SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/smp_lock.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/workqueue.h>
  50 #include <linux/bitops.h>
  51
  52 #include "nfs4_fs.h"
  53 #include "callback.h"
  54 #include "delegation.h"
  55 #include "internal.h"
  56
  57 #define OPENOWNER_POOL_SIZE     8
  58
  59 const nfs4_stateid zero_stateid;
  60
  61 static LIST_HEAD(nfs4_clientid_list);
  62
  63 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  64 {
  65         unsigned short port;
  66         int status;
  67
  68         port = nfs_callback_tcpport;
  69         if (clp->cl_addr.ss_family == AF_INET6)
  70                 port = nfs_callback_tcpport6;
  71
  72         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred);
  73         if (status == 0)
  74                 status = nfs4_proc_setclientid_confirm(clp, cred);
  75         if (status == 0)
  76                 nfs4_schedule_state_renewal(clp);
  77         return status;
  78 }
  79
  80 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  81 {
  82         struct rpc_cred *cred = NULL;
  83
  84         if (clp->cl_machine_cred != NULL)
  85                 cred = get_rpccred(clp->cl_machine_cred);
  86         return cred;
  87 }
  88
  89 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  90 {
  91         struct rpc_cred *cred;
  92
  93         spin_lock(&clp->cl_lock);
  94         cred = clp->cl_machine_cred;
  95         clp->cl_machine_cred = NULL;
  96         spin_unlock(&clp->cl_lock);
  97         if (cred != NULL)
  98                 put_rpccred(cred);
  99 }
 100
 101 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 102 {
 103         struct nfs4_state_owner *sp;
 104         struct rb_node *pos;
 105         struct rpc_cred *cred = NULL;
 106
 107         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 108                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 109                 if (list_empty(&sp->so_states))
 110                         continue;
 111                 cred = get_rpccred(sp->so_cred);
 112                 break;
 113         }
 114         return cred;
 115 }
 116
 117 #if defined(CONFIG_NFS_V4_1)
 118
 119 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 120 {
 121         int status;
 122         struct nfs_fsinfo fsinfo;
 123
 124         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 125         if (status == 0) {
 126                 /* Update lease time and schedule renewal */
 127                 spin_lock(&clp->cl_lock);
 128                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 129                 clp->cl_last_renewal = jiffies;
 130                 spin_unlock(&clp->cl_lock);
 131
 132                 nfs4_schedule_state_renewal(clp);
 133         }
 134
 135         return status;
 136 }
 137
 138 static void nfs4_end_drain_session(struct nfs_client *clp)
 139 {
 140         struct nfs4_session *ses = clp->cl_session;
 141         int max_slots;
 142
 143         if (test_and_clear_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state)) {
 144                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 145                 max_slots = ses->fc_slot_table.max_slots;
 146                 while (max_slots--) {
 147                         struct rpc_task *task;
 148
 149                         task = rpc_wake_up_next(&ses->fc_slot_table.
 150                                                 slot_tbl_waitq);
 151                         if (!task)
 152                                 break;
 153                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 154                 }
 155                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 156         }
 157 }
 158
 159 static int nfs4_begin_drain_session(struct nfs_client *clp)
 160 {
 161         struct nfs4_session *ses = clp->cl_session;
 162         struct nfs4_slot_table *tbl = &ses->fc_slot_table;
 163
 164         spin_lock(&tbl->slot_tbl_lock);
 165         set_bit(NFS4CLNT_SESSION_DRAINING, &clp->cl_state);
 166         if (tbl->highest_used_slotid != -1) {
 167                 INIT_COMPLETION(ses->complete);
 168                 spin_unlock(&tbl->slot_tbl_lock);
 169                 return wait_for_completion_interruptible(&ses->complete);
 170         }
 171         spin_unlock(&tbl->slot_tbl_lock);
 172         return 0;
 173 }
 174
 175 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 176 {
 177         int status;
 178
 179         status = nfs4_proc_exchange_id(clp, cred);
 180         if (status != 0)
 181                 goto out;
 182         status = nfs4_proc_create_session(clp);
 183         if (status != 0)
 184                 goto out;
 185         nfs41_setup_state_renewal(clp);
 186         nfs_mark_client_ready(clp, NFS_CS_READY);
 187 out:
 188         return status;
 189 }
 190
 191 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 192 {
 193         struct rpc_cred *cred;
 194
 195         spin_lock(&clp->cl_lock);
 196         cred = nfs4_get_machine_cred_locked(clp);
 197         spin_unlock(&clp->cl_lock);
 198         return cred;
 199 }
 200
 201 #endif /* CONFIG_NFS_V4_1 */
 202
 203 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 204 {
 205         struct nfs4_state_owner *sp;
 206         struct rb_node *pos;
