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 = kmalloc(sizeof(*sp),GFP_KERNEL);
 268         if (!sp)
 269                 return NULL;
 270         init_MUTEX(&sp->so_sema);
 271         sp->so_seqid = 0;                 /* arbitrary */
 272         INIT_LIST_HEAD(&sp->so_states);
 273         INIT_LIST_HEAD(&sp->so_delegations);
 274         atomic_set(&sp->so_count, 1);
 275         return sp;
 276 }
 277
 278 void
 279 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 280 {
 281         struct nfs4_client *clp = sp->so_client;
 282         spin_lock(&clp->cl_lock);
 283         list_del_init(&sp->so_list);
 284         spin_unlock(&clp->cl_lock);
 285 }
 286
 287 /*
 288  * Note: must be called with clp->cl_sem held in order to prevent races
 289  *       with reboot recovery!
 290  */
 291 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 292 {
 293         struct nfs4_client *clp = server->nfs4_state;
 294         struct nfs4_state_owner *sp, *new;
 295
 296         get_rpccred(cred);
 297         new = nfs4_alloc_state_owner();
 298         spin_lock(&clp->cl_lock);
 299         sp = nfs4_find_state_owner(clp, cred);
 300         if (sp == NULL)
 301                 sp = nfs4_client_grab_unused(clp, cred);
 302         if (sp == NULL && new != NULL) {
 303                 list_add(&new->so_list, &clp->cl_state_owners);
 304                 new->so_client = clp;
 305                 new->so_id = nfs4_alloc_lockowner_id(clp);
 306                 new->so_cred = cred;
 307                 sp = new;
 308                 new = NULL;
 309         }
 310         spin_unlock(&clp->cl_lock);
 311         if (new)
 312                 kfree(new);
 313         if (sp != NULL)
 314                 return sp;
 315         put_rpccred(cred);
 316         return NULL;
 317 }
 318
 319 /*
 320  * Must be called with clp->cl_sem held in order to avoid races
 321  * with state recovery...
 322  */
 323 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 324 {
 325         struct nfs4_client *clp = sp->so_client;
 326         struct rpc_cred *cred = sp->so_cred;
 327
 328         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 329                 return;
 330         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 331                 goto out_free;
 332         if (list_empty(&sp->so_list))
 333                 goto out_free;
 334         list_move(&sp->so_list, &clp->cl_unused);
 335         clp->cl_nunused++;
 336         spin_unlock(&clp->cl_lock);
 337         put_rpccred(cred);
 338         cred = NULL;
 339         return;
 340 out_free:
 341         list_del(&sp->so_list);
 342         spin_unlock(&clp->cl_lock);
 343         put_rpccred(cred);
 344         kfree(sp);
 345 }
 346
 347 static struct nfs4_state *
 348 nfs4_alloc_open_state(void)
 349 {
 350         struct nfs4_state *state;
 351
 352         state = kmalloc(sizeof(*state), GFP_KERNEL);
 353         if (!state)
 354                 return NULL;
 355         state->state = 0;
 356         state->nreaders = 0;
 357         state->nwriters = 0;
 358         state->flags = 0;
 359         memset(state->stateid.data, 0, sizeof(state->stateid.data));
 360         atomic_set(&state->count, 1);
 361         INIT_LIST_HEAD(&state->lock_states);
 362         init_MUTEX(&state->lock_sema);
 363         spin_lock_init(&state->state_lock);
 364         return state;
 365 }
 366
 367 static struct nfs4_state *
 368 __nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 369 {
 370         struct nfs_inode *nfsi = NFS_I(inode);
 371         struct nfs4_state *state;
 372
 373         mode &= (FMODE_READ|FMODE_WRITE);
 374         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 375                 if (state->owner->so_cred != cred)
 376                         continue;
 377                 if ((mode & FMODE_READ) != 0 && state->nreaders == 0)
 378                         continue;
 379                 if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0)
 380                         continue;
 381                 if ((state->state & mode) != mode)
 382                         continue;
 383                 atomic_inc(&state->count);
 384                 if (mode & FMODE_READ)
 385                         state->nreaders++;
 386                 if (mode & FMODE_WRITE)
 387                         state->nwriters++;
