Pileus Git - ~andy/linux/blob - fs/nfs/nfs4state.c

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