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 = kzalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
  95                 nfs_callback_down();
  96                 return NULL;
  97         }
  98         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
  99         init_rwsem(&clp->cl_sem);
 100         INIT_LIST_HEAD(&clp->cl_delegations);
 101         INIT_LIST_HEAD(&clp->cl_state_owners);
 102         INIT_LIST_HEAD(&clp->cl_unused);
 103         spin_lock_init(&clp->cl_lock);
 104         atomic_set(&clp->cl_count, 1);
 105         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
 106         INIT_LIST_HEAD(&clp->cl_superblocks);
 107         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
 108         clp->cl_rpcclient = ERR_PTR(-EINVAL);
 109         clp->cl_boot_time = CURRENT_TIME;
 110         clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
 111         return clp;
 112 }
 113
 114 static void
 115 nfs4_free_client(struct nfs4_client *clp)
 116 {
 117         struct nfs4_state_owner *sp;
 118
 119         while (!list_empty(&clp->cl_unused)) {
 120                 sp = list_entry(clp->cl_unused.next,
 121                                 struct nfs4_state_owner,
 122                                 so_list);
 123                 list_del(&sp->so_list);
 124                 kfree(sp);
 125         }
 126         BUG_ON(!list_empty(&clp->cl_state_owners));
 127         nfs_idmap_delete(clp);
 128         if (!IS_ERR(clp->cl_rpcclient))
 129                 rpc_shutdown_client(clp->cl_rpcclient);
 130         kfree(clp);
 131         nfs_callback_down();
 132 }
 133
 134 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 135 {
 136         struct nfs4_client *clp;
 137         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 138                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 139                         atomic_inc(&clp->cl_count);
 140                         return clp;
 141                 }
 142         }
 143         return NULL;
 144 }
 145
 146 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 147 {
 148         struct nfs4_client *clp;
 149         spin_lock(&state_spinlock);
 150         clp = __nfs4_find_client(addr);
 151         spin_unlock(&state_spinlock);
 152         return clp;
 153 }
 154
 155 struct nfs4_client *
 156 nfs4_get_client(struct in_addr *addr)
 157 {
 158         struct nfs4_client *clp, *new = NULL;
 159
 160         spin_lock(&state_spinlock);
 161         for (;;) {
 162                 clp = __nfs4_find_client(addr);
 163                 if (clp != NULL)
 164                         break;
 165                 clp = new;
 166                 if (clp != NULL) {
 167                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 168                         new = NULL;
 169                         break;
 170                 }
 171                 spin_unlock(&state_spinlock);
 172                 new = nfs4_alloc_client(addr);
 173                 spin_lock(&state_spinlock);
 174                 if (new == NULL)
 175                         break;
 176         }
 177         spin_unlock(&state_spinlock);
 178         if (new)
 179                 nfs4_free_client(new);
 180         return clp;
 181 }
 182
 183 void
 184 nfs4_put_client(struct nfs4_client *clp)
 185 {
 186         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 187                 return;
 188         list_del(&clp->cl_servers);
 189         spin_unlock(&state_spinlock);
 190         BUG_ON(!list_empty(&clp->cl_superblocks));
 191         rpc_wake_up(&clp->cl_rpcwaitq);
 192         nfs4_kill_renewd(clp);
 193         nfs4_free_client(clp);
 194 }
 195
 196 static int nfs4_init_client(struct nfs4_client *clp, struct rpc_cred *cred)
 197 {
 198         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
 199                         nfs_callback_tcpport, cred);
 200         if (status == 0)
 201                 status = nfs4_proc_setclientid_confirm(clp, cred);
 202         if (status == 0)
 203                 nfs4_schedule_state_renewal(clp);
 204         return status;
 205 }
 206
 207 u32
 208 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 209 {
 210         return clp->cl_lockowner_id ++;
 211 }
 212
