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/config.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/workqueue.h>
  49 #include <linux/bitops.h>
  50
  51 #include "nfs4_fs.h"
  52 #include "callback.h"
  53 #include "delegation.h"
  54
  55 #define OPENOWNER_POOL_SIZE     8
  56
  57 const nfs4_stateid zero_stateid;
  58
  59 static DEFINE_SPINLOCK(state_spinlock);
  60 static LIST_HEAD(nfs4_clientid_list);
  61
  62 void
  63 init_nfsv4_state(struct nfs_server *server)
  64 {
  65         server->nfs4_state = NULL;
  66         INIT_LIST_HEAD(&server->nfs4_siblings);
  67 }
  68
  69 void
  70 destroy_nfsv4_state(struct nfs_server *server)
  71 {
  72         kfree(server->mnt_path);
  73         server->mnt_path = NULL;
  74         if (server->nfs4_state) {
  75                 nfs4_put_client(server->nfs4_state);
  76                 server->nfs4_state = NULL;
  77         }
  78 }
  79
  80 /*
  81  * nfs4_get_client(): returns an empty client structure
  82  * nfs4_put_client(): drops reference to client structure
  83  *
  84  * Since these are allocated/deallocated very rarely, we don't
  85  * bother putting them in a slab cache...
  86  */
  87 static struct nfs4_client *
  88 nfs4_alloc_client(struct in_addr *addr)
  89 {
  90         struct nfs4_client *clp;
  91
  92         if (nfs_callback_up() < 0)
  93                 return NULL;
  94         if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
  95                 nfs_callback_down();
  96                 return NULL;
  97         }
  98         memset(clp, 0, sizeof(*clp));
  99         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
 100         init_rwsem(&clp->cl_sem);
 101         INIT_LIST_HEAD(&clp->cl_delegations);
 102         INIT_LIST_HEAD(&clp->cl_state_owners);
 103         INIT_LIST_HEAD(&clp->cl_unused);
 104         spin_lock_init(&clp->cl_lock);
 105         atomic_set(&clp->cl_count, 1);
 106         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
 107         INIT_LIST_HEAD(&clp->cl_superblocks);
 108         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
 109         clp->cl_rpcclient = ERR_PTR(-EINVAL);
 110         clp->cl_boot_time = CURRENT_TIME;
 111         clp->cl_state = 0;
 112         return clp;
 113 }
 114
 115 static void
 116 nfs4_free_client(struct nfs4_client *clp)
 117 {
 118         struct nfs4_state_owner *sp;
 119
 120         while (!list_empty(&clp->cl_unused)) {
 121                 sp = list_entry(clp->cl_unused.next,
 122                                 struct nfs4_state_owner,
 123                                 so_list);
 124                 list_del(&sp->so_list);
 125                 kfree(sp);
 126         }
 127         BUG_ON(!list_empty(&clp->cl_state_owners));
 128         if (clp->cl_cred)
 129                 put_rpccred(clp->cl_cred);
 130         nfs_idmap_delete(clp);
 131         if (!IS_ERR(clp->cl_rpcclient))
 132                 rpc_shutdown_client(clp->cl_rpcclient);
 133         kfree(clp);
 134         nfs_callback_down();
 135 }
 136
 137 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 138 {
 139         struct nfs4_client *clp;
 140         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 141                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 142                         atomic_inc(&clp->cl_count);
 143                         return clp;
 144                 }
 145         }
 146         return NULL;
 147 }
 148
 149 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 150 {
 151         struct nfs4_client *clp;
 152         spin_lock(&state_spinlock);
 153         clp = __nfs4_find_client(addr);
 154         spin_unlock(&state_spinlock);
 155         return clp;
 156 }
 157
 158 struct nfs4_client *
 159 nfs4_get_client(struct in_addr *addr)
 160 {
 161         struct nfs4_client *clp, *new = NULL;
 162
 163         spin_lock(&state_spinlock);
 164         for (;;) {
 165                 clp = __nfs4_find_client(addr);
 166                 if (clp != NULL)
 167                         break;
 168                 clp = new;
 169                 if (clp != NULL) {
 170                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 171                         new = NULL;
