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         mutex_init(&sp->so_delegreturn_mutex);
 523         return sp;
 524 }
 525
 526 static void
 527 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 528 {
 529         struct rb_node *rb_node = &sp->so_server_node;
 530
 531         if (!RB_EMPTY_NODE(rb_node)) {
 532                 struct nfs_server *server = sp->so_server;
 533                 struct nfs_client *clp = server->nfs_client;
 534
 535                 spin_lock(&clp->cl_lock);
 536                 if (!RB_EMPTY_NODE(rb_node)) {
 537                         rb_erase(rb_node, &server->state_owners);
 538                         RB_CLEAR_NODE(rb_node);
 539                 }
 540                 spin_unlock(&clp->cl_lock);
 541         }
 542 }
 543
 544 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 545 {
 546         nfs4_destroy_seqid_counter(&sp->so_seqid);
 547         put_rpccred(sp->so_cred);
 548         kfree(sp);
 549 }
 550
 551 static void nfs4_gc_state_owners(struct nfs_server *server)
 552 {
 553         struct nfs_client *clp = server->nfs_client;
 554         struct nfs4_state_owner *sp, *tmp;
 555         unsigned long time_min, time_max;
 556         LIST_HEAD(doomed);
 557
 558         spin_lock(&clp->cl_lock);
 559         time_max = jiffies;
 560         time_min = (long)time_max - (long)clp->cl_lease_time;
 561         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 562                 /* NB: LRU is sorted so that oldest is at the head */
 563                 if (time_in_range(sp->so_expires, time_min, time_max))
 564                         break;
 565                 list_move(&sp->so_lru, &doomed);
 566                 nfs4_remove_state_owner_locked(sp);
 567         }
 568         spin_unlock(&clp->cl_lock);
 569
 570         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 571                 list_del(&sp->so_lru);
 572                 nfs4_free_state_owner(sp);
 573         }
 574 }
 575
 576 /**
 577  * nfs4_get_state_owner - Look up a state owner given a credential
 578  * @server: nfs_server to search
 579  * @cred: RPC credential to match
 580  *
 581  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 582  */
 583 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 584                                               struct rpc_cred *cred,
 585                                               gfp_t gfp_flags)
 586 {
 587         struct nfs_client *clp = server->nfs_client;
 588         struct nfs4_state_owner *sp, *new;
 589
 590         spin_lock(&clp->cl_lock);
 591         sp = nfs4_find_state_owner_locked(server, cred);
 592         spin_unlock(&clp->cl_lock);
 593         if (sp != NULL)
 594                 goto out;
 595         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 596         if (new == NULL)
 597                 goto out;
 598         do {
 599                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 600                         break;
 601                 spin_lock(&clp->cl_lock);
 602                 sp = nfs4_insert_state_owner_locked(new);
 603                 spin_unlock(&clp->cl_lock);
 604         } while (sp == ERR_PTR(-EAGAIN));
 605         if (sp != new)
 606                 nfs4_free_state_owner(new);
 607 out:
 608         nfs4_gc_state_owners(server);
 609         return sp;
 610 }
 611
 612 /**
 613  * nfs4_put_state_owner - Release a nfs4_state_owner
 614  * @sp: state owner data to release
 615  *
 616  * Note that we keep released state owners on an LRU
 617  * list.
 618  * This caches valid state owners so that they can be
 619  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 620  * It also pins the uniquifier of dropped state owners for
 621  * a while, to ensure that those state owner names are
 622  * never reused.
 623  */
 624 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 625 {
 626         struct nfs_server *server = sp->so_server;
 627         struct nfs_client *clp = server->nfs_client;
 628
 629         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 630                 return;
 631
 632         sp->so_expires = jiffies;
 633         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 634         spin_unlock(&clp->cl_lock);
 635 }
 636
 637 /**
 638  * nfs4_purge_state_owners - Release all cached state owners
 639  * @server: nfs_server with cached state owners to release
 640  *
 641  * Called at umount time.  Remaining state owners will be on
 642  * the LRU with ref count of zero.
 643  */
 644 void nfs4_purge_state_owners(struct nfs_server *server)
 645 {
 646         struct nfs_client *clp = server->nfs_client;
 647         struct nfs4_state_owner *sp, *tmp;
 648         LIST_HEAD(doomed);
 649
 650         spin_lock(&clp->cl_lock);
 651         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 652                 list_move(&sp->so_lru, &doomed);
 653                 nfs4_remove_state_owner_locked(sp);
 654         }
 655         spin_unlock(&clp->cl_lock);
 656
 657         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 658                 list_del(&sp->so_lru);
 659                 nfs4_free_state_owner(sp);
 660         }
 661 }
 662
 663 static struct nfs4_state *
 664 nfs4_alloc_open_state(void)
 665 {
 666         struct nfs4_state *state;
 667
 668         state = kzalloc(sizeof(*state), GFP_NOFS);
 669         if (!state)
 670                 return NULL;
 671         atomic_set(&state->count, 1);
 672         INIT_LIST_HEAD(&state->lock_states);
 673         spin_lock_init(&state->state_lock);
 674         seqlock_init(&state->seqlock);
 675         return state;
 676 }
 677
 678 void
 679 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 680 {
 681         if (state->state == fmode)
 682                 return;
 683         /* NB! List reordering - see the reclaim code for why.  */
 684         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 685                 if (fmode & FMODE_WRITE)
 686                         list_move(&state->open_states, &state->owner->so_states);
 687                 else
 688                         list_move_tail(&state->open_states, &state->owner->so_states);
 689         }
 690         state->state = fmode;
 691 }
 692
 693 static struct nfs4_state *
 694 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 695 {
 696         struct nfs_inode *nfsi = NFS_I(inode);
 697         struct nfs4_state *state;
 698
 699         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 700                 if (state->owner != owner)
 701                         continue;
 702                 if (atomic_inc_not_zero(&state->count))
 703                         return state;
 704         }
 705         return NULL;
 706 }
 707
 708 static void
 709 nfs4_free_open_state(struct nfs4_state *state)
 710 {
 711         kfree(state);
 712 }
 713
 714 struct nfs4_state *
 715 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 716 {
 717         struct nfs4_state *state, *new;
 718         struct nfs_inode *nfsi = NFS_I(inode);
 719
 720         spin_lock(&inode->i_lock);
 721         state = __nfs4_find_state_byowner(inode, owner);
 722         spin_unlock(&inode->i_lock);
 723         if (state)
 724                 goto out;
 725         new = nfs4_alloc_open_state();
 726         spin_lock(&owner->so_lock);
 727         spin_lock(&inode->i_lock);
 728         state = __nfs4_find_state_byowner(inode, owner);
 729         if (state == NULL && new != NULL) {
 730                 state = new;
 731                 state->owner = owner;
 732                 atomic_inc(&owner->so_count);
 733                 list_add(&state->inode_states, &nfsi->open_states);
 734                 ihold(inode);
 735                 state->inode = inode;
 736                 spin_unlock(&inode->i_lock);
 737                 /* Note: The reclaim code dictates that we add stateless
 738                  * and read-only stateids to the end of the list */
 739                 list_add_tail(&state->open_states, &owner->so_states);
 740                 spin_unlock(&owner->so_lock);
 741         } else {
 742                 spin_unlock(&inode->i_lock);
 743                 spin_unlock(&owner->so_lock);
 744                 if (new)
 745                         nfs4_free_open_state(new);
 746         }
 747 out:
 748         return state;
 749 }
 750
 751 void nfs4_put_open_state(struct nfs4_state *state)
 752 {
 753         struct inode *inode = state->inode;
 754         struct nfs4_state_owner *owner = state->owner;
 755
 756         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 757                 return;
 758         spin_lock(&inode->i_lock);
 759         list_del(&state->inode_states);
 760         list_del(&state->open_states);
 761         spin_unlock(&inode->i_lock);
 762         spin_unlock(&owner->so_lock);
 763         iput(inode);
 764         nfs4_free_open_state(state);
 765         nfs4_put_state_owner(owner);
 766 }
 767
 768 /*
 769  * Close the current file.
