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 "nfs4_fs.h"
  55 #include "callback.h"
  56 #include "delegation.h"
  57 #include "internal.h"
  58 #include "pnfs.h"
  59
  60 #define OPENOWNER_POOL_SIZE     8
  61
  62 const nfs4_stateid zero_stateid;
  63
  64 static LIST_HEAD(nfs4_clientid_list);
  65
  66 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  67 {
  68         struct nfs4_setclientid_res clid = {
  69                 .clientid = clp->cl_clientid,
  70                 .confirm = clp->cl_confirm,
  71         };
  72         unsigned short port;
  73         int status;
  74
  75         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  76                 goto do_confirm;
  77         port = nfs_callback_tcpport;
  78         if (clp->cl_addr.ss_family == AF_INET6)
  79                 port = nfs_callback_tcpport6;
  80
  81         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  82         if (status != 0)
  83                 goto out;
  84         clp->cl_clientid = clid.clientid;
  85         clp->cl_confirm = clid.confirm;
  86         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  87 do_confirm:
  88         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  89         if (status != 0)
  90                 goto out;
  91         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  92         nfs4_schedule_state_renewal(clp);
  93 out:
  94         return status;
  95 }
  96
  97 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  98 {
  99         struct rpc_cred *cred = NULL;
 100
 101         if (clp->cl_machine_cred != NULL)
 102                 cred = get_rpccred(clp->cl_machine_cred);
 103         return cred;
 104 }
 105
 106 static void nfs4_clear_machine_cred(struct nfs_client *clp)
 107 {
 108         struct rpc_cred *cred;
 109
 110         spin_lock(&clp->cl_lock);
 111         cred = clp->cl_machine_cred;
 112         clp->cl_machine_cred = NULL;
 113         spin_unlock(&clp->cl_lock);
 114         if (cred != NULL)
 115                 put_rpccred(cred);
 116 }
 117
 118 static struct rpc_cred *
 119 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 120 {
 121         struct rpc_cred *cred = NULL;
 122         struct nfs4_state_owner *sp;
 123         struct rb_node *pos;
 124
 125         for (pos = rb_first(&server->state_owners);
 126              pos != NULL;
 127              pos = rb_next(pos)) {
 128                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 129                 if (list_empty(&sp->so_states))
 130                         continue;
 131                 cred = get_rpccred(sp->so_cred);
 132                 break;
 133         }
 134         return cred;
 135 }
 136
 137 /**
 138  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 139  * @clp: client state handle
 140  *
 141  * Returns an rpc_cred with reference count bumped, or NULL.
 142  * Caller must hold clp->cl_lock.
 143  */
 144 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 145 {
 146         struct rpc_cred *cred = NULL;
 147         struct nfs_server *server;
 148
 149         rcu_read_lock();
 150         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 151                 cred = nfs4_get_renew_cred_server_locked(server);
 152                 if (cred != NULL)
 153                         break;
 154         }
 155         rcu_read_unlock();
 156         return cred;
 157 }
 158
 159 #if defined(CONFIG_NFS_V4_1)
 160
 161 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 162 {
 163         int status;
 164         struct nfs_fsinfo fsinfo;
 165
 166         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 167                 nfs4_schedule_state_renewal(clp);
 168                 return 0;
 169         }
 170
 171         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 172         if (status == 0) {
 173                 /* Update lease time and schedule renewal */
 174                 spin_lock(&clp->cl_lock);
 175                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 176                 clp->cl_last_renewal = jiffies;
 177                 spin_unlock(&clp->cl_lock);
 178
 179                 nfs4_schedule_state_renewal(clp);
 180         }
 181
 182         return status;
 183 }
 184
 185 /*
 186  * Back channel returns NFS4ERR_DELAY for new requests when
 187  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 188  * is ended.
 189  */
 190 static void nfs4_end_drain_session(struct nfs_client *clp)
 191 {
 192         struct nfs4_session *ses = clp->cl_session;
 193         struct nfs4_slot_table *tbl;
 194         int max_slots;
 195
 196         if (ses == NULL)
 197                 return;
 198         tbl = &ses->fc_slot_table;
 199         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 200                 spin_lock(&tbl->slot_tbl_lock);
 201                 max_slots = tbl->max_slots;
 202                 while (max_slots--) {
 203                         if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 204                                                 nfs4_set_task_privileged,
 205                                                 NULL) == NULL)
 206                                 break;
 207                 }
 208                 spin_unlock(&tbl->slot_tbl_lock);
 209         }
 210 }
 211
 212 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 213 {
 214         spin_lock(&tbl->slot_tbl_lock);
 215         if (tbl->highest_used_slotid != -1) {
 216                 INIT_COMPLETION(tbl->complete);
 217                 spin_unlock(&tbl->slot_tbl_lock);
 218                 return wait_for_completion_interruptible(&tbl->complete);
 219         }
 220         spin_unlock(&tbl->slot_tbl_lock);
 221         return 0;
 222 }
 223
 224 static int nfs4_begin_drain_session(struct nfs_client *clp)
 225 {
 226         struct nfs4_session *ses = clp->cl_session;
 227         int ret = 0;
 228
 229         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 230         /* back channel */
 231         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 232         if (ret)
 233                 return ret;
 234         /* fore channel */
 235         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 236 }
 237
 238 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 239 {
 240         int status;
 241
 242         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 243                 goto do_confirm;
 244         nfs4_begin_drain_session(clp);
 245         status = nfs4_proc_exchange_id(clp, cred);
 246         if (status != 0)
 247                 goto out;
 248         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 249 do_confirm:
 250         status = nfs4_proc_create_session(clp);
 251         if (status != 0)
 252                 goto out;
 253         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 254         nfs41_setup_state_renewal(clp);
 255         nfs_mark_client_ready(clp, NFS_CS_READY);
 256 out:
 257         return status;
 258 }
 259
 260 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 261 {
 262         struct rpc_cred *cred;
 263
 264         spin_lock(&clp->cl_lock);
 265         cred = nfs4_get_machine_cred_locked(clp);
 266         spin_unlock(&clp->cl_lock);
 267         return cred;
 268 }
 269
 270 #endif /* CONFIG_NFS_V4_1 */
 271
 272 static struct rpc_cred *
 273 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 274 {
 275         struct nfs_client *clp = server->nfs_client;
 276         struct rpc_cred *cred = NULL;
 277         struct nfs4_state_owner *sp;
 278         struct rb_node *pos;
 279
 280         spin_lock(&clp->cl_lock);
 281         pos = rb_first(&server->state_owners);
 282         if (pos != NULL) {
 283                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 284                 cred = get_rpccred(sp->so_cred);
 285         }
 286         spin_unlock(&clp->cl_lock);
 287         return cred;
 288 }
 289
 290 /**
 291  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 292  * @clp: client state handle
 293  *
 294  * Returns an rpc_cred with reference count bumped, or NULL.