 207         struct rpc_cred *cred;
 208
 209         spin_lock(&clp->cl_lock);
 210         cred = nfs4_get_machine_cred_locked(clp);
 211         if (cred != NULL)
 212                 goto out;
 213         pos = rb_first(&clp->cl_state_owners);
 214         if (pos != NULL) {
 215                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 216                 cred = get_rpccred(sp->so_cred);
 217         }
 218 out:
 219         spin_unlock(&clp->cl_lock);
 220         return cred;
 221 }
 222
 223 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 224                 __u64 minval, int maxbits)
 225 {
 226         struct rb_node **p, *parent;
 227         struct nfs_unique_id *pos;
 228         __u64 mask = ~0ULL;
 229
 230         if (maxbits < 64)
 231                 mask = (1ULL << maxbits) - 1ULL;
 232
 233         /* Ensure distribution is more or less flat */
 234         get_random_bytes(&new->id, sizeof(new->id));
 235         new->id &= mask;
 236         if (new->id < minval)
 237                 new->id += minval;
 238 retry:
 239         p = &root->rb_node;
 240         parent = NULL;
 241
 242         while (*p != NULL) {
 243                 parent = *p;
 244                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 245
 246                 if (new->id < pos->id)
 247                         p = &(*p)->rb_left;
 248                 else if (new->id > pos->id)
 249                         p = &(*p)->rb_right;
 250                 else
 251                         goto id_exists;
 252         }
 253         rb_link_node(&new->rb_node, parent, p);
 254         rb_insert_color(&new->rb_node, root);
 255         return;
 256 id_exists:
 257         for (;;) {
 258                 new->id++;
 259                 if (new->id < minval || (new->id & mask) != new->id) {
 260                         new->id = minval;
 261                         break;
 262                 }
 263                 parent = rb_next(parent);
 264                 if (parent == NULL)
 265                         break;
 266                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 267                 if (new->id < pos->id)
 268                         break;
 269         }
 270         goto retry;
 271 }
 272
 273 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 274 {
 275         rb_erase(&id->rb_node, root);
 276 }
 277
 278 static struct nfs4_state_owner *
 279 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 280 {
 281         struct nfs_client *clp = server->nfs_client;
 282         struct rb_node **p = &clp->cl_state_owners.rb_node,
 283                        *parent = NULL;
 284         struct nfs4_state_owner *sp, *res = NULL;
 285
 286         while (*p != NULL) {
 287                 parent = *p;
 288                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 289
 290                 if (server < sp->so_server) {
 291                         p = &parent->rb_left;
 292                         continue;
 293                 }
 294                 if (server > sp->so_server) {
 295                         p = &parent->rb_right;
 296                         continue;
 297                 }
 298                 if (cred < sp->so_cred)
 299                         p = &parent->rb_left;
 300                 else if (cred > sp->so_cred)
 301                         p = &parent->rb_right;
 302                 else {
 303                         atomic_inc(&sp->so_count);
 304                         res = sp;
 305                         break;
 306                 }
 307         }
 308         return res;
 309 }
 310
 311 static struct nfs4_state_owner *
 312 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 313 {
 314         struct rb_node **p = &clp->cl_state_owners.rb_node,
 315                        *parent = NULL;
 316         struct nfs4_state_owner *sp;
 317
 318         while (*p != NULL) {
 319                 parent = *p;
 320                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 321
 322                 if (new->so_server < sp->so_server) {
 323                         p = &parent->rb_left;
 324                         continue;
 325                 }
 326                 if (new->so_server > sp->so_server) {
 327                         p = &parent->rb_right;
 328                         continue;
 329                 }
 330                 if (new->so_cred < sp->so_cred)
 331                         p = &parent->rb_left;
 332                 else if (new->so_cred > sp->so_cred)
 333                         p = &parent->rb_right;
 334                 else {
 335                         atomic_inc(&sp->so_count);
 336                         return sp;
 337                 }
 338         }
 339         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 340         rb_link_node(&new->so_client_node, parent, p);
 341         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 342         return new;
 343 }
 344
 345 static void
 346 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 347 {
 348         if (!RB_EMPTY_NODE(&sp->so_client_node))
 349                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 350         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 351 }
 352
 353 /*
 354  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 355  * create a new state_owner.