 770  */
 771 static void __nfs4_close(struct nfs4_state *state,
 772                 fmode_t fmode, gfp_t gfp_mask, int wait)
 773 {
 774         struct nfs4_state_owner *owner = state->owner;
 775         int call_close = 0;
 776         fmode_t newstate;
 777
 778         atomic_inc(&owner->so_count);
 779         /* Protect against nfs4_find_state() */
 780         spin_lock(&owner->so_lock);
 781         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 782                 case FMODE_READ:
 783                         state->n_rdonly--;
 784                         break;
 785                 case FMODE_WRITE:
 786                         state->n_wronly--;
 787                         break;
 788                 case FMODE_READ|FMODE_WRITE:
 789                         state->n_rdwr--;
 790         }
 791         newstate = FMODE_READ|FMODE_WRITE;
 792         if (state->n_rdwr == 0) {
 793                 if (state->n_rdonly == 0) {
 794                         newstate &= ~FMODE_READ;
 795                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 796                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 797                 }
 798                 if (state->n_wronly == 0) {
 799                         newstate &= ~FMODE_WRITE;
 800                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 801                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 802                 }
 803                 if (newstate == 0)
 804                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 805         }
 806         nfs4_state_set_mode_locked(state, newstate);
 807         spin_unlock(&owner->so_lock);
 808
 809         if (!call_close) {
 810                 nfs4_put_open_state(state);
 811                 nfs4_put_state_owner(owner);
 812         } else
 813                 nfs4_do_close(state, gfp_mask, wait);
 814 }
 815
 816 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 817 {
 818         __nfs4_close(state, fmode, GFP_NOFS, 0);
 819 }
 820
 821 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 822 {
 823         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 824 }
 825
 826 /*
 827  * Search the state->lock_states for an existing lock_owner
 828  * that is compatible with current->files
 829  */
 830 static struct nfs4_lock_state *
 831 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 832 {
 833         struct nfs4_lock_state *pos;
 834         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 835                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 836                         continue;
 837                 switch (pos->ls_owner.lo_type) {
 838                 case NFS4_POSIX_LOCK_TYPE:
 839                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 840                                 continue;
 841                         break;
 842                 case NFS4_FLOCK_LOCK_TYPE:
 843                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 844                                 continue;
 845                 }
 846                 atomic_inc(&pos->ls_count);
 847                 return pos;
 848         }
 849         return NULL;
 850 }
 851
 852 /*
 853  * Return a compatible lock_state. If no initialized lock_state structure
 854  * exists, return an uninitialized one.
 855  *
 856  */
 857 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)
 858 {
 859         struct nfs4_lock_state *lsp;
 860         struct nfs_server *server = state->owner->so_server;
 861
 862         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 863         if (lsp == NULL)
 864                 return NULL;
 865         nfs4_init_seqid_counter(&lsp->ls_seqid);
 866         atomic_set(&lsp->ls_count, 1);
 867         lsp->ls_state = state;
 868         lsp->ls_owner.lo_type = type;
 869         switch (lsp->ls_owner.lo_type) {
 870         case NFS4_FLOCK_LOCK_TYPE:
 871                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 872                 break;
 873         case NFS4_POSIX_LOCK_TYPE:
 874                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 875                 break;
 876         default:
 877                 goto out_free;
 878         }
 879         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 880         if (lsp->ls_seqid.owner_id < 0)
 881                 goto out_free;
 882         INIT_LIST_HEAD(&lsp->ls_locks);
 883         return lsp;
 884 out_free:
 885         kfree(lsp);
 886         return NULL;
 887 }
 888
 889 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 890 {
 891         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 892         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 893         kfree(lsp);
 894 }
 895
 896 /*
 897  * Return a compatible lock_state. If no initialized lock_state structure
 898  * exists, return an uninitialized one.
