git.oblomov.eu Git - 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 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
  64 {
  65         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
  66                         nfs_callback_tcpport, cred);
  67         if (status == 0)
  68                 status = nfs4_proc_setclientid_confirm(clp, cred);
  69         if (status == 0)
  70                 nfs4_schedule_state_renewal(clp);
  71         return status;
  72 }
  73
  74 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
  75 {
  76         struct nfs4_state_owner *sp;
  77         struct rb_node *pos;
  78         struct rpc_cred *cred = NULL;
  79
  80         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
  81                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
  82                 if (list_empty(&sp->so_states))
  83                         continue;
  84                 cred = get_rpccred(sp->so_cred);
  85                 break;
  86         }
  87         return cred;
  88 }
  89
  90 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
  91 {
  92         struct nfs4_state_owner *sp;
  93         struct rb_node *pos;
  94
  95         pos = rb_first(&clp->cl_state_owners);
  96         if (pos != NULL) {
  97                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
  98                 return get_rpccred(sp->so_cred);
  99         }
 100         return NULL;
 101 }
 102
 103 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 104                 __u64 minval, int maxbits)
 105 {
 106         struct rb_node **p, *parent;
 107         struct nfs_unique_id *pos;
 108         __u64 mask = ~0ULL;
 109
 110         if (maxbits < 64)
 111                 mask = (1ULL << maxbits) - 1ULL;
 112
 113         /* Ensure distribution is more or less flat */
 114         get_random_bytes(&new->id, sizeof(new->id));
 115         new->id &= mask;
 116         if (new->id < minval)
 117                 new->id += minval;
 118 retry:
 119         p = &root->rb_node;
 120         parent = NULL;
 121
 122         while (*p != NULL) {
 123                 parent = *p;
 124                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 125
 126                 if (new->id < pos->id)
 127                         p = &(*p)->rb_left;
 128                 else if (new->id > pos->id)
 129                         p = &(*p)->rb_right;
 130                 else
 131                         goto id_exists;
 132         }
 133         rb_link_node(&new->rb_node, parent, p);
 134         rb_insert_color(&new->rb_node, root);
 135         return;
 136 id_exists:
 137         for (;;) {
 138                 new->id++;
 139                 if (new->id < minval || (new->id & mask) != new->id) {
 140                         new->id = minval;
 141                         break;
 142                 }
 143                 parent = rb_next(parent);
 144                 if (parent == NULL)
 145                         break;
 146                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 147                 if (new->id < pos->id)
 148                         break;
 149         }
 150         goto retry;
 151 }
 152
 153 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 154 {
 155         rb_erase(&id->rb_node, root);
 156 }
 157
 158 static struct nfs4_state_owner *
 159 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 160 {
 161         struct nfs_client *clp = server->nfs_client;
 162         struct rb_node **p = &clp->cl_state_owners.rb_node,
 163                        *parent = NULL;
 164         struct nfs4_state_owner *sp, *res = NULL;
 165
 166         while (*p != NULL) {
 167                 parent = *p;
 168                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 169
 170                 if (server < sp->so_server) {
 171                         p = &parent->rb_left;
 172                         continue;
 173                 }
 174                 if (server > sp->so_server) {
 175                         p = &parent->rb_right;
 176                         continue;
 177                 }
 178                 if (cred < sp->so_cred)
 179                         p = &parent->rb_left;
 180                 else if (cred > sp->so_cred)
 181                         p = &parent->rb_right;
 182                 else {
 183                         atomic_inc(&sp->so_count);
 184                         res = sp;
 185                         break;
 186                 }
 187         }
 188         return res;
 189 }
 190
 191 static struct nfs4_state_owner *
 192 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 193 {
 194         struct rb_node **p = &clp->cl_state_owners.rb_node,
 195                        *parent = NULL;
 196         struct nfs4_state_owner *sp;
 197
 198         while (*p != NULL) {
 199                 parent = *p;
 200                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 201
 202                 if (new->so_server < sp->so_server) {
 203                         p = &parent->rb_left;
 204                         continue;
 205                 }
 206                 if (new->so_server > sp->so_server) {
 207                         p = &parent->rb_right;
 208                         continue;
 209                 }
 210                 if (new->so_cred < sp->so_cred)
 211                         p = &parent->rb_left;
 212                 else if (new->so_cred > sp->so_cred)
 213                         p = &parent->rb_right;
 214                 else {
 215                         atomic_inc(&sp->so_count);
 216                         return sp;
 217                 }
 218         }
 219         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 220         rb_link_node(&new->so_client_node, parent, p);
 221         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 222         return new;
 223 }
 224
 225 static void
 226 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 227 {
 228         if (!RB_EMPTY_NODE(&sp->so_client_node))
 229                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 230         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 231 }
 232
 233 /*
 234  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 235  * create a new state_owner.
