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         if (mode == 0)
 345                 list_del_init(&state->inode_states);
 346         state->state = mode;
 347 }
 348
 349 static struct nfs4_state *
 350 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 351 {
 352         struct nfs_inode *nfsi = NFS_I(inode);
 353         struct nfs4_state *state;
 354
 355         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 356                 if (state->owner != owner)
 357                         continue;
 358                 if (atomic_inc_not_zero(&state->count))
 359                         return state;
 360         }
 361         return NULL;
 362 }
 363
 364 static void
 365 nfs4_free_open_state(struct nfs4_state *state)
 366 {
 367         kfree(state);
 368 }
 369
 370 struct nfs4_state *
 371 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 372 {
 373         struct nfs4_state *state, *new;
 374         struct nfs_inode *nfsi = NFS_I(inode);
 375
 376         spin_lock(&inode->i_lock);
 377         state = __nfs4_find_state_byowner(inode, owner);
 378         spin_unlock(&inode->i_lock);
 379         if (state)
 380                 goto out;
 381         new = nfs4_alloc_open_state();
 382         spin_lock(&owner->so_lock);
 383         spin_lock(&inode->i_lock);
 384         state = __nfs4_find_state_byowner(inode, owner);
 385         if (state == NULL && new != NULL) {
 386                 state = new;
 387                 state->owner = owner;
 388                 atomic_inc(&owner->so_count);
 389                 list_add(&state->inode_states, &nfsi->open_states);
 390                 state->inode = igrab(inode);
 391                 spin_unlock(&inode->i_lock);
 392                 /* Note: The reclaim code dictates that we add stateless
 393                  * and read-only stateids to the end of the list */
 394                 list_add_tail(&state->open_states, &owner->so_states);
 395                 spin_unlock(&owner->so_lock);
 396         } else {
 397                 spin_unlock(&inode->i_lock);
 398                 spin_unlock(&owner->so_lock);
 399                 if (new)
 400                         nfs4_free_open_state(new);
 401         }
 402 out:
 403         return state;
 404 }
 405
 406 /*
 407  * Beware! Caller must be holding exactly one
 408  * reference to clp->cl_sem!
 409  */
 410 void nfs4_put_open_state(struct nfs4_state *state)
 411 {
 412         struct inode *inode = state->inode;
 413         struct nfs4_state_owner *owner = state->owner;
 414
 415         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 416                 return;
 417         spin_lock(&inode->i_lock);
 418         if (!list_empty(&state->inode_states))
 419                 list_del(&state->inode_states);
 420         list_del(&state->open_states);
 421         spin_unlock(&inode->i_lock);
 422         spin_unlock(&owner->so_lock);
 423         iput(inode);
 424         nfs4_free_open_state(state);
 425         nfs4_put_state_owner(owner);
 426 }
 427
 428 /*
 429  * Close the current file.
 430  */
 431 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
 432 {
 433         struct nfs4_state_owner *owner = state->owner;
 434         int call_close = 0;
 435         int newstate;
 436
 437         atomic_inc(&owner->so_count);
 438         /* Protect against nfs4_find_state() */
 439         spin_lock(&owner->so_lock);
 440         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 441                 case FMODE_READ:
 442                         state->n_rdonly--;
 443                         break;
 444                 case FMODE_WRITE:
 445                         state->n_wronly--;
 446                         break;
 447                 case FMODE_READ|FMODE_WRITE:
 448                         state->n_rdwr--;
 449         }
 450         newstate = FMODE_READ|FMODE_WRITE;
 451         if (state->n_rdwr == 0) {
 452                 if (state->n_rdonly == 0) {
 453                         newstate &= ~FMODE_READ;
 454                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 455                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 456                 }
 457                 if (state->n_wronly == 0) {
 458                         newstate &= ~FMODE_WRITE;
 459                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 460                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 461                 }
 462                 if (newstate == 0)
 463                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 464         }
 465         nfs4_state_set_mode_locked(state, newstate);
 466         spin_unlock(&owner->so_lock);
 467
 468         if (!call_close) {
 469                 nfs4_put_open_state(state);
 470                 nfs4_put_state_owner(owner);
 471         } else
 472                 nfs4_do_close(path, state);
 473 }
 474
 475 /*
 476  * Search the state->lock_states for an existing lock_owner
 477  * that is compatible with current->files
 478  */
 479 static struct nfs4_lock_state *
 480 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 481 {
 482         struct nfs4_lock_state *pos;
 483         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 484                 if (pos->ls_owner != fl_owner)
 485                         continue;
 486                 atomic_inc(&pos->ls_count);
 487                 return pos;
 488         }
 489         return NULL;
 490 }
 491
 492 /*
 493  * Return a compatible lock_state. If no initialized lock_state structure
 494  * exists, return an uninitialized one.