 213 static struct nfs4_state_owner *
 214 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 215 {
 216         struct nfs4_state_owner *sp = NULL;
 217
 218         if (!list_empty(&clp->cl_unused)) {
 219                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 220                 atomic_inc(&sp->so_count);
 221                 sp->so_cred = cred;
 222                 list_move(&sp->so_list, &clp->cl_state_owners);
 223                 clp->cl_nunused--;
 224         }
 225         return sp;
 226 }
 227
 228 struct rpc_cred *nfs4_get_renew_cred(struct nfs4_client *clp)
 229 {
 230         struct nfs4_state_owner *sp;
 231         struct rpc_cred *cred = NULL;
 232
 233         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 234                 if (list_empty(&sp->so_states))
 235                         continue;
 236                 cred = get_rpccred(sp->so_cred);
 237                 break;
 238         }
 239         return cred;
 240 }
 241
 242 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs4_client *clp)
 243 {
 244         struct nfs4_state_owner *sp;
 245
 246         if (!list_empty(&clp->cl_state_owners)) {
 247                 sp = list_entry(clp->cl_state_owners.next,
 248                                 struct nfs4_state_owner, so_list);
 249                 return get_rpccred(sp->so_cred);
 250         }
 251         return NULL;
 252 }
 253
 254 static struct nfs4_state_owner *
 255 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 256 {
 257         struct nfs4_state_owner *sp, *res = NULL;
 258
 259         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 260                 if (sp->so_cred != cred)
 261                         continue;
 262                 atomic_inc(&sp->so_count);
 263                 /* Move to the head of the list */
 264                 list_move(&sp->so_list, &clp->cl_state_owners);
 265                 res = sp;
 266                 break;
 267         }
 268         return res;
 269 }
 270
 271 /*
 272  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 273  * create a new state_owner.
 274  *
 275  */
 276 static struct nfs4_state_owner *
 277 nfs4_alloc_state_owner(void)
 278 {
 279         struct nfs4_state_owner *sp;
 280
 281         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 282         if (!sp)
 283                 return NULL;
 284         spin_lock_init(&sp->so_lock);
 285         INIT_LIST_HEAD(&sp->so_states);
 286         INIT_LIST_HEAD(&sp->so_delegations);
 287         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 288         sp->so_seqid.sequence = &sp->so_sequence;
 289         spin_lock_init(&sp->so_sequence.lock);
 290         INIT_LIST_HEAD(&sp->so_sequence.list);
 291         atomic_set(&sp->so_count, 1);
 292         return sp;
 293 }
 294
 295 void
 296 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 297 {
 298         struct nfs4_client *clp = sp->so_client;
 299         spin_lock(&clp->cl_lock);
 300         list_del_init(&sp->so_list);
 301         spin_unlock(&clp->cl_lock);
 302 }
 303
 304 /*
 305  * Note: must be called with clp->cl_sem held in order to prevent races
 306  *       with reboot recovery!
 307  */
 308 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 309 {
 310         struct nfs4_client *clp = server->nfs4_state;
 311         struct nfs4_state_owner *sp, *new;
 312
 313         get_rpccred(cred);
 314         new = nfs4_alloc_state_owner();
 315         spin_lock(&clp->cl_lock);
 316         sp = nfs4_find_state_owner(clp, cred);
 317         if (sp == NULL)
 318                 sp = nfs4_client_grab_unused(clp, cred);
 319         if (sp == NULL && new != NULL) {
 320                 list_add(&new->so_list, &clp->cl_state_owners);
 321                 new->so_client = clp;
 322                 new->so_id = nfs4_alloc_lockowner_id(clp);
 323                 new->so_cred = cred;
 324                 sp = new;
 325                 new = NULL;
 326         }
 327         spin_unlock(&clp->cl_lock);
 328         kfree(new);
 329         if (sp != NULL)
 330                 return sp;
 331         put_rpccred(cred);
 332         return NULL;
 333 }
 334
 335 /*
 336  * Must be called with clp->cl_sem held in order to avoid races
 337  * with state recovery...
