fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52 #include <linux/jiffies.h>
  53
  54 #include <linux/sunrpc/clnt.h>
  55
  56 #include "nfs4_fs.h"
  57 #include "callback.h"
  58 #include "delegation.h"
  59 #include "internal.h"
  60 #include "pnfs.h"
  61 #include "netns.h"
  62
  63 #define NFSDBG_FACILITY         NFSDBG_STATE
  64
  65 #define OPENOWNER_POOL_SIZE     8
  66
  67 const nfs4_stateid zero_stateid;
  68 static DEFINE_MUTEX(nfs_clid_init_mutex);
  69 static LIST_HEAD(nfs4_clientid_list);
  70
  71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  72 {
  73         struct nfs4_setclientid_res clid = {
  74                 .clientid = clp->cl_clientid,
  75                 .confirm = clp->cl_confirm,
  76         };
  77         unsigned short port;
  78         int status;
  79         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  80
  81         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  82                 goto do_confirm;
  83         port = nn->nfs_callback_tcpport;
  84         if (clp->cl_addr.ss_family == AF_INET6)
  85                 port = nn->nfs_callback_tcpport6;
  86
  87         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  88         if (status != 0)
  89                 goto out;
  90         clp->cl_clientid = clid.clientid;
  91         clp->cl_confirm = clid.confirm;
  92         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  93 do_confirm:
  94         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  95         if (status != 0)
  96                 goto out;
  97         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  98         nfs4_schedule_state_renewal(clp);
  99 out:
 100         return status;
 101 }
 102
 103 /**
 104  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 105  *
 106  * @clp: nfs_client under test
 107  * @result: OUT: found nfs_client, or clp
 108  * @cred: credential to use for trunking test
 109  *
 110  * Returns zero, a negative errno, or a negative NFS4ERR status.
 111  * If zero is returned, an nfs_client pointer is planted in
 112  * "result".
 113  *
 114  * Note: The returned client may not yet be marked ready.
 115  */
 116 int nfs40_discover_server_trunking(struct nfs_client *clp,
 117                                    struct nfs_client **result,
 118                                    struct rpc_cred *cred)
 119 {
 120         struct nfs4_setclientid_res clid = {
 121                 .clientid = clp->cl_clientid,
 122                 .confirm = clp->cl_confirm,
 123         };
 124         unsigned short port;
 125         int status;
 126
 127         port = nfs_callback_tcpport;
 128         if (clp->cl_addr.ss_family == AF_INET6)
 129                 port = nfs_callback_tcpport6;
 130
 131         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 132         if (status != 0)
 133                 goto out;
 134         clp->cl_clientid = clid.clientid;
 135         clp->cl_confirm = clid.confirm;
 136
 137         status = nfs40_walk_client_list(clp, result, cred);
 138         switch (status) {
 139         case -NFS4ERR_STALE_CLIENTID:
 140                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 141         case 0:
 142                 /* Sustain the lease, even if it's empty.  If the clientid4
 143                  * goes stale it's of no use for trunking discovery. */
 144                 nfs4_schedule_state_renewal(*result);
 145                 break;
 146         }
 147
 148 out:
 149         return status;
 150 }
 151
 152 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 153 {
 154         struct rpc_cred *cred = NULL;
 155
 156         if (clp->cl_machine_cred != NULL)
 157                 cred = get_rpccred(clp->cl_machine_cred);
 158         return cred;
 159 }
 160
 161 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 162 {
 163         struct rpc_cred *cred;
 164
 165         spin_lock(&clp->cl_lock);
 166         cred = clp->cl_machine_cred;
 167         clp->cl_machine_cred = NULL;
 168         spin_unlock(&clp->cl_lock);
 169         if (cred != NULL)
 170                 put_rpccred(cred);
 171 }
 172
 173 static struct rpc_cred *
 174 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 175 {
 176         struct rpc_cred *cred = NULL;
 177         struct nfs4_state_owner *sp;
 178         struct rb_node *pos;
 179
 180         for (pos = rb_first(&server->state_owners);
 181              pos != NULL;
 182              pos = rb_next(pos)) {
 183                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 184                 if (list_empty(&sp->so_states))
 185                         continue;
 186                 cred = get_rpccred(sp->so_cred);
 187                 break;
 188         }
 189         return cred;
 190 }
 191
 192 /**
 193  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 194  * @clp: client state handle
 195  *
 196  * Returns an rpc_cred with reference count bumped, or NULL.
 197  * Caller must hold clp->cl_lock.
 198  */
 199 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 200 {
 201         struct rpc_cred *cred = NULL;
 202         struct nfs_server *server;
 203
 204         /* Use machine credentials if available */
 205         cred = nfs4_get_machine_cred_locked(clp);
 206         if (cred != NULL)
 207                 goto out;
 208
 209         rcu_read_lock();
 210         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 211                 cred = nfs4_get_renew_cred_server_locked(server);
 212                 if (cred != NULL)
 213                         break;
 214         }
 215         rcu_read_unlock();
 216
 217 out:
 218         return cred;
 219 }
 220
 221 #if defined(CONFIG_NFS_V4_1)
 222
 223 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 224 {
 225         int status;
 226         struct nfs_fsinfo fsinfo;
 227
 228         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 229                 nfs4_schedule_state_renewal(clp);
 230                 return 0;
 231         }
 232
 233         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 234         if (status == 0) {
 235                 /* Update lease time and schedule renewal */
 236                 spin_lock(&clp->cl_lock);
 237                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 238                 clp->cl_last_renewal = jiffies;
 239                 spin_unlock(&clp->cl_lock);
 240
 241                 nfs4_schedule_state_renewal(clp);
 242         }
 243
 244         return status;
 245 }
 246
 247 /*
 248  * Back channel returns NFS4ERR_DELAY for new requests when
 249  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 250  * is ended.
 251  */
 252 static void nfs4_end_drain_session(struct nfs_client *clp)
 253 {
 254         struct nfs4_session *ses = clp->cl_session;
 255         struct nfs4_slot_table *tbl;
 256         int max_slots;
 257
 258         if (ses == NULL)
 259                 return;
 260         tbl = &ses->fc_slot_table;
 261         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 262                 spin_lock(&tbl->slot_tbl_lock);
 263                 max_slots = tbl->max_slots;
 264                 while (max_slots--) {
 265                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 266                                                 nfs4_set_task_privileged,
 267                                                 NULL) == NULL)
 268                                 break;
 269                 }
 270                 spin_unlock(&tbl->slot_tbl_lock);
 271         }
 272 }
 273
 274 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 275 {
 276         spin_lock(&tbl->slot_tbl_lock);
 277         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 278                 INIT_COMPLETION(tbl->complete);
 279                 spin_unlock(&tbl->slot_tbl_lock);
 280                 return wait_for_completion_interruptible(&tbl->complete);
 281         }
 282         spin_unlock(&tbl->slot_tbl_lock);
 283         return 0;
 284 }
 285
 286 static int nfs4_begin_drain_session(struct nfs_client *clp)
 287 {
 288         struct nfs4_session *ses = clp->cl_session;
 289         int ret = 0;
 290
 291         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 292         /* back channel */
 293         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 294         if (ret)
 295                 return ret;
 296         /* fore channel */
 297         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 298 }
 299
 300 static void nfs41_finish_session_reset(struct nfs_client *clp)
 301 {
 302         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 303         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 304         /* create_session negotiated new slot table */
 305         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
 306         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 307         nfs41_setup_state_renewal(clp);
 308 }
 309
 310 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 311 {
 312         int status;
 313
 314         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 315                 goto do_confirm;
 316         nfs4_begin_drain_session(clp);
 317         status = nfs4_proc_exchange_id(clp, cred);
 318         if (status != 0)
 319                 goto out;
 320         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 321 do_confirm:
 322         status = nfs4_proc_create_session(clp, cred);
 323         if (status != 0)
 324                 goto out;
 325         nfs41_finish_session_reset(clp);
 326         nfs_mark_client_ready(clp, NFS_CS_READY);
 327 out:
 328         return status;
 329 }
 330
 331 /**
 332  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 333  *
 334  * @clp: nfs_client under test
 335  * @result: OUT: found nfs_client, or clp
 336  * @cred: credential to use for trunking test
 337  *
 338  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 339  * If NFS4_OK is returned, an nfs_client pointer is planted in
 340  * "result".
