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         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         switch (status) {
 140         case -NFS4ERR_STALE_CLIENTID:
 141                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 142         case 0:
 143                 /* Sustain the lease, even if it's empty.  If the clientid4
 144                  * goes stale it's of no use for trunking discovery. */
 145                 nfs4_schedule_state_renewal(*result);
 146                 break;
 147         }
 148
 149 out:
 150         return status;
 151 }
 152
 153 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 154 {
 155         struct rpc_cred *cred = NULL;
 156
 157         if (clp->cl_machine_cred != NULL)
 158                 cred = get_rpccred(clp->cl_machine_cred);
 159         return cred;
 160 }
 161
 162 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 163 {
 164         struct rpc_cred *cred;
 165
 166         spin_lock(&clp->cl_lock);
 167         cred = clp->cl_machine_cred;
 168         clp->cl_machine_cred = NULL;
 169         spin_unlock(&clp->cl_lock);
 170         if (cred != NULL)
 171                 put_rpccred(cred);
 172 }
 173
 174 static struct rpc_cred *
 175 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 176 {
 177         struct rpc_cred *cred = NULL;
 178         struct nfs4_state_owner *sp;
 179         struct rb_node *pos;
 180
 181         for (pos = rb_first(&server->state_owners);
 182              pos != NULL;
 183              pos = rb_next(pos)) {
 184                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 185                 if (list_empty(&sp->so_states))
 186                         continue;
 187                 cred = get_rpccred(sp->so_cred);
 188                 break;
 189         }
 190         return cred;
 191 }
 192
 193 /**
 194  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 195  * @clp: client state handle
 196  *
 197  * Returns an rpc_cred with reference count bumped, or NULL.
 198  * Caller must hold clp->cl_lock.
 199  */
 200 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 201 {
 202         struct rpc_cred *cred = NULL;
 203         struct nfs_server *server;
 204
 205         /* Use machine credentials if available */
 206         cred = nfs4_get_machine_cred_locked(clp);
 207         if (cred != NULL)
 208                 goto out;
 209
 210         rcu_read_lock();
 211         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 212                 cred = nfs4_get_renew_cred_server_locked(server);
 213                 if (cred != NULL)
 214                         break;
 215         }
 216         rcu_read_unlock();
 217
 218 out:
 219         return cred;
 220 }
 221
 222 #if defined(CONFIG_NFS_V4_1)
 223
 224 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 225 {
 226         int status;
 227         struct nfs_fsinfo fsinfo;
 228
 229         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 230                 nfs4_schedule_state_renewal(clp);
 231                 return 0;
 232         }
 233
 234         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 235         if (status == 0) {
 236                 /* Update lease time and schedule renewal */
 237                 spin_lock(&clp->cl_lock);
 238                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 239                 clp->cl_last_renewal = jiffies;
 240                 spin_unlock(&clp->cl_lock);
 241
 242                 nfs4_schedule_state_renewal(clp);
 243         }
 244
 245         return status;
 246 }
 247
 248 /*
 249  * Back channel returns NFS4ERR_DELAY for new requests when
 250  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 251  * is ended.
 252  */
 253 static void nfs4_end_drain_session(struct nfs_client *clp)
 254 {
 255         struct nfs4_session *ses = clp->cl_session;
 256         struct nfs4_slot_table *tbl;
 257         int max_slots;
 258
 259         if (ses == NULL)
 260                 return;
 261         tbl = &ses->fc_slot_table;
 262         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 263                 spin_lock(&tbl->slot_tbl_lock);
 264                 max_slots = tbl->max_slots;
 265                 while (max_slots--) {
 266                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 267                                                 nfs4_set_task_privileged,
 268                                                 NULL) == NULL)
 269                                 break;
 270                 }
 271                 spin_unlock(&tbl->slot_tbl_lock);
 272         }
 273 }
 274
 275 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 276 {
 277         spin_lock(&tbl->slot_tbl_lock);
 278         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 279                 INIT_COMPLETION(tbl->complete);
 280                 spin_unlock(&tbl->slot_tbl_lock);
 281                 return wait_for_completion_interruptible(&tbl->complete);
 282         }
 283         spin_unlock(&tbl->slot_tbl_lock);
 284         return 0;
 285 }
 286
 287 static int nfs4_begin_drain_session(struct nfs_client *clp)
 288 {
 289         struct nfs4_session *ses = clp->cl_session;
 290         int ret = 0;
 291
 292         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 293         /* back channel */
 294         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 295         if (ret)
 296                 return ret;
 297         /* fore channel */
 298         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 299 }
 300
 301 static void nfs41_finish_session_reset(struct nfs_client *clp)
 302 {
 303         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 304         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 305         /* create_session negotiated new slot table */
 306         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
 307         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 308         nfs41_setup_state_renewal(clp);
 309 }
 310
 311 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 312 {
 313         int status;
 314
 315         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 316                 goto do_confirm;
 317         nfs4_begin_drain_session(clp);
 318         status = nfs4_proc_exchange_id(clp, cred);
 319         if (status != 0)
 320                 goto out;
 321         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 322 do_confirm:
 323         status = nfs4_proc_create_session(clp, cred);
 324         if (status != 0)
 325                 goto out;
 326         nfs41_finish_session_reset(clp);
 327         nfs_mark_client_ready(clp, NFS_CS_READY);
 328 out:
 329         return status;
 330 }
 331
 332 /**
 333  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 334  *
 335  * @clp: nfs_client under test
 336  * @result: OUT: found nfs_client, or clp
 337  * @cred: credential to use for trunking test
 338  *
 339  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 340  * If NFS4_OK is returned, an nfs_client pointer is planted in
 341  * "result".
 342  *
 343  * Note: The returned client may not yet be marked ready.
 344  */
 345 int nfs41_discover_server_trunking(struct nfs_client *clp,
 346                                    struct nfs_client **result,
 347                                    struct rpc_cred *cred)
 348 {
 349         int status;
 350
 351         status = nfs4_proc_exchange_id(clp, cred);
 352         if (status != NFS4_OK)
 353                 return status;
 354         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 355
 356         return nfs41_walk_client_list(clp, result, cred);
 357 }
 358
 359 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 360 {
 361         struct rpc_cred *cred;
 362
 363         spin_lock(&clp->cl_lock);
 364         cred = nfs4_get_machine_cred_locked(clp);
 365         spin_unlock(&clp->cl_lock);
 366         return cred;
 367 }
 368
 369 #endif /* CONFIG_NFS_V4_1 */
 370
 371 static struct rpc_cred *
 372 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 373 {
 374         struct nfs_client *clp = server->nfs_client;
 375         struct rpc_cred *cred = NULL;
 376         struct nfs4_state_owner *sp;
 377         struct rb_node *pos;
 378
 379         spin_lock(&clp->cl_lock);
 380         pos = rb_first(&server->state_owners);
 381         if (pos != NULL) {
 382                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 383                 cred = get_rpccred(sp->so_cred);
 384         }
 385         spin_unlock(&clp->cl_lock);
 386         return cred;
 387 }
 388
 389 /**
 390  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 391  * @clp: client state handle
 392  *
 393  * Returns an rpc_cred with reference count bumped, or NULL.