 338  */
 339 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 340 {
 341         struct nfs4_client *clp = sp->so_client;
 342         struct rpc_cred *cred = sp->so_cred;
 343
 344         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 345                 return;
 346         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 347                 goto out_free;
 348         if (list_empty(&sp->so_list))
 349                 goto out_free;
 350         list_move(&sp->so_list, &clp->cl_unused);
 351         clp->cl_nunused++;
 352         spin_unlock(&clp->cl_lock);
 353         put_rpccred(cred);
 354         cred = NULL;
 355         return;
 356 out_free:
 357         list_del(&sp->so_list);
 358         spin_unlock(&clp->cl_lock);
 359         put_rpccred(cred);
 360         kfree(sp);
 361 }
 362
 363 static struct nfs4_state *
 364 nfs4_alloc_open_state(void)
 365 {
 366         struct nfs4_state *state;
 367
 368         state = kzalloc(sizeof(*state), GFP_KERNEL);
 369         if (!state)
 370                 return NULL;
 371         atomic_set(&state->count, 1);
 372         INIT_LIST_HEAD(&state->lock_states);
 373         spin_lock_init(&state->state_lock);
 374         return state;
 375 }
 376
 377 void
 378 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 379 {
 380         if (state->state == mode)
 381                 return;
 382         /* NB! List reordering - see the reclaim code for why.  */
 383         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 384                 if (mode & FMODE_WRITE)
 385                         list_move(&state->open_states, &state->owner->so_states);
 386                 else
 387                         list_move_tail(&state->open_states, &state->owner->so_states);
 388         }
 389         if (mode == 0)
 390                 list_del_init(&state->inode_states);
 391         state->state = mode;
 392 }
 393
 394 static struct nfs4_state *
 395 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 396 {
 397         struct nfs_inode *nfsi = NFS_I(inode);
 398         struct nfs4_state *state;
 399
 400         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 401                 /* Is this in the process of being freed? */
 402                 if (state->state == 0)
 403                         continue;
 404                 if (state->owner == owner) {
 405                         atomic_inc(&state->count);
 406                         return state;
 407                 }
 408         }
 409         return NULL;
 410 }
 411
 412 static void
 413 nfs4_free_open_state(struct nfs4_state *state)
 414 {
 415         kfree(state);
 416 }
 417
 418 struct nfs4_state *
 419 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 420 {
 421         struct nfs4_state *state, *new;
 422         struct nfs_inode *nfsi = NFS_I(inode);
 423
 424         spin_lock(&inode->i_lock);
 425         state = __nfs4_find_state_byowner(inode, owner);
 426         spin_unlock(&inode->i_lock);
 427         if (state)
 428                 goto out;
 429         new = nfs4_alloc_open_state();
 430         spin_lock(&owner->so_lock);
 431         spin_lock(&inode->i_lock);
 432         state = __nfs4_find_state_byowner(inode, owner);
 433         if (state == NULL && new != NULL) {
 434                 state = new;
 435                 state->owner = owner;
 436                 atomic_inc(&owner->so_count);
 437                 list_add(&state->inode_states, &nfsi->open_states);
 438                 state->inode = igrab(inode);
 439                 spin_unlock(&inode->i_lock);
 440                 /* Note: The reclaim code dictates that we add stateless
 441                  * and read-only stateids to the end of the list */
 442                 list_add_tail(&state->open_states, &owner->so_states);
 443                 spin_unlock(&owner->so_lock);
 444         } else {
 445                 spin_unlock(&inode->i_lock);
 446                 spin_unlock(&owner->so_lock);
 447                 if (new)
 448                         nfs4_free_open_state(new);
 449         }
 450 out:
 451         return state;
 452 }
 453
 454 /*
 455  * Beware! Caller must be holding exactly one
 456  * reference to clp->cl_sem!
 457  */
 458 void nfs4_put_open_state(struct nfs4_state *state)
 459 {
 460         struct inode *inode = state->inode;
 461         struct nfs4_state_owner *owner = state->owner;
 462
 463         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 464                 return;
 465         spin_lock(&inode->i_lock);
 466         if (!list_empty(&state->inode_states))
 467                 list_del(&state->inode_states);
 468         list_del(&state->open_states);
 469         spin_unlock(&inode->i_lock);
 470         spin_unlock(&owner->so_lock);
 471         iput(inode);
 472         nfs4_free_open_state(state);
 473         nfs4_put_state_owner(owner);
 474 }
 475
 476 /*
 477  * Close the current file.
