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         if (server->mnt_path) {
  73                 kfree(server->mnt_path);
  74                 server->mnt_path = NULL;
  75         }
  76         if (server->nfs4_state) {
  77                 nfs4_put_client(server->nfs4_state);
  78                 server->nfs4_state = NULL;
  79         }
  80 }
  81
  82 /*
  83  * nfs4_get_client(): returns an empty client structure
  84  * nfs4_put_client(): drops reference to client structure
  85  *
  86  * Since these are allocated/deallocated very rarely, we don't
  87  * bother putting them in a slab cache...
  88  */
  89 static struct nfs4_client *
  90 nfs4_alloc_client(struct in_addr *addr)
  91 {
  92         struct nfs4_client *clp;
  93
  94         if (nfs_callback_up() < 0)
  95                 return NULL;
  96         if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
  97                 nfs_callback_down();
  98                 return NULL;
  99         }
 100         memset(clp, 0, sizeof(*clp));
 101         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
 102         init_rwsem(&clp->cl_sem);
 103         INIT_LIST_HEAD(&clp->cl_delegations);
 104         INIT_LIST_HEAD(&clp->cl_state_owners);
 105         INIT_LIST_HEAD(&clp->cl_unused);
 106         spin_lock_init(&clp->cl_lock);
 107         atomic_set(&clp->cl_count, 1);
 108         INIT_WORK(&clp->cl_recoverd, nfs4_recover_state, clp);
 109         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
 110         INIT_LIST_HEAD(&clp->cl_superblocks);
 111         init_waitqueue_head(&clp->cl_waitq);
 112         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
 113         clp->cl_rpcclient = ERR_PTR(-EINVAL);
 114         clp->cl_boot_time = CURRENT_TIME;
 115         clp->cl_state = 1 << NFS4CLNT_OK;
 116         return clp;
 117 }
 118
 119 static void
 120 nfs4_free_client(struct nfs4_client *clp)
 121 {
 122         struct nfs4_state_owner *sp;
 123
 124         while (!list_empty(&clp->cl_unused)) {
 125                 sp = list_entry(clp->cl_unused.next,
 126                                 struct nfs4_state_owner,
 127                                 so_list);
 128                 list_del(&sp->so_list);
 129                 kfree(sp);
 130         }
 131         BUG_ON(!list_empty(&clp->cl_state_owners));
 132         if (clp->cl_cred)
 133                 put_rpccred(clp->cl_cred);
 134         nfs_idmap_delete(clp);
 135         if (!IS_ERR(clp->cl_rpcclient))
 136                 rpc_shutdown_client(clp->cl_rpcclient);
 137         kfree(clp);
 138         nfs_callback_down();
 139 }
 140
 141 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 142 {
 143         struct nfs4_client *clp;
 144         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 145                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 146                         atomic_inc(&clp->cl_count);
 147                         return clp;
 148                 }
 149         }
 150         return NULL;
 151 }
 152
 153 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 154 {
 155         struct nfs4_client *clp;
 156         spin_lock(&state_spinlock);
 157         clp = __nfs4_find_client(addr);
 158         spin_unlock(&state_spinlock);
 159         return clp;
 160 }
 161
 162 struct nfs4_client *
 163 nfs4_get_client(struct in_addr *addr)
 164 {
 165         struct nfs4_client *clp, *new = NULL;
 166
 167         spin_lock(&state_spinlock);
 168         for (;;) {
 169                 clp = __nfs4_find_client(addr);
 170                 if (clp != NULL)
 171                         break;
 172                 clp = new;
 173                 if (clp != NULL) {
 174                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 175                         new = NULL;
 176                         break;
 177                 }
 178                 spin_unlock(&state_spinlock);
 179                 new = nfs4_alloc_client(addr);
 180                 spin_lock(&state_spinlock);
 181                 if (new == NULL)
 182                         break;
 183         }
 184         spin_unlock(&state_spinlock);
 185         if (new)
 186                 nfs4_free_client(new);
 187         return clp;
 188 }
 189
 190 void
 191 nfs4_put_client(struct nfs4_client *clp)
 192 {