 495  *
 496  */
 497 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 498 {
 499         struct nfs4_lock_state *lsp;
 500         struct nfs_client *clp = state->owner->so_client;
 501
 502         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 503         if (lsp == NULL)
 504                 return NULL;
 505         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 506         atomic_set(&lsp->ls_count, 1);
 507         lsp->ls_owner = fl_owner;
 508         spin_lock(&clp->cl_lock);
 509         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 510         spin_unlock(&clp->cl_lock);
 511         INIT_LIST_HEAD(&lsp->ls_locks);
 512         return lsp;
 513 }
 514
 515 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 516 {
 517         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 518
 519         spin_lock(&clp->cl_lock);
 520         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 521         spin_unlock(&clp->cl_lock);
 522         kfree(lsp);
 523 }
 524
 525 /*
 526  * Return a compatible lock_state. If no initialized lock_state structure
 527  * exists, return an uninitialized one.
 528  *
 529  * The caller must be holding clp->cl_sem
 530  */
 531 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 532 {
 533         struct nfs4_lock_state *lsp, *new = NULL;
 534
 535         for(;;) {
 536                 spin_lock(&state->state_lock);
 537                 lsp = __nfs4_find_lock_state(state, owner);
 538                 if (lsp != NULL)
 539                         break;
 540                 if (new != NULL) {
 541                         new->ls_state = state;
 542                         list_add(&new->ls_locks, &state->lock_states);
 543                         set_bit(LK_STATE_IN_USE, &state->flags);
 544                         lsp = new;
 545                         new = NULL;
 546                         break;
 547                 }
 548                 spin_unlock(&state->state_lock);
 549                 new = nfs4_alloc_lock_state(state, owner);
 550                 if (new == NULL)
 551                         return NULL;
 552         }
 553         spin_unlock(&state->state_lock);
 554         if (new != NULL)
 555                 nfs4_free_lock_state(new);
 556         return lsp;
 557 }
 558
 559 /*
 560  * Release reference to lock_state, and free it if we see that
 561  * it is no longer in use
 562  */
 563 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 564 {
 565         struct nfs4_state *state;
 566
 567         if (lsp == NULL)
 568                 return;
 569         state = lsp->ls_state;
 570         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 571                 return;
 572         list_del(&lsp->ls_locks);
 573         if (list_empty(&state->lock_states))
 574                 clear_bit(LK_STATE_IN_USE, &state->flags);
 575         spin_unlock(&state->state_lock);
 576         nfs4_free_lock_state(lsp);
 577 }
 578
 579 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 580 {
 581         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 582
 583         dst->fl_u.nfs4_fl.owner = lsp;
 584         atomic_inc(&lsp->ls_count);
 585 }
 586
 587 static void nfs4_fl_release_lock(struct file_lock *fl)
 588 {
 589         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 590 }
 591
 592 static struct file_lock_operations nfs4_fl_lock_ops = {
 593         .fl_copy_lock = nfs4_fl_copy_lock,
 594         .fl_release_private = nfs4_fl_release_lock,
 595 };
 596
 597 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 598 {
 599         struct nfs4_lock_state *lsp;
 600
 601         if (fl->fl_ops != NULL)
 602                 return 0;
 603         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 604         if (lsp == NULL)
 605                 return -ENOMEM;
 606         fl->fl_u.nfs4_fl.owner = lsp;
 607         fl->fl_ops = &nfs4_fl_lock_ops;
 608         return 0;
 609 }
 610
 611 /*
 612  * Byte-range lock aware utility to initialize the stateid of read/write
 613  * requests.
