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_boot_time = CURRENT_TIME;
 114         clp->cl_state = 1 << NFS4CLNT_OK;
 115         return clp;
 116 }
 117
 118 static void
 119 nfs4_free_client(struct nfs4_client *clp)
 120 {
 121         struct nfs4_state_owner *sp;
 122
 123         while (!list_empty(&clp->cl_unused)) {
 124                 sp = list_entry(clp->cl_unused.next,
 125                                 struct nfs4_state_owner,
 126                                 so_list);
 127                 list_del(&sp->so_list);
 128                 kfree(sp);
 129         }
 130         BUG_ON(!list_empty(&clp->cl_state_owners));
 131         if (clp->cl_cred)
 132                 put_rpccred(clp->cl_cred);
 133         nfs_idmap_delete(clp);
 134         if (clp->cl_rpcclient)
 135                 rpc_shutdown_client(clp->cl_rpcclient);
 136         kfree(clp);
 137         nfs_callback_down();
 138 }
 139
 140 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 141 {
 142         struct nfs4_client *clp;
 143         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 144                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 145                         atomic_inc(&clp->cl_count);
 146                         return clp;
 147                 }
 148         }
 149         return NULL;
 150 }
 151
 152 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 153 {
 154         struct nfs4_client *clp;
 155         spin_lock(&state_spinlock);
 156         clp = __nfs4_find_client(addr);
 157         spin_unlock(&state_spinlock);
 158         return clp;
 159 }
 160
 161 struct nfs4_client *
 162 nfs4_get_client(struct in_addr *addr)
 163 {
 164         struct nfs4_client *clp, *new = NULL;
 165
 166         spin_lock(&state_spinlock);
 167         for (;;) {
 168                 clp = __nfs4_find_client(addr);
 169                 if (clp != NULL)
 170                         break;
 171                 clp = new;
 172                 if (clp != NULL) {
 173                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 174                         new = NULL;
 175                         break;
 176                 }
 177                 spin_unlock(&state_spinlock);
 178                 new = nfs4_alloc_client(addr);
 179                 spin_lock(&state_spinlock);
 180                 if (new == NULL)
 181                         break;
 182         }
 183         spin_unlock(&state_spinlock);
 184         if (new)
 185                 nfs4_free_client(new);
 186         return clp;
 187 }
 188
 189 void
 190 nfs4_put_client(struct nfs4_client *clp)
 191 {
 192         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 193                 return;
 194         list_del(&clp->cl_servers);
 195         spin_unlock(&state_spinlock);
 196         BUG_ON(!list_empty(&clp->cl_superblocks));
 197         wake_up_all(&clp->cl_waitq);
 198         rpc_wake_up(&clp->cl_rpcwaitq);
 199         nfs4_kill_renewd(clp);
 200         nfs4_free_client(clp);
 201 }
 202
 203 static int __nfs4_init_client(struct nfs4_client *clp)
 204 {
 205         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
 206         if (status == 0)
 207                 status = nfs4_proc_setclientid_confirm(clp);
 208         if (status == 0)
 209                 nfs4_schedule_state_renewal(clp);
 210         return status;
 211 }
 212
 213 int nfs4_init_client(struct nfs4_client *clp)
 214 {
 215         return nfs4_map_errors(__nfs4_init_client(clp));
 216 }
 217
 218 u32
 219 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 220 {
 221         return clp->cl_lockowner_id ++;
 222 }
 223
 224 static struct nfs4_state_owner *
 225 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 226 {
 227         struct nfs4_state_owner *sp = NULL;
 228
 229         if (!list_empty(&clp->cl_unused)) {
 230                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 231                 atomic_inc(&sp->so_count);
 232                 sp->so_cred = cred;
 233                 list_move(&sp->so_list, &clp->cl_state_owners);
 234                 clp->cl_nunused--;
 235         }
 236         return sp;
 237 }
 238
 239 static struct nfs4_state_owner *
 240 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 241 {
 242         struct nfs4_state_owner *sp, *res = NULL;
 243
 244         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 245                 if (sp->so_cred != cred)
 246                         continue;
 247                 atomic_inc(&sp->so_count);
 248                 /* Move to the head of the list */
 249                 list_move(&sp->so_list, &clp->cl_state_owners);
 250                 res = sp;
 251                 break;
 252         }
 253         return res;
 254 }
 255
 256 /*
 257  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 258  * create a new state_owner.
