SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/smp_lock.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51
  52 #include "nfs4_fs.h"
  53 #include "callback.h"
  54 #include "delegation.h"
  55 #include "internal.h"
  56
  57 #define OPENOWNER_POOL_SIZE     8
  58
  59 const nfs4_stateid zero_stateid;
  60
  61 static LIST_HEAD(nfs4_clientid_list);
  62
  63 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         seqlock_init(&state->seqlock);
 311         return state;
 312 }
 313
 314 void
 315 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 316 {
 317         if (state->state == mode)
 318                 return;
 319         /* NB! List reordering - see the reclaim code for why.  */
 320         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 321                 if (mode & FMODE_WRITE)
 322                         list_move(&state->open_states, &state->owner->so_states);
 323                 else
 324                         list_move_tail(&state->open_states, &state->owner->so_states);
 325         }
 326         if (mode == 0)
 327                 list_del_init(&state->inode_states);
 328         state->state = mode;
 329 }
 330
 331 static struct nfs4_state *
 332 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 333 {
 334         struct nfs_inode *nfsi = NFS_I(inode);
 335         struct nfs4_state *state;
 336
 337         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 338                 if (state->owner != owner)
 339                         continue;
 340                 if (atomic_inc_not_zero(&state->count))
 341                         return state;
 342         }
 343         return NULL;
 344 }
 345
 346 static void
 347 nfs4_free_open_state(struct nfs4_state *state)
 348 {
 349         kfree(state);
 350 }
 351
 352 struct nfs4_state *
 353 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 354 {
 355         struct nfs4_state *state, *new;
 356         struct nfs_inode *nfsi = NFS_I(inode);
 357
 358         spin_lock(&inode->i_lock);
 359         state = __nfs4_find_state_byowner(inode, owner);
 360         spin_unlock(&inode->i_lock);
 361         if (state)
 362                 goto out;
 363         new = nfs4_alloc_open_state();
 364         spin_lock(&owner->so_lock);
 365         spin_lock(&inode->i_lock);
 366         state = __nfs4_find_state_byowner(inode, owner);
 367         if (state == NULL && new != NULL) {
 368                 state = new;
 369                 state->owner = owner;
 370                 atomic_inc(&owner->so_count);
 371                 list_add(&state->inode_states, &nfsi->open_states);
 372                 state->inode = igrab(inode);
 373                 spin_unlock(&inode->i_lock);
 374                 /* Note: The reclaim code dictates that we add stateless
 375                  * and read-only stateids to the end of the list */
 376                 list_add_tail(&state->open_states, &owner->so_states);
 377                 spin_unlock(&owner->so_lock);
 378         } else {
 379                 spin_unlock(&inode->i_lock);
 380                 spin_unlock(&owner->so_lock);
 381                 if (new)
 382                         nfs4_free_open_state(new);
 383         }
 384 out:
 385         return state;
 386 }
 387
 388 /*
 389  * Beware! Caller must be holding exactly one
 390  * reference to clp->cl_sem!
 391  */
 392 void nfs4_put_open_state(struct nfs4_state *state)
 393 {
 394         struct inode *inode = state->inode;
 395         struct nfs4_state_owner *owner = state->owner;
 396
 397         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 398                 return;
 399         spin_lock(&inode->i_lock);
 400         if (!list_empty(&state->inode_states))
 401                 list_del(&state->inode_states);
 402         list_del(&state->open_states);
 403         spin_unlock(&inode->i_lock);
 404         spin_unlock(&owner->so_lock);
 405         iput(inode);
 406         nfs4_free_open_state(state);
 407         nfs4_put_state_owner(owner);
 408 }
 409
 410 /*
 411  * Close the current file.