 193         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 194                 return;
 195         list_del(&clp->cl_servers);
 196         spin_unlock(&state_spinlock);
 197         BUG_ON(!list_empty(&clp->cl_superblocks));
 198         wake_up_all(&clp->cl_waitq);
 199         rpc_wake_up(&clp->cl_rpcwaitq);
 200         nfs4_kill_renewd(clp);
 201         nfs4_free_client(clp);
 202 }
 203
 204 static int __nfs4_init_client(struct nfs4_client *clp)
 205 {
 206         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
 207         if (status == 0)
 208                 status = nfs4_proc_setclientid_confirm(clp);
 209         if (status == 0)
 210                 nfs4_schedule_state_renewal(clp);
 211         return status;
 212 }
 213
 214 int nfs4_init_client(struct nfs4_client *clp)
 215 {
 216         return nfs4_map_errors(__nfs4_init_client(clp));
 217 }
 218
 219 u32
 220 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 221 {
 222         return clp->cl_lockowner_id ++;
 223 }
 224
 225 static struct nfs4_state_owner *
 226 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 227 {
 228         struct nfs4_state_owner *sp = NULL;
 229
 230         if (!list_empty(&clp->cl_unused)) {
 231                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 232                 atomic_inc(&sp->so_count);
 233                 sp->so_cred = cred;
 234                 list_move(&sp->so_list, &clp->cl_state_owners);
 235                 clp->cl_nunused--;
 236         }
 237         return sp;
 238 }
 239
 240 static struct nfs4_state_owner *
 241 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 242 {
 243         struct nfs4_state_owner *sp, *res = NULL;
 244
 245         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 246                 if (sp->so_cred != cred)
 247                         continue;
 248                 atomic_inc(&sp->so_count);
 249                 /* Move to the head of the list */
 250                 list_move(&sp->so_list, &clp->cl_state_owners);
 251                 res = sp;
 252                 break;
 253         }
 254         return res;
 255 }
 256
 257 /*
 258  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 259  * create a new state_owner.
 260  *
 261  */
 262 static struct nfs4_state_owner *
 263 nfs4_alloc_state_owner(void)
 264 {
 265         struct nfs4_state_owner *sp;
 266
 267         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 268         if (!sp)
 269                 return NULL;
 270         INIT_LIST_HEAD(&sp->so_states);
 271         INIT_LIST_HEAD(&sp->so_delegations);
 272         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 273         sp->so_seqid.sequence = &sp->so_sequence;
 274         spin_lock_init(&sp->so_sequence.lock);
 275         INIT_LIST_HEAD(&sp->so_sequence.list);
 276         atomic_set(&sp->so_count, 1);
 277         return sp;
 278 }
 279
 280 void
 281 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 282 {
 283         struct nfs4_client *clp = sp->so_client;
 284         spin_lock(&clp->cl_lock);
 285         list_del_init(&sp->so_list);
 286         spin_unlock(&clp->cl_lock);
 287 }
 288
 289 /*
 290  * Note: must be called with clp->cl_sem held in order to prevent races
 291  *       with reboot recovery!
 292  */
 293 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 294 {
 295         struct nfs4_client *clp = server->nfs4_state;
 296         struct nfs4_state_owner *sp, *new;
 297
 298         get_rpccred(cred);
 299         new = nfs4_alloc_state_owner();
 300         spin_lock(&clp->cl_lock);
 301         sp = nfs4_find_state_owner(clp, cred);
 302         if (sp == NULL)
 303                 sp = nfs4_client_grab_unused(clp, cred);
 304         if (sp == NULL && new != NULL) {
 305                 list_add(&new->so_list, &clp->cl_state_owners);
 306                 new->so_client = clp;
 307                 new->so_id = nfs4_alloc_lockowner_id(clp);
 308                 new->so_cred = cred;
 309                 sp = new;
 310                 new = NULL;
 311         }
 312         spin_unlock(&clp->cl_lock);
 313         if (new)
 314                 kfree(new);
 315         if (sp != NULL)
 316                 return sp;
 317         put_rpccred(cred);
 318         return NULL;
 319 }
 320
 321 /*
 322  * Must be called with clp->cl_sem held in order to avoid races
 323  * with state recovery...