 388                 return state;
 389         }
 390         return NULL;
 391 }
 392
 393 static struct nfs4_state *
 394 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 395 {
 396         struct nfs_inode *nfsi = NFS_I(inode);
 397         struct nfs4_state *state;
 398
 399         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 400                 /* Is this in the process of being freed? */
 401                 if (state->nreaders == 0 && state->nwriters == 0)
 402                         continue;
 403                 if (state->owner == owner) {
 404                         atomic_inc(&state->count);
 405                         return state;
 406                 }
 407         }
 408         return NULL;
 409 }
 410
 411 struct nfs4_state *
 412 nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 413 {
 414         struct nfs4_state *state;
 415
 416         spin_lock(&inode->i_lock);
 417         state = __nfs4_find_state(inode, cred, mode);
 418         spin_unlock(&inode->i_lock);
 419         return state;
 420 }
 421
 422 static void
 423 nfs4_free_open_state(struct nfs4_state *state)
 424 {
 425         kfree(state);
 426 }
 427
 428 struct nfs4_state *
 429 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 430 {
 431         struct nfs4_state *state, *new;
 432         struct nfs_inode *nfsi = NFS_I(inode);
 433
 434         spin_lock(&inode->i_lock);
 435         state = __nfs4_find_state_byowner(inode, owner);
 436         spin_unlock(&inode->i_lock);
 437         if (state)
 438                 goto out;
 439         new = nfs4_alloc_open_state();
 440         spin_lock(&inode->i_lock);
 441         state = __nfs4_find_state_byowner(inode, owner);
 442         if (state == NULL && new != NULL) {
 443                 state = new;
 444                 /* Caller *must* be holding owner->so_sem */
 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 and owner->so_sema!
 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  * Beware! Caller must be holding no references to clp->cl_sem!
 485  * of owner->so_sema!
 486  */
 487 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 488 {
 489         struct inode *inode = state->inode;
 490         struct nfs4_state_owner *owner = state->owner;
 491         struct nfs4_client *clp = owner->so_client;
 492         int newstate;
 493
 494         atomic_inc(&owner->so_count);
 495         down_read(&clp->cl_sem);
 496         down(&owner->so_sema);
 497         /* Protect against nfs4_find_state() */
 498         spin_lock(&inode->i_lock);
 499         if (mode & FMODE_READ)
 500                 state->nreaders--;
 501         if (mode & FMODE_WRITE)
 502                 state->nwriters--;
 503         if (state->nwriters == 0) {
 504                 if (state->nreaders == 0)
 505                         list_del_init(&state->inode_states);
 506                 /* See reclaim code */
 507                 list_move_tail(&state->open_states, &owner->so_states);
 508         }
 509         spin_unlock(&inode->i_lock);
 510         newstate = 0;
 511         if (state->state != 0) {
 512                 if (state->nreaders)
 513                         newstate |= FMODE_READ;
 514                 if (state->nwriters)
 515                         newstate |= FMODE_WRITE;
 516                 if (state->state == newstate)
 517                         goto out;
 518                 if (nfs4_do_close(inode, state, newstate) == -EINPROGRESS)
 519                         return;
 520         }
 521 out:
 522         nfs4_put_open_state(state);
 523         up(&owner->so_sema);
 524         nfs4_put_state_owner(owner);
 525         up_read(&clp->cl_sem);
 526 }
 527
 528 /*
 529  * Search the state->lock_states for an existing lock_owner
 530  * that is compatible with current->files
 531  */
 532 static struct nfs4_lock_state *
 533 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 534 {
 535         struct nfs4_lock_state *pos;
 536         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 537                 if (pos->ls_owner != fl_owner)
 538                         continue;
 539                 atomic_inc(&pos->ls_count);
 540                 return pos;
 541         }
 542         return NULL;
 543 }
 544
 545 /*
 546  * Return a compatible lock_state. If no initialized lock_state structure
 547  * exists, return an uninitialized one.