 356  *
 357  */
 358 static struct nfs4_state_owner *
 359 nfs4_alloc_state_owner(void)
 360 {
 361         struct nfs4_state_owner *sp;
 362
 363         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 364         if (!sp)
 365                 return NULL;
 366         spin_lock_init(&sp->so_lock);
 367         INIT_LIST_HEAD(&sp->so_states);
 368         INIT_LIST_HEAD(&sp->so_delegations);
 369         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 370         sp->so_seqid.sequence = &sp->so_sequence;
 371         spin_lock_init(&sp->so_sequence.lock);
 372         INIT_LIST_HEAD(&sp->so_sequence.list);
 373         atomic_set(&sp->so_count, 1);
 374         return sp;
 375 }
 376
 377 static void
 378 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 379 {
 380         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 381                 struct nfs_client *clp = sp->so_client;
 382
 383                 spin_lock(&clp->cl_lock);
 384                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 385                 RB_CLEAR_NODE(&sp->so_client_node);
 386                 spin_unlock(&clp->cl_lock);
 387         }
 388 }
 389
 390 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 391 {
 392         struct nfs_client *clp = server->nfs_client;
 393         struct nfs4_state_owner *sp, *new;
 394
 395         spin_lock(&clp->cl_lock);
 396         sp = nfs4_find_state_owner(server, cred);
 397         spin_unlock(&clp->cl_lock);
 398         if (sp != NULL)
 399                 return sp;
 400         new = nfs4_alloc_state_owner();
 401         if (new == NULL)
 402                 return NULL;
 403         new->so_client = clp;
 404         new->so_server = server;
 405         new->so_cred = cred;
 406         spin_lock(&clp->cl_lock);
 407         sp = nfs4_insert_state_owner(clp, new);
 408         spin_unlock(&clp->cl_lock);
 409         if (sp == new)
 410                 get_rpccred(cred);
 411         else {
 412                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 413                 kfree(new);
 414         }
 415         return sp;
 416 }
 417
 418 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 419 {
 420         struct nfs_client *clp = sp->so_client;
 421         struct rpc_cred *cred = sp->so_cred;
 422
 423         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 424                 return;
 425         nfs4_remove_state_owner(clp, sp);
 426         spin_unlock(&clp->cl_lock);
 427         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 428         put_rpccred(cred);
 429         kfree(sp);
 430 }
 431
 432 static struct nfs4_state *
 433 nfs4_alloc_open_state(void)
 434 {
 435         struct nfs4_state *state;
 436
 437         state = kzalloc(sizeof(*state), GFP_KERNEL);
 438         if (!state)
 439                 return NULL;
 440         atomic_set(&state->count, 1);
 441         INIT_LIST_HEAD(&state->lock_states);
 442         spin_lock_init(&state->state_lock);
 443         seqlock_init(&state->seqlock);
 444         return state;
 445 }
 446
 447 void
 448 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 449 {
 450         if (state->state == fmode)
 451                 return;
 452         /* NB! List reordering - see the reclaim code for why.  */
 453         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 454                 if (fmode & FMODE_WRITE)
 455                         list_move(&state->open_states, &state->owner->so_states);
 456                 else
 457                         list_move_tail(&state->open_states, &state->owner->so_states);
 458         }
 459         state->state = fmode;
 460 }
 461
 462 static struct nfs4_state *
 463 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 464 {
 465         struct nfs_inode *nfsi = NFS_I(inode);
 466         struct nfs4_state *state;
 467
 468         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 469                 if (state->owner != owner)
 470                         continue;
 471                 if (atomic_inc_not_zero(&state->count))
 472                         return state;
 473         }
 474         return NULL;
 475 }
 476
 477 static void
 478 nfs4_free_open_state(struct nfs4_state *state)
 479 {
 480         kfree(state);
 481 }
 482
 483 struct nfs4_state *
 484 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 485 {
 486         struct nfs4_state *state, *new;
 487         struct nfs_inode *nfsi = NFS_I(inode);
 488
 489         spin_lock(&inode->i_lock);
 490         state = __nfs4_find_state_byowner(inode, owner);
 491         spin_unlock(&inode->i_lock);
 492         if (state)
 493                 goto out;
 494         new = nfs4_alloc_open_state();
 495         spin_lock(&owner->so_lock);
 496         spin_lock(&inode->i_lock);
 497         state = __nfs4_find_state_byowner(inode, owner);
 498         if (state == NULL && new != NULL) {
 499                 state = new;
 500                 state->owner = owner;
 501                 atomic_inc(&owner->so_count);
 502                 list_add(&state->inode_states, &nfsi->open_states);
 503                 state->inode = igrab(inode);
 504                 spin_unlock(&inode->i_lock);
 505                 /* Note: The reclaim code dictates that we add stateless
 506                  * and read-only stateids to the end of the list */
 507                 list_add_tail(&state->open_states, &owner->so_states);
 508                 spin_unlock(&owner->so_lock);
 509         } else {
 510                 spin_unlock(&inode->i_lock);
 511                 spin_unlock(&owner->so_lock);
 512                 if (new)
 513                         nfs4_free_open_state(new);
 514         }
 515 out:
 516         return state;
 517 }
 518
 519 void nfs4_put_open_state(struct nfs4_state *state)
 520 {
 521         struct inode *inode = state->inode;
 522         struct nfs4_state_owner *owner = state->owner;
 523
 524         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 525                 return;
 526         spin_lock(&inode->i_lock);
 527         list_del(&state->inode_states);
 528         list_del(&state->open_states);
 529         spin_unlock(&inode->i_lock);
 530         spin_unlock(&owner->so_lock);
 531         iput(inode);
 532         nfs4_free_open_state(state);
 533         nfs4_put_state_owner(owner);
 534 }
 535
 536 /*
 537  * Close the current file.