 341  *
 342  * Note: The returned client may not yet be marked ready.
 343  */
 344 int nfs41_discover_server_trunking(struct nfs_client *clp,
 345                                    struct nfs_client **result,
 346                                    struct rpc_cred *cred)
 347 {
 348         int status;
 349
 350         status = nfs4_proc_exchange_id(clp, cred);
 351         if (status != NFS4_OK)
 352                 return status;
 353
 354         return nfs41_walk_client_list(clp, result, cred);
 355 }
 356
 357 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 358 {
 359         struct rpc_cred *cred;
 360
 361         spin_lock(&clp->cl_lock);
 362         cred = nfs4_get_machine_cred_locked(clp);
 363         spin_unlock(&clp->cl_lock);
 364         return cred;
 365 }
 366
 367 #endif /* CONFIG_NFS_V4_1 */
 368
 369 static struct rpc_cred *
 370 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 371 {
 372         struct nfs_client *clp = server->nfs_client;
 373         struct rpc_cred *cred = NULL;
 374         struct nfs4_state_owner *sp;
 375         struct rb_node *pos;
 376
 377         spin_lock(&clp->cl_lock);
 378         pos = rb_first(&server->state_owners);
 379         if (pos != NULL) {
 380                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 381                 cred = get_rpccred(sp->so_cred);
 382         }
 383         spin_unlock(&clp->cl_lock);
 384         return cred;
 385 }
 386
 387 /**
 388  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 389  * @clp: client state handle
 390  *
 391  * Returns an rpc_cred with reference count bumped, or NULL.
 392  */
 393 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 394 {
 395         struct nfs_server *server;
 396         struct rpc_cred *cred;
 397
 398         spin_lock(&clp->cl_lock);
 399         cred = nfs4_get_machine_cred_locked(clp);
 400         spin_unlock(&clp->cl_lock);
 401         if (cred != NULL)
 402                 goto out;
 403
 404         rcu_read_lock();
 405         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 406                 cred = nfs4_get_setclientid_cred_server(server);
 407                 if (cred != NULL)
 408                         break;
 409         }
 410         rcu_read_unlock();
 411
 412 out:
 413         return cred;
 414 }
 415
 416 static struct nfs4_state_owner *
 417 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 418 {
 419         struct rb_node **p = &server->state_owners.rb_node,
 420                        *parent = NULL;
 421         struct nfs4_state_owner *sp;
 422
 423         while (*p != NULL) {
 424                 parent = *p;
 425                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 426
 427                 if (cred < sp->so_cred)
 428                         p = &parent->rb_left;
 429                 else if (cred > sp->so_cred)
 430                         p = &parent->rb_right;
 431                 else {
 432                         if (!list_empty(&sp->so_lru))
 433                                 list_del_init(&sp->so_lru);
 434                         atomic_inc(&sp->so_count);
 435                         return sp;
 436                 }
 437         }
 438         return NULL;
 439 }
 440
 441 static struct nfs4_state_owner *
 442 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 443 {
 444         struct nfs_server *server = new->so_server;
 445         struct rb_node **p = &server->state_owners.rb_node,
 446                        *parent = NULL;
 447         struct nfs4_state_owner *sp;
 448         int err;
 449
 450         while (*p != NULL) {
 451                 parent = *p;
 452                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 453
 454                 if (new->so_cred < sp->so_cred)
 455                         p = &parent->rb_left;
 456                 else if (new->so_cred > sp->so_cred)
 457                         p = &parent->rb_right;
 458                 else {
 459                         if (!list_empty(&sp->so_lru))
 460                                 list_del_init(&sp->so_lru);
 461                         atomic_inc(&sp->so_count);
 462                         return sp;
 463                 }
 464         }
 465         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 466         if (err)
 467                 return ERR_PTR(err);
 468         rb_link_node(&new->so_server_node, parent, p);
 469         rb_insert_color(&new->so_server_node, &server->state_owners);
 470         return new;
 471 }
 472
 473 static void
 474 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 475 {
 476         struct nfs_server *server = sp->so_server;
 477
 478         if (!RB_EMPTY_NODE(&sp->so_server_node))
 479                 rb_erase(&sp->so_server_node, &server->state_owners);
 480         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 481 }
 482
 483 static void
 484 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 485 {
 486         sc->create_time = ktime_get();
 487         sc->flags = 0;
 488         sc->counter = 0;
 489         spin_lock_init(&sc->lock);
 490         INIT_LIST_HEAD(&sc->list);
 491         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 492 }
 493
 494 static void
 495 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 496 {
 497         rpc_destroy_wait_queue(&sc->wait);
 498 }
 499
 500 /*
 501  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 502  * create a new state_owner.
 503  *
 504  */
 505 static struct nfs4_state_owner *
 506 nfs4_alloc_state_owner(struct nfs_server *server,
 507                 struct rpc_cred *cred,
 508                 gfp_t gfp_flags)
 509 {
 510         struct nfs4_state_owner *sp;
 511
 512         sp = kzalloc(sizeof(*sp), gfp_flags);
 513         if (!sp)
 514                 return NULL;
 515         sp->so_server = server;
 516         sp->so_cred = get_rpccred(cred);
 517         spin_lock_init(&sp->so_lock);
 518         INIT_LIST_HEAD(&sp->so_states);
 519         nfs4_init_seqid_counter(&sp->so_seqid);
 520         atomic_set(&sp->so_count, 1);
 521         INIT_LIST_HEAD(&sp->so_lru);
 522         return sp;
 523 }
 524
 525 static void
 526 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 527 {
 528         struct rb_node *rb_node = &sp->so_server_node;
 529
 530         if (!RB_EMPTY_NODE(rb_node)) {
 531                 struct nfs_server *server = sp->so_server;
 532                 struct nfs_client *clp = server->nfs_client;
 533
 534                 spin_lock(&clp->cl_lock);
 535                 if (!RB_EMPTY_NODE(rb_node)) {
 536                         rb_erase(rb_node, &server->state_owners);
 537                         RB_CLEAR_NODE(rb_node);
 538                 }
 539                 spin_unlock(&clp->cl_lock);
 540         }
 541 }
 542
 543 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 544 {
 545         nfs4_destroy_seqid_counter(&sp->so_seqid);
 546         put_rpccred(sp->so_cred);
 547         kfree(sp);
 548 }
 549
 550 static void nfs4_gc_state_owners(struct nfs_server *server)
 551 {
 552         struct nfs_client *clp = server->nfs_client;
 553         struct nfs4_state_owner *sp, *tmp;
 554         unsigned long time_min, time_max;
 555         LIST_HEAD(doomed);
 556
 557         spin_lock(&clp->cl_lock);
 558         time_max = jiffies;
 559         time_min = (long)time_max - (long)clp->cl_lease_time;
 560         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 561                 /* NB: LRU is sorted so that oldest is at the head */
 562                 if (time_in_range(sp->so_expires, time_min, time_max))
 563                         break;
 564                 list_move(&sp->so_lru, &doomed);
 565                 nfs4_remove_state_owner_locked(sp);
 566         }
 567         spin_unlock(&clp->cl_lock);
 568
 569         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 570                 list_del(&sp->so_lru);
 571                 nfs4_free_state_owner(sp);
 572         }
 573 }
 574
 575 /**
 576  * nfs4_get_state_owner - Look up a state owner given a credential
 577  * @server: nfs_server to search
 578  * @cred: RPC credential to match
 579  *
 580  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 581  */
 582 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 583                                               struct rpc_cred *cred,
 584                                               gfp_t gfp_flags)
 585 {
 586         struct nfs_client *clp = server->nfs_client;
 587         struct nfs4_state_owner *sp, *new;
 588
 589         spin_lock(&clp->cl_lock);
 590         sp = nfs4_find_state_owner_locked(server, cred);
 591         spin_unlock(&clp->cl_lock);
 592         if (sp != NULL)
 593                 goto out;
 594         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 595         if (new == NULL)
 596                 goto out;
 597         do {
 598                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 599                         break;
 600                 spin_lock(&clp->cl_lock);
 601                 sp = nfs4_insert_state_owner_locked(new);
 602                 spin_unlock(&clp->cl_lock);
 603         } while (sp == ERR_PTR(-EAGAIN));
 604         if (sp != new)
 605                 nfs4_free_state_owner(new);
 606 out:
 607         nfs4_gc_state_owners(server);
 608         return sp;
 609 }
 610
 611 /**
 612  * nfs4_put_state_owner - Release a nfs4_state_owner
 613  * @sp: state owner data to release
 614  *
 615  * Note that we keep released state owners on an LRU
 616  * list.
