 }
1631
1632 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1633 {
1634         nfs_delegation_clear_all(clp);
1635         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1636 }
1637
1638 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1639 {
1640         switch (error) {
1641                 case 0:
1642                         break;
1643                 case -NFS4ERR_CB_PATH_DOWN:
1644                         nfs40_handle_cb_pathdown(clp);
1645                         break;
1646                 case -NFS4ERR_NO_GRACE:
1647                         nfs4_state_end_reclaim_reboot(clp);
1648                         break;
1649                 case -NFS4ERR_STALE_CLIENTID:
1650                 case -NFS4ERR_LEASE_MOVED:
1651                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1652                         nfs4_state_clear_reclaim_reboot(clp);
1653                         nfs4_state_start_reclaim_reboot(clp);
1654                         break;
1655                 case -NFS4ERR_EXPIRED:
1656                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1657                         nfs4_state_start_reclaim_nograce(clp);
1658                         break;
1659                 case -NFS4ERR_BADSESSION:
1660                 case -NFS4ERR_BADSLOT:
1661                 case -NFS4ERR_BAD_HIGH_SLOT:
1662                 case -NFS4ERR_DEADSESSION:
1663                 case -NFS4ERR_SEQ_FALSE_RETRY:
1664                 case -NFS4ERR_SEQ_MISORDERED:
1665                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1666                         /* Zero session reset errors */
1667                         break;
1668                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1669                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1670                         break;
1671                 default:
1672                         dprintk("%s: failed to handle error %d for server %s\n",
1673                                         __func__, error, clp->cl_hostname);
1674                         return error;
1675         }
1676         dprintk("%s: handled error %d for server %s\n", __func__, error,
1677                         clp->cl_hostname);
1678         return 0;
1679 }
1680
1681 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1682 {
1683         struct nfs4_state_owner *sp;
1684         struct nfs_server *server;
1685         struct rb_node *pos;
1686         int status = 0;
1687
1688 restart:
1689         rcu_read_lock();
1690         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1691                 nfs4_purge_state_owners(server);
1692                 spin_lock(&clp->cl_lock);
1693                 for (pos = rb_first(&server->state_owners);
1694                      pos != NULL;
1695                      pos = rb_next(pos)) {
1696                         sp = rb_entry(pos,
1697                                 struct nfs4_state_owner, so_server_node);
1698                         if (!test_and_clear_bit(ops->owner_flag_bit,
1699                                                         &sp->so_flags))
1700                                 continue;
1701                         atomic_inc(&sp->so_count);
1702                         spin_unlock(&clp->cl_lock);
1703                         rcu_read_unlock();
1704
1705                         status = nfs4_reclaim_open_state(sp, ops);
1706                         if (status < 0) {
1707                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1708                                 nfs4_put_state_owner(sp);
1709                                 return nfs4_recovery_handle_error(clp, status);
1710                         }
1711
1712                         nfs4_put_state_owner(sp);
1713                         goto restart;
1714                 }
1715                 spin_unlock(&clp->cl_lock);
1716         }
1717         rcu_read_unlock();
1718         return status;
1719 }
1720
1721 static int nfs4_check_lease(struct nfs_client *clp)
1722 {
1723         struct rpc_cred *cred;
1724         const struct nfs4_state_maintenance_ops *ops =
1725                 clp->cl_mvops->state_renewal_ops;
1726         int status;
1727
1728         /* Is the client already known to have an expired lease? */
1729         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1730                 return 0;
1731         spin_lock(&clp->cl_lock);
1732         cred = ops->get_state_renewal_cred_locked(clp);
1733         spin_unlock(&clp->cl_lock);
1734         if (cred == NULL) {
1735                 cred = nfs4_get_setclientid_cred(clp);
1736                 status = -ENOKEY;
1737                 if (cred == NULL)
1738                         goto out;
1739         }
1740         status = ops->renew_lease(clp, cred);
1741         put_rpccred(cred);