 538  */
 539 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
 540 {
 541         struct nfs4_state_owner *owner = state->owner;
 542         int call_close = 0;
 543         fmode_t newstate;
 544
 545         atomic_inc(&owner->so_count);
 546         /* Protect against nfs4_find_state() */
 547         spin_lock(&owner->so_lock);
 548         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 549                 case FMODE_READ:
 550                         state->n_rdonly--;
 551                         break;
 552                 case FMODE_WRITE:
 553                         state->n_wronly--;
 554                         break;
 555                 case FMODE_READ|FMODE_WRITE:
 556                         state->n_rdwr--;
 557         }
 558         newstate = FMODE_READ|FMODE_WRITE;
 559         if (state->n_rdwr == 0) {
 560                 if (state->n_rdonly == 0) {
 561                         newstate &= ~FMODE_READ;
 562                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 563                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 564                 }
 565                 if (state->n_wronly == 0) {
 566                         newstate &= ~FMODE_WRITE;
 567                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 568                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 569                 }
 570                 if (newstate == 0)
 571                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 572         }
 573         nfs4_state_set_mode_locked(state, newstate);
 574         spin_unlock(&owner->so_lock);
 575
 576         if (!call_close) {
 577                 nfs4_put_open_state(state);
 578                 nfs4_put_state_owner(owner);
 579         } else
 580                 nfs4_do_close(path, state, wait);
 581 }
 582
 583 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 584 {
 585         __nfs4_close(path, state, fmode, 0);
 586 }
 587
 588 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 589 {
 590         __nfs4_close(path, state, fmode, 1);
 591 }
 592
 593 /*
 594  * Search the state->lock_states for an existing lock_owner
 595  * that is compatible with current->files
 596  */
 597 static struct nfs4_lock_state *
 598 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 599 {
 600         struct nfs4_lock_state *pos;
 601         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 602                 if (pos->ls_owner != fl_owner)
 603                         continue;
 604                 atomic_inc(&pos->ls_count);
 605                 return pos;
 606         }
 607         return NULL;
 608 }
 609
 610 /*
 611  * Return a compatible lock_state. If no initialized lock_state structure
 612  * exists, return an uninitialized one.
 613  *
 614  */
 615 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 616 {
 617         struct nfs4_lock_state *lsp;
 618         struct nfs_client *clp = state->owner->so_client;
 619
 620         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 621         if (lsp == NULL)
 622                 return NULL;
 623         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 624         spin_lock_init(&lsp->ls_sequence.lock);
 625         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 626         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 627         atomic_set(&lsp->ls_count, 1);
 628         lsp->ls_state = state;
 629         lsp->ls_owner = fl_owner;
 630         spin_lock(&clp->cl_lock);
 631         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 632         spin_unlock(&clp->cl_lock);
 633         INIT_LIST_HEAD(&lsp->ls_locks);
 634         return lsp;
 635 }
 636
 637 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 638 {
 639         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 640
 641         spin_lock(&clp->cl_lock);
 642         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 643         spin_unlock(&clp->cl_lock);
 644         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 645         kfree(lsp);
 646 }
 647
 648 /*
 649  * Return a compatible lock_state. If no initialized lock_state structure
 650  * exists, return an uninitialized one.
 651  *
 652  */
 653 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 654 {
 655         struct nfs4_lock_state *lsp, *new = NULL;
 656
 657         for(;;) {
 658                 spin_lock(&state->state_lock);
 659                 lsp = __nfs4_find_lock_state(state, owner);
 660                 if (lsp != NULL)
 661                         break;
 662                 if (new != NULL) {
 663                         list_add(&new->ls_locks, &state->lock_states);
 664                         set_bit(LK_STATE_IN_USE, &state->flags);
 665                         lsp = new;
 666                         new = NULL;
 667                         break;
 668                 }
 669                 spin_unlock(&state->state_lock);
 670                 new = nfs4_alloc_lock_state(state, owner);
 671                 if (new == NULL)
 672                         return NULL;
 673         }
 674         spin_unlock(&state->state_lock);
 675         if (new != NULL)
 676                 nfs4_free_lock_state(new);
 677         return lsp;
 678 }
 679
 680 /*
 681  * Release reference to lock_state, and free it if we see that
 682  * it is no longer in use
 683  */
 684 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 685 {
 686         struct nfs4_state *state;
 687
 688         if (lsp == NULL)
 689                 return;
 690         state = lsp->ls_state;
 691         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 692                 return;
 693         list_del(&lsp->ls_locks);
 694         if (list_empty(&state->lock_states))
 695                 clear_bit(LK_STATE_IN_USE, &state->flags);
 696         spin_unlock(&state->state_lock);
 697         nfs4_free_lock_state(lsp);
 698 }
 699
 700 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 701 {
 702         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 703
 704         dst->fl_u.nfs4_fl.owner = lsp;
 705         atomic_inc(&lsp->ls_count);
 706 }
 707
 708 static void nfs4_fl_release_lock(struct file_lock *fl)
 709 {
 710         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 711 }
 712
 713 static const struct file_lock_operations nfs4_fl_lock_ops = {
 714         .fl_copy_lock = nfs4_fl_copy_lock,
 715         .fl_release_private = nfs4_fl_release_lock,
 716 };
 717
 718 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 719 {
 720         struct nfs4_lock_state *lsp;
 721
 722         if (fl->fl_ops != NULL)
 723                 return 0;
 724         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 725         if (lsp == NULL)
 726                 return -ENOMEM;
 727         fl->fl_u.nfs4_fl.owner = lsp;
 728         fl->fl_ops = &nfs4_fl_lock_ops;
 729         return 0;
 730 }
 731
 732 /*
 733  * Byte-range lock aware utility to initialize the stateid of read/write
 734  * requests.