 295  */
 296 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 297 {
 298         struct nfs_server *server;
 299         struct rpc_cred *cred;
 300
 301         spin_lock(&clp->cl_lock);
 302         cred = nfs4_get_machine_cred_locked(clp);
 303         spin_unlock(&clp->cl_lock);
 304         if (cred != NULL)
 305                 goto out;
 306
 307         rcu_read_lock();
 308         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 309                 cred = nfs4_get_setclientid_cred_server(server);
 310                 if (cred != NULL)
 311                         break;
 312         }
 313         rcu_read_unlock();
 314
 315 out:
 316         return cred;
 317 }
 318
 319 static struct nfs4_state_owner *
 320 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 321 {
 322         struct rb_node **p = &server->state_owners.rb_node,
 323                        *parent = NULL;
 324         struct nfs4_state_owner *sp;
 325
 326         while (*p != NULL) {
 327                 parent = *p;
 328                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 329
 330                 if (cred < sp->so_cred)
 331                         p = &parent->rb_left;
 332                 else if (cred > sp->so_cred)
 333                         p = &parent->rb_right;
 334                 else {
 335                         if (!list_empty(&sp->so_lru))
 336                                 list_del_init(&sp->so_lru);
 337                         atomic_inc(&sp->so_count);
 338                         return sp;
 339                 }
 340         }
 341         return NULL;
 342 }
 343
 344 static struct nfs4_state_owner *
 345 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 346 {
 347         struct nfs_server *server = new->so_server;
 348         struct rb_node **p = &server->state_owners.rb_node,
 349                        *parent = NULL;
 350         struct nfs4_state_owner *sp;
 351         int err;
 352
 353         while (*p != NULL) {
 354                 parent = *p;
 355                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 356
 357                 if (new->so_cred < sp->so_cred)
 358                         p = &parent->rb_left;
 359                 else if (new->so_cred > sp->so_cred)
 360                         p = &parent->rb_right;
 361                 else {
 362                         if (!list_empty(&sp->so_lru))
 363                                 list_del_init(&sp->so_lru);
 364                         atomic_inc(&sp->so_count);
 365                         return sp;
 366                 }
 367         }
 368         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 369         if (err)
 370                 return ERR_PTR(err);
 371         rb_link_node(&new->so_server_node, parent, p);
 372         rb_insert_color(&new->so_server_node, &server->state_owners);
 373         return new;
 374 }
 375
 376 static void
 377 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 378 {
 379         struct nfs_server *server = sp->so_server;
 380
 381         if (!RB_EMPTY_NODE(&sp->so_server_node))
 382                 rb_erase(&sp->so_server_node, &server->state_owners);
 383         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 384 }
 385
 386 static void
 387 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 388 {
 389         sc->flags = 0;
 390         sc->counter = 0;
 391         spin_lock_init(&sc->lock);
 392         INIT_LIST_HEAD(&sc->list);
 393         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 394 }
 395
 396 static void
 397 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 398 {
 399         rpc_destroy_wait_queue(&sc->wait);
 400 }
 401
 402 /*
 403  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 404  * create a new state_owner.
 405  *
 406  */
 407 static struct nfs4_state_owner *
 408 nfs4_alloc_state_owner(struct nfs_server *server,
 409                 struct rpc_cred *cred,
 410                 gfp_t gfp_flags)
 411 {
 412         struct nfs4_state_owner *sp;
 413
 414         sp = kzalloc(sizeof(*sp), gfp_flags);
 415         if (!sp)
 416                 return NULL;
 417         sp->so_server = server;
 418         sp->so_cred = get_rpccred(cred);
 419         spin_lock_init(&sp->so_lock);
 420         INIT_LIST_HEAD(&sp->so_states);
 421         nfs4_init_seqid_counter(&sp->so_seqid);
 422         atomic_set(&sp->so_count, 1);
 423         INIT_LIST_HEAD(&sp->so_lru);
 424         return sp;
 425 }
 426
 427 static void
 428 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 429 {
 430         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 431                 struct nfs_server *server = sp->so_server;
 432                 struct nfs_client *clp = server->nfs_client;
 433
 434                 spin_lock(&clp->cl_lock);
 435                 rb_erase(&sp->so_server_node, &server->state_owners);
 436                 RB_CLEAR_NODE(&sp->so_server_node);
 437                 spin_unlock(&clp->cl_lock);
 438         }
 439 }
 440
 441 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 442 {
 443         nfs4_destroy_seqid_counter(&sp->so_seqid);
 444         put_rpccred(sp->so_cred);
 445         kfree(sp);
 446 }
 447
 448 static void nfs4_gc_state_owners(struct nfs_server *server)
 449 {
 450         struct nfs_client *clp = server->nfs_client;
 451         struct nfs4_state_owner *sp, *tmp;
 452         unsigned long time_min, time_max;
 453         LIST_HEAD(doomed);
 454
 455         spin_lock(&clp->cl_lock);
 456         time_max = jiffies;
 457         time_min = (long)time_max - (long)clp->cl_lease_time;
 458         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 459                 /* NB: LRU is sorted so that oldest is at the head */
 460                 if (time_in_range(sp->so_expires, time_min, time_max))
 461                         break;
 462                 list_move(&sp->so_lru, &doomed);
 463                 nfs4_remove_state_owner_locked(sp);
 464         }
 465         spin_unlock(&clp->cl_lock);
 466
 467         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 468                 list_del(&sp->so_lru);
 469                 nfs4_free_state_owner(sp);
 470         }
 471 }
 472
 473 /**
 474  * nfs4_get_state_owner - Look up a state owner given a credential
 475  * @server: nfs_server to search
 476  * @cred: RPC credential to match
 477  *
 478  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 479  */
 480 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 481                                               struct rpc_cred *cred,
 482                                               gfp_t gfp_flags)
 483 {
 484         struct nfs_client *clp = server->nfs_client;
 485         struct nfs4_state_owner *sp, *new;
 486
 487         spin_lock(&clp->cl_lock);
 488         sp = nfs4_find_state_owner_locked(server, cred);
 489         spin_unlock(&clp->cl_lock);
 490         if (sp != NULL)
 491                 goto out;
 492         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 493         if (new == NULL)
 494                 goto out;
 495         do {
 496                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 497                         break;
 498                 spin_lock(&clp->cl_lock);
 499                 sp = nfs4_insert_state_owner_locked(new);
 500                 spin_unlock(&clp->cl_lock);
 501         } while (sp == ERR_PTR(-EAGAIN));
 502         if (sp != new)
 503                 nfs4_free_state_owner(new);
 504 out:
 505         nfs4_gc_state_owners(server);
 506         return sp;
 507 }
 508
 509 /**
 510  * nfs4_put_state_owner - Release a nfs4_state_owner
 511  * @sp: state owner data to release
 512  */
 513 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 514 {
 515         struct nfs_server *server = sp->so_server;
 516         struct nfs_client *clp = server->nfs_client;
 517
 518         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 519                 return;
 520
 521         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 522                 sp->so_expires = jiffies;
 523                 list_add_tail(&sp->so_lru, &server->state_owners_lru);
 524                 spin_unlock(&clp->cl_lock);
 525         } else {
 526                 nfs4_remove_state_owner_locked(sp);
 527                 spin_unlock(&clp->cl_lock);
 528                 nfs4_free_state_owner(sp);
 529         }
 530 }
 531
 532 /**
 533  * nfs4_purge_state_owners - Release all cached state owners
 534  * @server: nfs_server with cached state owners to release
 535  *
 536  * Called at umount time.  Remaining state owners will be on
 537  * the LRU with ref count of zero.
