fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52 #include <linux/jiffies.h>
  53
  54 #include <linux/sunrpc/clnt.h>
  55
  56 #include "nfs4_fs.h"
  57 #include "callback.h"
  58 #include "delegation.h"
  59 #include "internal.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid;
  69 static DEFINE_MUTEX(nfs_clid_init_mutex);
  70
  71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  72 {
  73         struct nfs4_setclientid_res clid = {
  74                 .clientid = clp->cl_clientid,
  75                 .confirm = clp->cl_confirm,
  76         };
  77         unsigned short port;
  78         int status;
  79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  80
  81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  82                 goto do_confirm;
  83         port = nn->nfs_callback_tcpport;
  84         if (clp->cl_addr.ss_family == AF_INET6)
  85                 port = nn->nfs_callback_tcpport6;
  86
  87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  88         if (status != 0)
  89                 goto out;
  90         clp->cl_clientid = clid.clientid;
  91         clp->cl_confirm = clid.confirm;
  92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  93 do_confirm:
  94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  95         if (status != 0)
  96                 goto out;
  97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98         nfs4_schedule_state_renewal(clp);
  99 out:
 100         return status;
 101 }
 102
 103 /**
 104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 105  *
 106  * @clp: nfs_client under test
 107  * @result: OUT: found nfs_client, or clp
 108  * @cred: credential to use for trunking test
 109  *
 110  * Returns zero, a negative errno, or a negative NFS4ERR status.
 111  * If zero is returned, an nfs_client pointer is planted in
 112  * "result".
 113  *
 114  * Note: The returned client may not yet be marked ready.
 115  */
 116 int nfs40_discover_server_trunking(struct nfs_client *clp,
 117                                    struct nfs_client **result,
 118                                    struct rpc_cred *cred)
 119 {
 120         struct nfs4_setclientid_res clid = {
 121                 .clientid = clp->cl_clientid,
 122                 .confirm = clp->cl_confirm,
 123         };
 124         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 125         unsigned short port;
 126         int status;
 127
 128         port = nn->nfs_callback_tcpport;
 129         if (clp->cl_addr.ss_family == AF_INET6)
 130                 port = nn->nfs_callback_tcpport6;
 131
 132         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 133         if (status != 0)
 134                 goto out;
 135         clp->cl_clientid = clid.clientid;
 136         clp->cl_confirm = clid.confirm;
 137
 138         status = nfs40_walk_client_list(clp, result, cred);
 139         if (status == 0) {
 140                 /* Sustain the lease, even if it's empty.  If the clientid4
 141                  * goes stale it's of no use for trunking discovery. */
 142                 nfs4_schedule_state_renewal(*result);
 143         }
 144 out:
 145         return status;
 146 }
 147
 148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 149 {
 150         struct rpc_cred *cred = NULL;
 151
 152         if (clp->cl_machine_cred != NULL)
 153                 cred = get_rpccred(clp->cl_machine_cred);
 154         return cred;
 155 }
 156
 157 static struct rpc_cred *
 158 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 159 {
 160         struct rpc_cred *cred = NULL;
 161         struct nfs4_state_owner *sp;
 162         struct rb_node *pos;
 163
 164         for (pos = rb_first(&server->state_owners);
 165              pos != NULL;
 166              pos = rb_next(pos)) {
 167                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 168                 if (list_empty(&sp->so_states))
 169                         continue;
 170                 cred = get_rpccred(sp->so_cred);
 171                 break;
 172         }
 173         return cred;
 174 }
 175
 176 /**
 177  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 178  * @clp: client state handle
 179  *
 180  * Returns an rpc_cred with reference count bumped, or NULL.
 181  * Caller must hold clp->cl_lock.
 182  */
 183 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 184 {
 185         struct rpc_cred *cred = NULL;
 186         struct nfs_server *server;
 187
 188         /* Use machine credentials if available */
 189         cred = nfs4_get_machine_cred_locked(clp);
 190         if (cred != NULL)
 191                 goto out;
 192
 193         rcu_read_lock();
 194         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 195                 cred = nfs4_get_renew_cred_server_locked(server);
 196                 if (cred != NULL)
 197                         break;
 198         }
 199         rcu_read_unlock();
 200
 201 out:
 202         return cred;
 203 }
 204
 205 #if defined(CONFIG_NFS_V4_1)
 206
 207 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 208 {
 209         int status;
 210         struct nfs_fsinfo fsinfo;
 211
 212         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 213                 nfs4_schedule_state_renewal(clp);
 214                 return 0;
 215         }
 216
 217         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 218         if (status == 0) {
 219                 /* Update lease time and schedule renewal */
 220                 spin_lock(&clp->cl_lock);
 221                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 222                 clp->cl_last_renewal = jiffies;
 223                 spin_unlock(&clp->cl_lock);
 224
 225                 nfs4_schedule_state_renewal(clp);
 226         }
 227
 228         return status;
 229 }
 230
 231 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 232 {
 233         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 234                 spin_lock(&tbl->slot_tbl_lock);
 235                 nfs41_wake_slot_table(tbl);
 236                 spin_unlock(&tbl->slot_tbl_lock);
 237         }
 238 }
 239
 240 static void nfs4_end_drain_session(struct nfs_client *clp)
 241 {
 242         struct nfs4_session *ses = clp->cl_session;
 243
 244         if (ses != NULL) {
 245                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 246                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 247         }
 248 }
 249
 250 /*
 251  * Signal state manager thread if session fore channel is drained
 252  */
 253 void nfs4_slot_tbl_drain_complete(struct nfs4_slot_table *tbl)
 254 {
 255         if (nfs4_slot_tbl_draining(tbl))
 256                 complete(&tbl->complete);
 257 }
 258
 259 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 260 {
 261         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 262         spin_lock(&tbl->slot_tbl_lock);
 263         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 264                 INIT_COMPLETION(tbl->complete);
 265                 spin_unlock(&tbl->slot_tbl_lock);
 266                 return wait_for_completion_interruptible(&tbl->complete);
 267         }
 268         spin_unlock(&tbl->slot_tbl_lock);
 269         return 0;
 270 }
 271
 272 static int nfs4_begin_drain_session(struct nfs_client *clp)
 273 {
 274         struct nfs4_session *ses = clp->cl_session;
 275         int ret = 0;
 276
 277         /* back channel */
 278         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 279         if (ret)
 280                 return ret;
 281         /* fore channel */
 282         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 283 }
 284
 285 static void nfs41_finish_session_reset(struct nfs_client *clp)
 286 {
 287         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 288         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 289         /* create_session negotiated new slot table */
 290         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 291         nfs41_setup_state_renewal(clp);
 292 }
 293
 294 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 295 {
 296         int status;
 297
 298         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 299                 goto do_confirm;
 300         nfs4_begin_drain_session(clp);
 301         status = nfs4_proc_exchange_id(clp, cred);
 302         if (status != 0)
 303                 goto out;
 304         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 305 do_confirm:
 306         status = nfs4_proc_create_session(clp, cred);
 307         if (status != 0)
 308                 goto out;
 309         nfs41_finish_session_reset(clp);
 310         nfs_mark_client_ready(clp, NFS_CS_READY);
 311 out:
 312         return status;
 313 }
 314
 315 /**
 316  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 317  *
 318  * @clp: nfs_client under test
 319  * @result: OUT: found nfs_client, or clp
 320  * @cred: credential to use for trunking test
 321  *
 322  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 323  * If NFS4_OK is returned, an nfs_client pointer is planted in
 324  * "result".
 325  *
 326  * Note: The returned client may not yet be marked ready.
