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         }
 920         return new;
 921 }
 922
 923 void nfs_release_seqid(struct nfs_seqid *seqid)
 924 {
 925         if (!list_empty(&seqid->list)) {
 926                 struct nfs_seqid_counter *sequence = seqid->sequence;
 927
 928                 spin_lock(&sequence->lock);
 929                 list_del_init(&seqid->list);
 930                 spin_unlock(&sequence->lock);
 931                 rpc_wake_up(&sequence->wait);
 932         }
 933 }
 934
 935 void nfs_free_seqid(struct nfs_seqid *seqid)
 936 {
 937         nfs_release_seqid(seqid);
 938         kfree(seqid);
 939 }
 940
 941 /*
 942  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 943  * failed with a seqid incrementing error -
 944  * see comments nfs_fs.h:seqid_mutating_error()
 945  */
 946 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 947 {
 948         BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
 949         switch (status) {
 950                 case 0:
 951                         break;
 952                 case -NFS4ERR_BAD_SEQID:
 953                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 954                                 return;
 955                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 956                                         " sequence-id error on an"
 957                                         " unconfirmed sequence %p!\n",
 958                                         seqid->sequence);
 959                 case -NFS4ERR_STALE_CLIENTID:
 960                 case -NFS4ERR_STALE_STATEID:
 961                 case -NFS4ERR_BAD_STATEID:
 962                 case -NFS4ERR_BADXDR:
 963                 case -NFS4ERR_RESOURCE:
 964                 case -NFS4ERR_NOFILEHANDLE:
 965                         /* Non-seqid mutating errors */
 966                         return;
 967         };
 968         /*
 969          * Note: no locking needed as we are guaranteed to be first
 970          * on the sequence list
 971          */
 972         seqid->sequence->counter++;
 973 }
 974
 975 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 976 {
 977         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 978                                         struct nfs4_state_owner, so_seqid);
 979         struct nfs_server *server = sp->so_server;
 980
 981         if (status == -NFS4ERR_BAD_SEQID)
 982                 nfs4_drop_state_owner(sp);
 983         if (!nfs4_has_session(server->nfs_client))
 984                 nfs_increment_seqid(status, seqid);
 985 }
 986
 987 /*
 988  * Increment the seqid if the LOCK/LOCKU succeeded, or
 989  * failed with a seqid incrementing error -
 990  * see comments nfs_fs.h:seqid_mutating_error()
 991  */
 992 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 993 {
 994         nfs_increment_seqid(status, seqid);
 995 }
 996
 997 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 998 {
 999         struct nfs_seqid_counter *sequence = seqid->sequence;
1000         int status = 0;
1001
1002         spin_lock(&sequence->lock);
1003         if (list_empty(&seqid->list))
1004                 list_add_tail(&seqid->list, &sequence->list);
1005         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1006                 goto unlock;
1007         rpc_sleep_on(&sequence->wait, task, NULL);
1008         status = -EAGAIN;
1009 unlock:
1010         spin_unlock(&sequence->lock);
1011         return status;
1012 }
1013
1014 static int nfs4_run_state_manager(void *);
1015
1016 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1017 {
1018         smp_mb__before_clear_bit();
1019         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1020         smp_mb__after_clear_bit();
1021         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1022         rpc_wake_up(&clp->cl_rpcwaitq);
1023 }
1024
1025 /*
1026  * Schedule the nfs_client asynchronous state management routine
1027  */
1028 void nfs4_schedule_state_manager(struct nfs_client *clp)
1029 {
1030         struct task_struct *task;
1031
1032         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1033                 return;
1034         __module_get(THIS_MODULE);
1035         atomic_inc(&clp->cl_count);
1036         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1037                                 rpc_peeraddr2str(clp->cl_rpcclient,
1038                                                         RPC_DISPLAY_ADDR));
1039         if (!IS_ERR(task))
1040                 return;
1041         nfs4_clear_state_manager_bit(clp);
1042         nfs_put_client(clp);
1043         module_put(THIS_MODULE);
1044 }
1045
1046 /*
1047  * Schedule a lease recovery attempt
1048  */
1049 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1050 {
1051         if (!clp)
1052                 return;
1053         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1054                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1055         nfs4_schedule_state_manager(clp);
1056 }
1057
1058 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1059 {
1060         nfs_handle_cb_pathdown(clp);
1061         nfs4_schedule_state_manager(clp);
1062 }
1063
1064 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1065 {
1066
1067         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1068         /* Don't recover state that expired before the reboot */
1069         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1070                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1071                 return 0;
1072         }
1073         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1074         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1075         return 1;
1076 }
1077
1078 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1079 {
