err.no Git - linux-2.6/blob - fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/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/workqueue.h>
  47 #include <linux/bitops.h>
  48
  49 #include "nfs4_fs.h"
  50 #include "callback.h"
  51 #include "delegation.h"
  52
  53 #define OPENOWNER_POOL_SIZE     8
  54
  55 const nfs4_stateid zero_stateid;
  56
  57 static DEFINE_SPINLOCK(state_spinlock);
  58 static LIST_HEAD(nfs4_clientid_list);
  59
  60 static void nfs4_recover_state(void *);
  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_recoverd, nfs4_recover_state, clp);
 107         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
 108         INIT_LIST_HEAD(&clp->cl_superblocks);
 109         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
 110         clp->cl_rpcclient = ERR_PTR(-EINVAL);
 111         clp->cl_boot_time = CURRENT_TIME;
 112         clp->cl_state = 0;
 113         return clp;
 114 }
 115
 116 static void
 117 nfs4_free_client(struct nfs4_client *clp)
 118 {
 119         struct nfs4_state_owner *sp;
 120
 121         while (!list_empty(&clp->cl_unused)) {
 122                 sp = list_entry(clp->cl_unused.next,
 123                                 struct nfs4_state_owner,
 124                                 so_list);
 125                 list_del(&sp->so_list);
 126                 kfree(sp);
 127         }
 128         BUG_ON(!list_empty(&clp->cl_state_owners));
 129         if (clp->cl_cred)
 130                 put_rpccred(clp->cl_cred);
 131         nfs_idmap_delete(clp);
 132         if (!IS_ERR(clp->cl_rpcclient))
 133                 rpc_shutdown_client(clp->cl_rpcclient);
 134         kfree(clp);
 135         nfs_callback_down();
 136 }
 137
 138 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 139 {
 140         struct nfs4_client *clp;
 141         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 142                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 143                         atomic_inc(&clp->cl_count);
 144                         return clp;
 145                 }
 146         }
 147         return NULL;
 148 }
 149
 150 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 151 {
 152         struct nfs4_client *clp;
 153         spin_lock(&state_spinlock);
 154         clp = __nfs4_find_client(addr);
 155         spin_unlock(&state_spinlock);
 156         return clp;
 157 }
 158
 159 struct nfs4_client *
 160 nfs4_get_client(struct in_addr *addr)
 161 {
 162         struct nfs4_client *clp, *new = NULL;
 163
 164         spin_lock(&state_spinlock);
 165         for (;;) {
 166                 clp = __nfs4_find_client(addr);
 167                 if (clp != NULL)
 168                         break;
 169                 clp = new;
 170                 if (clp != NULL) {
 171                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 172                         new = NULL;
 173                         break;
 174                 }
 175                 spin_unlock(&state_spinlock);
 176                 new = nfs4_alloc_client(addr);
 177                 spin_lock(&state_spinlock);
 178                 if (new == NULL)
 179                         break;
 180         }
 181         spin_unlock(&state_spinlock);
 182         if (new)
 183                 nfs4_free_client(new);
 184         return clp;
 185 }
 186
 187 void
 188 nfs4_put_client(struct nfs4_client *clp)
 189 {
 190         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 191                 return;
 192         list_del(&clp->cl_servers);
 193         spin_unlock(&state_spinlock);
 194         BUG_ON(!list_empty(&clp->cl_superblocks));
 195         rpc_wake_up(&clp->cl_rpcwaitq);
 196         nfs4_kill_renewd(clp);
 197         nfs4_free_client(clp);
 198 }
 199
 200 static int __nfs4_init_client(struct nfs4_client *clp)
 201 {
 202         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
 203         if (status == 0)
 204                 status = nfs4_proc_setclientid_confirm(clp);
 205         if (status == 0)
 206                 nfs4_schedule_state_renewal(clp);
 207         return status;
 208 }
 209
 210 int nfs4_init_client(struct nfs4_client *clp)
 211 {
 212         return nfs4_map_errors(__nfs4_init_client(clp));
 213 }
 214
 215 u32
 216 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 217 {
 218         return clp->cl_lockowner_id ++;
 219 }
 220
 221 static struct nfs4_state_owner *
 222 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 223 {
 224         struct nfs4_state_owner *sp = NULL;
 225
 226         if (!list_empty(&clp->cl_unused)) {
 227                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 228                 atomic_inc(&sp->so_count);
 229                 sp->so_cred = cred;
 230                 list_move(&sp->so_list, &clp->cl_state_owners);
 231                 clp->cl_nunused--;
 232         }
 233         return sp;
 234 }
 235
 236 static struct nfs4_state_owner *
 237 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 238 {
 239         struct nfs4_state_owner *sp, *res = NULL;
 240
 241         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 242                 if (sp->so_cred != cred)
 243                         continue;
 244                 atomic_inc(&sp->so_count);
 245                 /* Move to the head of the list */
 246                 list_move(&sp->so_list, &clp->cl_state_owners);
 247                 res = sp;
 248                 break;
 249         }
 250         return res;
 251 }
 252
 253 /*
 254  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 255  * create a new state_owner.