 478  */
 479 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 480 {
 481         struct inode *inode = state->inode;
 482         struct nfs4_state_owner *owner = state->owner;
 483         int oldstate, newstate = 0;
 484
 485         atomic_inc(&owner->so_count);
 486         /* Protect against nfs4_find_state() */
 487         spin_lock(&owner->so_lock);
 488         spin_lock(&inode->i_lock);
 489         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 490                 case FMODE_READ:
 491                         state->n_rdonly--;
 492                         break;
 493                 case FMODE_WRITE:
 494                         state->n_wronly--;
 495                         break;
 496                 case FMODE_READ|FMODE_WRITE:
 497                         state->n_rdwr--;
 498         }
 499         oldstate = newstate = state->state;
 500         if (state->n_rdwr == 0) {
 501                 if (state->n_rdonly == 0)
 502                         newstate &= ~FMODE_READ;
 503                 if (state->n_wronly == 0)
 504                         newstate &= ~FMODE_WRITE;
 505         }
 506         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 507                 nfs4_state_set_mode_locked(state, newstate);
 508                 oldstate = newstate;
 509         }
 510         spin_unlock(&inode->i_lock);
 511         spin_unlock(&owner->so_lock);
 512
 513         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
 514                 return;
 515         nfs4_put_open_state(state);
 516         nfs4_put_state_owner(owner);
 517 }
 518
 519 /*
 520  * Search the state->lock_states for an existing lock_owner
 521  * that is compatible with current->files
 522  */
 523 static struct nfs4_lock_state *
 524 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 525 {
 526         struct nfs4_lock_state *pos;
 527         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 528                 if (pos->ls_owner != fl_owner)
 529                         continue;
 530                 atomic_inc(&pos->ls_count);
 531                 return pos;
 532         }
 533         return NULL;
 534 }
 535
 536 /*
 537  * Return a compatible lock_state. If no initialized lock_state structure
 538  * exists, return an uninitialized one.
 539  *
 540  */
 541 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 542 {
 543         struct nfs4_lock_state *lsp;
 544         struct nfs4_client *clp = state->owner->so_client;
 545
 546         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 547         if (lsp == NULL)
 548                 return NULL;
 549         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 550         atomic_set(&lsp->ls_count, 1);
 551         lsp->ls_owner = fl_owner;
 552         spin_lock(&clp->cl_lock);
 553         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 554         spin_unlock(&clp->cl_lock);
 555         INIT_LIST_HEAD(&lsp->ls_locks);
 556         return lsp;
 557 }
 558
 559 /*
 560  * Return a compatible lock_state. If no initialized lock_state structure
 561  * exists, return an uninitialized one.
 562  *
 563  * The caller must be holding clp->cl_sem
 564  */
 565 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 566 {
 567         struct nfs4_lock_state *lsp, *new = NULL;
 568
 569         for(;;) {
 570                 spin_lock(&state->state_lock);
 571                 lsp = __nfs4_find_lock_state(state, owner);
 572                 if (lsp != NULL)
 573                         break;
 574                 if (new != NULL) {
 575                         new->ls_state = state;
 576                         list_add(&new->ls_locks, &state->lock_states);
 577                         set_bit(LK_STATE_IN_USE, &state->flags);
 578                         lsp = new;
 579                         new = NULL;
 580                         break;
 581                 }
 582                 spin_unlock(&state->state_lock);
 583                 new = nfs4_alloc_lock_state(state, owner);
 584                 if (new == NULL)
 585                         return NULL;
 586         }
 587         spin_unlock(&state->state_lock);
 588         kfree(new);
 589         return lsp;
 590 }
 591
 592 /*
 593  * Release reference to lock_state, and free it if we see that
 594  * it is no longer in use
 595  */
 596 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 597 {
 598         struct nfs4_state *state;
 599
 600         if (lsp == NULL)
 601                 return;
 602         state = lsp->ls_state;
 603         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 604                 return;
 605         list_del(&lsp->ls_locks);
 606         if (list_empty(&state->lock_states))
 607                 clear_bit(LK_STATE_IN_USE, &state->flags);
 608         spin_unlock(&state->state_lock);
 609         kfree(lsp);
 610 }
 611
 612 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 613 {
 614         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 615
 616         dst->fl_u.nfs4_fl.owner = lsp;
 617         atomic_inc(&lsp->ls_count);
 618 }
 619
 620 static void nfs4_fl_release_lock(struct file_lock *fl)
 621 {
 622         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 623 }
 624
 625 static struct file_lock_operations nfs4_fl_lock_ops = {
 626         .fl_copy_lock = nfs4_fl_copy_lock,
 627         .fl_release_private = nfs4_fl_release_lock,
 628 };
 629
 630 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 631 {
 632         struct nfs4_lock_state *lsp;
 633
 634         if (fl->fl_ops != NULL)
 635                 return 0;
 636         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 637         if (lsp == NULL)
 638                 return -ENOMEM;
 639         fl->fl_u.nfs4_fl.owner = lsp;
 640         fl->fl_ops = &nfs4_fl_lock_ops;
 641         return 0;
 642 }
 643
 644 /*
 645  * Byte-range lock aware utility to initialize the stateid of read/write
 646  * requests.