 548  *
 549  * The caller must be holding state->lock_sema
 550  */
 551 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 552 {
 553         struct nfs4_lock_state *lsp;
 554         struct nfs4_client *clp = state->owner->so_client;
 555
 556         lsp = kmalloc(sizeof(*lsp), GFP_KERNEL);
 557         if (lsp == NULL)
 558                 return NULL;
 559         lsp->ls_flags = 0;
 560         lsp->ls_seqid = 0;      /* arbitrary */
 561         memset(lsp->ls_stateid.data, 0, sizeof(lsp->ls_stateid.data));
 562         atomic_set(&lsp->ls_count, 1);
 563         lsp->ls_owner = fl_owner;
 564         spin_lock(&clp->cl_lock);
 565         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 566         spin_unlock(&clp->cl_lock);
 567         INIT_LIST_HEAD(&lsp->ls_locks);
 568         return lsp;
 569 }
 570
 571 /*
 572  * Return a compatible lock_state. If no initialized lock_state structure
 573  * exists, return an uninitialized one.
 574  *
 575  * The caller must be holding state->lock_sema and clp->cl_sem
 576  */
 577 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 578 {
 579         struct nfs4_lock_state *lsp, *new = NULL;
 580
 581         for(;;) {
 582                 spin_lock(&state->state_lock);
 583                 lsp = __nfs4_find_lock_state(state, owner);
 584                 if (lsp != NULL)
 585                         break;
 586                 if (new != NULL) {
 587                         new->ls_state = state;
 588                         list_add(&new->ls_locks, &state->lock_states);
 589                         set_bit(LK_STATE_IN_USE, &state->flags);
 590                         lsp = new;
 591                         new = NULL;
 592                         break;
 593                 }
 594                 spin_unlock(&state->state_lock);
 595                 new = nfs4_alloc_lock_state(state, owner);
 596                 if (new == NULL)
 597                         return NULL;
 598         }
 599         spin_unlock(&state->state_lock);
 600         kfree(new);
 601         return lsp;
 602 }
 603
 604 /*
 605  * Release reference to lock_state, and free it if we see that
 606  * it is no longer in use
 607  */
 608 static void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 609 {
 610         struct nfs4_state *state;
 611
 612         if (lsp == NULL)
 613                 return;
 614         state = lsp->ls_state;
 615         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 616                 return;
 617         list_del(&lsp->ls_locks);
 618         if (list_empty(&state->lock_states))
 619                 clear_bit(LK_STATE_IN_USE, &state->flags);
 620         spin_unlock(&state->state_lock);
 621         kfree(lsp);
 622 }
 623
 624 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 625 {
 626         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 627
 628         dst->fl_u.nfs4_fl.owner = lsp;
 629         atomic_inc(&lsp->ls_count);
 630 }
 631
 632 static void nfs4_fl_release_lock(struct file_lock *fl)
 633 {
 634         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 635 }
 636
 637 static struct file_lock_operations nfs4_fl_lock_ops = {
 638         .fl_copy_lock = nfs4_fl_copy_lock,
 639         .fl_release_private = nfs4_fl_release_lock,
 640 };
 641
 642 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 643 {
 644         struct nfs4_lock_state *lsp;
 645
 646         if (fl->fl_ops != NULL)
 647                 return 0;
 648         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 649         if (lsp == NULL)
 650                 return -ENOMEM;
 651         fl->fl_u.nfs4_fl.owner = lsp;
 652         fl->fl_ops = &nfs4_fl_lock_ops;
 653         return 0;
 654 }
 655
 656 /*
 657  * Byte-range lock aware utility to initialize the stateid of read/write
 658  * requests.