 256  *
 257  */
 258 static struct nfs4_state_owner *
 259 nfs4_alloc_state_owner(void)
 260 {
 261         struct nfs4_state_owner *sp;
 262
 263         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 264         if (!sp)
 265                 return NULL;
 266         spin_lock_init(&sp->so_lock);
 267         INIT_LIST_HEAD(&sp->so_states);
 268         INIT_LIST_HEAD(&sp->so_delegations);
 269         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 270         sp->so_seqid.sequence = &sp->so_sequence;
 271         spin_lock_init(&sp->so_sequence.lock);
 272         INIT_LIST_HEAD(&sp->so_sequence.list);
 273         atomic_set(&sp->so_count, 1);
 274         return sp;
 275 }
 276
 277 void
 278 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 279 {
 280         struct nfs4_client *clp = sp->so_client;
 281         spin_lock(&clp->cl_lock);
 282         list_del_init(&sp->so_list);
 283         spin_unlock(&clp->cl_lock);
 284 }
 285
 286 /*
 287  * Note: must be called with clp->cl_sem held in order to prevent races
 288  *       with reboot recovery!
 289  */
 290 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 291 {
 292         struct nfs4_client *clp = server->nfs4_state;
 293         struct nfs4_state_owner *sp, *new;
 294
 295         get_rpccred(cred);
 296         new = nfs4_alloc_state_owner();
 297         spin_lock(&clp->cl_lock);
 298         sp = nfs4_find_state_owner(clp, cred);
 299         if (sp == NULL)
 300                 sp = nfs4_client_grab_unused(clp, cred);
 301         if (sp == NULL && new != NULL) {
 302                 list_add(&new->so_list, &clp->cl_state_owners);
 303                 new->so_client = clp;
 304                 new->so_id = nfs4_alloc_lockowner_id(clp);
 305                 new->so_cred = cred;
 306                 sp = new;
 307                 new = NULL;
 308         }
 309         spin_unlock(&clp->cl_lock);
 310         kfree(new);
 311         if (sp != NULL)
 312                 return sp;
 313         put_rpccred(cred);
 314         return NULL;
 315 }
 316
 317 /*
 318  * Must be called with clp->cl_sem held in order to avoid races
 319  * with state recovery...