1742 out:
1743         return nfs4_recovery_handle_error(clp, status);
1744 }
1745
1746 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1747  * and for recoverable errors on EXCHANGE_ID for v4.1
1748  */
1749 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1750 {
1751         switch (status) {
1752         case -NFS4ERR_SEQ_MISORDERED:
1753                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1754                         return -ESERVERFAULT;
1755                 /* Lease confirmation error: retry after purging the lease */
1756                 ssleep(1);
1757                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1758                 break;
1759         case -NFS4ERR_STALE_CLIENTID:
1760                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1761                 nfs4_state_clear_reclaim_reboot(clp);
1762                 nfs4_state_start_reclaim_reboot(clp);
1763                 break;
1764         case -NFS4ERR_CLID_INUSE:
1765                 pr_err("NFS: Server %s reports our clientid is in use\n",
1766                         clp->cl_hostname);
1767                 nfs_mark_client_ready(clp, -EPERM);
1768                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1769                 return -EPERM;
1770         case -EACCES:
1771                 if (clp->cl_machine_cred == NULL)
1772                         return -EACCES;
1773                 /* Handle case where the user hasn't set up machine creds */
1774                 nfs4_clear_machine_cred(clp);
1775         case -NFS4ERR_DELAY:
1776         case -ETIMEDOUT:
1777         case -EAGAIN:
1778                 ssleep(1);
1779                 break;
1780
1781         case -NFS4ERR_MINOR_VERS_MISMATCH:
1782                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1783                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1784                 dprintk("%s: exit with error %d for server %s\n",
1785                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1786                 return -EPROTONOSUPPORT;
1787         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1788                                  * in nfs4_exchange_id */
1789         default:
1790                 dprintk("%s: exit with error %d for server %s\n", __func__,
1791                                 status, clp->cl_hostname);
1792                 return status;
1793         }
1794         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1795         dprintk("%s: handled error %d for server %s\n", __func__, status,
1796                         clp->cl_hostname);
1797         return 0;
1798 }
1799
1800 static int nfs4_establish_lease(struct nfs_client *clp)
1801 {
1802         struct rpc_cred *cred;
1803         const struct nfs4_state_recovery_ops *ops =
1804                 clp->cl_mvops->reboot_recovery_ops;
1805         int status;
1806
1807         cred = ops->get_clid_cred(clp);
1808         if (cred == NULL)
1809                 return -ENOENT;
1810         status = ops->establish_clid(clp, cred);
1811         put_rpccred(cred);
1812         if (status != 0)
1813                 return status;
1814         pnfs_destroy_all_layouts(clp);
1815         return 0;
1816 }
1817
1818 /*
1819  * Returns zero or a negative errno.  NFS4ERR values are converted
1820  * to local errno values.
1821  */
1822 static int nfs4_reclaim_lease(struct nfs_client *clp)
1823 {
1824         int status;
1825
1826         status = nfs4_establish_lease(clp);
1827         if (status < 0)
1828                 return nfs4_handle_reclaim_lease_error(clp, status);
1829         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1830                 nfs4_state_start_reclaim_nograce(clp);
1831         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1832                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1833         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1834         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1835         return 0;
1836 }
1837
1838 static int nfs4_purge_lease(struct nfs_client *clp)
1839 {
1840         int status;
1841
1842         status = nfs4_establish_lease(clp);
1843         if (status < 0)
1844                 return nfs4_handle_reclaim_lease_error(clp, status);
1845         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1846         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1847         nfs4_state_start_reclaim_nograce(clp);
1848         return 0;
1849 }
1850
1851 /**
1852  * nfs4_discover_server_trunking - Detect server IP address trunking
1853  *
1854  * @clp: nfs_client under test
1855  * @result: OUT: found nfs_client, or clp
1856  *
1857  * Returns zero or a negative errno.  If zero is returned,
1858  * an nfs_client pointer is planted in "result".