1080         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1081         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1082         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1083         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1084         return 1;
1085 }
1086
1087 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1088 {
1089         struct nfs_client *clp = server->nfs_client;
1090
1091         nfs4_state_mark_reclaim_nograce(clp, state);
1092         nfs4_schedule_state_manager(clp);
1093 }
1094
1095 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1096 {
1097         struct inode *inode = state->inode;
1098         struct nfs_inode *nfsi = NFS_I(inode);
1099         struct file_lock *fl;
1100         int status = 0;
1101
1102         if (inode->i_flock == NULL)
1103                 return 0;
1104
1105         /* Guard against delegation returns and new lock/unlock calls */
1106         down_write(&nfsi->rwsem);
1107         /* Protect inode->i_flock using the BKL */
1108         lock_flocks();
1109         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1110                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1111                         continue;
1112                 if (nfs_file_open_context(fl->fl_file)->state != state)
1113                         continue;
1114                 unlock_flocks();
1115                 status = ops->recover_lock(state, fl);
1116                 switch (status) {
1117                         case 0:
1118                                 break;
1119                         case -ESTALE:
1120                         case -NFS4ERR_ADMIN_REVOKED:
1121                         case -NFS4ERR_STALE_STATEID:
1122                         case -NFS4ERR_BAD_STATEID:
1123                         case -NFS4ERR_EXPIRED:
1124                         case -NFS4ERR_NO_GRACE:
1125                         case -NFS4ERR_STALE_CLIENTID:
1126                         case -NFS4ERR_BADSESSION:
1127                         case -NFS4ERR_BADSLOT:
1128                         case -NFS4ERR_BAD_HIGH_SLOT:
1129                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1130                                 goto out;
1131                         default:
1132                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1133                                                 __func__, status);
1134                         case -ENOMEM:
1135                         case -NFS4ERR_DENIED:
1136                         case -NFS4ERR_RECLAIM_BAD:
1137                         case -NFS4ERR_RECLAIM_CONFLICT:
1138                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1139                                 status = 0;
1140                 }
1141                 lock_flocks();
1142         }
1143         unlock_flocks();
1144 out:
1145         up_write(&nfsi->rwsem);
1146         return status;
1147 }
1148
1149 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1150 {
1151         struct nfs4_state *state;
1152         struct nfs4_lock_state *lock;
1153         int status = 0;
1154
1155         /* Note: we rely on the sp->so_states list being ordered
1156          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1157          * states first.
1158          * This is needed to ensure that the server won't give us any
1159          * read delegations that we have to return if, say, we are
1160          * recovering after a network partition or a reboot from a
1161          * server that doesn't support a grace period.
1162          */
1163 restart:
1164         spin_lock(&sp->so_lock);
1165         list_for_each_entry(state, &sp->so_states, open_states) {
1166                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1167                         continue;
1168                 if (state->state == 0)
1169                         continue;
1170                 atomic_inc(&state->count);
1171                 spin_unlock(&sp->so_lock);
1172                 status = ops->recover_open(sp, state);
1173                 if (status >= 0) {
1174                         status = nfs4_reclaim_locks(state, ops);
1175                         if (status >= 0) {
1176                                 spin_lock(&state->state_lock);
1177                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1178                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1179                                                 printk("%s: Lock reclaim failed!\n",
1180                                                         __func__);
1181                                 }
1182                                 spin_unlock(&state->state_lock);
1183                                 nfs4_put_open_state(state);
1184                                 goto restart;
1185                         }
1186                 }
1187                 switch (status) {
1188                         default:
1189                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1190                                                 __func__, status);
1191                         case -ENOENT:
1192                         case -ENOMEM:
1193                         case -ESTALE:
1194                                 /*
1195                                  * Open state on this file cannot be recovered
1196                                  * All we can do is revert to using the zero stateid.
1197                                  */
1198                                 memset(state->stateid.data, 0,
1199                                         sizeof(state->stateid.data));
1200                                 /* Mark the file as being 'closed' */
1201                                 state->state = 0;
1202                                 break;
1203                         case -EKEYEXPIRED:
1204                                 /*
1205                                  * User RPCSEC_GSS context has expired.