 172                         break;
 173                 }
 174                 spin_unlock(&state_spinlock);
 175                 new = nfs4_alloc_client(addr);
 176                 spin_lock(&state_spinlock);
 177                 if (new == NULL)
 178                         break;
 179         }
 180         spin_unlock(&state_spinlock);
 181         if (new)
 182                 nfs4_free_client(new);
 183         return clp;
 184 }
 185
 186 void
 187 nfs4_put_client(struct nfs4_client *clp)
 188 {
 189         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 190                 return;
 191         list_del(&clp->cl_servers);
 192         spin_unlock(&state_spinlock);
 193         BUG_ON(!list_empty(&clp->cl_superblocks));
 194         rpc_wake_up(&clp->cl_rpcwaitq);
 195         nfs4_kill_renewd(clp);
 196         nfs4_free_client(clp);
 197 }
 198
 199 static int __nfs4_init_client(struct nfs4_client *clp)
 200 {
 201         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
 202         if (status == 0)
 203                 status = nfs4_proc_setclientid_confirm(clp);
 204         if (status == 0)
 205                 nfs4_schedule_state_renewal(clp);
 206         return status;
 207 }
 208
 209 int nfs4_init_client(struct nfs4_client *clp)
 210 {
 211         return nfs4_map_errors(__nfs4_init_client(clp));
 212 }
 213
 214 u32
 215 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 216 {
 217         return clp->cl_lockowner_id ++;
 218 }
 219
 220 static struct nfs4_state_owner *
 221 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 222 {
 223         struct nfs4_state_owner *sp = NULL;
 224
 225         if (!list_empty(&clp->cl_unused)) {
 226                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 227                 atomic_inc(&sp->so_count);
 228                 sp->so_cred = cred;
 229                 list_move(&sp->so_list, &clp->cl_state_owners);
 230                 clp->cl_nunused--;
 231         }
 232         return sp;
 233 }
 234
 235 struct rpc_cred *nfs4_get_renew_cred(struct nfs4_client *clp)
 236 {
 237         struct nfs4_state_owner *sp;
 238         struct rpc_cred *cred = NULL;
 239
 240         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 241                 if (list_empty(&sp->so_states))
 242                         continue;
 243                 cred = get_rpccred(sp->so_cred);
 244                 break;
 245         }
 246         return cred;
 247 }
 248
 249 static struct nfs4_state_owner *
 250 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 251 {
 252         struct nfs4_state_owner *sp, *res = NULL;
 253
 254         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 255                 if (sp->so_cred != cred)
 256                         continue;
 257                 atomic_inc(&sp->so_count);
 258                 /* Move to the head of the list */
 259                 list_move(&sp->so_list, &clp->cl_state_owners);
 260                 res = sp;
 261                 break;
 262         }
 263         return res;
 264 }
 265
 266 /*
 267  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 268  * create a new state_owner.
 269  *
 270  */
 271 static struct nfs4_state_owner *
 272 nfs4_alloc_state_owner(void)
 273 {
 274         struct nfs4_state_owner *sp;
 275
 276         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 277         if (!sp)
 278                 return NULL;
 279         spin_lock_init(&sp->so_lock);
 280         INIT_LIST_HEAD(&sp->so_states);
 281         INIT_LIST_HEAD(&sp->so_delegations);
 282         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 283         sp->so_seqid.sequence = &sp->so_sequence;
 284         spin_lock_init(&sp->so_sequence.lock);
 285         INIT_LIST_HEAD(&sp->so_sequence.list);
 286         atomic_set(&sp->so_count, 1);
 287         return sp;
 288 }
 289
 290 void
 291 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 292 {
 293         struct nfs4_client *clp = sp->so_client;
 294         spin_lock(&clp->cl_lock);
 295         list_del_init(&sp->so_list);
 296         spin_unlock(&clp->cl_lock);
 297 }
 298
 299 /*
 300  * Note: must be called with clp->cl_sem held in order to prevent races
 301  *       with reboot recovery!