 617  * This caches valid state owners so that they can be
 618  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 619  * It also pins the uniquifier of dropped state owners for
 620  * a while, to ensure that those state owner names are
 621  * never reused.
 622  */
 623 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 624 {
 625         struct nfs_server *server = sp->so_server;
 626         struct nfs_client *clp = server->nfs_client;
 627
 628         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 629                 return;
 630
 631         sp->so_expires = jiffies;
 632         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 633         spin_unlock(&clp->cl_lock);
 634 }
 635
 636 /**
 637  * nfs4_purge_state_owners - Release all cached state owners
 638  * @server: nfs_server with cached state owners to release
 639  *
 640  * Called at umount time.  Remaining state owners will be on
 641  * the LRU with ref count of zero.
 642  */
 643 void nfs4_purge_state_owners(struct nfs_server *server)
 644 {
 645         struct nfs_client *clp = server->nfs_client;
 646         struct nfs4_state_owner *sp, *tmp;
 647         LIST_HEAD(doomed);
 648
 649         spin_lock(&clp->cl_lock);
 650         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 651                 list_move(&sp->so_lru, &doomed);
 652                 nfs4_remove_state_owner_locked(sp);
 653         }
 654         spin_unlock(&clp->cl_lock);
 655
 656         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 657                 list_del(&sp->so_lru);
 658                 nfs4_free_state_owner(sp);
 659         }
 660 }
 661
 662 static struct nfs4_state *
 663 nfs4_alloc_open_state(void)
 664 {
 665         struct nfs4_state *state;
 666
 667         state = kzalloc(sizeof(*state), GFP_NOFS);
 668         if (!state)
 669                 return NULL;
 670         atomic_set(&state->count, 1);
 671         INIT_LIST_HEAD(&state->lock_states);
 672         spin_lock_init(&state->state_lock);
 673         seqlock_init(&state->seqlock);
 674         return state;
 675 }
 676
 677 void
 678 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 679 {
 680         if (state->state == fmode)
 681                 return;
 682         /* NB! List reordering - see the reclaim code for why.  */
 683         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 684                 if (fmode & FMODE_WRITE)
 685                         list_move(&state->open_states, &state->owner->so_states);
 686                 else
 687                         list_move_tail(&state->open_states, &state->owner->so_states);
 688         }
 689         state->state = fmode;
 690 }
 691
 692 static struct nfs4_state *
 693 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 694 {
 695         struct nfs_inode *nfsi = NFS_I(inode);
 696         struct nfs4_state *state;
 697
 698         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 699                 if (state->owner != owner)
 700                         continue;
 701                 if (atomic_inc_not_zero(&state->count))
 702                         return state;
 703         }
 704         return NULL;
 705 }
 706
 707 static void
 708 nfs4_free_open_state(struct nfs4_state *state)
 709 {
 710         kfree(state);
 711 }
 712
 713 struct nfs4_state *
 714 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 715 {
 716         struct nfs4_state *state, *new;
 717         struct nfs_inode *nfsi = NFS_I(inode);
 718
 719         spin_lock(&inode->i_lock);
 720         state = __nfs4_find_state_byowner(inode, owner);
 721         spin_unlock(&inode->i_lock);
 722         if (state)
 723                 goto out;
 724         new = nfs4_alloc_open_state();
 725         spin_lock(&owner->so_lock);
 726         spin_lock(&inode->i_lock);
 727         state = __nfs4_find_state_byowner(inode, owner);
 728         if (state == NULL && new != NULL) {
 729                 state = new;
 730                 state->owner = owner;
 731                 atomic_inc(&owner->so_count);
 732                 list_add(&state->inode_states, &nfsi->open_states);
 733                 ihold(inode);
 734                 state->inode = inode;
 735                 spin_unlock(&inode->i_lock);
 736                 /* Note: The reclaim code dictates that we add stateless
 737                  * and read-only stateids to the end of the list */
 738                 list_add_tail(&state->open_states, &owner->so_states);
 739                 spin_unlock(&owner->so_lock);
 740         } else {
 741                 spin_unlock(&inode->i_lock);
 742                 spin_unlock(&owner->so_lock);
 743                 if (new)
 744                         nfs4_free_open_state(new);
 745         }
 746 out:
 747         return state;
 748 }
 749
 750 void nfs4_put_open_state(struct nfs4_state *state)
 751 {
 752         struct inode *inode = state->inode;
 753         struct nfs4_state_owner *owner = state->owner;
 754
 755         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 756                 return;
 757         spin_lock(&inode->i_lock);
 758         list_del(&state->inode_states);
 759         list_del(&state->open_states);
 760         spin_unlock(&inode->i_lock);
 761         spin_unlock(&owner->so_lock);
 762         iput(inode);
 763         nfs4_free_open_state(state);
 764         nfs4_put_state_owner(owner);
 765 }
 766
 767 /*
 768  * Close the current file.
 769  */
 770 static void __nfs4_close(struct nfs4_state *state,
 771                 fmode_t fmode, gfp_t gfp_mask, int wait)
 772 {
 773         struct nfs4_state_owner *owner = state->owner;
 774         int call_close = 0;
 775         fmode_t newstate;
 776
 777         atomic_inc(&owner->so_count);
 778         /* Protect against nfs4_find_state() */
 779         spin_lock(&owner->so_lock);
 780         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 781                 case FMODE_READ:
 782                         state->n_rdonly--;
 783                         break;
 784                 case FMODE_WRITE:
 785                         state->n_wronly--;
 786                         break;
 787                 case FMODE_READ|FMODE_WRITE:
 788                         state->n_rdwr--;
 789         }
 790         newstate = FMODE_READ|FMODE_WRITE;
 791         if (state->n_rdwr == 0) {
 792                 if (state->n_rdonly == 0) {
 793                         newstate &= ~FMODE_READ;
 794                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 795                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 796                 }
 797                 if (state->n_wronly == 0) {
 798                         newstate &= ~FMODE_WRITE;
 799                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 800                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 801                 }
 802                 if (newstate == 0)
 803                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 804         }
 805         nfs4_state_set_mode_locked(state, newstate);
 806         spin_unlock(&owner->so_lock);
 807
 808         if (!call_close) {
 809                 nfs4_put_open_state(state);
 810                 nfs4_put_state_owner(owner);
 811         } else
 812                 nfs4_do_close(state, gfp_mask, wait);
 813 }
 814
 815 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 816 {
 817         __nfs4_close(state, fmode, GFP_NOFS, 0);
 818 }
 819
 820 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 821 {
 822         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 823 }
 824
 825 /*
 826  * Search the state->lock_states for an existing lock_owner
 827  * that is compatible with current->files
 828  */
 829 static struct nfs4_lock_state *
 830 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 831 {
 832         struct nfs4_lock_state *pos;
 833         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 834                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 835                         continue;
 836                 switch (pos->ls_owner.lo_type) {
 837                 case NFS4_POSIX_LOCK_TYPE:
 838                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 839                                 continue;
 840                         break;
 841                 case NFS4_FLOCK_LOCK_TYPE:
 842                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 843                                 continue;
 844                 }
 845                 atomic_inc(&pos->ls_count);
 846                 return pos;
 847         }
 848         return NULL;
 849 }
 850
 851 /*
 852  * Return a compatible lock_state. If no initialized lock_state structure
 853  * exists, return an uninitialized one.
