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 static struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  75 {
  76         struct rpc_cred *cred = NULL;
  77
  78         if (clp->cl_machine_cred != NULL)
  79                 cred = get_rpccred(clp->cl_machine_cred);
  80         return cred;
  81 }
  82
  83 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  84 {
  85         struct rpc_cred *cred;
  86
  87         spin_lock(&clp->cl_lock);
  88         cred = clp->cl_machine_cred;
  89         clp->cl_machine_cred = NULL;
  90         spin_unlock(&clp->cl_lock);
  91         if (cred != NULL)
  92                 put_rpccred(cred);
  93 }
  94
  95 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
  96 {
  97         struct nfs4_state_owner *sp;
  98         struct rb_node *pos;
  99         struct rpc_cred *cred = NULL;
 100
 101         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 102                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 103                 if (list_empty(&sp->so_states))
 104                         continue;
 105                 cred = get_rpccred(sp->so_cred);
 106                 break;
 107         }
 108         return cred;
 109 }
 110
 111 static struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
 112 {
 113         struct rpc_cred *cred;
 114
 115         spin_lock(&clp->cl_lock);
 116         cred = nfs4_get_renew_cred_locked(clp);
 117         spin_unlock(&clp->cl_lock);
 118         return cred;
 119 }
 120
 121 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 122 {
 123         struct nfs4_state_owner *sp;
 124         struct rb_node *pos;
 125         struct rpc_cred *cred;
 126
 127         spin_lock(&clp->cl_lock);
 128         cred = nfs4_get_machine_cred_locked(clp);
 129         if (cred != NULL)
 130                 goto out;
 131         pos = rb_first(&clp->cl_state_owners);
 132         if (pos != NULL) {
 133                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 134                 cred = get_rpccred(sp->so_cred);
 135         }
 136 out:
 137         spin_unlock(&clp->cl_lock);
 138         return cred;
 139 }
 140
 141 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 142                 __u64 minval, int maxbits)
 143 {
 144         struct rb_node **p, *parent;
 145         struct nfs_unique_id *pos;
 146         __u64 mask = ~0ULL;
 147
 148         if (maxbits < 64)
 149                 mask = (1ULL << maxbits) - 1ULL;
 150
 151         /* Ensure distribution is more or less flat */
 152         get_random_bytes(&new->id, sizeof(new->id));
 153         new->id &= mask;
 154         if (new->id < minval)
 155                 new->id += minval;
 156 retry:
 157         p = &root->rb_node;
 158         parent = NULL;
 159
 160         while (*p != NULL) {
 161                 parent = *p;
 162                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 163
 164                 if (new->id < pos->id)
 165                         p = &(*p)->rb_left;
 166                 else if (new->id > pos->id)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         goto id_exists;
 170         }
 171         rb_link_node(&new->rb_node, parent, p);
 172         rb_insert_color(&new->rb_node, root);
 173         return;
 174 id_exists:
 175         for (;;) {
 176                 new->id++;
 177                 if (new->id < minval || (new->id & mask) != new->id) {
 178                         new->id = minval;
 179                         break;
 180                 }
 181                 parent = rb_next(parent);
 182                 if (parent == NULL)
 183                         break;
 184                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 185                 if (new->id < pos->id)
 186                         break;
 187         }
 188         goto retry;
 189 }
 190
 191 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 192 {
 193         rb_erase(&id->rb_node, root);
 194 }
 195
 196 static struct nfs4_state_owner *
 197 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 198 {
 199         struct nfs_client *clp = server->nfs_client;
 200         struct rb_node **p = &clp->cl_state_owners.rb_node,
 201                        *parent = NULL;
 202         struct nfs4_state_owner *sp, *res = NULL;
 203
 204         while (*p != NULL) {
 205                 parent = *p;
 206                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 207
 208                 if (server < sp->so_server) {
 209                         p = &parent->rb_left;
 210                         continue;
 211                 }
 212                 if (server > sp->so_server) {
 213                         p = &parent->rb_right;
 214                         continue;
 215                 }
 216                 if (cred < sp->so_cred)
 217                         p = &parent->rb_left;
 218                 else if (cred > sp->so_cred)
 219                         p = &parent->rb_right;
 220                 else {
 221                         atomic_inc(&sp->so_count);
 222                         res = sp;
 223                         break;
 224                 }
 225         }
 226         return res;
 227 }
 228
 229 static struct nfs4_state_owner *
 230 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 231 {
 232         struct rb_node **p = &clp->cl_state_owners.rb_node,
 233                        *parent = NULL;
 234         struct nfs4_state_owner *sp;
 235
 236         while (*p != NULL) {
 237                 parent = *p;
 238                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 239
 240                 if (new->so_server < sp->so_server) {
 241                         p = &parent->rb_left;
 242                         continue;
 243                 }
 244                 if (new->so_server > sp->so_server) {
 245                         p = &parent->rb_right;
 246                         continue;
 247                 }
 248                 if (new->so_cred < sp->so_cred)
 249                         p = &parent->rb_left;
 250                 else if (new->so_cred > sp->so_cred)
 251                         p = &parent->rb_right;
 252                 else {
 253                         atomic_inc(&sp->so_count);
 254                         return sp;
 255                 }
 256         }
 257         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 258         rb_link_node(&new->so_client_node, parent, p);
 259         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 260         return new;
 261 }
 262
 263 static void
 264 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 265 {
 266         if (!RB_EMPTY_NODE(&sp->so_client_node))
 267                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 268         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 269 }
 270
 271 /*
 272  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 273  * create a new state_owner.