 659  */
 660 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 661 {
 662         struct nfs4_lock_state *lsp;
 663
 664         memcpy(dst, &state->stateid, sizeof(*dst));
 665         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 666                 return;
 667
 668         spin_lock(&state->state_lock);
 669         lsp = __nfs4_find_lock_state(state, fl_owner);
 670         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 671                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 672         spin_unlock(&state->state_lock);
 673         nfs4_put_lock_state(lsp);
 674 }
 675
 676 /*
 677 * Called with state->lock_sema and clp->cl_sem held.
 678 */
 679 void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *lsp)
 680 {
 681         if (status == NFS_OK || seqid_mutating_err(-status))
 682                 lsp->ls_seqid++;
 683 }
 684
 685 /*
 686 * Called with sp->so_sema and clp->cl_sem held.
 687 *
 688 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 689 * failed with a seqid incrementing error -
 690 * see comments nfs_fs.h:seqid_mutating_error()
 691 */
 692 void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp)
 693 {
 694         if (status == NFS_OK || seqid_mutating_err(-status))
 695                 sp->so_seqid++;
 696         /* If the server returns BAD_SEQID, unhash state_owner here */
 697         if (status == -NFS4ERR_BAD_SEQID)
 698                 nfs4_drop_state_owner(sp);
 699 }
 700
 701 static int reclaimer(void *);
 702 struct reclaimer_args {
 703         struct nfs4_client *clp;
 704         struct completion complete;
 705 };
 706
 707 /*
 708  * State recovery routine
 709  */
 710 void
 711 nfs4_recover_state(void *data)
 712 {
 713         struct nfs4_client *clp = (struct nfs4_client *)data;
 714         struct reclaimer_args args = {
 715                 .clp = clp,
 716         };
 717         might_sleep();
 718
 719         init_completion(&args.complete);
 720
 721         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
 722                 goto out_failed_clear;
 723         wait_for_completion(&args.complete);
 724         return;
 725 out_failed_clear:
 726         set_bit(NFS4CLNT_OK, &clp->cl_state);
 727         wake_up_all(&clp->cl_waitq);
 728         rpc_wake_up(&clp->cl_rpcwaitq);
 729 }
 730
 731 /*
 732  * Schedule a state recovery attempt
 733  */
 734 void
 735 nfs4_schedule_state_recovery(struct nfs4_client *clp)
 736 {
 737         if (!clp)
 738                 return;
 739         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
 740                 schedule_work(&clp->cl_recoverd);
 741 }
 742
 743 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 744 {
 745         struct inode *inode = state->inode;
 746         struct file_lock *fl;
 747         int status = 0;
 748
 749         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 750                 if (!(fl->fl_flags & FL_POSIX))
 751                         continue;
 752                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 753                         continue;
 754                 status = ops->recover_lock(state, fl);
 755                 if (status >= 0)
 756                         continue;
 757                 switch (status) {
 758                         default:
 759                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 760                                                 __FUNCTION__, status);
 761                         case -NFS4ERR_EXPIRED:
 762                         case -NFS4ERR_NO_GRACE:
 763                         case -NFS4ERR_RECLAIM_BAD:
 764                         case -NFS4ERR_RECLAIM_CONFLICT:
 765                                 /* kill_proc(fl->fl_owner, SIGLOST, 1); */
 766                                 break;
 767                         case -NFS4ERR_STALE_CLIENTID:
 768                                 goto out_err;
 769                 }
 770         }
 771         return 0;
 772 out_err:
 773         return status;
 774 }
 775
 776 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 777 {
 778         struct nfs4_state *state;
 779         struct nfs4_lock_state *lock;
 780         int status = 0;
 781
 782         /* Note: we rely on the sp->so_states list being ordered
 783          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 784          * states first.
 785          * This is needed to ensure that the server won't give us any
 786          * read delegations that we have to return if, say, we are
 787          * recovering after a network partition or a reboot from a
 788          * server that doesn't support a grace period.