 320  */
 321 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 322 {
 323         struct nfs4_client *clp = sp->so_client;
 324         struct rpc_cred *cred = sp->so_cred;
 325
 326         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 327                 return;
 328         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 329                 goto out_free;
 330         if (list_empty(&sp->so_list))
 331                 goto out_free;
 332         list_move(&sp->so_list, &clp->cl_unused);
 333         clp->cl_nunused++;
 334         spin_unlock(&clp->cl_lock);
 335         put_rpccred(cred);
 336         cred = NULL;
 337         return;
 338 out_free:
 339         list_del(&sp->so_list);
 340         spin_unlock(&clp->cl_lock);
 341         put_rpccred(cred);
 342         kfree(sp);
 343 }
 344
 345 static struct nfs4_state *
 346 nfs4_alloc_open_state(void)
 347 {
 348         struct nfs4_state *state;
 349
 350         state = kzalloc(sizeof(*state), GFP_KERNEL);
 351         if (!state)
 352                 return NULL;
 353         atomic_set(&state->count, 1);
 354         INIT_LIST_HEAD(&state->lock_states);
 355         spin_lock_init(&state->state_lock);
 356         return state;
 357 }
 358
 359 void
 360 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 361 {
 362         if (state->state == mode)
 363                 return;
 364         /* NB! List reordering - see the reclaim code for why.  */
 365         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 366                 if (mode & FMODE_WRITE)
 367                         list_move(&state->open_states, &state->owner->so_states);
 368                 else
 369                         list_move_tail(&state->open_states, &state->owner->so_states);
 370         }
 371         if (mode == 0)
 372                 list_del_init(&state->inode_states);
 373         state->state = mode;
 374 }
 375
 376 static struct nfs4_state *
 377 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 378 {
 379         struct nfs_inode *nfsi = NFS_I(inode);
 380         struct nfs4_state *state;
 381
 382         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 383                 /* Is this in the process of being freed? */
 384                 if (state->state == 0)
 385                         continue;
 386                 if (state->owner == owner) {
 387                         atomic_inc(&state->count);
 388                         return state;
 389                 }
 390         }
 391         return NULL;
 392 }
 393
 394 static void
 395 nfs4_free_open_state(struct nfs4_state *state)
 396 {
 397         kfree(state);
 398 }
 399
 400 struct nfs4_state *
 401 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 402 {
 403         struct nfs4_state *state, *new;
 404         struct nfs_inode *nfsi = NFS_I(inode);
 405
 406         spin_lock(&inode->i_lock);
 407         state = __nfs4_find_state_byowner(inode, owner);
 408         spin_unlock(&inode->i_lock);
 409         if (state)
 410                 goto out;
 411         new = nfs4_alloc_open_state();
 412         spin_lock(&owner->so_lock);
 413         spin_lock(&inode->i_lock);
 414         state = __nfs4_find_state_byowner(inode, owner);
 415         if (state == NULL && new != NULL) {
 416                 state = new;
 417                 state->owner = owner;
 418                 atomic_inc(&owner->so_count);
 419                 list_add(&state->inode_states, &nfsi->open_states);
 420                 state->inode = igrab(inode);
 421                 spin_unlock(&inode->i_lock);
 422                 /* Note: The reclaim code dictates that we add stateless
 423                  * and read-only stateids to the end of the list */
 424                 list_add_tail(&state->open_states, &owner->so_states);
 425                 spin_unlock(&owner->so_lock);
 426         } else {
 427                 spin_unlock(&inode->i_lock);
 428                 spin_unlock(&owner->so_lock);
 429                 if (new)
 430                         nfs4_free_open_state(new);
 431         }
 432 out:
 433         return state;
 434 }
 435
 436 /*
 437  * Beware! Caller must be holding exactly one
 438  * reference to clp->cl_sem!
 439  */
 440 void nfs4_put_open_state(struct nfs4_state *state)
 441 {
 442         struct inode *inode = state->inode;
 443         struct nfs4_state_owner *owner = state->owner;
 444
 445         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 446                 return;
 447         spin_lock(&inode->i_lock);
 448         if (!list_empty(&state->inode_states))
 449                 list_del(&state->inode_states);
 450         list_del(&state->open_states);
 451         spin_unlock(&inode->i_lock);
 452         spin_unlock(&owner->so_lock);
 453         iput(inode);
 454         nfs4_free_open_state(state);
 455         nfs4_put_state_owner(owner);
 456 }
 457
 458 /*
 459  * Close the current file.