 538  */
 539 void nfs4_purge_state_owners(struct nfs_server *server)
 540 {
 541         struct nfs_client *clp = server->nfs_client;
 542         struct nfs4_state_owner *sp, *tmp;
 543         LIST_HEAD(doomed);
 544
 545         spin_lock(&clp->cl_lock);
 546         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 547                 list_move(&sp->so_lru, &doomed);
 548                 nfs4_remove_state_owner_locked(sp);
 549         }
 550         spin_unlock(&clp->cl_lock);
 551
 552         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 553                 list_del(&sp->so_lru);
 554                 nfs4_free_state_owner(sp);
 555         }
 556 }
 557
 558 static struct nfs4_state *
 559 nfs4_alloc_open_state(void)
 560 {
 561         struct nfs4_state *state;
 562
 563         state = kzalloc(sizeof(*state), GFP_NOFS);
 564         if (!state)
 565                 return NULL;
 566         atomic_set(&state->count, 1);
 567         INIT_LIST_HEAD(&state->lock_states);
 568         spin_lock_init(&state->state_lock);
 569         seqlock_init(&state->seqlock);
 570         return state;
 571 }
 572
 573 void
 574 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 575 {
 576         if (state->state == fmode)
 577                 return;
 578         /* NB! List reordering - see the reclaim code for why.  */
 579         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 580                 if (fmode & FMODE_WRITE)
 581                         list_move(&state->open_states, &state->owner->so_states);
 582                 else
 583                         list_move_tail(&state->open_states, &state->owner->so_states);
 584         }
 585         state->state = fmode;
 586 }
 587
 588 static struct nfs4_state *
 589 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 590 {
 591         struct nfs_inode *nfsi = NFS_I(inode);
 592         struct nfs4_state *state;
 593
 594         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 595                 if (state->owner != owner)
 596                         continue;
 597                 if (atomic_inc_not_zero(&state->count))
 598                         return state;
 599         }
 600         return NULL;
 601 }
 602
 603 static void
 604 nfs4_free_open_state(struct nfs4_state *state)
 605 {
 606         kfree(state);
 607 }
 608
 609 struct nfs4_state *
 610 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 611 {
 612         struct nfs4_state *state, *new;
 613         struct nfs_inode *nfsi = NFS_I(inode);
 614
 615         spin_lock(&inode->i_lock);
 616         state = __nfs4_find_state_byowner(inode, owner);
 617         spin_unlock(&inode->i_lock);
 618         if (state)
 619                 goto out;
 620         new = nfs4_alloc_open_state();
 621         spin_lock(&owner->so_lock);
 622         spin_lock(&inode->i_lock);
 623         state = __nfs4_find_state_byowner(inode, owner);
 624         if (state == NULL && new != NULL) {
 625                 state = new;
 626                 state->owner = owner;
 627                 atomic_inc(&owner->so_count);
 628                 list_add(&state->inode_states, &nfsi->open_states);
 629                 ihold(inode);
 630                 state->inode = inode;
 631                 spin_unlock(&inode->i_lock);
 632                 /* Note: The reclaim code dictates that we add stateless
 633                  * and read-only stateids to the end of the list */
 634                 list_add_tail(&state->open_states, &owner->so_states);
 635                 spin_unlock(&owner->so_lock);
 636         } else {
 637                 spin_unlock(&inode->i_lock);
 638                 spin_unlock(&owner->so_lock);
 639                 if (new)
 640                         nfs4_free_open_state(new);
 641         }
 642 out:
 643         return state;
 644 }
 645
 646 void nfs4_put_open_state(struct nfs4_state *state)
 647 {
 648         struct inode *inode = state->inode;
 649         struct nfs4_state_owner *owner = state->owner;
 650
 651         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 652                 return;
 653         spin_lock(&inode->i_lock);
 654         list_del(&state->inode_states);
 655         list_del(&state->open_states);
 656         spin_unlock(&inode->i_lock);
 657         spin_unlock(&owner->so_lock);
 658         iput(inode);
 659         nfs4_free_open_state(state);
 660         nfs4_put_state_owner(owner);
 661 }
 662
 663 /*
 664  * Close the current file.