 789          */
 790         list_for_each_entry(state, &sp->so_states, open_states) {
 791                 if (state->state == 0)
 792                         continue;
 793                 status = ops->recover_open(sp, state);
 794                 list_for_each_entry(lock, &state->lock_states, ls_locks)
 795                         lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 796                 if (status >= 0) {
 797                         status = nfs4_reclaim_locks(ops, state);
 798                         if (status < 0)
 799                                 goto out_err;
 800                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 801                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 802                                         printk("%s: Lock reclaim failed!\n",
 803                                                         __FUNCTION__);
 804                         }
 805                         continue;
 806                 }
 807                 switch (status) {
 808                         default:
 809                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 810                                                 __FUNCTION__, status);
 811                         case -ENOENT:
 812                         case -NFS4ERR_RECLAIM_BAD:
 813                         case -NFS4ERR_RECLAIM_CONFLICT:
 814                                 /*
 815                                  * Open state on this file cannot be recovered
 816                                  * All we can do is revert to using the zero stateid.
 817                                  */
 818                                 memset(state->stateid.data, 0,
 819                                         sizeof(state->stateid.data));
 820                                 /* Mark the file as being 'closed' */
 821                                 state->state = 0;
 822                                 break;
 823                         case -NFS4ERR_EXPIRED:
 824                         case -NFS4ERR_NO_GRACE:
 825                         case -NFS4ERR_STALE_CLIENTID:
 826                                 goto out_err;
 827                 }
 828         }
 829         return 0;
 830 out_err:
 831         return status;
 832 }
 833
 834 static int reclaimer(void *ptr)
 835 {
 836         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
 837         struct nfs4_client *clp = args->clp;
 838         struct nfs4_state_owner *sp;
 839         struct nfs4_state_recovery_ops *ops;
 840         int status = 0;
 841
 842         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
 843         allow_signal(SIGKILL);
 844
 845         atomic_inc(&clp->cl_count);
 846         complete(&args->complete);
 847
 848         /* Ensure exclusive access to NFSv4 state */
 849         lock_kernel();
 850         down_write(&clp->cl_sem);
 851         /* Are there any NFS mounts out there? */
 852         if (list_empty(&clp->cl_superblocks))
 853                 goto out;
 854 restart_loop:
 855         status = nfs4_proc_renew(clp);
 856         switch (status) {
 857                 case 0:
 858                 case -NFS4ERR_CB_PATH_DOWN:
 859                         goto out;
 860                 case -NFS4ERR_STALE_CLIENTID:
 861                 case -NFS4ERR_LEASE_MOVED:
 862                         ops = &nfs4_reboot_recovery_ops;
 863                         break;
 864                 default:
 865                         ops = &nfs4_network_partition_recovery_ops;
 866         };
 867         status = __nfs4_init_client(clp);
 868         if (status)
 869                 goto out_error;
 870         /* Mark all delegations for reclaim */
 871         nfs_delegation_mark_reclaim(clp);
 872         /* Note: list is protected by exclusive lock on cl->cl_sem */
 873         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 874                 status = nfs4_reclaim_open_state(ops, sp);
 875                 if (status < 0) {
 876                         if (status == -NFS4ERR_NO_GRACE) {
 877                                 ops = &nfs4_network_partition_recovery_ops;
 878                                 status = nfs4_reclaim_open_state(ops, sp);
 879                         }
 880                         if (status == -NFS4ERR_STALE_CLIENTID)
 881                                 goto restart_loop;
 882                         if (status == -NFS4ERR_EXPIRED)
 883                                 goto restart_loop;
 884                 }
 885         }
 886         nfs_delegation_reap_unclaimed(clp);
 887 out:
 888         set_bit(NFS4CLNT_OK, &clp->cl_state);
 889         up_write(&clp->cl_sem);
 890         unlock_kernel();
 891         wake_up_all(&clp->cl_waitq);
 892         rpc_wake_up(&clp->cl_rpcwaitq);
 893         if (status == -NFS4ERR_CB_PATH_DOWN)
 894                 nfs_handle_cb_pathdown(clp);
 895         nfs4_put_client(clp);
 896         return 0;
 897 out_error:
 898         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 899                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 900         goto out;
 901 }
 902
 903 /*
 904  * Local variables:
 905  *  c-basic-offset: 8
 906  * End:
 907  */