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

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/config.h>
  42 #include <linux/slab.h>
  43 #include <linux/smp_lock.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/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_MUTEX(&sp->so_sema);
 271         INIT_LIST_HEAD(&sp->so_states);
 272         INIT_LIST_HEAD(&sp->so_delegations);
 273         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 274         sp->so_seqid.sequence = &sp->so_sequence;
 275         spin_lock_init(&sp->so_sequence.lock);
 276         INIT_LIST_HEAD(&sp->so_sequence.list);
 277         atomic_set(&sp->so_count, 1);
 278         return sp;
 279 }
 280
 281 void
 282 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 283 {
 284         struct nfs4_client *clp = sp->so_client;
 285         spin_lock(&clp->cl_lock);
 286         list_del_init(&sp->so_list);
 287         spin_unlock(&clp->cl_lock);
 288 }
 289
 290 /*
 291  * Note: must be called with clp->cl_sem held in order to prevent races
 292  *       with reboot recovery!
 293  */
 294 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 295 {
 296         struct nfs4_client *clp = server->nfs4_state;
 297         struct nfs4_state_owner *sp, *new;
 298
 299         get_rpccred(cred);
 300         new = nfs4_alloc_state_owner();
 301         spin_lock(&clp->cl_lock);
 302         sp = nfs4_find_state_owner(clp, cred);
 303         if (sp == NULL)
 304                 sp = nfs4_client_grab_unused(clp, cred);
 305         if (sp == NULL && new != NULL) {
 306                 list_add(&new->so_list, &clp->cl_state_owners);
 307                 new->so_client = clp;
 308                 new->so_id = nfs4_alloc_lockowner_id(clp);
 309                 new->so_cred = cred;
 310                 sp = new;
 311                 new = NULL;
 312         }
 313         spin_unlock(&clp->cl_lock);
 314         if (new)
 315                 kfree(new);
 316         if (sp != NULL)
 317                 return sp;
 318         put_rpccred(cred);
 319         return NULL;
 320 }
 321
 322 /*
 323  * Must be called with clp->cl_sem held in order to avoid races
 324  * with state recovery...
 325  */
 326 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 327 {
 328         struct nfs4_client *clp = sp->so_client;
 329         struct rpc_cred *cred = sp->so_cred;
 330
 331         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 332                 return;
 333         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 334                 goto out_free;
 335         if (list_empty(&sp->so_list))
 336                 goto out_free;
 337         list_move(&sp->so_list, &clp->cl_unused);
 338         clp->cl_nunused++;
 339         spin_unlock(&clp->cl_lock);
 340         put_rpccred(cred);
 341         cred = NULL;
 342         return;
 343 out_free:
 344         list_del(&sp->so_list);
 345         spin_unlock(&clp->cl_lock);
 346         put_rpccred(cred);
 347         kfree(sp);
 348 }
 349
 350 static struct nfs4_state *
 351 nfs4_alloc_open_state(void)
 352 {
 353         struct nfs4_state *state;
 354
 355         state = kmalloc(sizeof(*state), GFP_KERNEL);
 356         if (!state)
 357                 return NULL;
 358         state->state = 0;
 359         state->nreaders = 0;
 360         state->nwriters = 0;
 361         state->flags = 0;
 362         memset(state->stateid.data, 0, sizeof(state->stateid.data));
 363         atomic_set(&state->count, 1);
 364         INIT_LIST_HEAD(&state->lock_states);
 365         init_MUTEX(&state->lock_sema);
 366         spin_lock_init(&state->state_lock);
 367         return state;
 368 }
 369
 370 static struct nfs4_state *
 371 __nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 372 {
 373         struct nfs_inode *nfsi = NFS_I(inode);
 374         struct nfs4_state *state;
 375
 376         mode &= (FMODE_READ|FMODE_WRITE);
 377         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 378                 if (state->owner->so_cred != cred)