 665  */
 666 static void __nfs4_close(struct nfs4_state *state,
 667                 fmode_t fmode, gfp_t gfp_mask, int wait)
 668 {
 669         struct nfs4_state_owner *owner = state->owner;
 670         int call_close = 0;
 671         fmode_t newstate;
 672
 673         atomic_inc(&owner->so_count);
 674         /* Protect against nfs4_find_state() */
 675         spin_lock(&owner->so_lock);
 676         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 677                 case FMODE_READ:
 678                         state->n_rdonly--;
 679                         break;
 680                 case FMODE_WRITE:
 681                         state->n_wronly--;
 682                         break;
 683                 case FMODE_READ|FMODE_WRITE:
 684                         state->n_rdwr--;
 685         }
 686         newstate = FMODE_READ|FMODE_WRITE;
 687         if (state->n_rdwr == 0) {
 688                 if (state->n_rdonly == 0) {
 689                         newstate &= ~FMODE_READ;
 690                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 691                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 692                 }
 693                 if (state->n_wronly == 0) {
 694                         newstate &= ~FMODE_WRITE;
 695                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 696                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 697                 }
 698                 if (newstate == 0)
 699                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 700         }
 701         nfs4_state_set_mode_locked(state, newstate);
 702         spin_unlock(&owner->so_lock);
 703
 704         if (!call_close) {
 705                 nfs4_put_open_state(state);
 706                 nfs4_put_state_owner(owner);
 707         } else {
 708                 bool roc = pnfs_roc(state->inode);
 709
 710                 nfs4_do_close(state, gfp_mask, wait, roc);
 711         }
 712 }
 713
 714 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 715 {
 716         __nfs4_close(state, fmode, GFP_NOFS, 0);
 717 }
 718
 719 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 720 {
 721         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 722 }
 723
 724 /*
 725  * Search the state->lock_states for an existing lock_owner
 726  * that is compatible with current->files
 727  */
 728 static struct nfs4_lock_state *
 729 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 730 {
 731         struct nfs4_lock_state *pos;
 732         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 733                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 734                         continue;
 735                 switch (pos->ls_owner.lo_type) {
 736                 case NFS4_POSIX_LOCK_TYPE:
 737                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 738                                 continue;
 739                         break;
 740                 case NFS4_FLOCK_LOCK_TYPE:
 741                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 742                                 continue;
 743                 }
 744                 atomic_inc(&pos->ls_count);
 745                 return pos;
 746         }
 747         return NULL;
 748 }
 749
 750 /*
 751  * Return a compatible lock_state. If no initialized lock_state structure
 752  * exists, return an uninitialized one.
 753  *
 754  */
 755 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)
 756 {
 757         struct nfs4_lock_state *lsp;
 758         struct nfs_server *server = state->owner->so_server;
 759
 760         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 761         if (lsp == NULL)
 762                 return NULL;
 763         nfs4_init_seqid_counter(&lsp->ls_seqid);
 764         atomic_set(&lsp->ls_count, 1);
 765         lsp->ls_state = state;
 766         lsp->ls_owner.lo_type = type;
 767         switch (lsp->ls_owner.lo_type) {
 768         case NFS4_FLOCK_LOCK_TYPE:
 769                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 770                 break;
 771         case NFS4_POSIX_LOCK_TYPE:
 772                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 773                 break;
 774         default:
 775                 goto out_free;
 776         }
 777         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 778         if (lsp->ls_seqid.owner_id < 0)
 779                 goto out_free;
 780         INIT_LIST_HEAD(&lsp->ls_locks);
 781         return lsp;
 782 out_free:
 783         kfree(lsp);
 784         return NULL;
 785 }
 786
 787 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 788 {
 789         struct nfs_server *server = lsp->ls_state->owner->so_server;
 790
 791         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 792         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 793         kfree(lsp);
 794 }
 795
 796 /*
 797  * Return a compatible lock_state. If no initialized lock_state structure
 798  * exists, return an uninitialized one.
 799  *
 800  */
 801 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 802 {
 803         struct nfs4_lock_state *lsp, *new = NULL;
 804
 805         for(;;) {
 806                 spin_lock(&state->state_lock);
 807                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 808                 if (lsp != NULL)
 809                         break;
 810                 if (new != NULL) {
 811                         list_add(&new->ls_locks, &state->lock_states);
 812                         set_bit(LK_STATE_IN_USE, &state->flags);
 813                         lsp = new;
 814                         new = NULL;
 815                         break;
 816                 }
 817                 spin_unlock(&state->state_lock);
 818                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 819                 if (new == NULL)
 820                         return NULL;
 821         }
 822         spin_unlock(&state->state_lock);
 823         if (new != NULL)
 824                 nfs4_free_lock_state(new);
 825         return lsp;
 826 }
 827
 828 /*
 829  * Release reference to lock_state, and free it if we see that
 830  * it is no longer in use
 831  */
 832 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 833 {
 834         struct nfs4_state *state;
 835
 836         if (lsp == NULL)
 837                 return;
 838         state = lsp->ls_state;
 839         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 840                 return;
 841         list_del(&lsp->ls_locks);
 842         if (list_empty(&state->lock_states))
 843                 clear_bit(LK_STATE_IN_USE, &state->flags);
 844         spin_unlock(&state->state_lock);
 845         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 846                 nfs4_release_lockowner(lsp);
 847         nfs4_free_lock_state(lsp);
 848 }
 849
 850 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 851 {
 852         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 853
 854         dst->fl_u.nfs4_fl.owner = lsp;
 855         atomic_inc(&lsp->ls_count);
 856 }
 857
 858 static void nfs4_fl_release_lock(struct file_lock *fl)
 859 {
 860         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 861 }
 862
 863 static const struct file_lock_operations nfs4_fl_lock_ops = {
 864         .fl_copy_lock = nfs4_fl_copy_lock,
 865         .fl_release_private = nfs4_fl_release_lock,
 866 };
 867
 868 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 869 {
 870         struct nfs4_lock_state *lsp;
 871
 872         if (fl->fl_ops != NULL)
 873                 return 0;
 874         if (fl->fl_flags & FL_POSIX)
 875                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 876         else if (fl->fl_flags & FL_FLOCK)
 877                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 878         else
 879                 return -EINVAL;
 880         if (lsp == NULL)
 881                 return -ENOMEM;
 882         fl->fl_u.nfs4_fl.owner = lsp;
 883         fl->fl_ops = &nfs4_fl_lock_ops;
 884         return 0;
 885 }
 886
 887 /*
 888  * Byte-range lock aware utility to initialize the stateid of read/write
 889  * requests.