 614  */
 615 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 616 {
 617         struct nfs4_lock_state *lsp;
 618         int seq;
 619
 620         do {
 621                 seq = read_seqbegin(&state->seqlock);
 622                 memcpy(dst, &state->stateid, sizeof(*dst));
 623         } while (read_seqretry(&state->seqlock, seq));
 624         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 625                 return;
 626
 627         spin_lock(&state->state_lock);
 628         lsp = __nfs4_find_lock_state(state, fl_owner);
 629         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 630                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 631         spin_unlock(&state->state_lock);
 632         nfs4_put_lock_state(lsp);
 633 }
 634
 635 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 636 {
 637         struct rpc_sequence *sequence = counter->sequence;
 638         struct nfs_seqid *new;
 639
 640         new = kmalloc(sizeof(*new), GFP_KERNEL);
 641         if (new != NULL) {
 642                 new->sequence = counter;
 643                 spin_lock(&sequence->lock);
 644                 list_add_tail(&new->list, &sequence->list);
 645                 spin_unlock(&sequence->lock);
 646         }
 647         return new;
 648 }
 649
 650 void nfs_free_seqid(struct nfs_seqid *seqid)
 651 {
 652         struct rpc_sequence *sequence = seqid->sequence->sequence;
 653
 654         spin_lock(&sequence->lock);
 655         list_del(&seqid->list);
 656         spin_unlock(&sequence->lock);
 657         rpc_wake_up(&sequence->wait);
 658         kfree(seqid);
 659 }
 660
 661 /*
 662  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 663  * failed with a seqid incrementing error -
 664  * see comments nfs_fs.h:seqid_mutating_error()
 665  */
 666 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 667 {
 668         switch (status) {
 669                 case 0:
 670                         break;
 671                 case -NFS4ERR_BAD_SEQID:
 672                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 673                                 return;
 674                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 675                                         "sequence-id error on an"
 676                                         "unconfirmed sequence %p!\n",
 677                                         seqid->sequence);
 678                 case -NFS4ERR_STALE_CLIENTID:
 679                 case -NFS4ERR_STALE_STATEID:
 680                 case -NFS4ERR_BAD_STATEID:
 681                 case -NFS4ERR_BADXDR:
 682                 case -NFS4ERR_RESOURCE:
 683                 case -NFS4ERR_NOFILEHANDLE:
 684                         /* Non-seqid mutating errors */
 685                         return;
 686         };
 687         /*
 688          * Note: no locking needed as we are guaranteed to be first
 689          * on the sequence list
 690          */
 691         seqid->sequence->counter++;
 692 }
 693
 694 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 695 {
 696         if (status == -NFS4ERR_BAD_SEQID) {
 697                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 698                                 struct nfs4_state_owner, so_seqid);
 699                 nfs4_drop_state_owner(sp);
 700         }
 701         nfs_increment_seqid(status, seqid);
 702 }
 703
 704 /*
 705  * Increment the seqid if the LOCK/LOCKU succeeded, or
 706  * failed with a seqid incrementing error -
 707  * see comments nfs_fs.h:seqid_mutating_error()
 708  */
 709 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 710 {
 711         nfs_increment_seqid(status, seqid);
 712 }
 713
 714 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 715 {
 716         struct rpc_sequence *sequence = seqid->sequence->sequence;
 717         int status = 0;
 718
 719         if (sequence->list.next == &seqid->list)
 720                 goto out;
 721         spin_lock(&sequence->lock);
 722         if (sequence->list.next != &seqid->list) {
 723                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 724                 status = -EAGAIN;
 725         }
 726         spin_unlock(&sequence->lock);
 727 out:
 728         return status;
 729 }
 730
 731 static int reclaimer(void *);
 732
 733 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 734 {
 735         smp_mb__before_clear_bit();
 736         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 737         smp_mb__after_clear_bit();