 647  */
 648 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 649 {
 650         struct nfs4_lock_state *lsp;
 651
 652         memcpy(dst, &state->stateid, sizeof(*dst));
 653         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 654                 return;
 655
 656         spin_lock(&state->state_lock);
 657         lsp = __nfs4_find_lock_state(state, fl_owner);
 658         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 659                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 660         spin_unlock(&state->state_lock);
 661         nfs4_put_lock_state(lsp);
 662 }
 663
 664 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 665 {
 666         struct rpc_sequence *sequence = counter->sequence;
 667         struct nfs_seqid *new;
 668
 669         new = kmalloc(sizeof(*new), GFP_KERNEL);
 670         if (new != NULL) {
 671                 new->sequence = counter;
 672                 spin_lock(&sequence->lock);
 673                 list_add_tail(&new->list, &sequence->list);
 674                 spin_unlock(&sequence->lock);
 675         }
 676         return new;
 677 }
 678
 679 void nfs_free_seqid(struct nfs_seqid *seqid)
 680 {
 681         struct rpc_sequence *sequence = seqid->sequence->sequence;
 682
 683         spin_lock(&sequence->lock);
 684         list_del(&seqid->list);
 685         spin_unlock(&sequence->lock);
 686         rpc_wake_up(&sequence->wait);
 687         kfree(seqid);
 688 }
 689
 690 /*
 691  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 692  * failed with a seqid incrementing error -
 693  * see comments nfs_fs.h:seqid_mutating_error()
 694  */
 695 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 696 {
 697         switch (status) {
 698                 case 0:
 699                         break;
 700                 case -NFS4ERR_BAD_SEQID:
 701                 case -NFS4ERR_STALE_CLIENTID:
 702                 case -NFS4ERR_STALE_STATEID:
 703                 case -NFS4ERR_BAD_STATEID:
 704                 case -NFS4ERR_BADXDR:
 705                 case -NFS4ERR_RESOURCE:
 706                 case -NFS4ERR_NOFILEHANDLE:
 707                         /* Non-seqid mutating errors */
 708                         return;
 709         };
 710         /*
 711          * Note: no locking needed as we are guaranteed to be first
 712          * on the sequence list
 713          */
 714         seqid->sequence->counter++;
 715 }
 716
 717 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 718 {
 719         if (status == -NFS4ERR_BAD_SEQID) {
 720                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 721                                 struct nfs4_state_owner, so_seqid);
 722                 nfs4_drop_state_owner(sp);
 723         }
 724         return nfs_increment_seqid(status, seqid);
 725 }
 726
 727 /*
 728  * Increment the seqid if the LOCK/LOCKU succeeded, or
 729  * failed with a seqid incrementing error -
 730  * see comments nfs_fs.h:seqid_mutating_error()
 731  */
 732 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 733 {
 734         return nfs_increment_seqid(status, seqid);
 735 }
 736
 737 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 738 {
 739         struct rpc_sequence *sequence = seqid->sequence->sequence;
 740         int status = 0;
 741
 742         if (sequence->list.next == &seqid->list)
 743                 goto out;
 744         spin_lock(&sequence->lock);
 745         if (sequence->list.next != &seqid->list) {
 746                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 747                 status = -EAGAIN;
 748         }
 749         spin_unlock(&sequence->lock);
 750 out:
 751         return status;
 752 }
 753
 754 static int reclaimer(void *);
 755
 756 static inline void nfs4_clear_recover_bit(struct nfs4_client *clp)
 757 {
 758         smp_mb__before_clear_bit();
 759         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 760         smp_mb__after_clear_bit();
 761         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 762         rpc_wake_up(&clp->cl_rpcwaitq);
 763 }
 764
 765 /*
 766  * State recovery routine
 767  */
 768 static void nfs4_recover_state(struct nfs4_client *clp)
