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 "pnfs.h"
  61 #include "netns.h"
  62
  63 #define NFSDBG_FACILITY         NFSDBG_STATE
  64
  65 #define OPENOWNER_POOL_SIZE     8
  66
  67 const nfs4_stateid zero_stateid;
  68 static DEFINE_MUTEX(nfs_clid_init_mutex);
  69 static LIST_HEAD(nfs4_clientid_list);
  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         switch (status) {
 140         case -NFS4ERR_STALE_CLIENTID:
 141                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 142         case 0:
 143                 /* Sustain the lease, even if it's empty.  If the clientid4
 144                  * goes stale it's of no use for trunking discovery. */
 145                 nfs4_schedule_state_renewal(*result);
 146                 break;
 147         }
 148
 149 out:
 150         return status;
 151 }
 152
 153 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 154 {
 155         struct rpc_cred *cred = NULL;
 156
 157         if (clp->cl_machine_cred != NULL)
 158                 cred = get_rpccred(clp->cl_machine_cred);
 159         return cred;
 160 }
 161
 162 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 163 {
 164         struct rpc_cred *cred;
 165
 166         spin_lock(&clp->cl_lock);
 167         cred = clp->cl_machine_cred;
 168         clp->cl_machine_cred = NULL;
 169         spin_unlock(&clp->cl_lock);
 170         if (cred != NULL)
 171                 put_rpccred(cred);
 172 }
 173
 174 static struct rpc_cred *
 175 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 176 {
 177         struct rpc_cred *cred = NULL;
 178         struct nfs4_state_owner *sp;
 179         struct rb_node *pos;
 180
 181         for (pos = rb_first(&server->state_owners);
 182              pos != NULL;
 183              pos = rb_next(pos)) {
 184                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 185                 if (list_empty(&sp->so_states))
 186                         continue;
 187                 cred = get_rpccred(sp->so_cred);
 188                 break;
 189         }
 190         return cred;
 191 }
 192
 193 /**
 194  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 195  * @clp: client state handle
 196  *
 197  * Returns an rpc_cred with reference count bumped, or NULL.
 198  * Caller must hold clp->cl_lock.
 199  */
 200 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 201 {
 202         struct rpc_cred *cred = NULL;
 203         struct nfs_server *server;
 204
 205         /* Use machine credentials if available */
 206         cred = nfs4_get_machine_cred_locked(clp);
 207         if (cred != NULL)
 208                 goto out;
 209
 210         rcu_read_lock();
 211         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 212                 cred = nfs4_get_renew_cred_server_locked(server);
 213                 if (cred != NULL)
 214                         break;
 215         }
 216         rcu_read_unlock();
 217
 218 out:
 219         return cred;
 220 }
 221
 222 #if defined(CONFIG_NFS_V4_1)
 223
 224 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 225 {
 226         int status;
 227         struct nfs_fsinfo fsinfo;
 228
 229         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 230                 nfs4_schedule_state_renewal(clp);
 231                 return 0;
 232         }
 233
 234         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 235         if (status == 0) {
 236                 /* Update lease time and schedule renewal */
 237                 spin_lock(&clp->cl_lock);
 238                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 239                 clp->cl_last_renewal = jiffies;
 240                 spin_unlock(&clp->cl_lock);
 241
 242                 nfs4_schedule_state_renewal(clp);
 243         }
 244
 245         return status;
 246 }
 247
 248 /*
 249  * Back channel returns NFS4ERR_DELAY for new requests when
 250  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 251  * is ended.
 252  */
 253 static void nfs4_end_drain_session(struct nfs_client *clp)
 254 {
 255         struct nfs4_session *ses = clp->cl_session;
 256         struct nfs4_slot_table *tbl;
 257         unsigned int i;
 258
 259         if (ses == NULL)
 260                 return;
 261         tbl = &ses->fc_slot_table;
 262         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 263                 spin_lock(&tbl->slot_tbl_lock);
 264                 for (i = 0; i <= tbl->max_slotid; i++) {
 265                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 266                                                 nfs4_set_task_privileged,
 267                                                 NULL) == NULL)
 268                                 break;
 269                 }
 270                 spin_unlock(&tbl->slot_tbl_lock);
 271         }
 272 }
 273
 274 /*
 275  * Signal state manager thread if session fore channel is drained
 276  */
 277 void nfs4_session_drain_complete(struct nfs4_session *session,
 278                 struct nfs4_slot_table *tbl)
 279 {
 280         if (nfs4_session_draining(session))
 281                 complete(&tbl->complete);
 282 }
 283
 284 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 285 {
 286         spin_lock(&tbl->slot_tbl_lock);
 287         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 288                 INIT_COMPLETION(tbl->complete);
 289                 spin_unlock(&tbl->slot_tbl_lock);
 290                 return wait_for_completion_interruptible(&tbl->complete);
 291         }
 292         spin_unlock(&tbl->slot_tbl_lock);
 293         return 0;
 294 }
 295
 296 static int nfs4_begin_drain_session(struct nfs_client *clp)
 297 {
 298         struct nfs4_session *ses = clp->cl_session;
 299         int ret = 0;
 300
 301         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 302         /* back channel */
 303         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 304         if (ret)
 305                 return ret;
 306         /* fore channel */
 307         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 308 }
 309
 310 static void nfs41_finish_session_reset(struct nfs_client *clp)
 311 {
 312         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 313         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 314         /* create_session negotiated new slot table */
 315         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 316         nfs41_setup_state_renewal(clp);
 317 }
 318
 319 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 320 {
 321         int status;
 322
 323         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 324                 goto do_confirm;
 325         nfs4_begin_drain_session(clp);
 326         status = nfs4_proc_exchange_id(clp, cred);
 327         if (status != 0)
 328                 goto out;
 329         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 330 do_confirm:
 331         status = nfs4_proc_create_session(clp, cred);
 332         if (status != 0)
 333                 goto out;
 334         nfs41_finish_session_reset(clp);
 335         nfs_mark_client_ready(clp, NFS_CS_READY);
 336 out:
 337         return status;
 338 }
 339
 340 /**
 341  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 342  *
 343  * @clp: nfs_client under test
 344  * @result: OUT: found nfs_client, or clp
 345  * @cred: credential to use for trunking test
 346  *
 347  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 348  * If NFS4_OK is returned, an nfs_client pointer is planted in
 349  * "result".
 350  *
 351  * Note: The returned client may not yet be marked ready.
 352  */
 353 int nfs41_discover_server_trunking(struct nfs_client *clp,
 354                                    struct nfs_client **result,
 355                                    struct rpc_cred *cred)
 356 {
 357         int status;
 358
 359         status = nfs4_proc_exchange_id(clp, cred);
 360         if (status != NFS4_OK)
 361                 return status;
 362         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 363
 364         return nfs41_walk_client_list(clp, result, cred);
 365 }
 366
 367 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 368 {
 369         struct rpc_cred *cred;
 370
 371         spin_lock(&clp->cl_lock);
 372         cred = nfs4_get_machine_cred_locked(clp);
 373         spin_unlock(&clp->cl_lock);
 374         return cred;
 375 }
 376
 377 #endif /* CONFIG_NFS_V4_1 */
 378
 379 static struct rpc_cred *
 380 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 381 {
 382         struct nfs_client *clp = server->nfs_client;
 383         struct rpc_cred *cred = NULL;
 384         struct nfs4_state_owner *sp;
 385         struct rb_node *pos;
 386
 387         spin_lock(&clp->cl_lock);
 388         pos = rb_first(&server->state_owners);
 389         if (pos != NULL) {
 390                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 391                 cred = get_rpccred(sp->so_cred);
 392         }
 393         spin_unlock(&clp->cl_lock);
 394         return cred;
 395 }
 396
 397 /**
 398  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 399  * @clp: client state handle
 400  *
 401  * Returns an rpc_cred with reference count bumped, or NULL.