 899  *
 900  */
 901 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 902 {
 903         struct nfs4_lock_state *lsp, *new = NULL;
 904
 905         for(;;) {
 906                 spin_lock(&state->state_lock);
 907                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 908                 if (lsp != NULL)
 909                         break;
 910                 if (new != NULL) {
 911                         list_add(&new->ls_locks, &state->lock_states);
 912                         set_bit(LK_STATE_IN_USE, &state->flags);
 913                         lsp = new;
 914                         new = NULL;
 915                         break;
 916                 }
 917                 spin_unlock(&state->state_lock);
 918                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 919                 if (new == NULL)
 920                         return NULL;
 921         }
 922         spin_unlock(&state->state_lock);
 923         if (new != NULL)
 924                 nfs4_free_lock_state(state->owner->so_server, new);
 925         return lsp;
 926 }
 927
 928 /*
 929  * Release reference to lock_state, and free it if we see that
 930  * it is no longer in use
 931  */
 932 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 933 {
 934         struct nfs4_state *state;
 935
 936         if (lsp == NULL)
 937                 return;
 938         state = lsp->ls_state;
 939         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 940                 return;
 941         list_del(&lsp->ls_locks);
 942         if (list_empty(&state->lock_states))
 943                 clear_bit(LK_STATE_IN_USE, &state->flags);
 944         spin_unlock(&state->state_lock);
 945         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 946                 if (nfs4_release_lockowner(lsp) == 0)
 947                         return;
 948         }
 949         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 950 }
 951
 952 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 953 {
 954         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 955
 956         dst->fl_u.nfs4_fl.owner = lsp;
 957         atomic_inc(&lsp->ls_count);
 958 }
 959
 960 static void nfs4_fl_release_lock(struct file_lock *fl)
 961 {
 962         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 963 }
 964
 965 static const struct file_lock_operations nfs4_fl_lock_ops = {
 966         .fl_copy_lock = nfs4_fl_copy_lock,
 967         .fl_release_private = nfs4_fl_release_lock,
 968 };
 969
 970 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 971 {
 972         struct nfs4_lock_state *lsp;
 973
 974         if (fl->fl_ops != NULL)
 975                 return 0;
 976         if (fl->fl_flags & FL_POSIX)
 977                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 978         else if (fl->fl_flags & FL_FLOCK)
 979                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 980                                 NFS4_FLOCK_LOCK_TYPE);
 981         else
 982                 return -EINVAL;
 983         if (lsp == NULL)
 984                 return -ENOMEM;
 985         fl->fl_u.nfs4_fl.owner = lsp;
 986         fl->fl_ops = &nfs4_fl_lock_ops;
 987         return 0;
 988 }
 989
 990 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 991                 const struct nfs_lockowner *lockowner)
 992 {
 993         struct nfs4_lock_state *lsp;
 994         fl_owner_t fl_owner;
 995         pid_t fl_pid;
 996         bool ret = false;
 997
 998
 999         if (lockowner == NULL)
1000                 goto out;
1001
1002         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1003                 goto out;
1004
1005         fl_owner = lockowner->l_owner;
1006         fl_pid = lockowner->l_pid;
1007         spin_lock(&state->state_lock);
1008         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1009         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1010                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1011                 ret = true;
1012         }
1013         spin_unlock(&state->state_lock);
1014         nfs4_put_lock_state(lsp);
1015 out:
1016         return ret;
1017 }
1018
1019 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1020 {
1021         int seq;
1022
1023         do {
1024                 seq = read_seqbegin(&state->seqlock);
1025                 nfs4_stateid_copy(dst, &state->stateid);
1026         } while (read_seqretry(&state->seqlock, seq));
1027 }
1028
1029 /*
1030  * Byte-range lock aware utility to initialize the stateid of read/write
1031  * requests.
1032  */
1033 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1034                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1035 {
1036         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1037                 return;
1038         if (nfs4_copy_lock_stateid(dst, state, lockowner))
1039                 return;
1040         nfs4_copy_open_stateid(dst, state);
1041 }
1042
1043 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1044 {
1045         struct nfs_seqid *new;
1046
1047         new = kmalloc(sizeof(*new), gfp_mask);
1048         if (new != NULL) {
1049                 new->sequence = counter;
1050                 INIT_LIST_HEAD(&new->list);
1051                 new->task = NULL;
1052         }
1053         return new;
1054 }
1055
1056 void nfs_release_seqid(struct nfs_seqid *seqid)
1057 {
1058         struct nfs_seqid_counter *sequence;
1059
1060         if (list_empty(&seqid->list))
1061                 return;
1062         sequence = seqid->sequence;
1063         spin_lock(&sequence->lock);
1064         list_del_init(&seqid->list);
1065         if (!list_empty(&sequence->list)) {
1066                 struct nfs_seqid *next;
1067
1068                 next = list_first_entry(&sequence->list,
1069                                 struct nfs_seqid, list);
1070                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1071         }
1072         spin_unlock(&sequence->lock);
1073 }
1074
1075 void nfs_free_seqid(struct nfs_seqid *seqid)
1076 {
1077         nfs_release_seqid(seqid);
1078         kfree(seqid);
1079 }
1080
1081 /*
1082  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1083  * failed with a seqid incrementing error -
1084  * see comments nfs_fs.h:seqid_mutating_error()
1085  */
1086 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1087 {
1088         switch (status) {
1089                 case 0:
1090                         break;
1091                 case -NFS4ERR_BAD_SEQID:
1092                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1093                                 return;
1094                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1095                                         " sequence-id error on an"
1096                                         " unconfirmed sequence %p!\n",
1097                                         seqid->sequence);
1098                 case -NFS4ERR_STALE_CLIENTID:
1099                 case -NFS4ERR_STALE_STATEID:
1100                 case -NFS4ERR_BAD_STATEID:
1101                 case -NFS4ERR_BADXDR:
1102                 case -NFS4ERR_RESOURCE:
1103                 case -NFS4ERR_NOFILEHANDLE:
1104                         /* Non-seqid mutating errors */
1105                         return;
1106         };
1107         /*
1108          * Note: no locking needed as we are guaranteed to be first
1109          * on the sequence list
1110          */
1111         seqid->sequence->counter++;
1112 }
1113
1114 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1115 {
1116         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1117                                         struct nfs4_state_owner, so_seqid);
1118         struct nfs_server *server = sp->so_server;
1119
1120         if (status == -NFS4ERR_BAD_SEQID)
1121                 nfs4_drop_state_owner(sp);
1122         if (!nfs4_has_session(server->nfs_client))
1123                 nfs_increment_seqid(status, seqid);
1124 }
1125
1126 /*
1127  * Increment the seqid if the LOCK/LOCKU succeeded, or
1128  * failed with a seqid incrementing error -
1129  * see comments nfs_fs.h:seqid_mutating_error()
1130  */
1131 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1132 {
1133         nfs_increment_seqid(status, seqid);
1134 }
1135
1136 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1137 {
1138         struct nfs_seqid_counter *sequence = seqid->sequence;
1139         int status = 0;
1140
1141         spin_lock(&sequence->lock);
1142         seqid->task = task;
1143         if (list_empty(&seqid->list))
1144                 list_add_tail(&seqid->list, &sequence->list);
1145         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1146                 goto unlock;
1147         rpc_sleep_on(&sequence->wait, task, NULL);
1148         status = -EAGAIN;
1149 unlock:
1150         spin_unlock(&sequence->lock);
1151         return status;
1152 }
1153
1154 static int nfs4_run_state_manager(void *);
1155
1156 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1157 {