 324  */
 325 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 326 {
 327         struct nfs4_client *clp = sp->so_client;
 328         struct rpc_cred *cred = sp->so_cred;
 329
 330         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 331                 return;
 332         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 333                 goto out_free;
 334         if (list_empty(&sp->so_list))
 335                 goto out_free;
 336         list_move(&sp->so_list, &clp->cl_unused);
 337         clp->cl_nunused++;
 338         spin_unlock(&clp->cl_lock);
 339         put_rpccred(cred);
 340         cred = NULL;
 341         return;
 342 out_free:
 343         list_del(&sp->so_list);
 344         spin_unlock(&clp->cl_lock);
 345         put_rpccred(cred);
 346         kfree(sp);
 347 }
 348
 349 static struct nfs4_state *
 350 nfs4_alloc_open_state(void)
 351 {
 352         struct nfs4_state *state;
 353
 354         state = kmalloc(sizeof(*state), GFP_KERNEL);
 355         if (!state)
 356                 return NULL;
 357         state->state = 0;
 358         state->nreaders = 0;
 359         state->nwriters = 0;
 360         state->flags = 0;
 361         memset(state->stateid.data, 0, sizeof(state->stateid.data));
 362         atomic_set(&state->count, 1);
 363         INIT_LIST_HEAD(&state->lock_states);
 364         spin_lock_init(&state->state_lock);
 365         return state;
 366 }
 367
 368 static struct nfs4_state *
 369 __nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 370 {
 371         struct nfs_inode *nfsi = NFS_I(inode);
 372         struct nfs4_state *state;
 373
 374         mode &= (FMODE_READ|FMODE_WRITE);
 375         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 376                 if (state->owner->so_cred != cred)
 377                         continue;
 378                 if ((mode & FMODE_READ) != 0 && state->nreaders == 0)
 379                         continue;
 380                 if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0)
 381                         continue;
 382                 if ((state->state & mode) != mode)
 383                         continue;
 384                 atomic_inc(&state->count);
 385                 if (mode & FMODE_READ)
 386                         state->nreaders++;
 387                 if (mode & FMODE_WRITE)
 388                         state->nwriters++;
 389                 return state;
 390         }
 391         return NULL;
 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->nreaders == 0 && state->nwriters == 0)
 403                         continue;
 404                 if (state->owner == owner) {
 405                         atomic_inc(&state->count);
 406                         return state;
 407                 }
 408         }
 409         return NULL;
 410 }
 411
 412 struct nfs4_state *
 413 nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 414 {
 415         struct nfs4_state *state;
 416
 417         spin_lock(&inode->i_lock);
 418         state = __nfs4_find_state(inode, cred, mode);
 419         spin_unlock(&inode->i_lock);
 420         return state;
 421 }
 422
 423 static void
 424 nfs4_free_open_state(struct nfs4_state *state)
 425 {
 426         kfree(state);
 427 }
 428
 429 struct nfs4_state *
 430 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 431 {
 432         struct nfs4_state *state, *new;
 433         struct nfs_inode *nfsi = NFS_I(inode);
 434
 435         spin_lock(&inode->i_lock);
 436         state = __nfs4_find_state_byowner(inode, owner);
 437         spin_unlock(&inode->i_lock);
 438         if (state)
 439                 goto out;
 440         new = nfs4_alloc_open_state();
 441         spin_lock(&inode->i_lock);
 442         state = __nfs4_find_state_byowner(inode, owner);
 443         if (state == NULL && new != NULL) {
 444                 state = new;
 445                 /* Note: The reclaim code dictates that we add stateless
 446                  * and read-only stateids to the end of the list */
 447                 list_add_tail(&state->open_states, &owner->so_states);
 448                 state->owner = owner;
 449                 atomic_inc(&owner->so_count);
 450                 list_add(&state->inode_states, &nfsi->open_states);
 451                 state->inode = igrab(inode);
 452                 spin_unlock(&inode->i_lock);
 453         } else {
 454                 spin_unlock(&inode->i_lock);
 455                 if (new)
 456                         nfs4_free_open_state(new);
 457         }
 458 out:
 459         return state;
 460 }
 461
 462 /*
 463  * Beware! Caller must be holding exactly one
 464  * reference to clp->cl_sem!