 854  *
 855  */
 856 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 857 {
 858         struct nfs4_lock_state *lsp;
 859         struct nfs_server *server = state->owner->so_server;
 860
 861         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 862         if (lsp == NULL)
 863                 return NULL;
 864         nfs4_init_seqid_counter(&lsp->ls_seqid);
 865         atomic_set(&lsp->ls_count, 1);
 866         lsp->ls_state = state;
 867         lsp->ls_owner.lo_type = type;
 868         switch (lsp->ls_owner.lo_type) {
 869         case NFS4_FLOCK_LOCK_TYPE:
 870                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 871                 break;
 872         case NFS4_POSIX_LOCK_TYPE:
 873                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 874                 break;
 875         default:
 876                 goto out_free;
 877         }
 878         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 879         if (lsp->ls_seqid.owner_id < 0)
 880                 goto out_free;
 881         INIT_LIST_HEAD(&lsp->ls_locks);
 882         return lsp;
 883 out_free:
 884         kfree(lsp);
 885         return NULL;
 886 }
 887
 888 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 889 {
 890         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 891         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 892         kfree(lsp);
 893 }
 894
 895 /*
 896  * Return a compatible lock_state. If no initialized lock_state structure
 897  * exists, return an uninitialized one.
 898  *
 899  */
 900 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 901 {
 902         struct nfs4_lock_state *lsp, *new = NULL;
 903
 904         for(;;) {
 905                 spin_lock(&state->state_lock);
 906                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 907                 if (lsp != NULL)
 908                         break;
 909                 if (new != NULL) {
 910                         list_add(&new->ls_locks, &state->lock_states);
 911                         set_bit(LK_STATE_IN_USE, &state->flags);
 912                         lsp = new;
 913                         new = NULL;
 914                         break;
 915                 }
 916                 spin_unlock(&state->state_lock);
 917                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 918                 if (new == NULL)
 919                         return NULL;
 920         }
 921         spin_unlock(&state->state_lock);
 922         if (new != NULL)
 923                 nfs4_free_lock_state(state->owner->so_server, new);
 924         return lsp;
 925 }
 926
 927 /*
 928  * Release reference to lock_state, and free it if we see that
 929  * it is no longer in use
 930  */
 931 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 932 {
 933         struct nfs4_state *state;
 934
 935         if (lsp == NULL)
 936                 return;
 937         state = lsp->ls_state;
 938         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 939                 return;
 940         list_del(&lsp->ls_locks);
 941         if (list_empty(&state->lock_states))
 942                 clear_bit(LK_STATE_IN_USE, &state->flags);
 943         spin_unlock(&state->state_lock);
 944         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 945                 if (nfs4_release_lockowner(lsp) == 0)
 946                         return;
 947         }
 948         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 949 }
 950
 951 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 952 {
 953         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 954
 955         dst->fl_u.nfs4_fl.owner = lsp;
 956         atomic_inc(&lsp->ls_count);
 957 }
 958
 959 static void nfs4_fl_release_lock(struct file_lock *fl)
 960 {
 961         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 962 }
 963
 964 static const struct file_lock_operations nfs4_fl_lock_ops = {
 965         .fl_copy_lock = nfs4_fl_copy_lock,
 966         .fl_release_private = nfs4_fl_release_lock,
 967 };
 968
 969 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 970 {
 971         struct nfs4_lock_state *lsp;
 972
 973         if (fl->fl_ops != NULL)
 974                 return 0;
 975         if (fl->fl_flags & FL_POSIX)
 976                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 977         else if (fl->fl_flags & FL_FLOCK)
 978                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 979                                 NFS4_FLOCK_LOCK_TYPE);
 980         else
 981                 return -EINVAL;
 982         if (lsp == NULL)
 983                 return -ENOMEM;
 984         fl->fl_u.nfs4_fl.owner = lsp;
 985         fl->fl_ops = &nfs4_fl_lock_ops;
 986         return 0;
 987 }
 988
 989 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 990                 const struct nfs_lockowner *lockowner)
 991 {
 992         struct nfs4_lock_state *lsp;
 993         fl_owner_t fl_owner;
 994         pid_t fl_pid;
 995         bool ret = false;
 996
 997
 998         if (lockowner == NULL)
 999                 goto out;
1000
1001         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1002                 goto out;
1003
1004         fl_owner = lockowner->l_owner;
1005         fl_pid = lockowner->l_pid;
1006         spin_lock(&state->state_lock);
1007         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1008         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1009                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1010                 ret = true;
1011         }
1012         spin_unlock(&state->state_lock);
1013         nfs4_put_lock_state(lsp);
1014 out:
1015         return ret;
1016 }
1017
1018 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1019 {
1020         int seq;
1021
1022         do {
1023                 seq = read_seqbegin(&state->seqlock);
1024                 nfs4_stateid_copy(dst, &state->stateid);
1025         } while (read_seqretry(&state->seqlock, seq));
1026 }
1027
1028 /*
1029  * Byte-range lock aware utility to initialize the stateid of read/write
1030  * requests.
1031  */
1032 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1033                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1034 {
1035         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1036                 return;
1037         if (nfs4_copy_lock_stateid(dst, state, lockowner))
1038                 return;
1039         nfs4_copy_open_stateid(dst, state);
1040 }
1041
1042 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1043 {
1044         struct nfs_seqid *new;
1045
1046         new = kmalloc(sizeof(*new), gfp_mask);
1047         if (new != NULL) {
1048                 new->sequence = counter;
1049                 INIT_LIST_HEAD(&new->list);
1050                 new->task = NULL;
1051         }
1052         return new;
1053 }
1054
1055 void nfs_release_seqid(struct nfs_seqid *seqid)
1056 {
1057         struct nfs_seqid_counter *sequence;
1058
1059         if (list_empty(&seqid->list))
1060                 return;
1061         sequence = seqid->sequence;
1062         spin_lock(&sequence->lock);
1063         list_del_init(&seqid->list);
1064         if (!list_empty(&sequence->list)) {
1065                 struct nfs_seqid *next;
1066
1067                 next = list_first_entry(&sequence->list,
1068                                 struct nfs_seqid, list);
1069                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1070         }
1071         spin_unlock(&sequence->lock);
1072 }
1073
1074 void nfs_free_seqid(struct nfs_seqid *seqid)
1075 {
1076         nfs_release_seqid(seqid);
1077         kfree(seqid);
1078 }
1079
1080 /*
1081  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1082  * failed with a seqid incrementing error -
1083  * see comments nfs_fs.h:seqid_mutating_error()
1084  */
1085 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1086 {
1087         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
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 /*
1211  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1212  * @clp: client to process
1213  *
1214  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1215  * resend of the SETCLIENTID and hence re-establish the
1216  * callback channel. Then return all existing delegations.