 302  */
 303 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 304 {
 305         struct nfs4_client *clp = server->nfs4_state;
 306         struct nfs4_state_owner *sp, *new;
 307
 308         get_rpccred(cred);
 309         new = nfs4_alloc_state_owner();
 310         spin_lock(&clp->cl_lock);
 311         sp = nfs4_find_state_owner(clp, cred);
 312         if (sp == NULL)
 313                 sp = nfs4_client_grab_unused(clp, cred);
 314         if (sp == NULL && new != NULL) {
 315                 list_add(&new->so_list, &clp->cl_state_owners);
 316                 new->so_client = clp;
 317                 new->so_id = nfs4_alloc_lockowner_id(clp);
 318                 new->so_cred = cred;
 319                 sp = new;
 320                 new = NULL;
 321         }
 322         spin_unlock(&clp->cl_lock);
 323         kfree(new);
 324         if (sp != NULL)
 325                 return sp;
 326         put_rpccred(cred);
 327         return NULL;
 328 }
 329
 330 /*
 331  * Must be called with clp->cl_sem held in order to avoid races
 332  * with state recovery...
 333  */
 334 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 335 {
 336         struct nfs4_client *clp = sp->so_client;
 337         struct rpc_cred *cred = sp->so_cred;
 338
 339         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 340                 return;
 341         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 342                 goto out_free;
 343         if (list_empty(&sp->so_list))
 344                 goto out_free;
 345         list_move(&sp->so_list, &clp->cl_unused);
 346         clp->cl_nunused++;
 347         spin_unlock(&clp->cl_lock);
 348         put_rpccred(cred);
 349         cred = NULL;
 350         return;
 351 out_free:
 352         list_del(&sp->so_list);
 353         spin_unlock(&clp->cl_lock);
 354         put_rpccred(cred);
 355         kfree(sp);
 356 }
 357
 358 static struct nfs4_state *
 359 nfs4_alloc_open_state(void)
 360 {
 361         struct nfs4_state *state;
 362
 363         state = kzalloc(sizeof(*state), GFP_KERNEL);
 364         if (!state)
 365                 return NULL;
 366         atomic_set(&state->count, 1);
 367         INIT_LIST_HEAD(&state->lock_states);
 368         spin_lock_init(&state->state_lock);
 369         return state;
 370 }
 371
 372 void
 373 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 374 {
 375         if (state->state == mode)
 376                 return;
 377         /* NB! List reordering - see the reclaim code for why.  */
 378         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 379                 if (mode & FMODE_WRITE)
 380                         list_move(&state->open_states, &state->owner->so_states);
 381                 else
 382                         list_move_tail(&state->open_states, &state->owner->so_states);
 383         }
 384         if (mode == 0)
 385                 list_del_init(&state->inode_states);
 386         state->state = mode;
 387 }
 388
 389 static struct nfs4_state *
 390 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 391 {
 392         struct nfs_inode *nfsi = NFS_I(inode);
 393         struct nfs4_state *state;
 394
 395         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 396                 /* Is this in the process of being freed? */
 397                 if (state->state == 0)
 398                         continue;
 399                 if (state->owner == owner) {
 400                         atomic_inc(&state->count);
 401                         return state;
 402                 }
 403         }
 404         return NULL;
 405 }
 406
 407 static void
 408 nfs4_free_open_state(struct nfs4_state *state)
 409 {
 410         kfree(state);
 411 }
 412
 413 struct nfs4_state *
 414 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 415 {
 416         struct nfs4_state *state, *new;
 417         struct nfs_inode *nfsi = NFS_I(inode);
 418
 419         spin_lock(&inode->i_lock);
 420         state = __nfs4_find_state_byowner(inode, owner);
 421         spin_unlock(&inode->i_lock);
 422         if (state)
 423                 goto out;
 424         new = nfs4_alloc_open_state();
 425         spin_lock(&owner->so_lock);
 426         spin_lock(&inode->i_lock);
 427         state = __nfs4_find_state_byowner(inode, owner);
 428         if (state == NULL && new != NULL) {
 429                 state = new;
 430                 state->owner = owner;
 431                 atomic_inc(&owner->so_count);
 432                 list_add(&state->inode_states, &nfsi->open_states);
 433                 state->inode = igrab(inode);
 434                 spin_unlock(&inode->i_lock);
 435                 /* Note: The reclaim code dictates that we add stateless
 436                  * and read-only stateids to the end of the list */
 437                 list_add_tail(&state->open_states, &owner->so_states);
 438                 spin_unlock(&owner->so_lock);
 439         } else {
 440                 spin_unlock(&inode->i_lock);
 441                 spin_unlock(&owner->so_lock);
 442                 if (new)
 443                         nfs4_free_open_state(new);
 444         }
 445 out:
 446         return state;
 447 }
 448
 449 /*
 450  * Beware! Caller must be holding exactly one
 451  * reference to clp->cl_sem!