 394  */
 395 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 396 {
 397         struct nfs_server *server;
 398         struct rpc_cred *cred;
 399
 400         spin_lock(&clp->cl_lock);
 401         cred = nfs4_get_machine_cred_locked(clp);
 402         spin_unlock(&clp->cl_lock);
 403         if (cred != NULL)
 404                 goto out;
 405
 406         rcu_read_lock();
 407         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 408                 cred = nfs4_get_setclientid_cred_server(server);
 409                 if (cred != NULL)
 410                         break;
 411         }
 412         rcu_read_unlock();
 413
 414 out:
 415         return cred;
 416 }
 417
 418 static struct nfs4_state_owner *
 419 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 420 {
 421         struct rb_node **p = &server->state_owners.rb_node,
 422                        *parent = NULL;
 423         struct nfs4_state_owner *sp;
 424
 425         while (*p != NULL) {
 426                 parent = *p;
 427                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 428
 429                 if (cred < sp->so_cred)
 430                         p = &parent->rb_left;
 431                 else if (cred > sp->so_cred)
 432                         p = &parent->rb_right;
 433                 else {
 434                         if (!list_empty(&sp->so_lru))
 435                                 list_del_init(&sp->so_lru);
 436                         atomic_inc(&sp->so_count);
 437                         return sp;
 438                 }
 439         }
 440         return NULL;
 441 }
 442
 443 static struct nfs4_state_owner *
 444 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 445 {
 446         struct nfs_server *server = new->so_server;
 447         struct rb_node **p = &server->state_owners.rb_node,
 448                        *parent = NULL;
 449         struct nfs4_state_owner *sp;
 450         int err;
 451
 452         while (*p != NULL) {
 453                 parent = *p;
 454                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 455
 456                 if (new->so_cred < sp->so_cred)
 457                         p = &parent->rb_left;
 458                 else if (new->so_cred > sp->so_cred)
 459                         p = &parent->rb_right;
 460                 else {
 461                         if (!list_empty(&sp->so_lru))
 462                                 list_del_init(&sp->so_lru);
 463                         atomic_inc(&sp->so_count);
 464                         return sp;
 465                 }
 466         }
 467         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 468         if (err)
 469                 return ERR_PTR(err);
 470         rb_link_node(&new->so_server_node, parent, p);
 471         rb_insert_color(&new->so_server_node, &server->state_owners);
 472         return new;
 473 }
 474
 475 static void
 476 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 477 {
 478         struct nfs_server *server = sp->so_server;
 479
 480         if (!RB_EMPTY_NODE(&sp->so_server_node))
 481                 rb_erase(&sp->so_server_node, &server->state_owners);
 482         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 483 }
 484
 485 static void
 486 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 487 {
 488         sc->create_time = ktime_get();
 489         sc->flags = 0;
 490         sc->counter = 0;
 491         spin_lock_init(&sc->lock);
 492         INIT_LIST_HEAD(&sc->list);
 493         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 494 }
 495
 496 static void
 497 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 498 {
 499         rpc_destroy_wait_queue(&sc->wait);
 500 }
 501
 502 /*
 503  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 504  * create a new state_owner.
 505  *
 506  */
 507 static struct nfs4_state_owner *
 508 nfs4_alloc_state_owner(struct nfs_server *server,
 509                 struct rpc_cred *cred,
 510                 gfp_t gfp_flags)
 511 {
 512         struct nfs4_state_owner *sp;
 513
 514         sp = kzalloc(sizeof(*sp), gfp_flags);
 515         if (!sp)
 516                 return NULL;
 517         sp->so_server = server;
 518         sp->so_cred = get_rpccred(cred);
 519         spin_lock_init(&sp->so_lock);
 520         INIT_LIST_HEAD(&sp->so_states);
 521         nfs4_init_seqid_counter(&sp->so_seqid);
 522         atomic_set(&sp->so_count, 1);
 523         INIT_LIST_HEAD(&sp->so_lru);
 524         return sp;
 525 }
 526
 527 static void
 528 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 529 {
 530         struct rb_node *rb_node = &sp->so_server_node;
 531
 532         if (!RB_EMPTY_NODE(rb_node)) {
 533                 struct nfs_server *server = sp->so_server;
 534                 struct nfs_client *clp = server->nfs_client;
 535
 536                 spin_lock(&clp->cl_lock);
 537                 if (!RB_EMPTY_NODE(rb_node)) {
 538                         rb_erase(rb_node, &server->state_owners);
 539                         RB_CLEAR_NODE(rb_node);
 540                 }
 541                 spin_unlock(&clp->cl_lock);
 542         }
 543 }
 544
 545 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 546 {
 547         nfs4_destroy_seqid_counter(&sp->so_seqid);
 548         put_rpccred(sp->so_cred);
 549         kfree(sp);
 550 }
 551
 552 static void nfs4_gc_state_owners(struct nfs_server *server)
 553 {
 554         struct nfs_client *clp = server->nfs_client;
 555         struct nfs4_state_owner *sp, *tmp;
 556         unsigned long time_min, time_max;
 557         LIST_HEAD(doomed);
 558
 559         spin_lock(&clp->cl_lock);
 560         time_max = jiffies;
 561         time_min = (long)time_max - (long)clp->cl_lease_time;
 562         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 563                 /* NB: LRU is sorted so that oldest is at the head */
 564                 if (time_in_range(sp->so_expires, time_min, time_max))
 565                         break;
 566                 list_move(&sp->so_lru, &doomed);
 567                 nfs4_remove_state_owner_locked(sp);
 568         }
 569         spin_unlock(&clp->cl_lock);
 570
 571         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 572                 list_del(&sp->so_lru);
 573                 nfs4_free_state_owner(sp);
 574         }
 575 }
 576
 577 /**
 578  * nfs4_get_state_owner - Look up a state owner given a credential
 579  * @server: nfs_server to search
 580  * @cred: RPC credential to match
 581  *
 582  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 583  */
 584 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 585                                               struct rpc_cred *cred,
 586                                               gfp_t gfp_flags)
 587 {
 588         struct nfs_client *clp = server->nfs_client;
 589         struct nfs4_state_owner *sp, *new;
 590
 591         spin_lock(&clp->cl_lock);
 592         sp = nfs4_find_state_owner_locked(server, cred);
 593         spin_unlock(&clp->cl_lock);
 594         if (sp != NULL)
 595                 goto out;
 596         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 597         if (new == NULL)
 598                 goto out;
 599         do {
 600                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 601                         break;
 602                 spin_lock(&clp->cl_lock);
 603                 sp = nfs4_insert_state_owner_locked(new);
 604                 spin_unlock(&clp->cl_lock);
 605         } while (sp == ERR_PTR(-EAGAIN));
 606         if (sp != new)
 607                 nfs4_free_state_owner(new);
 608 out:
 609         nfs4_gc_state_owners(server);
 610         return sp;
 611 }
 612
 613 /**
 614  * nfs4_put_state_owner - Release a nfs4_state_owner
 615  * @sp: state owner data to release
 616  *
 617  * Note that we keep released state owners on an LRU
 618  * list.
 619  * This caches valid state owners so that they can be
 620  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 621  * It also pins the uniquifier of dropped state owners for
 622  * a while, to ensure that those state owner names are
 623  * never reused.
 624  */
 625 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 626 {
 627         struct nfs_server *server = sp->so_server;
 628         struct nfs_client *clp = server->nfs_client;
 629
 630         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 631                 return;
 632
 633         sp->so_expires = jiffies;
 634         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 635         spin_unlock(&clp->cl_lock);
 636 }
 637
 638 /**
 639  * nfs4_purge_state_owners - Release all cached state owners
 640  * @server: nfs_server with cached state owners to release
 641  *
 642  * Called at umount time.  Remaining state owners will be on
 643  * the LRU with ref count of zero.