1217  */
1218 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1219 {
1220         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1221         nfs_expire_all_delegations(clp);
1222         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1223                         clp->cl_hostname);
1224 }
1225
1226 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1227 {
1228         nfs40_handle_cb_pathdown(clp);
1229         nfs4_schedule_state_manager(clp);
1230 }
1231
1232 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1233 {
1234
1235         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1236         /* Don't recover state that expired before the reboot */
1237         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1238                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1239                 return 0;
1240         }
1241         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1242         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1243         return 1;
1244 }
1245
1246 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1247 {
1248         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1249         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1250         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1251         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1252         return 1;
1253 }
1254
1255 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1256 {
1257         struct nfs_client *clp = server->nfs_client;
1258
1259         nfs4_state_mark_reclaim_nograce(clp, state);
1260         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1261                         clp->cl_hostname);
1262         nfs4_schedule_state_manager(clp);
1263 }
1264 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1265
1266 void nfs_inode_find_state_and_recover(struct inode *inode,
1267                 const nfs4_stateid *stateid)
1268 {
1269         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1270         struct nfs_inode *nfsi = NFS_I(inode);
1271         struct nfs_open_context *ctx;
1272         struct nfs4_state *state;
1273         bool found = false;
1274
1275         spin_lock(&inode->i_lock);
1276         list_for_each_entry(ctx, &nfsi->open_files, list) {
1277                 state = ctx->state;
1278                 if (state == NULL)
1279                         continue;
1280                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1281                         continue;
1282                 if (!nfs4_stateid_match(&state->stateid, stateid))
1283                         continue;
1284                 nfs4_state_mark_reclaim_nograce(clp, state);
1285                 found = true;
1286         }
1287         spin_unlock(&inode->i_lock);
1288         if (found)
1289                 nfs4_schedule_state_manager(clp);
1290 }
1291
1292
1293 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1294 {
1295         struct inode *inode = state->inode;
1296         struct nfs_inode *nfsi = NFS_I(inode);
1297         struct file_lock *fl;
1298         int status = 0;
1299
1300         if (inode->i_flock == NULL)
1301                 return 0;
1302
1303         /* Guard against delegation returns and new lock/unlock calls */
1304         down_write(&nfsi->rwsem);
1305         /* Protect inode->i_flock using the BKL */
1306         lock_flocks();
1307         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1308                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1309                         continue;
1310                 if (nfs_file_open_context(fl->fl_file)->state != state)
1311                         continue;
1312                 unlock_flocks();
1313                 status = ops->recover_lock(state, fl);
1314                 switch (status) {
1315                         case 0:
1316                                 break;
1317                         case -ESTALE:
1318                         case -NFS4ERR_ADMIN_REVOKED:
1319                         case -NFS4ERR_STALE_STATEID:
1320                         case -NFS4ERR_BAD_STATEID:
1321                         case -NFS4ERR_EXPIRED:
1322                         case -NFS4ERR_NO_GRACE:
1323                         case -NFS4ERR_STALE_CLIENTID:
1324                         case -NFS4ERR_BADSESSION:
1325                         case -NFS4ERR_BADSLOT:
1326                         case -NFS4ERR_BAD_HIGH_SLOT:
1327                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1328                                 goto out;
1329                         default:
1330                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1331                                         "Zeroing state\n", __func__, status);
1332                         case -ENOMEM:
1333                         case -NFS4ERR_DENIED:
1334                         case -NFS4ERR_RECLAIM_BAD:
1335                         case -NFS4ERR_RECLAIM_CONFLICT:
1336                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1337                                 status = 0;
1338                 }
1339                 lock_flocks();
1340         }
1341         unlock_flocks();
1342 out:
1343         up_write(&nfsi->rwsem);
1344         return status;
1345 }
1346
1347 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1348 {
1349         struct nfs4_state *state;
1350         struct nfs4_lock_state *lock;
1351         int status = 0;
1352
1353         /* Note: we rely on the sp->so_states list being ordered
1354          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1355          * states first.
1356          * This is needed to ensure that the server won't give us any
1357          * read delegations that we have to return if, say, we are
1358          * recovering after a network partition or a reboot from a
1359          * server that doesn't support a grace period.
1360          */
1361 restart:
1362         spin_lock(&sp->so_lock);
1363         list_for_each_entry(state, &sp->so_states, open_states) {
1364                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1365                         continue;
1366                 if (state->state == 0)
1367                         continue;
1368                 atomic_inc(&state->count);
1369                 spin_unlock(&sp->so_lock);
1370                 status = ops->recover_open(sp, state);
1371                 if (status >= 0) {
1372                         status = nfs4_reclaim_locks(state, ops);
1373                         if (status >= 0) {
1374                                 spin_lock(&state->state_lock);
1375                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1376                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1377                                                 pr_warn_ratelimited("NFS: "
1378                                                         "%s: Lock reclaim "
1379                                                         "failed!\n", __func__);
1380                                 }
1381                                 spin_unlock(&state->state_lock);
1382                                 nfs4_put_open_state(state);
1383                                 goto restart;
1384                         }
1385                 }
1386                 switch (status) {
1387                         default:
1388                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1389                                         "Zeroing state\n", __func__, status);
1390                         case -ENOENT:
1391                         case -ENOMEM:
1392                         case -ESTALE:
1393                                 /*
1394                                  * Open state on this file cannot be recovered
1395                                  * All we can do is revert to using the zero stateid.
1396                                  */
1397                                 memset(&state->stateid, 0,
1398                                         sizeof(state->stateid));
1399                                 /* Mark the file as being 'closed' */
1400                                 state->state = 0;
1401                                 break;
1402                         case -EKEYEXPIRED:
1403                                 /*
1404                                  * User RPCSEC_GSS context has expired.
1405                                  * We cannot recover this stateid now, so
1406                                  * skip it and allow recovery thread to
1407                                  * proceed.