 402  */
 403 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 404 {
 405         struct nfs_server *server;
 406         struct rpc_cred *cred;
 407
 408         spin_lock(&clp->cl_lock);
 409         cred = nfs4_get_machine_cred_locked(clp);
 410         spin_unlock(&clp->cl_lock);
 411         if (cred != NULL)
 412                 goto out;
 413
 414         rcu_read_lock();
 415         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 416                 cred = nfs4_get_setclientid_cred_server(server);
 417                 if (cred != NULL)
 418                         break;
 419         }
 420         rcu_read_unlock();
 421
 422 out:
 423         return cred;
 424 }
 425
 426 static struct nfs4_state_owner *
 427 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 428 {
 429         struct rb_node **p = &server->state_owners.rb_node,
 430                        *parent = NULL;
 431         struct nfs4_state_owner *sp;
 432
 433         while (*p != NULL) {
 434                 parent = *p;
 435                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 436
 437                 if (cred < sp->so_cred)
 438                         p = &parent->rb_left;
 439                 else if (cred > sp->so_cred)
 440                         p = &parent->rb_right;
 441                 else {
 442                         if (!list_empty(&sp->so_lru))
 443                                 list_del_init(&sp->so_lru);
 444                         atomic_inc(&sp->so_count);
 445                         return sp;
 446                 }
 447         }
 448         return NULL;
 449 }
 450
 451 static struct nfs4_state_owner *
 452 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 453 {
 454         struct nfs_server *server = new->so_server;
 455         struct rb_node **p = &server->state_owners.rb_node,
 456                        *parent = NULL;
 457         struct nfs4_state_owner *sp;
 458         int err;
 459
 460         while (*p != NULL) {
 461                 parent = *p;
 462                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 463
 464                 if (new->so_cred < sp->so_cred)
 465                         p = &parent->rb_left;
 466                 else if (new->so_cred > sp->so_cred)
 467                         p = &parent->rb_right;
 468                 else {
 469                         if (!list_empty(&sp->so_lru))
 470                                 list_del_init(&sp->so_lru);
 471                         atomic_inc(&sp->so_count);
 472                         return sp;
 473                 }
 474         }
 475         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 476         if (err)
 477                 return ERR_PTR(err);
 478         rb_link_node(&new->so_server_node, parent, p);
 479         rb_insert_color(&new->so_server_node, &server->state_owners);
 480         return new;
 481 }
 482
 483 static void
 484 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 485 {
 486         struct nfs_server *server = sp->so_server;
 487
 488         if (!RB_EMPTY_NODE(&sp->so_server_node))
 489                 rb_erase(&sp->so_server_node, &server->state_owners);
 490         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 491 }
 492
 493 static void
 494 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 495 {
 496         sc->create_time = ktime_get();
 497         sc->flags = 0;
 498         sc->counter = 0;
 499         spin_lock_init(&sc->lock);
 500         INIT_LIST_HEAD(&sc->list);
 501         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 502 }
 503
 504 static void
 505 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 506 {
 507         rpc_destroy_wait_queue(&sc->wait);
 508 }
 509
 510 /*
 511  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 512  * create a new state_owner.
 513  *
 514  */
 515 static struct nfs4_state_owner *
 516 nfs4_alloc_state_owner(struct nfs_server *server,
 517                 struct rpc_cred *cred,
 518                 gfp_t gfp_flags)
 519 {
 520         struct nfs4_state_owner *sp;
 521
 522         sp = kzalloc(sizeof(*sp), gfp_flags);
 523         if (!sp)
 524                 return NULL;
 525         sp->so_server = server;
 526         sp->so_cred = get_rpccred(cred);
 527         spin_lock_init(&sp->so_lock);
 528         INIT_LIST_HEAD(&sp->so_states);
 529         nfs4_init_seqid_counter(&sp->so_seqid);
 530         atomic_set(&sp->so_count, 1);
 531         INIT_LIST_HEAD(&sp->so_lru);
 532         return sp;
 533 }
 534
 535 static void
 536 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 537 {
 538         struct rb_node *rb_node = &sp->so_server_node;
 539
 540         if (!RB_EMPTY_NODE(rb_node)) {
 541                 struct nfs_server *server = sp->so_server;
 542                 struct nfs_client *clp = server->nfs_client;
 543
 544                 spin_lock(&clp->cl_lock);
 545                 if (!RB_EMPTY_NODE(rb_node)) {
 546                         rb_erase(rb_node, &server->state_owners);
 547                         RB_CLEAR_NODE(rb_node);
 548                 }
 549                 spin_unlock(&clp->cl_lock);
 550         }
 551 }
 552
 553 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 554 {
 555         nfs4_destroy_seqid_counter(&sp->so_seqid);
 556         put_rpccred(sp->so_cred);
 557         kfree(sp);
 558 }
 559
 560 static void nfs4_gc_state_owners(struct nfs_server *server)
 561 {
 562         struct nfs_client *clp = server->nfs_client;
 563         struct nfs4_state_owner *sp, *tmp;
 564         unsigned long time_min, time_max;
 565         LIST_HEAD(doomed);
 566
 567         spin_lock(&clp->cl_lock);
 568         time_max = jiffies;
 569         time_min = (long)time_max - (long)clp->cl_lease_time;
 570         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 571                 /* NB: LRU is sorted so that oldest is at the head */
 572                 if (time_in_range(sp->so_expires, time_min, time_max))
 573                         break;
 574                 list_move(&sp->so_lru, &doomed);
 575                 nfs4_remove_state_owner_locked(sp);
 576         }
 577         spin_unlock(&clp->cl_lock);
 578
 579         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 580                 list_del(&sp->so_lru);
 581                 nfs4_free_state_owner(sp);
 582         }
 583 }
 584
 585 /**
 586  * nfs4_get_state_owner - Look up a state owner given a credential
 587  * @server: nfs_server to search
 588  * @cred: RPC credential to match
 589  *
 590  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 591  */
 592 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 593                                               struct rpc_cred *cred,
 594                                               gfp_t gfp_flags)
 595 {
 596         struct nfs_client *clp = server->nfs_client;
 597         struct nfs4_state_owner *sp, *new;
 598
 599         spin_lock(&clp->cl_lock);
 600         sp = nfs4_find_state_owner_locked(server, cred);
 601         spin_unlock(&clp->cl_lock);
 602         if (sp != NULL)
 603                 goto out;
 604         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 605         if (new == NULL)
 606                 goto out;
 607         do {
 608                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 609                         break;
 610                 spin_lock(&clp->cl_lock);
 611                 sp = nfs4_insert_state_owner_locked(new);
 612                 spin_unlock(&clp->cl_lock);
 613         } while (sp == ERR_PTR(-EAGAIN));
 614         if (sp != new)
 615                 nfs4_free_state_owner(new);
 616 out:
 617         nfs4_gc_state_owners(server);
 618         return sp;
 619 }
 620
 621 /**
 622  * nfs4_put_state_owner - Release a nfs4_state_owner
 623  * @sp: state owner data to release
 624  *
 625  * Note that we keep released state owners on an LRU
 626  * list.
 627  * This caches valid state owners so that they can be
 628  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 629  * It also pins the uniquifier of dropped state owners for
 630  * a while, to ensure that those state owner names are
 631  * never reused.
