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_SLOT_TBL_DRAINING, &tbl->slot_tbl_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_slot_tbl_drain_complete(struct nfs4_slot_table *tbl)
 255 {
 256         if (nfs4_slot_tbl_draining(tbl))
 257                 complete(&tbl->complete);
 258 }
 259
 260 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 261 {
 262         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 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         /* back channel */
 279         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 280         if (ret)
 281                 return ret;
 282         /* fore channel */
 283         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 284 }
 285
 286 static void nfs41_finish_session_reset(struct nfs_client *clp)
 287 {
 288         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 289         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 290         /* create_session negotiated new slot table */
 291         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 292         nfs41_setup_state_renewal(clp);
 293 }
 294
 295 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 296 {
 297         int status;
 298
 299         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 300                 goto do_confirm;
 301         nfs4_begin_drain_session(clp);
 302         status = nfs4_proc_exchange_id(clp, cred);
 303         if (status != 0)
 304                 goto out;
 305         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 306 do_confirm:
 307         status = nfs4_proc_create_session(clp, cred);
 308         if (status != 0)
 309                 goto out;
 310         nfs41_finish_session_reset(clp);
 311         nfs_mark_client_ready(clp, NFS_CS_READY);
 312 out:
 313         return status;
 314 }
 315
 316 /**
 317  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 318  *
 319  * @clp: nfs_client under test
 320  * @result: OUT: found nfs_client, or clp
 321  * @cred: credential to use for trunking test
 322  *
 323  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 324  * If NFS4_OK is returned, an nfs_client pointer is planted in
 325  * "result".
 326  *
 327  * Note: The returned client may not yet be marked ready.
 328  */
 329 int nfs41_discover_server_trunking(struct nfs_client *clp,
 330                                    struct nfs_client **result,
 331                                    struct rpc_cred *cred)
 332 {
 333         int status;
 334
 335         status = nfs4_proc_exchange_id(clp, cred);
 336         if (status != NFS4_OK)
 337                 return status;
 338         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 339
 340         return nfs41_walk_client_list(clp, result, cred);
 341 }
 342
 343 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 344 {
 345         struct rpc_cred *cred;
 346
 347         spin_lock(&clp->cl_lock);
 348         cred = nfs4_get_machine_cred_locked(clp);
 349         spin_unlock(&clp->cl_lock);
 350         return cred;
 351 }
 352
 353 #endif /* CONFIG_NFS_V4_1 */
 354
 355 static struct rpc_cred *
 356 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 357 {
 358         struct nfs_client *clp = server->nfs_client;
 359         struct rpc_cred *cred = NULL;
 360         struct nfs4_state_owner *sp;
 361         struct rb_node *pos;
 362
 363         spin_lock(&clp->cl_lock);
 364         pos = rb_first(&server->state_owners);
 365         if (pos != NULL) {
 366                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 367                 cred = get_rpccred(sp->so_cred);
 368         }
 369         spin_unlock(&clp->cl_lock);
 370         return cred;
 371 }
 372
 373 /**
 374  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 375  * @clp: client state handle
 376  *
 377  * Returns an rpc_cred with reference count bumped, or NULL.
 378  */
 379 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 380 {
 381         struct nfs_server *server;
 382         struct rpc_cred *cred;
 383
 384         spin_lock(&clp->cl_lock);
 385         cred = nfs4_get_machine_cred_locked(clp);
 386         spin_unlock(&clp->cl_lock);
 387         if (cred != NULL)
 388                 goto out;
 389
 390         rcu_read_lock();
 391         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 392                 cred = nfs4_get_setclientid_cred_server(server);
 393                 if (cred != NULL)
 394                         break;
 395         }
 396         rcu_read_unlock();
 397
 398 out:
 399         return cred;
 400 }
 401
 402 static struct nfs4_state_owner *
 403 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 404 {
 405         struct rb_node **p = &server->state_owners.rb_node,
 406                        *parent = NULL;
 407         struct nfs4_state_owner *sp;
 408
 409         while (*p != NULL) {
 410                 parent = *p;
 411                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 412
 413                 if (cred < sp->so_cred)
 414                         p = &parent->rb_left;
 415                 else if (cred > sp->so_cred)
 416                         p = &parent->rb_right;
 417                 else {
 418                         if (!list_empty(&sp->so_lru))
 419                                 list_del_init(&sp->so_lru);
 420                         atomic_inc(&sp->so_count);
 421                         return sp;
 422                 }
 423         }
 424         return NULL;
 425 }
 426
 427 static struct nfs4_state_owner *
 428 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 429 {
 430         struct nfs_server *server = new->so_server;
 431         struct rb_node **p = &server->state_owners.rb_node,
 432                        *parent = NULL;
 433         struct nfs4_state_owner *sp;
 434         int err;
 435
 436         while (*p != NULL) {
 437                 parent = *p;
 438                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 439
 440                 if (new->so_cred < sp->so_cred)
 441                         p = &parent->rb_left;
 442                 else if (new->so_cred > sp->so_cred)
 443                         p = &parent->rb_right;
 444                 else {
 445                         if (!list_empty(&sp->so_lru))
 446                                 list_del_init(&sp->so_lru);
 447                         atomic_inc(&sp->so_count);
 448                         return sp;
 449                 }
 450         }
 451         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 452         if (err)
 453                 return ERR_PTR(err);
 454         rb_link_node(&new->so_server_node, parent, p);
 455         rb_insert_color(&new->so_server_node, &server->state_owners);
 456         return new;
 457 }
 458
 459 static void
 460 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 461 {
 462         struct nfs_server *server = sp->so_server;
 463
 464         if (!RB_EMPTY_NODE(&sp->so_server_node))
 465                 rb_erase(&sp->so_server_node, &server->state_owners);
 466         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 467 }
 468
 469 static void
 470 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 471 {
 472         sc->create_time = ktime_get();
 473         sc->flags = 0;
 474         sc->counter = 0;
 475         spin_lock_init(&sc->lock);
 476         INIT_LIST_HEAD(&sc->list);
 477         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 478 }
 479
 480 static void
 481 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 482 {
 483         rpc_destroy_wait_queue(&sc->wait);
 484 }
 485
 486 /*
 487  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 488  * create a new state_owner.