 465  */
 466 void nfs4_put_open_state(struct nfs4_state *state)
 467 {
 468         struct inode *inode = state->inode;
 469         struct nfs4_state_owner *owner = state->owner;
 470
 471         if (!atomic_dec_and_lock(&state->count, &inode->i_lock))
 472                 return;
 473         if (!list_empty(&state->inode_states))
 474                 list_del(&state->inode_states);
 475         spin_unlock(&inode->i_lock);
 476         list_del(&state->open_states);
 477         iput(inode);
 478         BUG_ON (state->state != 0);
 479         nfs4_free_open_state(state);
 480         nfs4_put_state_owner(owner);
 481 }
 482
 483 /*
 484  * Close the current file.
 485  */
 486 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 487 {
 488         struct inode *inode = state->inode;
 489         struct nfs4_state_owner *owner = state->owner;
 490         int newstate;
 491
 492         atomic_inc(&owner->so_count);
 493         /* Protect against nfs4_find_state() */
 494         spin_lock(&inode->i_lock);
 495         if (mode & FMODE_READ)
 496                 state->nreaders--;
 497         if (mode & FMODE_WRITE)
 498                 state->nwriters--;
 499         if (state->nwriters == 0) {
 500                 if (state->nreaders == 0)
 501                         list_del_init(&state->inode_states);
 502                 /* See reclaim code */
 503                 list_move_tail(&state->open_states, &owner->so_states);
 504         }
 505         spin_unlock(&inode->i_lock);
 506         newstate = 0;
 507         if (state->state != 0) {
 508                 if (state->nreaders)
 509                         newstate |= FMODE_READ;
 510                 if (state->nwriters)
 511                         newstate |= FMODE_WRITE;
 512                 if (state->state == newstate)
 513                         goto out;
 514                 if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 515                         state->state = newstate;
 516                         goto out;
 517                 }
 518                 if (nfs4_do_close(inode, state, newstate) == 0)
 519                         return;
 520         }
 521 out:
 522         nfs4_put_open_state(state);
 523         nfs4_put_state_owner(owner);
 524 }
 525
 526 /*
 527  * Search the state->lock_states for an existing lock_owner
 528  * that is compatible with current->files
 529  */
 530 static struct nfs4_lock_state *
 531 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 532 {
 533         struct nfs4_lock_state *pos;
 534         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 535                 if (pos->ls_owner != fl_owner)
 536                         continue;
 537                 atomic_inc(&pos->ls_count);
 538                 return pos;
 539         }
 540         return NULL;
 541 }
 542
 543 /*
 544  * Return a compatible lock_state. If no initialized lock_state structure
 545  * exists, return an uninitialized one.
 546  *
 547  */
 548 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 549 {
 550         struct nfs4_lock_state *lsp;
 551         struct nfs4_client *clp = state->owner->so_client;
 552
 553         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 554         if (lsp == NULL)
 555                 return NULL;
 556         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 557         atomic_set(&lsp->ls_count, 1);
 558         lsp->ls_owner = fl_owner;
 559         spin_lock(&clp->cl_lock);
 560         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 561         spin_unlock(&clp->cl_lock);
 562         INIT_LIST_HEAD(&lsp->ls_locks);
 563         return lsp;
 564 }
 565
 566 /*
 567  * Return a compatible lock_state. If no initialized lock_state structure
 568  * exists, return an uninitialized one.
