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
  53 #include "nfs4_fs.h"
  54 #include "callback.h"
  55 #include "delegation.h"
  56 #include "internal.h"
  57 #include "pnfs.h"
  58
  59 #define OPENOWNER_POOL_SIZE     8
  60
  61 const nfs4_stateid zero_stateid;
  62
  63 static LIST_HEAD(nfs4_clientid_list);
  64
  65 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  66 {
  67         struct nfs4_setclientid_res clid;
  68         unsigned short port;
  69         int status;
  70
  71         port = nfs_callback_tcpport;
  72         if (clp->cl_addr.ss_family == AF_INET6)
  73                 port = nfs_callback_tcpport6;
  74
  75         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  76         if (status != 0)
  77                 goto out;
  78         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  79         if (status != 0)
  80                 goto out;
  81         clp->cl_clientid = clid.clientid;
  82         nfs4_schedule_state_renewal(clp);
  83 out:
  84         return status;
  85 }
  86
  87 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  88 {
  89         struct rpc_cred *cred = NULL;
  90
  91         if (clp->cl_machine_cred != NULL)
  92                 cred = get_rpccred(clp->cl_machine_cred);
  93         return cred;
  94 }
  95
  96 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  97 {
  98         struct rpc_cred *cred;
  99
 100         spin_lock(&clp->cl_lock);
 101         cred = clp->cl_machine_cred;
 102         clp->cl_machine_cred = NULL;
 103         spin_unlock(&clp->cl_lock);
 104         if (cred != NULL)
 105                 put_rpccred(cred);
 106 }
 107
 108 static struct rpc_cred *
 109 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 110 {
 111         struct rpc_cred *cred = NULL;
 112         struct nfs4_state_owner *sp;
 113         struct rb_node *pos;
 114
 115         for (pos = rb_first(&server->state_owners);
 116              pos != NULL;
 117              pos = rb_next(pos)) {
 118                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 119                 if (list_empty(&sp->so_states))
 120                         continue;
 121                 cred = get_rpccred(sp->so_cred);
 122                 break;
 123         }
 124         return cred;
 125 }
 126
 127 /**
 128  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 129  * @clp: client state handle
 130  *
 131  * Returns an rpc_cred with reference count bumped, or NULL.
 132  * Caller must hold clp->cl_lock.
 133  */
 134 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 135 {
 136         struct rpc_cred *cred = NULL;
 137         struct nfs_server *server;
 138
 139         rcu_read_lock();
 140         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 141                 cred = nfs4_get_renew_cred_server_locked(server);
 142                 if (cred != NULL)
 143                         break;
 144         }
 145         rcu_read_unlock();
 146         return cred;
 147 }
 148
 149 #if defined(CONFIG_NFS_V4_1)
 150
 151 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 152 {
 153         int status;
 154         struct nfs_fsinfo fsinfo;
 155
 156         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 157                 nfs4_schedule_state_renewal(clp);
 158                 return 0;
 159         }
 160
 161         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 162         if (status == 0) {
 163                 /* Update lease time and schedule renewal */
 164                 spin_lock(&clp->cl_lock);
 165                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 166                 clp->cl_last_renewal = jiffies;
 167                 spin_unlock(&clp->cl_lock);
 168
 169                 nfs4_schedule_state_renewal(clp);
 170         }
 171
 172         return status;
 173 }
 174
 175 /*
 176  * Back channel returns NFS4ERR_DELAY for new requests when
 177  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 178  * is ended.
 179  */
 180 static void nfs4_end_drain_session(struct nfs_client *clp)
 181 {
 182         struct nfs4_session *ses = clp->cl_session;
 183         int max_slots;
 184
 185         if (ses == NULL)
 186                 return;
 187         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 188                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 189                 max_slots = ses->fc_slot_table.max_slots;
 190                 while (max_slots--) {
 191                         struct rpc_task *task;
 192
 193                         task = rpc_wake_up_next(&ses->fc_slot_table.
 194                                                 slot_tbl_waitq);
 195                         if (!task)
 196                                 break;
 197                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 198                 }
 199                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 200         }
 201 }
 202
 203 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 204 {
 205         spin_lock(&tbl->slot_tbl_lock);
 206         if (tbl->highest_used_slotid != -1) {
 207                 INIT_COMPLETION(tbl->complete);
 208                 spin_unlock(&tbl->slot_tbl_lock);
 209                 return wait_for_completion_interruptible(&tbl->complete);
 210         }
 211         spin_unlock(&tbl->slot_tbl_lock);
 212         return 0;
 213 }
 214
 215 static int nfs4_begin_drain_session(struct nfs_client *clp)
 216 {
 217         struct nfs4_session *ses = clp->cl_session;
 218         int ret = 0;
 219
 220         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 221         /* back channel */
 222         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 223         if (ret)
 224                 return ret;
 225         /* fore channel */
 226         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 227 }
 228
 229 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 230 {
 231         int status;
 232
 233         nfs4_begin_drain_session(clp);
 234         status = nfs4_proc_exchange_id(clp, cred);
 235         if (status != 0)
 236                 goto out;
 237         status = nfs4_proc_create_session(clp);
 238         if (status != 0)
 239                 goto out;
 240         nfs41_setup_state_renewal(clp);
 241         nfs_mark_client_ready(clp, NFS_CS_READY);
 242 out:
 243         return status;
 244 }
 245
 246 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 247 {
 248         struct rpc_cred *cred;
 249
 250         spin_lock(&clp->cl_lock);
 251         cred = nfs4_get_machine_cred_locked(clp);
 252         spin_unlock(&clp->cl_lock);
 253         return cred;
 254 }
 255
 256 #endif /* CONFIG_NFS_V4_1 */
 257
 258 static struct rpc_cred *
 259 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 260 {
 261         struct nfs_client *clp = server->nfs_client;
 262         struct rpc_cred *cred = NULL;
 263         struct nfs4_state_owner *sp;
 264         struct rb_node *pos;
 265
 266         spin_lock(&clp->cl_lock);
 267         pos = rb_first(&server->state_owners);
 268         if (pos != NULL) {
 269                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 270                 cred = get_rpccred(sp->so_cred);
 271         }
 272         spin_unlock(&clp->cl_lock);
 273         return cred;
 274 }
 275
 276 /**
 277  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 278  * @clp: client state handle
 279  *
 280  * Returns an rpc_cred with reference count bumped, or NULL.