 460  */
 461 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 462 {
 463         struct inode *inode = state->inode;
 464         struct nfs4_state_owner *owner = state->owner;
 465         int oldstate, newstate = 0;
 466
 467         atomic_inc(&owner->so_count);
 468         /* Protect against nfs4_find_state() */
 469         spin_lock(&owner->so_lock);
 470         spin_lock(&inode->i_lock);
 471         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 472                 case FMODE_READ:
 473                         state->n_rdonly--;
 474                         break;
 475                 case FMODE_WRITE:
 476                         state->n_wronly--;
 477                         break;
 478                 case FMODE_READ|FMODE_WRITE:
 479                         state->n_rdwr--;
 480         }
 481         oldstate = newstate = state->state;
 482         if (state->n_rdwr == 0) {
 483                 if (state->n_rdonly == 0)
 484                         newstate &= ~FMODE_READ;
 485                 if (state->n_wronly == 0)
 486                         newstate &= ~FMODE_WRITE;
 487         }
 488         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 489                 nfs4_state_set_mode_locked(state, newstate);
 490                 oldstate = newstate;
 491         }
 492         spin_unlock(&inode->i_lock);
 493         spin_unlock(&owner->so_lock);
 494
 495         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
 496                 return;
 497         nfs4_put_open_state(state);
 498         nfs4_put_state_owner(owner);
 499 }
 500
 501 /*
 502  * Search the state->lock_states for an existing lock_owner
 503  * that is compatible with current->files
 504  */
 505 static struct nfs4_lock_state *
 506 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 507 {
 508         struct nfs4_lock_state *pos;
 509         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 510                 if (pos->ls_owner != fl_owner)
 511                         continue;
 512                 atomic_inc(&pos->ls_count);
 513                 return pos;
 514         }
 515         return NULL;
 516 }
 517
 518 /*
 519  * Return a compatible lock_state. If no initialized lock_state structure
 520  * exists, return an uninitialized one.
 521  *
 522  */
 523 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 524 {
 525         struct nfs4_lock_state *lsp;
 526         struct nfs4_client *clp = state->owner->so_client;
 527
 528         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 529         if (lsp == NULL)
 530                 return NULL;
 531         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 532         atomic_set(&lsp->ls_count, 1);
 533         lsp->ls_owner = fl_owner;
 534         spin_lock(&clp->cl_lock);
 535         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 536         spin_unlock(&clp->cl_lock);
 537         INIT_LIST_HEAD(&lsp->ls_locks);
 538         return lsp;
 539 }
 540
 541 /*
 542  * Return a compatible lock_state. If no initialized lock_state structure
 543  * exists, return an uninitialized one.
 544  *
 545  * The caller must be holding clp->cl_sem
 546  */
 547 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 548 {
 549         struct nfs4_lock_state *lsp, *new = NULL;
 550
 551         for(;;) {
 552                 spin_lock(&state->state_lock);
 553                 lsp = __nfs4_find_lock_state(state, owner);
 554                 if (lsp != NULL)
 555                         break;
 556                 if (new != NULL) {
 557                         new->ls_state = state;
 558                         list_add(&new->ls_locks, &state->lock_states);
 559                         set_bit(LK_STATE_IN_USE, &state->flags);
 560                         lsp = new;
 561                         new = NULL;
 562                         break;
 563                 }
 564                 spin_unlock(&state->state_lock);
 565                 new = nfs4_alloc_lock_state(state, owner);
 566                 if (new == NULL)
 567                         return NULL;
 568         }
 569         spin_unlock(&state->state_lock);
 570         kfree(new);
 571         return lsp;
 572 }
 573
 574 /*
 575  * Release reference to lock_state, and free it if we see that
 576  * it is no longer in use
 577  */
 578 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 579 {
 580         struct nfs4_state *state;
 581
 582         if (lsp == NULL)
 583                 return;
 584         state = lsp->ls_state;
 585         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 586                 return;
 587         list_del(&lsp->ls_locks);
 588         if (list_empty(&state->lock_states))
 589                 clear_bit(LK_STATE_IN_USE, &state->flags);
 590         spin_unlock(&state->state_lock);
 591         kfree(lsp);
