fs/nfs/nfs4state.c

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