 489  *
 490  */
 491 static struct nfs4_state_owner *
 492 nfs4_alloc_state_owner(struct nfs_server *server,
 493                 struct rpc_cred *cred,
 494                 gfp_t gfp_flags)
 495 {
 496         struct nfs4_state_owner *sp;
 497
 498         sp = kzalloc(sizeof(*sp), gfp_flags);
 499         if (!sp)
 500                 return NULL;
 501         sp->so_server = server;
 502         sp->so_cred = get_rpccred(cred);
 503         spin_lock_init(&sp->so_lock);
 504         INIT_LIST_HEAD(&sp->so_states);
 505         nfs4_init_seqid_counter(&sp->so_seqid);
 506         atomic_set(&sp->so_count, 1);
 507         INIT_LIST_HEAD(&sp->so_lru);
 508         seqcount_init(&sp->so_reclaim_seqcount);
 509         mutex_init(&sp->so_delegreturn_mutex);
 510         return sp;
 511 }
 512
 513 static void
 514 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 515 {
 516         struct rb_node *rb_node = &sp->so_server_node;
 517
 518         if (!RB_EMPTY_NODE(rb_node)) {
 519                 struct nfs_server *server = sp->so_server;
 520                 struct nfs_client *clp = server->nfs_client;
 521
 522                 spin_lock(&clp->cl_lock);
 523                 if (!RB_EMPTY_NODE(rb_node)) {
 524                         rb_erase(rb_node, &server->state_owners);
 525                         RB_CLEAR_NODE(rb_node);
 526                 }
 527                 spin_unlock(&clp->cl_lock);
 528         }
 529 }
 530
 531 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 532 {
 533         nfs4_destroy_seqid_counter(&sp->so_seqid);
 534         put_rpccred(sp->so_cred);
 535         kfree(sp);
 536 }
 537
 538 static void nfs4_gc_state_owners(struct nfs_server *server)
 539 {
 540         struct nfs_client *clp = server->nfs_client;
 541         struct nfs4_state_owner *sp, *tmp;
 542         unsigned long time_min, time_max;
 543         LIST_HEAD(doomed);
 544
 545         spin_lock(&clp->cl_lock);
 546         time_max = jiffies;
 547         time_min = (long)time_max - (long)clp->cl_lease_time;
 548         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 549                 /* NB: LRU is sorted so that oldest is at the head */
 550                 if (time_in_range(sp->so_expires, time_min, time_max))
 551                         break;
 552                 list_move(&sp->so_lru, &doomed);
 553                 nfs4_remove_state_owner_locked(sp);
 554         }
 555         spin_unlock(&clp->cl_lock);
 556
 557         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 558                 list_del(&sp->so_lru);
 559                 nfs4_free_state_owner(sp);
 560         }
 561 }
 562
 563 /**
 564  * nfs4_get_state_owner - Look up a state owner given a credential
 565  * @server: nfs_server to search
 566  * @cred: RPC credential to match
 567  *
 568  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 569  */
 570 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 571                                               struct rpc_cred *cred,
 572                                               gfp_t gfp_flags)
 573 {
 574         struct nfs_client *clp = server->nfs_client;
 575         struct nfs4_state_owner *sp, *new;
 576
 577         spin_lock(&clp->cl_lock);
 578         sp = nfs4_find_state_owner_locked(server, cred);
 579         spin_unlock(&clp->cl_lock);
 580         if (sp != NULL)
 581                 goto out;
 582         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 583         if (new == NULL)
 584                 goto out;
 585         do {
 586                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 587                         break;
 588                 spin_lock(&clp->cl_lock);
 589                 sp = nfs4_insert_state_owner_locked(new);
 590                 spin_unlock(&clp->cl_lock);
 591         } while (sp == ERR_PTR(-EAGAIN));
 592         if (sp != new)
 593                 nfs4_free_state_owner(new);
 594 out:
 595         nfs4_gc_state_owners(server);
 596         return sp;
 597 }
 598
 599 /**
 600  * nfs4_put_state_owner - Release a nfs4_state_owner
 601  * @sp: state owner data to release
 602  *
 603  * Note that we keep released state owners on an LRU
 604  * list.
 605  * This caches valid state owners so that they can be
 606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 607  * It also pins the uniquifier of dropped state owners for
 608  * a while, to ensure that those state owner names are
 609  * never reused.
 610  */
 611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 612 {
 613         struct nfs_server *server = sp->so_server;
 614         struct nfs_client *clp = server->nfs_client;
 615
 616         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 617                 return;
 618
 619         sp->so_expires = jiffies;
 620         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 621         spin_unlock(&clp->cl_lock);
 622 }
 623
 624 /**
 625  * nfs4_purge_state_owners - Release all cached state owners
 626  * @server: nfs_server with cached state owners to release
 627  *
 628  * Called at umount time.  Remaining state owners will be on
 629  * the LRU with ref count of zero.
 630  */
 631 void nfs4_purge_state_owners(struct nfs_server *server)
 632 {
 633         struct nfs_client *clp = server->nfs_client;
 634         struct nfs4_state_owner *sp, *tmp;
 635         LIST_HEAD(doomed);
 636
 637         spin_lock(&clp->cl_lock);
 638         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 639                 list_move(&sp->so_lru, &doomed);
 640                 nfs4_remove_state_owner_locked(sp);
 641         }
 642         spin_unlock(&clp->cl_lock);
 643
 644         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 645                 list_del(&sp->so_lru);
 646                 nfs4_free_state_owner(sp);
 647         }
 648 }
 649
 650 static struct nfs4_state *
 651 nfs4_alloc_open_state(void)
 652 {
 653         struct nfs4_state *state;
 654
 655         state = kzalloc(sizeof(*state), GFP_NOFS);
 656         if (!state)
 657                 return NULL;
 658         atomic_set(&state->count, 1);
 659         INIT_LIST_HEAD(&state->lock_states);
 660         spin_lock_init(&state->state_lock);
 661         seqlock_init(&state->seqlock);
 662         return state;
 663 }
 664
 665 void
 666 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 667 {
 668         if (state->state == fmode)
 669                 return;
 670         /* NB! List reordering - see the reclaim code for why.  */
 671         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 672                 if (fmode & FMODE_WRITE)
 673                         list_move(&state->open_states, &state->owner->so_states);
 674                 else
 675                         list_move_tail(&state->open_states, &state->owner->so_states);
 676         }
 677         state->state = fmode;
 678 }
 679
 680 static struct nfs4_state *
 681 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 682 {
 683         struct nfs_inode *nfsi = NFS_I(inode);
 684         struct nfs4_state *state;
 685
 686         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 687                 if (state->owner != owner)
 688                         continue;
 689                 if (!nfs4_valid_open_stateid(state))
 690                         continue;
 691                 if (atomic_inc_not_zero(&state->count))
 692                         return state;
 693         }
 694         return NULL;
 695 }
 696
 697 static void
 698 nfs4_free_open_state(struct nfs4_state *state)
 699 {
 700         kfree(state);
 701 }
 702
 703 struct nfs4_state *
 704 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 705 {
 706         struct nfs4_state *state, *new;
 707         struct nfs_inode *nfsi = NFS_I(inode);
 708
 709         spin_lock(&inode->i_lock);
 710         state = __nfs4_find_state_byowner(inode, owner);
 711         spin_unlock(&inode->i_lock);
 712         if (state)
 713                 goto out;
 714         new = nfs4_alloc_open_state();
 715         spin_lock(&owner->so_lock);
 716         spin_lock(&inode->i_lock);
 717         state = __nfs4_find_state_byowner(inode, owner);
 718         if (state == NULL && new != NULL) {
 719                 state = new;
 720                 state->owner = owner;
 721                 atomic_inc(&owner->so_count);
 722                 list_add(&state->inode_states, &nfsi->open_states);
 723                 ihold(inode);
 724                 state->inode = inode;
 725                 spin_unlock(&inode->i_lock);
 726                 /* Note: The reclaim code dictates that we add stateless
 727                  * and read-only stateids to the end of the list */
 728                 list_add_tail(&state->open_states, &owner->so_states);
 729                 spin_unlock(&owner->so_lock);
 730         } else {
 731                 spin_unlock(&inode->i_lock);
 732                 spin_unlock(&owner->so_lock);
 733                 if (new)
 734                         nfs4_free_open_state(new);
 735         }
 736 out:
 737         return state;
 738 }
 739
 740 void nfs4_put_open_state(struct nfs4_state *state)
 741 {
 742         struct inode *inode = state->inode;
 743         struct nfs4_state_owner *owner = state->owner;
 744
 745         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 746                 return;
 747         spin_lock(&inode->i_lock);
 748         list_del(&state->inode_states);
 749         list_del(&state->open_states);
 750         spin_unlock(&inode->i_lock);
 751         spin_unlock(&owner->so_lock);
 752         iput(inode);
 753         nfs4_free_open_state(state);
 754         nfs4_put_state_owner(owner);
 755 }
 756
 757 /*
 758  * Close the current file.