 592 }
 593
 594 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 595 {
 596         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 597
 598         dst->fl_u.nfs4_fl.owner = lsp;
 599         atomic_inc(&lsp->ls_count);
 600 }
 601
 602 static void nfs4_fl_release_lock(struct file_lock *fl)
 603 {
 604         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 605 }
 606
 607 static struct file_lock_operations nfs4_fl_lock_ops = {
 608         .fl_copy_lock = nfs4_fl_copy_lock,
 609         .fl_release_private = nfs4_fl_release_lock,
 610 };
 611
 612 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 613 {
 614         struct nfs4_lock_state *lsp;
 615
 616         if (fl->fl_ops != NULL)
 617                 return 0;
 618         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 619         if (lsp == NULL)
 620                 return -ENOMEM;
 621         fl->fl_u.nfs4_fl.owner = lsp;
 622         fl->fl_ops = &nfs4_fl_lock_ops;
 623         return 0;
 624 }
 625
 626 /*
 627  * Byte-range lock aware utility to initialize the stateid of read/write
 628  * requests.
 629  */
 630 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 631 {
 632         struct nfs4_lock_state *lsp;
 633
 634         memcpy(dst, &state->stateid, sizeof(*dst));
 635         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 636                 return;
 637
 638         spin_lock(&state->state_lock);
 639         lsp = __nfs4_find_lock_state(state, fl_owner);
 640         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 641                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 642         spin_unlock(&state->state_lock);
 643         nfs4_put_lock_state(lsp);
 644 }
 645
 646 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 647 {
 648         struct rpc_sequence *sequence = counter->sequence;
 649         struct nfs_seqid *new;
 650
 651         new = kmalloc(sizeof(*new), GFP_KERNEL);
 652         if (new != NULL) {
 653                 new->sequence = counter;
 654                 spin_lock(&sequence->lock);
 655                 list_add_tail(&new->list, &sequence->list);
 656                 spin_unlock(&sequence->lock);
 657         }
 658         return new;
 659 }
 660
 661 void nfs_free_seqid(struct nfs_seqid *seqid)
 662 {
 663         struct rpc_sequence *sequence = seqid->sequence->sequence;
 664
 665         spin_lock(&sequence->lock);
 666         list_del(&seqid->list);
 667         spin_unlock(&sequence->lock);
 668         rpc_wake_up(&sequence->wait);
 669         kfree(seqid);
 670 }
 671
 672 /*
 673  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 674  * failed with a seqid incrementing error -
 675  * see comments nfs_fs.h:seqid_mutating_error()
 676  */
 677 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 678 {
 679         switch (status) {
 680                 case 0:
 681                         break;
 682                 case -NFS4ERR_BAD_SEQID:
 683                 case -NFS4ERR_STALE_CLIENTID:
 684                 case -NFS4ERR_STALE_STATEID:
 685                 case -NFS4ERR_BAD_STATEID:
 686                 case -NFS4ERR_BADXDR:
 687                 case -NFS4ERR_RESOURCE:
 688                 case -NFS4ERR_NOFILEHANDLE:
 689                         /* Non-seqid mutating errors */
 690                         return;
 691         };
 692         /*
 693          * Note: no locking needed as we are guaranteed to be first
 694          * on the sequence list
 695          */
 696         seqid->sequence->counter++;
 697 }
 698
 699 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 700 {
 701         if (status == -NFS4ERR_BAD_SEQID) {
 702                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 703                                 struct nfs4_state_owner, so_seqid);
 704                 nfs4_drop_state_owner(sp);
 705         }
 706         return nfs_increment_seqid(status, seqid);
 707 }
 708
 709 /*
 710  * Increment the seqid if the LOCK/LOCKU succeeded, or
 711  * failed with a seqid incrementing error -
 712  * see comments nfs_fs.h:seqid_mutating_error()
 713  */
 714 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 715 {
 716         return nfs_increment_seqid(status, seqid);
 717 }
 718
 719 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 720 {
 721         struct rpc_sequence *sequence = seqid->sequence->sequence;
 722         int status = 0;
 723
 724         if (sequence->list.next == &seqid->list)
 725                 goto out;
 726         spin_lock(&sequence->lock);
 727         if (sequence->list.next != &seqid->list) {
 728                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 729                 status = -EAGAIN;
 730         }
 731         spin_unlock(&sequence->lock);
 732 out:
 733         return status;
 734 }
 735
 736 static int reclaimer(void *);