 412  */
 413 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
 414 {
 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         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 423                 case FMODE_READ:
 424                         state->n_rdonly--;
 425                         break;
 426                 case FMODE_WRITE:
 427                         state->n_wronly--;
 428                         break;
 429                 case FMODE_READ|FMODE_WRITE:
 430                         state->n_rdwr--;
 431         }
 432         newstate = FMODE_READ|FMODE_WRITE;
 433         if (state->n_rdwr == 0) {
 434                 if (state->n_rdonly == 0) {
 435                         newstate &= ~FMODE_READ;
 436                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 437                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 438                 }
 439                 if (state->n_wronly == 0) {
 440                         newstate &= ~FMODE_WRITE;
 441                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 442                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 443                 }
 444                 if (newstate == 0)
 445                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 446         }
 447         nfs4_state_set_mode_locked(state, newstate);
 448         spin_unlock(&owner->so_lock);
 449
 450         if (!call_close) {
 451                 nfs4_put_open_state(state);
 452                 nfs4_put_state_owner(owner);
 453         } else
 454                 nfs4_do_close(path, state);
 455 }
 456
 457 /*
 458  * Search the state->lock_states for an existing lock_owner
 459  * that is compatible with current->files
 460  */
 461 static struct nfs4_lock_state *
 462 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 463 {
 464         struct nfs4_lock_state *pos;
 465         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 466                 if (pos->ls_owner != fl_owner)
 467                         continue;
 468                 atomic_inc(&pos->ls_count);
 469                 return pos;
 470         }
 471         return NULL;
 472 }
 473
 474 /*
 475  * Return a compatible lock_state. If no initialized lock_state structure
 476  * exists, return an uninitialized one.
 477  *
 478  */
 479 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 480 {
 481         struct nfs4_lock_state *lsp;
 482         struct nfs_client *clp = state->owner->so_client;
 483
 484         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 485         if (lsp == NULL)
 486                 return NULL;
 487         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 488         atomic_set(&lsp->ls_count, 1);
 489         lsp->ls_owner = fl_owner;
 490         spin_lock(&clp->cl_lock);
 491         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 492         spin_unlock(&clp->cl_lock);
 493         INIT_LIST_HEAD(&lsp->ls_locks);
 494         return lsp;
 495 }
 496
 497 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 498 {
 499         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 500
 501         spin_lock(&clp->cl_lock);
 502         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 503         spin_unlock(&clp->cl_lock);
 504         kfree(lsp);
 505 }
 506
 507 /*
 508  * Return a compatible lock_state. If no initialized lock_state structure
 509  * exists, return an uninitialized one.
 510  *
 511  * The caller must be holding clp->cl_sem
 512  */
 513 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 514 {
 515         struct nfs4_lock_state *lsp, *new = NULL;
 516
 517         for(;;) {
 518                 spin_lock(&state->state_lock);
 519                 lsp = __nfs4_find_lock_state(state, owner);
 520                 if (lsp != NULL)
 521                         break;
 522                 if (new != NULL) {
 523                         new->ls_state = state;
 524                         list_add(&new->ls_locks, &state->lock_states);
 525                         set_bit(LK_STATE_IN_USE, &state->flags);
 526                         lsp = new;
 527                         new = NULL;
 528                         break;
 529                 }
 530                 spin_unlock(&state->state_lock);
 531                 new = nfs4_alloc_lock_state(state, owner);
 532                 if (new == NULL)
 533                         return NULL;
 534         }
 535         spin_unlock(&state->state_lock);
 536         if (new != NULL)
 537                 nfs4_free_lock_state(new);
 538         return lsp;
 539 }
 540
 541 /*
 542  * Release reference to lock_state, and free it if we see that
 543  * it is no longer in use
 544  */
 545 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 546 {
 547         struct nfs4_state *state;
 548
 549         if (lsp == NULL)
 550                 return;
 551         state = lsp->ls_state;
 552         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 553                 return;
 554         list_del(&lsp->ls_locks);
 555         if (list_empty(&state->lock_states))
 556                 clear_bit(LK_STATE_IN_USE, &state->flags);
 557         spin_unlock(&state->state_lock);
 558         nfs4_free_lock_state(lsp);
 559 }
 560
 561 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 562 {
 563         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 564
 565         dst->fl_u.nfs4_fl.owner = lsp;
 566         atomic_inc(&lsp->ls_count);
 567 }
 568
 569 static void nfs4_fl_release_lock(struct file_lock *fl)
 570 {
 571         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 572 }
 573
 574 static struct file_lock_operations nfs4_fl_lock_ops = {
 575         .fl_copy_lock = nfs4_fl_copy_lock,
 576         .fl_release_private = nfs4_fl_release_lock,
 577 };
 578
 579 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 580 {
 581         struct nfs4_lock_state *lsp;
 582
 583         if (fl->fl_ops != NULL)
 584                 return 0;
 585         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 586         if (lsp == NULL)
 587                 return -ENOMEM;
 588         fl->fl_u.nfs4_fl.owner = lsp;
 589         fl->fl_ops = &nfs4_fl_lock_ops;
 590         return 0;
 591 }
 592
 593 /*
 594  * Byte-range lock aware utility to initialize the stateid of read/write
 595  * requests.