 632  */
 633 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 634 {
 635         struct nfs_server *server = sp->so_server;
 636         struct nfs_client *clp = server->nfs_client;
 637
 638         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 639                 return;
 640
 641         sp->so_expires = jiffies;
 642         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 643         spin_unlock(&clp->cl_lock);
 644 }
 645
 646 /**
 647  * nfs4_purge_state_owners - Release all cached state owners
 648  * @server: nfs_server with cached state owners to release
 649  *
 650  * Called at umount time.  Remaining state owners will be on
 651  * the LRU with ref count of zero.
 652  */
 653 void nfs4_purge_state_owners(struct nfs_server *server)
 654 {
 655         struct nfs_client *clp = server->nfs_client;
 656         struct nfs4_state_owner *sp, *tmp;
 657         LIST_HEAD(doomed);
 658
 659         spin_lock(&clp->cl_lock);
 660         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 661                 list_move(&sp->so_lru, &doomed);
 662                 nfs4_remove_state_owner_locked(sp);
 663         }
 664         spin_unlock(&clp->cl_lock);
 665
 666         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 667                 list_del(&sp->so_lru);
 668                 nfs4_free_state_owner(sp);
 669         }
 670 }
 671
 672 static struct nfs4_state *
 673 nfs4_alloc_open_state(void)
 674 {
 675         struct nfs4_state *state;
 676
 677         state = kzalloc(sizeof(*state), GFP_NOFS);
 678         if (!state)
 679                 return NULL;
 680         atomic_set(&state->count, 1);
 681         INIT_LIST_HEAD(&state->lock_states);
 682         spin_lock_init(&state->state_lock);
 683         seqlock_init(&state->seqlock);
 684         return state;
 685 }
 686
 687 void
 688 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 689 {
 690         if (state->state == fmode)
 691                 return;
 692         /* NB! List reordering - see the reclaim code for why.  */
 693         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 694                 if (fmode & FMODE_WRITE)
 695                         list_move(&state->open_states, &state->owner->so_states);
 696                 else
 697                         list_move_tail(&state->open_states, &state->owner->so_states);
 698         }
 699         state->state = fmode;
 700 }
 701
 702 static struct nfs4_state *
 703 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 704 {
 705         struct nfs_inode *nfsi = NFS_I(inode);
 706         struct nfs4_state *state;
 707
 708         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 709                 if (state->owner != owner)
 710                         continue;
 711                 if (atomic_inc_not_zero(&state->count))
 712                         return state;
 713         }
 714         return NULL;
 715 }
 716
 717 static void
 718 nfs4_free_open_state(struct nfs4_state *state)
 719 {
 720         kfree(state);
 721 }
 722
 723 struct nfs4_state *
 724 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 725 {
 726         struct nfs4_state *state, *new;
 727         struct nfs_inode *nfsi = NFS_I(inode);
 728
 729         spin_lock(&inode->i_lock);
 730         state = __nfs4_find_state_byowner(inode, owner);
 731         spin_unlock(&inode->i_lock);
 732         if (state)
 733                 goto out;
 734         new = nfs4_alloc_open_state();
 735         spin_lock(&owner->so_lock);
 736         spin_lock(&inode->i_lock);
 737         state = __nfs4_find_state_byowner(inode, owner);
 738         if (state == NULL && new != NULL) {
 739                 state = new;
 740                 state->owner = owner;
 741                 atomic_inc(&owner->so_count);
 742                 list_add(&state->inode_states, &nfsi->open_states);
 743                 ihold(inode);
 744                 state->inode = inode;
 745                 spin_unlock(&inode->i_lock);
 746                 /* Note: The reclaim code dictates that we add stateless
 747                  * and read-only stateids to the end of the list */
 748                 list_add_tail(&state->open_states, &owner->so_states);
 749                 spin_unlock(&owner->so_lock);
 750         } else {
 751                 spin_unlock(&inode->i_lock);
 752                 spin_unlock(&owner->so_lock);
 753                 if (new)
 754                         nfs4_free_open_state(new);
 755         }
 756 out:
 757         return state;
 758 }
 759
 760 void nfs4_put_open_state(struct nfs4_state *state)
 761 {
 762         struct inode *inode = state->inode;
 763         struct nfs4_state_owner *owner = state->owner;
 764
 765         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 766                 return;
 767         spin_lock(&inode->i_lock);
 768         list_del(&state->inode_states);
 769         list_del(&state->open_states);
 770         spin_unlock(&inode->i_lock);
 771         spin_unlock(&owner->so_lock);
 772         iput(inode);
 773         nfs4_free_open_state(state);
 774         nfs4_put_state_owner(owner);
 775 }
 776
 777 /*
 778  * Close the current file.
 779  */
 780 static void __nfs4_close(struct nfs4_state *state,
 781                 fmode_t fmode, gfp_t gfp_mask, int wait)
 782 {
 783         struct nfs4_state_owner *owner = state->owner;
 784         int call_close = 0;
 785         fmode_t newstate;
 786
 787         atomic_inc(&owner->so_count);
 788         /* Protect against nfs4_find_state() */
 789         spin_lock(&owner->so_lock);
 790         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 791                 case FMODE_READ:
 792                         state->n_rdonly--;
 793                         break;
 794                 case FMODE_WRITE:
 795                         state->n_wronly--;
 796                         break;
 797                 case FMODE_READ|FMODE_WRITE:
 798                         state->n_rdwr--;
 799         }
 800         newstate = FMODE_READ|FMODE_WRITE;
 801         if (state->n_rdwr == 0) {
 802                 if (state->n_rdonly == 0) {
 803                         newstate &= ~FMODE_READ;
 804                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 805                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 806                 }
 807                 if (state->n_wronly == 0) {
 808                         newstate &= ~FMODE_WRITE;
 809                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 810                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 811                 }
 812                 if (newstate == 0)
 813                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 814         }
 815         nfs4_state_set_mode_locked(state, newstate);
 816         spin_unlock(&owner->so_lock);
 817
 818         if (!call_close) {
 819                 nfs4_put_open_state(state);
 820                 nfs4_put_state_owner(owner);
 821         } else
 822                 nfs4_do_close(state, gfp_mask, wait);
 823 }
 824
 825 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 826 {
 827         __nfs4_close(state, fmode, GFP_NOFS, 0);
 828 }
 829
 830 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 831 {
 832         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 833 }
 834
 835 /*
 836  * Search the state->lock_states for an existing lock_owner
 837  * that is compatible with current->files
 838  */
 839 static struct nfs4_lock_state *
 840 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 841 {
 842         struct nfs4_lock_state *pos;
 843         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 844                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 845                         continue;
 846                 switch (pos->ls_owner.lo_type) {
 847                 case NFS4_POSIX_LOCK_TYPE:
 848                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 849                                 continue;
 850                         break;
 851                 case NFS4_FLOCK_LOCK_TYPE:
 852                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 853                                 continue;
 854                 }
 855                 atomic_inc(&pos->ls_count);
 856                 return pos;
 857         }
 858         return NULL;
 859 }
 860
 861 /*
 862  * Return a compatible lock_state. If no initialized lock_state structure
 863  * exists, return an uninitialized one.