 236  *
 237  */
 238 static struct nfs4_state_owner *
 239 nfs4_alloc_state_owner(void)
 240 {
 241         struct nfs4_state_owner *sp;
 242
 243         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 244         if (!sp)
 245                 return NULL;
 246         spin_lock_init(&sp->so_lock);
 247         INIT_LIST_HEAD(&sp->so_states);
 248         INIT_LIST_HEAD(&sp->so_delegations);
 249         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 250         sp->so_seqid.sequence = &sp->so_sequence;
 251         spin_lock_init(&sp->so_sequence.lock);
 252         INIT_LIST_HEAD(&sp->so_sequence.list);
 253         atomic_set(&sp->so_count, 1);
 254         return sp;
 255 }
 256
 257 void
 258 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 259 {
 260         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 261                 struct nfs_client *clp = sp->so_client;
 262
 263                 spin_lock(&clp->cl_lock);
 264                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 265                 RB_CLEAR_NODE(&sp->so_client_node);
 266                 spin_unlock(&clp->cl_lock);
 267         }
 268 }
 269
 270 /*
 271  * Note: must be called with clp->cl_sem held in order to prevent races
 272  *       with reboot recovery!
 273  */
 274 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 275 {
 276         struct nfs_client *clp = server->nfs_client;
 277         struct nfs4_state_owner *sp, *new;
 278
 279         spin_lock(&clp->cl_lock);
 280         sp = nfs4_find_state_owner(server, cred);
 281         spin_unlock(&clp->cl_lock);
 282         if (sp != NULL)
 283                 return sp;
 284         new = nfs4_alloc_state_owner();
 285         if (new == NULL)
 286                 return NULL;
 287         new->so_client = clp;
 288         new->so_server = server;
 289         new->so_cred = cred;
 290         spin_lock(&clp->cl_lock);
 291         sp = nfs4_insert_state_owner(clp, new);
 292         spin_unlock(&clp->cl_lock);
 293         if (sp == new)
 294                 get_rpccred(cred);
 295         else
 296                 kfree(new);
 297         return sp;
 298 }
 299
 300 /*
 301  * Must be called with clp->cl_sem held in order to avoid races
 302  * with state recovery...