 281  */
 282 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 283 {
 284         struct nfs_server *server;
 285         struct rpc_cred *cred;
 286
 287         spin_lock(&clp->cl_lock);
 288         cred = nfs4_get_machine_cred_locked(clp);
 289         spin_unlock(&clp->cl_lock);
 290         if (cred != NULL)
 291                 goto out;
 292
 293         rcu_read_lock();
 294         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 295                 cred = nfs4_get_setclientid_cred_server(server);
 296                 if (cred != NULL)
 297                         break;
 298         }
 299         rcu_read_unlock();
 300
 301 out:
 302         return cred;
 303 }
 304
 305 static void nfs_alloc_unique_id_locked(struct rb_root *root,
 306                                        struct nfs_unique_id *new,
 307                                        __u64 minval, int maxbits)
 308 {
 309         struct rb_node **p, *parent;
 310         struct nfs_unique_id *pos;
 311         __u64 mask = ~0ULL;
 312
 313         if (maxbits < 64)
 314                 mask = (1ULL << maxbits) - 1ULL;
 315
 316         /* Ensure distribution is more or less flat */
 317         get_random_bytes(&new->id, sizeof(new->id));
 318         new->id &= mask;
 319         if (new->id < minval)
 320                 new->id += minval;
 321 retry:
 322         p = &root->rb_node;
 323         parent = NULL;
 324
 325         while (*p != NULL) {
 326                 parent = *p;
 327                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 328
 329                 if (new->id < pos->id)
 330                         p = &(*p)->rb_left;
 331                 else if (new->id > pos->id)
 332                         p = &(*p)->rb_right;
 333                 else
 334                         goto id_exists;
 335         }
 336         rb_link_node(&new->rb_node, parent, p);
 337         rb_insert_color(&new->rb_node, root);
 338         return;
 339 id_exists:
 340         for (;;) {
 341                 new->id++;
 342                 if (new->id < minval || (new->id & mask) != new->id) {
 343                         new->id = minval;
 344                         break;
 345                 }
 346                 parent = rb_next(parent);
 347                 if (parent == NULL)
 348                         break;
 349                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 350                 if (new->id < pos->id)
 351                         break;
 352         }
 353         goto retry;
 354 }
 355
 356 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 357 {
 358         rb_erase(&id->rb_node, root);
 359 }
 360
 361 static struct nfs4_state_owner *
 362 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 363 {
 364         struct rb_node **p = &server->state_owners.rb_node,
 365                        *parent = NULL;
 366         struct nfs4_state_owner *sp, *res = NULL;
 367
 368         while (*p != NULL) {
 369                 parent = *p;
 370                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 371
 372                 if (server < sp->so_server) {
 373                         p = &parent->rb_left;
 374                         continue;
 375                 }
 376                 if (server > sp->so_server) {
 377                         p = &parent->rb_right;
 378                         continue;
 379                 }
 380                 if (cred < sp->so_cred)
 381                         p = &parent->rb_left;
 382                 else if (cred > sp->so_cred)
 383                         p = &parent->rb_right;
 384                 else {
 385                         atomic_inc(&sp->so_count);
 386                         res = sp;
 387                         break;
 388                 }
 389         }
 390         return res;
 391 }
 392
 393 static struct nfs4_state_owner *
 394 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 395 {
 396         struct nfs_server *server = new->so_server;
 397         struct rb_node **p = &server->state_owners.rb_node,
 398                        *parent = NULL;
 399         struct nfs4_state_owner *sp;
 400
 401         while (*p != NULL) {
 402                 parent = *p;
 403                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 404
 405                 if (new->so_cred < sp->so_cred)
 406                         p = &parent->rb_left;
 407                 else if (new->so_cred > sp->so_cred)
 408                         p = &parent->rb_right;
 409                 else {
 410                         atomic_inc(&sp->so_count);
 411                         return sp;
 412                 }
 413         }
 414         nfs_alloc_unique_id_locked(&server->openowner_id,
 415                                         &new->so_owner_id, 1, 64);
 416         rb_link_node(&new->so_server_node, parent, p);
 417         rb_insert_color(&new->so_server_node, &server->state_owners);
 418         return new;
 419 }
 420
 421 static void
 422 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 423 {
 424         struct nfs_server *server = sp->so_server;
 425
 426         if (!RB_EMPTY_NODE(&sp->so_server_node))
 427                 rb_erase(&sp->so_server_node, &server->state_owners);
 428         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
 429 }
 430
 431 /*
 432  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 433  * create a new state_owner.