 274  *
 275  */
 276 static struct nfs4_state_owner *
 277 nfs4_alloc_state_owner(void)
 278 {
 279         struct nfs4_state_owner *sp;
 280
 281         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 282         if (!sp)
 283                 return NULL;
 284         spin_lock_init(&sp->so_lock);
 285         INIT_LIST_HEAD(&sp->so_states);
 286         INIT_LIST_HEAD(&sp->so_delegations);
 287         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 288         sp->so_seqid.sequence = &sp->so_sequence;
 289         spin_lock_init(&sp->so_sequence.lock);
 290         INIT_LIST_HEAD(&sp->so_sequence.list);
 291         atomic_set(&sp->so_count, 1);
 292         return sp;
 293 }
 294
 295 static void
 296 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 297 {
 298         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 299                 struct nfs_client *clp = sp->so_client;
 300
 301                 spin_lock(&clp->cl_lock);
 302                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 303                 RB_CLEAR_NODE(&sp->so_client_node);
 304                 spin_unlock(&clp->cl_lock);
 305         }
 306 }
 307
 308 /*
 309  * Note: must be called with clp->cl_sem held in order to prevent races
 310  *       with reboot recovery!
 311  */
 312 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 313 {
 314         struct nfs_client *clp = server->nfs_client;
 315         struct nfs4_state_owner *sp, *new;
 316
 317         spin_lock(&clp->cl_lock);
 318         sp = nfs4_find_state_owner(server, cred);
 319         spin_unlock(&clp->cl_lock);
 320         if (sp != NULL)
 321                 return sp;
 322         new = nfs4_alloc_state_owner();
 323         if (new == NULL)
 324                 return NULL;
 325         new->so_client = clp;
 326         new->so_server = server;
 327         new->so_cred = cred;
 328         spin_lock(&clp->cl_lock);
 329         sp = nfs4_insert_state_owner(clp, new);
 330         spin_unlock(&clp->cl_lock);
 331         if (sp == new)
 332                 get_rpccred(cred);
 333         else {
 334                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 335                 kfree(new);
 336         }
 337         return sp;
 338 }
 339
 340 /*
 341  * Must be called with clp->cl_sem held in order to avoid races
 342  * with state recovery...