 569  *
 570  * The caller must be holding clp->cl_sem
 571  */
 572 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 573 {
 574         struct nfs4_lock_state *lsp, *new = NULL;
 575
 576         for(;;) {
 577                 spin_lock(&state->state_lock);
 578                 lsp = __nfs4_find_lock_state(state, owner);
 579                 if (lsp != NULL)
 580                         break;
 581                 if (new != NULL) {
 582                         new->ls_state = state;
 583                         list_add(&new->ls_locks, &state->lock_states);
 584                         set_bit(LK_STATE_IN_USE, &state->flags);
 585                         lsp = new;
 586                         new = NULL;
 587                         break;
 588                 }
 589                 spin_unlock(&state->state_lock);
 590                 new = nfs4_alloc_lock_state(state, owner);
 591                 if (new == NULL)
 592                         return NULL;
 593         }
 594         spin_unlock(&state->state_lock);
 595         kfree(new);
 596         return lsp;
 597 }
 598
 599 /*
 600  * Release reference to lock_state, and free it if we see that
 601  * it is no longer in use
 602  */
 603 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 604 {
 605         struct nfs4_state *state;
 606
 607         if (lsp == NULL)
 608                 return;
 609         state = lsp->ls_state;
 610         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 611                 return;
 612         list_del(&lsp->ls_locks);
 613         if (list_empty(&state->lock_states))
 614                 clear_bit(LK_STATE_IN_USE, &state->flags);
 615         spin_unlock(&state->state_lock);
 616         kfree(lsp);
 617 }
 618
 619 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 620 {
 621         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 622
 623         dst->fl_u.nfs4_fl.owner = lsp;
 624         atomic_inc(&lsp->ls_count);
 625 }
 626
 627 static void nfs4_fl_release_lock(struct file_lock *fl)
 628 {
 629         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 630 }
 631
 632 static struct file_lock_operations nfs4_fl_lock_ops = {
 633         .fl_copy_lock = nfs4_fl_copy_lock,
 634         .fl_release_private = nfs4_fl_release_lock,
 635 };
 636
 637 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 638 {
 639         struct nfs4_lock_state *lsp;
 640
 641         if (fl->fl_ops != NULL)
 642                 return 0;
 643         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 644         if (lsp == NULL)
 645                 return -ENOMEM;
 646         fl->fl_u.nfs4_fl.owner = lsp;
 647         fl->fl_ops = &nfs4_fl_lock_ops;
 648         return 0;
 649 }
 650
 651 /*
 652  * Byte-range lock aware utility to initialize the stateid of read/write
 653  * requests.
 654  */
 655 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 656 {
 657         struct nfs4_lock_state *lsp;
 658
 659         memcpy(dst, &state->stateid, sizeof(*dst));
 660         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 661                 return;
 662
 663         spin_lock(&state->state_lock);
 664         lsp = __nfs4_find_lock_state(state, fl_owner);
 665         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 666                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 667         spin_unlock(&state->state_lock);
 668         nfs4_put_lock_state(lsp);
 669 }
 670
 671 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 672 {
 673         struct rpc_sequence *sequence = counter->sequence;
 674         struct nfs_seqid *new;
 675
 676         new = kmalloc(sizeof(*new), GFP_KERNEL);
 677         if (new != NULL) {
 678                 new->sequence = counter;
 679                 spin_lock(&sequence->lock);
 680                 list_add_tail(&new->list, &sequence->list);
 681                 spin_unlock(&sequence->lock);
 682         }
 683         return new;
 684 }
 685
 686 void nfs_free_seqid(struct nfs_seqid *seqid)
 687 {
 688         struct rpc_sequence *sequence = seqid->sequence->sequence;
 689
 690         spin_lock(&sequence->lock);
 691         list_del(&seqid->list);
 692         rpc_wake_up(&sequence->wait);
 693         spin_unlock(&sequence->lock);
 694         kfree(seqid);
 695 }
 696
 697 /*
 698  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 699  * failed with a seqid incrementing error -
 700  * see comments nfs_fs.h:seqid_mutating_error()
 701  */
 702 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 703 {
 704         switch (status) {
 705                 case 0:
 706                         break;
 707                 case -NFS4ERR_BAD_SEQID:
 708                 case -NFS4ERR_STALE_CLIENTID:
 709                 case -NFS4ERR_STALE_STATEID:
 710                 case -NFS4ERR_BAD_STATEID:
 711                 case -NFS4ERR_BADXDR:
 712                 case -NFS4ERR_RESOURCE:
 713                 case -NFS4ERR_NOFILEHANDLE:
 714                         /* Non-seqid mutating errors */
 715                         return;
 716         };
 717         /*
 718          * Note: no locking needed as we are guaranteed to be first
 719          * on the sequence list
 720          */
 721         seqid->sequence->counter++;
 722 }
 723
 724 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 725 {
 726         if (status == -NFS4ERR_BAD_SEQID) {
 727                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 728                                 struct nfs4_state_owner, so_seqid);