 864  *
 865  */
 866 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)
 867 {
 868         struct nfs4_lock_state *lsp;
 869         struct nfs_server *server = state->owner->so_server;
 870
 871         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 872         if (lsp == NULL)
 873                 return NULL;
 874         nfs4_init_seqid_counter(&lsp->ls_seqid);
 875         atomic_set(&lsp->ls_count, 1);
 876         lsp->ls_state = state;
 877         lsp->ls_owner.lo_type = type;
 878         switch (lsp->ls_owner.lo_type) {
 879         case NFS4_FLOCK_LOCK_TYPE:
 880                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 881                 break;
 882         case NFS4_POSIX_LOCK_TYPE:
 883                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 884                 break;
 885         default:
 886                 goto out_free;
 887         }
 888         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 889         if (lsp->ls_seqid.owner_id < 0)
 890                 goto out_free;
 891         INIT_LIST_HEAD(&lsp->ls_locks);
 892         return lsp;
 893 out_free:
 894         kfree(lsp);
 895         return NULL;
 896 }
 897
 898 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 899 {
 900         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 901         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 902         kfree(lsp);
 903 }
 904
 905 /*
 906  * Return a compatible lock_state. If no initialized lock_state structure
 907  * exists, return an uninitialized one.
 908  *
 909  */
 910 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 911 {
 912         struct nfs4_lock_state *lsp, *new = NULL;
 913
 914         for(;;) {
 915                 spin_lock(&state->state_lock);
 916                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 917                 if (lsp != NULL)
 918                         break;
 919                 if (new != NULL) {
 920                         list_add(&new->ls_locks, &state->lock_states);
 921                         set_bit(LK_STATE_IN_USE, &state->flags);
 922                         lsp = new;
 923                         new = NULL;
 924                         break;
 925                 }
 926                 spin_unlock(&state->state_lock);
 927                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 928                 if (new == NULL)
 929                         return NULL;
 930         }
 931         spin_unlock(&state->state_lock);
 932         if (new != NULL)
 933                 nfs4_free_lock_state(state->owner->so_server, new);
 934         return lsp;
 935 }
 936
 937 /*
 938  * Release reference to lock_state, and free it if we see that
 939  * it is no longer in use
 940  */
 941 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 942 {
 943         struct nfs4_state *state;
 944
 945         if (lsp == NULL)
 946                 return;
 947         state = lsp->ls_state;
 948         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 949                 return;
 950         list_del(&lsp->ls_locks);
 951         if (list_empty(&state->lock_states))
 952                 clear_bit(LK_STATE_IN_USE, &state->flags);
 953         spin_unlock(&state->state_lock);
 954         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 955                 if (nfs4_release_lockowner(lsp) == 0)
 956                         return;
 957         }
 958         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 959 }
 960
 961 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 962 {
 963         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 964
 965         dst->fl_u.nfs4_fl.owner = lsp;
 966         atomic_inc(&lsp->ls_count);
 967 }
 968
 969 static void nfs4_fl_release_lock(struct file_lock *fl)
 970 {
 971         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 972 }
 973
 974 static const struct file_lock_operations nfs4_fl_lock_ops = {
 975         .fl_copy_lock = nfs4_fl_copy_lock,
 976         .fl_release_private = nfs4_fl_release_lock,
 977 };
 978
 979 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 980 {
 981         struct nfs4_lock_state *lsp;
 982
 983         if (fl->fl_ops != NULL)
 984                 return 0;
 985         if (fl->fl_flags & FL_POSIX)
 986                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 987         else if (fl->fl_flags & FL_FLOCK)
 988                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 989                                 NFS4_FLOCK_LOCK_TYPE);
 990         else
 991                 return -EINVAL;
 992         if (lsp == NULL)
 993                 return -ENOMEM;
 994         fl->fl_u.nfs4_fl.owner = lsp;
 995         fl->fl_ops = &nfs4_fl_lock_ops;
 996         return 0;
 997 }
 998
 999 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1000                 const struct nfs_lockowner *lockowner)
1001 {
1002         struct nfs4_lock_state *lsp;
1003         fl_owner_t fl_owner;
1004         pid_t fl_pid;
1005         bool ret = false;
1006
1007
1008         if (lockowner == NULL)
1009                 goto out;
1010
1011         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1012                 goto out;
1013
1014         fl_owner = lockowner->l_owner;
1015         fl_pid = lockowner->l_pid;
1016         spin_lock(&state->state_lock);
1017         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1018         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1019                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1020                 ret = true;
1021         }
1022         spin_unlock(&state->state_lock);
1023         nfs4_put_lock_state(lsp);
1024 out:
1025         return ret;
1026 }
1027
1028 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1029 {
1030         int seq;
1031
1032         do {
1033                 seq = read_seqbegin(&state->seqlock);
1034                 nfs4_stateid_copy(dst, &state->stateid);
1035         } while (read_seqretry(&state->seqlock, seq));
1036 }
1037
1038 /*
1039  * Byte-range lock aware utility to initialize the stateid of read/write
1040  * requests.
1041  */
1042 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1043                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1044 {
1045         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1046                 return;
1047         if (nfs4_copy_lock_stateid(dst, state, lockowner))
1048                 return;
1049         nfs4_copy_open_stateid(dst, state);
1050 }
1051
1052 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1053 {
1054         struct nfs_seqid *new;
1055
1056         new = kmalloc(sizeof(*new), gfp_mask);
1057         if (new != NULL) {
1058                 new->sequence = counter;
1059                 INIT_LIST_HEAD(&new->list);
1060                 new->task = NULL;
1061         }
1062         return new;
1063 }
1064
1065 void nfs_release_seqid(struct nfs_seqid *seqid)
1066 {
1067         struct nfs_seqid_counter *sequence;
1068
1069         if (list_empty(&seqid->list))
1070                 return;
1071         sequence = seqid->sequence;
1072         spin_lock(&sequence->lock);
1073         list_del_init(&seqid->list);
1074         if (!list_empty(&sequence->list)) {
1075                 struct nfs_seqid *next;
1076
1077                 next = list_first_entry(&sequence->list,
1078                                 struct nfs_seqid, list);
1079                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1080         }
1081         spin_unlock(&sequence->lock);
1082 }
1083
1084 void nfs_free_seqid(struct nfs_seqid *seqid)
1085 {
1086         nfs_release_seqid(seqid);
1087         kfree(seqid);
1088 }
1089
1090 /*
1091  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1092  * failed with a seqid incrementing error -
1093  * see comments nfs_fs.h:seqid_mutating_error()
1094  */
1095 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1096 {
1097         switch (status) {
1098                 case 0:
1099                         break;
1100                 case -NFS4ERR_BAD_SEQID:
1101                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1102                                 return;
1103                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1104                                         " sequence-id error on an"
1105                                         " unconfirmed sequence %p!\n",
1106                                         seqid->sequence);
1107                 case -NFS4ERR_STALE_CLIENTID:
1108                 case -NFS4ERR_STALE_STATEID:
1109                 case -NFS4ERR_BAD_STATEID:
1110                 case -NFS4ERR_BADXDR:
1111                 case -NFS4ERR_RESOURCE:
1112                 case -NFS4ERR_NOFILEHANDLE:
1113                         /* Non-seqid mutating errors */
1114                         return;
1115         };
1116         /*
1117          * Note: no locking needed as we are guaranteed to be first
1118          * on the sequence list
1119          */
1120         seqid->sequence->counter++;
1121 }
1122
1123 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1124 {
1125         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1126                                         struct nfs4_state_owner, so_seqid);
1127         struct nfs_server *server = sp->so_server;
1128
1129         if (status == -NFS4ERR_BAD_SEQID)
1130                 nfs4_drop_state_owner(sp);
1131         if (!nfs4_has_session(server->nfs_client))
1132                 nfs_increment_seqid(status, seqid);
1133 }
1134
1135 /*
1136  * Increment the seqid if the LOCK/LOCKU succeeded, or
1137  * failed with a seqid incrementing error -
1138  * see comments nfs_fs.h:seqid_mutating_error()
1139  */
1140 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1141 {
1142         nfs_increment_seqid(status, seqid);
1143 }
1144
1145 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1146 {