 379                         continue;
 380                 if ((mode & FMODE_READ) != 0 && state->nreaders == 0)
 381                         continue;
 382                 if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0)
 383                         continue;
 384                 if ((state->state & mode) != mode)
 385                         continue;
 386                 atomic_inc(&state->count);
 387                 if (mode & FMODE_READ)
 388                         state->nreaders++;
 389                 if (mode & FMODE_WRITE)
 390                         state->nwriters++;
 391                 return state;
 392         }
 393         return NULL;
 394 }
 395
 396 static struct nfs4_state *
 397 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 398 {
 399         struct nfs_inode *nfsi = NFS_I(inode);
 400         struct nfs4_state *state;
 401
 402         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 403                 /* Is this in the process of being freed? */
 404                 if (state->nreaders == 0 && state->nwriters == 0)
 405                         continue;
 406                 if (state->owner == owner) {
 407                         atomic_inc(&state->count);
 408                         return state;
 409                 }
 410         }
 411         return NULL;
 412 }
 413
 414 struct nfs4_state *
 415 nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 416 {
 417         struct nfs4_state *state;
 418
 419         spin_lock(&inode->i_lock);
 420         state = __nfs4_find_state(inode, cred, mode);
 421         spin_unlock(&inode->i_lock);
 422         return state;
 423 }
 424
 425 static void
 426 nfs4_free_open_state(struct nfs4_state *state)
 427 {
 428         kfree(state);
 429 }
 430
 431 struct nfs4_state *
 432 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 433 {
 434         struct nfs4_state *state, *new;
 435         struct nfs_inode *nfsi = NFS_I(inode);
 436
 437         spin_lock(&inode->i_lock);
 438         state = __nfs4_find_state_byowner(inode, owner);
 439         spin_unlock(&inode->i_lock);
 440         if (state)
 441                 goto out;
 442         new = nfs4_alloc_open_state();
 443         spin_lock(&inode->i_lock);
 444         state = __nfs4_find_state_byowner(inode, owner);
 445         if (state == NULL && new != NULL) {
 446                 state = new;
 447                 /* Caller *must* be holding owner->so_sem */
 448                 /* Note: The reclaim code dictates that we add stateless
 449                  * and read-only stateids to the end of the list */
 450                 list_add_tail(&state->open_states, &owner->so_states);
 451                 state->owner = owner;
 452                 atomic_inc(&owner->so_count);
 453                 list_add(&state->inode_states, &nfsi->open_states);
 454                 state->inode = igrab(inode);
 455                 spin_unlock(&inode->i_lock);
 456         } else {
 457                 spin_unlock(&inode->i_lock);
 458                 if (new)
 459                         nfs4_free_open_state(new);
 460         }
 461 out:
 462         return state;
 463 }
 464
 465 /*
 466  * Beware! Caller must be holding exactly one
 467  * reference to clp->cl_sem and owner->so_sema!
 468  */
 469 void nfs4_put_open_state(struct nfs4_state *state)
 470 {
 471         struct inode *inode = state->inode;
 472         struct nfs4_state_owner *owner = state->owner;
 473
 474         if (!atomic_dec_and_lock(&state->count, &inode->i_lock))
 475                 return;
 476         if (!list_empty(&state->inode_states))
 477                 list_del(&state->inode_states);
 478         spin_unlock(&inode->i_lock);
 479         list_del(&state->open_states);
 480         iput(inode);
 481         BUG_ON (state->state != 0);
 482         nfs4_free_open_state(state);
 483         nfs4_put_state_owner(owner);
 484 }
 485
 486 /*
 487  * Beware! Caller must be holding no references to clp->cl_sem!
 488  * of owner->so_sema!