 729                 nfs4_drop_state_owner(sp);
 730         }
 731         return nfs_increment_seqid(status, seqid);
 732 }
 733
 734 /*
 735  * Increment the seqid if the LOCK/LOCKU succeeded, or
 736  * failed with a seqid incrementing error -
 737  * see comments nfs_fs.h:seqid_mutating_error()
 738  */
 739 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 740 {
 741         return nfs_increment_seqid(status, seqid);
 742 }
 743
 744 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 745 {
 746         struct rpc_sequence *sequence = seqid->sequence->sequence;
 747         int status = 0;
 748
 749         spin_lock(&sequence->lock);
 750         if (sequence->list.next != &seqid->list) {
 751                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 752                 status = -EAGAIN;
 753         }
 754         spin_unlock(&sequence->lock);
 755         return status;
 756 }
 757
 758 static int reclaimer(void *);
 759 struct reclaimer_args {
 760         struct nfs4_client *clp;
 761         struct completion complete;
 762 };
 763
 764 /*
 765  * State recovery routine
 766  */
 767 void
 768 nfs4_recover_state(void *data)
 769 {
 770         struct nfs4_client *clp = (struct nfs4_client *)data;
 771         struct reclaimer_args args = {
 772                 .clp = clp,
 773         };
 774         might_sleep();
 775
 776         init_completion(&args.complete);
 777
 778         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
 779                 goto out_failed_clear;
 780         wait_for_completion(&args.complete);
 781         return;
 782 out_failed_clear:
 783         set_bit(NFS4CLNT_OK, &clp->cl_state);
 784         wake_up_all(&clp->cl_waitq);
 785         rpc_wake_up(&clp->cl_rpcwaitq);
 786 }
 787
 788 /*
 789  * Schedule a state recovery attempt
 790  */
 791 void
 792 nfs4_schedule_state_recovery(struct nfs4_client *clp)
 793 {
 794         if (!clp)
 795                 return;
 796         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
 797                 schedule_work(&clp->cl_recoverd);
 798 }
 799
 800 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 801 {
 802         struct inode *inode = state->inode;
 803         struct file_lock *fl;
 804         int status = 0;
 805
 806         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 807                 if (!(fl->fl_flags & FL_POSIX))
 808                         continue;
 809                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 810                         continue;
 811                 status = ops->recover_lock(state, fl);
 812                 if (status >= 0)
 813                         continue;
 814                 switch (status) {
 815                         default:
 816                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 817                                                 __FUNCTION__, status);
 818                         case -NFS4ERR_EXPIRED:
 819                         case -NFS4ERR_NO_GRACE:
 820                         case -NFS4ERR_RECLAIM_BAD:
 821                         case -NFS4ERR_RECLAIM_CONFLICT:
 822                                 /* kill_proc(fl->fl_owner, SIGLOST, 1); */
 823                                 break;
 824                         case -NFS4ERR_STALE_CLIENTID:
 825                                 goto out_err;
 826                 }
 827         }
 828         return 0;
 829 out_err:
 830         return status;
 831 }
 832
 833 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 834 {
 835         struct nfs4_state *state;
 836         struct nfs4_lock_state *lock;
 837         int status = 0;
 838
 839         /* Note: we rely on the sp->so_states list being ordered
 840          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 841          * states first.
 842          * This is needed to ensure that the server won't give us any
 843          * read delegations that we have to return if, say, we are
 844          * recovering after a network partition or a reboot from a
 845          * server that doesn't support a grace period.
 846          */
 847         list_for_each_entry(state, &sp->so_states, open_states) {
 848                 if (state->state == 0)
 849                         continue;
 850                 status = ops->recover_open(sp, state);
 851                 if (status >= 0) {
 852                         status = nfs4_reclaim_locks(ops, state);
 853                         if (status < 0)
 854                                 goto out_err;
 855                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 856                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 857                                         printk("%s: Lock reclaim failed!\n",
 858                                                         __FUNCTION__);
 859                         }
 860                         continue;
 861                 }
 862                 switch (status) {
 863                         default:
 864                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 865                                                 __FUNCTION__, status);
 866                         case -ENOENT:
 867                         case -NFS4ERR_RECLAIM_BAD:
 868                         case -NFS4ERR_RECLAIM_CONFLICT:
 869                                 /*
 870                                  * Open state on this file cannot be recovered
 871                                  * All we can do is revert to using the zero stateid.