 303  */
 304 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 305 {
 306         struct nfs_client *clp = sp->so_client;
 307         struct rpc_cred *cred = sp->so_cred;
 308
 309         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 310                 return;
 311         nfs4_remove_state_owner(clp, sp);
 312         spin_unlock(&clp->cl_lock);
 313         put_rpccred(cred);
 314         kfree(sp);
 315 }
 316
 317 static struct nfs4_state *
 318 nfs4_alloc_open_state(void)
 319 {
 320         struct nfs4_state *state;
 321
 322         state = kzalloc(sizeof(*state), GFP_KERNEL);
 323         if (!state)
 324                 return NULL;
 325         atomic_set(&state->count, 1);
 326         INIT_LIST_HEAD(&state->lock_states);
 327         spin_lock_init(&state->state_lock);
 328         seqlock_init(&state->seqlock);
 329         return state;
 330 }
 331
 332 void
 333 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 334 {
 335         if (state->state == mode)
 336                 return;
 337         /* NB! List reordering - see the reclaim code for why.  */
 338         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 339                 if (mode & FMODE_WRITE)
 340                         list_move(&state->open_states, &state->owner->so_states);
 341                 else
 342                         list_move_tail(&state->open_states, &state->owner->so_states);
 343         }
 344         state->state = mode;
 345 }
 346
 347 static struct nfs4_state *
 348 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 349 {
 350         struct nfs_inode *nfsi = NFS_I(inode);
 351         struct nfs4_state *state;
 352
 353         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 354                 if (state->owner != owner)
 355                         continue;
 356                 if (atomic_inc_not_zero(&state->count))
 357                         return state;
 358         }
 359         return NULL;
 360 }
 361
 362 static void
 363 nfs4_free_open_state(struct nfs4_state *state)
 364 {
 365         kfree(state);
 366 }
 367
 368 struct nfs4_state *
 369 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 370 {
 371         struct nfs4_state *state, *new;
 372         struct nfs_inode *nfsi = NFS_I(inode);
 373
 374         spin_lock(&inode->i_lock);
 375         state = __nfs4_find_state_byowner(inode, owner);
 376         spin_unlock(&inode->i_lock);
 377         if (state)
 378                 goto out;
 379         new = nfs4_alloc_open_state();
 380         spin_lock(&owner->so_lock);
 381         spin_lock(&inode->i_lock);
 382         state = __nfs4_find_state_byowner(inode, owner);
 383         if (state == NULL && new != NULL) {
 384                 state = new;
 385                 state->owner = owner;
 386                 atomic_inc(&owner->so_count);
 387                 list_add(&state->inode_states, &nfsi->open_states);
 388                 state->inode = igrab(inode);
 389                 spin_unlock(&inode->i_lock);
 390                 /* Note: The reclaim code dictates that we add stateless
 391                  * and read-only stateids to the end of the list */
 392                 list_add_tail(&state->open_states, &owner->so_states);
 393                 spin_unlock(&owner->so_lock);
 394         } else {
 395                 spin_unlock(&inode->i_lock);
 396                 spin_unlock(&owner->so_lock);
 397                 if (new)
 398                         nfs4_free_open_state(new);
 399         }
 400 out:
 401         return state;
 402 }
 403
 404 /*
 405  * Beware! Caller must be holding exactly one
 406  * reference to clp->cl_sem!
 407  */
 408 void nfs4_put_open_state(struct nfs4_state *state)
 409 {
 410         struct inode *inode = state->inode;
 411         struct nfs4_state_owner *owner = state->owner;
 412
 413         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 414                 return;
 415         spin_lock(&inode->i_lock);
 416         list_del(&state->inode_states);
 417         list_del(&state->open_states);
 418         spin_unlock(&inode->i_lock);
 419         spin_unlock(&owner->so_lock);
 420         iput(inode);
 421         nfs4_free_open_state(state);
 422         nfs4_put_state_owner(owner);
 423 }
 424
 425 /*
 426  * Close the current file.
 427  */
 428 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
 429 {
 430         struct nfs4_state_owner *owner = state->owner;
 431         int call_close = 0;
 432         int newstate;
 433
 434         atomic_inc(&owner->so_count);
 435         /* Protect against nfs4_find_state() */
 436         spin_lock(&owner->so_lock);
 437         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 438                 case FMODE_READ:
 439                         state->n_rdonly--;
 440                         break;
 441                 case FMODE_WRITE:
 442                         state->n_wronly--;
 443                         break;
 444                 case FMODE_READ|FMODE_WRITE:
 445                         state->n_rdwr--;
 446         }
 447         newstate = FMODE_READ|FMODE_WRITE;
 448         if (state->n_rdwr == 0) {
 449                 if (state->n_rdonly == 0) {
 450                         newstate &= ~FMODE_READ;
 451                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 452                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 453                 }
 454                 if (state->n_wronly == 0) {
 455                         newstate &= ~FMODE_WRITE;
 456                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 457                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 458                 }
 459                 if (newstate == 0)
 460                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 461         }
 462         nfs4_state_set_mode_locked(state, newstate);
 463         spin_unlock(&owner->so_lock);
 464
 465         if (!call_close) {
 466                 nfs4_put_open_state(state);
 467                 nfs4_put_state_owner(owner);
 468         } else
 469                 nfs4_do_close(path, state, wait);
 470 }
 471
 472 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
 473 {
 474         __nfs4_close(path, state, mode, 0);
 475 }
 476
 477 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
 478 {
 479         __nfs4_close(path, state, mode, 1);
 480 }
 481
 482 /*
 483  * Search the state->lock_states for an existing lock_owner
 484  * that is compatible with current->files
 485  */
 486 static struct nfs4_lock_state *
 487 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 488 {
 489         struct nfs4_lock_state *pos;
 490         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 491                 if (pos->ls_owner != fl_owner)
 492                         continue;
 493                 atomic_inc(&pos->ls_count);
 494                 return pos;
 495         }
 496         return NULL;
 497 }
 498
 499 /*
 500  * Return a compatible lock_state. If no initialized lock_state structure
 501  * exists, return an uninitialized one.