 343  */
 344 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 345 {
 346         struct nfs_client *clp = sp->so_client;
 347         struct rpc_cred *cred = sp->so_cred;
 348
 349         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 350                 return;
 351         nfs4_remove_state_owner(clp, sp);
 352         spin_unlock(&clp->cl_lock);
 353         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 354         put_rpccred(cred);
 355         kfree(sp);
 356 }
 357
 358 static struct nfs4_state *
 359 nfs4_alloc_open_state(void)
 360 {
 361         struct nfs4_state *state;
 362
 363         state = kzalloc(sizeof(*state), GFP_KERNEL);
 364         if (!state)
 365                 return NULL;
 366         atomic_set(&state->count, 1);
 367         INIT_LIST_HEAD(&state->lock_states);
 368         spin_lock_init(&state->state_lock);
 369         seqlock_init(&state->seqlock);
 370         return state;
 371 }
 372
 373 void
 374 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 375 {
 376         if (state->state == mode)
 377                 return;
 378         /* NB! List reordering - see the reclaim code for why.  */
 379         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 380                 if (mode & FMODE_WRITE)
 381                         list_move(&state->open_states, &state->owner->so_states);
 382                 else
 383                         list_move_tail(&state->open_states, &state->owner->so_states);
 384         }
 385         state->state = mode;
 386 }
 387
 388 static struct nfs4_state *
 389 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 390 {
 391         struct nfs_inode *nfsi = NFS_I(inode);
 392         struct nfs4_state *state;
 393
 394         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 395                 if (state->owner != owner)
 396                         continue;
 397                 if (atomic_inc_not_zero(&state->count))
 398                         return state;
 399         }
 400         return NULL;
 401 }
 402
 403 static void
 404 nfs4_free_open_state(struct nfs4_state *state)
 405 {
 406         kfree(state);
 407 }
 408
 409 struct nfs4_state *
 410 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 411 {
 412         struct nfs4_state *state, *new;
 413         struct nfs_inode *nfsi = NFS_I(inode);
 414
 415         spin_lock(&inode->i_lock);
 416         state = __nfs4_find_state_byowner(inode, owner);
 417         spin_unlock(&inode->i_lock);
 418         if (state)
 419                 goto out;
 420         new = nfs4_alloc_open_state();
 421         spin_lock(&owner->so_lock);
 422         spin_lock(&inode->i_lock);
 423         state = __nfs4_find_state_byowner(inode, owner);
 424         if (state == NULL && new != NULL) {
 425                 state = new;
 426                 state->owner = owner;
 427                 atomic_inc(&owner->so_count);
 428                 list_add(&state->inode_states, &nfsi->open_states);
 429                 state->inode = igrab(inode);
 430                 spin_unlock(&inode->i_lock);
 431                 /* Note: The reclaim code dictates that we add stateless
 432                  * and read-only stateids to the end of the list */
 433                 list_add_tail(&state->open_states, &owner->so_states);
 434                 spin_unlock(&owner->so_lock);
 435         } else {
 436                 spin_unlock(&inode->i_lock);
 437                 spin_unlock(&owner->so_lock);
 438                 if (new)
 439                         nfs4_free_open_state(new);
 440         }
 441 out:
 442         return state;
 443 }
 444
 445 /*
 446  * Beware! Caller must be holding exactly one
 447  * reference to clp->cl_sem!
 448  */
 449 void nfs4_put_open_state(struct nfs4_state *state)
 450 {
 451         struct inode *inode = state->inode;
 452         struct nfs4_state_owner *owner = state->owner;
 453
 454         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 455                 return;
 456         spin_lock(&inode->i_lock);
 457         list_del(&state->inode_states);
 458         list_del(&state->open_states);
 459         spin_unlock(&inode->i_lock);
 460         spin_unlock(&owner->so_lock);
 461         iput(inode);
 462         nfs4_free_open_state(state);
 463         nfs4_put_state_owner(owner);
 464 }
 465
 466 /*
 467  * Close the current file.
 468  */
 469 static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
 470 {
 471         struct nfs4_state_owner *owner = state->owner;
 472         int call_close = 0;
 473         int newstate;
 474
 475         atomic_inc(&owner->so_count);
 476         /* Protect against nfs4_find_state() */
 477         spin_lock(&owner->so_lock);
 478         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 479                 case FMODE_READ:
 480                         state->n_rdonly--;
 481                         break;
 482                 case FMODE_WRITE:
 483                         state->n_wronly--;
 484                         break;
 485                 case FMODE_READ|FMODE_WRITE:
 486                         state->n_rdwr--;
 487         }
 488         newstate = FMODE_READ|FMODE_WRITE;
 489         if (state->n_rdwr == 0) {
 490                 if (state->n_rdonly == 0) {
 491                         newstate &= ~FMODE_READ;
 492                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 493                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 494                 }
 495                 if (state->n_wronly == 0) {
 496                         newstate &= ~FMODE_WRITE;
 497                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 498                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 499                 }
 500                 if (newstate == 0)
 501                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 502         }
 503         nfs4_state_set_mode_locked(state, newstate);
 504         spin_unlock(&owner->so_lock);
 505
 506         if (!call_close) {
 507                 nfs4_put_open_state(state);
 508                 nfs4_put_state_owner(owner);
 509         } else
 510                 nfs4_do_close(path, state, wait);
 511 }
 512
 513 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
 514 {
 515         __nfs4_close(path, state, mode, 0);
 516 }
 517
 518 void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
 519 {
 520         __nfs4_close(path, state, mode, 1);
 521 }
 522
 523 /*
 524  * Search the state->lock_states for an existing lock_owner
 525  * that is compatible with current->files
 526  */
 527 static struct nfs4_lock_state *
 528 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 529 {
 530         struct nfs4_lock_state *pos;
 531         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 532                 if (pos->ls_owner != fl_owner)
 533                         continue;
 534                 atomic_inc(&pos->ls_count);
 535                 return pos;
 536         }
 537         return NULL;
 538 }
 539
 540 /*
 541  * Return a compatible lock_state. If no initialized lock_state structure
 542  * exists, return an uninitialized one.