 434  *
 435  */
 436 static struct nfs4_state_owner *
 437 nfs4_alloc_state_owner(void)
 438 {
 439         struct nfs4_state_owner *sp;
 440
 441         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 442         if (!sp)
 443                 return NULL;
 444         spin_lock_init(&sp->so_lock);
 445         INIT_LIST_HEAD(&sp->so_states);
 446         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 447         sp->so_seqid.sequence = &sp->so_sequence;
 448         spin_lock_init(&sp->so_sequence.lock);
 449         INIT_LIST_HEAD(&sp->so_sequence.list);
 450         atomic_set(&sp->so_count, 1);
 451         return sp;
 452 }
 453
 454 static void
 455 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 456 {
 457         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 458                 struct nfs_server *server = sp->so_server;
 459                 struct nfs_client *clp = server->nfs_client;
 460
 461                 spin_lock(&clp->cl_lock);
 462                 rb_erase(&sp->so_server_node, &server->state_owners);
 463                 RB_CLEAR_NODE(&sp->so_server_node);
 464                 spin_unlock(&clp->cl_lock);
 465         }
 466 }
 467
 468 /**
 469  * nfs4_get_state_owner - Look up a state owner given a credential
 470  * @server: nfs_server to search
 471  * @cred: RPC credential to match
 472  *
 473  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 474  */
 475 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 476                                               struct rpc_cred *cred)
 477 {
 478         struct nfs_client *clp = server->nfs_client;
 479         struct nfs4_state_owner *sp, *new;
 480
 481         spin_lock(&clp->cl_lock);
 482         sp = nfs4_find_state_owner_locked(server, cred);
 483         spin_unlock(&clp->cl_lock);
 484         if (sp != NULL)
 485                 return sp;
 486         new = nfs4_alloc_state_owner();
 487         if (new == NULL)
 488                 return NULL;
 489         new->so_server = server;
 490         new->so_cred = cred;
 491         spin_lock(&clp->cl_lock);
 492         sp = nfs4_insert_state_owner_locked(new);
 493         spin_unlock(&clp->cl_lock);
 494         if (sp == new)
 495                 get_rpccred(cred);
 496         else {
 497                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 498                 kfree(new);
 499         }
 500         return sp;
 501 }
 502
 503 /**
 504  * nfs4_put_state_owner - Release a nfs4_state_owner
 505  * @sp: state owner data to release
 506  *
 507  */
 508 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 509 {
 510         struct nfs_client *clp = sp->so_server->nfs_client;
 511         struct rpc_cred *cred = sp->so_cred;
 512
 513         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 514                 return;
 515         nfs4_remove_state_owner_locked(sp);
 516         spin_unlock(&clp->cl_lock);
 517         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 518         put_rpccred(cred);
 519         kfree(sp);
 520 }
 521
 522 static struct nfs4_state *
 523 nfs4_alloc_open_state(void)
 524 {
 525         struct nfs4_state *state;
 526
 527         state = kzalloc(sizeof(*state), GFP_NOFS);
 528         if (!state)
 529                 return NULL;
 530         atomic_set(&state->count, 1);
 531         INIT_LIST_HEAD(&state->lock_states);
 532         spin_lock_init(&state->state_lock);
 533         seqlock_init(&state->seqlock);
 534         return state;
 535 }
 536
 537 void
 538 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 539 {
 540         if (state->state == fmode)
 541                 return;
 542         /* NB! List reordering - see the reclaim code for why.  */
 543         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 544                 if (fmode & FMODE_WRITE)
 545                         list_move(&state->open_states, &state->owner->so_states);
 546                 else
 547                         list_move_tail(&state->open_states, &state->owner->so_states);
 548         }
 549         state->state = fmode;
 550 }
 551
 552 static struct nfs4_state *
 553 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 554 {
 555         struct nfs_inode *nfsi = NFS_I(inode);
 556         struct nfs4_state *state;
 557
 558         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 559                 if (state->owner != owner)
 560                         continue;
 561                 if (atomic_inc_not_zero(&state->count))
 562                         return state;
 563         }
 564         return NULL;
 565 }
 566
 567 static void
 568 nfs4_free_open_state(struct nfs4_state *state)
 569 {
 570         kfree(state);
 571 }
 572
 573 struct nfs4_state *
 574 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 575 {
 576         struct nfs4_state *state, *new;
 577         struct nfs_inode *nfsi = NFS_I(inode);
 578
 579         spin_lock(&inode->i_lock);
 580         state = __nfs4_find_state_byowner(inode, owner);
 581         spin_unlock(&inode->i_lock);
 582         if (state)
 583                 goto out;
 584         new = nfs4_alloc_open_state();
 585         spin_lock(&owner->so_lock);
 586         spin_lock(&inode->i_lock);
 587         state = __nfs4_find_state_byowner(inode, owner);
 588         if (state == NULL && new != NULL) {
 589                 state = new;
 590                 state->owner = owner;
 591                 atomic_inc(&owner->so_count);
 592                 list_add(&state->inode_states, &nfsi->open_states);
 593                 state->inode = igrab(inode);
 594                 spin_unlock(&inode->i_lock);
 595                 /* Note: The reclaim code dictates that we add stateless
 596                  * and read-only stateids to the end of the list */
 597                 list_add_tail(&state->open_states, &owner->so_states);
 598                 spin_unlock(&owner->so_lock);
 599         } else {
 600                 spin_unlock(&inode->i_lock);
 601                 spin_unlock(&owner->so_lock);
 602                 if (new)
 603                         nfs4_free_open_state(new);
 604         }
 605 out:
 606         return state;
 607 }
 608
 609 void nfs4_put_open_state(struct nfs4_state *state)
 610 {
 611         struct inode *inode = state->inode;
 612         struct nfs4_state_owner *owner = state->owner;
 613
 614         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 615                 return;
 616         spin_lock(&inode->i_lock);
 617         list_del(&state->inode_states);
 618         list_del(&state->open_states);
 619         spin_unlock(&inode->i_lock);
 620         spin_unlock(&owner->so_lock);
 621         iput(inode);
 622         nfs4_free_open_state(state);
 623         nfs4_put_state_owner(owner);
 624 }
 625
 626 /*
 627  * Close the current file.
 628  */
 629 static void __nfs4_close(struct path *path, struct nfs4_state *state,
 630                 fmode_t fmode, gfp_t gfp_mask, int wait)
 631 {
 632         struct nfs4_state_owner *owner = state->owner;
 633         int call_close = 0;
 634         fmode_t newstate;
 635
 636         atomic_inc(&owner->so_count);
 637         /* Protect against nfs4_find_state() */
 638         spin_lock(&owner->so_lock);
 639         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 640                 case FMODE_READ:
 641                         state->n_rdonly--;
 642                         break;
 643                 case FMODE_WRITE:
 644                         state->n_wronly--;
 645                         break;
 646                 case FMODE_READ|FMODE_WRITE:
 647                         state->n_rdwr--;
 648         }
 649         newstate = FMODE_READ|FMODE_WRITE;
 650         if (state->n_rdwr == 0) {
 651                 if (state->n_rdonly == 0) {
 652                         newstate &= ~FMODE_READ;
 653                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 654                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 655                 }
 656                 if (state->n_wronly == 0) {
 657                         newstate &= ~FMODE_WRITE;
 658                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 659                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 660                 }
 661                 if (newstate == 0)
 662                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 663         }
 664         nfs4_state_set_mode_locked(state, newstate);
 665         spin_unlock(&owner->so_lock);
 666
 667         if (!call_close) {
 668                 nfs4_put_open_state(state);
 669                 nfs4_put_state_owner(owner);
 670         } else {
 671                 bool roc = pnfs_roc(state->inode);
 672
 673                 nfs4_do_close(path, state, gfp_mask, wait, roc);
 674         }
 675 }
 676
 677 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 678 {
 679         __nfs4_close(path, state, fmode, GFP_NOFS, 0);
 680 }
 681
 682 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 683 {
 684         __nfs4_close(path, state, fmode, GFP_KERNEL, 1);
 685 }
 686
 687 /*
 688  * Search the state->lock_states for an existing lock_owner
 689  * that is compatible with current->files
 690  */
 691 static struct nfs4_lock_state *
 692 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 693 {
 694         struct nfs4_lock_state *pos;
 695         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 696                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 697                         continue;
 698                 switch (pos->ls_owner.lo_type) {
 699                 case NFS4_POSIX_LOCK_TYPE:
 700                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 701                                 continue;
 702                         break;
 703                 case NFS4_FLOCK_LOCK_TYPE:
 704                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 705                                 continue;
 706                 }
 707                 atomic_inc(&pos->ls_count);
 708                 return pos;
 709         }
 710         return NULL;
 711 }
 712
 713 /*
 714  * Return a compatible lock_state. If no initialized lock_state structure
 715  * exists, return an uninitialized one.