1158         smp_mb__before_clear_bit();
1159         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1160         smp_mb__after_clear_bit();
1161         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1162         rpc_wake_up(&clp->cl_rpcwaitq);
1163 }
1164
1165 /*
1166  * Schedule the nfs_client asynchronous state management routine
1167  */
1168 void nfs4_schedule_state_manager(struct nfs_client *clp)
1169 {
1170         struct task_struct *task;
1171         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1172
1173         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1174                 return;
1175         __module_get(THIS_MODULE);
1176         atomic_inc(&clp->cl_count);
1177
1178         /* The rcu_read_lock() is not strictly necessary, as the state
1179          * manager is the only thread that ever changes the rpc_xprt
1180          * after it's initialized.  At this point, we're single threaded. */
1181         rcu_read_lock();
1182         snprintf(buf, sizeof(buf), "%s-manager",
1183                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1184         rcu_read_unlock();
1185         task = kthread_run(nfs4_run_state_manager, clp, buf);
1186         if (IS_ERR(task)) {
1187                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1188                         __func__, PTR_ERR(task));
1189                 nfs4_clear_state_manager_bit(clp);
1190                 nfs_put_client(clp);
1191                 module_put(THIS_MODULE);
1192         }
1193 }
1194
1195 /*
1196  * Schedule a lease recovery attempt
1197  */
1198 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1199 {
1200         if (!clp)
1201                 return;
1202         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1203                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1204         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1205                         clp->cl_hostname);
1206         nfs4_schedule_state_manager(clp);
1207 }
1208 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1209
1210 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1211 {
1212         int res;
1213
1214         might_sleep();
1215
1216         res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1217                         nfs_wait_bit_killable, TASK_KILLABLE);
1218         if (res)
1219                 return res;
1220
1221         if (clp->cl_cons_state < 0)
1222                 return clp->cl_cons_state;
1223         return 0;
1224 }
1225
1226 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1227 {
1228         unsigned int loop;
1229         int ret;
1230
1231         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1232                 ret = nfs4_wait_clnt_recover(clp);
1233                 if (ret != 0)
1234                         break;
1235                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1236                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1237                         break;
1238                 nfs4_schedule_state_manager(clp);
1239                 ret = -EIO;
1240         }
1241         return ret;
1242 }
1243
1244 /*
1245  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1246  * @clp: client to process
1247  *
1248  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1249  * resend of the SETCLIENTID and hence re-establish the
1250  * callback channel. Then return all existing delegations.
1251  */
1252 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1253 {
1254         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1255         nfs_expire_all_delegations(clp);
1256         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1257                         clp->cl_hostname);
1258 }
1259
1260 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1261 {
1262         nfs40_handle_cb_pathdown(clp);
1263         nfs4_schedule_state_manager(clp);
1264 }
1265
1266 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1267 {
1268
1269         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1270         /* Don't recover state that expired before the reboot */
1271         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1272                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1273                 return 0;
1274         }
1275         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1276         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1277         return 1;
1278 }
1279
1280 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1281 {
1282         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1283         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1284         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1285         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1286         return 1;
1287 }
1288
1289 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1290 {
1291         struct nfs_client *clp = server->nfs_client;
1292
1293         nfs4_state_mark_reclaim_nograce(clp, state);
1294         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1295                         clp->cl_hostname);
1296         nfs4_schedule_state_manager(clp);
1297 }
1298 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1299
1300 void nfs_inode_find_state_and_recover(struct inode *inode,
1301                 const nfs4_stateid *stateid)
1302 {
1303         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1304         struct nfs_inode *nfsi = NFS_I(inode);
1305         struct nfs_open_context *ctx;
1306         struct nfs4_state *state;
1307         bool found = false;
1308
1309         spin_lock(&inode->i_lock);
1310         list_for_each_entry(ctx, &nfsi->open_files, list) {
1311                 state = ctx->state;
1312                 if (state == NULL)
1313                         continue;
1314                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1315                         continue;
1316                 if (!nfs4_stateid_match(&state->stateid, stateid))
1317                         continue;
1318                 nfs4_state_mark_reclaim_nograce(clp, state);
1319                 found = true;
1320         }
1321         spin_unlock(&inode->i_lock);
1322         if (found)
1323                 nfs4_schedule_state_manager(clp);
1324 }
1325
1326
1327 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1328 {
1329         struct inode *inode = state->inode;
1330         struct nfs_inode *nfsi = NFS_I(inode);
1331         struct file_lock *fl;
1332         int status = 0;
1333
1334         if (inode->i_flock == NULL)
1335                 return 0;
1336
1337         /* Guard against delegation returns and new lock/unlock calls */
1338         down_write(&nfsi->rwsem);
1339         /* Protect inode->i_flock using the BKL */
1340         lock_flocks();
1341         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1342                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1343                         continue;
1344                 if (nfs_file_open_context(fl->fl_file)->state != state)
1345                         continue;
1346                 unlock_flocks();
1347                 status = ops->recover_lock(state, fl);
1348                 switch (status) {
1349                         case 0:
1350                                 break;
1351                         case -ESTALE:
1352                         case -NFS4ERR_ADMIN_REVOKED:
1353                         case -NFS4ERR_STALE_STATEID:
1354                         case -NFS4ERR_BAD_STATEID:
1355                         case -NFS4ERR_EXPIRED:
1356                         case -NFS4ERR_NO_GRACE:
1357                         case -NFS4ERR_STALE_CLIENTID:
1358                         case -NFS4ERR_BADSESSION:
1359                         case -NFS4ERR_BADSLOT:
1360                         case -NFS4ERR_BAD_HIGH_SLOT:
1361                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1362                                 goto out;
1363                         default:
1364                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1365                                         "Zeroing state\n", __func__, status);
1366                         case -ENOMEM:
1367                         case -NFS4ERR_DENIED:
1368                         case -NFS4ERR_RECLAIM_BAD:
1369                         case -NFS4ERR_RECLAIM_CONFLICT:
1370                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1371                                 status = 0;
1372                 }
1373                 lock_flocks();
1374         }
1375         unlock_flocks();
1376 out:
1377         up_write(&nfsi->rwsem);
1378         return status;
1379 }
1380
1381 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1382 {
1383         struct nfs4_state *state;
1384         struct nfs4_lock_state *lock;
1385         int status = 0;
1386
1387         /* Note: we rely on the sp->so_states list being ordered
1388          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1389          * states first.
