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