1206                                  * We cannot recover this stateid now, so
1207                                  * skip it and allow recovery thread to
1208                                  * proceed.
1209                                  */
1210                                 break;
1211                         case -NFS4ERR_ADMIN_REVOKED:
1212                         case -NFS4ERR_STALE_STATEID:
1213                         case -NFS4ERR_BAD_STATEID:
1214                         case -NFS4ERR_RECLAIM_BAD:
1215                         case -NFS4ERR_RECLAIM_CONFLICT:
1216                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1217                                 break;
1218                         case -NFS4ERR_EXPIRED:
1219                         case -NFS4ERR_NO_GRACE:
1220                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1221                         case -NFS4ERR_STALE_CLIENTID:
1222                         case -NFS4ERR_BADSESSION:
1223                         case -NFS4ERR_BADSLOT:
1224                         case -NFS4ERR_BAD_HIGH_SLOT:
1225                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1226                                 goto out_err;
1227                 }
1228                 nfs4_put_open_state(state);
1229                 goto restart;
1230         }
1231         spin_unlock(&sp->so_lock);
1232         return 0;
1233 out_err:
1234         nfs4_put_open_state(state);
1235         return status;
1236 }
1237
1238 static void nfs4_clear_open_state(struct nfs4_state *state)
1239 {
1240         struct nfs4_lock_state *lock;
1241
1242         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1243         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1244         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1245         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1246         spin_lock(&state->state_lock);
1247         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1248                 lock->ls_seqid.flags = 0;
1249                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1250         }
1251         spin_unlock(&state->state_lock);
1252 }
1253
1254 static void nfs4_reset_seqids(struct nfs_server *server,
1255         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1256 {
1257         struct nfs_client *clp = server->nfs_client;
1258         struct nfs4_state_owner *sp;
1259         struct rb_node *pos;
1260         struct nfs4_state *state;
1261
1262         spin_lock(&clp->cl_lock);
1263         for (pos = rb_first(&server->state_owners);
1264              pos != NULL;
1265              pos = rb_next(pos)) {
1266                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1267                 sp->so_seqid.flags = 0;
1268                 spin_lock(&sp->so_lock);
1269                 list_for_each_entry(state, &sp->so_states, open_states) {
1270                         if (mark_reclaim(clp, state))
1271                                 nfs4_clear_open_state(state);
1272                 }
1273                 spin_unlock(&sp->so_lock);
1274         }
1275         spin_unlock(&clp->cl_lock);
1276 }
1277
1278 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1279         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1280 {
1281         struct nfs_server *server;
1282
1283         rcu_read_lock();
1284         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1285                 nfs4_reset_seqids(server, mark_reclaim);
1286         rcu_read_unlock();
1287 }
1288
1289 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1290 {
1291         /* Mark all delegations for reclaim */
1292         nfs_delegation_mark_reclaim(clp);
1293         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1294 }
1295
1296 static void nfs4_reclaim_complete(struct nfs_client *clp,
1297                                  const struct nfs4_state_recovery_ops *ops)
1298 {
1299         /* Notify the server we're done reclaiming our state */
1300         if (ops->reclaim_complete)
1301                 (void)ops->reclaim_complete(clp);
1302 }
1303
1304 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1305 {
1306         struct nfs_client *clp = server->nfs_client;
1307         struct nfs4_state_owner *sp;
1308         struct rb_node *pos;
1309         struct nfs4_state *state;
1310
1311         spin_lock(&clp->cl_lock);
1312         for (pos = rb_first(&server->state_owners);
1313              pos != NULL;
1314              pos = rb_next(pos)) {
1315                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1316                 spin_lock(&sp->so_lock);
1317                 list_for_each_entry(state, &sp->so_states, open_states) {
1318                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1319                                                 &state->flags))
1320                                 continue;
1321                         nfs4_state_mark_reclaim_nograce(clp, state);
1322                 }
1323                 spin_unlock(&sp->so_lock);
1324         }
1325         spin_unlock(&clp->cl_lock);
1326 }
1327
1328 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1329 {
1330         struct nfs_server *server;
1331
1332         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1333                 return 0;
1334
1335         rcu_read_lock();
1336         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1337                 nfs4_clear_reclaim_server(server);
1338         rcu_read_unlock();
1339
1340         nfs_delegation_reap_unclaimed(clp);
1341         return 1;
1342 }
1343
1344 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1345 {
1346         if (!nfs4_state_clear_reclaim_reboot(clp))
1347                 return;
1348         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1349 }
1350
1351 static void nfs_delegation_clear_all(struct nfs_client *clp)