 327  */
 328 int nfs41_discover_server_trunking(struct nfs_client *clp,
 329                                    struct nfs_client **result,
 330                                    struct rpc_cred *cred)
 331 {
 332         int status;
 333
 334         status = nfs4_proc_exchange_id(clp, cred);
 335         if (status != NFS4_OK)
 336                 return status;
 337         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 338
 339         return nfs41_walk_client_list(clp, result, cred);
 340 }
 341
 342 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 343 {
 344         struct rpc_cred *cred;
 345
 346         spin_lock(&clp->cl_lock);
 347         cred = nfs4_get_machine_cred_locked(clp);
 348         spin_unlock(&clp->cl_lock);
 349         return cred;
 350 }
 351
 352 #endif /* CONFIG_NFS_V4_1 */
 353
 354 static struct rpc_cred *
 355 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 356 {
 357         struct nfs_client *clp = server->nfs_client;
 358         struct rpc_cred *cred = NULL;
 359         struct nfs4_state_owner *sp;
 360         struct rb_node *pos;
 361
 362         spin_lock(&clp->cl_lock);
 363         pos = rb_first(&server->state_owners);
 364         if (pos != NULL) {
 365                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 366                 cred = get_rpccred(sp->so_cred);
 367         }
 368         spin_unlock(&clp->cl_lock);
 369         return cred;
 370 }
 371
 372 /**
 373  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 374  * @clp: client state handle
 375  *
 376  * Returns an rpc_cred with reference count bumped, or NULL.
 377  */
 378 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 379 {
 380         struct nfs_server *server;
 381         struct rpc_cred *cred;
 382
 383         spin_lock(&clp->cl_lock);
 384         cred = nfs4_get_machine_cred_locked(clp);
 385         spin_unlock(&clp->cl_lock);
 386         if (cred != NULL)
 387                 goto out;
 388
 389         rcu_read_lock();
 390         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 391                 cred = nfs4_get_setclientid_cred_server(server);
 392                 if (cred != NULL)
 393                         break;
 394         }
 395         rcu_read_unlock();
 396
 397 out:
 398         return cred;
 399 }
 400
 401 static struct nfs4_state_owner *
 402 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 403 {
 404         struct rb_node **p = &server->state_owners.rb_node,
 405                        *parent = NULL;
 406         struct nfs4_state_owner *sp;
 407
 408         while (*p != NULL) {
 409                 parent = *p;
 410                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 411
 412                 if (cred < sp->so_cred)
 413                         p = &parent->rb_left;
 414                 else if (cred > sp->so_cred)
 415                         p = &parent->rb_right;
 416                 else {
 417                         if (!list_empty(&sp->so_lru))
 418                                 list_del_init(&sp->so_lru);
 419                         atomic_inc(&sp->so_count);
 420                         return sp;
 421                 }
 422         }
 423         return NULL;
 424 }
 425
 426 static struct nfs4_state_owner *
 427 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 428 {
 429         struct nfs_server *server = new->so_server;
 430         struct rb_node **p = &server->state_owners.rb_node,
 431                        *parent = NULL;
 432         struct nfs4_state_owner *sp;
 433         int err;
 434
 435         while (*p != NULL) {
 436                 parent = *p;
 437                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 438
 439                 if (new->so_cred < sp->so_cred)
 440                         p = &parent->rb_left;
 441                 else if (new->so_cred > sp->so_cred)
 442                         p = &parent->rb_right;
 443                 else {
 444                         if (!list_empty(&sp->so_lru))
 445                                 list_del_init(&sp->so_lru);
 446                         atomic_inc(&sp->so_count);
 447                         return sp;
 448                 }
 449         }
 450         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 451         if (err)
 452                 return ERR_PTR(err);
 453         rb_link_node(&new->so_server_node, parent, p);
 454         rb_insert_color(&new->so_server_node, &server->state_owners);
 455         return new;
 456 }
 457
 458 static void
 459 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 460 {
 461         struct nfs_server *server = sp->so_server;
 462
 463         if (!RB_EMPTY_NODE(&sp->so_server_node))
 464                 rb_erase(&sp->so_server_node, &server->state_owners);
 465         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 466 }
 467
 468 static void
 469 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 470 {
 471         sc->create_time = ktime_get();
 472         sc->flags = 0;
 473         sc->counter = 0;
 474         spin_lock_init(&sc->lock);
 475         INIT_LIST_HEAD(&sc->list);
 476         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 477 }
 478
 479 static void
 480 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 481 {
 482         rpc_destroy_wait_queue(&sc->wait);
 483 }
 484
 485 /*
 486  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 487  * create a new state_owner.
 488  *
 489  */
 490 static struct nfs4_state_owner *
 491 nfs4_alloc_state_owner(struct nfs_server *server,
 492                 struct rpc_cred *cred,
 493                 gfp_t gfp_flags)
 494 {
 495         struct nfs4_state_owner *sp;
 496
 497         sp = kzalloc(sizeof(*sp), gfp_flags);
 498         if (!sp)
 499                 return NULL;
 500         sp->so_server = server;
 501         sp->so_cred = get_rpccred(cred);
 502         spin_lock_init(&sp->so_lock);
 503         INIT_LIST_HEAD(&sp->so_states);
 504         nfs4_init_seqid_counter(&sp->so_seqid);
 505         atomic_set(&sp->so_count, 1);
 506         INIT_LIST_HEAD(&sp->so_lru);
 507         seqcount_init(&sp->so_reclaim_seqcount);
 508         mutex_init(&sp->so_delegreturn_mutex);
 509         return sp;
 510 }
 511
 512 static void
 513 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 514 {
 515         struct rb_node *rb_node = &sp->so_server_node;
 516
 517         if (!RB_EMPTY_NODE(rb_node)) {
 518                 struct nfs_server *server = sp->so_server;
 519                 struct nfs_client *clp = server->nfs_client;
 520
 521                 spin_lock(&clp->cl_lock);
 522                 if (!RB_EMPTY_NODE(rb_node)) {
 523                         rb_erase(rb_node, &server->state_owners);
 524                         RB_CLEAR_NODE(rb_node);
 525                 }
 526                 spin_unlock(&clp->cl_lock);
 527         }
 528 }
 529
 530 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 531 {
 532         nfs4_destroy_seqid_counter(&sp->so_seqid);
 533         put_rpccred(sp->so_cred);
 534         kfree(sp);
 535 }
 536
 537 static void nfs4_gc_state_owners(struct nfs_server *server)
 538 {
 539         struct nfs_client *clp = server->nfs_client;
 540         struct nfs4_state_owner *sp, *tmp;
 541         unsigned long time_min, time_max;
 542         LIST_HEAD(doomed);
 543
 544         spin_lock(&clp->cl_lock);
 545         time_max = jiffies;
 546         time_min = (long)time_max - (long)clp->cl_lease_time;
 547         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 548                 /* NB: LRU is sorted so that oldest is at the head */
 549                 if (time_in_range(sp->so_expires, time_min, time_max))
 550                         break;
 551                 list_move(&sp->so_lru, &doomed);
 552                 nfs4_remove_state_owner_locked(sp);
 553         }
 554         spin_unlock(&clp->cl_lock);
 555
 556         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 557                 list_del(&sp->so_lru);
 558                 nfs4_free_state_owner(sp);
 559         }
 560 }
 561
 562 /**
 563  * nfs4_get_state_owner - Look up a state owner given a credential
 564  * @server: nfs_server to search
 565  * @cred: RPC credential to match
 566  *
 567  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 568  */
 569 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 570                                               struct rpc_cred *cred,
 571                                               gfp_t gfp_flags)
 572 {
 573         struct nfs_client *clp = server->nfs_client;
 574         struct nfs4_state_owner *sp, *new;
 575
 576         spin_lock(&clp->cl_lock);
 577         sp = nfs4_find_state_owner_locked(server, cred);
 578         spin_unlock(&clp->cl_lock);
 579         if (sp != NULL)
 580                 goto out;
 581         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 582         if (new == NULL)
 583                 goto out;
 584         do {
 585                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 586                         break;
 587                 spin_lock(&clp->cl_lock);
 588                 sp = nfs4_insert_state_owner_locked(new);
 589                 spin_unlock(&clp->cl_lock);
 590         } while (sp == ERR_PTR(-EAGAIN));
 591         if (sp != new)
 592                 nfs4_free_state_owner(new);
 593 out:
 594         nfs4_gc_state_owners(server);
 595         return sp;
 596 }
 597
 598 /**
 599  * nfs4_put_state_owner - Release a nfs4_state_owner
 600  * @sp: state owner data to release
 601  *
 602  * Note that we keep released state owners on an LRU
 603  * list.
 604  * This caches valid state owners so that they can be
 605  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 606  * It also pins the uniquifier of dropped state owners for
 607  * a while, to ensure that those state owner names are
 608  * never reused.