 890  */
 891 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 892 {
 893         struct nfs4_lock_state *lsp;
 894         int seq;
 895
 896         do {
 897                 seq = read_seqbegin(&state->seqlock);
 898                 memcpy(dst, &state->stateid, sizeof(*dst));
 899         } while (read_seqretry(&state->seqlock, seq));
 900         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 901                 return;
 902
 903         spin_lock(&state->state_lock);
 904         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 905         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 906                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 907         spin_unlock(&state->state_lock);
 908         nfs4_put_lock_state(lsp);
 909 }
 910
 911 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 912 {
 913         struct nfs_seqid *new;
 914
 915         new = kmalloc(sizeof(*new), gfp_mask);
 916         if (new != NULL) {
 917                 new->sequence = counter;
 918                 INIT_LIST_HEAD(&new->list);
 919                 new->task = NULL;
 920         }
 921         return new;
 922 }
 923
 924 void nfs_release_seqid(struct nfs_seqid *seqid)
 925 {
 926         struct nfs_seqid_counter *sequence;
 927
 928         if (list_empty(&seqid->list))
 929                 return;
 930         sequence = seqid->sequence;
 931         spin_lock(&sequence->lock);
 932         list_del_init(&seqid->list);
 933         if (!list_empty(&sequence->list)) {
 934                 struct nfs_seqid *next;
 935
 936                 next = list_first_entry(&sequence->list,
 937                                 struct nfs_seqid, list);
 938                 rpc_wake_up_queued_task(&sequence->wait, next->task);
 939         }
 940         spin_unlock(&sequence->lock);
 941 }
 942
 943 void nfs_free_seqid(struct nfs_seqid *seqid)
 944 {
 945         nfs_release_seqid(seqid);
 946         kfree(seqid);
 947 }
 948
 949 /*
 950  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 951  * failed with a seqid incrementing error -
 952  * see comments nfs_fs.h:seqid_mutating_error()
 953  */
 954 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 955 {
 956         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
 957         switch (status) {
 958                 case 0:
 959                         break;
 960                 case -NFS4ERR_BAD_SEQID:
 961                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 962                                 return;
 963                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 964                                         " sequence-id error on an"
 965                                         " unconfirmed sequence %p!\n",
 966                                         seqid->sequence);
 967                 case -NFS4ERR_STALE_CLIENTID:
 968                 case -NFS4ERR_STALE_STATEID:
 969                 case -NFS4ERR_BAD_STATEID:
 970                 case -NFS4ERR_BADXDR:
 971                 case -NFS4ERR_RESOURCE:
 972                 case -NFS4ERR_NOFILEHANDLE:
 973                         /* Non-seqid mutating errors */
 974                         return;
 975         };
 976         /*
 977          * Note: no locking needed as we are guaranteed to be first
 978          * on the sequence list
 979          */
 980         seqid->sequence->counter++;
 981 }
 982
 983 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 984 {
 985         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 986                                         struct nfs4_state_owner, so_seqid);
 987         struct nfs_server *server = sp->so_server;
 988
 989         if (status == -NFS4ERR_BAD_SEQID)
 990                 nfs4_drop_state_owner(sp);
 991         if (!nfs4_has_session(server->nfs_client))
 992                 nfs_increment_seqid(status, seqid);
 993 }
 994
 995 /*
 996  * Increment the seqid if the LOCK/LOCKU succeeded, or
 997  * failed with a seqid incrementing error -
 998  * see comments nfs_fs.h:seqid_mutating_error()
 999  */
1000 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1001 {
1002         nfs_increment_seqid(status, seqid);
1003 }
1004
1005 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1006 {
1007         struct nfs_seqid_counter *sequence = seqid->sequence;
1008         int status = 0;
1009
1010         spin_lock(&sequence->lock);
1011         seqid->task = task;
1012         if (list_empty(&seqid->list))
1013                 list_add_tail(&seqid->list, &sequence->list);
1014         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1015                 goto unlock;
1016         rpc_sleep_on(&sequence->wait, task, NULL);
1017         status = -EAGAIN;
1018 unlock:
1019         spin_unlock(&sequence->lock);
1020         return status;
1021 }
1022
1023 static int nfs4_run_state_manager(void *);
1024
1025 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1026 {
1027         smp_mb__before_clear_bit();
1028         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1029         smp_mb__after_clear_bit();
1030         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1031         rpc_wake_up(&clp->cl_rpcwaitq);
1032 }
1033
1034 /*
1035  * Schedule the nfs_client asynchronous state management routine
1036  */
1037 void nfs4_schedule_state_manager(struct nfs_client *clp)
1038 {
1039         struct task_struct *task;
1040
1041         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1042                 return;
1043         __module_get(THIS_MODULE);
1044         atomic_inc(&clp->cl_count);
1045         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1046                                 rpc_peeraddr2str(clp->cl_rpcclient,
1047                                                         RPC_DISPLAY_ADDR));
1048         if (!IS_ERR(task))
1049                 return;
1050         nfs4_clear_state_manager_bit(clp);
1051         nfs_put_client(clp);
1052         module_put(THIS_MODULE);
1053 }
1054
1055 /*
1056  * Schedule a lease recovery attempt
1057  */
1058 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1059 {
1060         if (!clp)
1061                 return;
1062         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1063                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1064         nfs4_schedule_state_manager(clp);
1065 }
1066
1067 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1068 {
1069         nfs_handle_cb_pathdown(clp);
1070         nfs4_schedule_state_manager(clp);
1071 }
1072
1073 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1074 {
1075
1076         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1077         /* Don't recover state that expired before the reboot */
1078         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1079                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1080                 return 0;
1081         }
1082         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1083         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1084         return 1;
1085 }
1086
1087 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1088 {
1089         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1090         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1091         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1092         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1093         return 1;
1094 }
1095
1096 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1097 {
1098         struct nfs_client *clp = server->nfs_client;
1099
1100         nfs4_state_mark_reclaim_nograce(clp, state);
1101         nfs4_schedule_state_manager(clp);
1102 }
1103
1104 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1105 {
1106         struct inode *inode = state->inode;
1107         struct nfs_inode *nfsi = NFS_I(inode);
1108         struct file_lock *fl;
1109         int status = 0;
1110
1111         if (inode->i_flock == NULL)
1112                 return 0;
1113
1114         /* Guard against delegation returns and new lock/unlock calls */
1115         down_write(&nfsi->rwsem);
1116         /* Protect inode->i_flock using the BKL */
1117         lock_flocks();
1118         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1119                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1120                         continue;
1121                 if (nfs_file_open_context(fl->fl_file)->state != state)
1122                         continue;
1123                 unlock_flocks();
1124                 status = ops->recover_lock(state, fl);
1125                 switch (status) {
1126                         case 0:
1127                                 break;
1128                         case -ESTALE:
1129                         case -NFS4ERR_ADMIN_REVOKED:
1130                         case -NFS4ERR_STALE_STATEID:
1131                         case -NFS4ERR_BAD_STATEID:
1132                         case -NFS4ERR_EXPIRED:
1133                         case -NFS4ERR_NO_GRACE:
1134                         case -NFS4ERR_STALE_CLIENTID:
1135                         case -NFS4ERR_BADSESSION:
1136                         case -NFS4ERR_BADSLOT:
1137                         case -NFS4ERR_BAD_HIGH_SLOT:
1138                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1139                                 goto out;
1140                         default:
1141                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1142                                                 __func__, status);
1143                         case -ENOMEM:
1144                         case -NFS4ERR_DENIED:
1145                         case -NFS4ERR_RECLAIM_BAD:
1146                         case -NFS4ERR_RECLAIM_CONFLICT:
1147                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1148                                 status = 0;
1149                 }
1150                 lock_flocks();
1151         }
1152         unlock_flocks();
1153 out:
1154         up_write(&nfsi->rwsem);
1155         return status;
1156 }
1157
1158 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1159 {
1160         struct nfs4_state *state;
1161         struct nfs4_lock_state *lock;
1162         int status = 0;
1163
1164         /* Note: we rely on the sp->so_states list being ordered
1165          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1166          * states first.