1352 {
1353         nfs_delegation_mark_reclaim(clp);
1354         nfs_delegation_reap_unclaimed(clp);
1355 }
1356
1357 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1358 {
1359         nfs_delegation_clear_all(clp);
1360         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1361 }
1362
1363 static void nfs4_warn_keyexpired(const char *s)
1364 {
1365         printk_ratelimited(KERN_WARNING "Error: state manager"
1366                         " encountered RPCSEC_GSS session"
1367                         " expired against NFSv4 server %s.\n",
1368                         s);
1369 }
1370
1371 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1372 {
1373         switch (error) {
1374                 case 0:
1375                         break;
1376                 case -NFS4ERR_CB_PATH_DOWN:
1377                         nfs_handle_cb_pathdown(clp);
1378                         break;
1379                 case -NFS4ERR_NO_GRACE:
1380                         nfs4_state_end_reclaim_reboot(clp);
1381                         break;
1382                 case -NFS4ERR_STALE_CLIENTID:
1383                 case -NFS4ERR_LEASE_MOVED:
1384                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1385                         nfs4_state_clear_reclaim_reboot(clp);
1386                         nfs4_state_start_reclaim_reboot(clp);
1387                         break;
1388                 case -NFS4ERR_EXPIRED:
1389                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1390                         nfs4_state_start_reclaim_nograce(clp);
1391                         break;
1392                 case -NFS4ERR_BADSESSION:
1393                 case -NFS4ERR_BADSLOT:
1394                 case -NFS4ERR_BAD_HIGH_SLOT:
1395                 case -NFS4ERR_DEADSESSION:
1396                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1397                 case -NFS4ERR_SEQ_FALSE_RETRY:
1398                 case -NFS4ERR_SEQ_MISORDERED:
1399                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1400                         /* Zero session reset errors */
1401                         break;
1402                 case -EKEYEXPIRED:
1403                         /* Nothing we can do */
1404                         nfs4_warn_keyexpired(clp->cl_hostname);
1405                         break;
1406                 default:
1407                         return error;
1408         }
1409         return 0;
1410 }
1411
1412 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1413 {
1414         struct nfs4_state_owner *sp;
1415         struct nfs_server *server;
1416         struct rb_node *pos;
1417         int status = 0;
1418
1419 restart:
1420         rcu_read_lock();
1421         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1422                 nfs4_purge_state_owners(server);
1423                 spin_lock(&clp->cl_lock);
1424                 for (pos = rb_first(&server->state_owners);
1425                      pos != NULL;
1426                      pos = rb_next(pos)) {
1427                         sp = rb_entry(pos,
1428                                 struct nfs4_state_owner, so_server_node);
1429                         if (!test_and_clear_bit(ops->owner_flag_bit,
1430                                                         &sp->so_flags))
1431                                 continue;
1432                         atomic_inc(&sp->so_count);
1433                         spin_unlock(&clp->cl_lock);
1434                         rcu_read_unlock();
1435
1436                         status = nfs4_reclaim_open_state(sp, ops);
1437                         if (status < 0) {
1438                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1439                                 nfs4_put_state_owner(sp);
1440                                 return nfs4_recovery_handle_error(clp, status);
1441                         }
1442
1443                         nfs4_put_state_owner(sp);
1444                         goto restart;
1445                 }
1446                 spin_unlock(&clp->cl_lock);
1447         }
1448         rcu_read_unlock();
1449         return status;
1450 }
1451
1452 static int nfs4_check_lease(struct nfs_client *clp)
1453 {
1454         struct rpc_cred *cred;
1455         const struct nfs4_state_maintenance_ops *ops =
1456                 clp->cl_mvops->state_renewal_ops;
1457         int status;
1458
1459         /* Is the client already known to have an expired lease? */
1460         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1461                 return 0;
1462         spin_lock(&clp->cl_lock);
1463         cred = ops->get_state_renewal_cred_locked(clp);
1464         spin_unlock(&clp->cl_lock);
1465         if (cred == NULL) {
1466                 cred = nfs4_get_setclientid_cred(clp);
1467                 status = -ENOKEY;
1468                 if (cred == NULL)
1469                         goto out;
1470         }
1471         status = ops->renew_lease(clp, cred);
1472         put_rpccred(cred);
1473 out:
1474         return nfs4_recovery_handle_error(clp, status);
1475 }
1476
1477 static int nfs4_reclaim_lease(struct nfs_client *clp)
1478 {
1479         struct rpc_cred *cred;
1480         const struct nfs4_state_recovery_ops *ops =
1481                 clp->cl_mvops->reboot_recovery_ops;
1482         int status = -ENOENT;
1483
1484         cred = ops->get_clid_cred(clp);
1485         if (cred != NULL) {
1486                 status = ops->establish_clid(clp, cred);
1487                 put_rpccred(cred);
1488                 /* Handle case where the user hasn't set up machine creds */
1489                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1490                         nfs4_clear_machine_cred(clp);
1491                         status = -EAGAIN;
1492                 }