 872                                  */
 873                                 memset(state->stateid.data, 0,
 874                                         sizeof(state->stateid.data));
 875                                 /* Mark the file as being 'closed' */
 876                                 state->state = 0;
 877                                 break;
 878                         case -NFS4ERR_EXPIRED:
 879                         case -NFS4ERR_NO_GRACE:
 880                         case -NFS4ERR_STALE_CLIENTID:
 881                                 goto out_err;
 882                 }
 883         }
 884         return 0;
 885 out_err:
 886         return status;
 887 }
 888
 889 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 890 {
 891         struct nfs4_state_owner *sp;
 892         struct nfs4_state *state;
 893         struct nfs4_lock_state *lock;
 894
 895         /* Reset all sequence ids to zero */
 896         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 897                 sp->so_seqid.counter = 0;
 898                 sp->so_seqid.flags = 0;
 899                 list_for_each_entry(state, &sp->so_states, open_states) {
 900                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 901                                 lock->ls_seqid.counter = 0;
 902                                 lock->ls_seqid.flags = 0;
 903                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 904                         }
 905                 }
 906         }
 907 }
 908
 909 static int reclaimer(void *ptr)
 910 {
 911         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
 912         struct nfs4_client *clp = args->clp;
 913         struct nfs4_state_owner *sp;
 914         struct nfs4_state_recovery_ops *ops;
 915         int status = 0;
 916
 917         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
 918         allow_signal(SIGKILL);
 919
 920         atomic_inc(&clp->cl_count);
 921         complete(&args->complete);
 922
 923         /* Ensure exclusive access to NFSv4 state */
 924         lock_kernel();
 925         down_write(&clp->cl_sem);
 926         /* Are there any NFS mounts out there? */
 927         if (list_empty(&clp->cl_superblocks))
 928                 goto out;
 929 restart_loop:
 930         status = nfs4_proc_renew(clp);
 931         switch (status) {
 932                 case 0:
 933                 case -NFS4ERR_CB_PATH_DOWN:
 934                         goto out;
 935                 case -NFS4ERR_STALE_CLIENTID:
 936                 case -NFS4ERR_LEASE_MOVED:
 937                         ops = &nfs4_reboot_recovery_ops;
 938                         break;
 939                 default:
 940                         ops = &nfs4_network_partition_recovery_ops;
 941         };
 942         nfs4_state_mark_reclaim(clp);
 943         status = __nfs4_init_client(clp);
 944         if (status)
 945                 goto out_error;
 946         /* Mark all delegations for reclaim */
 947         nfs_delegation_mark_reclaim(clp);
 948         /* Note: list is protected by exclusive lock on cl->cl_sem */
 949         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 950                 status = nfs4_reclaim_open_state(ops, sp);
 951                 if (status < 0) {
 952                         if (status == -NFS4ERR_NO_GRACE) {
 953                                 ops = &nfs4_network_partition_recovery_ops;
 954                                 status = nfs4_reclaim_open_state(ops, sp);
 955                         }
 956                         if (status == -NFS4ERR_STALE_CLIENTID)
 957                                 goto restart_loop;
 958                         if (status == -NFS4ERR_EXPIRED)
 959                                 goto restart_loop;
 960                 }
 961         }
 962         nfs_delegation_reap_unclaimed(clp);
 963 out:
 964         set_bit(NFS4CLNT_OK, &clp->cl_state);
 965         up_write(&clp->cl_sem);
 966         unlock_kernel();
 967         wake_up_all(&clp->cl_waitq);
 968         rpc_wake_up(&clp->cl_rpcwaitq);
 969         if (status == -NFS4ERR_CB_PATH_DOWN)
 970                 nfs_handle_cb_pathdown(clp);
 971         nfs4_put_client(clp);
 972         return 0;
 973 out_error:
 974         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 975                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 976         goto out;
 977 }
 978
 979 /*
 980  * Local variables:
 981  *  c-basic-offset: 8
 982  * End:
 983  */