 543  *
 544  */
 545 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 546 {
 547         struct nfs4_lock_state *lsp;
 548         struct nfs_client *clp = state->owner->so_client;
 549
 550         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 551         if (lsp == NULL)
 552                 return NULL;
 553         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 554         spin_lock_init(&lsp->ls_sequence.lock);
 555         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 556         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 557         atomic_set(&lsp->ls_count, 1);
 558         lsp->ls_owner = fl_owner;
 559         spin_lock(&clp->cl_lock);
 560         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 561         spin_unlock(&clp->cl_lock);
 562         INIT_LIST_HEAD(&lsp->ls_locks);
 563         return lsp;
 564 }
 565
 566 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 567 {
 568         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 569
 570         spin_lock(&clp->cl_lock);
 571         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 572         spin_unlock(&clp->cl_lock);
 573         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 574         kfree(lsp);
 575 }
 576
 577 /*
 578  * Return a compatible lock_state. If no initialized lock_state structure
 579  * exists, return an uninitialized one.
 580  *
 581  * The caller must be holding clp->cl_sem
 582  */
 583 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 584 {
 585         struct nfs4_lock_state *lsp, *new = NULL;
 586
 587         for(;;) {
 588                 spin_lock(&state->state_lock);
 589                 lsp = __nfs4_find_lock_state(state, owner);
 590                 if (lsp != NULL)
 591                         break;
 592                 if (new != NULL) {
 593                         new->ls_state = state;
 594                         list_add(&new->ls_locks, &state->lock_states);
 595                         set_bit(LK_STATE_IN_USE, &state->flags);
 596                         lsp = new;
 597                         new = NULL;
 598                         break;
 599                 }
 600                 spin_unlock(&state->state_lock);
 601                 new = nfs4_alloc_lock_state(state, owner);
 602                 if (new == NULL)
 603                         return NULL;
 604         }
 605         spin_unlock(&state->state_lock);
 606         if (new != NULL)
 607                 nfs4_free_lock_state(new);
 608         return lsp;
 609 }
 610
 611 /*
 612  * Release reference to lock_state, and free it if we see that
 613  * it is no longer in use
 614  */
 615 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 616 {
 617         struct nfs4_state *state;
 618
 619         if (lsp == NULL)
 620                 return;
 621         state = lsp->ls_state;
 622         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 623                 return;
 624         list_del(&lsp->ls_locks);
 625         if (list_empty(&state->lock_states))
 626                 clear_bit(LK_STATE_IN_USE, &state->flags);
 627         spin_unlock(&state->state_lock);
 628         nfs4_free_lock_state(lsp);
 629 }
 630
 631 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 632 {
 633         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 634
 635         dst->fl_u.nfs4_fl.owner = lsp;
 636         atomic_inc(&lsp->ls_count);
 637 }
 638
 639 static void nfs4_fl_release_lock(struct file_lock *fl)
 640 {
 641         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 642 }
 643
 644 static struct file_lock_operations nfs4_fl_lock_ops = {
 645         .fl_copy_lock = nfs4_fl_copy_lock,
 646         .fl_release_private = nfs4_fl_release_lock,
 647 };
 648
 649 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 650 {
 651         struct nfs4_lock_state *lsp;
 652
 653         if (fl->fl_ops != NULL)
 654                 return 0;
 655         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 656         if (lsp == NULL)
 657                 return -ENOMEM;
 658         fl->fl_u.nfs4_fl.owner = lsp;
 659         fl->fl_ops = &nfs4_fl_lock_ops;
 660         return 0;
 661 }
 662
 663 /*
 664  * Byte-range lock aware utility to initialize the stateid of read/write
 665  * requests.