 502  *
 503  */
 504 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 505 {
 506         struct nfs4_lock_state *lsp;
 507         struct nfs_client *clp = state->owner->so_client;
 508
 509         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 510         if (lsp == NULL)
 511                 return NULL;
 512         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 513         spin_lock_init(&lsp->ls_sequence.lock);
 514         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 515         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 516         atomic_set(&lsp->ls_count, 1);
 517         lsp->ls_owner = fl_owner;
 518         spin_lock(&clp->cl_lock);
 519         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 520         spin_unlock(&clp->cl_lock);
 521         INIT_LIST_HEAD(&lsp->ls_locks);
 522         return lsp;
 523 }
 524
 525 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 526 {
 527         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 528
 529         spin_lock(&clp->cl_lock);
 530         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 531         spin_unlock(&clp->cl_lock);
 532         kfree(lsp);
 533 }
 534
 535 /*
 536  * Return a compatible lock_state. If no initialized lock_state structure
 537  * exists, return an uninitialized one.
 538  *
 539  * The caller must be holding clp->cl_sem
 540  */
 541 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 542 {
 543         struct nfs4_lock_state *lsp, *new = NULL;
 544
 545         for(;;) {
 546                 spin_lock(&state->state_lock);
 547                 lsp = __nfs4_find_lock_state(state, owner);
 548                 if (lsp != NULL)
 549                         break;
 550                 if (new != NULL) {
 551                         new->ls_state = state;
 552                         list_add(&new->ls_locks, &state->lock_states);
 553                         set_bit(LK_STATE_IN_USE, &state->flags);
 554                         lsp = new;
 555                         new = NULL;
 556                         break;
 557                 }
 558                 spin_unlock(&state->state_lock);
 559                 new = nfs4_alloc_lock_state(state, owner);
 560                 if (new == NULL)
 561                         return NULL;
 562         }
 563         spin_unlock(&state->state_lock);
 564         if (new != NULL)
 565                 nfs4_free_lock_state(new);
 566         return lsp;
 567 }
 568
 569 /*
 570  * Release reference to lock_state, and free it if we see that
 571  * it is no longer in use
 572  */
 573 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 574 {
 575         struct nfs4_state *state;
 576
 577         if (lsp == NULL)
 578                 return;
 579         state = lsp->ls_state;
 580         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 581                 return;
 582         list_del(&lsp->ls_locks);
 583         if (list_empty(&state->lock_states))
 584                 clear_bit(LK_STATE_IN_USE, &state->flags);
 585         spin_unlock(&state->state_lock);
 586         nfs4_free_lock_state(lsp);
 587 }
 588
 589 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 590 {
 591         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 592
 593         dst->fl_u.nfs4_fl.owner = lsp;
 594         atomic_inc(&lsp->ls_count);
 595 }
 596
 597 static void nfs4_fl_release_lock(struct file_lock *fl)
 598 {
 599         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 600 }
 601
 602 static struct file_lock_operations nfs4_fl_lock_ops = {
 603         .fl_copy_lock = nfs4_fl_copy_lock,
 604         .fl_release_private = nfs4_fl_release_lock,
 605 };
 606
 607 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 608 {
 609         struct nfs4_lock_state *lsp;
 610
 611         if (fl->fl_ops != NULL)
 612                 return 0;
 613         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 614         if (lsp == NULL)
 615                 return -ENOMEM;
 616         fl->fl_u.nfs4_fl.owner = lsp;
 617         fl->fl_ops = &nfs4_fl_lock_ops;
 618         return 0;
 619 }
 620
 621 /*
 622  * Byte-range lock aware utility to initialize the stateid of read/write
 623  * requests.