1390          * This is needed to ensure that the server won't give us any
1391          * read delegations that we have to return if, say, we are
1392          * recovering after a network partition or a reboot from a
1393          * server that doesn't support a grace period.
1394          */
1395         spin_lock(&sp->so_lock);
1396         write_seqcount_begin(&sp->so_reclaim_seqcount);
1397 restart:
1398         list_for_each_entry(state, &sp->so_states, open_states) {
1399                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1400                         continue;
1401                 if (state->state == 0)
1402                         continue;
1403                 atomic_inc(&state->count);
1404                 spin_unlock(&sp->so_lock);
1405                 status = ops->recover_open(sp, state);
1406                 if (status >= 0) {
1407                         status = nfs4_reclaim_locks(state, ops);
1408                         if (status >= 0) {
1409                                 spin_lock(&state->state_lock);
1410                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1411                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1412                                                 pr_warn_ratelimited("NFS: "
1413                                                         "%s: Lock reclaim "
1414                                                         "failed!\n", __func__);
1415                                 }
1416                                 spin_unlock(&state->state_lock);
1417                                 nfs4_put_open_state(state);
1418                                 spin_lock(&sp->so_lock);
1419                                 goto restart;
1420                         }
1421                 }
1422                 switch (status) {
1423                         default:
1424                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1425                                         "Zeroing state\n", __func__, status);
1426                         case -ENOENT:
1427                         case -ENOMEM:
1428                         case -ESTALE:
1429                                 /*
1430                                  * Open state on this file cannot be recovered
1431                                  * All we can do is revert to using the zero stateid.
1432                                  */
1433                                 memset(&state->stateid, 0,
1434                                         sizeof(state->stateid));
1435                                 /* Mark the file as being 'closed' */
1436                                 state->state = 0;
1437                                 break;
1438                         case -NFS4ERR_ADMIN_REVOKED:
1439                         case -NFS4ERR_STALE_STATEID:
1440                         case -NFS4ERR_BAD_STATEID:
1441                         case -NFS4ERR_RECLAIM_BAD:
1442                         case -NFS4ERR_RECLAIM_CONFLICT:
1443                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1444                                 break;
1445                         case -NFS4ERR_EXPIRED:
1446                         case -NFS4ERR_NO_GRACE:
1447                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1448                         case -NFS4ERR_STALE_CLIENTID:
1449                         case -NFS4ERR_BADSESSION:
1450                         case -NFS4ERR_BADSLOT:
1451                         case -NFS4ERR_BAD_HIGH_SLOT:
1452                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1453                                 goto out_err;
1454                 }
1455                 nfs4_put_open_state(state);
1456                 spin_lock(&sp->so_lock);
1457                 goto restart;
1458         }
1459         write_seqcount_end(&sp->so_reclaim_seqcount);
1460         spin_unlock(&sp->so_lock);
1461         return 0;
1462 out_err:
1463         nfs4_put_open_state(state);
1464         spin_lock(&sp->so_lock);
1465         write_seqcount_end(&sp->so_reclaim_seqcount);
1466         spin_unlock(&sp->so_lock);
1467         return status;
1468 }
1469
1470 static void nfs4_clear_open_state(struct nfs4_state *state)
1471 {
1472         struct nfs4_lock_state *lock;
1473
1474         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1475         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1476         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1477         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1478         spin_lock(&state->state_lock);
1479         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1480                 lock->ls_seqid.flags = 0;
1481                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1482         }
1483         spin_unlock(&state->state_lock);
1484 }
1485
1486 static void nfs4_reset_seqids(struct nfs_server *server,
1487         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1488 {
1489         struct nfs_client *clp = server->nfs_client;
1490         struct nfs4_state_owner *sp;
1491         struct rb_node *pos;
1492         struct nfs4_state *state;
1493
1494         spin_lock(&clp->cl_lock);
1495         for (pos = rb_first(&server->state_owners);
1496              pos != NULL;
1497              pos = rb_next(pos)) {
1498                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1499                 sp->so_seqid.flags = 0;
1500                 spin_lock(&sp->so_lock);
1501                 list_for_each_entry(state, &sp->so_states, open_states) {
1502                         if (mark_reclaim(clp, state))
1503                                 nfs4_clear_open_state(state);
1504                 }
1505                 spin_unlock(&sp->so_lock);
1506         }
1507         spin_unlock(&clp->cl_lock);
1508 }
1509
1510 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1511         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1512 {
1513         struct nfs_server *server;
1514
1515         rcu_read_lock();
1516         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1517                 nfs4_reset_seqids(server, mark_reclaim);
1518         rcu_read_unlock();
1519 }
1520
1521 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1522 {
1523         /* Mark all delegations for reclaim */
1524         nfs_delegation_mark_reclaim(clp);
1525         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1526 }
1527
1528 static void nfs4_reclaim_complete(struct nfs_client *clp,
1529                                  const struct nfs4_state_recovery_ops *ops)
1530 {
1531         /* Notify the server we're done reclaiming our state */
1532         if (ops->reclaim_complete)
1533                 (void)ops->reclaim_complete(clp);
1534 }
1535
1536 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1537 {
1538         struct nfs_client *clp = server->nfs_client;
1539         struct nfs4_state_owner *sp;
1540         struct rb_node *pos;
1541         struct nfs4_state *state;
1542
1543         spin_lock(&clp->cl_lock);
1544         for (pos = rb_first(&server->state_owners);
1545              pos != NULL;
1546              pos = rb_next(pos)) {
1547                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1548                 spin_lock(&sp->so_lock);
1549                 list_for_each_entry(state, &sp->so_states, open_states) {
1550                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1551                                                 &state->flags))
1552                                 continue;
1553                         nfs4_state_mark_reclaim_nograce(clp, state);
1554                 }
1555                 spin_unlock(&sp->so_lock);
1556         }
1557         spin_unlock(&clp->cl_lock);
1558 }
1559
1560 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1561 {
1562         struct nfs_server *server;
1563
1564         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1565                 return 0;
1566
1567         rcu_read_lock();
1568         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1569                 nfs4_clear_reclaim_server(server);
1570         rcu_read_unlock();
1571
1572         nfs_delegation_reap_unclaimed(clp);
1573         return 1;
1574 }
1575
1576 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1577 {
1578         if (!nfs4_state_clear_reclaim_reboot(clp))
1579                 return;