1167          * This is needed to ensure that the server won't give us any
1168          * read delegations that we have to return if, say, we are
1169          * recovering after a network partition or a reboot from a
1170          * server that doesn't support a grace period.
1171          */
1172 restart:
1173         spin_lock(&sp->so_lock);
1174         list_for_each_entry(state, &sp->so_states, open_states) {
1175                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1176                         continue;
1177                 if (state->state == 0)
1178                         continue;
1179                 atomic_inc(&state->count);
1180                 spin_unlock(&sp->so_lock);
1181                 status = ops->recover_open(sp, state);
1182                 if (status >= 0) {
1183                         status = nfs4_reclaim_locks(state, ops);
1184                         if (status >= 0) {
1185                                 spin_lock(&state->state_lock);
1186                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1187                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1188                                                 printk("%s: Lock reclaim failed!\n",
1189                                                         __func__);
1190                                 }
1191                                 spin_unlock(&state->state_lock);
1192                                 nfs4_put_open_state(state);
1193                                 goto restart;
1194                         }
1195                 }
1196                 switch (status) {
1197                         default:
1198                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1199                                                 __func__, status);
1200                         case -ENOENT:
1201                         case -ENOMEM:
1202                         case -ESTALE:
1203                                 /*
1204                                  * Open state on this file cannot be recovered
1205                                  * All we can do is revert to using the zero stateid.
1206                                  */
1207                                 memset(state->stateid.data, 0,
1208                                         sizeof(state->stateid.data));
1209                                 /* Mark the file as being 'closed' */
1210                                 state->state = 0;
1211                                 break;
1212                         case -EKEYEXPIRED:
1213                                 /*
1214                                  * User RPCSEC_GSS context has expired.
1215                                  * We cannot recover this stateid now, so
1216                                  * skip it and allow recovery thread to
1217                                  * proceed.
1218                                  */
1219                                 break;
1220                         case -NFS4ERR_ADMIN_REVOKED:
1221                         case -NFS4ERR_STALE_STATEID:
1222                         case -NFS4ERR_BAD_STATEID:
1223                         case -NFS4ERR_RECLAIM_BAD:
1224                         case -NFS4ERR_RECLAIM_CONFLICT:
1225                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1226                                 break;
1227                         case -NFS4ERR_EXPIRED:
1228                         case -NFS4ERR_NO_GRACE:
1229                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1230                         case -NFS4ERR_STALE_CLIENTID:
1231                         case -NFS4ERR_BADSESSION:
1232                         case -NFS4ERR_BADSLOT:
1233                         case -NFS4ERR_BAD_HIGH_SLOT:
1234                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1235                                 goto out_err;
1236                 }
1237                 nfs4_put_open_state(state);
1238                 goto restart;
1239         }
1240         spin_unlock(&sp->so_lock);
1241         return 0;
1242 out_err:
1243         nfs4_put_open_state(state);
1244         return status;
1245 }
1246
1247 static void nfs4_clear_open_state(struct nfs4_state *state)
1248 {
1249         struct nfs4_lock_state *lock;
1250
1251         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1252         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1253         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1254         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1255         spin_lock(&state->state_lock);
1256         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1257                 lock->ls_seqid.flags = 0;
1258                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1259         }
1260         spin_unlock(&state->state_lock);
1261 }
1262
1263 static void nfs4_reset_seqids(struct nfs_server *server,
1264         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1265 {
1266         struct nfs_client *clp = server->nfs_client;
1267         struct nfs4_state_owner *sp;
1268         struct rb_node *pos;
1269         struct nfs4_state *state;
1270
1271         spin_lock(&clp->cl_lock);
1272         for (pos = rb_first(&server->state_owners);
1273              pos != NULL;
1274              pos = rb_next(pos)) {
1275                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1276                 sp->so_seqid.flags = 0;
1277                 spin_lock(&sp->so_lock);
1278                 list_for_each_entry(state, &sp->so_states, open_states) {
1279                         if (mark_reclaim(clp, state))
1280                                 nfs4_clear_open_state(state);
1281                 }
1282                 spin_unlock(&sp->so_lock);
1283         }
1284         spin_unlock(&clp->cl_lock);
1285 }
1286
1287 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1288         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1289 {
1290         struct nfs_server *server;
1291
1292         rcu_read_lock();
1293         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1294                 nfs4_reset_seqids(server, mark_reclaim);
1295         rcu_read_unlock();
1296 }
1297
1298 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1299 {
1300         /* Mark all delegations for reclaim */
1301         nfs_delegation_mark_reclaim(clp);
1302         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1303 }
1304
1305 static void nfs4_reclaim_complete(struct nfs_client *clp,
1306                                  const struct nfs4_state_recovery_ops *ops)
1307 {
1308         /* Notify the server we're done reclaiming our state */
1309         if (ops->reclaim_complete)
1310                 (void)ops->reclaim_complete(clp);
1311 }
1312
1313 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1314 {
1315         struct nfs_client *clp = server->nfs_client;
1316         struct nfs4_state_owner *sp;
1317         struct rb_node *pos;
1318         struct nfs4_state *state;
1319
1320         spin_lock(&clp->cl_lock);
1321         for (pos = rb_first(&server->state_owners);
1322              pos != NULL;
1323              pos = rb_next(pos)) {
1324                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1325                 spin_lock(&sp->so_lock);
1326                 list_for_each_entry(state, &sp->so_states, open_states) {
1327                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1328                                                 &state->flags))
1329                                 continue;
1330                         nfs4_state_mark_reclaim_nograce(clp, state);
1331                 }
1332                 spin_unlock(&sp->so_lock);
1333         }
1334         spin_unlock(&clp->cl_lock);
1335 }
1336
1337 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1338 {
1339         struct nfs_server *server;
1340
1341         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1342                 return 0;
1343
1344         rcu_read_lock();
1345         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1346                 nfs4_clear_reclaim_server(server);
1347         rcu_read_unlock();
1348
1349         nfs_delegation_reap_unclaimed(clp);
1350         return 1;
1351 }
1352
1353 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1354 {
1355         if (!nfs4_state_clear_reclaim_reboot(clp))
1356                 return;
1357         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1358 }
1359
1360 static void nfs_delegation_clear_all(struct nfs_client *clp)
1361 {
1362         nfs_delegation_mark_reclaim(clp);
1363         nfs_delegation_reap_unclaimed(clp);
1364 }
1365