 596  */
 597 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 598 {
 599         struct nfs4_lock_state *lsp;
 600         int seq;
 601
 602         do {
 603                 seq = read_seqbegin(&state->seqlock);
 604                 memcpy(dst, &state->stateid, sizeof(*dst));
 605         } while (read_seqretry(&state->seqlock, seq));
 606         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 607                 return;
 608
 609         spin_lock(&state->state_lock);
 610         lsp = __nfs4_find_lock_state(state, fl_owner);
 611         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 612                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 613         spin_unlock(&state->state_lock);
 614         nfs4_put_lock_state(lsp);
 615 }
 616
 617 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 618 {
 619         struct rpc_sequence *sequence = counter->sequence;
 620         struct nfs_seqid *new;
 621
 622         new = kmalloc(sizeof(*new), GFP_KERNEL);
 623         if (new != NULL) {
 624                 new->sequence = counter;
 625                 spin_lock(&sequence->lock);
 626                 list_add_tail(&new->list, &sequence->list);
 627                 spin_unlock(&sequence->lock);
 628         }
 629         return new;
 630 }
 631
 632 void nfs_free_seqid(struct nfs_seqid *seqid)
 633 {
 634         struct rpc_sequence *sequence = seqid->sequence->sequence;
 635
 636         spin_lock(&sequence->lock);
 637         list_del(&seqid->list);
 638         spin_unlock(&sequence->lock);
 639         rpc_wake_up(&sequence->wait);
 640         kfree(seqid);
 641 }
 642
 643 /*
 644  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 645  * failed with a seqid incrementing error -
 646  * see comments nfs_fs.h:seqid_mutating_error()
 647  */
 648 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 649 {
 650         switch (status) {
 651                 case 0:
 652                         break;
 653                 case -NFS4ERR_BAD_SEQID:
 654                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 655                                 return;
 656                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 657                                         "sequence-id error on an"
 658                                         "unconfirmed sequence %p!\n",
 659                                         seqid->sequence);
 660                 case -NFS4ERR_STALE_CLIENTID:
 661                 case -NFS4ERR_STALE_STATEID:
 662                 case -NFS4ERR_BAD_STATEID:
 663                 case -NFS4ERR_BADXDR:
 664                 case -NFS4ERR_RESOURCE:
 665                 case -NFS4ERR_NOFILEHANDLE:
 666                         /* Non-seqid mutating errors */
 667                         return;
 668         };
 669         /*
 670          * Note: no locking needed as we are guaranteed to be first
 671          * on the sequence list
 672          */
 673         seqid->sequence->counter++;
 674 }
 675
 676 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 677 {
 678         if (status == -NFS4ERR_BAD_SEQID) {
 679                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 680                                 struct nfs4_state_owner, so_seqid);
 681                 nfs4_drop_state_owner(sp);
 682         }
 683         nfs_increment_seqid(status, seqid);
 684 }
 685
 686 /*
 687  * Increment the seqid if the LOCK/LOCKU succeeded, or
 688  * failed with a seqid incrementing error -
 689  * see comments nfs_fs.h:seqid_mutating_error()
 690  */
 691 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 692 {
 693         nfs_increment_seqid(status, seqid);
 694 }
 695
 696 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 697 {
 698         struct rpc_sequence *sequence = seqid->sequence->sequence;
 699         int status = 0;
 700
 701         if (sequence->list.next == &seqid->list)
 702                 goto out;
 703         spin_lock(&sequence->lock);
 704         if (sequence->list.next != &seqid->list) {
 705                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 706                 status = -EAGAIN;
 707         }
 708         spin_unlock(&sequence->lock);
 709 out:
 710         return status;
 711 }
 712
 713 static int reclaimer(void *);
 714
 715 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 716 {
 717         smp_mb__before_clear_bit();
 718         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 719         smp_mb__after_clear_bit();