 609  */
 610 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 611 {
 612         struct nfs_server *server = sp->so_server;
 613         struct nfs_client *clp = server->nfs_client;
 614
 615         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 616                 return;
 617
 618         sp->so_expires = jiffies;
 619         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 620         spin_unlock(&clp->cl_lock);
 621 }
 622
 623 /**
 624  * nfs4_purge_state_owners - Release all cached state owners
 625  * @server: nfs_server with cached state owners to release
 626  *
 627  * Called at umount time.  Remaining state owners will be on
 628  * the LRU with ref count of zero.
 629  */
 630 void nfs4_purge_state_owners(struct nfs_server *server)
 631 {
 632         struct nfs_client *clp = server->nfs_client;
 633         struct nfs4_state_owner *sp, *tmp;
 634         LIST_HEAD(doomed);
 635
 636         spin_lock(&clp->cl_lock);
 637         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 638                 list_move(&sp->so_lru, &doomed);
 639                 nfs4_remove_state_owner_locked(sp);
 640         }
 641         spin_unlock(&clp->cl_lock);
 642
 643         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 644                 list_del(&sp->so_lru);
 645                 nfs4_free_state_owner(sp);
 646         }
 647 }
 648
 649 static struct nfs4_state *
 650 nfs4_alloc_open_state(void)
 651 {
 652         struct nfs4_state *state;
 653
 654         state = kzalloc(sizeof(*state), GFP_NOFS);
 655         if (!state)
 656                 return NULL;
 657         atomic_set(&state->count, 1);
 658         INIT_LIST_HEAD(&state->lock_states);
 659         spin_lock_init(&state->state_lock);
 660         seqlock_init(&state->seqlock);
 661         return state;
 662 }
 663
 664 void
 665 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 666 {
 667         if (state->state == fmode)
 668                 return;
 669         /* NB! List reordering - see the reclaim code for why.  */
 670         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 671                 if (fmode & FMODE_WRITE)
 672                         list_move(&state->open_states, &state->owner->so_states);
 673                 else
 674                         list_move_tail(&state->open_states, &state->owner->so_states);
 675         }
 676         state->state = fmode;
 677 }
 678
 679 static struct nfs4_state *
 680 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 681 {
 682         struct nfs_inode *nfsi = NFS_I(inode);
 683         struct nfs4_state *state;
 684
 685         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 686                 if (state->owner != owner)
 687                         continue;
 688                 if (!nfs4_valid_open_stateid(state))
 689                         continue;
 690                 if (atomic_inc_not_zero(&state->count))
 691                         return state;
 692         }
 693         return NULL;
 694 }
 695
 696 static void
 697 nfs4_free_open_state(struct nfs4_state *state)
 698 {
 699         kfree(state);
 700 }
 701
 702 struct nfs4_state *
 703 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 704 {
 705         struct nfs4_state *state, *new;
 706         struct nfs_inode *nfsi = NFS_I(inode);
 707
 708         spin_lock(&inode->i_lock);
 709         state = __nfs4_find_state_byowner(inode, owner);
 710         spin_unlock(&inode->i_lock);
 711         if (state)
 712                 goto out;
 713         new = nfs4_alloc_open_state();
 714         spin_lock(&owner->so_lock);
 715         spin_lock(&inode->i_lock);
 716         state = __nfs4_find_state_byowner(inode, owner);
 717         if (state == NULL && new != NULL) {
 718                 state = new;
 719                 state->owner = owner;
 720                 atomic_inc(&owner->so_count);
 721                 list_add(&state->inode_states, &nfsi->open_states);
 722                 ihold(inode);
 723                 state->inode = inode;
 724                 spin_unlock(&inode->i_lock);
 725                 /* Note: The reclaim code dictates that we add stateless
 726                  * and read-only stateids to the end of the list */
 727                 list_add_tail(&state->open_states, &owner->so_states);
 728                 spin_unlock(&owner->so_lock);
 729         } else {
 730                 spin_unlock(&inode->i_lock);
 731                 spin_unlock(&owner->so_lock);
 732                 if (new)
 733                         nfs4_free_open_state(new);
 734         }
 735 out:
 736         return state;
 737 }
 738
 739 void nfs4_put_open_state(struct nfs4_state *state)
 740 {
 741         struct inode *inode = state->inode;
 742         struct nfs4_state_owner *owner = state->owner;
 743
 744         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 745                 return;
 746         spin_lock(&inode->i_lock);
 747         list_del(&state->inode_states);
 748         list_del(&state->open_states);
 749         spin_unlock(&inode->i_lock);
 750         spin_unlock(&owner->so_lock);
 751         iput(inode);
 752         nfs4_free_open_state(state);
 753         nfs4_put_state_owner(owner);
 754 }
 755
 756 /*
 757  * Close the current file.
 758  */
 759 static void __nfs4_close(struct nfs4_state *state,
 760                 fmode_t fmode, gfp_t gfp_mask, int wait)
 761 {
 762         struct nfs4_state_owner *owner = state->owner;
 763         int call_close = 0;
 764         fmode_t newstate;
 765
 766         atomic_inc(&owner->so_count);
 767         /* Protect against nfs4_find_state() */
 768         spin_lock(&owner->so_lock);
 769         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 770                 case FMODE_READ:
 771                         state->n_rdonly--;
 772                         break;
 773                 case FMODE_WRITE:
 774                         state->n_wronly--;
 775                         break;
 776                 case FMODE_READ|FMODE_WRITE:
 777                         state->n_rdwr--;
 778         }
 779         newstate = FMODE_READ|FMODE_WRITE;
 780         if (state->n_rdwr == 0) {
 781                 if (state->n_rdonly == 0) {
 782                         newstate &= ~FMODE_READ;
 783                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 784                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 785                 }
 786                 if (state->n_wronly == 0) {
 787                         newstate &= ~FMODE_WRITE;
 788                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 789                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 790                 }
 791                 if (newstate == 0)
 792                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 793         }
 794         nfs4_state_set_mode_locked(state, newstate);
 795         spin_unlock(&owner->so_lock);
 796
 797         if (!call_close) {
 798                 nfs4_put_open_state(state);
 799                 nfs4_put_state_owner(owner);
 800         } else
 801                 nfs4_do_close(state, gfp_mask, wait);
 802 }
 803
 804 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 805 {
 806         __nfs4_close(state, fmode, GFP_NOFS, 0);
 807 }
 808
 809 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 810 {
 811         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 812 }
 813
 814 /*
 815  * Search the state->lock_states for an existing lock_owner
 816  * that is compatible with current->files
 817  */
 818 static struct nfs4_lock_state *
 819 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 820 {
 821         struct nfs4_lock_state *pos;
 822         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 823                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 824                         continue;
 825                 switch (pos->ls_owner.lo_type) {
 826                 case NFS4_POSIX_LOCK_TYPE:
 827                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 828                                 continue;
 829                         break;
 830                 case NFS4_FLOCK_LOCK_TYPE:
 831                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 832                                 continue;
 833                 }
 834                 atomic_inc(&pos->ls_count);
 835                 return pos;
 836         }
 837         return NULL;
 838 }
 839
 840 /*
 841  * Return a compatible lock_state. If no initialized lock_state structure
 842  * exists, return an uninitialized one.
 843  *
 844  */
 845 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)
 846 {
 847         struct nfs4_lock_state *lsp;
 848         struct nfs_server *server = state->owner->so_server;
 849
 850         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 851         if (lsp == NULL)
 852                 return NULL;
 853         nfs4_init_seqid_counter(&lsp->ls_seqid);
 854         atomic_set(&lsp->ls_count, 1);
 855         lsp->ls_state = state;
 856         lsp->ls_owner.lo_type = type;
 857         switch (lsp->ls_owner.lo_type) {
 858         case NFS4_FLOCK_LOCK_TYPE:
 859                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 860                 break;
 861         case NFS4_POSIX_LOCK_TYPE:
 862                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 863                 break;
 864         default:
 865                 goto out_free;
 866         }
 867         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 868         if (lsp->ls_seqid.owner_id < 0)
 869                 goto out_free;
 870         INIT_LIST_HEAD(&lsp->ls_locks);
 871         return lsp;
 872 out_free:
 873         kfree(lsp);
 874         return NULL;
 875 }
 876
 877 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 878 {
 879         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 880         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 881         kfree(lsp);
 882 }
 883
 884 /*
 885  * Return a compatible lock_state. If no initialized lock_state structure
 886  * exists, return an uninitialized one.