 769 {
 770         struct task_struct *task;
 771
 772         __module_get(THIS_MODULE);
 773         atomic_inc(&clp->cl_count);
 774         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 775                         NIPQUAD(clp->cl_addr));
 776         if (!IS_ERR(task))
 777                 return;
 778         nfs4_clear_recover_bit(clp);
 779         nfs4_put_client(clp);
 780         module_put(THIS_MODULE);
 781 }
 782
 783 /*
 784  * Schedule a state recovery attempt
 785  */
 786 void nfs4_schedule_state_recovery(struct nfs4_client *clp)
 787 {
 788         if (!clp)
 789                 return;
 790         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 791                 nfs4_recover_state(clp);
 792 }
 793
 794 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 795 {
 796         struct inode *inode = state->inode;
 797         struct file_lock *fl;
 798         int status = 0;
 799
 800         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 801                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 802                         continue;
 803                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 804                         continue;
 805                 status = ops->recover_lock(state, fl);
 806                 if (status >= 0)
 807                         continue;
 808                 switch (status) {
 809                         default:
 810                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 811                                                 __FUNCTION__, status);
 812                         case -NFS4ERR_EXPIRED:
 813                         case -NFS4ERR_NO_GRACE:
 814                         case -NFS4ERR_RECLAIM_BAD:
 815                         case -NFS4ERR_RECLAIM_CONFLICT:
 816                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 817                                 break;
 818                         case -NFS4ERR_STALE_CLIENTID:
 819                                 goto out_err;
 820                 }
 821         }
 822         return 0;
 823 out_err:
 824         return status;
 825 }
 826
 827 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 828 {
 829         struct nfs4_state *state;
 830         struct nfs4_lock_state *lock;
 831         int status = 0;
 832
 833         /* Note: we rely on the sp->so_states list being ordered
 834          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 835          * states first.
 836          * This is needed to ensure that the server won't give us any
 837          * read delegations that we have to return if, say, we are
 838          * recovering after a network partition or a reboot from a
 839          * server that doesn't support a grace period.
 840          */
 841         list_for_each_entry(state, &sp->so_states, open_states) {
 842                 if (state->state == 0)
 843                         continue;
 844                 status = ops->recover_open(sp, state);
 845                 if (status >= 0) {
 846                         status = nfs4_reclaim_locks(ops, state);
 847                         if (status < 0)
 848                                 goto out_err;
 849                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 850                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 851                                         printk("%s: Lock reclaim failed!\n",
 852                                                         __FUNCTION__);
 853                         }
 854                         continue;
 855                 }
 856                 switch (status) {
 857                         default:
 858                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 859                                                 __FUNCTION__, status);
 860                         case -ENOENT:
 861                         case -NFS4ERR_RECLAIM_BAD:
 862                         case -NFS4ERR_RECLAIM_CONFLICT:
 863                                 /*
 864                                  * Open state on this file cannot be recovered
 865                                  * All we can do is revert to using the zero stateid.