1859  *
1860  * Note: since we are invoked in process context, and
1861  * not from inside the state manager, we cannot use
1862  * nfs4_handle_reclaim_lease_error().
1863  */
1864 int nfs4_discover_server_trunking(struct nfs_client *clp,
1865                                   struct nfs_client **result)
1866 {
1867         const struct nfs4_state_recovery_ops *ops =
1868                                 clp->cl_mvops->reboot_recovery_ops;
1869         rpc_authflavor_t *flavors, flav, save;
1870         struct rpc_clnt *clnt;
1871         struct rpc_cred *cred;
1872         int i, len, status;
1873
1874         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1875
1876         len = NFS_MAX_SECFLAVORS;
1877         flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1878         if (flavors == NULL) {
1879                 status = -ENOMEM;
1880                 goto out;
1881         }
1882         len = rpcauth_list_flavors(flavors, len);
1883         if (len < 0) {
1884                 status = len;
1885                 goto out_free;
1886         }
1887         clnt = clp->cl_rpcclient;
1888         save = clnt->cl_auth->au_flavor;
1889         i = 0;
1890
1891         mutex_lock(&nfs_clid_init_mutex);
1892         status  = -ENOENT;
1893 again:
1894         cred = ops->get_clid_cred(clp);
1895         if (cred == NULL)
1896                 goto out_unlock;
1897
1898         status = ops->detect_trunking(clp, result, cred);
1899         put_rpccred(cred);
1900         switch (status) {
1901         case 0:
1902                 break;
1903
1904         case -EACCES:
1905                 if (clp->cl_machine_cred == NULL)
1906                         break;
1907                 /* Handle case where the user hasn't set up machine creds */
1908                 nfs4_clear_machine_cred(clp);
1909         case -NFS4ERR_DELAY:
1910         case -ETIMEDOUT:
1911         case -EAGAIN:
1912                 ssleep(1);
1913         case -NFS4ERR_STALE_CLIENTID:
1914                 dprintk("NFS: %s after status %d, retrying\n",
1915                         __func__, status);
1916                 goto again;
1917
1918         case -NFS4ERR_CLID_INUSE:
1919         case -NFS4ERR_WRONGSEC:
1920                 status = -EPERM;
1921                 if (i >= len)
1922                         break;
1923
1924                 flav = flavors[i++];
1925                 if (flav == save)
1926                         flav = flavors[i++];
1927                 clnt = rpc_clone_client_set_auth(clnt, flav);
1928                 if (IS_ERR(clnt)) {
1929                         status = PTR_ERR(clnt);
1930                         break;
1931                 }
1932                 clp->cl_rpcclient = clnt;
1933                 goto again;
1934
1935         case -NFS4ERR_MINOR_VERS_MISMATCH:
1936                 status = -EPROTONOSUPPORT;
1937                 break;
1938
1939         case -EKEYEXPIRED:
1940         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1941                                  * in nfs4_exchange_id */
1942                 status = -EKEYEXPIRED;
1943         }
1944
1945 out_unlock:
1946         mutex_unlock(&nfs_clid_init_mutex);
1947 out_free:
1948         kfree(flavors);
1949 out:
1950         dprintk("NFS: %s: status = %d\n", __func__, status);
1951         return status;
1952 }
1953
1954 #ifdef CONFIG_NFS_V4_1
1955 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1956 {
1957         struct nfs_client *clp = session->clp;
1958
1959         switch (err) {
1960         default:
1961                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1962                 break;
1963         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1964                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1965         }
1966         nfs4_schedule_lease_recovery(clp);
1967 }
1968 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1969
1970 static void nfs41_ping_server(struct nfs_client *clp)
1971 {
1972         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1973         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1974         nfs4_schedule_state_manager(clp);
1975 }
1976
1977 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1978 {
1979         nfs41_ping_server(clp);
1980 }
1981
1982 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1983 {
1984         nfs41_ping_server(clp);
1985 }
1986
1987 static void nfs4_reset_all_state(struct nfs_client *clp)
1988 {
1989         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1990                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1991                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1992                 nfs4_state_start_reclaim_nograce(clp);
1993                 dprintk("%s: scheduling reset of all state for server %s!\n",
1994                                 __func__, clp->cl_hostname);
1995                 nfs4_schedule_state_manager(clp);
1996         }
1997 }
1998
1999 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2000 {
2001         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2002                 nfs4_state_start_reclaim_reboot(clp);
2003                 dprintk("%s: server %s rebooted!\n", __func__,
2004                                 clp->cl_hostname);