1408                                  */
1409                                 break;
1410                         case -NFS4ERR_ADMIN_REVOKED:
1411                         case -NFS4ERR_STALE_STATEID:
1412                         case -NFS4ERR_BAD_STATEID:
1413                         case -NFS4ERR_RECLAIM_BAD:
1414                         case -NFS4ERR_RECLAIM_CONFLICT:
1415                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1416                                 break;
1417                         case -NFS4ERR_EXPIRED:
1418                         case -NFS4ERR_NO_GRACE:
1419                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1420                         case -NFS4ERR_STALE_CLIENTID:
1421                         case -NFS4ERR_BADSESSION:
1422                         case -NFS4ERR_BADSLOT:
1423                         case -NFS4ERR_BAD_HIGH_SLOT:
1424                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1425                                 goto out_err;
1426                 }
1427                 nfs4_put_open_state(state);
1428                 goto restart;
1429         }
1430         spin_unlock(&sp->so_lock);
1431         return 0;
1432 out_err:
1433         nfs4_put_open_state(state);
1434         return status;
1435 }
1436
1437 static void nfs4_clear_open_state(struct nfs4_state *state)
1438 {
1439         struct nfs4_lock_state *lock;
1440
1441         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1442         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1443         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1444         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1445         spin_lock(&state->state_lock);
1446         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1447                 lock->ls_seqid.flags = 0;
1448                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1449         }
1450         spin_unlock(&state->state_lock);
1451 }
1452
1453 static void nfs4_reset_seqids(struct nfs_server *server,
1454         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1455 {
1456         struct nfs_client *clp = server->nfs_client;
1457         struct nfs4_state_owner *sp;
1458         struct rb_node *pos;
1459         struct nfs4_state *state;
1460
1461         spin_lock(&clp->cl_lock);
1462         for (pos = rb_first(&server->state_owners);
1463              pos != NULL;
1464              pos = rb_next(pos)) {
1465                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1466                 sp->so_seqid.flags = 0;
1467                 spin_lock(&sp->so_lock);
1468                 list_for_each_entry(state, &sp->so_states, open_states) {
1469                         if (mark_reclaim(clp, state))
1470                                 nfs4_clear_open_state(state);
1471                 }
1472                 spin_unlock(&sp->so_lock);
1473         }
1474         spin_unlock(&clp->cl_lock);
1475 }
1476
1477 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1478         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1479 {
1480         struct nfs_server *server;
1481
1482         rcu_read_lock();
1483         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1484                 nfs4_reset_seqids(server, mark_reclaim);
1485         rcu_read_unlock();
1486 }
1487
1488 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1489 {
1490         /* Mark all delegations for reclaim */
1491         nfs_delegation_mark_reclaim(clp);
1492         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1493 }
1494
1495 static void nfs4_reclaim_complete(struct nfs_client *clp,
1496                                  const struct nfs4_state_recovery_ops *ops)
1497 {
1498         /* Notify the server we're done reclaiming our state */
1499         if (ops->reclaim_complete)
1500                 (void)ops->reclaim_complete(clp);
1501 }
1502
1503 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1504 {
1505         struct nfs_client *clp = server->nfs_client;
1506         struct nfs4_state_owner *sp;
1507         struct rb_node *pos;
1508         struct nfs4_state *state;
1509
1510         spin_lock(&clp->cl_lock);
1511         for (pos = rb_first(&server->state_owners);
1512              pos != NULL;
1513              pos = rb_next(pos)) {
1514                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1515                 spin_lock(&sp->so_lock);
1516                 list_for_each_entry(state, &sp->so_states, open_states) {
1517                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1518                                                 &state->flags))
1519                                 continue;
1520                         nfs4_state_mark_reclaim_nograce(clp, state);
1521                 }
1522                 spin_unlock(&sp->so_lock);
1523         }
1524         spin_unlock(&clp->cl_lock);
1525 }
1526
1527 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1528 {
1529         struct nfs_server *server;
1530
1531         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1532                 return 0;
1533
1534         rcu_read_lock();
1535         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1536                 nfs4_clear_reclaim_server(server);
1537         rcu_read_unlock();
1538
1539         nfs_delegation_reap_unclaimed(clp);
1540         return 1;
1541 }
1542
1543 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1544 {
1545         if (!nfs4_state_clear_reclaim_reboot(clp))
1546                 return;
1547         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1548 }
1549
1550 static void nfs_delegation_clear_all(struct nfs_client *clp)
1551 {
1552         nfs_delegation_mark_reclaim(clp);
1553         nfs_delegation_reap_unclaimed(clp);
1554 }
1555
1556 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1557 {
1558         nfs_delegation_clear_all(clp);
1559         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1560 }
1561
1562 static void nfs4_warn_keyexpired(const char *s)
1563 {
1564         printk_ratelimited(KERN_WARNING "Error: state manager"
1565                         " encountered RPCSEC_GSS session"
1566                         " expired against NFSv4 server %s.\n",
1567                         s);
1568 }
1569
1570 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1571 {
1572         switch (error) {
1573                 case 0:
1574                         break;
1575                 case -NFS4ERR_CB_PATH_DOWN:
1576                         nfs40_handle_cb_pathdown(clp);
1577                         break;
1578                 case -NFS4ERR_NO_GRACE:
1579                         nfs4_state_end_reclaim_reboot(clp);
1580                         break;
1581                 case -NFS4ERR_STALE_CLIENTID:
1582                 case -NFS4ERR_LEASE_MOVED:
1583                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1584                         nfs4_state_clear_reclaim_reboot(clp);
1585                         nfs4_state_start_reclaim_reboot(clp);
1586                         break;
1587                 case -NFS4ERR_EXPIRED:
1588                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1589                         nfs4_state_start_reclaim_nograce(clp);
1590                         break;
1591                 case -NFS4ERR_BADSESSION:
1592                 case -NFS4ERR_BADSLOT:
1593                 case -NFS4ERR_BAD_HIGH_SLOT:
1594                 case -NFS4ERR_DEADSESSION:
1595                 case -NFS4ERR_SEQ_FALSE_RETRY:
1596                 case -NFS4ERR_SEQ_MISORDERED:
1597                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1598                         /* Zero session reset errors */
1599                         break;
1600                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1601                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1602                         break;
1603                 case -EKEYEXPIRED:
1604                         /* Nothing we can do */
1605                         nfs4_warn_keyexpired(clp->cl_hostname);
1606                         break;
1607                 default:
1608                         dprintk("%s: failed to handle error %d for server %s\n",
1609                                         __func__, error, clp->cl_hostname);
1610                         return error;
1611         }
1612         dprintk("%s: handled error %d for server %s\n", __func__, error,
1613                         clp->cl_hostname);
1614         return 0;
1615 }
1616
1617 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1618 {
1619         struct nfs4_state_owner *sp;
1620         struct nfs_server *server;
1621         struct rb_node *pos;
1622         int status = 0;
1623
1624 restart:
1625         rcu_read_lock();
1626         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1627                 nfs4_purge_state_owners(server);
1628                 spin_lock(&clp->cl_lock);
1629                 for (pos = rb_first(&server->state_owners);
1630                      pos != NULL;
1631                      pos = rb_next(pos)) {
1632                         sp = rb_entry(pos,
1633                                 struct nfs4_state_owner, so_server_node);
1634                         if (!test_and_clear_bit(ops->owner_flag_bit,
1635                                                         &sp->so_flags))
1636                                 continue;
1637                         atomic_inc(&sp->so_count);
1638                         spin_unlock(&clp->cl_lock);
1639                         rcu_read_unlock();
1640
1641                         status = nfs4_reclaim_open_state(sp, ops);
1642                         if (status < 0) {
1643                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1644                                 nfs4_put_state_owner(sp);
1645                                 return nfs4_recovery_handle_error(clp, status);
1646                         }
1647
1648                         nfs4_put_state_owner(sp);
1649                         goto restart;
1650                 }
1651                 spin_unlock(&clp->cl_lock);
1652         }
1653         rcu_read_unlock();
1654         return status;
1655 }
1656
1657 static int nfs4_check_lease(struct nfs_client *clp)
1658 {
1659         struct rpc_cred *cred;
1660         const struct nfs4_state_maintenance_ops *ops =
1661                 clp->cl_mvops->state_renewal_ops;
1662         int status;
1663
1664         /* Is the client already known to have an expired lease? */
1665         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1666                 return 0;
1667         spin_lock(&clp->cl_lock);
1668         cred = ops->get_state_renewal_cred_locked(clp);
1669         spin_unlock(&clp->cl_lock);
1670         if (cred == NULL) {
1671                 cred = nfs4_get_setclientid_cred(clp);
1672                 status = -ENOKEY;
1673                 if (cred == NULL)
1674                         goto out;
1675         }
1676         status = ops->renew_lease(clp, cred);
1677         put_rpccred(cred);
1678 out:
1679         return nfs4_recovery_handle_error(clp, status);
1680 }
1681
1682 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1683  * on EXCHANGE_ID for v4.1
1684  */
1685 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1686 {
1687         switch (status) {
1688         case -NFS4ERR_SEQ_MISORDERED:
1689                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1690                         return -ESERVERFAULT;
1691                 /* Lease confirmation error: retry after purging the lease */
1692                 ssleep(1);
1693         case -NFS4ERR_STALE_CLIENTID:
1694                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1695                 break;
1696         case -NFS4ERR_CLID_INUSE:
1697                 pr_err("NFS: Server %s reports our clientid is in use\n",
1698                         clp->cl_hostname);
1699                 nfs_mark_client_ready(clp, -EPERM);
1700                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1701                 return -EPERM;
1702         case -EACCES:
1703                 if (clp->cl_machine_cred == NULL)
1704                         return -EACCES;
1705                 /* Handle case where the user hasn't set up machine creds */
1706                 nfs4_clear_machine_cred(clp);
1707         case -NFS4ERR_DELAY:
1708         case -ETIMEDOUT:
1709         case -EAGAIN:
1710                 ssleep(1);
1711                 break;
1712
1713         case -NFS4ERR_MINOR_VERS_MISMATCH:
1714                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1715                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1716                 dprintk("%s: exit with error %d for server %s\n",
1717                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1718                 return -EPROTONOSUPPORT;
1719         case -EKEYEXPIRED:
1720                 nfs4_warn_keyexpired(clp->cl_hostname);
1721         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1722                                  * in nfs4_exchange_id */
1723         default:
1724                 dprintk("%s: exit with error %d for server %s\n", __func__,
1725                                 status, clp->cl_hostname);
1726                 return status;
1727         }
1728         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1729         dprintk("%s: handled error %d for server %s\n", __func__, status,
1730                         clp->cl_hostname);
1731         return 0;
1732 }
1733
1734 static int nfs4_establish_lease(struct nfs_client *clp)
1735 {
1736         struct rpc_cred *cred;
1737         const struct nfs4_state_recovery_ops *ops =
1738                 clp->cl_mvops->reboot_recovery_ops;
1739         int status;
1740
1741         cred = ops->get_clid_cred(clp);
1742         if (cred == NULL)
1743                 return -ENOENT;
1744         status = ops->establish_clid(clp, cred);
1745         put_rpccred(cred);
1746         if (status != 0)
1747                 return status;
1748         pnfs_destroy_all_layouts(clp);
1749         return 0;
1750 }
1751
1752 /*
1753  * Returns zero or a negative errno.  NFS4ERR values are converted
1754  * to local errno values.