1147         struct nfs_seqid_counter *sequence = seqid->sequence;
1148         int status = 0;
1149
1150         spin_lock(&sequence->lock);
1151         seqid->task = task;
1152         if (list_empty(&seqid->list))
1153                 list_add_tail(&seqid->list, &sequence->list);
1154         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1155                 goto unlock;
1156         rpc_sleep_on(&sequence->wait, task, NULL);
1157         status = -EAGAIN;
1158 unlock:
1159         spin_unlock(&sequence->lock);
1160         return status;
1161 }
1162
1163 static int nfs4_run_state_manager(void *);
1164
1165 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1166 {
1167         smp_mb__before_clear_bit();
1168         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1169         smp_mb__after_clear_bit();
1170         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1171         rpc_wake_up(&clp->cl_rpcwaitq);
1172 }
1173
1174 /*
1175  * Schedule the nfs_client asynchronous state management routine
1176  */
1177 void nfs4_schedule_state_manager(struct nfs_client *clp)
1178 {
1179         struct task_struct *task;
1180         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1181
1182         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1183                 return;
1184         __module_get(THIS_MODULE);
1185         atomic_inc(&clp->cl_count);
1186
1187         /* The rcu_read_lock() is not strictly necessary, as the state
1188          * manager is the only thread that ever changes the rpc_xprt
1189          * after it's initialized.  At this point, we're single threaded. */
1190         rcu_read_lock();
1191         snprintf(buf, sizeof(buf), "%s-manager",
1192                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1193         rcu_read_unlock();
1194         task = kthread_run(nfs4_run_state_manager, clp, buf);
1195         if (IS_ERR(task)) {
1196                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1197                         __func__, PTR_ERR(task));
1198                 nfs4_clear_state_manager_bit(clp);
1199                 nfs_put_client(clp);
1200                 module_put(THIS_MODULE);
1201         }
1202 }
1203
1204 /*
1205  * Schedule a lease recovery attempt
1206  */
1207 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1208 {
1209         if (!clp)
1210                 return;
1211         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1212                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1213         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1214                         clp->cl_hostname);
1215         nfs4_schedule_state_manager(clp);
1216 }
1217 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1218
1219 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1220 {
1221         int res;
1222
1223         might_sleep();
1224
1225         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1226                         nfs_wait_bit_killable, TASK_KILLABLE);
1227         if (res)
1228                 return res;
1229
1230         if (clp->cl_cons_state < 0)
1231                 return clp->cl_cons_state;
1232         return 0;
1233 }
1234
1235 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1236 {
1237         unsigned int loop;
1238         int ret;
1239
1240         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1241                 ret = nfs4_wait_clnt_recover(clp);
1242                 if (ret != 0)
1243                         break;
1244                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1245                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1246                         break;
1247                 nfs4_schedule_state_manager(clp);
1248                 ret = -EIO;
1249         }
1250         return ret;
1251 }
1252
1253 /*
1254  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1255  * @clp: client to process
1256  *
1257  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1258  * resend of the SETCLIENTID and hence re-establish the
1259  * callback channel. Then return all existing delegations.
1260  */
1261 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1262 {
1263         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1264         nfs_expire_all_delegations(clp);
1265         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1266                         clp->cl_hostname);
1267 }
1268
1269 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1270 {
1271         nfs40_handle_cb_pathdown(clp);
1272         nfs4_schedule_state_manager(clp);
1273 }
1274
1275 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1276 {
1277
1278         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1279         /* Don't recover state that expired before the reboot */
1280         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1281                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1282                 return 0;
1283         }
1284         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1285         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1286         return 1;
1287 }
1288
1289 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1290 {
1291         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1292         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1293         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1294         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1295         return 1;
1296 }
1297
1298 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1299 {
1300         struct nfs_client *clp = server->nfs_client;
1301
1302         nfs4_state_mark_reclaim_nograce(clp, state);
1303         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1304                         clp->cl_hostname);
1305         nfs4_schedule_state_manager(clp);
1306 }
1307 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1308
1309 void nfs_inode_find_state_and_recover(struct inode *inode,
1310                 const nfs4_stateid *stateid)
1311 {
1312         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1313         struct nfs_inode *nfsi = NFS_I(inode);
1314         struct nfs_open_context *ctx;
1315         struct nfs4_state *state;
1316         bool found = false;
1317
1318         spin_lock(&inode->i_lock);
1319         list_for_each_entry(ctx, &nfsi->open_files, list) {
1320                 state = ctx->state;
1321                 if (state == NULL)
1322                         continue;
1323                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1324                         continue;
1325                 if (!nfs4_stateid_match(&state->stateid, stateid))
1326                         continue;
1327                 nfs4_state_mark_reclaim_nograce(clp, state);
1328                 found = true;
1329         }
1330         spin_unlock(&inode->i_lock);
1331         if (found)
1332                 nfs4_schedule_state_manager(clp);
1333 }
1334
1335
1336 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1337 {
1338         struct inode *inode = state->inode;
1339         struct nfs_inode *nfsi = NFS_I(inode);
1340         struct file_lock *fl;
1341         int status = 0;
1342
1343         if (inode->i_flock == NULL)
1344                 return 0;
1345
1346         /* Guard against delegation returns and new lock/unlock calls */
1347         down_write(&nfsi->rwsem);
1348         /* Protect inode->i_flock using the BKL */
1349         lock_flocks();
1350         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1351                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1352                         continue;
1353                 if (nfs_file_open_context(fl->fl_file)->state != state)
1354                         continue;
1355                 unlock_flocks();
1356                 status = ops->recover_lock(state, fl);
1357                 switch (status) {
1358                         case 0:
1359                                 break;
1360                         case -ESTALE:
1361                         case -NFS4ERR_ADMIN_REVOKED:
1362                         case -NFS4ERR_STALE_STATEID:
1363                         case -NFS4ERR_BAD_STATEID:
1364                         case -NFS4ERR_EXPIRED:
1365                         case -NFS4ERR_NO_GRACE:
1366                         case -NFS4ERR_STALE_CLIENTID:
1367                         case -NFS4ERR_BADSESSION:
1368                         case -NFS4ERR_BADSLOT:
1369                         case -NFS4ERR_BAD_HIGH_SLOT:
1370                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1371                                 goto out;
1372                         default:
1373                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1374                                         "Zeroing state\n", __func__, status);
1375                         case -ENOMEM:
1376                         case -NFS4ERR_DENIED:
1377                         case -NFS4ERR_RECLAIM_BAD:
1378                         case -NFS4ERR_RECLAIM_CONFLICT:
1379                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1380                                 status = 0;
1381                 }
1382                 lock_flocks();
1383         }
1384         unlock_flocks();
1385 out:
1386         up_write(&nfsi->rwsem);
1387         return status;
1388 }
1389
1390 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1391 {
1392         struct nfs4_state *state;
1393         struct nfs4_lock_state *lock;
1394         int status = 0;
1395
1396         /* Note: we rely on the sp->so_states list being ordered
1397          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1398          * states first.
1399          * This is needed to ensure that the server won't give us any
1400          * read delegations that we have to return if, say, we are
1401          * recovering after a network partition or a reboot from a
1402          * server that doesn't support a grace period.