 720         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 721         rpc_wake_up(&clp->cl_rpcwaitq);
 722 }
 723
 724 /*
 725  * State recovery routine
 726  */
 727 static void nfs4_recover_state(struct nfs_client *clp)
 728 {
 729         struct task_struct *task;
 730
 731         __module_get(THIS_MODULE);
 732         atomic_inc(&clp->cl_count);
 733         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 734                         NIPQUAD(clp->cl_addr.sin_addr));
 735         if (!IS_ERR(task))
 736                 return;
 737         nfs4_clear_recover_bit(clp);
 738         nfs_put_client(clp);
 739         module_put(THIS_MODULE);
 740 }
 741
 742 /*
 743  * Schedule a state recovery attempt
 744  */
 745 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 746 {
 747         if (!clp)
 748                 return;
 749         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 750                 nfs4_recover_state(clp);
 751 }
 752
 753 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 754 {
 755         struct inode *inode = state->inode;
 756         struct file_lock *fl;
 757         int status = 0;
 758
 759         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 760                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 761                         continue;
 762                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 763                         continue;
 764                 status = ops->recover_lock(state, fl);
 765                 if (status >= 0)
 766                         continue;
 767                 switch (status) {
 768                         default:
 769                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 770                                                 __FUNCTION__, status);
 771                         case -NFS4ERR_EXPIRED:
 772                         case -NFS4ERR_NO_GRACE:
 773                         case -NFS4ERR_RECLAIM_BAD:
 774                         case -NFS4ERR_RECLAIM_CONFLICT:
 775                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 776                                 break;
 777                         case -NFS4ERR_STALE_CLIENTID:
 778                                 goto out_err;
 779                 }
 780         }
 781         return 0;
 782 out_err:
 783         return status;
 784 }
 785
 786 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 787 {
 788         struct nfs4_state *state;
 789         struct nfs4_lock_state *lock;
 790         int status = 0;
 791
 792         /* Note: we rely on the sp->so_states list being ordered
 793          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 794          * states first.
 795          * This is needed to ensure that the server won't give us any
 796          * read delegations that we have to return if, say, we are
 797          * recovering after a network partition or a reboot from a
 798          * server that doesn't support a grace period.
 799          */
 800         list_for_each_entry(state, &sp->so_states, open_states) {
 801                 if (state->state == 0)
 802                         continue;
 803                 status = ops->recover_open(sp, state);
 804                 if (status >= 0) {
 805                         status = nfs4_reclaim_locks(ops, state);
 806                         if (status < 0)
 807                                 goto out_err;
 808                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 809                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 810                                         printk("%s: Lock reclaim failed!\n",
 811                                                         __FUNCTION__);
 812                         }
 813                         continue;
 814                 }
 815                 switch (status) {
 816                         default:
 817                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 818                                                 __FUNCTION__, status);
 819                         case -ENOENT:
 820                         case -NFS4ERR_RECLAIM_BAD:
 821                         case -NFS4ERR_RECLAIM_CONFLICT:
 822                                 /*
 823                                  * Open state on this file cannot be recovered
 824                                  * All we can do is revert to using the zero stateid.