1580         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1581 }
1582
1583 static void nfs_delegation_clear_all(struct nfs_client *clp)
1584 {
1585         nfs_delegation_mark_reclaim(clp);
1586         nfs_delegation_reap_unclaimed(clp);
1587 }
1588
1589 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1590 {
1591         nfs_delegation_clear_all(clp);
1592         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1593 }
1594
1595 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1596 {
1597         switch (error) {
1598                 case 0:
1599                         break;
1600                 case -NFS4ERR_CB_PATH_DOWN:
1601                         nfs40_handle_cb_pathdown(clp);
1602                         break;
1603                 case -NFS4ERR_NO_GRACE:
1604                         nfs4_state_end_reclaim_reboot(clp);
1605                         break;
1606                 case -NFS4ERR_STALE_CLIENTID:
1607                 case -NFS4ERR_LEASE_MOVED:
1608                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1609                         nfs4_state_clear_reclaim_reboot(clp);
1610                         nfs4_state_start_reclaim_reboot(clp);
1611                         break;
1612                 case -NFS4ERR_EXPIRED:
1613                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1614                         nfs4_state_start_reclaim_nograce(clp);
1615                         break;
1616                 case -NFS4ERR_BADSESSION:
1617                 case -NFS4ERR_BADSLOT:
1618                 case -NFS4ERR_BAD_HIGH_SLOT:
1619                 case -NFS4ERR_DEADSESSION:
1620                 case -NFS4ERR_SEQ_FALSE_RETRY:
1621                 case -NFS4ERR_SEQ_MISORDERED:
1622                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1623                         /* Zero session reset errors */
1624                         break;
1625                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1626                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1627                         break;
1628                 default:
1629                         dprintk("%s: failed to handle error %d for server %s\n",
1630                                         __func__, error, clp->cl_hostname);
1631                         return error;
1632         }
1633         dprintk("%s: handled error %d for server %s\n", __func__, error,
1634                         clp->cl_hostname);
1635         return 0;
1636 }
1637
1638 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1639 {
1640         struct nfs4_state_owner *sp;
1641         struct nfs_server *server;
1642         struct rb_node *pos;
1643         int status = 0;
1644
1645 restart:
1646         rcu_read_lock();
1647         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1648                 nfs4_purge_state_owners(server);
1649                 spin_lock(&clp->cl_lock);
1650                 for (pos = rb_first(&server->state_owners);
1651                      pos != NULL;
1652                      pos = rb_next(pos)) {
1653                         sp = rb_entry(pos,
1654                                 struct nfs4_state_owner, so_server_node);
1655                         if (!test_and_clear_bit(ops->owner_flag_bit,
1656                                                         &sp->so_flags))
1657                                 continue;
1658                         atomic_inc(&sp->so_count);
1659                         spin_unlock(&clp->cl_lock);
1660                         rcu_read_unlock();
1661
1662                         status = nfs4_reclaim_open_state(sp, ops);
1663                         if (status < 0) {
1664                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1665                                 nfs4_put_state_owner(sp);
1666                                 return nfs4_recovery_handle_error(clp, status);
1667                         }
1668
1669                         nfs4_put_state_owner(sp);
1670                         goto restart;
1671                 }
1672                 spin_unlock(&clp->cl_lock);
1673         }
1674         rcu_read_unlock();
1675         return status;
1676 }
1677
1678 static int nfs4_check_lease(struct nfs_client *clp)
1679 {
1680         struct rpc_cred *cred;
1681         const struct nfs4_state_maintenance_ops *ops =
1682                 clp->cl_mvops->state_renewal_ops;
1683         int status;
1684
1685         /* Is the client already known to have an expired lease? */
1686         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1687                 return 0;
1688         spin_lock(&clp->cl_lock);
1689         cred = ops->get_state_renewal_cred_locked(clp);
1690         spin_unlock(&clp->cl_lock);
1691         if (cred == NULL) {
1692                 cred = nfs4_get_setclientid_cred(clp);
1693                 status = -ENOKEY;
1694                 if (cred == NULL)
1695                         goto out;
1696         }
1697         status = ops->renew_lease(clp, cred);
1698         put_rpccred(cred);
1699 out:
1700         return nfs4_recovery_handle_error(clp, status);
1701 }
1702
1703 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1704  * and for recoverable errors on EXCHANGE_ID for v4.1
1705  */
1706 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1707 {
1708         switch (status) {
1709         case -NFS4ERR_SEQ_MISORDERED:
1710                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1711                         return -ESERVERFAULT;
1712                 /* Lease confirmation error: retry after purging the lease */
1713                 ssleep(1);
1714                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1715                 break;
1716         case -NFS4ERR_STALE_CLIENTID:
1717                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1718                 nfs4_state_clear_reclaim_reboot(clp);
1719                 nfs4_state_start_reclaim_reboot(clp);
1720                 break;
1721         case -NFS4ERR_CLID_INUSE:
1722                 pr_err("NFS: Server %s reports our clientid is in use\n",
1723                         clp->cl_hostname);
1724                 nfs_mark_client_ready(clp, -EPERM);
1725                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1726                 return -EPERM;
1727         case -EACCES:
1728                 if (clp->cl_machine_cred == NULL)
1729                         return -EACCES;
1730                 /* Handle case where the user hasn't set up machine creds */
1731                 nfs4_clear_machine_cred(clp);
1732         case -NFS4ERR_DELAY:
1733         case -ETIMEDOUT:
1734         case -EAGAIN:
1735                 ssleep(1);
1736                 break;
1737
1738         case -NFS4ERR_MINOR_VERS_MISMATCH:
1739                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1740                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1741                 dprintk("%s: exit with error %d for server %s\n",
1742                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1743                 return -EPROTONOSUPPORT;
1744         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1745                                  * in nfs4_exchange_id */
1746         default:
1747                 dprintk("%s: exit with error %d for server %s\n", __func__,
1748                                 status, clp->cl_hostname);
1749                 return status;
1750         }
1751         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1752         dprintk("%s: handled error %d for server %s\n", __func__, status,
1753                         clp->cl_hostname);
1754         return 0;
1755 }
1756
1757 static int nfs4_establish_lease(struct nfs_client *clp)
1758 {
1759         struct rpc_cred *cred;
1760         const struct nfs4_state_recovery_ops *ops =
1761                 clp->cl_mvops->reboot_recovery_ops;
1762         int status;
1763
1764         cred = ops->get_clid_cred(clp);
1765         if (cred == NULL)
1766                 return -ENOENT;
1767         status = ops->establish_clid(clp, cred);
1768         put_rpccred(cred);
1769         if (status != 0)
1770                 return status;
1771         pnfs_destroy_all_layouts(clp);
1772         return 0;
1773 }
1774
1775 /*
1776  * Returns zero or a negative errno.  NFS4ERR values are converted
1777  * to local errno values.