1366 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1367 {
1368         nfs_delegation_clear_all(clp);
1369         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1370 }
1371
1372 static void nfs4_warn_keyexpired(const char *s)
1373 {
1374         printk_ratelimited(KERN_WARNING "Error: state manager"
1375                         " encountered RPCSEC_GSS session"
1376                         " expired against NFSv4 server %s.\n",
1377                         s);
1378 }
1379
1380 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1381 {
1382         switch (error) {
1383                 case 0:
1384                         break;
1385                 case -NFS4ERR_CB_PATH_DOWN:
1386                         nfs_handle_cb_pathdown(clp);
1387                         break;
1388                 case -NFS4ERR_NO_GRACE:
1389                         nfs4_state_end_reclaim_reboot(clp);
1390                         break;
1391                 case -NFS4ERR_STALE_CLIENTID:
1392                 case -NFS4ERR_LEASE_MOVED:
1393                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1394                         nfs4_state_clear_reclaim_reboot(clp);
1395                         nfs4_state_start_reclaim_reboot(clp);
1396                         break;
1397                 case -NFS4ERR_EXPIRED:
1398                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1399                         nfs4_state_start_reclaim_nograce(clp);
1400                         break;
1401                 case -NFS4ERR_BADSESSION:
1402                 case -NFS4ERR_BADSLOT:
1403                 case -NFS4ERR_BAD_HIGH_SLOT:
1404                 case -NFS4ERR_DEADSESSION:
1405                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1406                 case -NFS4ERR_SEQ_FALSE_RETRY:
1407                 case -NFS4ERR_SEQ_MISORDERED:
1408                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1409                         /* Zero session reset errors */
1410                         break;
1411                 case -EKEYEXPIRED:
1412                         /* Nothing we can do */
1413                         nfs4_warn_keyexpired(clp->cl_hostname);
1414                         break;
1415                 default:
1416                         return error;
1417         }
1418         return 0;
1419 }
1420
1421 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1422 {
1423         struct nfs4_state_owner *sp;
1424         struct nfs_server *server;
1425         struct rb_node *pos;
1426         int status = 0;
1427
1428 restart:
1429         rcu_read_lock();
1430         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1431                 nfs4_purge_state_owners(server);
1432                 spin_lock(&clp->cl_lock);
1433                 for (pos = rb_first(&server->state_owners);
1434                      pos != NULL;
1435                      pos = rb_next(pos)) {
1436                         sp = rb_entry(pos,
1437                                 struct nfs4_state_owner, so_server_node);
1438                         if (!test_and_clear_bit(ops->owner_flag_bit,
1439                                                         &sp->so_flags))
1440                                 continue;
1441                         atomic_inc(&sp->so_count);
1442                         spin_unlock(&clp->cl_lock);
1443                         rcu_read_unlock();
1444
1445                         status = nfs4_reclaim_open_state(sp, ops);
1446                         if (status < 0) {
1447                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1448                                 nfs4_put_state_owner(sp);
1449                                 return nfs4_recovery_handle_error(clp, status);
1450                         }
1451
1452                         nfs4_put_state_owner(sp);
1453                         goto restart;
1454                 }
1455                 spin_unlock(&clp->cl_lock);
1456         }
1457         rcu_read_unlock();
1458         return status;
1459 }
1460
1461 static int nfs4_check_lease(struct nfs_client *clp)
1462 {
1463         struct rpc_cred *cred;
1464         const struct nfs4_state_maintenance_ops *ops =
1465                 clp->cl_mvops->state_renewal_ops;
1466         int status;
1467
1468         /* Is the client already known to have an expired lease? */
1469         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1470                 return 0;
1471         spin_lock(&clp->cl_lock);
1472         cred = ops->get_state_renewal_cred_locked(clp);
1473         spin_unlock(&clp->cl_lock);
1474         if (cred == NULL) {
1475                 cred = nfs4_get_setclientid_cred(clp);
1476                 status = -ENOKEY;
1477                 if (cred == NULL)
1478                         goto out;
1479         }
1480         status = ops->renew_lease(clp, cred);
1481         put_rpccred(cred);
1482 out:
1483         return nfs4_recovery_handle_error(clp, status);
1484 }
1485
1486 static int nfs4_reclaim_lease(struct nfs_client *clp)
1487 {
1488         struct rpc_cred *cred;
1489         const struct nfs4_state_recovery_ops *ops =
1490                 clp->cl_mvops->reboot_recovery_ops;
1491         int status = -ENOENT;
1492
1493         cred = ops->get_clid_cred(clp);
1494         if (cred != NULL) {
1495                 status = ops->establish_clid(clp, cred);
1496                 put_rpccred(cred);
1497                 /* Handle case where the user hasn't set up machine creds */
1498                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1499                         nfs4_clear_machine_cred(clp);
1500                         status = -EAGAIN;
1501                 }
1502                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1503                         status = -EPROTONOSUPPORT;
1504         }
1505         return status;
1506 }
1507
1508 #ifdef CONFIG_NFS_V4_1
1509 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1510 {
1511         struct nfs_client *clp = session->clp;
1512
1513         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1514         nfs4_schedule_lease_recovery(clp);
1515 }
1516 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1517
1518 void nfs41_handle_recall_slot(struct nfs_client *clp)
1519 {
1520         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1521         nfs4_schedule_state_manager(clp);
1522 }
1523
1524 static void nfs4_reset_all_state(struct nfs_client *clp)
1525 {
1526         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1527                 clp->cl_boot_time = CURRENT_TIME;
1528                 nfs4_state_start_reclaim_nograce(clp);
1529                 nfs4_schedule_state_manager(clp);
1530         }
1531 }
1532
1533 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1534 {
1535         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1536                 nfs4_state_start_reclaim_reboot(clp);
1537                 nfs4_schedule_state_manager(clp);
1538         }
1539 }
1540
1541 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1542 {
1543         /* Temporary */
1544         nfs4_reset_all_state(clp);
1545 }
1546
1547 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1548 {
1549         /* This will need to handle layouts too */
1550         nfs_expire_all_delegations(clp);
1551 }
1552
1553 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1554 {
1555         nfs_expire_all_delegations(clp);
1556         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1557                 nfs4_schedule_state_manager(clp);
1558 }
1559
1560 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1561 {
1562         if (!flags)
1563                 return;
1564         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1565                 nfs41_handle_server_reboot(clp);
1566         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1567                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1568                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1569                             SEQ4_STATUS_LEASE_MOVED))
1570                 nfs41_handle_state_revoked(clp);
1571         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1572                 nfs41_handle_recallable_state_revoked(clp);
1573         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1574                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1575                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1576                 nfs41_handle_cb_path_down(clp);
1577 }
1578
1579 static int nfs4_reset_session(struct nfs_client *clp)
1580 {
1581         int status;
1582
1583         nfs4_begin_drain_session(clp);