 666  */
 667 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 668 {
 669         struct nfs4_lock_state *lsp;
 670         int seq;
 671
 672         do {
 673                 seq = read_seqbegin(&state->seqlock);
 674                 memcpy(dst, &state->stateid, sizeof(*dst));
 675         } while (read_seqretry(&state->seqlock, seq));
 676         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 677                 return;
 678
 679         spin_lock(&state->state_lock);
 680         lsp = __nfs4_find_lock_state(state, fl_owner);
 681         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 682                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 683         spin_unlock(&state->state_lock);
 684         nfs4_put_lock_state(lsp);
 685 }
 686
 687 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 688 {
 689         struct nfs_seqid *new;
 690
 691         new = kmalloc(sizeof(*new), GFP_KERNEL);
 692         if (new != NULL) {
 693                 new->sequence = counter;
 694                 INIT_LIST_HEAD(&new->list);
 695         }
 696         return new;
 697 }
 698
 699 void nfs_free_seqid(struct nfs_seqid *seqid)
 700 {
 701         if (!list_empty(&seqid->list)) {
 702                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 703
 704                 spin_lock(&sequence->lock);
 705                 list_del(&seqid->list);
 706                 spin_unlock(&sequence->lock);
 707                 rpc_wake_up(&sequence->wait);
 708         }
 709         kfree(seqid);
 710 }
 711
 712 /*
 713  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 714  * failed with a seqid incrementing error -
 715  * see comments nfs_fs.h:seqid_mutating_error()
 716  */
 717 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 718 {
 719         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 720         switch (status) {
 721                 case 0:
 722                         break;
 723                 case -NFS4ERR_BAD_SEQID:
 724                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 725                                 return;
 726                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 727                                         " sequence-id error on an"
 728                                         " unconfirmed sequence %p!\n",
 729                                         seqid->sequence);
 730                 case -NFS4ERR_STALE_CLIENTID:
 731                 case -NFS4ERR_STALE_STATEID:
 732                 case -NFS4ERR_BAD_STATEID:
 733                 case -NFS4ERR_BADXDR:
 734                 case -NFS4ERR_RESOURCE:
 735                 case -NFS4ERR_NOFILEHANDLE:
 736                         /* Non-seqid mutating errors */
 737                         return;
 738         };
 739         /*
 740          * Note: no locking needed as we are guaranteed to be first
 741          * on the sequence list
 742          */
 743         seqid->sequence->counter++;
 744 }
 745
 746 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 747 {
 748         if (status == -NFS4ERR_BAD_SEQID) {
 749                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 750                                 struct nfs4_state_owner, so_seqid);
 751                 nfs4_drop_state_owner(sp);
 752         }
 753         nfs_increment_seqid(status, seqid);
 754 }
 755
 756 /*
 757  * Increment the seqid if the LOCK/LOCKU succeeded, or
 758  * failed with a seqid incrementing error -
 759  * see comments nfs_fs.h:seqid_mutating_error()
 760  */
 761 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 762 {
 763         nfs_increment_seqid(status, seqid);
 764 }
 765
 766 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 767 {
 768         struct rpc_sequence *sequence = seqid->sequence->sequence;
 769         int status = 0;
 770
 771         spin_lock(&sequence->lock);
 772         if (list_empty(&seqid->list))
 773                 list_add_tail(&seqid->list, &sequence->list);
 774         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 775                 goto unlock;
 776         rpc_sleep_on(&sequence->wait, task, NULL);
 777         status = -EAGAIN;
 778 unlock:
 779         spin_unlock(&sequence->lock);
 780         return status;
 781 }
 782
 783 static int reclaimer(void *);
 784
 785 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 786 {
 787         smp_mb__before_clear_bit();
 788         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 789         smp_mb__after_clear_bit();