 735  */
 736 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 737 {
 738         struct nfs4_lock_state *lsp;
 739         int seq;
 740
 741         do {
 742                 seq = read_seqbegin(&state->seqlock);
 743                 memcpy(dst, &state->stateid, sizeof(*dst));
 744         } while (read_seqretry(&state->seqlock, seq));
 745         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 746                 return;
 747
 748         spin_lock(&state->state_lock);
 749         lsp = __nfs4_find_lock_state(state, fl_owner);
 750         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 751                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 752         spin_unlock(&state->state_lock);
 753         nfs4_put_lock_state(lsp);
 754 }
 755
 756 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 757 {
 758         struct nfs_seqid *new;
 759
 760         new = kmalloc(sizeof(*new), GFP_KERNEL);
 761         if (new != NULL) {
 762                 new->sequence = counter;
 763                 INIT_LIST_HEAD(&new->list);
 764         }
 765         return new;
 766 }
 767
 768 void nfs_free_seqid(struct nfs_seqid *seqid)
 769 {
 770         if (!list_empty(&seqid->list)) {
 771                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 772
 773                 spin_lock(&sequence->lock);
 774                 list_del(&seqid->list);
 775                 spin_unlock(&sequence->lock);
 776                 rpc_wake_up(&sequence->wait);
 777         }
 778         kfree(seqid);
 779 }
 780
 781 /*
 782  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 783  * failed with a seqid incrementing error -
 784  * see comments nfs_fs.h:seqid_mutating_error()
 785  */
 786 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 787 {
 788         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 789         switch (status) {
 790                 case 0:
 791                         break;
 792                 case -NFS4ERR_BAD_SEQID:
 793                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 794                                 return;
 795                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 796                                         " sequence-id error on an"
 797                                         " unconfirmed sequence %p!\n",
 798                                         seqid->sequence);
 799                 case -NFS4ERR_STALE_CLIENTID:
 800                 case -NFS4ERR_STALE_STATEID:
 801                 case -NFS4ERR_BAD_STATEID:
 802                 case -NFS4ERR_BADXDR:
 803                 case -NFS4ERR_RESOURCE:
 804                 case -NFS4ERR_NOFILEHANDLE:
 805                         /* Non-seqid mutating errors */
 806                         return;
 807         };
 808         /*
 809          * Note: no locking needed as we are guaranteed to be first
 810          * on the sequence list
 811          */
 812         seqid->sequence->counter++;
 813 }
 814
 815 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 816 {
 817         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 818                                         struct nfs4_state_owner, so_seqid);
 819         struct nfs_server *server = sp->so_server;
 820
 821         if (status == -NFS4ERR_BAD_SEQID)
 822                 nfs4_drop_state_owner(sp);
 823         if (!nfs4_has_session(server->nfs_client))
 824                 nfs_increment_seqid(status, seqid);
 825 }
 826
 827 /*
 828  * Increment the seqid if the LOCK/LOCKU succeeded, or
 829  * failed with a seqid incrementing error -
 830  * see comments nfs_fs.h:seqid_mutating_error()
 831  */
 832 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 833 {
 834         nfs_increment_seqid(status, seqid);
 835 }
 836
 837 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 838 {
 839         struct rpc_sequence *sequence = seqid->sequence->sequence;
 840         int status = 0;
 841
 842         spin_lock(&sequence->lock);
 843         if (list_empty(&seqid->list))
 844                 list_add_tail(&seqid->list, &sequence->list);
 845         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 846                 goto unlock;
 847         rpc_sleep_on(&sequence->wait, task, NULL);
 848         status = -EAGAIN;
 849 unlock:
 850         spin_unlock(&sequence->lock);
 851         return status;
 852 }
 853
 854 static int nfs4_run_state_manager(void *);
 855
 856 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 857 {
 858         smp_mb__before_clear_bit();
 859         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 860         smp_mb__after_clear_bit();
 861         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 862         rpc_wake_up(&clp->cl_rpcwaitq);
 863 }
 864
 865 /*
 866  * Schedule the nfs_client asynchronous state management routine
 867  */
 868 void nfs4_schedule_state_manager(struct nfs_client *clp)
 869 {
 870         struct task_struct *task;
 871
 872         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
 873                 return;
 874         __module_get(THIS_MODULE);
 875         atomic_inc(&clp->cl_count);
 876         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
 877                                 rpc_peeraddr2str(clp->cl_rpcclient,
 878                                                         RPC_DISPLAY_ADDR));
 879         if (!IS_ERR(task))
 880                 return;
 881         nfs4_clear_state_manager_bit(clp);
 882         nfs_put_client(clp);
 883         module_put(THIS_MODULE);
 884 }
 885
 886 /*
 887  * Schedule a state recovery attempt
 888  */
 889 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 890 {
 891         if (!clp)
 892                 return;
 893         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
 894                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 895         nfs4_schedule_state_manager(clp);
 896 }
 897
 898 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
 899 {
 900
 901         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 902         /* Don't recover state that expired before the reboot */
 903         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
 904                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 905                 return 0;
 906         }
 907         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
 908         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
 909         return 1;
 910 }
 911
 912 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
 913 {
 914         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
 915         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 916         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
 917         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
 918         return 1;
 919 }
 920
 921 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 922 {
 923         struct inode *inode = state->inode;
 924         struct nfs_inode *nfsi = NFS_I(inode);
 925         struct file_lock *fl;
 926         int status = 0;
 927
 928         if (inode->i_flock == NULL)
 929                 return 0;
 930
 931         /* Guard against delegation returns and new lock/unlock calls */