1778  */
1779 static int nfs4_reclaim_lease(struct nfs_client *clp)
1780 {
1781         int status;
1782
1783         status = nfs4_establish_lease(clp);
1784         if (status < 0)
1785                 return nfs4_handle_reclaim_lease_error(clp, status);
1786         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1787                 nfs4_state_start_reclaim_nograce(clp);
1788         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1789                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1790         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1791         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1792         return 0;
1793 }
1794
1795 static int nfs4_purge_lease(struct nfs_client *clp)
1796 {
1797         int status;
1798
1799         status = nfs4_establish_lease(clp);
1800         if (status < 0)
1801                 return nfs4_handle_reclaim_lease_error(clp, status);
1802         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1803         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1804         nfs4_state_start_reclaim_nograce(clp);
1805         return 0;
1806 }
1807
1808 /**
1809  * nfs4_discover_server_trunking - Detect server IP address trunking
1810  *
1811  * @clp: nfs_client under test
1812  * @result: OUT: found nfs_client, or clp
1813  *
1814  * Returns zero or a negative errno.  If zero is returned,
1815  * an nfs_client pointer is planted in "result".
1816  *
1817  * Note: since we are invoked in process context, and
1818  * not from inside the state manager, we cannot use
1819  * nfs4_handle_reclaim_lease_error().
1820  */
1821 int nfs4_discover_server_trunking(struct nfs_client *clp,
1822                                   struct nfs_client **result)
1823 {
1824         const struct nfs4_state_recovery_ops *ops =
1825                                 clp->cl_mvops->reboot_recovery_ops;
1826         rpc_authflavor_t *flavors, flav, save;
1827         struct rpc_clnt *clnt;
1828         struct rpc_cred *cred;
1829         int i, len, status;
1830
1831         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1832
1833         len = NFS_MAX_SECFLAVORS;
1834         flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1835         if (flavors == NULL) {
1836                 status = -ENOMEM;
1837                 goto out;
1838         }
1839         len = rpcauth_list_flavors(flavors, len);
1840         if (len < 0) {
1841                 status = len;
1842                 goto out_free;
1843         }
1844         clnt = clp->cl_rpcclient;
1845         save = clnt->cl_auth->au_flavor;
1846         i = 0;
1847
1848         mutex_lock(&nfs_clid_init_mutex);
1849         status  = -ENOENT;
1850 again:
1851         cred = ops->get_clid_cred(clp);
1852         if (cred == NULL)
1853                 goto out_unlock;
1854
1855         status = ops->detect_trunking(clp, result, cred);
1856         put_rpccred(cred);
1857         switch (status) {
1858         case 0:
1859                 break;
1860
1861         case -EACCES:
1862                 if (clp->cl_machine_cred == NULL)
1863                         break;
1864                 /* Handle case where the user hasn't set up machine creds */
1865                 nfs4_clear_machine_cred(clp);
1866         case -NFS4ERR_DELAY:
1867         case -ETIMEDOUT:
1868         case -EAGAIN:
1869                 ssleep(1);
1870         case -NFS4ERR_STALE_CLIENTID:
1871                 dprintk("NFS: %s after status %d, retrying\n",
1872                         __func__, status);
1873                 goto again;
1874
1875         case -NFS4ERR_CLID_INUSE:
1876         case -NFS4ERR_WRONGSEC:
1877                 status = -EPERM;
1878                 if (i >= len)
1879                         break;
1880
1881                 flav = flavors[i++];
1882                 if (flav == save)
1883                         flav = flavors[i++];
1884                 clnt = rpc_clone_client_set_auth(clnt, flav);
1885                 if (IS_ERR(clnt)) {
1886                         status = PTR_ERR(clnt);
1887                         break;
1888                 }
1889                 clp->cl_rpcclient = clnt;
1890                 goto again;
1891
1892         case -NFS4ERR_MINOR_VERS_MISMATCH:
1893                 status = -EPROTONOSUPPORT;
1894                 break;
1895
1896         case -EKEYEXPIRED:
1897         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1898                                  * in nfs4_exchange_id */
1899                 status = -EKEYEXPIRED;
1900         }
1901
1902 out_unlock:
1903         mutex_unlock(&nfs_clid_init_mutex);
1904 out_free:
1905         kfree(flavors);
1906 out:
1907         dprintk("NFS: %s: status = %d\n", __func__, status);
1908         return status;
1909 }
1910
1911 #ifdef CONFIG_NFS_V4_1
1912 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1913 {
1914         struct nfs_client *clp = session->clp;
1915
1916         switch (err) {
1917         default:
1918                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1919                 break;
1920         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1921                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1922         }
1923         nfs4_schedule_lease_recovery(clp);
1924 }
1925 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1926
1927 static void nfs41_ping_server(struct nfs_client *clp)
1928 {
1929         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
1930         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1931         nfs4_schedule_state_manager(clp);
1932 }
1933
1934 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
1935 {
1936         nfs41_ping_server(clp);
1937 }
1938
1939 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
1940 {
1941         nfs41_ping_server(clp);
1942 }
1943
1944 static void nfs4_reset_all_state(struct nfs_client *clp)
1945 {
1946         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1947                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1948                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1949                 nfs4_state_start_reclaim_nograce(clp);
1950                 dprintk("%s: scheduling reset of all state for server %s!\n",
1951                                 __func__, clp->cl_hostname);
1952                 nfs4_schedule_state_manager(clp);
1953         }
1954 }
1955
1956 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1957 {
1958         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1959                 nfs4_state_start_reclaim_reboot(clp);
1960                 dprintk("%s: server %s rebooted!\n", __func__,
1961                                 clp->cl_hostname);
1962                 nfs4_schedule_state_manager(clp);
1963         }
1964 }
1965
1966 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1967 {
1968         nfs4_reset_all_state(clp);
1969         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1970 }
1971
1972 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1973 {
1974         /* This will need to handle layouts too */
1975         nfs_expire_all_delegations(clp);
1976         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1977                         clp->cl_hostname);
1978 }
1979
1980 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1981 {
1982         nfs_expire_all_delegations(clp);
1983         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1984                 nfs4_schedule_state_manager(clp);
1985         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1986                         clp->cl_hostname);
1987 }
1988
1989 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1990 {
1991         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1992                 &clp->cl_state) == 0)
1993                 nfs4_schedule_state_manager(clp);
1994 }
1995
1996 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1997 {
1998         if (!flags)
1999                 return;
2000
2001         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2002                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2003
2004         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2005                 nfs41_handle_server_reboot(clp);
2006         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
2007                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2008                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
2009                             SEQ4_STATUS_LEASE_MOVED))
2010                 nfs41_handle_state_revoked(clp);