 716  *
 717  */
 718 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)
 719 {
 720         struct nfs4_lock_state *lsp;
 721         struct nfs_server *server = state->owner->so_server;
 722         struct nfs_client *clp = server->nfs_client;
 723
 724         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 725         if (lsp == NULL)
 726                 return NULL;
 727         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 728         spin_lock_init(&lsp->ls_sequence.lock);
 729         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 730         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 731         atomic_set(&lsp->ls_count, 1);
 732         lsp->ls_state = state;
 733         lsp->ls_owner.lo_type = type;
 734         switch (lsp->ls_owner.lo_type) {
 735         case NFS4_FLOCK_LOCK_TYPE:
 736                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 737                 break;
 738         case NFS4_POSIX_LOCK_TYPE:
 739                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 740                 break;
 741         default:
 742                 kfree(lsp);
 743                 return NULL;
 744         }
 745         spin_lock(&clp->cl_lock);
 746         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
 747         spin_unlock(&clp->cl_lock);
 748         INIT_LIST_HEAD(&lsp->ls_locks);
 749         return lsp;
 750 }
 751
 752 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 753 {
 754         struct nfs_server *server = lsp->ls_state->owner->so_server;
 755         struct nfs_client *clp = server->nfs_client;
 756
 757         spin_lock(&clp->cl_lock);
 758         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
 759         spin_unlock(&clp->cl_lock);
 760         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 761         kfree(lsp);
 762 }
 763
 764 /*
 765  * Return a compatible lock_state. If no initialized lock_state structure
 766  * exists, return an uninitialized one.
 767  *
 768  */
 769 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 770 {
 771         struct nfs4_lock_state *lsp, *new = NULL;
 772
 773         for(;;) {
 774                 spin_lock(&state->state_lock);
 775                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 776                 if (lsp != NULL)
 777                         break;
 778                 if (new != NULL) {
 779                         list_add(&new->ls_locks, &state->lock_states);
 780                         set_bit(LK_STATE_IN_USE, &state->flags);
 781                         lsp = new;
 782                         new = NULL;
 783                         break;
 784                 }
 785                 spin_unlock(&state->state_lock);
 786                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 787                 if (new == NULL)
 788                         return NULL;
 789         }
 790         spin_unlock(&state->state_lock);
 791         if (new != NULL)
 792                 nfs4_free_lock_state(new);
 793         return lsp;
 794 }
 795
 796 /*
 797  * Release reference to lock_state, and free it if we see that
 798  * it is no longer in use
 799  */
 800 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 801 {
 802         struct nfs4_state *state;
 803
 804         if (lsp == NULL)
 805                 return;
 806         state = lsp->ls_state;
 807         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 808                 return;
 809         list_del(&lsp->ls_locks);
 810         if (list_empty(&state->lock_states))
 811                 clear_bit(LK_STATE_IN_USE, &state->flags);
 812         spin_unlock(&state->state_lock);
 813         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 814                 nfs4_release_lockowner(lsp);
 815         nfs4_free_lock_state(lsp);
 816 }
 817
 818 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 819 {
 820         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 821
 822         dst->fl_u.nfs4_fl.owner = lsp;
 823         atomic_inc(&lsp->ls_count);
 824 }
 825
 826 static void nfs4_fl_release_lock(struct file_lock *fl)
 827 {
 828         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 829 }
 830
 831 static const struct file_lock_operations nfs4_fl_lock_ops = {
 832         .fl_copy_lock = nfs4_fl_copy_lock,
 833         .fl_release_private = nfs4_fl_release_lock,
 834 };
 835
 836 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 837 {
 838         struct nfs4_lock_state *lsp;
 839
 840         if (fl->fl_ops != NULL)
 841                 return 0;
 842         if (fl->fl_flags & FL_POSIX)
 843                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 844         else if (fl->fl_flags & FL_FLOCK)
 845                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 846         else
 847                 return -EINVAL;
 848         if (lsp == NULL)
 849                 return -ENOMEM;
 850         fl->fl_u.nfs4_fl.owner = lsp;
 851         fl->fl_ops = &nfs4_fl_lock_ops;
 852         return 0;
 853 }
 854
 855 /*
 856  * Byte-range lock aware utility to initialize the stateid of read/write
 857  * requests.