 644  */
 645 void nfs4_purge_state_owners(struct nfs_server *server)
 646 {
 647         struct nfs_client *clp = server->nfs_client;
 648         struct nfs4_state_owner *sp, *tmp;
 649         LIST_HEAD(doomed);
 650
 651         spin_lock(&clp->cl_lock);
 652         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 653                 list_move(&sp->so_lru, &doomed);
 654                 nfs4_remove_state_owner_locked(sp);
 655         }
 656         spin_unlock(&clp->cl_lock);
 657
 658         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 659                 list_del(&sp->so_lru);
 660                 nfs4_free_state_owner(sp);
 661         }
 662 }
 663
 664 static struct nfs4_state *
 665 nfs4_alloc_open_state(void)
 666 {
 667         struct nfs4_state *state;
 668
 669         state = kzalloc(sizeof(*state), GFP_NOFS);
 670         if (!state)
 671                 return NULL;
 672         atomic_set(&state->count, 1);
 673         INIT_LIST_HEAD(&state->lock_states);
 674         spin_lock_init(&state->state_lock);
 675         seqlock_init(&state->seqlock);
 676         return state;
 677 }
 678
 679 void
 680 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 681 {
 682         if (state->state == fmode)
 683                 return;
 684         /* NB! List reordering - see the reclaim code for why.  */
 685         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 686                 if (fmode & FMODE_WRITE)
 687                         list_move(&state->open_states, &state->owner->so_states);
 688                 else
 689                         list_move_tail(&state->open_states, &state->owner->so_states);
 690         }
 691         state->state = fmode;
 692 }
 693
 694 static struct nfs4_state *
 695 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 696 {
 697         struct nfs_inode *nfsi = NFS_I(inode);
 698         struct nfs4_state *state;
 699
 700         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 701                 if (state->owner != owner)
 702                         continue;
 703                 if (atomic_inc_not_zero(&state->count))
 704                         return state;
 705         }
 706         return NULL;
 707 }
 708
 709 static void
 710 nfs4_free_open_state(struct nfs4_state *state)
 711 {
 712         kfree(state);
 713 }
 714
 715 struct nfs4_state *
 716 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 717 {
 718         struct nfs4_state *state, *new;
 719         struct nfs_inode *nfsi = NFS_I(inode);
 720
 721         spin_lock(&inode->i_lock);
 722         state = __nfs4_find_state_byowner(inode, owner);
 723         spin_unlock(&inode->i_lock);
 724         if (state)
 725                 goto out;
 726         new = nfs4_alloc_open_state();
 727         spin_lock(&owner->so_lock);
 728         spin_lock(&inode->i_lock);
 729         state = __nfs4_find_state_byowner(inode, owner);
 730         if (state == NULL && new != NULL) {
 731                 state = new;
 732                 state->owner = owner;
 733                 atomic_inc(&owner->so_count);
 734                 list_add(&state->inode_states, &nfsi->open_states);
 735                 ihold(inode);
 736                 state->inode = inode;
 737                 spin_unlock(&inode->i_lock);
 738                 /* Note: The reclaim code dictates that we add stateless
 739                  * and read-only stateids to the end of the list */
 740                 list_add_tail(&state->open_states, &owner->so_states);
 741                 spin_unlock(&owner->so_lock);
 742         } else {
 743                 spin_unlock(&inode->i_lock);
 744                 spin_unlock(&owner->so_lock);
 745                 if (new)
 746                         nfs4_free_open_state(new);
 747         }
 748 out:
 749         return state;
 750 }
 751
 752 void nfs4_put_open_state(struct nfs4_state *state)
 753 {
 754         struct inode *inode = state->inode;
 755         struct nfs4_state_owner *owner = state->owner;
 756
 757         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 758                 return;
 759         spin_lock(&inode->i_lock);
 760         list_del(&state->inode_states);
 761         list_del(&state->open_states);
 762         spin_unlock(&inode->i_lock);
 763         spin_unlock(&owner->so_lock);
 764         iput(inode);
 765         nfs4_free_open_state(state);
 766         nfs4_put_state_owner(owner);
 767 }
 768
 769 /*
 770  * Close the current file.
 771  */
 772 static void __nfs4_close(struct nfs4_state *state,
 773                 fmode_t fmode, gfp_t gfp_mask, int wait)
 774 {
 775         struct nfs4_state_owner *owner = state->owner;
 776         int call_close = 0;
 777         fmode_t newstate;
 778
 779         atomic_inc(&owner->so_count);
 780         /* Protect against nfs4_find_state() */
 781         spin_lock(&owner->so_lock);
 782         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 783                 case FMODE_READ:
 784                         state->n_rdonly--;
 785                         break;
 786                 case FMODE_WRITE:
 787                         state->n_wronly--;
 788                         break;
 789                 case FMODE_READ|FMODE_WRITE:
 790                         state->n_rdwr--;
 791         }
 792         newstate = FMODE_READ|FMODE_WRITE;
 793         if (state->n_rdwr == 0) {
 794                 if (state->n_rdonly == 0) {
 795                         newstate &= ~FMODE_READ;
 796                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 797                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 798                 }
 799                 if (state->n_wronly == 0) {
 800                         newstate &= ~FMODE_WRITE;
 801                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 802                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 803                 }
 804                 if (newstate == 0)
 805                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 806         }
 807         nfs4_state_set_mode_locked(state, newstate);
 808         spin_unlock(&owner->so_lock);
 809
 810         if (!call_close) {
 811                 nfs4_put_open_state(state);
 812                 nfs4_put_state_owner(owner);
 813         } else
 814                 nfs4_do_close(state, gfp_mask, wait);
 815 }
 816
 817 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 818 {
 819         __nfs4_close(state, fmode, GFP_NOFS, 0);
 820 }
 821
 822 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 823 {
 824         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 825 }
 826
 827 /*
 828  * Search the state->lock_states for an existing lock_owner
 829  * that is compatible with current->files
 830  */
 831 static struct nfs4_lock_state *
 832 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 833 {
 834         struct nfs4_lock_state *pos;
 835         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 836                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 837                         continue;
 838                 switch (pos->ls_owner.lo_type) {
 839                 case NFS4_POSIX_LOCK_TYPE:
 840                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 841                                 continue;
 842                         break;
 843                 case NFS4_FLOCK_LOCK_TYPE:
 844                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 845                                 continue;
 846                 }
 847                 atomic_inc(&pos->ls_count);
 848                 return pos;
 849         }
 850         return NULL;
 851 }
 852
 853 /*
 854  * Return a compatible lock_state. If no initialized lock_state structure
 855  * exists, return an uninitialized one.
 856  *
 857  */
 858 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)
 859 {
 860         struct nfs4_lock_state *lsp;
 861         struct nfs_server *server = state->owner->so_server;
 862
 863         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 864         if (lsp == NULL)
 865                 return NULL;
 866         nfs4_init_seqid_counter(&lsp->ls_seqid);
 867         atomic_set(&lsp->ls_count, 1);
 868         lsp->ls_state = state;
 869         lsp->ls_owner.lo_type = type;
 870         switch (lsp->ls_owner.lo_type) {
 871         case NFS4_FLOCK_LOCK_TYPE:
 872                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 873                 break;
 874         case NFS4_POSIX_LOCK_TYPE:
 875                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 876                 break;
 877         default:
 878                 goto out_free;
 879         }
 880         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 881         if (lsp->ls_seqid.owner_id < 0)
 882                 goto out_free;
 883         INIT_LIST_HEAD(&lsp->ls_locks);
 884         return lsp;
 885 out_free:
 886         kfree(lsp);
 887         return NULL;
 888 }
 889
 890 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 891 {
 892         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 893         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 894         kfree(lsp);
 895 }
 896
 897 /*
 898  * Return a compatible lock_state. If no initialized lock_state structure
 899  * exists, return an uninitialized one.
