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