 624  */
 625 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 626 {
 627         struct nfs4_lock_state *lsp;
 628         int seq;
 629
 630         do {
 631                 seq = read_seqbegin(&state->seqlock);
 632                 memcpy(dst, &state->stateid, sizeof(*dst));
 633         } while (read_seqretry(&state->seqlock, seq));
 634         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 635                 return;
 636
 637         spin_lock(&state->state_lock);
 638         lsp = __nfs4_find_lock_state(state, fl_owner);
 639         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 640                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 641         spin_unlock(&state->state_lock);
 642         nfs4_put_lock_state(lsp);
 643 }
 644
 645 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 646 {
 647         struct rpc_sequence *sequence = counter->sequence;
 648         struct nfs_seqid *new;
 649
 650         new = kmalloc(sizeof(*new), GFP_KERNEL);
 651         if (new != NULL) {
 652                 new->sequence = counter;
 653                 spin_lock(&sequence->lock);
 654                 list_add_tail(&new->list, &sequence->list);
 655                 spin_unlock(&sequence->lock);
 656         }
 657         return new;
 658 }
 659
 660 void nfs_free_seqid(struct nfs_seqid *seqid)
 661 {
 662         struct rpc_sequence *sequence = seqid->sequence->sequence;
 663
 664         spin_lock(&sequence->lock);
 665         list_del(&seqid->list);
 666         spin_unlock(&sequence->lock);
 667         rpc_wake_up(&sequence->wait);
 668         kfree(seqid);
 669 }
 670
 671 /*
 672  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 673  * failed with a seqid incrementing error -
 674  * see comments nfs_fs.h:seqid_mutating_error()
 675  */
 676 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 677 {
 678         switch (status) {
 679                 case 0:
 680                         break;
 681                 case -NFS4ERR_BAD_SEQID:
 682                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 683                                 return;
 684                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 685                                         "sequence-id error on an"
 686                                         "unconfirmed sequence %p!\n",
 687                                         seqid->sequence);
 688                 case -NFS4ERR_STALE_CLIENTID:
 689                 case -NFS4ERR_STALE_STATEID:
 690                 case -NFS4ERR_BAD_STATEID:
 691                 case -NFS4ERR_BADXDR:
 692                 case -NFS4ERR_RESOURCE:
 693                 case -NFS4ERR_NOFILEHANDLE:
 694                         /* Non-seqid mutating errors */
 695                         return;
 696         };
 697         /*
 698          * Note: no locking needed as we are guaranteed to be first
 699          * on the sequence list
 700          */
 701         seqid->sequence->counter++;
 702 }
 703
 704 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 705 {
 706         if (status == -NFS4ERR_BAD_SEQID) {
 707                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 708                                 struct nfs4_state_owner, so_seqid);
 709                 nfs4_drop_state_owner(sp);
 710         }
 711         nfs_increment_seqid(status, seqid);
 712 }
 713
 714 /*
 715  * Increment the seqid if the LOCK/LOCKU succeeded, or
 716  * failed with a seqid incrementing error -
 717  * see comments nfs_fs.h:seqid_mutating_error()
 718  */
 719 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 720 {
 721         nfs_increment_seqid(status, seqid);
 722 }
 723
 724 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 725 {
 726         struct rpc_sequence *sequence = seqid->sequence->sequence;
 727         int status = 0;
 728
 729         if (sequence->list.next == &seqid->list)
 730                 goto out;
 731         spin_lock(&sequence->lock);
 732         if (sequence->list.next != &seqid->list) {
 733                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 734                 status = -EAGAIN;
 735         }
 736         spin_unlock(&sequence->lock);
 737 out:
 738         return status;
 739 }
 740
 741 static int reclaimer(void *);
 742
 743 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 744 {
 745         smp_mb__before_clear_bit();
 746         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 747         smp_mb__after_clear_bit();