 887  *
 888  */
 889 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 890 {
 891         struct nfs4_lock_state *lsp, *new = NULL;
 892
 893         for(;;) {
 894                 spin_lock(&state->state_lock);
 895                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 896                 if (lsp != NULL)
 897                         break;
 898                 if (new != NULL) {
 899                         list_add(&new->ls_locks, &state->lock_states);
 900                         set_bit(LK_STATE_IN_USE, &state->flags);
 901                         lsp = new;
 902                         new = NULL;
 903                         break;
 904                 }
 905                 spin_unlock(&state->state_lock);
 906                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 907                 if (new == NULL)
 908                         return NULL;
 909         }
 910         spin_unlock(&state->state_lock);
 911         if (new != NULL)
 912                 nfs4_free_lock_state(state->owner->so_server, new);
 913         return lsp;
 914 }
 915
 916 /*
 917  * Release reference to lock_state, and free it if we see that
 918  * it is no longer in use
 919  */
 920 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 921 {
 922         struct nfs_server *server;
 923         struct nfs4_state *state;
 924
 925         if (lsp == NULL)
 926                 return;
 927         state = lsp->ls_state;
 928         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 929                 return;
 930         list_del(&lsp->ls_locks);
 931         if (list_empty(&state->lock_states))
 932                 clear_bit(LK_STATE_IN_USE, &state->flags);
 933         spin_unlock(&state->state_lock);
 934         server = state->owner->so_server;
 935         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 936                 struct nfs_client *clp = server->nfs_client;
 937
 938                 clp->cl_mvops->free_lock_state(server, lsp);
 939         } else
 940                 nfs4_free_lock_state(server, lsp);
 941 }
 942
 943 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 944 {
 945         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 946
 947         dst->fl_u.nfs4_fl.owner = lsp;
 948         atomic_inc(&lsp->ls_count);
 949 }
 950
 951 static void nfs4_fl_release_lock(struct file_lock *fl)
 952 {
 953         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 954 }
 955
 956 static const struct file_lock_operations nfs4_fl_lock_ops = {
 957         .fl_copy_lock = nfs4_fl_copy_lock,
 958         .fl_release_private = nfs4_fl_release_lock,
 959 };
 960
 961 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 962 {
 963         struct nfs4_lock_state *lsp;
 964
 965         if (fl->fl_ops != NULL)
 966                 return 0;
 967         if (fl->fl_flags & FL_POSIX)
 968                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 969         else if (fl->fl_flags & FL_FLOCK)
 970                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 971                                 NFS4_FLOCK_LOCK_TYPE);
 972         else
 973                 return -EINVAL;
 974         if (lsp == NULL)
 975                 return -ENOMEM;
 976         fl->fl_u.nfs4_fl.owner = lsp;
 977         fl->fl_ops = &nfs4_fl_lock_ops;
 978         return 0;
 979 }
 980
 981 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 982                 struct nfs4_state *state,
 983                 const struct nfs_lockowner *lockowner)
 984 {
 985         struct nfs4_lock_state *lsp;
 986         fl_owner_t fl_owner;
 987         pid_t fl_pid;
 988         int ret = -ENOENT;
 989
 990
 991         if (lockowner == NULL)
 992                 goto out;
 993
 994         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 995                 goto out;
 996
 997         fl_owner = lockowner->l_owner;
 998         fl_pid = lockowner->l_pid;
 999         spin_lock(&state->state_lock);
1000         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1001         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1002                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1003                 ret = 0;
1004                 smp_rmb();
1005                 if (!list_empty(&lsp->ls_seqid.list))
1006                         ret = -EWOULDBLOCK;
1007         }
1008         spin_unlock(&state->state_lock);
1009         nfs4_put_lock_state(lsp);
1010 out:
1011         return ret;
1012 }
1013
1014 static int nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1015 {
1016         const nfs4_stateid *src;
1017         int ret;
1018         int seq;
1019
1020         do {
1021                 src = &zero_stateid;
1022                 seq = read_seqbegin(&state->seqlock);
1023                 if (test_bit(NFS_OPEN_STATE, &state->flags))
1024                         src = &state->open_stateid;
1025                 nfs4_stateid_copy(dst, src);
1026                 ret = 0;
1027                 smp_rmb();
1028                 if (!list_empty(&state->owner->so_seqid.list))
1029                         ret = -EWOULDBLOCK;
1030         } while (read_seqretry(&state->seqlock, seq));
1031         return ret;
1032 }
1033
1034 /*
1035  * Byte-range lock aware utility to initialize the stateid of read/write
1036  * requests.
1037  */
1038 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1039                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1040 {
1041         int ret = 0;
1042         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1043                 goto out;
1044         ret = nfs4_copy_lock_stateid(dst, state, lockowner);
1045         if (ret != -ENOENT)
1046                 goto out;
1047         ret = nfs4_copy_open_stateid(dst, state);
1048 out:
1049         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1050                 dst->seqid = 0;
1051         return ret;
1052 }
1053
1054 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1055 {
1056         struct nfs_seqid *new;
1057
1058         new = kmalloc(sizeof(*new), gfp_mask);
1059         if (new != NULL) {
1060                 new->sequence = counter;
1061                 INIT_LIST_HEAD(&new->list);
1062                 new->task = NULL;
1063         }
1064         return new;
1065 }
1066
1067 void nfs_release_seqid(struct nfs_seqid *seqid)
1068 {
1069         struct nfs_seqid_counter *sequence;
1070
1071         if (list_empty(&seqid->list))
1072                 return;
1073         sequence = seqid->sequence;
1074         spin_lock(&sequence->lock);
1075         list_del_init(&seqid->list);
1076         if (!list_empty(&sequence->list)) {
1077                 struct nfs_seqid *next;
1078
1079                 next = list_first_entry(&sequence->list,
1080                                 struct nfs_seqid, list);
1081                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1082         }
1083         spin_unlock(&sequence->lock);
1084 }
1085
1086 void nfs_free_seqid(struct nfs_seqid *seqid)
1087 {
1088         nfs_release_seqid(seqid);
1089         kfree(seqid);
1090 }
1091
1092 /*
1093  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1094  * failed with a seqid incrementing error -
1095  * see comments nfs_fs.h:seqid_mutating_error()
1096  */
1097 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1098 {
1099         switch (status) {
1100                 case 0:
1101                         break;
1102                 case -NFS4ERR_BAD_SEQID:
1103                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1104                                 return;
1105                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1106                                         " sequence-id error on an"
1107                                         " unconfirmed sequence %p!\n",
1108                                         seqid->sequence);
1109                 case -NFS4ERR_STALE_CLIENTID:
1110                 case -NFS4ERR_STALE_STATEID:
1111                 case -NFS4ERR_BAD_STATEID:
1112                 case -NFS4ERR_BADXDR:
1113                 case -NFS4ERR_RESOURCE:
1114                 case -NFS4ERR_NOFILEHANDLE:
1115                         /* Non-seqid mutating errors */
1116                         return;
1117         };
1118         /*
1119          * Note: no locking needed as we are guaranteed to be first
1120          * on the sequence list
1121          */
1122         seqid->sequence->counter++;
1123 }
1124
1125 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1126 {
1127         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1128                                         struct nfs4_state_owner, so_seqid);
1129         struct nfs_server *server = sp->so_server;
1130
1131         if (status == -NFS4ERR_BAD_SEQID)
1132                 nfs4_drop_state_owner(sp);
1133         if (!nfs4_has_session(server->nfs_client))
1134                 nfs_increment_seqid(status, seqid);
1135 }
1136
1137 /*
1138  * Increment the seqid if the LOCK/LOCKU succeeded, or
1139  * failed with a seqid incrementing error -
1140  * see comments nfs_fs.h:seqid_mutating_error()
1141  */
1142 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1143 {
1144         nfs_increment_seqid(status, seqid);
1145 }
1146
1147 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1148 {
1149         struct nfs_seqid_counter *sequence = seqid->sequence;
1150         int status = 0;
1151
1152         spin_lock(&sequence->lock);
1153         seqid->task = task;
1154         if (list_empty(&seqid->list))
1155                 list_add_tail(&seqid->list, &sequence->list);
1156         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1157                 goto unlock;
1158         rpc_sleep_on(&sequence->wait, task, NULL);
1159         status = -EAGAIN;
1160 unlock:
1161         spin_unlock(&sequence->lock);
1162         return status;
1163 }
1164
1165 static int nfs4_run_state_manager(void *);
1166
1167 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1168 {
1169         smp_mb__before_clear_bit();