 790         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 791         rpc_wake_up(&clp->cl_rpcwaitq);
 792 }
 793
 794 /*
 795  * State recovery routine
 796  */
 797 static void nfs4_recover_state(struct nfs_client *clp)
 798 {
 799         struct task_struct *task;
 800
 801         __module_get(THIS_MODULE);
 802         atomic_inc(&clp->cl_count);
 803         task = kthread_run(reclaimer, clp, "%s-reclaim",
 804                                 rpc_peeraddr2str(clp->cl_rpcclient,
 805                                                         RPC_DISPLAY_ADDR));
 806         if (!IS_ERR(task))
 807                 return;
 808         nfs4_clear_recover_bit(clp);
 809         nfs_put_client(clp);
 810         module_put(THIS_MODULE);
 811 }
 812
 813 /*
 814  * Schedule a state recovery attempt
 815  */
 816 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 817 {
 818         if (!clp)
 819                 return;
 820         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 821                 nfs4_recover_state(clp);
 822 }
 823
 824 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 825 {
 826         struct inode *inode = state->inode;
 827         struct file_lock *fl;
 828         int status = 0;
 829
 830         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 831                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 832                         continue;
 833                 if (nfs_file_open_context(fl->fl_file)->state != state)
 834                         continue;
 835                 status = ops->recover_lock(state, fl);
 836                 if (status >= 0)
 837                         continue;
 838                 switch (status) {
 839                         default:
 840                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 841                                                 __func__, status);
 842                         case -NFS4ERR_EXPIRED:
 843                         case -NFS4ERR_NO_GRACE:
 844                         case -NFS4ERR_RECLAIM_BAD:
 845                         case -NFS4ERR_RECLAIM_CONFLICT:
 846                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 847                                 break;
 848                         case -NFS4ERR_STALE_CLIENTID:
 849                                 goto out_err;
 850                 }
 851         }
 852         return 0;
 853 out_err:
 854         return status;
 855 }
 856
 857 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
 858 {
 859         struct nfs4_state *state;
 860         struct nfs4_lock_state *lock;
 861         int status = 0;
 862
 863         /* Note: we rely on the sp->so_states list being ordered
 864          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 865          * states first.
 866          * This is needed to ensure that the server won't give us any
 867          * read delegations that we have to return if, say, we are
 868          * recovering after a network partition or a reboot from a
 869          * server that doesn't support a grace period.
 870          */
 871         list_for_each_entry(state, &sp->so_states, open_states) {
 872                 if (state->state == 0)
 873                         continue;
 874                 status = ops->recover_open(sp, state);
 875                 if (status >= 0) {
 876                         status = nfs4_reclaim_locks(state, ops);
 877                         if (status >= 0) {
 878                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
 879                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 880                                                 printk("%s: Lock reclaim failed!\n",
 881                                                         __func__);
 882                                 }
 883                                 continue;
 884                         }
 885                 }
 886                 switch (status) {
 887                         default:
 888                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 889                                                 __func__, status);
 890                         case -ENOENT:
 891                         case -NFS4ERR_RECLAIM_BAD:
 892                         case -NFS4ERR_RECLAIM_CONFLICT:
 893                                 /*
 894                                  * Open state on this file cannot be recovered
 895                                  * All we can do is revert to using the zero stateid.