 748         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 749         rpc_wake_up(&clp->cl_rpcwaitq);
 750 }
 751
 752 /*
 753  * State recovery routine
 754  */
 755 static void nfs4_recover_state(struct nfs_client *clp)
 756 {
 757         struct task_struct *task;
 758
 759         __module_get(THIS_MODULE);
 760         atomic_inc(&clp->cl_count);
 761         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 762                         NIPQUAD(clp->cl_addr.sin_addr));
 763         if (!IS_ERR(task))
 764                 return;
 765         nfs4_clear_recover_bit(clp);
 766         nfs_put_client(clp);
 767         module_put(THIS_MODULE);
 768 }
 769
 770 /*
 771  * Schedule a state recovery attempt
 772  */
 773 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 774 {
 775         if (!clp)
 776                 return;
 777         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 778                 nfs4_recover_state(clp);
 779 }
 780
 781 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 782 {
 783         struct inode *inode = state->inode;
 784         struct file_lock *fl;
 785         int status = 0;
 786
 787         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 788                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 789                         continue;
 790                 if (nfs_file_open_context(fl->fl_file)->state != state)
 791                         continue;
 792                 status = ops->recover_lock(state, fl);
 793                 if (status >= 0)
 794                         continue;
 795                 switch (status) {
 796                         default:
 797                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 798                                                 __FUNCTION__, status);
 799                         case -NFS4ERR_EXPIRED:
 800                         case -NFS4ERR_NO_GRACE:
 801                         case -NFS4ERR_RECLAIM_BAD:
 802                         case -NFS4ERR_RECLAIM_CONFLICT:
 803                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 804                                 break;
 805                         case -NFS4ERR_STALE_CLIENTID:
 806                                 goto out_err;
 807                 }
 808         }
 809         return 0;
 810 out_err:
 811         return status;
 812 }
 813
 814 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 815 {
 816         struct nfs4_state *state;
 817         struct nfs4_lock_state *lock;
 818         int status = 0;
 819
 820         /* Note: we rely on the sp->so_states list being ordered
 821          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 822          * states first.
 823          * This is needed to ensure that the server won't give us any
 824          * read delegations that we have to return if, say, we are
 825          * recovering after a network partition or a reboot from a
 826          * server that doesn't support a grace period.
 827          */
 828         list_for_each_entry(state, &sp->so_states, open_states) {
 829                 if (state->state == 0)
 830                         continue;
 831                 status = ops->recover_open(sp, state);
 832                 if (status >= 0) {
 833                         status = nfs4_reclaim_locks(ops, state);
 834                         if (status < 0)
 835                                 goto out_err;
 836                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 837                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 838                                         printk("%s: Lock reclaim failed!\n",
 839                                                         __FUNCTION__);
 840                         }
 841                         continue;
 842                 }
 843                 switch (status) {
 844                         default:
 845                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 846                                                 __FUNCTION__, status);
 847                         case -ENOENT:
 848                         case -NFS4ERR_RECLAIM_BAD:
 849                         case -NFS4ERR_RECLAIM_CONFLICT:
 850                                 /*
 851                                  * Open state on this file cannot be recovered
 852                                  * All we can do is revert to using the zero stateid.