1493                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1494                         status = -EPROTONOSUPPORT;
1495         }
1496         return status;
1497 }
1498
1499 #ifdef CONFIG_NFS_V4_1
1500 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1501 {
1502         struct nfs_client *clp = session->clp;
1503
1504         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1505         nfs4_schedule_lease_recovery(clp);
1506 }
1507 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1508
1509 void nfs41_handle_recall_slot(struct nfs_client *clp)
1510 {
1511         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1512         nfs4_schedule_state_manager(clp);
1513 }
1514
1515 static void nfs4_reset_all_state(struct nfs_client *clp)
1516 {
1517         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1518                 clp->cl_boot_time = CURRENT_TIME;
1519                 nfs4_state_start_reclaim_nograce(clp);
1520                 nfs4_schedule_state_manager(clp);
1521         }
1522 }
1523
1524 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1525 {
1526         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1527                 nfs4_state_start_reclaim_reboot(clp);
1528                 nfs4_schedule_state_manager(clp);
1529         }
1530 }
1531
1532 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1533 {
1534         /* Temporary */
1535         nfs4_reset_all_state(clp);
1536 }
1537
1538 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1539 {
1540         /* This will need to handle layouts too */
1541         nfs_expire_all_delegations(clp);
1542 }
1543
1544 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1545 {
1546         nfs_expire_all_delegations(clp);
1547         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1548                 nfs4_schedule_state_manager(clp);
1549 }
1550
1551 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1552 {
1553         if (!flags)
1554                 return;
1555         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1556                 nfs41_handle_server_reboot(clp);
1557         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1558                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1559                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1560                             SEQ4_STATUS_LEASE_MOVED))
1561                 nfs41_handle_state_revoked(clp);
1562         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1563                 nfs41_handle_recallable_state_revoked(clp);
1564         if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1565                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1566                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1567                 nfs41_handle_cb_path_down(clp);
1568 }
1569
1570 static int nfs4_reset_session(struct nfs_client *clp)
1571 {
1572         int status;
1573
1574         nfs4_begin_drain_session(clp);
1575         status = nfs4_proc_destroy_session(clp->cl_session);
1576         if (status && status != -NFS4ERR_BADSESSION &&
1577             status != -NFS4ERR_DEADSESSION) {
1578                 status = nfs4_recovery_handle_error(clp, status);
1579                 goto out;
1580         }
1581
1582         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1583         status = nfs4_proc_create_session(clp);
1584         if (status) {
1585                 status = nfs4_recovery_handle_error(clp, status);
1586                 goto out;
1587         }
1588         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1589         /* create_session negotiated new slot table */
1590         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1591
1592          /* Let the state manager reestablish state */
1593         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1594                 nfs41_setup_state_renewal(clp);
1595 out:
1596         return status;
1597 }
1598
1599 static int nfs4_recall_slot(struct nfs_client *clp)
1600 {
1601         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1602         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1603         struct nfs4_slot *new, *old;
1604         int i;
1605
1606         nfs4_begin_drain_session(clp);
1607         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1608                       GFP_NOFS);
1609         if (!new)
1610                 return -ENOMEM;
1611
1612         spin_lock(&fc_tbl->slot_tbl_lock);
1613         for (i = 0; i < fc_tbl->target_max_slots; i++)
1614                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1615         old = fc_tbl->slots;
1616         fc_tbl->slots = new;
1617         fc_tbl->max_slots = fc_tbl->target_max_slots;
1618         fc_tbl->target_max_slots = 0;
1619         fc_attrs->max_reqs = fc_tbl->max_slots;
1620         spin_unlock(&fc_tbl->slot_tbl_lock);
1621
1622         kfree(old);
1623         nfs4_end_drain_session(clp);
1624         return 0;
1625 }
1626
1627 #else /* CONFIG_NFS_V4_1 */
1628 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1629 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1630 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1631 #endif /* CONFIG_NFS_V4_1 */
1632
1633 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1634  * on EXCHANGE_ID for v4.1
1635  */
1636 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1637 {
1638         switch (status) {
1639         case -NFS4ERR_CLID_INUSE:
1640         case -NFS4ERR_STALE_CLIENTID:
1641                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1642                 break;
1643         case -NFS4ERR_DELAY:
1644         case -ETIMEDOUT:
1645         case -EAGAIN:
1646                 ssleep(1);
1647                 break;
1648
1649         case -EKEYEXPIRED:
1650                 nfs4_warn_keyexpired(clp->cl_hostname);