 452  */
 453 void nfs4_put_open_state(struct nfs4_state *state)
 454 {
 455         struct inode *inode = state->inode;
 456         struct nfs4_state_owner *owner = state->owner;
 457
 458         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 459                 return;
 460         spin_lock(&inode->i_lock);
 461         if (!list_empty(&state->inode_states))
 462                 list_del(&state->inode_states);
 463         list_del(&state->open_states);
 464         spin_unlock(&inode->i_lock);
 465         spin_unlock(&owner->so_lock);
 466         iput(inode);
 467         nfs4_free_open_state(state);
 468         nfs4_put_state_owner(owner);
 469 }
 470
 471 /*
 472  * Close the current file.
 473  */
 474 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 475 {
 476         struct inode *inode = state->inode;
 477         struct nfs4_state_owner *owner = state->owner;
 478         int oldstate, newstate = 0;
 479
 480         atomic_inc(&owner->so_count);
 481         /* Protect against nfs4_find_state() */
 482         spin_lock(&owner->so_lock);
 483         spin_lock(&inode->i_lock);
 484         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 485                 case FMODE_READ:
 486                         state->n_rdonly--;
 487                         break;
 488                 case FMODE_WRITE:
 489                         state->n_wronly--;
 490                         break;
 491                 case FMODE_READ|FMODE_WRITE:
 492                         state->n_rdwr--;
 493         }
 494         oldstate = newstate = state->state;
 495         if (state->n_rdwr == 0) {
 496                 if (state->n_rdonly == 0)
 497                         newstate &= ~FMODE_READ;
 498                 if (state->n_wronly == 0)
 499                         newstate &= ~FMODE_WRITE;
 500         }
 501         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 502                 nfs4_state_set_mode_locked(state, newstate);
 503                 oldstate = newstate;
 504         }
 505         spin_unlock(&inode->i_lock);
 506         spin_unlock(&owner->so_lock);
 507
 508         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
 509                 return;
 510         nfs4_put_open_state(state);
 511         nfs4_put_state_owner(owner);
 512 }
 513
 514 /*
 515  * Search the state->lock_states for an existing lock_owner
 516  * that is compatible with current->files
 517  */
 518 static struct nfs4_lock_state *
 519 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 520 {
 521         struct nfs4_lock_state *pos;
 522         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 523                 if (pos->ls_owner != fl_owner)
 524                         continue;
 525                 atomic_inc(&pos->ls_count);
 526                 return pos;
 527         }
 528         return NULL;
 529 }
 530
 531 /*
 532  * Return a compatible lock_state. If no initialized lock_state structure
 533  * exists, return an uninitialized one.
 534  *
 535  */
 536 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 537 {
 538         struct nfs4_lock_state *lsp;
 539         struct nfs4_client *clp = state->owner->so_client;
 540
 541         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 542         if (lsp == NULL)
 543                 return NULL;
 544         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 545         atomic_set(&lsp->ls_count, 1);
 546         lsp->ls_owner = fl_owner;
 547         spin_lock(&clp->cl_lock);
 548         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 549         spin_unlock(&clp->cl_lock);
 550         INIT_LIST_HEAD(&lsp->ls_locks);
 551         return lsp;
 552 }
 553
 554 /*
 555  * Return a compatible lock_state. If no initialized lock_state structure
 556  * exists, return an uninitialized one.