 858  */
 859 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 860 {
 861         struct nfs4_lock_state *lsp;
 862         int seq;
 863
 864         do {
 865                 seq = read_seqbegin(&state->seqlock);
 866                 memcpy(dst, &state->stateid, sizeof(*dst));
 867         } while (read_seqretry(&state->seqlock, seq));
 868         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 869                 return;
 870
 871         spin_lock(&state->state_lock);
 872         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 873         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 874                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 875         spin_unlock(&state->state_lock);
 876         nfs4_put_lock_state(lsp);
 877 }
 878
 879 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 880 {
 881         struct nfs_seqid *new;
 882
 883         new = kmalloc(sizeof(*new), gfp_mask);
 884         if (new != NULL) {
 885                 new->sequence = counter;
 886                 INIT_LIST_HEAD(&new->list);
 887         }
 888         return new;
 889 }
 890
 891 void nfs_release_seqid(struct nfs_seqid *seqid)
 892 {
 893         if (!list_empty(&seqid->list)) {
 894                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 895
 896                 spin_lock(&sequence->lock);
 897                 list_del_init(&seqid->list);
 898                 spin_unlock(&sequence->lock);
 899                 rpc_wake_up(&sequence->wait);
 900         }
 901 }
 902
 903 void nfs_free_seqid(struct nfs_seqid *seqid)
 904 {
 905         nfs_release_seqid(seqid);
 906         kfree(seqid);
 907 }
 908
 909 /*
 910  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 911  * failed with a seqid incrementing error -
 912  * see comments nfs_fs.h:seqid_mutating_error()
 913  */
 914 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 915 {
 916         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 917         switch (status) {
 918                 case 0:
 919                         break;
 920                 case -NFS4ERR_BAD_SEQID:
 921                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 922                                 return;
 923                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 924                                         " sequence-id error on an"
 925                                         " unconfirmed sequence %p!\n",
 926                                         seqid->sequence);
 927                 case -NFS4ERR_STALE_CLIENTID:
 928                 case -NFS4ERR_STALE_STATEID:
 929                 case -NFS4ERR_BAD_STATEID:
 930                 case -NFS4ERR_BADXDR:
 931                 case -NFS4ERR_RESOURCE:
 932                 case -NFS4ERR_NOFILEHANDLE:
 933                         /* Non-seqid mutating errors */
 934                         return;
 935         };
 936         /*
 937          * Note: no locking needed as we are guaranteed to be first
 938          * on the sequence list
 939          */
 940         seqid->sequence->counter++;
 941 }
 942
 943 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 944 {
 945         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 946                                         struct nfs4_state_owner, so_seqid);
 947         struct nfs_server *server = sp->so_server;
 948
 949         if (status == -NFS4ERR_BAD_SEQID)
 950                 nfs4_drop_state_owner(sp);
 951         if (!nfs4_has_session(server->nfs_client))
 952                 nfs_increment_seqid(status, seqid);
 953 }
 954
 955 /*
 956  * Increment the seqid if the LOCK/LOCKU succeeded, or
 957  * failed with a seqid incrementing error -
 958  * see comments nfs_fs.h:seqid_mutating_error()
 959  */
 960 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 961 {
 962         nfs_increment_seqid(status, seqid);
 963 }
 964
 965 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 966 {
 967         struct rpc_sequence *sequence = seqid->sequence->sequence;
 968         int status = 0;
 969
 970         spin_lock(&sequence->lock);
 971         if (list_empty(&seqid->list))
 972                 list_add_tail(&seqid->list, &sequence->list);
 973         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 974                 goto unlock;
 975         rpc_sleep_on(&sequence->wait, task, NULL);
 976         status = -EAGAIN;
 977 unlock:
 978         spin_unlock(&sequence->lock);
 979         return status;
 980 }
 981
 982 static int nfs4_run_state_manager(void *);
 983
 984 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 985 {
 986         smp_mb__before_clear_bit();
 987         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 988         smp_mb__after_clear_bit();
 989         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 990         rpc_wake_up(&clp->cl_rpcwaitq);
 991 }
 992
 993 /*
 994  * Schedule the nfs_client asynchronous state management routine
 995  */
 996 void nfs4_schedule_state_manager(struct nfs_client *clp)
 997 {
 998         struct task_struct *task;
 999
1000         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1001                 return;
1002         __module_get(THIS_MODULE);
1003         atomic_inc(&clp->cl_count);
1004         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1005                                 rpc_peeraddr2str(clp->cl_rpcclient,
1006                                                         RPC_DISPLAY_ADDR));
1007         if (!IS_ERR(task))
1008                 return;
1009         nfs4_clear_state_manager_bit(clp);
1010         nfs_put_client(clp);
1011         module_put(THIS_MODULE);
1012 }
1013
1014 /*
1015  * Schedule a lease recovery attempt
1016  */
1017 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1018 {
1019         if (!clp)
1020                 return;
1021         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1022                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1023         nfs4_schedule_state_manager(clp);
1024 }
1025
1026 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1027 {
1028
1029         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1030         /* Don't recover state that expired before the reboot */
1031         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1032                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1033                 return 0;
1034         }
1035         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1036         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1037         return 1;
1038 }
1039
1040 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1041 {
1042         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1043         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1044         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1045         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1046         return 1;
1047 }
1048
1049 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1050 {
1051         struct nfs_client *clp = server->nfs_client;
1052
1053         nfs4_state_mark_reclaim_nograce(clp, state);
1054         nfs4_schedule_state_manager(clp);
1055 }
1056
1057 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1058 {
1059         struct inode *inode = state->inode;
1060         struct nfs_inode *nfsi = NFS_I(inode);
1061         struct file_lock *fl;
1062         int status = 0;
1063
1064         if (inode->i_flock == NULL)
1065                 return 0;
1066
1067         /* Guard against delegation returns and new lock/unlock calls */
1068         down_write(&nfsi->rwsem);
1069         /* Protect inode->i_flock using the BKL */
1070         lock_flocks();
1071         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1072                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1073                         continue;
1074                 if (nfs_file_open_context(fl->fl_file)->state != state)
1075                         continue;
1076                 unlock_flocks();
1077                 status = ops->recover_lock(state, fl);
1078                 switch (status) {
1079                         case 0:
1080                                 break;
1081                         case -ESTALE:
1082                         case -NFS4ERR_ADMIN_REVOKED:
1083                         case -NFS4ERR_STALE_STATEID:
1084                         case -NFS4ERR_BAD_STATEID:
1085                         case -NFS4ERR_EXPIRED:
1086                         case -NFS4ERR_NO_GRACE:
1087                         case -NFS4ERR_STALE_CLIENTID:
1088                         case -NFS4ERR_BADSESSION:
1089                         case -NFS4ERR_BADSLOT:
1090                         case -NFS4ERR_BAD_HIGH_SLOT:
1091                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1092                                 goto out;
1093                         default:
1094                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1095                                                 __func__, status);
1096                         case -ENOMEM:
1097                         case -NFS4ERR_DENIED:
1098                         case -NFS4ERR_RECLAIM_BAD:
1099                         case -NFS4ERR_RECLAIM_CONFLICT:
1100                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1101                                 status = 0;
1102                 }
1103                 lock_flocks();
1104         }
1105         unlock_flocks();
1106 out:
1107         up_write(&nfsi->rwsem);
1108         return status;
1109 }
1110
1111 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1112 {
1113         struct nfs4_state *state;
1114         struct nfs4_lock_state *lock;
1115         int status = 0;
1116
1117         /* Note: we rely on the sp->so_states list being ordered
1118          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1119          * states first.