1651         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1652                                  * in nfs4_exchange_id */
1653         default:
1654                 return;
1655         }
1656         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1657 }
1658
1659 static void nfs4_state_manager(struct nfs_client *clp)
1660 {
1661         int status = 0;
1662
1663         /* Ensure exclusive access to NFSv4 state */
1664         do {
1665                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1666                         /* We're going to have to re-establish a clientid */
1667                         status = nfs4_reclaim_lease(clp);
1668                         if (status) {
1669                                 nfs4_set_lease_expired(clp, status);
1670                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1671                                                         &clp->cl_state))
1672                                         continue;
1673                                 if (clp->cl_cons_state ==
1674                                                         NFS_CS_SESSION_INITING)
1675                                         nfs_mark_client_ready(clp, status);
1676                                 goto out_error;
1677                         }
1678                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1679
1680                         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1681                                                &clp->cl_state))
1682                                 nfs4_state_start_reclaim_nograce(clp);
1683                         else
1684                                 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1685                                         &clp->cl_state);
1686
1687                         pnfs_destroy_all_layouts(clp);
1688                 }
1689
1690                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1691                         status = nfs4_check_lease(clp);
1692                         if (status < 0)
1693                                 goto out_error;
1694                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1695                                 continue;
1696                 }
1697
1698                 /* Initialize or reset the session */
1699                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1700                    && nfs4_has_session(clp)) {
1701                         status = nfs4_reset_session(clp);
1702                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1703                                 continue;
1704                         if (status < 0)
1705                                 goto out_error;
1706                 }
1707
1708                 /* First recover reboot state... */
1709                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1710                         status = nfs4_do_reclaim(clp,
1711                                 clp->cl_mvops->reboot_recovery_ops);
1712                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1713                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1714                                 continue;
1715                         nfs4_state_end_reclaim_reboot(clp);
1716                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1717                                 continue;
1718                         if (status < 0)
1719                                 goto out_error;
1720                 }
1721
1722                 /* Now recover expired state... */
1723                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1724                         status = nfs4_do_reclaim(clp,
1725                                 clp->cl_mvops->nograce_recovery_ops);
1726                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1727                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1728                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1729                                 continue;
1730                         if (status < 0)
1731                                 goto out_error;
1732                 }
1733
1734                 nfs4_end_drain_session(clp);
1735                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1736                         nfs_client_return_marked_delegations(clp);
1737                         continue;
1738                 }
1739                 /* Recall session slots */
1740                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1741                    && nfs4_has_session(clp)) {
1742                         status = nfs4_recall_slot(clp);
1743                         if (status < 0)
1744                                 goto out_error;
1745                         continue;
1746                 }
1747
1748
1749                 nfs4_clear_state_manager_bit(clp);
1750                 /* Did we race with an attempt to give us more work? */
1751                 if (clp->cl_state == 0)
1752                         break;
1753                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1754                         break;
1755         } while (atomic_read(&clp->cl_count) > 1);
1756         return;
1757 out_error:
1758         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1759                         " with error %d\n", clp->cl_hostname, -status);
1760         nfs4_end_drain_session(clp);
1761         nfs4_clear_state_manager_bit(clp);
1762 }
1763
1764 static int nfs4_run_state_manager(void *ptr)
1765 {
1766         struct nfs_client *clp = ptr;
1767
1768         allow_signal(SIGKILL);
1769         nfs4_state_manager(clp);
1770         nfs_put_client(clp);
1771         module_put_and_exit(0);
1772         return 0;
1773 }
1774
1775 /*
1776  * Local variables:
1777  *  c-basic-offset: 8
1778  * End:
1779  */