 759  */
 760 static void __nfs4_close(struct nfs4_state *state,
 761                 fmode_t fmode, gfp_t gfp_mask, int wait)
 762 {
 763         struct nfs4_state_owner *owner = state->owner;
 764         int call_close = 0;
 765         fmode_t newstate;
 766
 767         atomic_inc(&owner->so_count);
 768         /* Protect against nfs4_find_state() */
 769         spin_lock(&owner->so_lock);
 770         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 771                 case FMODE_READ:
 772                         state->n_rdonly--;
 773                         break;
 774                 case FMODE_WRITE:
 775                         state->n_wronly--;
 776                         break;
 777                 case FMODE_READ|FMODE_WRITE:
 778                         state->n_rdwr--;
 779         }
 780         newstate = FMODE_READ|FMODE_WRITE;
 781         if (state->n_rdwr == 0) {
 782                 if (state->n_rdonly == 0) {
 783                         newstate &= ~FMODE_READ;
 784                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 785                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 786                 }
 787                 if (state->n_wronly == 0) {
 788                         newstate &= ~FMODE_WRITE;
 789                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 790                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 791                 }
 792                 if (newstate == 0)
 793                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 794         }
 795         nfs4_state_set_mode_locked(state, newstate);
 796         spin_unlock(&owner->so_lock);
 797
 798         if (!call_close) {
 799                 nfs4_put_open_state(state);
 800                 nfs4_put_state_owner(owner);
 801         } else
 802                 nfs4_do_close(state, gfp_mask, wait);
 803 }
 804
 805 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 806 {
 807         __nfs4_close(state, fmode, GFP_NOFS, 0);
 808 }
 809
 810 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 811 {
 812         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 813 }
 814
 815 /*
 816  * Search the state->lock_states for an existing lock_owner
 817  * that is compatible with current->files
 818  */
 819 static struct nfs4_lock_state *
 820 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 821 {
 822         struct nfs4_lock_state *pos;
 823         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 824                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 825                         continue;
 826                 switch (pos->ls_owner.lo_type) {
 827                 case NFS4_POSIX_LOCK_TYPE:
 828                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 829                                 continue;
 830                         break;
 831                 case NFS4_FLOCK_LOCK_TYPE:
 832                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 833                                 continue;
 834                 }
 835                 atomic_inc(&pos->ls_count);
 836                 return pos;
 837         }
 838         return NULL;
 839 }
 840
 841 /*
 842  * Return a compatible lock_state. If no initialized lock_state structure
 843  * exists, return an uninitialized one.
 844  *
 845  */
 846 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)
 847 {
 848         struct nfs4_lock_state *lsp;
 849         struct nfs_server *server = state->owner->so_server;
 850
 851         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 852         if (lsp == NULL)
 853                 return NULL;
 854         nfs4_init_seqid_counter(&lsp->ls_seqid);
 855         atomic_set(&lsp->ls_count, 1);
 856         lsp->ls_state = state;
 857         lsp->ls_owner.lo_type = type;
 858         switch (lsp->ls_owner.lo_type) {
 859         case NFS4_FLOCK_LOCK_TYPE:
 860                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 861                 break;
 862         case NFS4_POSIX_LOCK_TYPE:
 863                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 864                 break;
 865         default:
 866                 goto out_free;
 867         }
 868         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 869         if (lsp->ls_seqid.owner_id < 0)
 870                 goto out_free;
 871         INIT_LIST_HEAD(&lsp->ls_locks);
 872         return lsp;
 873 out_free:
 874         kfree(lsp);
 875         return NULL;
 876 }
 877
 878 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 879 {
 880         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 881         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 882         kfree(lsp);
 883 }
 884
 885 /*
 886  * Return a compatible lock_state. If no initialized lock_state structure
 887  * exists, return an uninitialized one.
 888  *
 889  */
 890 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 891 {
 892         struct nfs4_lock_state *lsp, *new = NULL;
 893
 894         for(;;) {
 895                 spin_lock(&state->state_lock);
 896                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 897                 if (lsp != NULL)
 898                         break;
 899                 if (new != NULL) {
 900                         list_add(&new->ls_locks, &state->lock_states);
 901                         set_bit(LK_STATE_IN_USE, &state->flags);
 902                         lsp = new;
 903                         new = NULL;
 904                         break;
 905                 }
 906                 spin_unlock(&state->state_lock);
 907                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 908                 if (new == NULL)
 909                         return NULL;
 910         }
 911         spin_unlock(&state->state_lock);
 912         if (new != NULL)
 913                 nfs4_free_lock_state(state->owner->so_server, new);
 914         return lsp;
 915 }
 916
 917 /*
 918  * Release reference to lock_state, and free it if we see that
 919  * it is no longer in use
 920  */
 921 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 922 {
 923         struct nfs_server *server;
 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         server = state->owner->so_server;
 936         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 937                 struct nfs_client *clp = server->nfs_client;
 938
 939                 clp->cl_mvops->free_lock_state(server, lsp);
 940         } else
 941                 nfs4_free_lock_state(server, lsp);
 942 }
 943
 944 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 945 {
 946         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 947
 948         dst->fl_u.nfs4_fl.owner = lsp;
 949         atomic_inc(&lsp->ls_count);
 950 }
 951
 952 static void nfs4_fl_release_lock(struct file_lock *fl)
 953 {
 954         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 955 }
 956
 957 static const struct file_lock_operations nfs4_fl_lock_ops = {
 958         .fl_copy_lock = nfs4_fl_copy_lock,
 959         .fl_release_private = nfs4_fl_release_lock,
 960 };
 961
 962 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 963 {
 964         struct nfs4_lock_state *lsp;
 965
 966         if (fl->fl_ops != NULL)
 967                 return 0;
 968         if (fl->fl_flags & FL_POSIX)
 969                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 970         else if (fl->fl_flags & FL_FLOCK)
 971                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 972                                 NFS4_FLOCK_LOCK_TYPE);
 973         else
 974                 return -EINVAL;
 975         if (lsp == NULL)
 976                 return -ENOMEM;
 977         fl->fl_u.nfs4_fl.owner = lsp;
 978         fl->fl_ops = &nfs4_fl_lock_ops;
 979         return 0;
 980 }
 981
 982 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 983                 struct nfs4_state *state,
 984                 const struct nfs_lockowner *lockowner)
 985 {
 986         struct nfs4_lock_state *lsp;
 987         fl_owner_t fl_owner;
 988         pid_t fl_pid;
 989         int ret = -ENOENT;
 990
 991
 992         if (lockowner == NULL)
 993                 goto out;
 994
 995         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 996                 goto out;
 997
 998         fl_owner = lockowner->l_owner;
 999         fl_pid = lockowner->l_pid;
1000         spin_lock(&state->state_lock);
1001         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1002         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1003                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1004                 ret = 0;
1005                 smp_rmb();
1006                 if (!list_empty(&lsp->ls_seqid.list))
1007                         ret = -EWOULDBLOCK;
1008         }
1009         spin_unlock(&state->state_lock);
1010         nfs4_put_lock_state(lsp);
1011 out:
1012         return ret;
1013 }
1014
1015 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1016 {
1017         const nfs4_stateid *src;
1018         int ret;
1019         int seq;
1020
1021         do {
1022                 src = &zero_stateid;
1023                 seq = read_seqbegin(&state->seqlock);
1024                 if (test_bit(NFS_OPEN_STATE, &state->flags))
1025                         src = &state->open_stateid;
1026                 nfs4_stateid_copy(dst, src);
1027                 ret = 0;
1028                 smp_rmb();
1029                 if (!list_empty(&state->owner->so_seqid.list))
1030                         ret = -EWOULDBLOCK;
1031         } while (read_seqretry(&state->seqlock, seq));
1032         return ret;
1033 }
1034
1035 /*
1036  * Byte-range lock aware utility to initialize the stateid of read/write
1037  * requests.