1120          * This is needed to ensure that the server won't give us any
1121          * read delegations that we have to return if, say, we are
1122          * recovering after a network partition or a reboot from a
1123          * server that doesn't support a grace period.
1124          */
1125 restart:
1126         spin_lock(&sp->so_lock);
1127         list_for_each_entry(state, &sp->so_states, open_states) {
1128                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1129                         continue;
1130                 if (state->state == 0)
1131                         continue;
1132                 atomic_inc(&state->count);
1133                 spin_unlock(&sp->so_lock);
1134                 status = ops->recover_open(sp, state);
1135                 if (status >= 0) {
1136                         status = nfs4_reclaim_locks(state, ops);
1137                         if (status >= 0) {
1138                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1139                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1140                                                 printk("%s: Lock reclaim failed!\n",
1141                                                         __func__);
1142                                 }
1143                                 nfs4_put_open_state(state);
1144                                 goto restart;
1145                         }
1146                 }
1147                 switch (status) {
1148                         default:
1149                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1150                                                 __func__, status);
1151                         case -ENOENT:
1152                         case -ENOMEM:
1153                         case -ESTALE:
1154                                 /*
1155                                  * Open state on this file cannot be recovered
1156                                  * All we can do is revert to using the zero stateid.
1157                                  */
1158                                 memset(state->stateid.data, 0,
1159                                         sizeof(state->stateid.data));
1160                                 /* Mark the file as being 'closed' */
1161                                 state->state = 0;
1162                                 break;
1163                         case -EKEYEXPIRED:
1164                                 /*
1165                                  * User RPCSEC_GSS context has expired.
1166                                  * We cannot recover this stateid now, so
1167                                  * skip it and allow recovery thread to
1168                                  * proceed.
1169                                  */
1170                                 break;
1171                         case -NFS4ERR_ADMIN_REVOKED:
1172                         case -NFS4ERR_STALE_STATEID:
1173                         case -NFS4ERR_BAD_STATEID:
1174                         case -NFS4ERR_RECLAIM_BAD:
1175                         case -NFS4ERR_RECLAIM_CONFLICT:
1176                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1177                                 break;
1178                         case -NFS4ERR_EXPIRED:
1179                         case -NFS4ERR_NO_GRACE:
1180                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1181                         case -NFS4ERR_STALE_CLIENTID:
1182                         case -NFS4ERR_BADSESSION:
1183                         case -NFS4ERR_BADSLOT:
1184                         case -NFS4ERR_BAD_HIGH_SLOT:
1185                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1186                                 goto out_err;
1187                 }
1188                 nfs4_put_open_state(state);
1189                 goto restart;
1190         }
1191         spin_unlock(&sp->so_lock);
1192         return 0;
1193 out_err:
1194         nfs4_put_open_state(state);
1195         return status;
1196 }
1197
1198 static void nfs4_clear_open_state(struct nfs4_state *state)
1199 {
1200         struct nfs4_lock_state *lock;
1201
1202         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1203         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1204         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1205         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1206         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1207                 lock->ls_seqid.flags = 0;
1208                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1209         }
1210 }
1211
1212 static void nfs4_reset_seqids(struct nfs_server *server,
1213         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1214 {
1215         struct nfs_client *clp = server->nfs_client;
1216         struct nfs4_state_owner *sp;
1217         struct rb_node *pos;
1218         struct nfs4_state *state;
1219
1220         spin_lock(&clp->cl_lock);
1221         for (pos = rb_first(&server->state_owners);
1222              pos != NULL;
1223              pos = rb_next(pos)) {
1224                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1225                 sp->so_seqid.flags = 0;
1226                 spin_lock(&sp->so_lock);
1227                 list_for_each_entry(state, &sp->so_states, open_states) {
1228                         if (mark_reclaim(clp, state))
1229                                 nfs4_clear_open_state(state);
1230                 }
1231                 spin_unlock(&sp->so_lock);
1232         }
1233         spin_unlock(&clp->cl_lock);
1234 }
1235
1236 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1237         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1238 {
1239         struct nfs_server *server;
1240
1241         rcu_read_lock();
1242         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1243                 nfs4_reset_seqids(server, mark_reclaim);
1244         rcu_read_unlock();
1245 }
1246
1247 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1248 {
1249         /* Mark all delegations for reclaim */
1250         nfs_delegation_mark_reclaim(clp);
1251         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1252 }
1253
1254 static void nfs4_reclaim_complete(struct nfs_client *clp,
1255                                  const struct nfs4_state_recovery_ops *ops)
1256 {
1257         /* Notify the server we're done reclaiming our state */
1258         if (ops->reclaim_complete)
1259                 (void)ops->reclaim_complete(clp);
1260 }
1261
1262 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1263 {
1264         struct nfs_client *clp = server->nfs_client;
1265         struct nfs4_state_owner *sp;
1266         struct rb_node *pos;
1267         struct nfs4_state *state;
1268
1269         spin_lock(&clp->cl_lock);
1270         for (pos = rb_first(&server->state_owners);
1271              pos != NULL;
1272              pos = rb_next(pos)) {
1273                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1274                 spin_lock(&sp->so_lock);
1275                 list_for_each_entry(state, &sp->so_states, open_states) {
1276                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1277                                                 &state->flags))
1278                                 continue;
1279                         nfs4_state_mark_reclaim_nograce(clp, state);
1280                 }
1281                 spin_unlock(&sp->so_lock);
1282         }
1283         spin_unlock(&clp->cl_lock);
1284 }
1285
1286 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1287 {
1288         struct nfs_server *server;
1289
1290         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1291                 return 0;
1292
1293         rcu_read_lock();
1294         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1295                 nfs4_clear_reclaim_server(server);
1296         rcu_read_unlock();
1297
1298         nfs_delegation_reap_unclaimed(clp);
1299         return 1;
1300 }
1301
1302 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1303 {
1304         if (!nfs4_state_clear_reclaim_reboot(clp))
1305                 return;
1306         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1307 }
1308
1309 static void nfs_delegation_clear_all(struct nfs_client *clp)
1310 {
1311         nfs_delegation_mark_reclaim(clp);