1170         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1171         smp_mb__after_clear_bit();
1172         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1173         rpc_wake_up(&clp->cl_rpcwaitq);
1174 }
1175
1176 /*
1177  * Schedule the nfs_client asynchronous state management routine
1178  */
1179 void nfs4_schedule_state_manager(struct nfs_client *clp)
1180 {
1181         struct task_struct *task;
1182         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1183
1184         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1185                 return;
1186         __module_get(THIS_MODULE);
1187         atomic_inc(&clp->cl_count);
1188
1189         /* The rcu_read_lock() is not strictly necessary, as the state
1190          * manager is the only thread that ever changes the rpc_xprt
1191          * after it's initialized.  At this point, we're single threaded. */
1192         rcu_read_lock();
1193         snprintf(buf, sizeof(buf), "%s-manager",
1194                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1195         rcu_read_unlock();
1196         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1197         if (IS_ERR(task)) {
1198                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1199                         __func__, PTR_ERR(task));
1200                 nfs4_clear_state_manager_bit(clp);
1201                 nfs_put_client(clp);
1202                 module_put(THIS_MODULE);
1203         }
1204 }
1205
1206 /*
1207  * Schedule a lease recovery attempt
1208  */
1209 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1210 {
1211         if (!clp)
1212                 return;
1213         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1214                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1215         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1216                         clp->cl_hostname);
1217         nfs4_schedule_state_manager(clp);
1218 }
1219 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1220
1221 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1222 {
1223         int res;
1224
1225         might_sleep();
1226
1227         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1228                         nfs_wait_bit_killable, TASK_KILLABLE);
1229         if (res)
1230                 return res;
1231
1232         if (clp->cl_cons_state < 0)
1233                 return clp->cl_cons_state;
1234         return 0;
1235 }
1236
1237 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1238 {
1239         unsigned int loop;
1240         int ret;
1241
1242         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1243                 ret = nfs4_wait_clnt_recover(clp);
1244                 if (ret != 0)
1245                         break;
1246                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1247                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1248                         break;
1249                 nfs4_schedule_state_manager(clp);
1250                 ret = -EIO;
1251         }
1252         return ret;
1253 }
1254
1255 /*
1256  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1257  * @clp: client to process
1258  *
1259  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1260  * resend of the SETCLIENTID and hence re-establish the
1261  * callback channel. Then return all existing delegations.
1262  */
1263 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1264 {
1265         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1266         nfs_expire_all_delegations(clp);
1267         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1268                         clp->cl_hostname);
1269 }
1270
1271 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1272 {
1273         nfs40_handle_cb_pathdown(clp);
1274         nfs4_schedule_state_manager(clp);
1275 }
1276
1277 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1278 {
1279
1280         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1281         /* Don't recover state that expired before the reboot */
1282         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1283                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1284                 return 0;
1285         }
1286         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1287         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1288         return 1;
1289 }
1290
1291 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1292 {
1293         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1294         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1295         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1296         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1297         return 1;
1298 }
1299
1300 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1301 {
1302         struct nfs_client *clp = server->nfs_client;
1303
1304         if (!nfs4_valid_open_stateid(state))
1305                 return -EBADF;
1306         nfs4_state_mark_reclaim_nograce(clp, state);
1307         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1308                         clp->cl_hostname);
1309         nfs4_schedule_state_manager(clp);
1310         return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1313
1314 void nfs_inode_find_state_and_recover(struct inode *inode,
1315                 const nfs4_stateid *stateid)
1316 {
1317         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1318         struct nfs_inode *nfsi = NFS_I(inode);
1319         struct nfs_open_context *ctx;
1320         struct nfs4_state *state;
1321         bool found = false;
1322
1323         spin_lock(&inode->i_lock);
1324         list_for_each_entry(ctx, &nfsi->open_files, list) {
1325                 state = ctx->state;
1326                 if (state == NULL)
1327                         continue;
1328                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1329                         continue;
1330                 if (!nfs4_stateid_match(&state->stateid, stateid))
1331                         continue;
1332                 nfs4_state_mark_reclaim_nograce(clp, state);
1333                 found = true;
1334         }
1335         spin_unlock(&inode->i_lock);
1336         if (found)
1337                 nfs4_schedule_state_manager(clp);
1338 }
1339
1340 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1341 {
1342         struct inode *inode = state->inode;
1343         struct nfs_inode *nfsi = NFS_I(inode);
1344         struct nfs_open_context *ctx;
1345
1346         spin_lock(&inode->i_lock);
1347         list_for_each_entry(ctx, &nfsi->open_files, list) {
1348                 if (ctx->state != state)
1349                         continue;
1350                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1351         }
1352         spin_unlock(&inode->i_lock);
1353 }
1354
1355 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1356 {
1357         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1358         nfs4_state_mark_open_context_bad(state);
1359 }
1360
1361
1362 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1363 {
1364         struct inode *inode = state->inode;
1365         struct nfs_inode *nfsi = NFS_I(inode);
1366         struct file_lock *fl;
1367         int status = 0;
1368
1369         if (inode->i_flock == NULL)
1370                 return 0;
1371
1372         /* Guard against delegation returns and new lock/unlock calls */
1373         down_write(&nfsi->rwsem);
1374         /* Protect inode->i_flock using the BKL */
1375         spin_lock(&inode->i_lock);
1376         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1377                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1378                         continue;
1379                 if (nfs_file_open_context(fl->fl_file)->state != state)
1380                         continue;
1381                 spin_unlock(&inode->i_lock);
1382                 status = ops->recover_lock(state, fl);
1383                 switch (status) {
1384                         case 0:
1385                                 break;
1386                         case -ESTALE:
1387                         case -NFS4ERR_ADMIN_REVOKED:
1388                         case -NFS4ERR_STALE_STATEID:
1389                         case -NFS4ERR_BAD_STATEID:
1390                         case -NFS4ERR_EXPIRED:
1391                         case -NFS4ERR_NO_GRACE:
1392                         case -NFS4ERR_STALE_CLIENTID:
1393                         case -NFS4ERR_BADSESSION:
1394                         case -NFS4ERR_BADSLOT:
1395                         case -NFS4ERR_BAD_HIGH_SLOT:
1396                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1397                                 goto out;
1398                         default:
1399                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1400                                         "Zeroing state\n", __func__, status);
1401                         case -ENOMEM:
1402                         case -NFS4ERR_DENIED:
1403                         case -NFS4ERR_RECLAIM_BAD:
1404                         case -NFS4ERR_RECLAIM_CONFLICT:
1405                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1406                                 status = 0;
1407                 }
1408                 spin_lock(&inode->i_lock);
1409         }
1410         spin_unlock(&inode->i_lock);
1411 out:
1412         up_write(&nfsi->rwsem);
1413         return status;
1414 }
1415
1416 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1417 {
1418         struct nfs4_state *state;
1419         struct nfs4_lock_state *lock;
1420         int status = 0;
1421
1422         /* Note: we rely on the sp->so_states list being ordered
1423          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1424          * states first.
1425          * This is needed to ensure that the server won't give us any
1426          * read delegations that we have to return if, say, we are
1427          * recovering after a network partition or a reboot from a
1428          * server that doesn't support a grace period.