2005                 nfs4_schedule_state_manager(clp);
2006         }
2007 }
2008
2009 static void nfs41_handle_state_revoked(struct nfs_client *clp)
2010 {
2011         nfs4_reset_all_state(clp);
2012         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2013 }
2014
2015 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2016 {
2017         /* This will need to handle layouts too */
2018         nfs_expire_all_delegations(clp);
2019         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2020                         clp->cl_hostname);
2021 }
2022
2023 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2024 {
2025         nfs_expire_all_delegations(clp);
2026         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2027                 nfs4_schedule_state_manager(clp);
2028         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2029                         clp->cl_hostname);
2030 }
2031
2032 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2033 {
2034         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2035                 &clp->cl_state) == 0)
2036                 nfs4_schedule_state_manager(clp);
2037 }
2038
2039 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2040 {
2041         if (!flags)
2042                 return;
2043
2044         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2045                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2046
2047         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2048                 nfs41_handle_server_reboot(clp);
2049         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2050                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2051                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2052                             SEQ4_STATUS_LEASE_MOVED))
2053                 nfs41_handle_state_revoked(clp);
2054         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2055                 nfs41_handle_recallable_state_revoked(clp);
2056         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2057                 nfs41_handle_backchannel_fault(clp);
2058         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2059                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2060                 nfs41_handle_cb_path_down(clp);
2061 }
2062
2063 static int nfs4_reset_session(struct nfs_client *clp)
2064 {
2065         struct rpc_cred *cred;
2066         int status;
2067
2068         if (!nfs4_has_session(clp))
2069                 return 0;
2070         nfs4_begin_drain_session(clp);
2071         cred = nfs4_get_exchange_id_cred(clp);
2072         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2073         switch (status) {
2074         case 0:
2075         case -NFS4ERR_BADSESSION:
2076         case -NFS4ERR_DEADSESSION:
2077                 break;
2078         case -NFS4ERR_BACK_CHAN_BUSY:
2079         case -NFS4ERR_DELAY:
2080                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2081                 status = 0;
2082                 ssleep(1);
2083                 goto out;
2084         default:
2085                 status = nfs4_recovery_handle_error(clp, status);
2086                 goto out;
2087         }
2088
2089         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2090         status = nfs4_proc_create_session(clp, cred);
2091         if (status) {
2092                 dprintk("%s: session reset failed with status %d for server %s!\n",
2093                         __func__, status, clp->cl_hostname);
2094                 status = nfs4_handle_reclaim_lease_error(clp, status);
2095                 goto out;
2096         }
2097         nfs41_finish_session_reset(clp);
2098         dprintk("%s: session reset was successful for server %s!\n",
2099                         __func__, clp->cl_hostname);
2100 out:
2101         if (cred)
2102                 put_rpccred(cred);
2103         return status;
2104 }
2105
2106 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2107 {
2108         struct rpc_cred *cred;
2109         int ret;
2110
2111         if (!nfs4_has_session(clp))
2112                 return 0;
2113         nfs4_begin_drain_session(clp);
2114         cred = nfs4_get_exchange_id_cred(clp);
2115         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2116         if (cred)
2117                 put_rpccred(cred);
2118         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2119         switch (ret) {
2120         case 0:
2121                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2122                         __func__, clp->cl_hostname);
2123                 break;
2124         case -NFS4ERR_DELAY:
2125                 ssleep(1);
2126                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2127                 break;
2128         default:
2129                 return nfs4_recovery_handle_error(clp, ret);
2130         }
2131         return 0;
2132 }
2133 #else /* CONFIG_NFS_V4_1 */
2134 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2135 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2136
2137 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2138 {
2139         return 0;
2140 }
2141 #endif /* CONFIG_NFS_V4_1 */
2142
2143 static void nfs4_state_manager(struct nfs_client *clp)