1038  */
1039 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1040                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1041 {
1042         int ret = 0;
1043         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1044                 goto out;
1045         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1046         if (ret != -ENOENT)
1047                 goto out;
1048         ret = nfs4_copy_open_stateid(dst, state);
1049 out:
1050         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1051                 dst->seqid = 0;
1052         return ret;
1053 }
1054
1055 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1056 {
1057         struct nfs_seqid *new;
1058
1059         new = kmalloc(sizeof(*new), gfp_mask);
1060         if (new != NULL) {
1061                 new->sequence = counter;
1062                 INIT_LIST_HEAD(&new->list);
1063                 new->task = NULL;
1064         }
1065         return new;
1066 }
1067
1068 void nfs_release_seqid(struct nfs_seqid *seqid)
1069 {
1070         struct nfs_seqid_counter *sequence;
1071
1072         if (list_empty(&seqid->list))
1073                 return;
1074         sequence = seqid->sequence;
1075         spin_lock(&sequence->lock);
1076         list_del_init(&seqid->list);
1077         if (!list_empty(&sequence->list)) {
1078                 struct nfs_seqid *next;
1079
1080                 next = list_first_entry(&sequence->list,
1081                                 struct nfs_seqid, list);
1082                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1083         }
1084         spin_unlock(&sequence->lock);
1085 }
1086
1087 void nfs_free_seqid(struct nfs_seqid *seqid)
1088 {
1089         nfs_release_seqid(seqid);
1090         kfree(seqid);
1091 }
1092
1093 /*
1094  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1095  * failed with a seqid incrementing error -
1096  * see comments nfs_fs.h:seqid_mutating_error()
1097  */
1098 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1099 {
1100         switch (status) {
1101                 case 0:
1102                         break;
1103                 case -NFS4ERR_BAD_SEQID:
1104                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1105                                 return;
1106                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1107                                         " sequence-id error on an"
1108                                         " unconfirmed sequence %p!\n",
1109                                         seqid->sequence);
1110                 case -NFS4ERR_STALE_CLIENTID:
1111                 case -NFS4ERR_STALE_STATEID:
1112                 case -NFS4ERR_BAD_STATEID:
1113                 case -NFS4ERR_BADXDR:
1114                 case -NFS4ERR_RESOURCE:
1115                 case -NFS4ERR_NOFILEHANDLE:
1116                         /* Non-seqid mutating errors */
1117                         return;
1118         };
1119         /*
1120          * Note: no locking needed as we are guaranteed to be first
1121          * on the sequence list
1122          */
1123         seqid->sequence->counter++;
1124 }
1125
1126 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1127 {
1128         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1129                                         struct nfs4_state_owner, so_seqid);
1130         struct nfs_server *server = sp->so_server;
1131
1132         if (status == -NFS4ERR_BAD_SEQID)
1133                 nfs4_drop_state_owner(sp);
1134         if (!nfs4_has_session(server->nfs_client))
1135                 nfs_increment_seqid(status, seqid);
1136 }
1137
1138 /*
1139  * Increment the seqid if the LOCK/LOCKU succeeded, or
1140  * failed with a seqid incrementing error -
1141  * see comments nfs_fs.h:seqid_mutating_error()
1142  */
1143 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1144 {
1145         nfs_increment_seqid(status, seqid);
1146 }
1147
1148 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1149 {
1150         struct nfs_seqid_counter *sequence = seqid->sequence;
1151         int status = 0;
1152
1153         spin_lock(&sequence->lock);
1154         seqid->task = task;
1155         if (list_empty(&seqid->list))
1156                 list_add_tail(&seqid->list, &sequence->list);
1157         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1158                 goto unlock;
1159         rpc_sleep_on(&sequence->wait, task, NULL);
1160         status = -EAGAIN;
1161 unlock:
1162         spin_unlock(&sequence->lock);
1163         return status;
1164 }
1165
1166 static int nfs4_run_state_manager(void *);
1167
1168 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1169 {
1170         smp_mb__before_clear_bit();
1171         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1172         smp_mb__after_clear_bit();
1173         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1174         rpc_wake_up(&clp->cl_rpcwaitq);
1175 }
1176
1177 /*
1178  * Schedule the nfs_client asynchronous state management routine
1179  */
1180 void nfs4_schedule_state_manager(struct nfs_client *clp)
1181 {
1182         struct task_struct *task;
1183         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1184
1185         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1186                 return;
1187         __module_get(THIS_MODULE);
1188         atomic_inc(&clp->cl_count);
1189
1190         /* The rcu_read_lock() is not strictly necessary, as the state
1191          * manager is the only thread that ever changes the rpc_xprt
1192          * after it's initialized.  At this point, we're single threaded. */
1193         rcu_read_lock();
1194         snprintf(buf, sizeof(buf), "%s-manager",
1195                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1196         rcu_read_unlock();
1197         task = kthread_run(nfs4_run_state_manager, clp, buf);
1198         if (IS_ERR(task)) {
1199                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1200                         __func__, PTR_ERR(task));
1201                 nfs4_clear_state_manager_bit(clp);
1202                 nfs_put_client(clp);
1203                 module_put(THIS_MODULE);
1204         }
1205 }
1206
1207 /*
1208  * Schedule a lease recovery attempt
1209  */
1210 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1211 {
1212         if (!clp)
1213                 return;
1214         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1215                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1216         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1217                         clp->cl_hostname);
1218         nfs4_schedule_state_manager(clp);
1219 }
1220 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1221
1222 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1223 {
1224         int res;
1225
1226         might_sleep();
1227
1228         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1229                         nfs_wait_bit_killable, TASK_KILLABLE);
1230         if (res)
1231                 return res;
1232
1233         if (clp->cl_cons_state < 0)
1234                 return clp->cl_cons_state;
1235         return 0;
1236 }
1237
1238 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1239 {
1240         unsigned int loop;
1241         int ret;
1242
1243         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1244                 ret = nfs4_wait_clnt_recover(clp);
1245                 if (ret != 0)
1246                         break;
1247                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1248                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1249                         break;
1250                 nfs4_schedule_state_manager(clp);
1251                 ret = -EIO;
1252         }
1253         return ret;
1254 }
1255
1256 /*
1257  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1258  * @clp: client to process
1259  *
1260  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1261  * resend of the SETCLIENTID and hence re-establish the
1262  * callback channel. Then return all existing delegations.