 557  *
 558  * The caller must be holding clp->cl_sem
 559  */
 560 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 561 {
 562         struct nfs4_lock_state *lsp, *new = NULL;
 563
 564         for(;;) {
 565                 spin_lock(&state->state_lock);
 566                 lsp = __nfs4_find_lock_state(state, owner);
 567                 if (lsp != NULL)
 568                         break;
 569                 if (new != NULL) {
 570                         new->ls_state = state;
 571                         list_add(&new->ls_locks, &state->lock_states);
 572                         set_bit(LK_STATE_IN_USE, &state->flags);
 573                         lsp = new;
 574                         new = NULL;
 575                         break;
 576                 }
 577                 spin_unlock(&state->state_lock);
 578                 new = nfs4_alloc_lock_state(state, owner);
 579                 if (new == NULL)
 580                         return NULL;
 581         }
 582         spin_unlock(&state->state_lock);
 583         kfree(new);
 584         return lsp;
 585 }
 586
 587 /*
 588  * Release reference to lock_state, and free it if we see that
 589  * it is no longer in use
 590  */
 591 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 592 {
 593         struct nfs4_state *state;
 594
 595         if (lsp == NULL)
 596                 return;
 597         state = lsp->ls_state;
 598         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 599                 return;
 600         list_del(&lsp->ls_locks);
 601         if (list_empty(&state->lock_states))
 602                 clear_bit(LK_STATE_IN_USE, &state->flags);
 603         spin_unlock(&state->state_lock);
 604         kfree(lsp);
 605 }
 606
 607 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 608 {
 609         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 610
 611         dst->fl_u.nfs4_fl.owner = lsp;
 612         atomic_inc(&lsp->ls_count);
 613 }
 614
 615 static void nfs4_fl_release_lock(struct file_lock *fl)
 616 {
 617         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 618 }
 619
 620 static struct file_lock_operations nfs4_fl_lock_ops = {
 621         .fl_copy_lock = nfs4_fl_copy_lock,
 622         .fl_release_private = nfs4_fl_release_lock,
 623 };
 624
 625 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 626 {
 627         struct nfs4_lock_state *lsp;
 628
 629         if (fl->fl_ops != NULL)
 630                 return 0;
 631         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 632         if (lsp == NULL)
 633                 return -ENOMEM;
 634         fl->fl_u.nfs4_fl.owner = lsp;
 635         fl->fl_ops = &nfs4_fl_lock_ops;
 636         return 0;
 637 }
 638
 639 /*
 640  * Byte-range lock aware utility to initialize the stateid of read/write
 641  * requests.
 642  */
 643 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 644 {
 645         struct nfs4_lock_state *lsp;
 646
 647         memcpy(dst, &state->stateid, sizeof(*dst));
 648         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 649                 return;
 650
 651         spin_lock(&state->state_lock);
 652         lsp = __nfs4_find_lock_state(state, fl_owner);
 653         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 654                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 655         spin_unlock(&state->state_lock);
 656         nfs4_put_lock_state(lsp);
 657 }
 658
 659 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 660 {
 661         struct rpc_sequence *sequence = counter->sequence;
 662         struct nfs_seqid *new;
 663
 664         new = kmalloc(sizeof(*new), GFP_KERNEL);
 665         if (new != NULL) {
 666                 new->sequence = counter;
 667                 spin_lock(&sequence->lock);
 668                 list_add_tail(&new->list, &sequence->list);
 669                 spin_unlock(&sequence->lock);
 670         }
 671         return new;
 672 }
 673
 674 void nfs_free_seqid(struct nfs_seqid *seqid)
 675 {
 676         struct rpc_sequence *sequence = seqid->sequence->sequence;
 677
 678         spin_lock(&sequence->lock);
 679         list_del(&seqid->list);
 680         spin_unlock(&sequence->lock);
 681         rpc_wake_up(&sequence->wait);
 682         kfree(seqid);
 683 }
 684
 685 /*
 686  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 687  * failed with a seqid incrementing error -
 688  * see comments nfs_fs.h:seqid_mutating_error()
 689  */
 690 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 691 {
 692         switch (status) {
 693                 case 0:
 694                         break;
 695                 case -NFS4ERR_BAD_SEQID:
 696                 case -NFS4ERR_STALE_CLIENTID:
 697                 case -NFS4ERR_STALE_STATEID:
 698                 case -NFS4ERR_BAD_STATEID:
 699                 case -NFS4ERR_BADXDR:
 700                 case -NFS4ERR_RESOURCE:
 701                 case -NFS4ERR_NOFILEHANDLE:
 702                         /* Non-seqid mutating errors */
 703                         return;
 704         };
 705         /*
 706          * Note: no locking needed as we are guaranteed to be first
 707          * on the sequence list
 708          */
 709         seqid->sequence->counter++;
 710 }
 711
 712 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 713 {
 714         if (status == -NFS4ERR_BAD_SEQID) {
 715                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 716                                 struct nfs4_state_owner, so_seqid);
 717                 nfs4_drop_state_owner(sp);
 718         }