1584         status = nfs4_proc_destroy_session(clp->cl_session);
1585         if (status && status != -NFS4ERR_BADSESSION &&
1586             status != -NFS4ERR_DEADSESSION) {
1587                 status = nfs4_recovery_handle_error(clp, status);
1588                 goto out;
1589         }
1590
1591         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1592         status = nfs4_proc_create_session(clp);
1593         if (status) {
1594                 status = nfs4_recovery_handle_error(clp, status);
1595                 goto out;
1596         }
1597         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1598         /* create_session negotiated new slot table */
1599         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1600
1601          /* Let the state manager reestablish state */
1602         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1603                 nfs41_setup_state_renewal(clp);
1604 out:
1605         return status;
1606 }
1607
1608 static int nfs4_recall_slot(struct nfs_client *clp)
1609 {
1610         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1611         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1612         struct nfs4_slot *new, *old;
1613         int i;
1614
1615         nfs4_begin_drain_session(clp);
1616         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1617                       GFP_NOFS);
1618         if (!new)
1619                 return -ENOMEM;
1620
1621         spin_lock(&fc_tbl->slot_tbl_lock);
1622         for (i = 0; i < fc_tbl->target_max_slots; i++)
1623                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1624         old = fc_tbl->slots;
1625         fc_tbl->slots = new;
1626         fc_tbl->max_slots = fc_tbl->target_max_slots;
1627         fc_tbl->target_max_slots = 0;
1628         fc_attrs->max_reqs = fc_tbl->max_slots;
1629         spin_unlock(&fc_tbl->slot_tbl_lock);
1630
1631         kfree(old);
1632         nfs4_end_drain_session(clp);
1633         return 0;
1634 }
1635
1636 #else /* CONFIG_NFS_V4_1 */
1637 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1638 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1639 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1640 #endif /* CONFIG_NFS_V4_1 */
1641
1642 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1643  * on EXCHANGE_ID for v4.1
1644  */
1645 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1646 {
1647         switch (status) {
1648         case -NFS4ERR_CLID_INUSE:
1649         case -NFS4ERR_STALE_CLIENTID:
1650                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1651                 break;
1652         case -NFS4ERR_DELAY:
1653         case -ETIMEDOUT:
1654         case -EAGAIN:
1655                 ssleep(1);
1656                 break;
1657
1658         case -EKEYEXPIRED:
1659                 nfs4_warn_keyexpired(clp->cl_hostname);
1660         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1661                                  * in nfs4_exchange_id */
1662         default:
1663                 return;
1664         }
1665         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1666 }
1667
1668 static void nfs4_state_manager(struct nfs_client *clp)
1669 {
1670         int status = 0;
1671
1672         /* Ensure exclusive access to NFSv4 state */
1673         do {
1674                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1675                         /* We're going to have to re-establish a clientid */
1676                         status = nfs4_reclaim_lease(clp);
1677                         if (status) {
1678                                 nfs4_set_lease_expired(clp, status);
1679                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1680                                                         &clp->cl_state))
1681                                         continue;
1682                                 if (clp->cl_cons_state ==
1683                                                         NFS_CS_SESSION_INITING)
1684                                         nfs_mark_client_ready(clp, status);
1685                                 goto out_error;
1686                         }
1687                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1688
1689                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1690                                                &clp->cl_state))
1691                                 nfs4_state_start_reclaim_nograce(clp);
1692                         else
1693                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1694                                         &clp->cl_state);
1695
1696                         pnfs_destroy_all_layouts(clp);
1697                 }
1698
1699                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1700                         status = nfs4_check_lease(clp);
1701                         if (status < 0)
1702                                 goto out_error;
1703                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1704                                 continue;
1705                 }
1706
1707                 /* Initialize or reset the session */
1708                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1709                    && nfs4_has_session(clp)) {
1710                         status = nfs4_reset_session(clp);
1711                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1712                                 continue;
1713                         if (status < 0)
1714                                 goto out_error;
1715                 }
1716
1717                 /* First recover reboot state... */
1718                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1719                         status = nfs4_do_reclaim(clp,
1720                                 clp->cl_mvops->reboot_recovery_ops);
1721                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1722                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1723                                 continue;
1724                         nfs4_state_end_reclaim_reboot(clp);
1725                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1726                                 continue;
1727                         if (status < 0)
1728                                 goto out_error;
1729                 }
1730
1731                 /* Now recover expired state... */
1732                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1733                         status = nfs4_do_reclaim(clp,
1734                                 clp->cl_mvops->nograce_recovery_ops);
1735                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1736                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1737                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1738                                 continue;
1739                         if (status < 0)
1740                                 goto out_error;
1741                 }
1742
1743                 nfs4_end_drain_session(clp);
1744                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1745                         nfs_client_return_marked_delegations(clp);
1746                         continue;
1747                 }
1748                 /* Recall session slots */
1749                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1750                    && nfs4_has_session(clp)) {
1751                         status = nfs4_recall_slot(clp);
1752                         if (status < 0)
1753                                 goto out_error;
1754                         continue;
1755                 }
1756
1757
1758                 nfs4_clear_state_manager_bit(clp);
1759                 /* Did we race with an attempt to give us more work? */
1760                 if (clp->cl_state == 0)
1761                         break;
1762                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1763                         break;
1764         } while (atomic_read(&clp->cl_count) > 1);
1765         return;
1766 out_error:
1767         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1768                         " with error %d\n", clp->cl_hostname, -status);
1769         nfs4_end_drain_session(clp);
1770         nfs4_clear_state_manager_bit(clp);
1771 }
1772
1773 static int nfs4_run_state_manager(void *ptr)
1774 {
1775         struct nfs_client *clp = ptr;
1776
1777         allow_signal(SIGKILL);
1778         nfs4_state_manager(clp);
1779         nfs_put_client(clp);
1780         module_put_and_exit(0);
1781         return 0;
1782 }
1783
1784 /*
1785  * Local variables:
1786  *  c-basic-offset: 8
1787  * End:
1788  */