 737 struct reclaimer_args {
 738         struct nfs4_client *clp;
 739         struct completion complete;
 740 };
 741
 742 static inline void nfs4_clear_recover_bit(struct nfs4_client *clp)
 743 {
 744         smp_mb__before_clear_bit();
 745         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 746         smp_mb__after_clear_bit();
 747         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 748         rpc_wake_up(&clp->cl_rpcwaitq);
 749 }
 750
 751 /*
 752  * State recovery routine
 753  */
 754 void
 755 nfs4_recover_state(void *data)
 756 {
 757         struct nfs4_client *clp = (struct nfs4_client *)data;
 758         struct reclaimer_args args = {
 759                 .clp = clp,
 760         };
 761         might_sleep();
 762
 763         init_completion(&args.complete);
 764
 765         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
 766                 goto out_failed_clear;
 767         wait_for_completion(&args.complete);
 768         return;
 769 out_failed_clear:
 770         nfs4_clear_recover_bit(clp);
 771 }
 772
 773 /*
 774  * Schedule a state recovery attempt
 775  */
 776 void
 777 nfs4_schedule_state_recovery(struct nfs4_client *clp)
 778 {
 779         if (!clp)
 780                 return;
 781         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 782                 schedule_work(&clp->cl_recoverd);
 783 }
 784
 785 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 786 {
 787         struct inode *inode = state->inode;
 788         struct file_lock *fl;
 789         int status = 0;
 790
 791         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 792                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 793                         continue;
 794                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 795                         continue;
 796                 status = ops->recover_lock(state, fl);
 797                 if (status >= 0)
 798                         continue;
 799                 switch (status) {
 800                         default:
 801                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 802                                                 __FUNCTION__, status);
 803                         case -NFS4ERR_EXPIRED:
 804                         case -NFS4ERR_NO_GRACE:
 805                         case -NFS4ERR_RECLAIM_BAD:
 806                         case -NFS4ERR_RECLAIM_CONFLICT:
 807                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 808                                 break;
 809                         case -NFS4ERR_STALE_CLIENTID:
 810                                 goto out_err;
 811                 }
 812         }
 813         return 0;
 814 out_err:
 815         return status;
 816 }
 817
 818 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 819 {
 820         struct nfs4_state *state;
 821         struct nfs4_lock_state *lock;
 822         int status = 0;
 823
 824         /* Note: we rely on the sp->so_states list being ordered
 825          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 826          * states first.
 827          * This is needed to ensure that the server won't give us any
 828          * read delegations that we have to return if, say, we are
 829          * recovering after a network partition or a reboot from a
 830          * server that doesn't support a grace period.
 831          */
 832         list_for_each_entry(state, &sp->so_states, open_states) {
 833                 if (state->state == 0)
 834                         continue;
 835                 status = ops->recover_open(sp, state);
 836                 if (status >= 0) {
 837                         status = nfs4_reclaim_locks(ops, state);
 838                         if (status < 0)
 839                                 goto out_err;
 840                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 841                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 842                                         printk("%s: Lock reclaim failed!\n",
 843                                                         __FUNCTION__);
 844                         }
 845                         continue;
 846                 }
 847                 switch (status) {
 848                         default:
 849                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 850                                                 __FUNCTION__, status);
 851                         case -ENOENT:
 852                         case -NFS4ERR_RECLAIM_BAD:
 853                         case -NFS4ERR_RECLAIM_CONFLICT:
 854                                 /*
 855                                  * Open state on this file cannot be recovered
 856                                  * All we can do is revert to using the zero stateid.