 259  *
 260  */
 261 static struct nfs4_state_owner *
 262 nfs4_alloc_state_owner(void)
 263 {
 264         struct nfs4_state_owner *sp;
 265
 266         sp = kmalloc(sizeof(*sp),GFP_KERNEL);
 267         if (!sp)
 268                 return NULL;
 269         init_MUTEX(&sp->so_sema);
 270         sp->so_seqid = 0;                 /* arbitrary */
 271         INIT_LIST_HEAD(&sp->so_states);
 272         INIT_LIST_HEAD(&sp->so_delegations);
 273         atomic_set(&sp->so_count, 1);
 274         return sp;
 275 }
 276
 277 void
 278 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 279 {
 280         struct nfs4_client *clp = sp->so_client;
 281         spin_lock(&clp->cl_lock);
 282         list_del_init(&sp->so_list);
 283         spin_unlock(&clp->cl_lock);
 284 }
 285
 286 /*
 287  * Note: must be called with clp->cl_sem held in order to prevent races
 288  *       with reboot recovery!
 289  */
 290 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 291 {
 292         struct nfs4_client *clp = server->nfs4_state;
 293         struct nfs4_state_owner *sp, *new;
 294
 295         get_rpccred(cred);
 296         new = nfs4_alloc_state_owner();
 297         spin_lock(&clp->cl_lock);
 298         sp = nfs4_find_state_owner(clp, cred);
 299         if (sp == NULL)
 300                 sp = nfs4_client_grab_unused(clp, cred);
 301         if (sp == NULL && new != NULL) {
 302                 list_add(&new->so_list, &clp->cl_state_owners);
 303                 new->so_client = clp;
 304                 new->so_id = nfs4_alloc_lockowner_id(clp);
 305                 new->so_cred = cred;
 306                 sp = new;
 307                 new = NULL;
 308         }
 309         spin_unlock(&clp->cl_lock);
 310         if (new)
 311                 kfree(new);
 312         if (sp != NULL)
 313                 return sp;
 314         put_rpccred(cred);
 315         return NULL;
 316 }
 317
 318 /*
 319  * Must be called with clp->cl_sem held in order to avoid races
 320  * with state recovery...
 321  */
 322 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 323 {
 324         struct nfs4_client *clp = sp->so_client;
 325         struct rpc_cred *cred = sp->so_cred;
 326
 327         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 328                 return;
 329         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 330                 goto out_free;
 331         if (list_empty(&sp->so_list))
 332                 goto out_free;
 333         list_move(&sp->so_list, &clp->cl_unused);
 334         clp->cl_nunused++;
 335         spin_unlock(&clp->cl_lock);
 336         put_rpccred(cred);
 337         cred = NULL;
 338         return;
 339 out_free:
 340         list_del(&sp->so_list);
 341         spin_unlock(&clp->cl_lock);
 342         put_rpccred(cred);
 343         kfree(sp);
 344 }
 345
 346 static struct nfs4_state *
 347 nfs4_alloc_open_state(void)
 348 {
 349         struct nfs4_state *state;
 350
 351         state = kmalloc(sizeof(*state), GFP_KERNEL);
 352         if (!state)
 353                 return NULL;
 354         state->state = 0;
 355         state->nreaders = 0;
 356         state->nwriters = 0;
 357         state->flags = 0;
 358         memset(state->stateid.data, 0, sizeof(state->stateid.data));
 359         atomic_set(&state->count, 1);
 360         INIT_LIST_HEAD(&state->lock_states);
 361         init_MUTEX(&state->lock_sema);
 362         rwlock_init(&state->state_lock);
 363         return state;
 364 }
 365
 366 static struct nfs4_state *
 367 __nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 368 {
 369         struct nfs_inode *nfsi = NFS_I(inode);
 370         struct nfs4_state *state;
 371
 372         mode &= (FMODE_READ|FMODE_WRITE);
 373         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 374                 if (state->owner->so_cred != cred)
 375                         continue;
 376                 if ((mode & FMODE_READ) != 0 && state->nreaders == 0)
 377                         continue;
 378                 if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0)
 379                         continue;
 380                 if ((state->state & mode) != mode)
 381                         continue;
 382                 atomic_inc(&state->count);
 383                 if (mode & FMODE_READ)
 384                         state->nreaders++;
 385                 if (mode & FMODE_WRITE)
 386                         state->nwriters++;
 387                 return state;
 388         }
 389         return NULL;
 390 }
 391
 392 static struct nfs4_state *
 393 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 394 {
 395         struct nfs_inode *nfsi = NFS_I(inode);
 396         struct nfs4_state *state;
 397
 398         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 399                 /* Is this in the process of being freed? */