 825                                  */
 826                                 memset(state->stateid.data, 0,
 827                                         sizeof(state->stateid.data));
 828                                 /* Mark the file as being 'closed' */
 829                                 state->state = 0;
 830                                 break;
 831                         case -NFS4ERR_EXPIRED:
 832                         case -NFS4ERR_NO_GRACE:
 833                         case -NFS4ERR_STALE_CLIENTID:
 834                                 goto out_err;
 835                 }
 836         }
 837         return 0;
 838 out_err:
 839         return status;
 840 }
 841
 842 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 843 {
 844         struct nfs4_state_owner *sp;
 845         struct rb_node *pos;
 846         struct nfs4_state *state;
 847         struct nfs4_lock_state *lock;
 848
 849         /* Reset all sequence ids to zero */
 850         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 851                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 852                 sp->so_seqid.counter = 0;
 853                 sp->so_seqid.flags = 0;
 854                 spin_lock(&sp->so_lock);
 855                 list_for_each_entry(state, &sp->so_states, open_states) {
 856                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 857                         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
 858                         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
 859                         clear_bit(NFS_O_RDWR_STATE, &state->flags);
 860                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 861                                 lock->ls_seqid.counter = 0;
 862                                 lock->ls_seqid.flags = 0;
 863                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 864                         }
 865                 }
 866                 spin_unlock(&sp->so_lock);
 867         }
 868 }
 869
 870 static int reclaimer(void *ptr)
 871 {
 872         struct nfs_client *clp = ptr;
 873         struct nfs4_state_owner *sp;
 874         struct rb_node *pos;
 875         struct nfs4_state_recovery_ops *ops;
 876         struct rpc_cred *cred;
 877         int status = 0;
 878
 879         allow_signal(SIGKILL);
 880
 881         /* Ensure exclusive access to NFSv4 state */
 882         lock_kernel();
 883         down_write(&clp->cl_sem);
 884         /* Are there any NFS mounts out there? */
 885         if (list_empty(&clp->cl_superblocks))
 886                 goto out;
 887 restart_loop:
 888         ops = &nfs4_network_partition_recovery_ops;
 889         /* Are there any open files on this volume? */
 890         cred = nfs4_get_renew_cred(clp);
 891         if (cred != NULL) {
 892                 /* Yes there are: try to renew the old lease */
 893                 status = nfs4_proc_renew(clp, cred);
 894                 switch (status) {
 895                         case 0:
 896                         case -NFS4ERR_CB_PATH_DOWN:
 897                                 put_rpccred(cred);
 898                                 goto out;
 899                         case -NFS4ERR_STALE_CLIENTID:
 900                         case -NFS4ERR_LEASE_MOVED:
 901                                 ops = &nfs4_reboot_recovery_ops;
 902                 }
 903         } else {
 904                 /* "reboot" to ensure we clear all state on the server */
 905                 clp->cl_boot_time = CURRENT_TIME;
 906                 cred = nfs4_get_setclientid_cred(clp);
 907         }
 908         /* We're going to have to re-establish a clientid */
 909         nfs4_state_mark_reclaim(clp);
 910         status = -ENOENT;
 911         if (cred != NULL) {
 912                 status = nfs4_init_client(clp, cred);
 913                 put_rpccred(cred);
 914         }
 915         if (status)
 916                 goto out_error;
 917         /* Mark all delegations for reclaim */
 918         nfs_delegation_mark_reclaim(clp);
 919         /* Note: list is protected by exclusive lock on cl->cl_sem */
 920         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 921                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 922                 status = nfs4_reclaim_open_state(ops, sp);
 923                 if (status < 0) {
 924                         if (status == -NFS4ERR_NO_GRACE) {
 925                                 ops = &nfs4_network_partition_recovery_ops;
 926                                 status = nfs4_reclaim_open_state(ops, sp);
 927                         }
 928                         if (status == -NFS4ERR_STALE_CLIENTID)
 929                                 goto restart_loop;
 930                         if (status == -NFS4ERR_EXPIRED)
 931                                 goto restart_loop;
 932                 }
 933         }
 934         nfs_delegation_reap_unclaimed(clp);
 935 out:
 936         up_write(&clp->cl_sem);
 937         unlock_kernel();
 938         if (status == -NFS4ERR_CB_PATH_DOWN)
 939                 nfs_handle_cb_pathdown(clp);
 940         nfs4_clear_recover_bit(clp);
 941         nfs_put_client(clp);
 942         module_put_and_exit(0);
 943         return 0;
 944 out_error:
 945         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 946                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
 947         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 948         goto out;
 949 }
 950
 951 /*
 952  * Local variables:
 953  *  c-basic-offset: 8
 954  * End:
 955  */