 900  *
 901  */
 902 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 903 {
 904         struct nfs4_lock_state *lsp, *new = NULL;
 905
 906         for(;;) {
 907                 spin_lock(&state->state_lock);
 908                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 909                 if (lsp != NULL)
 910                         break;
 911                 if (new != NULL) {
 912                         list_add(&new->ls_locks, &state->lock_states);
 913                         set_bit(LK_STATE_IN_USE, &state->flags);
 914                         lsp = new;
 915                         new = NULL;
 916                         break;
 917                 }
 918                 spin_unlock(&state->state_lock);
 919                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 920                 if (new == NULL)
 921                         return NULL;
 922         }
 923         spin_unlock(&state->state_lock);
 924         if (new != NULL)
 925                 nfs4_free_lock_state(state->owner->so_server, new);
 926         return lsp;
 927 }
 928
 929 /*
 930  * Release reference to lock_state, and free it if we see that
 931  * it is no longer in use
 932  */
 933 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 934 {
 935         struct nfs4_state *state;
 936
 937         if (lsp == NULL)
 938                 return;
 939         state = lsp->ls_state;
 940         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 941                 return;
 942         list_del(&lsp->ls_locks);
 943         if (list_empty(&state->lock_states))
 944                 clear_bit(LK_STATE_IN_USE, &state->flags);
 945         spin_unlock(&state->state_lock);
 946         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 947                 if (nfs4_release_lockowner(lsp) == 0)
 948                         return;
 949         }
 950         nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp);
 951 }
 952
 953 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 954 {
 955         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 956
 957         dst->fl_u.nfs4_fl.owner = lsp;
 958         atomic_inc(&lsp->ls_count);
 959 }
 960
 961 static void nfs4_fl_release_lock(struct file_lock *fl)
 962 {
 963         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 964 }
 965
 966 static const struct file_lock_operations nfs4_fl_lock_ops = {
 967         .fl_copy_lock = nfs4_fl_copy_lock,
 968         .fl_release_private = nfs4_fl_release_lock,
 969 };
 970
 971 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 972 {
 973         struct nfs4_lock_state *lsp;
 974
 975         if (fl->fl_ops != NULL)
 976                 return 0;
 977         if (fl->fl_flags & FL_POSIX)
 978                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 979         else if (fl->fl_flags & FL_FLOCK)
 980                 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid,
 981                                 NFS4_FLOCK_LOCK_TYPE);
 982         else
 983                 return -EINVAL;
 984         if (lsp == NULL)
 985                 return -ENOMEM;
 986         fl->fl_u.nfs4_fl.owner = lsp;
 987         fl->fl_ops = &nfs4_fl_lock_ops;
 988         return 0;
 989 }
 990
 991 static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state,
 992                 const struct nfs_lockowner *lockowner)
 993 {
 994         struct nfs4_lock_state *lsp;
 995         fl_owner_t fl_owner;
 996         pid_t fl_pid;
 997         bool ret = false;
 998
 999
1000         if (lockowner == NULL)
1001                 goto out;
1002
1003         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1004                 goto out;
1005
1006         fl_owner = lockowner->l_owner;
1007         fl_pid = lockowner->l_pid;
1008         spin_lock(&state->state_lock);
1009         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
1010         if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1011                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1012                 ret = true;
1013         }
1014         spin_unlock(&state->state_lock);
1015         nfs4_put_lock_state(lsp);
1016 out:
1017         return ret;
1018 }
1019
1020 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1021 {
1022         int seq;
1023
1024         do {
1025                 seq = read_seqbegin(&state->seqlock);
1026                 nfs4_stateid_copy(dst, &state->stateid);
1027         } while (read_seqretry(&state->seqlock, seq));
1028 }
1029
1030 /*
1031  * Byte-range lock aware utility to initialize the stateid of read/write
1032  * requests.
1033  */
1034 void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state,
1035                 fmode_t fmode, const struct nfs_lockowner *lockowner)
1036 {
1037         if (nfs4_copy_delegation_stateid(dst, state->inode, fmode))
1038                 return;
1039         if (nfs4_copy_lock_stateid(dst, state, lockowner))
1040                 return;
1041         nfs4_copy_open_stateid(dst, state);
1042 }
1043
1044 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1045 {
1046         struct nfs_seqid *new;
1047
1048         new = kmalloc(sizeof(*new), gfp_mask);
1049         if (new != NULL) {
1050                 new->sequence = counter;
1051                 INIT_LIST_HEAD(&new->list);
1052                 new->task = NULL;
1053         }
1054         return new;
1055 }
1056
1057 void nfs_release_seqid(struct nfs_seqid *seqid)
1058 {
1059         struct nfs_seqid_counter *sequence;
1060
1061         if (list_empty(&seqid->list))
1062                 return;
1063         sequence = seqid->sequence;
1064         spin_lock(&sequence->lock);
1065         list_del_init(&seqid->list);
1066         if (!list_empty(&sequence->list)) {
1067                 struct nfs_seqid *next;
1068
1069                 next = list_first_entry(&sequence->list,
1070                                 struct nfs_seqid, list);
1071                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1072         }
1073         spin_unlock(&sequence->lock);
1074 }
1075
1076 void nfs_free_seqid(struct nfs_seqid *seqid)
1077 {
1078         nfs_release_seqid(seqid);
1079         kfree(seqid);
1080 }
1081
1082 /*
1083  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1084  * failed with a seqid incrementing error -
1085  * see comments nfs_fs.h:seqid_mutating_error()
1086  */
1087 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1088 {
1089         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
1090         switch (status) {
1091                 case 0:
1092                         break;
1093                 case -NFS4ERR_BAD_SEQID:
1094                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1095                                 return;
1096                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1097                                         " sequence-id error on an"
1098                                         " unconfirmed sequence %p!\n",
1099                                         seqid->sequence);
1100                 case -NFS4ERR_STALE_CLIENTID:
1101                 case -NFS4ERR_STALE_STATEID:
1102                 case -NFS4ERR_BAD_STATEID:
1103                 case -NFS4ERR_BADXDR:
1104                 case -NFS4ERR_RESOURCE:
1105                 case -NFS4ERR_NOFILEHANDLE:
1106                         /* Non-seqid mutating errors */
1107                         return;
1108         };
1109         /*
1110          * Note: no locking needed as we are guaranteed to be first
1111          * on the sequence list
1112          */
1113         seqid->sequence->counter++;
1114 }
1115
1116 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1117 {
1118         struct nfs4_state_owner *sp = container_of(seqid->sequence,
1119                                         struct nfs4_state_owner, so_seqid);
1120         struct nfs_server *server = sp->so_server;
1121
1122         if (status == -NFS4ERR_BAD_SEQID)
1123                 nfs4_drop_state_owner(sp);
1124         if (!nfs4_has_session(server->nfs_client))
1125                 nfs_increment_seqid(status, seqid);
1126 }
1127
1128 /*
1129  * Increment the seqid if the LOCK/LOCKU succeeded, or
1130  * failed with a seqid incrementing error -
1131  * see comments nfs_fs.h:seqid_mutating_error()
1132  */
1133 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1134 {
1135         nfs_increment_seqid(status, seqid);
1136 }
1137
1138 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1139 {
1140         struct nfs_seqid_counter *sequence = seqid->sequence;
1141         int status = 0;
1142
1143         spin_lock(&sequence->lock);
1144         seqid->task = task;
1145         if (list_empty(&seqid->list))
1146                 list_add_tail(&seqid->list, &sequence->list);
1147         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1148                 goto unlock;
1149         rpc_sleep_on(&sequence->wait, task, NULL);
1150         status = -EAGAIN;
1151 unlock:
1152         spin_unlock(&sequence->lock);
1153         return status;
1154 }
1155
1156 static int nfs4_run_state_manager(void *);
1157
1158 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1159 {
1160         smp_mb__before_clear_bit();
1161         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1162         smp_mb__after_clear_bit();
1163         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1164         rpc_wake_up(&clp->cl_rpcwaitq);
1165 }
1166
1167 /*
1168  * Schedule the nfs_client asynchronous state management routine
1169  */
1170 void nfs4_schedule_state_manager(struct nfs_client *clp)
1171 {
1172         struct task_struct *task;
1173         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1174
1175         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1176                 return;
1177         __module_get(THIS_MODULE);
1178         atomic_inc(&clp->cl_count);
1179
1180         /* The rcu_read_lock() is not strictly necessary, as the state
1181          * manager is the only thread that ever changes the rpc_xprt
1182          * after it's initialized.  At this point, we're single threaded. */
1183         rcu_read_lock();
1184         snprintf(buf, sizeof(buf), "%s-manager",
1185                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1186         rcu_read_unlock();
1187         task = kthread_run(nfs4_run_state_manager, clp, buf);
1188         if (IS_ERR(task)) {
1189                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1190                         __func__, PTR_ERR(task));
1191                 nfs4_clear_state_manager_bit(clp);
1192                 nfs_put_client(clp);
1193                 module_put(THIS_MODULE);
1194         }
1195 }
1196
1197 /*
1198  * Schedule a lease recovery attempt
1199  */
1200 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1201 {
1202         if (!clp)
1203                 return;
1204         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1205                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1206         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1207                         clp->cl_hostname);
1208         nfs4_schedule_state_manager(clp);
1209 }
1210 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1211
1212 /*
1213  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1214  * @clp: client to process
1215  *
1216  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1217  * resend of the SETCLIENTID and hence re-establish the
1218  * callback channel. Then return all existing delegations.