 932         down_write(&nfsi->rwsem);
 933         /* Protect inode->i_flock using the BKL */
 934         lock_kernel();
 935         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 936                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 937                         continue;
 938                 if (nfs_file_open_context(fl->fl_file)->state != state)
 939                         continue;
 940                 unlock_kernel();
 941                 status = ops->recover_lock(state, fl);
 942                 switch (status) {
 943                         case 0:
 944                                 break;
 945                         case -ESTALE:
 946                         case -NFS4ERR_ADMIN_REVOKED:
 947                         case -NFS4ERR_STALE_STATEID:
 948                         case -NFS4ERR_BAD_STATEID:
 949                         case -NFS4ERR_EXPIRED:
 950                         case -NFS4ERR_NO_GRACE:
 951                         case -NFS4ERR_STALE_CLIENTID:
 952                         case -NFS4ERR_BADSESSION:
 953                         case -NFS4ERR_BADSLOT:
 954                         case -NFS4ERR_BAD_HIGH_SLOT:
 955                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 956                                 goto out;
 957                         default:
 958                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 959                                                 __func__, status);
 960                         case -ENOMEM:
 961                         case -NFS4ERR_DENIED:
 962                         case -NFS4ERR_RECLAIM_BAD:
 963                         case -NFS4ERR_RECLAIM_CONFLICT:
 964                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 965                                 status = 0;
 966                 }
 967                 lock_kernel();
 968         }
 969         unlock_kernel();
 970 out:
 971         up_write(&nfsi->rwsem);
 972         return status;
 973 }
 974
 975 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
 976 {
 977         struct nfs4_state *state;
 978         struct nfs4_lock_state *lock;
 979         int status = 0;
 980
 981         /* Note: we rely on the sp->so_states list being ordered
 982          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 983          * states first.
 984          * This is needed to ensure that the server won't give us any
 985          * read delegations that we have to return if, say, we are
 986          * recovering after a network partition or a reboot from a
 987          * server that doesn't support a grace period.
 988          */
 989 restart:
 990         spin_lock(&sp->so_lock);
 991         list_for_each_entry(state, &sp->so_states, open_states) {
 992                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
 993                         continue;
 994                 if (state->state == 0)
 995                         continue;
 996                 atomic_inc(&state->count);
 997                 spin_unlock(&sp->so_lock);
 998                 status = ops->recover_open(sp, state);
 999                 if (status >= 0) {
1000                         status = nfs4_reclaim_locks(state, ops);
1001                         if (status >= 0) {
1002                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1003                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1004                                                 printk("%s: Lock reclaim failed!\n",
1005                                                         __func__);
1006                                 }
1007                                 nfs4_put_open_state(state);
1008                                 goto restart;
1009                         }
1010                 }
1011                 switch (status) {
1012                         default:
1013                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1014                                                 __func__, status);
1015                         case -ENOENT:
1016                         case -ENOMEM:
1017                         case -ESTALE:
1018                                 /*
1019                                  * Open state on this file cannot be recovered
1020                                  * All we can do is revert to using the zero stateid.
1021                                  */
1022                                 memset(state->stateid.data, 0,
1023                                         sizeof(state->stateid.data));
1024                                 /* Mark the file as being 'closed' */
1025                                 state->state = 0;
1026                                 break;
1027                         case -NFS4ERR_ADMIN_REVOKED:
1028                         case -NFS4ERR_STALE_STATEID:
1029                         case -NFS4ERR_BAD_STATEID:
1030                         case -NFS4ERR_RECLAIM_BAD:
1031                         case -NFS4ERR_RECLAIM_CONFLICT:
1032                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1033                                 break;
1034                         case -NFS4ERR_EXPIRED:
1035                         case -NFS4ERR_NO_GRACE:
1036                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
1037                         case -NFS4ERR_STALE_CLIENTID:
1038                         case -NFS4ERR_BADSESSION:
1039                         case -NFS4ERR_BADSLOT:
1040                         case -NFS4ERR_BAD_HIGH_SLOT:
1041                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1042                                 goto out_err;
1043                 }
1044                 nfs4_put_open_state(state);
1045                 goto restart;
1046         }
1047         spin_unlock(&sp->so_lock);
1048         return 0;
1049 out_err:
1050         nfs4_put_open_state(state);
1051         return status;
1052 }
1053
1054 static void nfs4_clear_open_state(struct nfs4_state *state)
1055 {
1056         struct nfs4_lock_state *lock;
1057
1058         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1059         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1060         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1061         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1062         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1063                 lock->ls_seqid.flags = 0;
1064                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1065         }
1066 }
1067
1068 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1069 {
1070         struct nfs4_state_owner *sp;
1071         struct rb_node *pos;
1072         struct nfs4_state *state;
1073
1074         /* Reset all sequence ids to zero */
1075         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1076                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1077                 sp->so_seqid.flags = 0;
1078                 spin_lock(&sp->so_lock);
1079                 list_for_each_entry(state, &sp->so_states, open_states) {
1080                         if (mark_reclaim(clp, state))
1081                                 nfs4_clear_open_state(state);
1082                 }