 738         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 739         rpc_wake_up(&clp->cl_rpcwaitq);
 740 }
 741
 742 /*
 743  * State recovery routine
 744  */
 745 static void nfs4_recover_state(struct nfs_client *clp)
 746 {
 747         struct task_struct *task;
 748
 749         __module_get(THIS_MODULE);
 750         atomic_inc(&clp->cl_count);
 751         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 752                         NIPQUAD(clp->cl_addr.sin_addr));
 753         if (!IS_ERR(task))
 754                 return;
 755         nfs4_clear_recover_bit(clp);
 756         nfs_put_client(clp);
 757         module_put(THIS_MODULE);
 758 }
 759
 760 /*
 761  * Schedule a state recovery attempt
 762  */
 763 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 764 {
 765         if (!clp)
 766                 return;
 767         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 768                 nfs4_recover_state(clp);
 769 }
 770
 771 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 772 {
 773         struct inode *inode = state->inode;
 774         struct file_lock *fl;
 775         int status = 0;
 776
 777         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 778                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 779                         continue;
 780                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 781                         continue;
 782                 status = ops->recover_lock(state, fl);
 783                 if (status >= 0)
 784                         continue;
 785                 switch (status) {
 786                         default:
 787                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 788                                                 __FUNCTION__, status);
 789                         case -NFS4ERR_EXPIRED:
 790                         case -NFS4ERR_NO_GRACE:
 791                         case -NFS4ERR_RECLAIM_BAD:
 792                         case -NFS4ERR_RECLAIM_CONFLICT:
 793                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 794                                 break;
 795                         case -NFS4ERR_STALE_CLIENTID:
 796                                 goto out_err;
 797                 }
 798         }
 799         return 0;
 800 out_err:
 801         return status;
 802 }
 803
 804 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 805 {
 806         struct nfs4_state *state;
 807         struct nfs4_lock_state *lock;
 808         int status = 0;
 809
 810         /* Note: we rely on the sp->so_states list being ordered
 811          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 812          * states first.
 813          * This is needed to ensure that the server won't give us any
 814          * read delegations that we have to return if, say, we are
 815          * recovering after a network partition or a reboot from a
 816          * server that doesn't support a grace period.
 817          */
 818         list_for_each_entry(state, &sp->so_states, open_states) {
 819                 if (state->state == 0)
 820                         continue;
 821                 status = ops->recover_open(sp, state);
 822                 if (status >= 0) {
 823                         status = nfs4_reclaim_locks(ops, state);
 824                         if (status < 0)
 825                                 goto out_err;
 826                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 827                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 828                                         printk("%s: Lock reclaim failed!\n",
 829                                                         __FUNCTION__);
 830                         }
 831                         continue;
 832                 }
 833                 switch (status) {
 834                         default:
 835                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 836                                                 __FUNCTION__, status);
 837                         case -ENOENT:
 838                         case -NFS4ERR_RECLAIM_BAD:
 839                         case -NFS4ERR_RECLAIM_CONFLICT:
 840                                 /*
 841                                  * Open state on this file cannot be recovered
 842                                  * All we can do is revert to using the zero stateid.