1219  */
1220 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1221 {
1222         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1223         nfs_expire_all_delegations(clp);
1224         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1225                         clp->cl_hostname);
1226 }
1227
1228 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1229 {
1230         nfs40_handle_cb_pathdown(clp);
1231         nfs4_schedule_state_manager(clp);
1232 }
1233
1234 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1235 {
1236
1237         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1238         /* Don't recover state that expired before the reboot */
1239         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1240                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1241                 return 0;
1242         }
1243         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1244         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1245         return 1;
1246 }
1247
1248 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1249 {
1250         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1251         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1252         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1253         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1254         return 1;
1255 }
1256
1257 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1258 {
1259         struct nfs_client *clp = server->nfs_client;
1260
1261         nfs4_state_mark_reclaim_nograce(clp, state);
1262         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1263                         clp->cl_hostname);
1264         nfs4_schedule_state_manager(clp);
1265 }
1266 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1267
1268 void nfs_inode_find_state_and_recover(struct inode *inode,
1269                 const nfs4_stateid *stateid)
1270 {
1271         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1272         struct nfs_inode *nfsi = NFS_I(inode);
1273         struct nfs_open_context *ctx;
1274         struct nfs4_state *state;
1275         bool found = false;
1276
1277         spin_lock(&inode->i_lock);
1278         list_for_each_entry(ctx, &nfsi->open_files, list) {
1279                 state = ctx->state;
1280                 if (state == NULL)
1281                         continue;
1282                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
1283                         continue;
1284                 if (!nfs4_stateid_match(&state->stateid, stateid))
1285                         continue;
1286                 nfs4_state_mark_reclaim_nograce(clp, state);
1287                 found = true;
1288         }
1289         spin_unlock(&inode->i_lock);
1290         if (found)
1291                 nfs4_schedule_state_manager(clp);
1292 }
1293
1294
1295 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1296 {
1297         struct inode *inode = state->inode;
1298         struct nfs_inode *nfsi = NFS_I(inode);
1299         struct file_lock *fl;
1300         int status = 0;
1301
1302         if (inode->i_flock == NULL)
1303                 return 0;
1304
1305         /* Guard against delegation returns and new lock/unlock calls */
1306         down_write(&nfsi->rwsem);
1307         /* Protect inode->i_flock using the BKL */
1308         lock_flocks();
1309         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1310                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1311                         continue;
1312                 if (nfs_file_open_context(fl->fl_file)->state != state)
1313                         continue;
1314                 unlock_flocks();
1315                 status = ops->recover_lock(state, fl);
1316                 switch (status) {
1317                         case 0:
1318                                 break;
1319                         case -ESTALE:
1320                         case -NFS4ERR_ADMIN_REVOKED:
1321                         case -NFS4ERR_STALE_STATEID:
1322                         case -NFS4ERR_BAD_STATEID:
1323                         case -NFS4ERR_EXPIRED:
1324                         case -NFS4ERR_NO_GRACE:
1325                         case -NFS4ERR_STALE_CLIENTID:
1326                         case -NFS4ERR_BADSESSION:
1327                         case -NFS4ERR_BADSLOT:
1328                         case -NFS4ERR_BAD_HIGH_SLOT:
1329                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1330                                 goto out;
1331                         default:
1332                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1333                                         "Zeroing state\n", __func__, status);
1334                         case -ENOMEM:
1335                         case -NFS4ERR_DENIED:
1336                         case -NFS4ERR_RECLAIM_BAD:
1337                         case -NFS4ERR_RECLAIM_CONFLICT:
1338                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1339                                 status = 0;
1340                 }
1341                 lock_flocks();
1342         }
1343         unlock_flocks();
1344 out:
1345         up_write(&nfsi->rwsem);
1346         return status;
1347 }
1348
1349 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1350 {
1351         struct nfs4_state *state;
1352         struct nfs4_lock_state *lock;
1353         int status = 0;
1354
1355         /* Note: we rely on the sp->so_states list being ordered
1356          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1357          * states first.
1358          * This is needed to ensure that the server won't give us any
1359          * read delegations that we have to return if, say, we are
1360          * recovering after a network partition or a reboot from a
1361          * server that doesn't support a grace period.
1362          */
1363 restart:
1364         spin_lock(&sp->so_lock);
1365         list_for_each_entry(state, &sp->so_states, open_states) {
1366                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1367                         continue;
1368                 if (state->state == 0)
1369                         continue;
1370                 atomic_inc(&state->count);
1371                 spin_unlock(&sp->so_lock);
1372                 status = ops->recover_open(sp, state);
1373                 if (status >= 0) {
1374                         status = nfs4_reclaim_locks(state, ops);
1375                         if (status >= 0) {
1376                                 spin_lock(&state->state_lock);
1377                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1378                                         if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1379                                                 pr_warn_ratelimited("NFS: "
1380                                                         "%s: Lock reclaim "
1381                                                         "failed!\n", __func__);
1382                                 }
1383                                 spin_unlock(&state->state_lock);
1384                                 nfs4_put_open_state(state);
1385                                 goto restart;
1386                         }
1387                 }
1388                 switch (status) {
1389                         default:
1390                                 printk(KERN_ERR "NFS: %s: unhandled error %d. "
1391                                         "Zeroing state\n", __func__, status);
1392                         case -ENOENT:
1393                         case -ENOMEM:
1394                         case -ESTALE:
1395                                 /*
1396                                  * Open state on this file cannot be recovered
1397                                  * All we can do is revert to using the zero stateid.
1398                                  */
1399                                 memset(&state->stateid, 0,
1400                                         sizeof(state->stateid));
1401                                 /* Mark the file as being 'closed' */
1402                                 state->state = 0;
1403                                 break;
1404                         case -EKEYEXPIRED:
1405                                 /*
1406                                  * User RPCSEC_GSS context has expired.
1407                                  * We cannot recover this stateid now, so
1408                                  * skip it and allow recovery thread to
1409                                  * proceed.