 719         return nfs_increment_seqid(status, seqid);
 720 }
 721
 722 /*
 723  * Increment the seqid if the LOCK/LOCKU succeeded, or
 724  * failed with a seqid incrementing error -
 725  * see comments nfs_fs.h:seqid_mutating_error()
 726  */
 727 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 728 {
 729         return nfs_increment_seqid(status, seqid);
 730 }
 731
 732 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 733 {
 734         struct rpc_sequence *sequence = seqid->sequence->sequence;
 735         int status = 0;
 736
 737         if (sequence->list.next == &seqid->list)
 738                 goto out;
 739         spin_lock(&sequence->lock);
 740         if (sequence->list.next != &seqid->list) {
 741                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 742                 status = -EAGAIN;
 743         }
 744         spin_unlock(&sequence->lock);
 745 out:
 746         return status;
 747 }
 748
 749 static int reclaimer(void *);
 750
 751 static inline void nfs4_clear_recover_bit(struct nfs4_client *clp)
 752 {
 753         smp_mb__before_clear_bit();
 754         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 755         smp_mb__after_clear_bit();
 756         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 757         rpc_wake_up(&clp->cl_rpcwaitq);
 758 }
 759
 760 /*
 761  * State recovery routine
 762  */
 763 static void nfs4_recover_state(struct nfs4_client *clp)
 764 {
 765         struct task_struct *task;
 766
 767         __module_get(THIS_MODULE);
 768         atomic_inc(&clp->cl_count);
 769         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 770                         NIPQUAD(clp->cl_addr));
 771         if (!IS_ERR(task))
 772                 return;
 773         nfs4_clear_recover_bit(clp);
 774         nfs4_put_client(clp);
 775         module_put(THIS_MODULE);
 776 }
 777
 778 /*
 779  * Schedule a state recovery attempt
 780  */
 781 void nfs4_schedule_state_recovery(struct nfs4_client *clp)
 782 {
 783         if (!clp)
 784                 return;
 785         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 786                 nfs4_recover_state(clp);
 787 }
 788
 789 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 790 {
 791         struct inode *inode = state->inode;
 792         struct file_lock *fl;
 793         int status = 0;
 794
 795         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 796                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 797                         continue;
 798                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 799                         continue;
 800                 status = ops->recover_lock(state, fl);
 801                 if (status >= 0)
 802                         continue;
 803                 switch (status) {
 804                         default:
 805                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 806                                                 __FUNCTION__, status);
 807                         case -NFS4ERR_EXPIRED:
 808                         case -NFS4ERR_NO_GRACE:
 809                         case -NFS4ERR_RECLAIM_BAD:
 810                         case -NFS4ERR_RECLAIM_CONFLICT:
 811                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 812                                 break;
 813                         case -NFS4ERR_STALE_CLIENTID:
 814                                 goto out_err;
 815                 }
 816         }
 817         return 0;
 818 out_err:
 819         return status;
 820 }
 821
 822 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 823 {
 824         struct nfs4_state *state;
 825         struct nfs4_lock_state *lock;
 826         int status = 0;
 827
 828         /* Note: we rely on the sp->so_states list being ordered
 829          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 830          * states first.
 831          * This is needed to ensure that the server won't give us any
 832          * read delegations that we have to return if, say, we are
 833          * recovering after a network partition or a reboot from a
 834          * server that doesn't support a grace period.
 835          */
 836         list_for_each_entry(state, &sp->so_states, open_states) {
 837                 if (state->state == 0)
 838                         continue;
 839                 status = ops->recover_open(sp, state);
 840                 if (status >= 0) {
 841                         status = nfs4_reclaim_locks(ops, state);
 842                         if (status < 0)
 843                                 goto out_err;
 844                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 845                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 846                                         printk("%s: Lock reclaim failed!\n",
 847                                                         __FUNCTION__);
 848                         }
 849                         continue;
 850                 }
 851                 switch (status) {
 852                         default:
 853                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 854                                                 __FUNCTION__, status);
 855                         case -ENOENT:
 856                         case -NFS4ERR_RECLAIM_BAD:
 857                         case -NFS4ERR_RECLAIM_CONFLICT:
 858                                 /*
 859                                  * Open state on this file cannot be recovered
 860                                  * All we can do is revert to using the zero stateid.