 489  */
 490 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 491 {
 492         struct inode *inode = state->inode;
 493         struct nfs4_state_owner *owner = state->owner;
 494         struct nfs4_client *clp = owner->so_client;
 495         int newstate;
 496
 497         atomic_inc(&owner->so_count);
 498         down_read(&clp->cl_sem);
 499         down(&owner->so_sema);
 500         /* Protect against nfs4_find_state() */
 501         spin_lock(&inode->i_lock);
 502         if (mode & FMODE_READ)
 503                 state->nreaders--;
 504         if (mode & FMODE_WRITE)
 505                 state->nwriters--;
 506         if (state->nwriters == 0) {
 507                 if (state->nreaders == 0)
 508                         list_del_init(&state->inode_states);
 509                 /* See reclaim code */
 510                 list_move_tail(&state->open_states, &owner->so_states);
 511         }
 512         spin_unlock(&inode->i_lock);
 513         newstate = 0;
 514         if (state->state != 0) {
 515                 if (state->nreaders)
 516                         newstate |= FMODE_READ;
 517                 if (state->nwriters)
 518                         newstate |= FMODE_WRITE;
 519                 if (state->state == newstate)
 520                         goto out;
 521                 if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 522                         state->state = newstate;
 523                         goto out;
 524                 }
 525                 if (nfs4_do_close(inode, state, newstate) == 0)
 526                         return;
 527         }
 528 out:
 529         nfs4_put_open_state(state);
 530         up(&owner->so_sema);
 531         nfs4_put_state_owner(owner);
 532         up_read(&clp->cl_sem);
 533 }
 534
 535 /*
 536  * Search the state->lock_states for an existing lock_owner
 537  * that is compatible with current->files
 538  */
 539 static struct nfs4_lock_state *
 540 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 541 {
 542         struct nfs4_lock_state *pos;
 543         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 544                 if (pos->ls_owner != fl_owner)
 545                         continue;
 546                 atomic_inc(&pos->ls_count);
 547                 return pos;
 548         }
 549         return NULL;
 550 }
 551
 552 /*
 553  * Return a compatible lock_state. If no initialized lock_state structure
 554  * exists, return an uninitialized one.
 555  *
 556  * The caller must be holding state->lock_sema
 557  */
 558 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 559 {
 560         struct nfs4_lock_state *lsp;
 561         struct nfs4_client *clp = state->owner->so_client;
 562
 563         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 564         if (lsp == NULL)
 565                 return NULL;
 566         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 567         atomic_set(&lsp->ls_count, 1);
 568         lsp->ls_owner = fl_owner;
 569         spin_lock(&clp->cl_lock);
 570         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 571         spin_unlock(&clp->cl_lock);
 572         INIT_LIST_HEAD(&lsp->ls_locks);
 573         return lsp;
 574 }
 575
 576 /*
 577  * Return a compatible lock_state. If no initialized lock_state structure
 578  * exists, return an uninitialized one.
 579  *
 580  * The caller must be holding state->lock_sema and clp->cl_sem
 581  */
 582 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 583 {
 584         struct nfs4_lock_state *lsp, *new = NULL;
 585
 586         for(;;) {
 587                 spin_lock(&state->state_lock);
 588                 lsp = __nfs4_find_lock_state(state, owner);
 589                 if (lsp != NULL)
 590                         break;
 591                 if (new != NULL) {
 592                         new->ls_state = state;
 593                         list_add(&new->ls_locks, &state->lock_states);
 594                         set_bit(LK_STATE_IN_USE, &state->flags);
 595                         lsp = new;
 596                         new = NULL;
 597                         break;
 598                 }
 599                 spin_unlock(&state->state_lock);
 600                 new = nfs4_alloc_lock_state(state, owner);
 601                 if (new == NULL)
 602                         return NULL;
 603         }
 604         spin_unlock(&state->state_lock);
 605         kfree(new);
 606         return lsp;
 607 }
 608
 609 /*
 610  * Release reference to lock_state, and free it if we see that
 611  * it is no longer in use
 612  */
 613 static void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 614 {
 615         struct nfs4_state *state;
 616
 617         if (lsp == NULL)
 618                 return;
 619         state = lsp->ls_state;
 620         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 621                 return;
 622         list_del(&lsp->ls_locks);
 623         if (list_empty(&state->lock_states))
 624                 clear_bit(LK_STATE_IN_USE, &state->flags);
 625         spin_unlock(&state->state_lock);
 626         kfree(lsp);
 627 }
 628
 629 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 630 {
 631         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 632
 633         dst->fl_u.nfs4_fl.owner = lsp;
 634         atomic_inc(&lsp->ls_count);
 635 }
 636
 637 static void nfs4_fl_release_lock(struct file_lock *fl)
 638 {
 639         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 640 }
 641
 642 static struct file_lock_operations nfs4_fl_lock_ops = {
 643         .fl_copy_lock = nfs4_fl_copy_lock,
 644         .fl_release_private = nfs4_fl_release_lock,
 645 };
 646
 647 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 648 {
 649         struct nfs4_lock_state *lsp;
 650
 651         if (fl->fl_ops != NULL)
 652                 return 0;
 653         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 654         if (lsp == NULL)
 655                 return -ENOMEM;
 656         fl->fl_u.nfs4_fl.owner = lsp;
 657         fl->fl_ops = &nfs4_fl_lock_ops;
 658         return 0;
 659 }
 660
 661 /*
 662  * Byte-range lock aware utility to initialize the stateid of read/write
 663  * requests.