1083                 spin_unlock(&sp->so_lock);
1084         }
1085 }
1086
1087 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1088 {
1089         /* Mark all delegations for reclaim */
1090         nfs_delegation_mark_reclaim(clp);
1091         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1092 }
1093
1094 static void nfs4_reclaim_complete(struct nfs_client *clp,
1095                                  const struct nfs4_state_recovery_ops *ops)
1096 {
1097         /* Notify the server we're done reclaiming our state */
1098         if (ops->reclaim_complete)
1099                 (void)ops->reclaim_complete(clp);
1100 }
1101
1102 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1103 {
1104         struct nfs4_state_owner *sp;
1105         struct rb_node *pos;
1106         struct nfs4_state *state;
1107
1108         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1109                 return;
1110
1111         nfs4_reclaim_complete(clp,
1112                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1113
1114         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1115                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1116                 spin_lock(&sp->so_lock);
1117                 list_for_each_entry(state, &sp->so_states, open_states) {
1118                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1119                                 continue;
1120                         nfs4_state_mark_reclaim_nograce(clp, state);
1121                 }
1122                 spin_unlock(&sp->so_lock);
1123         }
1124
1125         nfs_delegation_reap_unclaimed(clp);
1126 }
1127
1128 static void nfs_delegation_clear_all(struct nfs_client *clp)
1129 {
1130         nfs_delegation_mark_reclaim(clp);
1131         nfs_delegation_reap_unclaimed(clp);
1132 }
1133
1134 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1135 {
1136         nfs_delegation_clear_all(clp);
1137         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1138 }
1139
1140 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1141 {
1142         switch (error) {
1143                 case -NFS4ERR_CB_PATH_DOWN:
1144                         nfs_handle_cb_pathdown(clp);
1145                         return 0;
1146                 case -NFS4ERR_NO_GRACE:
1147                         nfs4_state_end_reclaim_reboot(clp);
1148                         return 0;
1149                 case -NFS4ERR_STALE_CLIENTID:
1150                 case -NFS4ERR_LEASE_MOVED:
1151                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1152                         nfs4_state_end_reclaim_reboot(clp);
1153                         nfs4_state_start_reclaim_reboot(clp);
1154                         break;
1155                 case -NFS4ERR_EXPIRED:
1156                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1157                         nfs4_state_start_reclaim_nograce(clp);
1158                         break;
1159                 case -NFS4ERR_BADSESSION:
1160                 case -NFS4ERR_BADSLOT:
1161                 case -NFS4ERR_BAD_HIGH_SLOT:
1162                 case -NFS4ERR_DEADSESSION:
1163                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1164                 case -NFS4ERR_SEQ_FALSE_RETRY:
1165                 case -NFS4ERR_SEQ_MISORDERED:
1166                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1167                         /* Zero session reset errors */
1168                         return 0;
1169         }
1170         return error;
1171 }
1172
1173 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1174 {
1175         struct rb_node *pos;
1176         int status = 0;
1177
1178 restart:
1179         spin_lock(&clp->cl_lock);
1180         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1181                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1182                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1183                         continue;
1184                 atomic_inc(&sp->so_count);
1185                 spin_unlock(&clp->cl_lock);
1186                 status = nfs4_reclaim_open_state(sp, ops);
1187                 if (status < 0) {
1188                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1189                         nfs4_put_state_owner(sp);
1190                         return nfs4_recovery_handle_error(clp, status);
1191                 }
1192                 nfs4_put_state_owner(sp);
1193                 goto restart;
1194         }
1195         spin_unlock(&clp->cl_lock);
1196         return status;
1197 }
1198
1199 static int nfs4_check_lease(struct nfs_client *clp)
1200 {
1201         struct rpc_cred *cred;
1202         struct nfs4_state_maintenance_ops *ops =
1203                 nfs4_state_renewal_ops[clp->cl_minorversion];
1204         int status = -NFS4ERR_EXPIRED;
1205
1206         /* Is the client already known to have an expired lease? */
1207         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1208                 return 0;
1209         spin_lock(&clp->cl_lock);
1210         cred = ops->get_state_renewal_cred_locked(clp);
1211         spin_unlock(&clp->cl_lock);
1212         if (cred == NULL) {
1213                 cred = nfs4_get_setclientid_cred(clp);
1214                 if (cred == NULL)
1215                         goto out;
1216         }
1217         status = ops->renew_lease(clp, cred);
1218         put_rpccred(cred);
1219 out:
1220         return nfs4_recovery_handle_error(clp, status);
1221 }
1222
1223 static int nfs4_reclaim_lease(struct nfs_client *clp)
1224 {
1225         struct rpc_cred *cred;
1226         struct nfs4_state_recovery_ops *ops =
1227                 nfs4_reboot_recovery_ops[clp->cl_minorversion];
1228         int status = -ENOENT;
1229
1230         cred = ops->get_clid_cred(clp);
1231         if (cred != NULL) {
1232                 status = ops->establish_clid(clp, cred);
1233                 put_rpccred(cred);
1234                 /* Handle case where the user hasn't set up machine creds */
1235                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1236                         nfs4_clear_machine_cred(clp);
1237                         status = -EAGAIN;
1238                 }
1239                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1240                         status = -EPROTONOSUPPORT;
1241         }
1242         return status;
1243 }
1244
1245 #ifdef CONFIG_NFS_V4_1
1246 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1247 {
1248         if (!flags)
1249                 return;
1250         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED) {
1251                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1252                 nfs4_state_start_reclaim_reboot(clp);
1253                 nfs4_schedule_state_recovery(clp);
1254         } else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1255                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1256                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1257                             SEQ4_STATUS_RECALLABLE_STATE_REVOKED |
1258                             SEQ4_STATUS_LEASE_MOVED)) {