2011         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2012                 nfs41_handle_recallable_state_revoked(clp);
2013         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2014                 nfs41_handle_backchannel_fault(clp);
2015         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2016                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2017                 nfs41_handle_cb_path_down(clp);
2018 }
2019
2020 static int nfs4_reset_session(struct nfs_client *clp)
2021 {
2022         struct rpc_cred *cred;
2023         int status;
2024
2025         if (!nfs4_has_session(clp))
2026                 return 0;
2027         nfs4_begin_drain_session(clp);
2028         cred = nfs4_get_exchange_id_cred(clp);
2029         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2030         switch (status) {
2031         case 0:
2032         case -NFS4ERR_BADSESSION:
2033         case -NFS4ERR_DEADSESSION:
2034                 break;
2035         case -NFS4ERR_BACK_CHAN_BUSY:
2036         case -NFS4ERR_DELAY:
2037                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2038                 status = 0;
2039                 ssleep(1);
2040                 goto out;
2041         default:
2042                 status = nfs4_recovery_handle_error(clp, status);
2043                 goto out;
2044         }
2045
2046         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2047         status = nfs4_proc_create_session(clp, cred);
2048         if (status) {
2049                 dprintk("%s: session reset failed with status %d for server %s!\n",
2050                         __func__, status, clp->cl_hostname);
2051                 status = nfs4_handle_reclaim_lease_error(clp, status);
2052                 goto out;
2053         }
2054         nfs41_finish_session_reset(clp);
2055         dprintk("%s: session reset was successful for server %s!\n",
2056                         __func__, clp->cl_hostname);
2057 out:
2058         if (cred)
2059                 put_rpccred(cred);
2060         return status;
2061 }
2062
2063 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2064 {
2065         struct rpc_cred *cred;
2066         int ret;
2067
2068         if (!nfs4_has_session(clp))
2069                 return 0;
2070         nfs4_begin_drain_session(clp);
2071         cred = nfs4_get_exchange_id_cred(clp);
2072         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2073         if (cred)
2074                 put_rpccred(cred);
2075         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2076         switch (ret) {
2077         case 0:
2078                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2079                         __func__, clp->cl_hostname);
2080                 break;
2081         case -NFS4ERR_DELAY:
2082                 ssleep(1);
2083                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2084                 break;
2085         default:
2086                 return nfs4_recovery_handle_error(clp, ret);
2087         }
2088         return 0;
2089 }
2090 #else /* CONFIG_NFS_V4_1 */
2091 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2092 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2093
2094 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2095 {
2096         return 0;
2097 }
2098 #endif /* CONFIG_NFS_V4_1 */
2099
2100 static void nfs4_state_manager(struct nfs_client *clp)
2101 {
2102         int status = 0;
2103         const char *section = "", *section_sep = "";
2104
2105         /* Ensure exclusive access to NFSv4 state */
2106         do {
2107                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2108                         section = "purge state";
2109                         status = nfs4_purge_lease(clp);
2110                         if (status < 0)
2111                                 goto out_error;
2112                         continue;
2113                 }
2114
2115                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2116                         section = "lease expired";
2117                         /* We're going to have to re-establish a clientid */
2118                         status = nfs4_reclaim_lease(clp);
2119                         if (status < 0)
2120                                 goto out_error;
2121                         continue;
2122                 }
2123
2124                 /* Initialize or reset the session */
2125                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2126                         section = "reset session";
2127                         status = nfs4_reset_session(clp);
2128                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2129                                 continue;
2130                         if (status < 0)
2131                                 goto out_error;
2132                 }
2133
2134                 /* Send BIND_CONN_TO_SESSION */
2135                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2136                                 &clp->cl_state)) {
2137                         section = "bind conn to session";
2138                         status = nfs4_bind_conn_to_session(clp);
2139                         if (status < 0)
2140                                 goto out_error;
2141                         continue;
2142                 }
2143
2144                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2145                         section = "check lease";
2146                         status = nfs4_check_lease(clp);
2147                         if (status < 0)
2148                                 goto out_error;
2149                         continue;
2150                 }
2151
2152                 /* First recover reboot state... */
2153                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2154                         section = "reclaim reboot";
2155                         status = nfs4_do_reclaim(clp,
2156                                 clp->cl_mvops->reboot_recovery_ops);
2157                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2158                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2159                                 continue;
2160                         nfs4_state_end_reclaim_reboot(clp);
2161                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2162                                 continue;
2163                         if (status < 0)
2164                                 goto out_error;
2165                 }
2166
2167                 /* Now recover expired state... */
2168                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2169                         section = "reclaim nograce";
2170                         status = nfs4_do_reclaim(clp,
2171                                 clp->cl_mvops->nograce_recovery_ops);
2172                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2173                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2174                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2175                                 continue;
2176                         if (status < 0)
2177                                 goto out_error;
2178                 }
2179
2180                 nfs4_end_drain_session(clp);
2181                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2182                         nfs_client_return_marked_delegations(clp);
2183                         continue;
2184                 }
2185
2186                 nfs4_clear_state_manager_bit(clp);
2187                 /* Did we race with an attempt to give us more work? */
2188                 if (clp->cl_state == 0)
2189                         break;
2190                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2191                         break;
2192         } while (atomic_read(&clp->cl_count) > 1);
2193         return;
2194 out_error:
2195         if (strlen(section))
2196                 section_sep = ": ";
2197         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2198                         " with error %d\n", section_sep, section,
2199                         clp->cl_hostname, -status);
2200         ssleep(1);
2201         nfs4_end_drain_session(clp);
2202         nfs4_clear_state_manager_bit(clp);
2203 }
2204
2205 static int nfs4_run_state_manager(void *ptr)
2206 {
2207         struct nfs_client *clp = ptr;
2208
2209         allow_signal(SIGKILL);
2210         nfs4_state_manager(clp);
2211         nfs_put_client(clp);
2212         module_put_and_exit(0);
2213         return 0;
2214 }
2215
2216 /*
2217  * Local variables:
2218  *  c-basic-offset: 8
2219  * End:
2220  */