 853                                  */
 854                                 memset(state->stateid.data, 0,
 855                                         sizeof(state->stateid.data));
 856                                 /* Mark the file as being 'closed' */
 857                                 state->state = 0;
 858                                 break;
 859                         case -NFS4ERR_EXPIRED:
 860                         case -NFS4ERR_NO_GRACE:
 861                         case -NFS4ERR_STALE_CLIENTID:
 862                                 goto out_err;
 863                 }
 864         }
 865         return 0;
 866 out_err:
 867         return status;
 868 }
 869
 870 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 871 {
 872         struct nfs4_state_owner *sp;
 873         struct rb_node *pos;
 874         struct nfs4_state *state;
 875         struct nfs4_lock_state *lock;
 876
 877         /* Reset all sequence ids to zero */
 878         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 879                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 880                 sp->so_seqid.counter = 0;
 881                 sp->so_seqid.flags = 0;
 882                 spin_lock(&sp->so_lock);
 883                 list_for_each_entry(state, &sp->so_states, open_states) {
 884                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 885                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
 886                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
 887                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
 888                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 889                                 lock->ls_seqid.counter = 0;
 890                                 lock->ls_seqid.flags = 0;
 891                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 892                         }
 893                 }
 894                 spin_unlock(&sp->so_lock);
 895         }
 896 }
 897
 898 static int reclaimer(void *ptr)
 899 {
 900         struct nfs_client *clp = ptr;
 901         struct nfs4_state_owner *sp;
 902         struct rb_node *pos;
 903         struct nfs4_state_recovery_ops *ops;
 904         struct rpc_cred *cred;
 905         int status = 0;
 906
 907         allow_signal(SIGKILL);
 908
 909         /* Ensure exclusive access to NFSv4 state */
 910         lock_kernel();
 911         down_write(&clp->cl_sem);
 912         /* Are there any NFS mounts out there? */
 913         if (list_empty(&clp->cl_superblocks))
 914                 goto out;
 915 restart_loop:
 916         ops = &nfs4_network_partition_recovery_ops;
 917         /* Are there any open files on this volume? */
 918         cred = nfs4_get_renew_cred(clp);
 919         if (cred != NULL) {
 920                 /* Yes there are: try to renew the old lease */
 921                 status = nfs4_proc_renew(clp, cred);
 922                 switch (status) {
 923                         case 0:
 924                         case -NFS4ERR_CB_PATH_DOWN:
 925                                 put_rpccred(cred);
 926                                 goto out;
 927                         case -NFS4ERR_STALE_CLIENTID:
 928                         case -NFS4ERR_LEASE_MOVED:
 929                                 ops = &nfs4_reboot_recovery_ops;
 930                 }
 931         } else {
 932                 /* "reboot" to ensure we clear all state on the server */
 933                 clp->cl_boot_time = CURRENT_TIME;
 934                 cred = nfs4_get_setclientid_cred(clp);
 935         }
 936         /* We're going to have to re-establish a clientid */
 937         nfs4_state_mark_reclaim(clp);
 938         status = -ENOENT;
 939         if (cred != NULL) {
 940                 status = nfs4_init_client(clp, cred);
 941                 put_rpccred(cred);
 942         }
 943         if (status)
 944                 goto out_error;
 945         /* Mark all delegations for reclaim */
 946         nfs_delegation_mark_reclaim(clp);
 947         /* Note: list is protected by exclusive lock on cl->cl_sem */
 948         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 949                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 950                 status = nfs4_reclaim_open_state(ops, sp);
 951                 if (status < 0) {
 952                         if (status == -NFS4ERR_NO_GRACE) {
 953                                 ops = &nfs4_network_partition_recovery_ops;
 954                                 status = nfs4_reclaim_open_state(ops, sp);
 955                         }
 956                         if (status == -NFS4ERR_STALE_CLIENTID)
 957                                 goto restart_loop;
 958                         if (status == -NFS4ERR_EXPIRED)
 959                                 goto restart_loop;
 960                 }
 961         }
 962         nfs_delegation_reap_unclaimed(clp);
 963 out:
 964         up_write(&clp->cl_sem);
 965         unlock_kernel();
 966         if (status == -NFS4ERR_CB_PATH_DOWN)
 967                 nfs_handle_cb_pathdown(clp);
 968         nfs4_clear_recover_bit(clp);
 969         nfs_put_client(clp);
 970         module_put_and_exit(0);
 971         return 0;
 972 out_error:
 973         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 974                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
 975         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 976         goto out;
 977 }
 978
 979 /*
 980  * Local variables:
 981  *  c-basic-offset: 8
 982  * End:
 983  */