 861                                  */
 862                                 memset(state->stateid.data, 0,
 863                                         sizeof(state->stateid.data));
 864                                 /* Mark the file as being 'closed' */
 865                                 state->state = 0;
 866                                 break;
 867                         case -NFS4ERR_EXPIRED:
 868                         case -NFS4ERR_NO_GRACE:
 869                         case -NFS4ERR_STALE_CLIENTID:
 870                                 goto out_err;
 871                 }
 872         }
 873         return 0;
 874 out_err:
 875         return status;
 876 }
 877
 878 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 879 {
 880         struct nfs4_state_owner *sp;
 881         struct nfs4_state *state;
 882         struct nfs4_lock_state *lock;
 883
 884         /* Reset all sequence ids to zero */
 885         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 886                 sp->so_seqid.counter = 0;
 887                 sp->so_seqid.flags = 0;
 888                 spin_lock(&sp->so_lock);
 889                 list_for_each_entry(state, &sp->so_states, open_states) {
 890                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 891                                 lock->ls_seqid.counter = 0;
 892                                 lock->ls_seqid.flags = 0;
 893                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 894                         }
 895                 }
 896                 spin_unlock(&sp->so_lock);
 897         }
 898 }
 899
 900 static int reclaimer(void *ptr)
 901 {
 902         struct nfs4_client *clp = ptr;
 903         struct nfs4_state_owner *sp;
 904         struct nfs4_state_recovery_ops *ops;
 905         int status = 0;
 906
 907         allow_signal(SIGKILL);
 908
 909         /* Ensure exclusive access to NFSv4 state */
 910         lock_kernel();
 911         down_write(&clp->cl_sem);
 912         /* Are there any NFS mounts out there? */
 913         if (list_empty(&clp->cl_superblocks))
 914                 goto out;
 915 restart_loop:
 916         status = nfs4_proc_renew(clp, clp->cl_cred);
 917         switch (status) {
 918                 case 0:
 919                 case -NFS4ERR_CB_PATH_DOWN:
 920                         goto out;
 921                 case -NFS4ERR_STALE_CLIENTID:
 922                 case -NFS4ERR_LEASE_MOVED:
 923                         ops = &nfs4_reboot_recovery_ops;
 924                         break;
 925                 default:
 926                         ops = &nfs4_network_partition_recovery_ops;
 927         };
 928         nfs4_state_mark_reclaim(clp);
 929         status = __nfs4_init_client(clp);
 930         if (status)
 931                 goto out_error;
 932         /* Mark all delegations for reclaim */
 933         nfs_delegation_mark_reclaim(clp);
 934         /* Note: list is protected by exclusive lock on cl->cl_sem */
 935         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 936                 status = nfs4_reclaim_open_state(ops, sp);
 937                 if (status < 0) {
 938                         if (status == -NFS4ERR_NO_GRACE) {
 939                                 ops = &nfs4_network_partition_recovery_ops;
 940                                 status = nfs4_reclaim_open_state(ops, sp);
 941                         }
 942                         if (status == -NFS4ERR_STALE_CLIENTID)
 943                                 goto restart_loop;
 944                         if (status == -NFS4ERR_EXPIRED)
 945                                 goto restart_loop;
 946                 }
 947         }
 948         nfs_delegation_reap_unclaimed(clp);
 949 out:
 950         up_write(&clp->cl_sem);
 951         unlock_kernel();
 952         if (status == -NFS4ERR_CB_PATH_DOWN)
 953                 nfs_handle_cb_pathdown(clp);
 954         nfs4_clear_recover_bit(clp);
 955         nfs4_put_client(clp);
 956         module_put_and_exit(0);
 957         return 0;
 958 out_error:
 959         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 960                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 961         goto out;
 962 }
 963
 964 /*
 965  * Local variables:
 966  *  c-basic-offset: 8
 967  * End:
 968  */