1429          */
1430         spin_lock(&sp->so_lock);
1431         write_seqcount_begin(&sp->so_reclaim_seqcount);
1432 restart:
1433         list_for_each_entry(state, &sp->so_states, open_states) {
1434                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1435                         continue;
1436                 if (!nfs4_valid_open_stateid(state))
1437                         continue;
1438                 if (state->state == 0)
1439                         continue;
1440                 atomic_inc(&state->count);
1441                 spin_unlock(&sp->so_lock);
1442                 status = ops->recover_open(sp, state);
1443                 if (status >= 0) {
1444                         status = nfs4_reclaim_locks(state, ops);
1445                         if (status >= 0) {
1446                                 spin_lock(&state->state_lock);
1447                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1448                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1449                                                 pr_warn_ratelimited("NFS: "
1450                                                         "%s: Lock reclaim "
1451                                                         "failed!\n", __func__);
1452                                 }
1453                                 spin_unlock(&state->state_lock);
1454                                 nfs4_put_open_state(state);
1455                                 spin_lock(&sp->so_lock);
1456                                 goto restart;
1457                         }
1458                 }
1459                 switch (status) {
1460                         default:
1461                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1462                                         "Zeroing state\n", __func__, status);
1463                         case -ENOENT:
1464                         case -ENOMEM:
1465                         case -ESTALE:
1466                                 /*
1467                                  * Open state on this file cannot be recovered
1468                                  * All we can do is revert to using the zero stateid.
1469                                  */
1470                                 nfs4_state_mark_recovery_failed(state, status);
1471                                 break;
1472                         case -EAGAIN:
1473                                 ssleep(1);
1474                         case -NFS4ERR_ADMIN_REVOKED:
1475                         case -NFS4ERR_STALE_STATEID:
1476                         case -NFS4ERR_BAD_STATEID:
1477                         case -NFS4ERR_RECLAIM_BAD:
1478                         case -NFS4ERR_RECLAIM_CONFLICT:
1479                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1480                                 break;
1481                         case -NFS4ERR_EXPIRED:
1482                         case -NFS4ERR_NO_GRACE:
1483                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1484                         case -NFS4ERR_STALE_CLIENTID:
1485                         case -NFS4ERR_BADSESSION:
1486                         case -NFS4ERR_BADSLOT:
1487                         case -NFS4ERR_BAD_HIGH_SLOT:
1488                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1489                                 goto out_err;
1490                 }
1491                 nfs4_put_open_state(state);
1492                 spin_lock(&sp->so_lock);
1493                 goto restart;
1494         }
1495         write_seqcount_end(&sp->so_reclaim_seqcount);
1496         spin_unlock(&sp->so_lock);
1497         return 0;
1498 out_err:
1499         nfs4_put_open_state(state);
1500         spin_lock(&sp->so_lock);
1501         write_seqcount_end(&sp->so_reclaim_seqcount);
1502         spin_unlock(&sp->so_lock);
1503         return status;
1504 }
1505
1506 static void nfs4_clear_open_state(struct nfs4_state *state)
1507 {
1508         struct nfs4_lock_state *lock;
1509
1510         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1511         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1512         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1513         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1514         spin_lock(&state->state_lock);
1515         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1516                 lock->ls_seqid.flags = 0;
1517                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1518         }
1519         spin_unlock(&state->state_lock);
1520 }
1521
1522 static void nfs4_reset_seqids(struct nfs_server *server,
1523         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1524 {
1525         struct nfs_client *clp = server->nfs_client;
1526         struct nfs4_state_owner *sp;
1527         struct rb_node *pos;
1528         struct nfs4_state *state;
1529
1530         spin_lock(&clp->cl_lock);
1531         for (pos = rb_first(&server->state_owners);
1532              pos != NULL;
1533              pos = rb_next(pos)) {
1534                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1535                 sp->so_seqid.flags = 0;
1536                 spin_lock(&sp->so_lock);
1537                 list_for_each_entry(state, &sp->so_states, open_states) {
1538                         if (mark_reclaim(clp, state))
1539                                 nfs4_clear_open_state(state);
1540                 }
1541                 spin_unlock(&sp->so_lock);
1542         }
1543         spin_unlock(&clp->cl_lock);
1544 }
1545
1546 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1547         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1548 {
1549         struct nfs_server *server;
1550
1551         rcu_read_lock();
1552         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1553                 nfs4_reset_seqids(server, mark_reclaim);
1554         rcu_read_unlock();
1555 }
1556
1557 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1558 {
1559         /* Mark all delegations for reclaim */
1560         nfs_delegation_mark_reclaim(clp);
1561         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1562 }
1563
1564 static void nfs4_reclaim_complete(struct nfs_client *clp,
1565                                  const struct nfs4_state_recovery_ops *ops,
1566                                  struct rpc_cred *cred)
1567 {
1568         /* Notify the server we're done reclaiming our state */
1569         if (ops->reclaim_complete)
1570                 (void)ops->reclaim_complete(clp, cred);
1571 }
1572
1573 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1574 {
1575         struct nfs_client *clp = server->nfs_client;
1576         struct nfs4_state_owner *sp;
1577         struct rb_node *pos;
1578         struct nfs4_state *state;
1579
1580         spin_lock(&clp->cl_lock);
1581         for (pos = rb_first(&server->state_owners);
1582              pos != NULL;
1583              pos = rb_next(pos)) {
1584                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1585                 spin_lock(&sp->so_lock);
1586                 list_for_each_entry(state, &sp->so_states, open_states) {
1587                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1588                                                 &state->flags))
1589                                 continue;
1590                         nfs4_state_mark_reclaim_nograce(clp, state);
1591                 }
1592                 spin_unlock(&sp->so_lock);
1593         }
1594         spin_unlock(&clp->cl_lock);
1595 }
1596
1597 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1598 {
1599         struct nfs_server *server;
1600
1601         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1602                 return 0;
1603
1604         rcu_read_lock();
1605         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1606                 nfs4_clear_reclaim_server(server);
1607         rcu_read_unlock();
1608
1609         nfs_delegation_reap_unclaimed(clp);
1610         return 1;
1611 }
1612
1613 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1614 {
1615         const struct nfs4_state_recovery_ops *ops;
1616         struct rpc_cred *cred;
1617
1618         if (!nfs4_state_clear_reclaim_reboot(clp))
1619                 return;
1620         ops = clp->cl_mvops->reboot_recovery_ops;
1621         cred = ops->get_clid_cred(clp);
1622         nfs4_reclaim_complete(clp, ops, cred);
1623         put_rpccred(cred);
1624 }
1625
1626 static void nfs_delegation_clear_all(struct nfs_client *clp)
1627 {
1628         nfs_delegation_mark_reclaim(clp);
1629         nfs_delegation_reap_unclaimed(clp);
1630 }
1631
1632 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1633 {
1634         nfs_delegation_clear_all(clp);
1635         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1636 }
1637
1638 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1639 {
1640         switch (error) {
1641                 case 0:
1642                         break;
1643                 case -NFS4ERR_CB_PATH_DOWN:
1644                         nfs40_handle_cb_pathdown(clp);
1645                         break;
1646                 case -NFS4ERR_NO_GRACE:
1647                         nfs4_state_end_reclaim_reboot(clp);
1648                         break;
1649                 case -NFS4ERR_STALE_CLIENTID:
1650                 case -NFS4ERR_LEASE_MOVED:
1651                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1652                         nfs4_state_clear_reclaim_reboot(clp);
1653                         nfs4_state_start_reclaim_reboot(clp);
1654                         break;
1655                 case -NFS4ERR_EXPIRED:
1656                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1657                         nfs4_state_start_reclaim_nograce(clp);
1658                         break;
1659                 case -NFS4ERR_BADSESSION:
1660                 case -NFS4ERR_BADSLOT:
1661                 case -NFS4ERR_BAD_HIGH_SLOT:
1662                 case -NFS4ERR_DEADSESSION:
1663                 case -NFS4ERR_SEQ_FALSE_RETRY:
1664                 case -NFS4ERR_SEQ_MISORDERED:
1665                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1666                         /* Zero session reset errors */
1667                         break;
1668                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1669                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1670                         break;
1671                 default:
1672                         dprintk("%s: failed to handle error %d for server %s\n",
1673                                         __func__, error, clp->cl_hostname);
1674                         return error;
1675         }
1676         dprintk("%s: handled error %d for server %s\n", __func__, error,
1677                         clp->cl_hostname);
1678         return 0;
1679 }
1680
1681 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1682 {
1683         struct nfs4_state_owner *sp;
1684         struct nfs_server *server;
1685         struct rb_node *pos;
1686         int status = 0;
1687
1688 restart:
1689         rcu_read_lock();