 664  */
 665 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 666 {
 667         struct nfs4_lock_state *lsp;
 668
 669         memcpy(dst, &state->stateid, sizeof(*dst));
 670         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 671                 return;
 672
 673         spin_lock(&state->state_lock);
 674         lsp = __nfs4_find_lock_state(state, fl_owner);
 675         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 676                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 677         spin_unlock(&state->state_lock);
 678         nfs4_put_lock_state(lsp);
 679 }
 680
 681 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 682 {
 683         struct rpc_sequence *sequence = counter->sequence;
 684         struct nfs_seqid *new;
 685
 686         new = kmalloc(sizeof(*new), GFP_KERNEL);
 687         if (new != NULL) {
 688                 new->sequence = counter;
 689                 new->task = NULL;
 690                 spin_lock(&sequence->lock);
 691                 list_add_tail(&new->list, &sequence->list);
 692                 spin_unlock(&sequence->lock);
 693         }
 694         return new;
 695 }
 696
 697 void nfs_free_seqid(struct nfs_seqid *seqid)
 698 {
 699         struct rpc_sequence *sequence = seqid->sequence->sequence;
 700         struct rpc_task *next = NULL;
 701
 702         spin_lock(&sequence->lock);
 703         list_del(&seqid->list);
 704         if (!list_empty(&sequence->list)) {
 705                 next = list_entry(sequence->list.next, struct nfs_seqid, list)->task;
 706                 if (next)
 707                         rpc_wake_up_task(next);
 708         }
 709         spin_unlock(&sequence->lock);
 710         kfree(seqid);
 711 }
 712
 713 /*
 714  * Called with sp->so_sema and clp->cl_sem held.
 715  *
 716  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 717  * failed with a seqid incrementing error -
 718  * see comments nfs_fs.h:seqid_mutating_error()
 719  */
 720 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 721 {
 722         switch (status) {
 723                 case 0:
 724                         break;
 725                 case -NFS4ERR_BAD_SEQID:
 726                 case -NFS4ERR_STALE_CLIENTID:
 727                 case -NFS4ERR_STALE_STATEID:
 728                 case -NFS4ERR_BAD_STATEID:
 729                 case -NFS4ERR_BADXDR:
 730                 case -NFS4ERR_RESOURCE:
 731                 case -NFS4ERR_NOFILEHANDLE:
 732                         /* Non-seqid mutating errors */
 733                         return;
 734         };
 735         /*
 736          * Note: no locking needed as we are guaranteed to be first
 737          * on the sequence list
 738          */
 739         seqid->sequence->counter++;
 740 }
 741
 742 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 743 {
 744         if (status == -NFS4ERR_BAD_SEQID) {
 745                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 746                                 struct nfs4_state_owner, so_seqid);
 747                 nfs4_drop_state_owner(sp);
 748         }
 749         return nfs_increment_seqid(status, seqid);
 750 }
 751
 752 /*
 753  * Called with ls->lock_sema and clp->cl_sem held.