 866                                  */
 867                                 memset(state->stateid.data, 0,
 868                                         sizeof(state->stateid.data));
 869                                 /* Mark the file as being 'closed' */
 870                                 state->state = 0;
 871                                 break;
 872                         case -NFS4ERR_EXPIRED:
 873                         case -NFS4ERR_NO_GRACE:
 874                         case -NFS4ERR_STALE_CLIENTID:
 875                                 goto out_err;
 876                 }
 877         }
 878         return 0;
 879 out_err:
 880         return status;
 881 }
 882
 883 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 884 {
 885         struct nfs4_state_owner *sp;
 886         struct nfs4_state *state;
 887         struct nfs4_lock_state *lock;
 888
 889         /* Reset all sequence ids to zero */
 890         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 891                 sp->so_seqid.counter = 0;
 892                 sp->so_seqid.flags = 0;
 893                 spin_lock(&sp->so_lock);
 894                 list_for_each_entry(state, &sp->so_states, open_states) {
 895                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 896                                 lock->ls_seqid.counter = 0;
 897                                 lock->ls_seqid.flags = 0;
 898                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 899                         }
 900                 }
 901                 spin_unlock(&sp->so_lock);
 902         }
 903 }
 904
 905 static int reclaimer(void *ptr)
 906 {
 907         struct nfs4_client *clp = ptr;
 908         struct nfs4_state_owner *sp;
 909         struct nfs4_state_recovery_ops *ops;
 910         struct rpc_cred *cred;
 911         int status = 0;
 912
 913         allow_signal(SIGKILL);
 914
 915         /* Ensure exclusive access to NFSv4 state */
 916         lock_kernel();
 917         down_write(&clp->cl_sem);
 918         /* Are there any NFS mounts out there? */
 919         if (list_empty(&clp->cl_superblocks))
 920                 goto out;
 921 restart_loop:
 922         ops = &nfs4_network_partition_recovery_ops;
 923         /* Are there any open files on this volume? */
 924         cred = nfs4_get_renew_cred(clp);
 925         if (cred != NULL) {
 926                 /* Yes there are: try to renew the old lease */
 927                 status = nfs4_proc_renew(clp, cred);
 928                 switch (status) {
 929                         case 0:
 930                         case -NFS4ERR_CB_PATH_DOWN:
 931                                 put_rpccred(cred);
 932                                 goto out;
 933                         case -NFS4ERR_STALE_CLIENTID:
 934                         case -NFS4ERR_LEASE_MOVED:
 935                                 ops = &nfs4_reboot_recovery_ops;
 936                 }
 937         } else {
 938                 /* "reboot" to ensure we clear all state on the server */
 939                 clp->cl_boot_time = CURRENT_TIME;
 940                 cred = nfs4_get_setclientid_cred(clp);
 941         }
 942         /* We're going to have to re-establish a clientid */
 943         nfs4_state_mark_reclaim(clp);
 944         status = -ENOENT;
 945         if (cred != NULL) {
 946                 status = nfs4_init_client(clp, cred);
 947                 put_rpccred(cred);
 948         }
 949         if (status)
 950                 goto out_error;
 951         /* Mark all delegations for reclaim */
 952         nfs_delegation_mark_reclaim(clp);
 953         /* Note: list is protected by exclusive lock on cl->cl_sem */
 954         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 955                 status = nfs4_reclaim_open_state(ops, sp);
 956                 if (status < 0) {
 957                         if (status == -NFS4ERR_NO_GRACE) {
 958                                 ops = &nfs4_network_partition_recovery_ops;
 959                                 status = nfs4_reclaim_open_state(ops, sp);
 960                         }
 961                         if (status == -NFS4ERR_STALE_CLIENTID)
 962                                 goto restart_loop;
 963                         if (status == -NFS4ERR_EXPIRED)
 964                                 goto restart_loop;
 965                 }
 966         }
 967         nfs_delegation_reap_unclaimed(clp);
 968 out:
 969         up_write(&clp->cl_sem);
 970         unlock_kernel();
 971         if (status == -NFS4ERR_CB_PATH_DOWN)
 972                 nfs_handle_cb_pathdown(clp);
 973         nfs4_clear_recover_bit(clp);
 974         nfs4_put_client(clp);
 975         module_put_and_exit(0);
 976         return 0;
 977 out_error:
 978         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 979                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 980         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 981         goto out;
 982 }
 983
 984 /*
 985  * Local variables:
 986  *  c-basic-offset: 8
 987  * End:
 988  */