1403          */
1404 restart:
1405         spin_lock(&sp->so_lock);
1406         list_for_each_entry(state, &sp->so_states, open_states) {
1407                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1408                         continue;
1409                 if (state->state == 0)
1410                         continue;
1411                 atomic_inc(&state->count);
1412                 spin_unlock(&sp->so_lock);
1413                 status = ops->recover_open(sp, state);
1414                 if (status >= 0) {
1415                         status = nfs4_reclaim_locks(state, ops);
1416                         if (status >= 0) {
1417                                 spin_lock(&state->state_lock);
1418                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1419                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1420                                                 pr_warn_ratelimited("NFS: "
1421                                                         "%s: Lock reclaim "
1422                                                         "failed!\n", __func__);
1423                                 }
1424                                 spin_unlock(&state->state_lock);
1425                                 nfs4_put_open_state(state);
1426                                 goto restart;
1427                         }
1428                 }
1429                 switch (status) {
1430                         default:
1431                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1432                                         "Zeroing state\n", __func__, status);
1433                         case -ENOENT:
1434                         case -ENOMEM:
1435                         case -ESTALE:
1436                                 /*
1437                                  * Open state on this file cannot be recovered
1438                                  * All we can do is revert to using the zero stateid.
1439                                  */
1440                                 memset(&state->stateid, 0,
1441                                         sizeof(state->stateid));
1442                                 /* Mark the file as being 'closed' */
1443                                 state->state = 0;
1444                                 break;
1445                         case -EKEYEXPIRED:
1446                                 /*
1447                                  * User RPCSEC_GSS context has expired.
1448                                  * We cannot recover this stateid now, so
1449                                  * skip it and allow recovery thread to
1450                                  * proceed.
1451                                  */
1452                                 break;
1453                         case -NFS4ERR_ADMIN_REVOKED:
1454                         case -NFS4ERR_STALE_STATEID:
1455                         case -NFS4ERR_BAD_STATEID:
1456                         case -NFS4ERR_RECLAIM_BAD:
1457                         case -NFS4ERR_RECLAIM_CONFLICT:
1458                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1459                                 break;
1460                         case -NFS4ERR_EXPIRED:
1461                         case -NFS4ERR_NO_GRACE:
1462                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1463                         case -NFS4ERR_STALE_CLIENTID:
1464                         case -NFS4ERR_BADSESSION:
1465                         case -NFS4ERR_BADSLOT:
1466                         case -NFS4ERR_BAD_HIGH_SLOT:
1467                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1468                                 goto out_err;
1469                 }
1470                 nfs4_put_open_state(state);
1471                 goto restart;
1472         }
1473         spin_unlock(&sp->so_lock);
1474         return 0;
1475 out_err:
1476         nfs4_put_open_state(state);
1477         return status;
1478 }
1479
1480 static void nfs4_clear_open_state(struct nfs4_state *state)
1481 {
1482         struct nfs4_lock_state *lock;
1483
1484         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1485         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1486         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1487         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1488         spin_lock(&state->state_lock);
1489         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1490                 lock->ls_seqid.flags = 0;
1491                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1492         }
1493         spin_unlock(&state->state_lock);
1494 }
1495
1496 static void nfs4_reset_seqids(struct nfs_server *server,
1497         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1498 {
1499         struct nfs_client *clp = server->nfs_client;
1500         struct nfs4_state_owner *sp;
1501         struct rb_node *pos;
1502         struct nfs4_state *state;
1503
1504         spin_lock(&clp->cl_lock);
1505         for (pos = rb_first(&server->state_owners);
1506              pos != NULL;
1507              pos = rb_next(pos)) {
1508                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1509                 sp->so_seqid.flags = 0;
1510                 spin_lock(&sp->so_lock);
1511                 list_for_each_entry(state, &sp->so_states, open_states) {
1512                         if (mark_reclaim(clp, state))
1513                                 nfs4_clear_open_state(state);
1514                 }
1515                 spin_unlock(&sp->so_lock);
1516         }
1517         spin_unlock(&clp->cl_lock);
1518 }
1519
1520 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1521         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1522 {
1523         struct nfs_server *server;
1524
1525         rcu_read_lock();
1526         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1527                 nfs4_reset_seqids(server, mark_reclaim);
1528         rcu_read_unlock();
1529 }
1530
1531 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1532 {
1533         /* Mark all delegations for reclaim */
1534         nfs_delegation_mark_reclaim(clp);
1535         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1536 }
1537
1538 static void nfs4_reclaim_complete(struct nfs_client *clp,
1539                                  const struct nfs4_state_recovery_ops *ops)
1540 {
1541         /* Notify the server we're done reclaiming our state */
1542         if (ops->reclaim_complete)
1543                 (void)ops->reclaim_complete(clp);
1544 }
1545
1546 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1547 {
1548         struct nfs_client *clp = server->nfs_client;
1549         struct nfs4_state_owner *sp;
1550         struct rb_node *pos;
1551         struct nfs4_state *state;
1552
1553         spin_lock(&clp->cl_lock);
1554         for (pos = rb_first(&server->state_owners);
1555              pos != NULL;
1556              pos = rb_next(pos)) {
1557                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1558                 spin_lock(&sp->so_lock);
1559                 list_for_each_entry(state, &sp->so_states, open_states) {
1560                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1561                                                 &state->flags))
1562                                 continue;
1563                         nfs4_state_mark_reclaim_nograce(clp, state);
1564                 }
1565                 spin_unlock(&sp->so_lock);
1566         }
1567         spin_unlock(&clp->cl_lock);
1568 }
1569
1570 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1571 {
1572         struct nfs_server *server;
1573
1574         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1575                 return 0;
1576
1577         rcu_read_lock();
1578         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1579                 nfs4_clear_reclaim_server(server);
1580         rcu_read_unlock();
1581
1582         nfs_delegation_reap_unclaimed(clp);
1583         return 1;
1584 }
1585
1586 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1587 {
1588         if (!nfs4_state_clear_reclaim_reboot(clp))
1589                 return;
1590         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1591 }
1592
1593 static void nfs_delegation_clear_all(struct nfs_client *clp)
1594 {
1595         nfs_delegation_mark_reclaim(clp);
1596         nfs_delegation_reap_unclaimed(clp);
1597 }
1598
1599 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1600 {
1601         nfs_delegation_clear_all(clp);
1602         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1603 }
1604
1605 static void nfs4_warn_keyexpired(const char *s)
1606 {
1607         printk_ratelimited(KERN_WARNING "Error: state manager"
1608                         " encountered RPCSEC_GSS session"
1609                         " expired against NFSv4 server %s.\n",
1610                         s);
1611 }
1612
1613 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1614 {
1615         switch (error) {
1616                 case 0:
1617                         break;
1618                 case -NFS4ERR_CB_PATH_DOWN:
1619                         nfs40_handle_cb_pathdown(clp);
1620                         break;
1621                 case -NFS4ERR_NO_GRACE:
1622                         nfs4_state_end_reclaim_reboot(clp);
1623                         break;