1263  */
1264 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1265 {
1266         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1267         nfs_expire_all_delegations(clp);
1268         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1269                         clp->cl_hostname);
1270 }
1271
1272 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1273 {
1274         nfs40_handle_cb_pathdown(clp);
1275         nfs4_schedule_state_manager(clp);
1276 }
1277
1278 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1279 {
1280
1281         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1282         /* Don't recover state that expired before the reboot */
1283         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1284                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1285                 return 0;
1286         }
1287         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1288         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1289         return 1;
1290 }
1291
1292 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1293 {
1294         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1295         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1296         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1297         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1298         return 1;
1299 }
1300
1301 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1302 {
1303         struct nfs_client *clp = server->nfs_client;
1304
1305         if (!nfs4_valid_open_stateid(state))
1306                 return -EBADF;
1307         nfs4_state_mark_reclaim_nograce(clp, state);
1308         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1309                         clp->cl_hostname);
1310         nfs4_schedule_state_manager(clp);
1311         return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1314
1315 void nfs_inode_find_state_and_recover(struct inode *inode,
1316                 const nfs4_stateid *stateid)
1317 {
1318         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1319         struct nfs_inode *nfsi = NFS_I(inode);
1320         struct nfs_open_context *ctx;
1321         struct nfs4_state *state;
1322         bool found = false;
1323
1324         spin_lock(&inode->i_lock);
1325         list_for_each_entry(ctx, &nfsi->open_files, list) {
1326                 state = ctx->state;
1327                 if (state == NULL)
1328                         continue;
1329                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1330                         continue;
1331                 if (!nfs4_stateid_match(&state->stateid, stateid))
1332                         continue;
1333                 nfs4_state_mark_reclaim_nograce(clp, state);
1334                 found = true;
1335         }
1336         spin_unlock(&inode->i_lock);
1337         if (found)
1338                 nfs4_schedule_state_manager(clp);
1339 }
1340
1341 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1342 {
1343         struct inode *inode = state->inode;
1344         struct nfs_inode *nfsi = NFS_I(inode);
1345         struct nfs_open_context *ctx;
1346
1347         spin_lock(&inode->i_lock);
1348         list_for_each_entry(ctx, &nfsi->open_files, list) {
1349                 if (ctx->state != state)
1350                         continue;
1351                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1352         }
1353         spin_unlock(&inode->i_lock);
1354 }
1355
1356 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1357 {
1358         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1359         nfs4_state_mark_open_context_bad(state);
1360 }
1361
1362
1363 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1364 {
1365         struct inode *inode = state->inode;
1366         struct nfs_inode *nfsi = NFS_I(inode);
1367         struct file_lock *fl;
1368         int status = 0;
1369
1370         if (inode->i_flock == NULL)
1371                 return 0;
1372
1373         /* Guard against delegation returns and new lock/unlock calls */
1374         down_write(&nfsi->rwsem);
1375         /* Protect inode->i_flock using the BKL */
1376         lock_flocks();
1377         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1378                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1379                         continue;
1380                 if (nfs_file_open_context(fl->fl_file)->state != state)
1381                         continue;
1382                 unlock_flocks();
1383                 status = ops->recover_lock(state, fl);
1384                 switch (status) {
1385                         case 0:
1386                                 break;
1387                         case -ESTALE:
1388                         case -NFS4ERR_ADMIN_REVOKED:
1389                         case -NFS4ERR_STALE_STATEID:
1390                         case -NFS4ERR_BAD_STATEID:
1391                         case -NFS4ERR_EXPIRED:
1392                         case -NFS4ERR_NO_GRACE:
1393                         case -NFS4ERR_STALE_CLIENTID:
1394                         case -NFS4ERR_BADSESSION:
1395                         case -NFS4ERR_BADSLOT:
1396                         case -NFS4ERR_BAD_HIGH_SLOT:
1397                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1398                                 goto out;
1399                         default:
1400                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1401                                         "Zeroing state\n", __func__, status);
1402                         case -ENOMEM:
1403                         case -NFS4ERR_DENIED:
1404                         case -NFS4ERR_RECLAIM_BAD:
1405                         case -NFS4ERR_RECLAIM_CONFLICT:
1406                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1407                                 status = 0;
1408                 }
1409                 lock_flocks();
1410         }
1411         unlock_flocks();
1412 out:
1413         up_write(&nfsi->rwsem);
1414         return status;
1415 }
1416
1417 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1418 {
1419         struct nfs4_state *state;
1420         struct nfs4_lock_state *lock;
1421         int status = 0;
1422
1423         /* Note: we rely on the sp->so_states list being ordered
1424          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1425          * states first.
1426          * This is needed to ensure that the server won't give us any
1427          * read delegations that we have to return if, say, we are
1428          * recovering after a network partition or a reboot from a
1429          * server that doesn't support a grace period.
1430          */
1431         spin_lock(&sp->so_lock);
1432         write_seqcount_begin(&sp->so_reclaim_seqcount);
1433 restart:
1434         list_for_each_entry(state, &sp->so_states, open_states) {
1435                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1436                         continue;
1437                 if (!nfs4_valid_open_stateid(state))
1438                         continue;
1439                 if (state->state == 0)
1440                         continue;
1441                 atomic_inc(&state->count);
1442                 spin_unlock(&sp->so_lock);
1443                 status = ops->recover_open(sp, state);
1444                 if (status >= 0) {
1445                         status = nfs4_reclaim_locks(state, ops);
1446                         if (status >= 0) {
1447                                 spin_lock(&state->state_lock);
1448                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1449                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1450                                                 pr_warn_ratelimited("NFS: "
1451                                                         "%s: Lock reclaim "
1452                                                         "failed!\n", __func__);
1453                                 }
1454                                 spin_unlock(&state->state_lock);
1455                                 nfs4_put_open_state(state);
1456                                 spin_lock(&sp->so_lock);
1457                                 goto restart;
1458                         }
1459                 }
1460                 switch (status) {
1461                         default:
1462                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1463                                         "Zeroing state\n", __func__, status);
1464                         case -ENOENT:
1465                         case -ENOMEM:
1466                         case -ESTALE:
1467                                 /*
1468                                  * Open state on this file cannot be recovered
1469                                  * All we can do is revert to using the zero stateid.