 754  *
 755  * Increment the seqid if the LOCK/LOCKU succeeded, or
 756  * failed with a seqid incrementing error -
 757  * see comments nfs_fs.h:seqid_mutating_error()
 758  */
 759 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 760 {
 761         return nfs_increment_seqid(status, seqid);
 762 }
 763
 764 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 765 {
 766         struct rpc_sequence *sequence = seqid->sequence->sequence;
 767         int status = 0;
 768
 769         spin_lock(&sequence->lock);
 770         if (sequence->list.next != &seqid->list) {
 771                 seqid->task = task;
 772                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 773                 status = -EAGAIN;
 774         }
 775         spin_unlock(&sequence->lock);
 776         return status;
 777 }
 778
 779 static int reclaimer(void *);
 780 struct reclaimer_args {
 781         struct nfs4_client *clp;
 782         struct completion complete;
 783 };
 784
 785 /*
 786  * State recovery routine
 787  */
 788 void
 789 nfs4_recover_state(void *data)
 790 {
 791         struct nfs4_client *clp = (struct nfs4_client *)data;
 792         struct reclaimer_args args = {
 793                 .clp = clp,
 794         };
 795         might_sleep();
 796
 797         init_completion(&args.complete);
 798
 799         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
 800                 goto out_failed_clear;
 801         wait_for_completion(&args.complete);
 802         return;
 803 out_failed_clear:
 804         set_bit(NFS4CLNT_OK, &clp->cl_state);
 805         wake_up_all(&clp->cl_waitq);
 806         rpc_wake_up(&clp->cl_rpcwaitq);
 807 }
 808
 809 /*
 810  * Schedule a state recovery attempt
 811  */
 812 void
 813 nfs4_schedule_state_recovery(struct nfs4_client *clp)
 814 {
 815         if (!clp)
 816                 return;
 817         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
 818                 schedule_work(&clp->cl_recoverd);
 819 }
 820
 821 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 822 {
 823         struct inode *inode = state->inode;
 824         struct file_lock *fl;
 825         int status = 0;
 826
 827         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 828                 if (!(fl->fl_flags & FL_POSIX))
 829                         continue;
 830                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 831                         continue;
 832                 status = ops->recover_lock(state, fl);
 833                 if (status >= 0)
 834                         continue;
 835                 switch (status) {
 836                         default:
 837                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 838                                                 __FUNCTION__, status);
 839                         case -NFS4ERR_EXPIRED:
 840                         case -NFS4ERR_NO_GRACE:
 841                         case -NFS4ERR_RECLAIM_BAD:
 842                         case -NFS4ERR_RECLAIM_CONFLICT:
 843                                 /* kill_proc(fl->fl_owner, SIGLOST, 1); */
 844                                 break;
 845                         case -NFS4ERR_STALE_CLIENTID:
 846                                 goto out_err;
 847                 }
 848         }
 849         return 0;
 850 out_err:
 851         return status;
 852 }
 853
 854 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 855 {
 856         struct nfs4_state *state;
 857         struct nfs4_lock_state *lock;
 858         int status = 0;
 859
 860         /* Note: we rely on the sp->so_states list being ordered
 861          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 862          * states first.
 863          * This is needed to ensure that the server won't give us any
 864          * read delegations that we have to return if, say, we are
 865          * recovering after a network partition or a reboot from a
 866          * server that doesn't support a grace period.
 867          */
 868         list_for_each_entry(state, &sp->so_states, open_states) {
 869                 if (state->state == 0)
 870                         continue;
 871                 status = ops->recover_open(sp, state);
 872                 if (status >= 0) {
 873                         status = nfs4_reclaim_locks(ops, state);
 874                         if (status < 0)
 875                                 goto out_err;
 876                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 877                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 878                                         printk("%s: Lock reclaim failed!\n",
 879                                                         __FUNCTION__);
 880                         }
 881                         continue;
 882                 }
 883                 switch (status) {
 884                         default:
 885                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 886                                                 __FUNCTION__, status);
 887                         case -ENOENT:
 888                         case -NFS4ERR_RECLAIM_BAD:
 889                         case -NFS4ERR_RECLAIM_CONFLICT:
 890                                 /*
 891                                  * Open state on this file cannot be recovered
 892                                  * All we can do is revert to using the zero stateid.