1410                                  */
1411                                 break;
1412                         case -NFS4ERR_ADMIN_REVOKED:
1413                         case -NFS4ERR_STALE_STATEID:
1414                         case -NFS4ERR_BAD_STATEID:
1415                         case -NFS4ERR_RECLAIM_BAD:
1416                         case -NFS4ERR_RECLAIM_CONFLICT:
1417                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1418                                 break;
1419                         case -NFS4ERR_EXPIRED:
1420                         case -NFS4ERR_NO_GRACE:
1421                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1422                         case -NFS4ERR_STALE_CLIENTID:
1423                         case -NFS4ERR_BADSESSION:
1424                         case -NFS4ERR_BADSLOT:
1425                         case -NFS4ERR_BAD_HIGH_SLOT:
1426                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1427                                 goto out_err;
1428                 }
1429                 nfs4_put_open_state(state);
1430                 goto restart;
1431         }
1432         spin_unlock(&sp->so_lock);
1433         return 0;
1434 out_err:
1435         nfs4_put_open_state(state);
1436         return status;
1437 }
1438
1439 static void nfs4_clear_open_state(struct nfs4_state *state)
1440 {
1441         struct nfs4_lock_state *lock;
1442
1443         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1444         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1445         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1446         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1447         spin_lock(&state->state_lock);
1448         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1449                 lock->ls_seqid.flags = 0;
1450                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1451         }
1452         spin_unlock(&state->state_lock);
1453 }
1454
1455 static void nfs4_reset_seqids(struct nfs_server *server,
1456         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1457 {
1458         struct nfs_client *clp = server->nfs_client;
1459         struct nfs4_state_owner *sp;
1460         struct rb_node *pos;
1461         struct nfs4_state *state;
1462
1463         spin_lock(&clp->cl_lock);
1464         for (pos = rb_first(&server->state_owners);
1465              pos != NULL;
1466              pos = rb_next(pos)) {
1467                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1468                 sp->so_seqid.flags = 0;
1469                 spin_lock(&sp->so_lock);
1470                 list_for_each_entry(state, &sp->so_states, open_states) {
1471                         if (mark_reclaim(clp, state))
1472                                 nfs4_clear_open_state(state);
1473                 }
1474                 spin_unlock(&sp->so_lock);
1475         }
1476         spin_unlock(&clp->cl_lock);
1477 }
1478
1479 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1480         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1481 {
1482         struct nfs_server *server;
1483
1484         rcu_read_lock();
1485         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1486                 nfs4_reset_seqids(server, mark_reclaim);
1487         rcu_read_unlock();
1488 }
1489
1490 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1491 {
1492         /* Mark all delegations for reclaim */
1493         nfs_delegation_mark_reclaim(clp);
1494         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1495 }
1496
1497 static void nfs4_reclaim_complete(struct nfs_client *clp,
1498                                  const struct nfs4_state_recovery_ops *ops)
1499 {
1500         /* Notify the server we're done reclaiming our state */
1501         if (ops->reclaim_complete)
1502                 (void)ops->reclaim_complete(clp);
1503 }
1504
1505 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1506 {
1507         struct nfs_client *clp = server->nfs_client;
1508         struct nfs4_state_owner *sp;
1509         struct rb_node *pos;
1510         struct nfs4_state *state;
1511
1512         spin_lock(&clp->cl_lock);
1513         for (pos = rb_first(&server->state_owners);
1514              pos != NULL;
1515              pos = rb_next(pos)) {
1516                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1517                 spin_lock(&sp->so_lock);
1518                 list_for_each_entry(state, &sp->so_states, open_states) {
1519                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1520                                                 &state->flags))
1521                                 continue;
1522                         nfs4_state_mark_reclaim_nograce(clp, state);
1523                 }
1524                 spin_unlock(&sp->so_lock);
1525         }
1526         spin_unlock(&clp->cl_lock);
1527 }
1528
1529 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1530 {
1531         struct nfs_server *server;
1532
1533         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1534                 return 0;
1535
1536         rcu_read_lock();
1537         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1538                 nfs4_clear_reclaim_server(server);
1539         rcu_read_unlock();
1540
1541         nfs_delegation_reap_unclaimed(clp);
1542         return 1;
1543 }
1544
1545 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1546 {
1547         if (!nfs4_state_clear_reclaim_reboot(clp))
1548                 return;
1549         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1550 }
1551
1552 static void nfs_delegation_clear_all(struct nfs_client *clp)
1553 {
1554         nfs_delegation_mark_reclaim(clp);
1555         nfs_delegation_reap_unclaimed(clp);
1556 }
1557
1558 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1559 {
1560         nfs_delegation_clear_all(clp);
1561         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1562 }
1563
1564 static void nfs4_warn_keyexpired(const char *s)
1565 {
1566         printk_ratelimited(KERN_WARNING "Error: state manager"
1567                         " encountered RPCSEC_GSS session"
1568                         " expired against NFSv4 server %s.\n",
1569                         s);
1570 }
1571
1572 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1573 {
1574         switch (error) {
1575                 case 0:
1576                         break;
1577                 case -NFS4ERR_CB_PATH_DOWN:
1578                         nfs40_handle_cb_pathdown(clp);
1579                         break;
1580                 case -NFS4ERR_NO_GRACE:
1581                         nfs4_state_end_reclaim_reboot(clp);
1582                         break;
1583                 case -NFS4ERR_STALE_CLIENTID:
1584                 case -NFS4ERR_LEASE_MOVED:
1585                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1586                         nfs4_state_clear_reclaim_reboot(clp);
1587                         nfs4_state_start_reclaim_reboot(clp);
1588                         break;
1589                 case -NFS4ERR_EXPIRED:
1590                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1591                         nfs4_state_start_reclaim_nograce(clp);
1592                         break;
1593                 case -NFS4ERR_BADSESSION:
1594                 case -NFS4ERR_BADSLOT:
1595                 case -NFS4ERR_BAD_HIGH_SLOT:
1596                 case -NFS4ERR_DEADSESSION:
1597                 case -NFS4ERR_SEQ_FALSE_RETRY:
1598                 case -NFS4ERR_SEQ_MISORDERED:
1599                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1600                         /* Zero session reset errors */
1601                         break;
1602                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1603                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1604                         break;
1605                 case -EKEYEXPIRED:
1606                         /* Nothing we can do */
1607                         nfs4_warn_keyexpired(clp->cl_hostname);
1608                         break;
1609                 default:
1610                         dprintk("%s: failed to handle error %d for server %s\n",
1611                                         __func__, error, clp->cl_hostname);
1612                         return error;
1613         }
1614         dprintk("%s: handled error %d for server %s\n", __func__, error,
1615                         clp->cl_hostname);
1616         return 0;
1617 }
1618
1619 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1620 {
1621         struct nfs4_state_owner *sp;
1622         struct nfs_server *server;
1623         struct rb_node *pos;
1624         int status = 0;
1625
1626 restart:
1627         rcu_read_lock();
1628         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1629                 nfs4_purge_state_owners(server);
1630                 spin_lock(&clp->cl_lock);
1631                 for (pos = rb_first(&server->state_owners);
1632                      pos != NULL;
1633                      pos = rb_next(pos)) {
1634                         sp = rb_entry(pos,
1635                                 struct nfs4_state_owner, so_server_node);
1636                         if (!test_and_clear_bit(ops->owner_flag_bit,
1637                                                         &sp->so_flags))
1638                                 continue;
1639                         atomic_inc(&sp->so_count);
1640                         spin_unlock(&clp->cl_lock);
1641                         rcu_read_unlock();
1642
1643                         status = nfs4_reclaim_open_state(sp, ops);
1644                         if (status < 0) {
1645                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1646                                 nfs4_put_state_owner(sp);
1647                                 return nfs4_recovery_handle_error(clp, status);
1648                         }
1649
1650                         nfs4_put_state_owner(sp);
1651                         goto restart;
1652                 }
1653                 spin_unlock(&clp->cl_lock);
1654         }
1655         rcu_read_unlock();
1656         return status;
1657 }
1658
1659 static int nfs4_check_lease(struct nfs_client *clp)
1660 {
1661         struct rpc_cred *cred;
1662         const struct nfs4_state_maintenance_ops *ops =
1663                 clp->cl_mvops->state_renewal_ops;
1664         int status;
1665
1666         /* Is the client already known to have an expired lease? */
1667         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1668                 return 0;
1669         spin_lock(&clp->cl_lock);
1670         cred = ops->get_state_renewal_cred_locked(clp);
1671         spin_unlock(&clp->cl_lock);
1672         if (cred == NULL) {
1673                 cred = nfs4_get_setclientid_cred(clp);
1674                 status = -ENOKEY;
1675                 if (cred == NULL)
1676                         goto out;
1677         }
1678         status = ops->renew_lease(clp, cred);
1679         put_rpccred(cred);
1680 out:
1681         return nfs4_recovery_handle_error(clp, status);
1682 }
1683
1684 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1685  * and for recoverable errors on EXCHANGE_ID for v4.1
1686  */
1687 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1688 {
1689         switch (status) {
1690         case -NFS4ERR_SEQ_MISORDERED:
1691                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1692                         return -ESERVERFAULT;
1693                 /* Lease confirmation error: retry after purging the lease */
1694                 ssleep(1);
1695                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1696                 break;
1697         case -NFS4ERR_STALE_CLIENTID:
1698                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1699                 nfs4_state_clear_reclaim_reboot(clp);
1700                 nfs4_state_start_reclaim_reboot(clp);
1701                 break;
1702         case -NFS4ERR_CLID_INUSE:
1703                 pr_err("NFS: Server %s reports our clientid is in use\n",
1704                         clp->cl_hostname);
1705                 nfs_mark_client_ready(clp, -EPERM);
1706                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1707                 return -EPERM;
1708         case -EACCES:
1709                 if (clp->cl_machine_cred == NULL)
1710                         return -EACCES;
1711                 /* Handle case where the user hasn't set up machine creds */
1712                 nfs4_clear_machine_cred(clp);
1713         case -NFS4ERR_DELAY:
1714         case -ETIMEDOUT:
1715         case -EAGAIN:
1716                 ssleep(1);
1717                 break;
1718
1719         case -NFS4ERR_MINOR_VERS_MISMATCH:
1720                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1721                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1722                 dprintk("%s: exit with error %d for server %s\n",
1723                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1724                 return -EPROTONOSUPPORT;
1725         case -EKEYEXPIRED:
1726                 nfs4_warn_keyexpired(clp->cl_hostname);
1727         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1728                                  * in nfs4_exchange_id */
1729         default:
1730                 dprintk("%s: exit with error %d for server %s\n", __func__,
1731                                 status, clp->cl_hostname);
1732                 return status;
1733         }
1734         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1735         dprintk("%s: handled error %d for server %s\n", __func__, status,
1736                         clp->cl_hostname);
1737         return 0;
1738 }
1739
1740 static int nfs4_establish_lease(struct nfs_client *clp)
1741 {
1742         struct rpc_cred *cred;
1743         const struct nfs4_state_recovery_ops *ops =
1744                 clp->cl_mvops->reboot_recovery_ops;
1745         int status;
1746
1747         cred = ops->get_clid_cred(clp);
1748         if (cred == NULL)
1749                 return -ENOENT;
1750         status = ops->establish_clid(clp, cred);
1751         put_rpccred(cred);
1752         if (status != 0)
1753                 return status;
1754         pnfs_destroy_all_layouts(clp);
1755         return 0;
1756 }
1757
1758 /*
1759  * Returns zero or a negative errno.  NFS4ERR values are converted
1760  * to local errno values.