1312         nfs_delegation_reap_unclaimed(clp);
1313 }
1314
1315 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1316 {
1317         nfs_delegation_clear_all(clp);
1318         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1319 }
1320
1321 static void nfs4_warn_keyexpired(const char *s)
1322 {
1323         printk_ratelimited(KERN_WARNING "Error: state manager"
1324                         " encountered RPCSEC_GSS session"
1325                         " expired against NFSv4 server %s.\n",
1326                         s);
1327 }
1328
1329 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1330 {
1331         switch (error) {
1332                 case -NFS4ERR_CB_PATH_DOWN:
1333                         nfs_handle_cb_pathdown(clp);
1334                         return 0;
1335                 case -NFS4ERR_NO_GRACE:
1336                         nfs4_state_end_reclaim_reboot(clp);
1337                         return 0;
1338                 case -NFS4ERR_STALE_CLIENTID:
1339                 case -NFS4ERR_LEASE_MOVED:
1340                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1341                         nfs4_state_clear_reclaim_reboot(clp);
1342                         nfs4_state_start_reclaim_reboot(clp);
1343                         break;
1344                 case -NFS4ERR_EXPIRED:
1345                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1346                         nfs4_state_start_reclaim_nograce(clp);
1347                         break;
1348                 case -NFS4ERR_BADSESSION:
1349                 case -NFS4ERR_BADSLOT:
1350                 case -NFS4ERR_BAD_HIGH_SLOT:
1351                 case -NFS4ERR_DEADSESSION:
1352                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1353                 case -NFS4ERR_SEQ_FALSE_RETRY:
1354                 case -NFS4ERR_SEQ_MISORDERED:
1355                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1356                         /* Zero session reset errors */
1357                         return 0;
1358                 case -EKEYEXPIRED:
1359                         /* Nothing we can do */
1360                         nfs4_warn_keyexpired(clp->cl_hostname);
1361                         return 0;
1362         }
1363         return error;
1364 }
1365
1366 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1367 {
1368         struct nfs4_state_owner *sp;
1369         struct nfs_server *server;
1370         struct rb_node *pos;
1371         int status = 0;
1372
1373 restart:
1374         rcu_read_lock();
1375         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1376                 spin_lock(&clp->cl_lock);
1377                 for (pos = rb_first(&server->state_owners);
1378                      pos != NULL;
1379                      pos = rb_next(pos)) {
1380                         sp = rb_entry(pos,
1381                                 struct nfs4_state_owner, so_server_node);
1382                         if (!test_and_clear_bit(ops->owner_flag_bit,
1383                                                         &sp->so_flags))
1384                                 continue;
1385                         atomic_inc(&sp->so_count);
1386                         spin_unlock(&clp->cl_lock);
1387                         rcu_read_unlock();
1388
1389                         status = nfs4_reclaim_open_state(sp, ops);
1390                         if (status < 0) {
1391                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1392                                 nfs4_put_state_owner(sp);
1393                                 return nfs4_recovery_handle_error(clp, status);
1394                         }
1395
1396                         nfs4_put_state_owner(sp);
1397                         goto restart;
1398                 }
1399                 spin_unlock(&clp->cl_lock);
1400         }
1401         rcu_read_unlock();
1402         return status;
1403 }
1404
1405 static int nfs4_check_lease(struct nfs_client *clp)
1406 {
1407         struct rpc_cred *cred;
1408         const struct nfs4_state_maintenance_ops *ops =
1409                 clp->cl_mvops->state_renewal_ops;
1410         int status = -NFS4ERR_EXPIRED;
1411
1412         /* Is the client already known to have an expired lease? */
1413         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1414                 return 0;
1415         spin_lock(&clp->cl_lock);
1416         cred = ops->get_state_renewal_cred_locked(clp);
1417         spin_unlock(&clp->cl_lock);
1418         if (cred == NULL) {
1419                 cred = nfs4_get_setclientid_cred(clp);
1420                 if (cred == NULL)
1421                         goto out;
1422         }
1423         status = ops->renew_lease(clp, cred);
1424         put_rpccred(cred);
1425 out:
1426         return nfs4_recovery_handle_error(clp, status);
1427 }
1428
1429 static int nfs4_reclaim_lease(struct nfs_client *clp)
1430 {
1431         struct rpc_cred *cred;
1432         const struct nfs4_state_recovery_ops *ops =
1433                 clp->cl_mvops->reboot_recovery_ops;
1434         int status = -ENOENT;
1435
1436         cred = ops->get_clid_cred(clp);
1437         if (cred != NULL) {
1438                 status = ops->establish_clid(clp, cred);
1439                 put_rpccred(cred);
1440                 /* Handle case where the user hasn't set up machine creds */
1441                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1442                         nfs4_clear_machine_cred(clp);
1443                         status = -EAGAIN;
1444                 }
1445                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1446                         status = -EPROTONOSUPPORT;
1447         }
1448         return status;
1449 }
1450
1451 #ifdef CONFIG_NFS_V4_1
1452 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1453 {
1454         nfs4_schedule_lease_recovery(session->clp);
1455 }
1456 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1457
1458 void nfs41_handle_recall_slot(struct nfs_client *clp)
1459 {
1460         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1461         nfs4_schedule_state_manager(clp);
1462 }
1463
1464 static void nfs4_reset_all_state(struct nfs_client *clp)
1465 {
1466         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1467                 clp->cl_boot_time = CURRENT_TIME;
1468                 nfs4_state_start_reclaim_nograce(clp);
1469                 nfs4_schedule_state_manager(clp);
1470         }
1471 }
1472
1473 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1474 {
1475         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1476                 nfs4_state_start_reclaim_reboot(clp);
1477                 nfs4_schedule_state_manager(clp);
1478         }
1479 }
1480
1481 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1482 {
1483         /* Temporary */
1484         nfs4_reset_all_state(clp);
1485 }
1486
1487 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1488 {
1489         /* This will need to handle layouts too */
1490         nfs_expire_all_delegations(clp);
1491 }
1492
1493 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1494 {
1495         nfs_expire_all_delegations(clp);
1496         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1497                 nfs4_schedule_state_manager(clp);
1498 }
1499
1500 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1501 {
1502         if (!flags)
1503                 return;
1504         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1505                 nfs41_handle_server_reboot(clp);
1506         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1507                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1508                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1509                             SEQ4_STATUS_LEASE_MOVED))
1510                 nfs41_handle_state_revoked(clp);
1511         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1512                 nfs41_handle_recallable_state_revoked(clp);
1513         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1514                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1515                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1516                 nfs41_handle_cb_path_down(clp);
1517 }
1518
1519 static int nfs4_reset_session(struct nfs_client *clp)