1624                 case -NFS4ERR_STALE_CLIENTID:
1625                 case -NFS4ERR_LEASE_MOVED:
1626                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1627                         nfs4_state_clear_reclaim_reboot(clp);
1628                         nfs4_state_start_reclaim_reboot(clp);
1629                         break;
1630                 case -NFS4ERR_EXPIRED:
1631                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1632                         nfs4_state_start_reclaim_nograce(clp);
1633                         break;
1634                 case -NFS4ERR_BADSESSION:
1635                 case -NFS4ERR_BADSLOT:
1636                 case -NFS4ERR_BAD_HIGH_SLOT:
1637                 case -NFS4ERR_DEADSESSION:
1638                 case -NFS4ERR_SEQ_FALSE_RETRY:
1639                 case -NFS4ERR_SEQ_MISORDERED:
1640                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1641                         /* Zero session reset errors */
1642                         break;
1643                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1644                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1645                         break;
1646                 case -EKEYEXPIRED:
1647                         /* Nothing we can do */
1648                         nfs4_warn_keyexpired(clp->cl_hostname);
1649                         break;
1650                 default:
1651                         dprintk("%s: failed to handle error %d for server %s\n",
1652                                         __func__, error, clp->cl_hostname);
1653                         return error;
1654         }
1655         dprintk("%s: handled error %d for server %s\n", __func__, error,
1656                         clp->cl_hostname);
1657         return 0;
1658 }
1659
1660 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1661 {
1662         struct nfs4_state_owner *sp;
1663         struct nfs_server *server;
1664         struct rb_node *pos;
1665         int status = 0;
1666
1667 restart:
1668         rcu_read_lock();
1669         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1670                 nfs4_purge_state_owners(server);
1671                 spin_lock(&clp->cl_lock);
1672                 for (pos = rb_first(&server->state_owners);
1673                      pos != NULL;
1674                      pos = rb_next(pos)) {
1675                         sp = rb_entry(pos,
1676                                 struct nfs4_state_owner, so_server_node);
1677                         if (!test_and_clear_bit(ops->owner_flag_bit,
1678                                                         &sp->so_flags))
1679                                 continue;
1680                         atomic_inc(&sp->so_count);
1681                         spin_unlock(&clp->cl_lock);
1682                         rcu_read_unlock();
1683
1684                         status = nfs4_reclaim_open_state(sp, ops);
1685                         if (status < 0) {
1686                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1687                                 nfs4_put_state_owner(sp);
1688                                 return nfs4_recovery_handle_error(clp, status);
1689                         }
1690
1691                         nfs4_put_state_owner(sp);
1692                         goto restart;
1693                 }
1694                 spin_unlock(&clp->cl_lock);
1695         }
1696         rcu_read_unlock();
1697         return status;
1698 }
1699
1700 static int nfs4_check_lease(struct nfs_client *clp)
1701 {
1702         struct rpc_cred *cred;
1703         const struct nfs4_state_maintenance_ops *ops =
1704                 clp->cl_mvops->state_renewal_ops;
1705         int status;
1706
1707         /* Is the client already known to have an expired lease? */
1708         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1709                 return 0;
1710         spin_lock(&clp->cl_lock);
1711         cred = ops->get_state_renewal_cred_locked(clp);
1712         spin_unlock(&clp->cl_lock);
1713         if (cred == NULL) {
1714                 cred = nfs4_get_setclientid_cred(clp);
1715                 status = -ENOKEY;
1716                 if (cred == NULL)
1717                         goto out;
1718         }
1719         status = ops->renew_lease(clp, cred);
1720         put_rpccred(cred);
1721 out:
1722         return nfs4_recovery_handle_error(clp, status);
1723 }
1724
1725 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1726  * and for recoverable errors on EXCHANGE_ID for v4.1
1727  */
1728 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1729 {
1730         switch (status) {
1731         case -NFS4ERR_SEQ_MISORDERED:
1732                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1733                         return -ESERVERFAULT;
1734                 /* Lease confirmation error: retry after purging the lease */
1735                 ssleep(1);
1736                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1737                 break;
1738         case -NFS4ERR_STALE_CLIENTID:
1739                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1740                 nfs4_state_clear_reclaim_reboot(clp);
1741                 nfs4_state_start_reclaim_reboot(clp);
1742                 break;
1743         case -NFS4ERR_CLID_INUSE:
1744                 pr_err("NFS: Server %s reports our clientid is in use\n",
1745                         clp->cl_hostname);
1746                 nfs_mark_client_ready(clp, -EPERM);
1747                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1748                 return -EPERM;
1749         case -EACCES:
1750                 if (clp->cl_machine_cred == NULL)
1751                         return -EACCES;
1752                 /* Handle case where the user hasn't set up machine creds */
1753                 nfs4_clear_machine_cred(clp);
1754         case -NFS4ERR_DELAY:
1755         case -ETIMEDOUT:
1756         case -EAGAIN:
1757                 ssleep(1);
1758                 break;
1759
1760         case -NFS4ERR_MINOR_VERS_MISMATCH:
1761                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1762                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1763                 dprintk("%s: exit with error %d for server %s\n",
1764                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1765                 return -EPROTONOSUPPORT;
1766         case -EKEYEXPIRED:
1767                 nfs4_warn_keyexpired(clp->cl_hostname);
1768         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1769                                  * in nfs4_exchange_id */
1770         default:
1771                 dprintk("%s: exit with error %d for server %s\n", __func__,
1772                                 status, clp->cl_hostname);
1773                 return status;
1774         }
1775         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1776         dprintk("%s: handled error %d for server %s\n", __func__, status,
1777                         clp->cl_hostname);
1778         return 0;
1779 }
1780
1781 static int nfs4_establish_lease(struct nfs_client *clp)
1782 {
1783         struct rpc_cred *cred;
1784         const struct nfs4_state_recovery_ops *ops =
1785                 clp->cl_mvops->reboot_recovery_ops;
1786         int status;
1787
1788         cred = ops->get_clid_cred(clp);
1789         if (cred == NULL)
1790                 return -ENOENT;
1791         status = ops->establish_clid(clp, cred);
1792         put_rpccred(cred);
1793         if (status != 0)
1794                 return status;
1795         pnfs_destroy_all_layouts(clp);
1796         return 0;
1797 }
1798
1799 /*
1800  * Returns zero or a negative errno.  NFS4ERR values are converted
1801  * to local errno values.
1802  */
1803 static int nfs4_reclaim_lease(struct nfs_client *clp)
1804 {
1805         int status;
1806
1807         status = nfs4_establish_lease(clp);
1808         if (status < 0)
1809                 return nfs4_handle_reclaim_lease_error(clp, status);
1810         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1811                 nfs4_state_start_reclaim_nograce(clp);
1812         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1813                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1814         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1815         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1816         return 0;
1817 }
1818
1819 static int nfs4_purge_lease(struct nfs_client *clp)
1820 {
1821         int status;
1822
1823         status = nfs4_establish_lease(clp);
1824         if (status < 0)
1825                 return nfs4_handle_reclaim_lease_error(clp, status);
1826         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1827         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1828         nfs4_state_start_reclaim_nograce(clp);
1829         return 0;
1830 }
1831
1832 /**
1833  * nfs4_discover_server_trunking - Detect server IP address trunking
1834  *
1835  * @clp: nfs_client under test
1836  * @result: OUT: found nfs_client, or clp
1837  *
1838  * Returns zero or a negative errno.  If zero is returned,
1839  * an nfs_client pointer is planted in "result".
1840  *
1841  * Note: since we are invoked in process context, and
1842  * not from inside the state manager, we cannot use
1843  * nfs4_handle_reclaim_lease_error().