1470                                  */
1471                                 nfs4_state_mark_recovery_failed(state, status);
1472                                 break;
1473                         case -EAGAIN:
1474                                 ssleep(1);
1475                         case -NFS4ERR_ADMIN_REVOKED:
1476                         case -NFS4ERR_STALE_STATEID:
1477                         case -NFS4ERR_BAD_STATEID:
1478                         case -NFS4ERR_RECLAIM_BAD:
1479                         case -NFS4ERR_RECLAIM_CONFLICT:
1480                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1481                                 break;
1482                         case -NFS4ERR_EXPIRED:
1483                         case -NFS4ERR_NO_GRACE:
1484                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1485                         case -NFS4ERR_STALE_CLIENTID:
1486                         case -NFS4ERR_BADSESSION:
1487                         case -NFS4ERR_BADSLOT:
1488                         case -NFS4ERR_BAD_HIGH_SLOT:
1489                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1490                                 goto out_err;
1491                 }
1492                 nfs4_put_open_state(state);
1493                 spin_lock(&sp->so_lock);
1494                 goto restart;
1495         }
1496         write_seqcount_end(&sp->so_reclaim_seqcount);
1497         spin_unlock(&sp->so_lock);
1498         return 0;
1499 out_err:
1500         nfs4_put_open_state(state);
1501         spin_lock(&sp->so_lock);
1502         write_seqcount_end(&sp->so_reclaim_seqcount);
1503         spin_unlock(&sp->so_lock);
1504         return status;
1505 }
1506
1507 static void nfs4_clear_open_state(struct nfs4_state *state)
1508 {
1509         struct nfs4_lock_state *lock;
1510
1511         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1512         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1513         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1514         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1515         spin_lock(&state->state_lock);
1516         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1517                 lock->ls_seqid.flags = 0;
1518                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1519         }
1520         spin_unlock(&state->state_lock);
1521 }
1522
1523 static void nfs4_reset_seqids(struct nfs_server *server,
1524         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1525 {
1526         struct nfs_client *clp = server->nfs_client;
1527         struct nfs4_state_owner *sp;
1528         struct rb_node *pos;
1529         struct nfs4_state *state;
1530
1531         spin_lock(&clp->cl_lock);
1532         for (pos = rb_first(&server->state_owners);
1533              pos != NULL;
1534              pos = rb_next(pos)) {
1535                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1536                 sp->so_seqid.flags = 0;
1537                 spin_lock(&sp->so_lock);
1538                 list_for_each_entry(state, &sp->so_states, open_states) {
1539                         if (mark_reclaim(clp, state))
1540                                 nfs4_clear_open_state(state);
1541                 }
1542                 spin_unlock(&sp->so_lock);
1543         }
1544         spin_unlock(&clp->cl_lock);
1545 }
1546
1547 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1548         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1549 {
1550         struct nfs_server *server;
1551
1552         rcu_read_lock();
1553         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1554                 nfs4_reset_seqids(server, mark_reclaim);
1555         rcu_read_unlock();
1556 }
1557
1558 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1559 {
1560         /* Mark all delegations for reclaim */
1561         nfs_delegation_mark_reclaim(clp);
1562         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1563 }
1564
1565 static void nfs4_reclaim_complete(struct nfs_client *clp,
1566                                  const struct nfs4_state_recovery_ops *ops)
1567 {
1568         /* Notify the server we're done reclaiming our state */
1569         if (ops->reclaim_complete)
1570                 (void)ops->reclaim_complete(clp);
1571 }
1572
1573 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1574 {
1575         struct nfs_client *clp = server->nfs_client;
1576         struct nfs4_state_owner *sp;
1577         struct rb_node *pos;
1578         struct nfs4_state *state;
1579
1580         spin_lock(&clp->cl_lock);
1581         for (pos = rb_first(&server->state_owners);
1582              pos != NULL;
1583              pos = rb_next(pos)) {
1584                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1585                 spin_lock(&sp->so_lock);
1586                 list_for_each_entry(state, &sp->so_states, open_states) {
1587                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1588                                                 &state->flags))
1589                                 continue;
1590                         nfs4_state_mark_reclaim_nograce(clp, state);
1591                 }
1592                 spin_unlock(&sp->so_lock);
1593         }
1594         spin_unlock(&clp->cl_lock);
1595 }
1596
1597 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1598 {
1599         struct nfs_server *server;
1600
1601         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1602                 return 0;
1603
1604         rcu_read_lock();
1605         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1606                 nfs4_clear_reclaim_server(server);
1607         rcu_read_unlock();
1608
1609         nfs_delegation_reap_unclaimed(clp);
1610         return 1;
1611 }
1612
1613 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1614 {
1615         if (!nfs4_state_clear_reclaim_reboot(clp))
1616                 return;
1617         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1618 }
1619
1620 static void nfs_delegation_clear_all(struct nfs_client *clp)
1621 {
1622         nfs_delegation_mark_reclaim(clp);
1623         nfs_delegation_reap_unclaimed(clp);
1624 }
1625
1626 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1627 {
1628         nfs_delegation_clear_all(clp);
1629         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1630 }
1631
1632 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1633 {
1634         switch (error) {
1635                 case 0:
1636                         break;
1637                 case -NFS4ERR_CB_PATH_DOWN:
1638                         nfs40_handle_cb_pathdown(clp);
1639                         break;
1640                 case -NFS4ERR_NO_GRACE:
1641                         nfs4_state_end_reclaim_reboot(clp);
1642                         break;
1643                 case -NFS4ERR_STALE_CLIENTID:
1644                 case -NFS4ERR_LEASE_MOVED:
1645                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1646                         nfs4_state_clear_reclaim_reboot(clp);
1647                         nfs4_state_start_reclaim_reboot(clp);
1648                         break;
1649                 case -NFS4ERR_EXPIRED:
1650                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1651                         nfs4_state_start_reclaim_nograce(clp);
1652                         break;
1653                 case -NFS4ERR_BADSESSION:
1654                 case -NFS4ERR_BADSLOT:
1655                 case -NFS4ERR_BAD_HIGH_SLOT:
1656                 case -NFS4ERR_DEADSESSION:
1657                 case -NFS4ERR_SEQ_FALSE_RETRY:
1658                 case -NFS4ERR_SEQ_MISORDERED:
1659                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1660                         /* Zero session reset errors */
1661                         break;
1662                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1663                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1664                         break;
1665                 default:
1666                         dprintk("%s: failed to handle error %d for server %s\n",
1667                                         __func__, error, clp->cl_hostname);
1668                         return error;
1669         }
1670         dprintk("%s: handled error %d for server %s\n", __func__, error,
1671                         clp->cl_hostname);
1672         return 0;
1673 }
1674
1675 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1676 {
1677         struct nfs4_state_owner *sp;
1678         struct nfs_server *server;
1679         struct rb_node *pos;
1680         int status = 0;
1681
1682 restart:
1683         rcu_read_lock();
1684         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1685                 nfs4_purge_state_owners(server);
1686                 spin_lock(&clp->cl_lock);
1687                 for (pos = rb_first(&server->state_owners);
1688                      pos != NULL;
1689                      pos = rb_next(pos)) {
1690                         sp = rb_entry(pos,
1691                                 struct nfs4_state_owner, so_server_node);
1692                         if (!test_and_clear_bit(ops->owner_flag_bit,
1693                                                         &sp->so_flags))
1694                                 continue;
1695                         atomic_inc(&sp->so_count);
1696                         spin_unlock(&clp->cl_lock);
1697                         rcu_read_unlock();
1698
1699                         status = nfs4_reclaim_open_state(sp, ops);
1700                         if (status < 0) {
1701                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1702                                 nfs4_put_state_owner(sp);
1703                                 return nfs4_recovery_handle_error(clp, status);
1704                         }
1705
1706                         nfs4_put_state_owner(sp);
1707                         goto restart;
1708                 }
1709                 spin_unlock(&clp->cl_lock);
1710         }
1711         rcu_read_unlock();
1712         return status;
1713 }
1714
1715 static int nfs4_check_lease(struct nfs_client *clp)
1716 {
1717         struct rpc_cred *cred;
1718         const struct nfs4_state_maintenance_ops *ops =
1719                 clp->cl_mvops->state_renewal_ops;