 896                                  */
 897                                 memset(state->stateid.data, 0,
 898                                         sizeof(state->stateid.data));
 899                                 /* Mark the file as being 'closed' */
 900                                 state->state = 0;
 901                                 break;
 902                         case -NFS4ERR_EXPIRED:
 903                         case -NFS4ERR_NO_GRACE:
 904                         case -NFS4ERR_STALE_CLIENTID:
 905                                 goto out_err;
 906                 }
 907         }
 908         return 0;
 909 out_err:
 910         return status;
 911 }
 912
 913 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 914 {
 915         struct nfs4_state_owner *sp;
 916         struct rb_node *pos;
 917         struct nfs4_state *state;
 918         struct nfs4_lock_state *lock;
 919
 920         /* Reset all sequence ids to zero */
 921         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 922                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 923                 sp->so_seqid.counter = 0;
 924                 sp->so_seqid.flags = 0;
 925                 spin_lock(&sp->so_lock);
 926                 list_for_each_entry(state, &sp->so_states, open_states) {
 927                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 928                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
 929                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
 930                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
 931                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 932                                 lock->ls_seqid.counter = 0;
 933                                 lock->ls_seqid.flags = 0;
 934                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 935                         }
 936                 }
 937                 spin_unlock(&sp->so_lock);
 938         }
 939 }
 940
 941 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
 942 {
 943         struct rb_node *pos;
 944         int status = 0;
 945
 946         /* Note: list is protected by exclusive lock on cl->cl_sem */
 947         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 948                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 949                 status = nfs4_reclaim_open_state(sp, ops);
 950                 if (status < 0)
 951                         break;
 952         }
 953         return status;
 954 }
 955
 956 static int nfs4_check_lease(struct nfs_client *clp)
 957 {
 958         struct rpc_cred *cred;
 959         int status = -NFS4ERR_EXPIRED;
 960
 961         /* Are there any open files on this volume? */
 962         cred = nfs4_get_renew_cred(clp);
 963         if (cred != NULL) {
 964                 /* Yes there are: try to renew the old lease */
 965                 status = nfs4_proc_renew(clp, cred);
 966                 put_rpccred(cred);
 967                 return status;
 968         }
 969
 970         /* "reboot" to ensure we clear all state on the server */
 971         clp->cl_boot_time = CURRENT_TIME;
 972         return status;
 973 }
 974
 975 static int nfs4_reclaim_lease(struct nfs_client *clp)
 976 {
 977         struct rpc_cred *cred;
 978         int status = -ENOENT;
 979
 980         cred = nfs4_get_setclientid_cred(clp);
 981         if (cred != NULL) {
 982                 status = nfs4_init_client(clp, cred);
 983                 put_rpccred(cred);
 984                 /* Handle case where the user hasn't set up machine creds */
 985                 if (status == -EACCES && cred == clp->cl_machine_cred) {
 986                         nfs4_clear_machine_cred(clp);
 987                         status = -EAGAIN;
 988                 }
 989         }
 990         return status;
 991 }
 992
 993 static int reclaimer(void *ptr)
 994 {
 995         struct nfs_client *clp = ptr;
 996         const struct nfs4_state_recovery_ops *ops;
 997         int status = 0;
 998
 999         allow_signal(SIGKILL);
1000
1001         /* Ensure exclusive access to NFSv4 state */
1002         down_write(&clp->cl_sem);
1003         while (!list_empty(&clp->cl_superblocks)) {
1004                 ops = &nfs4_network_partition_recovery_ops;
1005                 status = nfs4_check_lease(clp);
1006                 switch (status) {
1007                         case 0:
1008                         case -NFS4ERR_CB_PATH_DOWN:
1009                                 goto out;
1010                         case -NFS4ERR_STALE_CLIENTID:
1011                         case -NFS4ERR_LEASE_MOVED:
1012                                 ops = &nfs4_reboot_recovery_ops;
1013                 }
1014
1015                 /* We're going to have to re-establish a clientid */
1016                 nfs4_state_mark_reclaim(clp);
1017
1018                 status = nfs4_reclaim_lease(clp);
1019                 if (status) {
1020                         if (status == -EAGAIN)
1021                                 continue;
1022                         goto out_error;
1023                 }
1024
1025                 /* Mark all delegations for reclaim */
1026                 nfs_delegation_mark_reclaim(clp);
1027                 /* Note: list is protected by exclusive lock on cl->cl_sem */
1028                 status = nfs4_do_reclaim(clp, ops);
1029                 if (status < 0) {
1030                         if (status == -NFS4ERR_NO_GRACE) {
1031                                 ops = &nfs4_network_partition_recovery_ops;
1032                                 status = nfs4_do_reclaim(clp, ops);
1033                         }
1034                         if (status == -NFS4ERR_STALE_CLIENTID)
1035                                 continue;
1036                         if (status == -NFS4ERR_EXPIRED)
1037                                 continue;
1038                 }
1039                 nfs_delegation_reap_unclaimed(clp);
1040                 break;
1041         }
1042 out:
1043         up_write(&clp->cl_sem);
1044         if (status == -NFS4ERR_CB_PATH_DOWN)
1045                 nfs_handle_cb_pathdown(clp);
1046         nfs4_clear_recover_bit(clp);
1047         nfs_put_client(clp);
1048         module_put_and_exit(0);
1049         return 0;
1050 out_error:
1051         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
1052                         " with error %d\n", clp->cl_hostname, -status);
1053         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1054         goto out;
1055 }
1056
1057 /*
1058  * Local variables:
1059  *  c-basic-offset: 8
1060  * End:
1061  */