 893                                  */
 894                                 memset(state->stateid.data, 0,
 895                                         sizeof(state->stateid.data));
 896                                 /* Mark the file as being 'closed' */
 897                                 state->state = 0;
 898                                 break;
 899                         case -NFS4ERR_EXPIRED:
 900                         case -NFS4ERR_NO_GRACE:
 901                         case -NFS4ERR_STALE_CLIENTID:
 902                                 goto out_err;
 903                 }
 904         }
 905         return 0;
 906 out_err:
 907         return status;
 908 }
 909
 910 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 911 {
 912         struct nfs4_state_owner *sp;
 913         struct nfs4_state *state;
 914         struct nfs4_lock_state *lock;
 915
 916         /* Reset all sequence ids to zero */
 917         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 918                 sp->so_seqid.counter = 0;
 919                 sp->so_seqid.flags = 0;
 920                 list_for_each_entry(state, &sp->so_states, open_states) {
 921                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 922                                 lock->ls_seqid.counter = 0;
 923                                 lock->ls_seqid.flags = 0;
 924                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 925                         }
 926                 }
 927         }
 928 }
 929
 930 static int reclaimer(void *ptr)
 931 {
 932         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
 933         struct nfs4_client *clp = args->clp;
 934         struct nfs4_state_owner *sp;
 935         struct nfs4_state_recovery_ops *ops;
 936         int status = 0;
 937
 938         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
 939         allow_signal(SIGKILL);
 940
 941         atomic_inc(&clp->cl_count);
 942         complete(&args->complete);
 943
 944         /* Ensure exclusive access to NFSv4 state */
 945         lock_kernel();
 946         down_write(&clp->cl_sem);
 947         /* Are there any NFS mounts out there? */
 948         if (list_empty(&clp->cl_superblocks))
 949                 goto out;
 950 restart_loop:
 951         status = nfs4_proc_renew(clp);
 952         switch (status) {
 953                 case 0:
 954                 case -NFS4ERR_CB_PATH_DOWN:
 955                         goto out;
 956                 case -NFS4ERR_STALE_CLIENTID:
 957                 case -NFS4ERR_LEASE_MOVED:
 958                         ops = &nfs4_reboot_recovery_ops;
 959                         break;
 960                 default:
 961                         ops = &nfs4_network_partition_recovery_ops;
 962         };
 963         nfs4_state_mark_reclaim(clp);
 964         status = __nfs4_init_client(clp);
 965         if (status)
 966                 goto out_error;
 967         /* Mark all delegations for reclaim */
 968         nfs_delegation_mark_reclaim(clp);
 969         /* Note: list is protected by exclusive lock on cl->cl_sem */
 970         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 971                 status = nfs4_reclaim_open_state(ops, sp);
 972                 if (status < 0) {
 973                         if (status == -NFS4ERR_NO_GRACE) {
 974                                 ops = &nfs4_network_partition_recovery_ops;
 975                                 status = nfs4_reclaim_open_state(ops, sp);
 976                         }
 977                         if (status == -NFS4ERR_STALE_CLIENTID)
 978                                 goto restart_loop;
 979                         if (status == -NFS4ERR_EXPIRED)
 980                                 goto restart_loop;
 981                 }
 982         }
 983         nfs_delegation_reap_unclaimed(clp);
 984 out:
 985         set_bit(NFS4CLNT_OK, &clp->cl_state);
 986         up_write(&clp->cl_sem);
 987         unlock_kernel();
 988         wake_up_all(&clp->cl_waitq);
 989         rpc_wake_up(&clp->cl_rpcwaitq);
 990         if (status == -NFS4ERR_CB_PATH_DOWN)
 991                 nfs_handle_cb_pathdown(clp);
 992         nfs4_put_client(clp);
 993         return 0;
 994 out_error:
 995         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 996                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 997         goto out;
 998 }
 999
1000 /*
1001  * Local variables:
1002  *  c-basic-offset: 8
1003  * End:
1004  */