1761  */
1762 static int nfs4_reclaim_lease(struct nfs_client *clp)
1763 {
1764         int status;
1765
1766         status = nfs4_establish_lease(clp);
1767         if (status < 0)
1768                 return nfs4_handle_reclaim_lease_error(clp, status);
1769         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1770                 nfs4_state_start_reclaim_nograce(clp);
1771         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1772                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1773         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1774         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1775         return 0;
1776 }
1777
1778 static int nfs4_purge_lease(struct nfs_client *clp)
1779 {
1780         int status;
1781
1782         status = nfs4_establish_lease(clp);
1783         if (status < 0)
1784                 return nfs4_handle_reclaim_lease_error(clp, status);
1785         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1786         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1787         nfs4_state_start_reclaim_nograce(clp);
1788         return 0;
1789 }
1790
1791 /**
1792  * nfs4_discover_server_trunking - Detect server IP address trunking
1793  *
1794  * @clp: nfs_client under test
1795  * @result: OUT: found nfs_client, or clp
1796  *
1797  * Returns zero or a negative errno.  If zero is returned,
1798  * an nfs_client pointer is planted in "result".
1799  *
1800  * Note: since we are invoked in process context, and
1801  * not from inside the state manager, we cannot use
1802  * nfs4_handle_reclaim_lease_error().
1803  */
1804 int nfs4_discover_server_trunking(struct nfs_client *clp,
1805                                   struct nfs_client **result)
1806 {
1807         const struct nfs4_state_recovery_ops *ops =
1808                                 clp->cl_mvops->reboot_recovery_ops;
1809         rpc_authflavor_t *flavors, flav, save;
1810         struct rpc_clnt *clnt;
1811         struct rpc_cred *cred;
1812         int i, len, status;
1813
1814         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
1815
1816         len = NFS_MAX_SECFLAVORS;
1817         flavors = kcalloc(len, sizeof(*flavors), GFP_KERNEL);
1818         if (flavors == NULL) {
1819                 status = -ENOMEM;
1820                 goto out;
1821         }
1822         len = rpcauth_list_flavors(flavors, len);
1823         if (len < 0) {
1824                 status = len;
1825                 goto out_free;
1826         }
1827         clnt = clp->cl_rpcclient;
1828         save = clnt->cl_auth->au_flavor;
1829         i = 0;
1830
1831         mutex_lock(&nfs_clid_init_mutex);
1832         status  = -ENOENT;
1833 again:
1834         cred = ops->get_clid_cred(clp);
1835         if (cred == NULL)
1836                 goto out_unlock;
1837
1838         status = ops->detect_trunking(clp, result, cred);
1839         put_rpccred(cred);
1840         switch (status) {
1841         case 0:
1842                 break;
1843
1844         case -EACCES:
1845                 if (clp->cl_machine_cred == NULL)
1846                         break;
1847                 /* Handle case where the user hasn't set up machine creds */
1848                 nfs4_clear_machine_cred(clp);
1849         case -NFS4ERR_DELAY:
1850         case -ETIMEDOUT:
1851         case -EAGAIN:
1852                 ssleep(1);
1853                 dprintk("NFS: %s after status %d, retrying\n",
1854                         __func__, status);
1855                 goto again;
1856
1857         case -NFS4ERR_CLID_INUSE:
1858         case -NFS4ERR_WRONGSEC:
1859                 status = -EPERM;
1860                 if (i >= len)
1861                         break;
1862
1863                 flav = flavors[i++];
1864                 if (flav == save)
1865                         flav = flavors[i++];
1866                 clnt = rpc_clone_client_set_auth(clnt, flav);
1867                 if (IS_ERR(clnt)) {
1868                         status = PTR_ERR(clnt);
1869                         break;
1870                 }
1871                 clp->cl_rpcclient = clnt;
1872                 goto again;
1873
1874         case -NFS4ERR_MINOR_VERS_MISMATCH:
1875                 status = -EPROTONOSUPPORT;
1876                 break;
1877
1878         case -EKEYEXPIRED:
1879                 nfs4_warn_keyexpired(clp->cl_hostname);
1880         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1881                                  * in nfs4_exchange_id */
1882                 status = -EKEYEXPIRED;
1883         }
1884
1885 out_unlock:
1886         mutex_unlock(&nfs_clid_init_mutex);
1887 out_free:
1888         kfree(flavors);
1889 out:
1890         dprintk("NFS: %s: status = %d\n", __func__, status);
1891         return status;
1892 }
1893
1894 #ifdef CONFIG_NFS_V4_1
1895 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
1896 {
1897         struct nfs_client *clp = session->clp;
1898
1899         switch (err) {
1900         default:
1901                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1902                 break;
1903         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1904                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1905         }
1906         nfs4_schedule_lease_recovery(clp);
1907 }
1908 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1909
1910 void nfs41_handle_recall_slot(struct nfs_client *clp)
1911 {
1912         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1913         dprintk("%s: scheduling slot recall for server %s\n", __func__,
1914                         clp->cl_hostname);
1915         nfs4_schedule_state_manager(clp);
1916 }
1917
1918 static void nfs4_reset_all_state(struct nfs_client *clp)
1919 {
1920         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1921                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1922                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1923                 nfs4_state_start_reclaim_nograce(clp);
1924                 dprintk("%s: scheduling reset of all state for server %s!\n",
1925                                 __func__, clp->cl_hostname);
1926                 nfs4_schedule_state_manager(clp);
1927         }
1928 }
1929
1930 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1931 {
1932         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1933                 nfs4_state_start_reclaim_reboot(clp);
1934                 dprintk("%s: server %s rebooted!\n", __func__,
1935                                 clp->cl_hostname);
1936                 nfs4_schedule_state_manager(clp);
1937         }
1938 }
1939
1940 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1941 {
1942         nfs4_reset_all_state(clp);
1943         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
1944 }
1945
1946 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1947 {
1948         /* This will need to handle layouts too */
1949         nfs_expire_all_delegations(clp);
1950         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
1951                         clp->cl_hostname);
1952 }
1953
1954 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
1955 {
1956         nfs_expire_all_delegations(clp);
1957         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1958                 nfs4_schedule_state_manager(clp);
1959         dprintk("%s: server %s declared a backchannel fault\n", __func__,
1960                         clp->cl_hostname);
1961 }
1962
1963 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1964 {
1965         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
1966                 &clp->cl_state) == 0)
1967                 nfs4_schedule_state_manager(clp);
1968 }
1969
1970 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1971 {
1972         if (!flags)
1973                 return;
1974
1975         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
1976                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
1977
1978         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1979                 nfs41_handle_server_reboot(clp);
1980         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1981                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1982                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1983                             SEQ4_STATUS_LEASE_MOVED))
1984                 nfs41_handle_state_revoked(clp);
1985         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1986                 nfs41_handle_recallable_state_revoked(clp);
1987         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
1988                 nfs41_handle_backchannel_fault(clp);
1989         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1990                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1991                 nfs41_handle_cb_path_down(clp);
1992 }
1993
1994 static int nfs4_reset_session(struct nfs_client *clp)
1995 {
1996         struct rpc_cred *cred;
1997         int status;
1998
1999         if (!nfs4_has_session(clp))
2000                 return 0;
2001         nfs4_begin_drain_session(clp);
2002         cred = nfs4_get_exchange_id_cred(clp);