2144 {
2145         int status = 0;
2146         const char *section = "", *section_sep = "";
2147
2148         /* Ensure exclusive access to NFSv4 state */
2149         do {
2150                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2151                         section = "purge state";
2152                         status = nfs4_purge_lease(clp);
2153                         if (status < 0)
2154                                 goto out_error;
2155                         continue;
2156                 }
2157
2158                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2159                         section = "lease expired";
2160                         /* We're going to have to re-establish a clientid */
2161                         status = nfs4_reclaim_lease(clp);
2162                         if (status < 0)
2163                                 goto out_error;
2164                         continue;
2165                 }
2166
2167                 /* Initialize or reset the session */
2168                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2169                         section = "reset session";
2170                         status = nfs4_reset_session(clp);
2171                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2172                                 continue;
2173                         if (status < 0)
2174                                 goto out_error;
2175                 }
2176
2177                 /* Send BIND_CONN_TO_SESSION */
2178                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2179                                 &clp->cl_state)) {
2180                         section = "bind conn to session";
2181                         status = nfs4_bind_conn_to_session(clp);
2182                         if (status < 0)
2183                                 goto out_error;
2184                         continue;
2185                 }
2186
2187                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2188                         section = "check lease";
2189                         status = nfs4_check_lease(clp);
2190                         if (status < 0)
2191                                 goto out_error;
2192                         continue;
2193                 }
2194
2195                 /* First recover reboot state... */
2196                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2197                         section = "reclaim reboot";
2198                         status = nfs4_do_reclaim(clp,
2199                                 clp->cl_mvops->reboot_recovery_ops);
2200                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2201                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2202                                 continue;
2203                         nfs4_state_end_reclaim_reboot(clp);
2204                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2205                                 continue;
2206                         if (status < 0)
2207                                 goto out_error;
2208                 }
2209
2210                 /* Now recover expired state... */
2211                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2212                         section = "reclaim nograce";
2213                         status = nfs4_do_reclaim(clp,
2214                                 clp->cl_mvops->nograce_recovery_ops);
2215                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2216                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2217                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2218                                 continue;
2219                         if (status < 0)
2220                                 goto out_error;
2221                 }
2222
2223                 nfs4_end_drain_session(clp);
2224                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2225                         nfs_client_return_marked_delegations(clp);
2226                         continue;
2227                 }
2228
2229                 nfs4_clear_state_manager_bit(clp);
2230                 /* Did we race with an attempt to give us more work? */
2231                 if (clp->cl_state == 0)
2232                         break;
2233                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2234                         break;
2235         } while (atomic_read(&clp->cl_count) > 1);
2236         return;
2237 out_error:
2238         if (strlen(section))
2239                 section_sep = ": ";
2240         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2241                         " with error %d\n", section_sep, section,
2242                         clp->cl_hostname, -status);
2243         ssleep(1);
2244         nfs4_end_drain_session(clp);
2245         nfs4_clear_state_manager_bit(clp);
2246 }
2247
2248 static int nfs4_run_state_manager(void *ptr)
2249 {
2250         struct nfs_client *clp = ptr;
2251
2252         allow_signal(SIGKILL);
2253         nfs4_state_manager(clp);
2254         nfs_put_client(clp);
2255         module_put_and_exit(0);
2256         return 0;
2257 }
2258
2259 /*
2260  * Local variables:
2261  *  c-basic-offset: 8
2262  * End:
2263  */