1755  */
1756 static int nfs4_reclaim_lease(struct nfs_client *clp)
1757 {
1758         int status;
1759
1760         status = nfs4_establish_lease(clp);
1761         if (status < 0)
1762                 return nfs4_handle_reclaim_lease_error(clp, status);
1763         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1764                 nfs4_state_start_reclaim_nograce(clp);
1765         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1766                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1767         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1768         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1769         return 0;
1770 }
1771
1772 static int nfs4_purge_lease(struct nfs_client *clp)
1773 {
1774         int status;
1775
1776         status = nfs4_establish_lease(clp);
1777         if (status < 0)
1778                 return nfs4_handle_reclaim_lease_error(clp, status);
1779         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1780         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1781         nfs4_state_start_reclaim_nograce(clp);
1782         return 0;
1783 }
1784
1785 /**
1786  * nfs4_discover_server_trunking - Detect server IP address trunking
1787  *
1788  * @clp: nfs_client under test
1789  * @result: OUT: found nfs_client, or clp
1790  *
1791  * Returns zero or a negative errno.  If zero is returned,
1792  * an nfs_client pointer is planted in "result".
1793  *
1794  * Note: since we are invoked in process context, and
1795  * not from inside the state manager, we cannot use
1796  * nfs4_handle_reclaim_lease_error().
1797  */
1798 int nfs4_discover_server_trunking(struct nfs_client *clp,
1799                                   struct nfs_client **result)
1800 {
1801         const struct nfs4_state_recovery_ops *ops =
1802                                 clp->cl_mvops->reboot_recovery_ops;
1803         rpc_authflavor_t *flavors, flav, save;
1804         struct rpc_clnt *clnt;
1805         struct rpc_cred *cred;
1806         int i, len, status;
1807
1808         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1809
1810         len = NFS_MAX_SECFLAVORS;
1811         flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1812         if (flavors == NULL) {
1813                 status = -ENOMEM;
1814                 goto out;
1815         }
1816         len = rpcauth_list_flavors(flavors, len);
1817         if (len < 0) {
1818                 status = len;
1819                 goto out_free;
1820         }
1821         clnt = clp->cl_rpcclient;
1822         save = clnt->cl_auth->au_flavor;
1823         i = 0;
1824
1825         mutex_lock(&nfs_clid_init_mutex);
1826         status  = -ENOENT;
1827 again:
1828         cred = ops->get_clid_cred(clp);
1829         if (cred == NULL)
1830                 goto out_unlock;
1831
1832         status = ops->detect_trunking(clp, result, cred);
1833         put_rpccred(cred);
1834         switch (status) {
1835         case 0:
1836                 break;
1837
1838         case -EACCES:
1839                 if (clp->cl_machine_cred == NULL)
1840                         break;
1841                 /* Handle case where the user hasn't set up machine creds */
1842                 nfs4_clear_machine_cred(clp);
1843         case -NFS4ERR_DELAY:
1844         case -ETIMEDOUT:
1845         case -EAGAIN:
1846                 ssleep(1);
1847                 dprintk("NFS: %s after status %d, retrying\n",
1848                         __func__, status);
1849                 goto again;
1850
1851         case -NFS4ERR_CLID_INUSE:
1852         case -NFS4ERR_WRONGSEC:
1853                 status = -EPERM;
1854                 if (i >= len)
1855                         break;
1856
1857                 flav = flavors[i++];
1858                 if (flav == save)
1859                         flav = flavors[i++];
1860                 clnt = rpc_clone_client_set_auth(clnt, flav);
1861                 if (IS_ERR(clnt)) {
1862                         status = PTR_ERR(clnt);
1863                         break;
1864                 }
1865                 clp->cl_rpcclient = clnt;
1866                 goto again;
1867
1868         case -NFS4ERR_MINOR_VERS_MISMATCH:
1869                 status = -EPROTONOSUPPORT;
1870                 break;
1871
1872         case -EKEYEXPIRED:
1873                 nfs4_warn_keyexpired(clp->cl_hostname);
1874         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1875                                  * in nfs4_exchange_id */
1876                 status = -EKEYEXPIRED;
1877         }
1878
1879 out_unlock:
1880         mutex_unlock(&nfs_clid_init_mutex);
1881 out_free:
1882         kfree(flavors);
1883 out:
1884         dprintk("NFS: %s: status = %d\n", __func__, status);
1885         return status;
1886 }
1887
1888 #ifdef CONFIG_NFS_V4_1
1889 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1890 {
1891         struct nfs_client *clp = session->clp;
1892
1893         switch (err) {
1894         default:
1895                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1896                 break;
1897         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1898                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1899         }
1900         nfs4_schedule_lease_recovery(clp);
1901 }
1902 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1903
1904 void nfs41_handle_recall_slot(struct nfs_client *clp)
1905 {
1906         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1907         dprintk("%s: scheduling slot recall for server %s\n", __func__,
1908                         clp->cl_hostname);
1909         nfs4_schedule_state_manager(clp);
1910 }
1911
1912 static void nfs4_reset_all_state(struct nfs_client *clp)
1913 {
1914         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1915                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1916                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1917                 nfs4_state_start_reclaim_nograce(clp);
1918                 dprintk("%s: scheduling reset of all state for server %s!\n",
1919                                 __func__, clp->cl_hostname);
1920                 nfs4_schedule_state_manager(clp);
1921         }
1922 }
1923
1924 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1925 {
1926         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1927                 nfs4_state_start_reclaim_reboot(clp);
1928                 dprintk("%s: server %s rebooted!\n", __func__,
1929                                 clp->cl_hostname);
1930                 nfs4_schedule_state_manager(clp);
1931         }
1932 }
1933
1934 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1935 {
1936         nfs4_reset_all_state(clp);
1937         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1938 }
1939
1940 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1941 {
1942         /* This will need to handle layouts too */
1943         nfs_expire_all_delegations(clp);
1944         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1945                         clp->cl_hostname);
1946 }
1947
1948 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1949 {
1950         nfs_expire_all_delegations(clp);
1951         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1952                 nfs4_schedule_state_manager(clp);
1953         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1954                         clp->cl_hostname);
1955 }
1956
1957 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1958 {
1959         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1960                 &clp->cl_state) == 0)
1961                 nfs4_schedule_state_manager(clp);
1962 }
1963
1964 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1965 {
1966         if (!flags)
1967                 return;
1968
1969         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1970                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
1971
1972         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1973                 nfs41_handle_server_reboot(clp);
1974         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1975                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1976                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1977                             SEQ4_STATUS_LEASE_MOVED))
1978                 nfs41_handle_state_revoked(clp);
1979         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1980                 nfs41_handle_recallable_state_revoked(clp);
1981         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1982                 nfs41_handle_backchannel_fault(clp);
1983         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1984                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1985                 nfs41_handle_cb_path_down(clp);
1986 }
1987
1988 static int nfs4_reset_session(struct nfs_client *clp)
1989 {
1990         struct rpc_cred *cred;
1991         int status;
1992
1993         if (!nfs4_has_session(clp))
1994                 return 0;
1995         nfs4_begin_drain_session(clp);
1996         cred = nfs4_get_exchange_id_cred(clp);