1259                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1260                 nfs4_state_start_reclaim_nograce(clp);
1261                 nfs4_schedule_state_recovery(clp);
1262         } else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1263                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1264                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1265                 nfs_expire_all_delegations(clp);
1266 }
1267
1268 static int nfs4_reset_session(struct nfs_client *clp)
1269 {
1270         int status;
1271
1272         status = nfs4_proc_destroy_session(clp->cl_session);
1273         if (status && status != -NFS4ERR_BADSESSION &&
1274             status != -NFS4ERR_DEADSESSION) {
1275                 status = nfs4_recovery_handle_error(clp, status);
1276                 goto out;
1277         }
1278
1279         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1280         status = nfs4_proc_create_session(clp);
1281         if (status)
1282                 status = nfs4_recovery_handle_error(clp, status);
1283
1284 out:
1285         /*
1286          * Let the state manager reestablish state
1287          */
1288         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1289             status == 0)
1290                 nfs41_setup_state_renewal(clp);
1291
1292         return status;
1293 }
1294
1295 #else /* CONFIG_NFS_V4_1 */
1296 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1297 static int nfs4_begin_drain_session(struct nfs_client *clp) { return 0; }
1298 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1299 #endif /* CONFIG_NFS_V4_1 */
1300
1301 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1302  * on EXCHANGE_ID for v4.1
1303  */
1304 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1305 {
1306         if (nfs4_has_session(clp)) {
1307                 switch (status) {
1308                 case -NFS4ERR_DELAY:
1309                 case -NFS4ERR_CLID_INUSE:
1310                 case -EAGAIN:
1311                         break;
1312
1313                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1314                                          * in nfs4_exchange_id */
1315                 default:
1316                         return;
1317                 }
1318         }
1319         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1320 }
1321
1322 static void nfs4_state_manager(struct nfs_client *clp)
1323 {
1324         int status = 0;
1325
1326         /* Ensure exclusive access to NFSv4 state */
1327         for(;;) {
1328                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1329                         /* We're going to have to re-establish a clientid */
1330                         nfs4_begin_drain_session(clp);
1331                         status = nfs4_reclaim_lease(clp);
1332                         if (status) {
1333                                 nfs4_set_lease_expired(clp, status);
1334                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1335                                                         &clp->cl_state))
1336                                         continue;
1337                                 if (clp->cl_cons_state ==
1338                                                         NFS_CS_SESSION_INITING)
1339                                         nfs_mark_client_ready(clp, status);
1340                                 goto out_error;
1341                         }
1342                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1343                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1344                 }
1345
1346                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1347                         status = nfs4_check_lease(clp);
1348                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1349                                 continue;
1350                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1351                                 goto out_error;
1352                 }
1353
1354                 /* Initialize or reset the session */
1355                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1356                    && nfs4_has_session(clp)) {
1357                         nfs4_begin_drain_session(clp);
1358                         status = nfs4_reset_session(clp);
1359                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1360                                 continue;
1361                         if (status < 0)
1362                                 goto out_error;
1363                 }
1364
1365                 /* First recover reboot state... */
1366                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1367                         status = nfs4_do_reclaim(clp,
1368                                 nfs4_reboot_recovery_ops[clp->cl_minorversion]);
1369                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1370                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1371                                 continue;
1372                         nfs4_state_end_reclaim_reboot(clp);
1373                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1374                                 continue;
1375                         if (status < 0)
1376                                 goto out_error;
1377                 }
1378
1379                 /* Now recover expired state... */
1380                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1381                         status = nfs4_do_reclaim(clp,
1382                                 nfs4_nograce_recovery_ops[clp->cl_minorversion]);
1383                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1384                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1385                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1386                                 continue;
1387                         if (status < 0)
1388                                 goto out_error;
1389                 }
1390
1391                 nfs4_end_drain_session(clp);
1392                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1393                         nfs_client_return_marked_delegations(clp);
1394                         continue;
1395                 }
1396
1397                 nfs4_clear_state_manager_bit(clp);
1398                 /* Did we race with an attempt to give us more work? */
1399                 if (clp->cl_state == 0)
1400                         break;
1401                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1402                         break;
1403         }
1404         return;
1405 out_error:
1406         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1407                         " with error %d\n", clp->cl_hostname, -status);
1408         nfs4_end_drain_session(clp);
1409         nfs4_clear_state_manager_bit(clp);
1410 }
1411
1412 static int nfs4_run_state_manager(void *ptr)
1413 {
1414         struct nfs_client *clp = ptr;
1415
1416         allow_signal(SIGKILL);
1417         nfs4_state_manager(clp);
1418         nfs_put_client(clp);
1419         module_put_and_exit(0);
1420         return 0;
1421 }
1422
1423 /*
1424  * Local variables:
1425  *  c-basic-offset: 8
1426  * End:
1427  */