1720         int status;
1721
1722         /* Is the client already known to have an expired lease? */
1723         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1724                 return 0;
1725         spin_lock(&clp->cl_lock);
1726         cred = ops->get_state_renewal_cred_locked(clp);
1727         spin_unlock(&clp->cl_lock);
1728         if (cred == NULL) {
1729                 cred = nfs4_get_setclientid_cred(clp);
1730                 status = -ENOKEY;
1731                 if (cred == NULL)
1732                         goto out;
1733         }
1734         status = ops->renew_lease(clp, cred);
1735         put_rpccred(cred);
1736         if (status == -ETIMEDOUT) {
1737                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1738                 return 0;
1739         }
1740 out:
1741         return nfs4_recovery_handle_error(clp, status);
1742 }
1743
1744 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1745  * and for recoverable errors on EXCHANGE_ID for v4.1
1746  */
1747 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1748 {
1749         switch (status) {
1750         case -NFS4ERR_SEQ_MISORDERED:
1751                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1752                         return -ESERVERFAULT;
1753                 /* Lease confirmation error: retry after purging the lease */
1754                 ssleep(1);
1755                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1756                 break;
1757         case -NFS4ERR_STALE_CLIENTID:
1758                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1759                 nfs4_state_clear_reclaim_reboot(clp);
1760                 nfs4_state_start_reclaim_reboot(clp);
1761                 break;
1762         case -NFS4ERR_CLID_INUSE:
1763                 pr_err("NFS: Server %s reports our clientid is in use\n",
1764                         clp->cl_hostname);
1765                 nfs_mark_client_ready(clp, -EPERM);
1766                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1767                 return -EPERM;
1768         case -EACCES:
1769         case -NFS4ERR_DELAY:
1770         case -ETIMEDOUT:
1771         case -EAGAIN:
1772                 ssleep(1);
1773                 break;
1774
1775         case -NFS4ERR_MINOR_VERS_MISMATCH:
1776                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1777                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1778                 dprintk("%s: exit with error %d for server %s\n",
1779                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1780                 return -EPROTONOSUPPORT;
1781         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1782                                  * in nfs4_exchange_id */
1783         default:
1784                 dprintk("%s: exit with error %d for server %s\n", __func__,
1785                                 status, clp->cl_hostname);
1786                 return status;
1787         }
1788         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1789         dprintk("%s: handled error %d for server %s\n", __func__, status,
1790                         clp->cl_hostname);
1791         return 0;
1792 }
1793
1794 static int nfs4_establish_lease(struct nfs_client *clp)
1795 {
1796         struct rpc_cred *cred;
1797         const struct nfs4_state_recovery_ops *ops =
1798                 clp->cl_mvops->reboot_recovery_ops;
1799         int status;
1800
1801         cred = ops->get_clid_cred(clp);
1802         if (cred == NULL)
1803                 return -ENOENT;
1804         status = ops->establish_clid(clp, cred);
1805         put_rpccred(cred);
1806         if (status != 0)
1807                 return status;
1808         pnfs_destroy_all_layouts(clp);
1809         return 0;
1810 }
1811
1812 /*
1813  * Returns zero or a negative errno.  NFS4ERR values are converted
1814  * to local errno values.
1815  */
1816 static int nfs4_reclaim_lease(struct nfs_client *clp)
1817 {
1818         int status;
1819
1820         status = nfs4_establish_lease(clp);
1821         if (status < 0)
1822                 return nfs4_handle_reclaim_lease_error(clp, status);
1823         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1824                 nfs4_state_start_reclaim_nograce(clp);
1825         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1826                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1827         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1828         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1829         return 0;
1830 }
1831
1832 static int nfs4_purge_lease(struct nfs_client *clp)
1833 {
1834         int status;
1835
1836         status = nfs4_establish_lease(clp);
1837         if (status < 0)
1838                 return nfs4_handle_reclaim_lease_error(clp, status);
1839         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1840         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1841         nfs4_state_start_reclaim_nograce(clp);
1842         return 0;
1843 }
1844
1845 /**
1846  * nfs4_discover_server_trunking - Detect server IP address trunking
1847  *
1848  * @clp: nfs_client under test
1849  * @result: OUT: found nfs_client, or clp
1850  *
1851  * Returns zero or a negative errno.  If zero is returned,
1852  * an nfs_client pointer is planted in "result".
1853  *
1854  * Note: since we are invoked in process context, and
1855  * not from inside the state manager, we cannot use
1856  * nfs4_handle_reclaim_lease_error().
1857  */
1858 int nfs4_discover_server_trunking(struct nfs_client *clp,
1859                                   struct nfs_client **result)
1860 {
1861         const struct nfs4_state_recovery_ops *ops =
1862                                 clp->cl_mvops->reboot_recovery_ops;
1863         struct rpc_clnt *clnt;
1864         struct rpc_cred *cred;
1865         int i, status;
1866
1867         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1868
1869         clnt = clp->cl_rpcclient;
1870         i = 0;
1871
1872         mutex_lock(&nfs_clid_init_mutex);
1873 again:
1874         status  = -ENOENT;
1875         cred = ops->get_clid_cred(clp);
1876         if (cred == NULL)
1877                 goto out_unlock;
1878
1879         status = ops->detect_trunking(clp, result, cred);
1880         put_rpccred(cred);
1881         switch (status) {
1882         case 0:
1883                 break;
1884         case -NFS4ERR_DELAY:
1885         case -ETIMEDOUT:
1886         case -EAGAIN:
1887                 ssleep(1);
1888         case -NFS4ERR_STALE_CLIENTID:
1889                 dprintk("NFS: %s after status %d, retrying\n",
1890                         __func__, status);
1891                 goto again;
1892         case -EACCES:
1893                 if (i++)
1894                         break;
1895         case -NFS4ERR_CLID_INUSE:
1896         case -NFS4ERR_WRONGSEC:
1897                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
1898                 if (IS_ERR(clnt)) {
1899                         status = PTR_ERR(clnt);
1900                         break;
1901                 }
1902                 /* Note: this is safe because we haven't yet marked the
1903                  * client as ready, so we are the only user of
1904                  * clp->cl_rpcclient
1905                  */
1906                 clnt = xchg(&clp->cl_rpcclient, clnt);
1907                 rpc_shutdown_client(clnt);
1908                 clnt = clp->cl_rpcclient;
1909                 goto again;
1910
1911         case -NFS4ERR_MINOR_VERS_MISMATCH:
1912                 status = -EPROTONOSUPPORT;
1913                 break;
1914
1915         case -EKEYEXPIRED:
1916         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1917                                  * in nfs4_exchange_id */
1918                 status = -EKEYEXPIRED;
1919                 break;
1920         default:
1921                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
1922                                 __func__, status);
1923                 status = -EIO;
1924         }
1925
1926 out_unlock:
1927         mutex_unlock(&nfs_clid_init_mutex);
1928         dprintk("NFS: %s: status = %d\n", __func__, status);
1929         return status;
1930 }
1931
1932 #ifdef CONFIG_NFS_V4_1
1933 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1934 {
1935         struct nfs_client *clp = session->clp;
1936
1937         switch (err) {
1938         default:
1939                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1940                 break;
1941         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1942                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1943         }
1944         nfs4_schedule_lease_recovery(clp);
1945 }
1946 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1947
1948 static void nfs41_ping_server(struct nfs_client *clp)
1949 {
1950         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1951         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1952         nfs4_schedule_state_manager(clp);
1953 }
1954
1955 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1956 {
1957         nfs41_ping_server(clp);
1958 }
1959
1960 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1961 {
1962         nfs41_ping_server(clp);
1963 }
1964
1965 static void nfs4_reset_all_state(struct nfs_client *clp)
1966 {
1967         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1968                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1969                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1970                 nfs4_state_start_reclaim_nograce(clp);
1971                 dprintk("%s: scheduling reset of all state for server %s!\n",
1972                                 __func__, clp->cl_hostname);
1973                 nfs4_schedule_state_manager(clp);
1974         }
1975 }
1976
1977 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1978 {
1979         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1980                 nfs4_state_start_reclaim_reboot(clp);
1981                 dprintk("%s: server %s rebooted!\n", __func__,
1982                                 clp->cl_hostname);
1983                 nfs4_schedule_state_manager(clp);
1984         }
1985 }
1986
1987 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1988 {