2003         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2004         if (status && status != -NFS4ERR_BADSESSION &&
2005             status != -NFS4ERR_DEADSESSION) {
2006                 status = nfs4_recovery_handle_error(clp, status);
2007                 goto out;
2008         }
2009
2010         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2011         status = nfs4_proc_create_session(clp, cred);
2012         if (status) {
2013                 dprintk("%s: session reset failed with status %d for server %s!\n",
2014                         __func__, status, clp->cl_hostname);
2015                 status = nfs4_handle_reclaim_lease_error(clp, status);
2016                 goto out;
2017         }
2018         nfs41_finish_session_reset(clp);
2019         dprintk("%s: session reset was successful for server %s!\n",
2020                         __func__, clp->cl_hostname);
2021 out:
2022         if (cred)
2023                 put_rpccred(cred);
2024         return status;
2025 }
2026
2027 static int nfs4_recall_slot(struct nfs_client *clp)
2028 {
2029         struct nfs4_slot_table *fc_tbl;
2030         struct nfs4_slot *new, *old;
2031         int i;
2032
2033         if (!nfs4_has_session(clp))
2034                 return 0;
2035         nfs4_begin_drain_session(clp);
2036         fc_tbl = &clp->cl_session->fc_slot_table;
2037         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
2038                       GFP_NOFS);
2039         if (!new)
2040                 return -ENOMEM;
2041
2042         spin_lock(&fc_tbl->slot_tbl_lock);
2043         for (i = 0; i < fc_tbl->target_max_slots; i++)
2044                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
2045         old = fc_tbl->slots;
2046         fc_tbl->slots = new;
2047         fc_tbl->max_slots = fc_tbl->target_max_slots;
2048         fc_tbl->target_max_slots = 0;
2049         clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
2050         spin_unlock(&fc_tbl->slot_tbl_lock);
2051
2052         kfree(old);
2053         return 0;
2054 }
2055
2056 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2057 {
2058         struct rpc_cred *cred;
2059         int ret;
2060
2061         if (!nfs4_has_session(clp))
2062                 return 0;
2063         nfs4_begin_drain_session(clp);
2064         cred = nfs4_get_exchange_id_cred(clp);
2065         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2066         if (cred)
2067                 put_rpccred(cred);
2068         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2069         switch (ret) {
2070         case 0:
2071                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2072                         __func__, clp->cl_hostname);
2073                 break;
2074         case -NFS4ERR_DELAY:
2075                 ssleep(1);
2076                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2077                 break;
2078         default:
2079                 return nfs4_recovery_handle_error(clp, ret);
2080         }
2081         return 0;
2082 }
2083 #else /* CONFIG_NFS_V4_1 */
2084 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2085 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
2086 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
2087
2088 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2089 {
2090         return 0;
2091 }
2092 #endif /* CONFIG_NFS_V4_1 */
2093
2094 static void nfs4_state_manager(struct nfs_client *clp)
2095 {
2096         int status = 0;
2097         const char *section = "", *section_sep = "";
2098
2099         /* Ensure exclusive access to NFSv4 state */
2100         do {
2101                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2102                         section = "purge state";
2103                         status = nfs4_purge_lease(clp);
2104                         if (status < 0)
2105                                 goto out_error;
2106                         continue;
2107                 }
2108
2109                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2110                         section = "lease expired";
2111                         /* We're going to have to re-establish a clientid */
2112                         status = nfs4_reclaim_lease(clp);
2113                         if (status < 0)
2114                                 goto out_error;
2115                         continue;
2116                 }
2117
2118                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2119                         section = "check lease";
2120                         status = nfs4_check_lease(clp);
2121                         if (status < 0)
2122                                 goto out_error;
2123                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2124                                 continue;
2125                 }
2126
2127                 /* Initialize or reset the session */
2128                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2129                         section = "reset session";
2130                         status = nfs4_reset_session(clp);
2131                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2132                                 continue;
2133                         if (status < 0)
2134                                 goto out_error;
2135                 }
2136
2137                 /* Send BIND_CONN_TO_SESSION */
2138                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2139                                 &clp->cl_state)) {
2140                         section = "bind conn to session";
2141                         status = nfs4_bind_conn_to_session(clp);
2142                         if (status < 0)
2143                                 goto out_error;
2144                         continue;
2145                 }
2146
2147                 /* Recall session slots */
2148                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)) {
2149                         section = "recall slot";
2150                         status = nfs4_recall_slot(clp);
2151                         if (status < 0)
2152                                 goto out_error;
2153                         continue;
2154                 }
2155
2156                 /* First recover reboot state... */
2157                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2158                         section = "reclaim reboot";
2159                         status = nfs4_do_reclaim(clp,
2160                                 clp->cl_mvops->reboot_recovery_ops);
2161                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2162                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
2163                                 continue;
2164                         nfs4_state_end_reclaim_reboot(clp);
2165                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2166                                 continue;
2167                         if (status < 0)
2168                                 goto out_error;
2169                 }
2170
2171                 /* Now recover expired state... */
2172                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2173                         section = "reclaim nograce";
2174                         status = nfs4_do_reclaim(clp,
2175                                 clp->cl_mvops->nograce_recovery_ops);
2176                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
2177                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
2178                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
2179                                 continue;
2180                         if (status < 0)
2181                                 goto out_error;
2182                 }
2183
2184                 nfs4_end_drain_session(clp);
2185                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2186                         nfs_client_return_marked_delegations(clp);
2187                         continue;
2188                 }
2189
2190                 nfs4_clear_state_manager_bit(clp);
2191                 /* Did we race with an attempt to give us more work? */
2192                 if (clp->cl_state == 0)
2193                         break;
2194                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2195                         break;
2196         } while (atomic_read(&clp->cl_count) > 1);
2197         return;
2198 out_error:
2199         if (strlen(section))
2200                 section_sep = ": ";
2201         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2202                         " with error %d\n", section_sep, section,
2203                         clp->cl_hostname, -status);
2204         ssleep(1);
2205         nfs4_end_drain_session(clp);
2206         nfs4_clear_state_manager_bit(clp);
2207 }
2208
2209 static int nfs4_run_state_manager(void *ptr)
2210 {
2211         struct nfs_client *clp = ptr;
2212
2213         allow_signal(SIGKILL);
2214         nfs4_state_manager(clp);
2215         nfs_put_client(clp);
2216         module_put_and_exit(0);
2217         return 0;
2218 }
2219
2220 /*
2221  * Local variables:
2222  *  c-basic-offset: 8
2223  * End:
2224  */