1520 {
1521         int status;
1522
1523         nfs4_begin_drain_session(clp);
1524         status = nfs4_proc_destroy_session(clp->cl_session);
1525         if (status && status != -NFS4ERR_BADSESSION &&
1526             status != -NFS4ERR_DEADSESSION) {
1527                 status = nfs4_recovery_handle_error(clp, status);
1528                 goto out;
1529         }
1530
1531         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1532         status = nfs4_proc_create_session(clp);
1533         if (status) {
1534                 status = nfs4_recovery_handle_error(clp, status);
1535                 goto out;
1536         }
1537         /* create_session negotiated new slot table */
1538         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1539
1540          /* Let the state manager reestablish state */
1541         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1542                 nfs41_setup_state_renewal(clp);
1543 out:
1544         return status;
1545 }
1546
1547 static int nfs4_recall_slot(struct nfs_client *clp)
1548 {
1549         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1550         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1551         struct nfs4_slot *new, *old;
1552         int i;
1553
1554         nfs4_begin_drain_session(clp);
1555         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1556                       GFP_NOFS);
1557         if (!new)
1558                 return -ENOMEM;
1559
1560         spin_lock(&fc_tbl->slot_tbl_lock);
1561         for (i = 0; i < fc_tbl->target_max_slots; i++)
1562                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1563         old = fc_tbl->slots;
1564         fc_tbl->slots = new;
1565         fc_tbl->max_slots = fc_tbl->target_max_slots;
1566         fc_tbl->target_max_slots = 0;
1567         fc_attrs->max_reqs = fc_tbl->max_slots;
1568         spin_unlock(&fc_tbl->slot_tbl_lock);
1569
1570         kfree(old);
1571         nfs4_end_drain_session(clp);
1572         return 0;
1573 }
1574
1575 #else /* CONFIG_NFS_V4_1 */
1576 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1577 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1578 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1579 #endif /* CONFIG_NFS_V4_1 */
1580
1581 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1582  * on EXCHANGE_ID for v4.1
1583  */
1584 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1585 {
1586         if (nfs4_has_session(clp)) {
1587                 switch (status) {
1588                 case -NFS4ERR_DELAY:
1589                 case -NFS4ERR_CLID_INUSE:
1590                 case -EAGAIN:
1591                         break;
1592
1593                 case -EKEYEXPIRED:
1594                         nfs4_warn_keyexpired(clp->cl_hostname);
1595                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1596                                          * in nfs4_exchange_id */
1597                 default:
1598                         return;
1599                 }
1600         }
1601         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1602 }
1603
1604 static void nfs4_state_manager(struct nfs_client *clp)
1605 {
1606         int status = 0;
1607
1608         /* Ensure exclusive access to NFSv4 state */
1609         for(;;) {
1610                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1611                         /* We're going to have to re-establish a clientid */
1612                         status = nfs4_reclaim_lease(clp);
1613                         if (status) {
1614                                 nfs4_set_lease_expired(clp, status);
1615                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1616                                                         &clp->cl_state))
1617                                         continue;
1618                                 if (clp->cl_cons_state ==
1619                                                         NFS_CS_SESSION_INITING)
1620                                         nfs_mark_client_ready(clp, status);
1621                                 goto out_error;
1622                         }
1623                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1624                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1625                         pnfs_destroy_all_layouts(clp);
1626                 }
1627
1628                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1629                         status = nfs4_check_lease(clp);
1630                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1631                                 continue;
1632                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1633                                 goto out_error;
1634                 }
1635
1636                 /* Initialize or reset the session */
1637                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1638                    && nfs4_has_session(clp)) {
1639                         status = nfs4_reset_session(clp);
1640                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1641                                 continue;
1642                         if (status < 0)
1643                                 goto out_error;
1644                 }
1645
1646                 /* First recover reboot state... */
1647                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1648                         status = nfs4_do_reclaim(clp,
1649                                 clp->cl_mvops->reboot_recovery_ops);
1650                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1651                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1652                                 continue;
1653                         nfs4_state_end_reclaim_reboot(clp);
1654                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1655                                 continue;
1656                         if (status < 0)
1657                                 goto out_error;
1658                 }
1659
1660                 /* Now recover expired state... */
1661                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1662                         status = nfs4_do_reclaim(clp,
1663                                 clp->cl_mvops->nograce_recovery_ops);
1664                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1665                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1666                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1667                                 continue;
1668                         if (status < 0)
1669                                 goto out_error;
1670                 }
1671
1672                 nfs4_end_drain_session(clp);
1673                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1674                         nfs_client_return_marked_delegations(clp);
1675                         continue;
1676                 }
1677                 /* Recall session slots */
1678                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1679                    && nfs4_has_session(clp)) {
1680                         status = nfs4_recall_slot(clp);
1681                         if (status < 0)
1682                                 goto out_error;
1683                         continue;
1684                 }
1685
1686
1687                 nfs4_clear_state_manager_bit(clp);
1688                 /* Did we race with an attempt to give us more work? */
1689                 if (clp->cl_state == 0)
1690                         break;
1691                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1692                         break;
1693         }
1694         return;
1695 out_error:
1696         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1697                         " with error %d\n", clp->cl_hostname, -status);
1698         nfs4_end_drain_session(clp);
1699         nfs4_clear_state_manager_bit(clp);
1700 }
1701
1702 static int nfs4_run_state_manager(void *ptr)
1703 {
1704         struct nfs_client *clp = ptr;
1705
1706         allow_signal(SIGKILL);
1707         nfs4_state_manager(clp);
1708         nfs_put_client(clp);
1709         module_put_and_exit(0);
1710         return 0;
1711 }
1712
1713 /*
1714  * Local variables:
1715  *  c-basic-offset: 8
1716  * End:
1717  */