1997         status = nfs4_proc_destroy_session(clp->cl_session, cred);
1998         if (status && status != -NFS4ERR_BADSESSION &&
1999             status != -NFS4ERR_DEADSESSION) {
2000                 status = nfs4_recovery_handle_error(clp, status);
2001                 goto out;
2002         }
2003
2004         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2005         status = nfs4_proc_create_session(clp, cred);
2006         if (status) {
2007                 dprintk("%s: session reset failed with status %d for server %s!\n",
2008                         __func__, status, clp->cl_hostname);
2009                 status = nfs4_handle_reclaim_lease_error(clp, status);
2010                 goto out;
2011         }
2012         nfs41_finish_session_reset(clp);
2013         dprintk("%s: session reset was successful for server %s!\n",
2014                         __func__, clp->cl_hostname);
2015 out:
2016         if (cred)
2017                 put_rpccred(cred);
2018         return status;
2019 }
2020
2021 static int nfs4_recall_slot(struct nfs_client *clp)
2022 {
2023         struct nfs4_slot_table *fc_tbl;
2024         struct nfs4_slot *new, *old;
2025         int i;
2026
2027         if (!nfs4_has_session(clp))
2028                 return 0;
2029         nfs4_begin_drain_session(clp);
2030         fc_tbl = &clp->cl_session->fc_slot_table;
2031         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
2032                       GFP_NOFS);
2033         if (!new)
2034                 return -ENOMEM;
2035
2036         spin_lock(&fc_tbl->slot_tbl_lock);
2037         for (i = 0; i < fc_tbl->target_max_slots; i++)
2038                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
2039         old = fc_tbl->slots;
2040         fc_tbl->slots = new;
2041         fc_tbl->max_slots = fc_tbl->target_max_slots;
2042         fc_tbl->target_max_slots = 0;
2043         clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
2044         spin_unlock(&fc_tbl->slot_tbl_lock);
2045
2046         kfree(old);
2047         return 0;
2048 }
2049
2050 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2051 {
2052         struct rpc_cred *cred;
2053         int ret;
2054
2055         if (!nfs4_has_session(clp))
2056                 return 0;
2057         nfs4_begin_drain_session(clp);
2058         cred = nfs4_get_exchange_id_cred(clp);
2059         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2060         if (cred)
2061                 put_rpccred(cred);
2062         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2063         switch (ret) {
2064         case 0:
2065                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2066                         __func__, clp->cl_hostname);
2067                 break;
2068         case -NFS4ERR_DELAY:
2069                 ssleep(1);
2070                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2071                 break;
2072         default:
2073                 return nfs4_recovery_handle_error(clp, ret);
2074         }
2075         return 0;
2076 }
2077 #else /* CONFIG_NFS_V4_1 */
2078 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2079 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2080 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
2081
2082 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2083 {
2084         return 0;
2085 }
2086 #endif /* CONFIG_NFS_V4_1 */
2087
2088 static void nfs4_state_manager(struct nfs_client *clp)
2089 {
2090         int status = 0;
2091         const char *section = "", *section_sep = "";
2092
2093         /* Ensure exclusive access to NFSv4 state */
2094         do {
2095                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2096                         section = "purge state";
2097                         status = nfs4_purge_lease(clp);
2098                         if (status < 0)
2099                                 goto out_error;
2100                         continue;
2101                 }
2102
2103                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2104                         section = "lease expired";
2105                         /* We're going to have to re-establish a clientid */
2106                         status = nfs4_reclaim_lease(clp);
2107                         if (status < 0)
2108                                 goto out_error;
2109                         continue;
2110                 }
2111
2112                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2113                         section = "check lease";
2114                         status = nfs4_check_lease(clp);
2115                         if (status < 0)
2116                                 goto out_error;
2117                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2118                                 continue;
2119                 }
2120
2121                 /* Initialize or reset the session */
2122                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2123                         section = "reset session";
2124                         status = nfs4_reset_session(clp);
2125                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2126                                 continue;
2127                         if (status < 0)
2128                                 goto out_error;
2129                 }
2130
2131                 /* Send BIND_CONN_TO_SESSION */
2132                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2133                                 &clp->cl_state)) {
2134                         section = "bind conn to session";
2135                         status = nfs4_bind_conn_to_session(clp);
2136                         if (status < 0)
2137                                 goto out_error;
2138                         continue;
2139                 }
2140
2141                 /* Recall session slots */
2142                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)) {
2143                         section = "recall slot";
2144                         status = nfs4_recall_slot(clp);
2145                         if (status < 0)
2146                                 goto out_error;
2147                         continue;
2148                 }
2149
2150                 /* First recover reboot state... */
2151                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2152                         section = "reclaim reboot";
2153                         status = nfs4_do_reclaim(clp,
2154                                 clp->cl_mvops->reboot_recovery_ops);
2155                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2156                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2157                                 continue;
2158                         nfs4_state_end_reclaim_reboot(clp);
2159                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2160                                 continue;
2161                         if (status < 0)
2162                                 goto out_error;
2163                 }
2164
2165                 /* Now recover expired state... */
2166                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2167                         section = "reclaim nograce";
2168                         status = nfs4_do_reclaim(clp,
2169                                 clp->cl_mvops->nograce_recovery_ops);
2170                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2171                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2172                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2173                                 continue;
2174                         if (status < 0)
2175                                 goto out_error;
2176                 }
2177
2178                 nfs4_end_drain_session(clp);
2179                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2180                         nfs_client_return_marked_delegations(clp);
2181                         continue;
2182                 }
2183
2184                 nfs4_clear_state_manager_bit(clp);
2185                 /* Did we race with an attempt to give us more work? */
2186                 if (clp->cl_state == 0)
2187                         break;
2188                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2189                         break;
2190         } while (atomic_read(&clp->cl_count) > 1);
2191         return;
2192 out_error:
2193         if (strlen(section))
2194                 section_sep = ": ";
2195         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2196                         " with error %d\n", section_sep, section,
2197                         clp->cl_hostname, -status);
2198         ssleep(1);
2199         nfs4_end_drain_session(clp);
2200         nfs4_clear_state_manager_bit(clp);
2201 }
2202
2203 static int nfs4_run_state_manager(void *ptr)
2204 {
2205         struct nfs_client *clp = ptr;
2206
2207         allow_signal(SIGKILL);
2208         nfs4_state_manager(clp);
2209         nfs_put_client(clp);
2210         module_put_and_exit(0);
2211         return 0;
2212 }
2213
2214 /*
2215  * Local variables:
2216  *  c-basic-offset: 8
2217  * End:
2218  */