1844  */
1845 int nfs4_discover_server_trunking(struct nfs_client *clp,
1846                                   struct nfs_client **result)
1847 {
1848         const struct nfs4_state_recovery_ops *ops =
1849                                 clp->cl_mvops->reboot_recovery_ops;
1850         rpc_authflavor_t *flavors, flav, save;
1851         struct rpc_clnt *clnt;
1852         struct rpc_cred *cred;
1853         int i, len, status;
1854
1855         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1856
1857         len = NFS_MAX_SECFLAVORS;
1858         flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1859         if (flavors == NULL) {
1860                 status = -ENOMEM;
1861                 goto out;
1862         }
1863         len = rpcauth_list_flavors(flavors, len);
1864         if (len < 0) {
1865                 status = len;
1866                 goto out_free;
1867         }
1868         clnt = clp->cl_rpcclient;
1869         save = clnt->cl_auth->au_flavor;
1870         i = 0;
1871
1872         mutex_lock(&nfs_clid_init_mutex);
1873         status  = -ENOENT;
1874 again:
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
1885         case -EACCES:
1886                 if (clp->cl_machine_cred == NULL)
1887                         break;
1888                 /* Handle case where the user hasn't set up machine creds */
1889                 nfs4_clear_machine_cred(clp);
1890         case -NFS4ERR_DELAY:
1891         case -ETIMEDOUT:
1892         case -EAGAIN:
1893                 ssleep(1);
1894                 dprintk("NFS: %s after status %d, retrying\n",
1895                         __func__, status);
1896                 goto again;
1897
1898         case -NFS4ERR_CLID_INUSE:
1899         case -NFS4ERR_WRONGSEC:
1900                 status = -EPERM;
1901                 if (i >= len)
1902                         break;
1903
1904                 flav = flavors[i++];
1905                 if (flav == save)
1906                         flav = flavors[i++];
1907                 clnt = rpc_clone_client_set_auth(clnt, flav);
1908                 if (IS_ERR(clnt)) {
1909                         status = PTR_ERR(clnt);
1910                         break;
1911                 }
1912                 clp->cl_rpcclient = clnt;
1913                 goto again;
1914
1915         case -NFS4ERR_MINOR_VERS_MISMATCH:
1916                 status = -EPROTONOSUPPORT;
1917                 break;
1918
1919         case -EKEYEXPIRED:
1920                 nfs4_warn_keyexpired(clp->cl_hostname);
1921         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1922                                  * in nfs4_exchange_id */
1923                 status = -EKEYEXPIRED;
1924         }
1925
1926 out_unlock:
1927         mutex_unlock(&nfs_clid_init_mutex);
1928 out_free:
1929         kfree(flavors);
1930 out:
1931         dprintk("NFS: %s: status = %d\n", __func__, status);
1932         return status;
1933 }
1934
1935 #ifdef CONFIG_NFS_V4_1
1936 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1937 {
1938         struct nfs_client *clp = session->clp;
1939
1940         switch (err) {
1941         default:
1942                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1943                 break;
1944         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1945                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1946         }
1947         nfs4_schedule_lease_recovery(clp);
1948 }
1949 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1950
1951 static void nfs41_ping_server(struct nfs_client *clp)
1952 {
1953         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1954         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1955         nfs4_schedule_state_manager(clp);
1956 }
1957
1958 void nfs41_server_notify_target_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         if (status && status != -NFS4ERR_BADSESSION &&
2050             status != -NFS4ERR_DEADSESSION) {
2051                 status = nfs4_recovery_handle_error(clp, status);
2052                 goto out;
2053         }
2054
2055         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2056         status = nfs4_proc_create_session(clp, cred);
2057         if (status) {
2058                 dprintk("%s: session reset failed with status %d for server %s!\n",
2059                         __func__, status, clp->cl_hostname);
2060                 status = nfs4_handle_reclaim_lease_error(clp, status);
2061                 goto out;
2062         }
2063         nfs41_finish_session_reset(clp);
2064         dprintk("%s: session reset was successful for server %s!\n",
2065                         __func__, clp->cl_hostname);
2066 out:
2067         if (cred)
2068                 put_rpccred(cred);
2069         return status;
2070 }
2071
2072 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2073 {
2074         struct rpc_cred *cred;
2075         int ret;
2076
2077         if (!nfs4_has_session(clp))
2078                 return 0;
2079         nfs4_begin_drain_session(clp);
2080         cred = nfs4_get_exchange_id_cred(clp);
2081         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2082         if (cred)
2083                 put_rpccred(cred);
2084         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2085         switch (ret) {
2086         case 0:
2087                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2088                         __func__, clp->cl_hostname);
2089                 break;
2090         case -NFS4ERR_DELAY:
2091                 ssleep(1);
2092                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2093                 break;
2094         default:
2095                 return nfs4_recovery_handle_error(clp, ret);
2096         }
2097         return 0;
2098 }
2099 #else /* CONFIG_NFS_V4_1 */
2100 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2101 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2102
2103 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2104 {
2105         return 0;
2106 }
2107 #endif /* CONFIG_NFS_V4_1 */
2108
2109 static void nfs4_state_manager(struct nfs_client *clp)
2110 {
2111         int status = 0;
2112         const char *section = "", *section_sep = "";
2113
2114         /* Ensure exclusive access to NFSv4 state */
2115         do {
2116                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2117                         section = "purge state";
2118                         status = nfs4_purge_lease(clp);
2119                         if (status < 0)
2120                                 goto out_error;
2121                         continue;
2122                 }
2123
2124                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2125                         section = "lease expired";
2126                         /* We're going to have to re-establish a clientid */
2127                         status = nfs4_reclaim_lease(clp);
2128                         if (status < 0)
2129                                 goto out_error;
2130                         continue;
2131                 }
2132
2133                 /* Initialize or reset the session */
2134                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2135                         section = "reset session";
2136                         status = nfs4_reset_session(clp);
2137                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2138                                 continue;
2139                         if (status < 0)
2140                                 goto out_error;
2141                 }
2142
2143                 /* Send BIND_CONN_TO_SESSION */
2144                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2145                                 &clp->cl_state)) {
2146                         section = "bind conn to session";
2147                         status = nfs4_bind_conn_to_session(clp);
2148                         if (status < 0)
2149                                 goto out_error;
2150                         continue;
2151                 }
2152
2153                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2154                         section = "check lease";
2155                         status = nfs4_check_lease(clp);
2156                         if (status < 0)
2157                                 goto out_error;
2158                         continue;
2159                 }
2160
2161                 /* First recover reboot state... */
2162                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2163                         section = "reclaim reboot";
2164                         status = nfs4_do_reclaim(clp,
2165                                 clp->cl_mvops->reboot_recovery_ops);
2166                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2167                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2168                                 continue;
2169                         nfs4_state_end_reclaim_reboot(clp);
2170                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2171                                 continue;
2172                         if (status < 0)
2173                                 goto out_error;
2174                 }
2175
2176                 /* Now recover expired state... */
2177                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2178                         section = "reclaim nograce";
2179                         status = nfs4_do_reclaim(clp,
2180                                 clp->cl_mvops->nograce_recovery_ops);
2181                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2182                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2183                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2184                                 continue;
2185                         if (status < 0)
2186                                 goto out_error;
2187                 }
2188
2189                 nfs4_end_drain_session(clp);
2190                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2191                         nfs_client_return_marked_delegations(clp);
2192                         continue;
2193                 }
2194
2195                 nfs4_clear_state_manager_bit(clp);
2196                 /* Did we race with an attempt to give us more work? */
2197                 if (clp->cl_state == 0)
2198                         break;
2199                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2200                         break;
2201         } while (atomic_read(&clp->cl_count) > 1);
2202         return;
2203 out_error:
2204         if (strlen(section))
2205                 section_sep = ": ";
2206         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2207                         " with error %d\n", section_sep, section,
2208                         clp->cl_hostname, -status);
2209         ssleep(1);
2210         nfs4_end_drain_session(clp);
2211         nfs4_clear_state_manager_bit(clp);
2212 }
2213
2214 static int nfs4_run_state_manager(void *ptr)
2215 {
2216         struct nfs_client *clp = ptr;
2217
2218         allow_signal(SIGKILL);
2219         nfs4_state_manager(clp);
2220         nfs_put_client(clp);
2221         module_put_and_exit(0);
2222         return 0;
2223 }
2224
2225 /*
2226  * Local variables:
2227  *  c-basic-offset: 8
2228  * End:
2229  */