1989         nfs4_reset_all_state(clp);
1990         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1991 }
1992
1993 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1994 {
1995         /* This will need to handle layouts too */
1996         nfs_expire_all_delegations(clp);
1997         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1998                         clp->cl_hostname);
1999 }
2000
2001 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2002 {
2003         nfs_expire_all_delegations(clp);
2004         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2005                 nfs4_schedule_state_manager(clp);
2006         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2007                         clp->cl_hostname);
2008 }
2009
2010 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2011 {
2012         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2013                 &clp->cl_state) == 0)
2014                 nfs4_schedule_state_manager(clp);
2015 }
2016
2017 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2018 {
2019         if (!flags)
2020                 return;
2021
2022         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2023                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2024
2025         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2026                 nfs41_handle_server_reboot(clp);
2027         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2028                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2029                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2030                             SEQ4_STATUS_LEASE_MOVED))
2031                 nfs41_handle_state_revoked(clp);
2032         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2033                 nfs41_handle_recallable_state_revoked(clp);
2034         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2035                 nfs41_handle_backchannel_fault(clp);
2036         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2037                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2038                 nfs41_handle_cb_path_down(clp);
2039 }
2040
2041 static int nfs4_reset_session(struct nfs_client *clp)
2042 {
2043         struct rpc_cred *cred;
2044         int status;
2045
2046         if (!nfs4_has_session(clp))
2047                 return 0;
2048         nfs4_begin_drain_session(clp);
2049         cred = nfs4_get_exchange_id_cred(clp);
2050         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2051         switch (status) {
2052         case 0:
2053         case -NFS4ERR_BADSESSION:
2054         case -NFS4ERR_DEADSESSION:
2055                 break;
2056         case -NFS4ERR_BACK_CHAN_BUSY:
2057         case -NFS4ERR_DELAY:
2058                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2059                 status = 0;
2060                 ssleep(1);
2061                 goto out;
2062         default:
2063                 status = nfs4_recovery_handle_error(clp, status);
2064                 goto out;
2065         }
2066
2067         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2068         status = nfs4_proc_create_session(clp, cred);
2069         if (status) {
2070                 dprintk("%s: session reset failed with status %d for server %s!\n",
2071                         __func__, status, clp->cl_hostname);
2072                 status = nfs4_handle_reclaim_lease_error(clp, status);
2073                 goto out;
2074         }
2075         nfs41_finish_session_reset(clp);
2076         dprintk("%s: session reset was successful for server %s!\n",
2077                         __func__, clp->cl_hostname);
2078 out:
2079         if (cred)
2080                 put_rpccred(cred);
2081         return status;
2082 }
2083
2084 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2085 {
2086         struct rpc_cred *cred;
2087         int ret;
2088
2089         if (!nfs4_has_session(clp))
2090                 return 0;
2091         nfs4_begin_drain_session(clp);
2092         cred = nfs4_get_exchange_id_cred(clp);
2093         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2094         if (cred)
2095                 put_rpccred(cred);
2096         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2097         switch (ret) {
2098         case 0:
2099                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2100                         __func__, clp->cl_hostname);
2101                 break;
2102         case -NFS4ERR_DELAY:
2103                 ssleep(1);
2104                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2105                 break;
2106         default:
2107                 return nfs4_recovery_handle_error(clp, ret);
2108         }
2109         return 0;
2110 }
2111 #else /* CONFIG_NFS_V4_1 */
2112 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2113 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2114
2115 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2116 {
2117         return 0;
2118 }
2119 #endif /* CONFIG_NFS_V4_1 */
2120
2121 static void nfs4_state_manager(struct nfs_client *clp)
2122 {
2123         int status = 0;
2124         const char *section = "", *section_sep = "";
2125
2126         /* Ensure exclusive access to NFSv4 state */
2127         do {
2128                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2129                         section = "purge state";
2130                         status = nfs4_purge_lease(clp);
2131                         if (status < 0)
2132                                 goto out_error;
2133                         continue;
2134                 }
2135
2136                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2137                         section = "lease expired";
2138                         /* We're going to have to re-establish a clientid */
2139                         status = nfs4_reclaim_lease(clp);
2140                         if (status < 0)
2141                                 goto out_error;
2142                         continue;
2143                 }
2144
2145                 /* Initialize or reset the session */
2146                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2147                         section = "reset session";
2148                         status = nfs4_reset_session(clp);
2149                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2150                                 continue;
2151                         if (status < 0)
2152                                 goto out_error;
2153                 }
2154
2155                 /* Send BIND_CONN_TO_SESSION */
2156                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2157                                 &clp->cl_state)) {
2158                         section = "bind conn to session";
2159                         status = nfs4_bind_conn_to_session(clp);
2160                         if (status < 0)
2161                                 goto out_error;
2162                         continue;
2163                 }
2164
2165                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2166                         section = "check lease";
2167                         status = nfs4_check_lease(clp);
2168                         if (status < 0)
2169                                 goto out_error;
2170                         continue;
2171                 }
2172
2173                 /* First recover reboot state... */
2174                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2175                         section = "reclaim reboot";
2176                         status = nfs4_do_reclaim(clp,
2177                                 clp->cl_mvops->reboot_recovery_ops);
2178                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2179                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2180                                 continue;
2181                         nfs4_state_end_reclaim_reboot(clp);
2182                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2183                                 continue;
2184                         if (status < 0)
2185                                 goto out_error;
2186                 }
2187
2188                 /* Now recover expired state... */
2189                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2190                         section = "reclaim nograce";
2191                         status = nfs4_do_reclaim(clp,
2192                                 clp->cl_mvops->nograce_recovery_ops);
2193                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2194                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2195                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2196                                 continue;
2197                         if (status < 0)
2198                                 goto out_error;
2199                 }
2200
2201                 nfs4_end_drain_session(clp);
2202                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2203                         nfs_client_return_marked_delegations(clp);
2204                         continue;
2205                 }
2206
2207                 nfs4_clear_state_manager_bit(clp);
2208                 /* Did we race with an attempt to give us more work? */
2209                 if (clp->cl_state == 0)
2210                         break;
2211                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2212                         break;
2213         } while (atomic_read(&clp->cl_count) > 1);
2214         return;
2215 out_error:
2216         if (strlen(section))
2217                 section_sep = ": ";
2218         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2219                         " with error %d\n", section_sep, section,
2220                         clp->cl_hostname, -status);
2221         ssleep(1);
2222         nfs4_end_drain_session(clp);
2223         nfs4_clear_state_manager_bit(clp);
2224 }
2225
2226 static int nfs4_run_state_manager(void *ptr)
2227 {
2228         struct nfs_client *clp = ptr;
2229
2230         allow_signal(SIGKILL);
2231         nfs4_state_manager(clp);
2232         nfs_put_client(clp);
2233         module_put_and_exit(0);
2234         return 0;
2235 }
2236
2237 /*
2238  * Local variables:
2239  *  c-basic-offset: 8
2240  * End:
2241  */