 857                                  */
 858                                 memset(state->stateid.data, 0,
 859                                         sizeof(state->stateid.data));
 860                                 /* Mark the file as being 'closed' */
 861                                 state->state = 0;
 862                                 break;
 863                         case -NFS4ERR_EXPIRED:
 864                         case -NFS4ERR_NO_GRACE:
 865                         case -NFS4ERR_STALE_CLIENTID:
 866                                 goto out_err;
 867                 }
 868         }
 869         return 0;
 870 out_err:
 871         return status;
 872 }
 873
 874 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 875 {
 876         struct nfs4_state_owner *sp;
 877         struct nfs4_state *state;
 878         struct nfs4_lock_state *lock;
 879
 880         /* Reset all sequence ids to zero */
 881         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 882                 sp->so_seqid.counter = 0;
 883                 sp->so_seqid.flags = 0;
 884                 spin_lock(&sp->so_lock);
 885                 list_for_each_entry(state, &sp->so_states, open_states) {
 886                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 887                                 lock->ls_seqid.counter = 0;
 888                                 lock->ls_seqid.flags = 0;
 889                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 890                         }
 891                 }
 892                 spin_unlock(&sp->so_lock);
 893         }
 894 }
 895
 896 static int reclaimer(void *ptr)
 897 {
 898         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
 899         struct nfs4_client *clp = args->clp;
 900         struct nfs4_state_owner *sp;
 901         struct nfs4_state_recovery_ops *ops;
 902         int status = 0;
 903
 904         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
 905         allow_signal(SIGKILL);
 906
 907         atomic_inc(&clp->cl_count);
 908         complete(&args->complete);
 909
 910         /* Ensure exclusive access to NFSv4 state */
 911         lock_kernel();
 912         down_write(&clp->cl_sem);
 913         /* Are there any NFS mounts out there? */
 914         if (list_empty(&clp->cl_superblocks))
 915                 goto out;
 916 restart_loop:
 917         status = nfs4_proc_renew(clp);
 918         switch (status) {
 919                 case 0:
 920                 case -NFS4ERR_CB_PATH_DOWN:
 921                         goto out;
 922                 case -NFS4ERR_STALE_CLIENTID:
 923                 case -NFS4ERR_LEASE_MOVED:
 924                         ops = &nfs4_reboot_recovery_ops;
 925                         break;
 926                 default:
 927                         ops = &nfs4_network_partition_recovery_ops;
 928         };
 929         nfs4_state_mark_reclaim(clp);
 930         status = __nfs4_init_client(clp);
 931         if (status)
 932                 goto out_error;
 933         /* Mark all delegations for reclaim */
 934         nfs_delegation_mark_reclaim(clp);
 935         /* Note: list is protected by exclusive lock on cl->cl_sem */
 936         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 937                 status = nfs4_reclaim_open_state(ops, sp);
 938                 if (status < 0) {
 939                         if (status == -NFS4ERR_NO_GRACE) {
 940                                 ops = &nfs4_network_partition_recovery_ops;
 941                                 status = nfs4_reclaim_open_state(ops, sp);
 942                         }
 943                         if (status == -NFS4ERR_STALE_CLIENTID)
 944                                 goto restart_loop;
 945                         if (status == -NFS4ERR_EXPIRED)
 946                                 goto restart_loop;
 947                 }
 948         }
 949         nfs_delegation_reap_unclaimed(clp);
 950 out:
 951         up_write(&clp->cl_sem);
 952         unlock_kernel();
 953         if (status == -NFS4ERR_CB_PATH_DOWN)
 954                 nfs_handle_cb_pathdown(clp);
 955         nfs4_clear_recover_bit(clp);
 956         nfs4_put_client(clp);
 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  */