 843                                  */
 844                                 memset(state->stateid.data, 0,
 845                                         sizeof(state->stateid.data));
 846                                 /* Mark the file as being 'closed' */
 847                                 state->state = 0;
 848                                 break;
 849                         case -NFS4ERR_EXPIRED:
 850                         case -NFS4ERR_NO_GRACE:
 851                         case -NFS4ERR_STALE_CLIENTID:
 852                                 goto out_err;
 853                 }
 854         }
 855         return 0;
 856 out_err:
 857         return status;
 858 }
 859
 860 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 861 {
 862         struct nfs4_state_owner *sp;
 863         struct rb_node *pos;
 864         struct nfs4_state *state;
 865         struct nfs4_lock_state *lock;
 866
 867         /* Reset all sequence ids to zero */
 868         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 869                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 870                 sp->so_seqid.counter = 0;
 871                 sp->so_seqid.flags = 0;
 872                 spin_lock(&sp->so_lock);
 873                 list_for_each_entry(state, &sp->so_states, open_states) {
 874                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 875                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
 876                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
 877                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
 878                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 879                                 lock->ls_seqid.counter = 0;
 880                                 lock->ls_seqid.flags = 0;
 881                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 882                         }
 883                 }
 884                 spin_unlock(&sp->so_lock);
 885         }
 886 }
 887
 888 static int reclaimer(void *ptr)
 889 {
 890         struct nfs_client *clp = ptr;
 891         struct nfs4_state_owner *sp;
 892         struct rb_node *pos;
 893         struct nfs4_state_recovery_ops *ops;
 894         struct rpc_cred *cred;
 895         int status = 0;
 896
 897         allow_signal(SIGKILL);
 898
 899         /* Ensure exclusive access to NFSv4 state */
 900         lock_kernel();
 901         down_write(&clp->cl_sem);
 902         /* Are there any NFS mounts out there? */
 903         if (list_empty(&clp->cl_superblocks))
 904                 goto out;
 905 restart_loop:
 906         ops = &nfs4_network_partition_recovery_ops;
 907         /* Are there any open files on this volume? */
 908         cred = nfs4_get_renew_cred(clp);
 909         if (cred != NULL) {
 910                 /* Yes there are: try to renew the old lease */
 911                 status = nfs4_proc_renew(clp, cred);
 912                 switch (status) {
 913                         case 0:
 914                         case -NFS4ERR_CB_PATH_DOWN:
 915                                 put_rpccred(cred);
 916                                 goto out;
 917                         case -NFS4ERR_STALE_CLIENTID:
 918                         case -NFS4ERR_LEASE_MOVED:
 919                                 ops = &nfs4_reboot_recovery_ops;
 920                 }
 921         } else {
 922                 /* "reboot" to ensure we clear all state on the server */
 923                 clp->cl_boot_time = CURRENT_TIME;
 924                 cred = nfs4_get_setclientid_cred(clp);
 925         }
 926         /* We're going to have to re-establish a clientid */
 927         nfs4_state_mark_reclaim(clp);
 928         status = -ENOENT;
 929         if (cred != NULL) {
 930                 status = nfs4_init_client(clp, cred);
 931                 put_rpccred(cred);
 932         }
 933         if (status)
 934                 goto out_error;
 935         /* Mark all delegations for reclaim */
 936         nfs_delegation_mark_reclaim(clp);
 937         /* Note: list is protected by exclusive lock on cl->cl_sem */
 938         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 939                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 940                 status = nfs4_reclaim_open_state(ops, sp);
 941                 if (status < 0) {
 942                         if (status == -NFS4ERR_NO_GRACE) {
 943                                 ops = &nfs4_network_partition_recovery_ops;
 944                                 status = nfs4_reclaim_open_state(ops, sp);
 945                         }
 946                         if (status == -NFS4ERR_STALE_CLIENTID)
 947                                 goto restart_loop;
 948                         if (status == -NFS4ERR_EXPIRED)
 949                                 goto restart_loop;
 950                 }
 951         }
 952         nfs_delegation_reap_unclaimed(clp);
 953 out:
 954         up_write(&clp->cl_sem);
 955         unlock_kernel();
 956         if (status == -NFS4ERR_CB_PATH_DOWN)
 957                 nfs_handle_cb_pathdown(clp);
 958         nfs4_clear_recover_bit(clp);
 959         nfs_put_client(clp);
 960         module_put_and_exit(0);
 961         return 0;
 962 out_error:
 963         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 964                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
 965         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 966         goto out;
 967 }
 968
 969 /*
 970  * Local variables:
 971  *  c-basic-offset: 8
 972  * End:
 973  */