1690         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1691                 nfs4_purge_state_owners(server);
1692                 spin_lock(&clp->cl_lock);
1693                 for (pos = rb_first(&server->state_owners);
1694                      pos != NULL;
1695                      pos = rb_next(pos)) {
1696                         sp = rb_entry(pos,
1697                                 struct nfs4_state_owner, so_server_node);
1698                         if (!test_and_clear_bit(ops->owner_flag_bit,
1699                                                         &sp->so_flags))
1700                                 continue;
1701                         atomic_inc(&sp->so_count);
1702                         spin_unlock(&clp->cl_lock);
1703                         rcu_read_unlock();
1704
1705                         status = nfs4_reclaim_open_state(sp, ops);
1706                         if (status < 0) {
1707                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1708                                 nfs4_put_state_owner(sp);
1709                                 return nfs4_recovery_handle_error(clp, status);
1710                         }
1711
1712                         nfs4_put_state_owner(sp);
1713                         goto restart;
1714                 }
1715                 spin_unlock(&clp->cl_lock);
1716         }
1717         rcu_read_unlock();
1718         return status;
1719 }
1720
1721 static int nfs4_check_lease(struct nfs_client *clp)
1722 {
1723         struct rpc_cred *cred;
1724         const struct nfs4_state_maintenance_ops *ops =
1725                 clp->cl_mvops->state_renewal_ops;
1726         int status;
1727
1728         /* Is the client already known to have an expired lease? */
1729         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1730                 return 0;
1731         spin_lock(&clp->cl_lock);
1732         cred = ops->get_state_renewal_cred_locked(clp);
1733         spin_unlock(&clp->cl_lock);
1734         if (cred == NULL) {
1735                 cred = nfs4_get_setclientid_cred(clp);
1736                 status = -ENOKEY;
1737                 if (cred == NULL)
1738                         goto out;
1739         }
1740         status = ops->renew_lease(clp, cred);
1741         put_rpccred(cred);
1742         if (status == -ETIMEDOUT) {
1743                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1744                 return 0;
1745         }
1746 out:
1747         return nfs4_recovery_handle_error(clp, status);
1748 }
1749
1750 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1751  * and for recoverable errors on EXCHANGE_ID for v4.1
1752  */
1753 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1754 {
1755         switch (status) {
1756         case -NFS4ERR_SEQ_MISORDERED:
1757                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1758                         return -ESERVERFAULT;
1759                 /* Lease confirmation error: retry after purging the lease */
1760                 ssleep(1);
1761                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1762                 break;
1763         case -NFS4ERR_STALE_CLIENTID:
1764                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1765                 nfs4_state_clear_reclaim_reboot(clp);
1766                 nfs4_state_start_reclaim_reboot(clp);
1767                 break;
1768         case -NFS4ERR_CLID_INUSE:
1769                 pr_err("NFS: Server %s reports our clientid is in use\n",
1770                         clp->cl_hostname);
1771                 nfs_mark_client_ready(clp, -EPERM);
1772                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1773                 return -EPERM;
1774         case -EACCES:
1775         case -NFS4ERR_DELAY:
1776         case -ETIMEDOUT:
1777         case -EAGAIN:
1778                 ssleep(1);
1779                 break;
1780
1781         case -NFS4ERR_MINOR_VERS_MISMATCH:
1782                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1783                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1784                 dprintk("%s: exit with error %d for server %s\n",
1785                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1786                 return -EPROTONOSUPPORT;
1787         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1788                                  * in nfs4_exchange_id */
1789         default:
1790                 dprintk("%s: exit with error %d for server %s\n", __func__,
1791                                 status, clp->cl_hostname);
1792                 return status;
1793         }
1794         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1795         dprintk("%s: handled error %d for server %s\n", __func__, status,
1796                         clp->cl_hostname);
1797         return 0;
1798 }
1799
1800 static int nfs4_establish_lease(struct nfs_client *clp)
1801 {
1802         struct rpc_cred *cred;
1803         const struct nfs4_state_recovery_ops *ops =
1804                 clp->cl_mvops->reboot_recovery_ops;
1805         int status;
1806
1807         cred = ops->get_clid_cred(clp);
1808         if (cred == NULL)
1809                 return -ENOENT;
1810         status = ops->establish_clid(clp, cred);
1811         put_rpccred(cred);
1812         if (status != 0)
1813                 return status;
1814         pnfs_destroy_all_layouts(clp);
1815         return 0;
1816 }
1817
1818 /*
1819  * Returns zero or a negative errno.  NFS4ERR values are converted
1820  * to local errno values.
1821  */
1822 static int nfs4_reclaim_lease(struct nfs_client *clp)
1823 {
1824         int status;
1825
1826         status = nfs4_establish_lease(clp);
1827         if (status < 0)
1828                 return nfs4_handle_reclaim_lease_error(clp, status);
1829         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1830                 nfs4_state_start_reclaim_nograce(clp);
1831         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1832                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1833         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1834         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1835         return 0;
1836 }
1837
1838 static int nfs4_purge_lease(struct nfs_client *clp)
1839 {
1840         int status;
1841
1842         status = nfs4_establish_lease(clp);
1843         if (status < 0)
1844                 return nfs4_handle_reclaim_lease_error(clp, status);
1845         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1846         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1847         nfs4_state_start_reclaim_nograce(clp);
1848         return 0;
1849 }
1850
1851 /**
1852  * nfs4_discover_server_trunking - Detect server IP address trunking
1853  *
1854  * @clp: nfs_client under test
1855  * @result: OUT: found nfs_client, or clp
1856  *
1857  * Returns zero or a negative errno.  If zero is returned,
1858  * an nfs_client pointer is planted in "result".
1859  *
1860  * Note: since we are invoked in process context, and
1861  * not from inside the state manager, we cannot use
1862  * nfs4_handle_reclaim_lease_error().
1863  */
1864 int nfs4_discover_server_trunking(struct nfs_client *clp,
1865                                   struct nfs_client **result)
1866 {
1867         const struct nfs4_state_recovery_ops *ops =
1868                                 clp->cl_mvops->reboot_recovery_ops;
1869         struct rpc_clnt *clnt;
1870         struct rpc_cred *cred;
1871         int i, status;
1872
1873         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1874
1875         clnt = clp->cl_rpcclient;
1876         i = 0;
1877
1878         mutex_lock(&nfs_clid_init_mutex);
1879 again:
1880         status  = -ENOENT;
1881         cred = ops->get_clid_cred(clp);
1882         if (cred == NULL)
1883                 goto out_unlock;
1884
1885         status = ops->detect_trunking(clp, result, cred);
1886         put_rpccred(cred);
1887         switch (status) {
1888         case 0:
1889                 break;
1890         case -NFS4ERR_DELAY:
1891         case -ETIMEDOUT:
1892         case -EAGAIN:
1893                 ssleep(1);
1894         case -NFS4ERR_STALE_CLIENTID:
1895                 dprintk("NFS: %s after status %d, retrying\n",
1896                         __func__, status);
1897                 goto again;
1898         case -EACCES:
1899                 if (i++)
1900                         break;
1901         case -NFS4ERR_CLID_INUSE:
1902         case -NFS4ERR_WRONGSEC:
1903                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
1904                 if (IS_ERR(clnt)) {
1905                         status = PTR_ERR(clnt);
1906                         break;
1907                 }
1908                 /* Note: this is safe because we haven't yet marked the
1909                  * client as ready, so we are the only user of
1910                  * clp->cl_rpcclient
1911                  */
1912                 clnt = xchg(&clp->cl_rpcclient, clnt);
1913                 rpc_shutdown_client(clnt);
1914                 clnt = clp->cl_rpcclient;
1915                 goto again;
1916
1917         case -NFS4ERR_MINOR_VERS_MISMATCH:
1918                 status = -EPROTONOSUPPORT;
1919                 break;
1920
1921         case -EKEYEXPIRED:
1922         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1923                                  * in nfs4_exchange_id */
1924                 status = -EKEYEXPIRED;
1925                 break;
1926         default:
1927                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
1928                                 __func__, status);
1929                 status = -EIO;
1930         }
1931
1932 out_unlock:
1933         mutex_unlock(&nfs_clid_init_mutex);
1934         dprintk("NFS: %s: status = %d\n", __func__, status);
1935         return status;
1936 }
1937
1938 #ifdef CONFIG_NFS_V4_1
1939 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1940 {
1941         struct nfs_client *clp = session->clp;
1942
1943         switch (err) {
1944         default:
1945                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1946                 break;
1947         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1948                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1949         }
1950         nfs4_schedule_lease_recovery(clp);
1951 }
1952 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1953
1954 static void nfs41_ping_server(struct nfs_client *clp)
1955 {
1956         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1957         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1958         nfs4_schedule_state_manager(clp);
1959 }
1960
1961 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1962 {
1963         nfs41_ping_server(clp);
1964 }
1965
1966 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1967 {
1968         nfs41_ping_server(clp);
1969 }
1970
1971 static void nfs4_reset_all_state(struct nfs_client *clp)
1972 {
1973         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1974                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1975                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1976                 nfs4_state_start_reclaim_nograce(clp);
1977                 dprintk("%s: scheduling reset of all state for server %s!\n",
1978                                 __func__, clp->cl_hostname);
1979                 nfs4_schedule_state_manager(clp);
1980         }
1981 }
1982
1983 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1984 {
1985         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1986                 nfs4_state_start_reclaim_reboot(clp);
1987                 dprintk("%s: server %s rebooted!\n", __func__,
1988                                 clp->cl_hostname);
1989                 nfs4_schedule_state_manager(clp);
1990         }
1991 }
1992
1993 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1994 {
1995         nfs4_reset_all_state(clp);
1996         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1997 }
1998
1999 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2000 {
2001         /* This will need to handle layouts too */
2002         nfs_expire_all_delegations(clp);
2003         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2004                         clp->cl_hostname);
2005 }
2006
2007 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2008 {
2009         nfs_expire_all_delegations(clp);
2010         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
2011                 nfs4_schedule_state_manager(clp);
2012         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2013                         clp->cl_hostname);
2014 }
2015
2016 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2017 {
2018         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2019                 &clp->cl_state) == 0)
2020                 nfs4_schedule_state_manager(clp);
2021 }
2022
2023 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
2024 {
2025         if (!flags)
2026                 return;
2027
2028         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2029                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2030
2031         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2032                 nfs41_handle_server_reboot(clp);
2033         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2034                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2035                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2036                             SEQ4_STATUS_LEASE_MOVED))
2037                 nfs41_handle_state_revoked(clp);
2038         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2039                 nfs41_handle_recallable_state_revoked(clp);
2040         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2041                 nfs41_handle_backchannel_fault(clp);
2042         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2043                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2044                 nfs41_handle_cb_path_down(clp);
2045 }
2046
2047 static int nfs4_reset_session(struct nfs_client *clp)
2048 {
2049         struct rpc_cred *cred;
2050         int status;
2051
2052         if (!nfs4_has_session(clp))
2053                 return 0;
2054         nfs4_begin_drain_session(clp);
2055         cred = nfs4_get_exchange_id_cred(clp);
2056         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2057         switch (status) {
2058         case 0:
2059         case -NFS4ERR_BADSESSION:
2060         case -NFS4ERR_DEADSESSION:
2061                 break;
2062         case -NFS4ERR_BACK_CHAN_BUSY:
2063         case -NFS4ERR_DELAY:
2064                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2065                 status = 0;
2066                 ssleep(1);
2067                 goto out;
2068         default:
2069                 status = nfs4_recovery_handle_error(clp, status);
2070                 goto out;
2071         }
2072
2073         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2074         status = nfs4_proc_create_session(clp, cred);
2075         if (status) {
2076                 dprintk("%s: session reset failed with status %d for server %s!\n",
2077                         __func__, status, clp->cl_hostname);
2078                 status = nfs4_handle_reclaim_lease_error(clp, status);
2079                 goto out;
2080         }
2081         nfs41_finish_session_reset(clp);
2082         dprintk("%s: session reset was successful for server %s!\n",
2083                         __func__, clp->cl_hostname);
2084 out:
2085         if (cred)
2086                 put_rpccred(cred);
2087         return status;
2088 }
2089
2090 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2091 {
2092         struct rpc_cred *cred;
2093         int ret;
2094
2095         if (!nfs4_has_session(clp))
2096                 return 0;
2097         nfs4_begin_drain_session(clp);
2098         cred = nfs4_get_exchange_id_cred(clp);
2099         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2100         if (cred)
2101                 put_rpccred(cred);
2102         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2103         switch (ret) {
2104         case 0:
2105                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2106                         __func__, clp->cl_hostname);
2107                 break;
2108         case -NFS4ERR_DELAY:
2109                 ssleep(1);
2110                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2111                 break;
2112         default:
2113                 return nfs4_recovery_handle_error(clp, ret);
2114         }
2115         return 0;
2116 }
2117 #else /* CONFIG_NFS_V4_1 */
2118 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2119 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2120
2121 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2122 {
2123         return 0;
2124 }
2125 #endif /* CONFIG_NFS_V4_1 */
2126
2127 static void nfs4_state_manager(struct nfs_client *clp)
2128 {
2129         int status = 0;
2130         const char *section = "", *section_sep = "";
2131
2132         /* Ensure exclusive access to NFSv4 state */
2133         do {
2134                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2135                         section = "purge state";
2136                         status = nfs4_purge_lease(clp);
2137                         if (status < 0)
2138                                 goto out_error;
2139                         continue;
2140                 }
2141
2142                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2143                         section = "lease expired";
2144                         /* We're going to have to re-establish a clientid */
2145                         status = nfs4_reclaim_lease(clp);
2146                         if (status < 0)
2147                                 goto out_error;
2148                         continue;
2149                 }
2150
2151                 /* Initialize or reset the session */
2152                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2153                         section = "reset session";
2154                         status = nfs4_reset_session(clp);
2155                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2156                                 continue;
2157                         if (status < 0)
2158                                 goto out_error;
2159                 }
2160
2161                 /* Send BIND_CONN_TO_SESSION */
2162                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2163                                 &clp->cl_state)) {
2164                         section = "bind conn to session";
2165                         status = nfs4_bind_conn_to_session(clp);
2166                         if (status < 0)
2167                                 goto out_error;
2168                         continue;
2169                 }
2170
2171                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2172                         section = "check lease";
2173                         status = nfs4_check_lease(clp);
2174                         if (status < 0)
2175                                 goto out_error;
2176                         continue;
2177                 }
2178
2179                 /* First recover reboot state... */
2180                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2181                         section = "reclaim reboot";
2182                         status = nfs4_do_reclaim(clp,
2183                                 clp->cl_mvops->reboot_recovery_ops);
2184                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2185                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2186                                 continue;
2187                         nfs4_state_end_reclaim_reboot(clp);
2188                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2189                                 continue;
2190                         if (status < 0)
2191                                 goto out_error;
2192                 }
2193
2194                 /* Now recover expired state... */
2195                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2196                         section = "reclaim nograce";
2197                         status = nfs4_do_reclaim(clp,
2198                                 clp->cl_mvops->nograce_recovery_ops);
2199                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2200                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2201                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2202                                 continue;
2203                         if (status < 0)
2204                                 goto out_error;
2205                 }
2206
2207                 nfs4_end_drain_session(clp);
2208                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2209                         nfs_client_return_marked_delegations(clp);
2210                         continue;
2211                 }
2212
2213                 nfs4_clear_state_manager_bit(clp);
2214                 /* Did we race with an attempt to give us more work? */
2215                 if (clp->cl_state == 0)
2216                         break;
2217                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2218                         break;
2219         } while (atomic_read(&clp->cl_count) > 1);
2220         return;
2221 out_error:
2222         if (strlen(section))
2223                 section_sep = ": ";
2224         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2225                         " with error %d\n", section_sep, section,
2226                         clp->cl_hostname, -status);
2227         ssleep(1);
2228         nfs4_end_drain_session(clp);
2229         nfs4_clear_state_manager_bit(clp);
2230 }
2231
2232 static int nfs4_run_state_manager(void *ptr)
2233 {
2234         struct nfs_client *clp = ptr;
2235
2236         allow_signal(SIGKILL);
2237         nfs4_state_manager(clp);
2238         nfs_put_client(clp);
2239         module_put_and_exit(0);
2240         return 0;
2241 }
2242
2243 /*
2244  * Local variables:
2245  *  c-basic-offset: 8
2246  * End:
2247  */