 400                 if (state->nreaders == 0 && state->nwriters == 0)
 401                         continue;
 402                 if (state->owner == owner) {
 403                         atomic_inc(&state->count);
 404                         return state;
 405                 }
 406         }
 407         return NULL;
 408 }
 409
 410 struct nfs4_state *
 411 nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode)
 412 {
 413         struct nfs4_state *state;
 414
 415         spin_lock(&inode->i_lock);
 416         state = __nfs4_find_state(inode, cred, mode);
 417         spin_unlock(&inode->i_lock);
 418         return state;
 419 }
 420
 421 static void
 422 nfs4_free_open_state(struct nfs4_state *state)
 423 {
 424         kfree(state);
 425 }
 426
 427 struct nfs4_state *
 428 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 429 {
 430         struct nfs4_state *state, *new;
 431         struct nfs_inode *nfsi = NFS_I(inode);
 432
 433         spin_lock(&inode->i_lock);
 434         state = __nfs4_find_state_byowner(inode, owner);
 435         spin_unlock(&inode->i_lock);
 436         if (state)
 437                 goto out;
 438         new = nfs4_alloc_open_state();
 439         spin_lock(&inode->i_lock);
 440         state = __nfs4_find_state_byowner(inode, owner);
 441         if (state == NULL && new != NULL) {
 442                 state = new;
 443                 /* Caller *must* be holding owner->so_sem */
 444                 /* Note: The reclaim code dictates that we add stateless
 445                  * and read-only stateids to the end of the list */
 446                 list_add_tail(&state->open_states, &owner->so_states);
 447                 state->owner = owner;
 448                 atomic_inc(&owner->so_count);
 449                 list_add(&state->inode_states, &nfsi->open_states);
 450                 state->inode = igrab(inode);
 451                 spin_unlock(&inode->i_lock);
 452         } else {
 453                 spin_unlock(&inode->i_lock);
 454                 if (new)
 455                         nfs4_free_open_state(new);
 456         }
 457 out:
 458         return state;
 459 }
 460
 461 /*
 462  * Beware! Caller must be holding exactly one
 463  * reference to clp->cl_sem and owner->so_sema!
 464  */
 465 void nfs4_put_open_state(struct nfs4_state *state)
 466 {
 467         struct inode *inode = state->inode;
 468         struct nfs4_state_owner *owner = state->owner;
 469
 470         if (!atomic_dec_and_lock(&state->count, &inode->i_lock))
 471                 return;
 472         if (!list_empty(&state->inode_states))
 473                 list_del(&state->inode_states);
 474         spin_unlock(&inode->i_lock);
 475         list_del(&state->open_states);
 476         iput(inode);
 477         BUG_ON (state->state != 0);
 478         nfs4_free_open_state(state);
 479         nfs4_put_state_owner(owner);
 480 }
 481
 482 /*
 483  * Beware! Caller must be holding no references to clp->cl_sem!
 484  * of owner->so_sema!
 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         struct nfs4_client *clp = owner->so_client;
 491         int newstate;
 492
 493         atomic_inc(&owner->so_count);
 494         down_read(&clp->cl_sem);
 495         down(&owner->so_sema);
 496         /* Protect against nfs4_find_state() */
 497         spin_lock(&inode->i_lock);
 498         if (mode & FMODE_READ)
 499                 state->nreaders--;
 500         if (mode & FMODE_WRITE)
 501                 state->nwriters--;
 502         if (state->nwriters == 0) {
 503                 if (state->nreaders == 0)
 504                         list_del_init(&state->inode_states);
 505                 /* See reclaim code */
 506                 list_move_tail(&state->open_states, &owner->so_states);
 507         }
 508         spin_unlock(&inode->i_lock);
 509         newstate = 0;
 510         if (state->state != 0) {
 511                 if (state->nreaders)
 512                         newstate |= FMODE_READ;
 513                 if (state->nwriters)
 514                         newstate |= FMODE_WRITE;
 515                 if (state->state == newstate)
 516                         goto out;
 517                 if (nfs4_do_close(inode, state, newstate) == -EINPROGRESS)
 518                         return;
 519         }
 520 out:
 521         nfs4_put_open_state(state);
 522         up(&owner->so_sema);
 523         nfs4_put_state_owner(owner);
 524         up_read(&clp->cl_sem);
 525 }
 526
 527 /*
 528  * Search the state->lock_states for an existing lock_owner
 529  * that is compatible with current->files
 530  */
 531 static struct nfs4_lock_state *
 532 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 533 {
 534         struct nfs4_lock_state *pos;
 535         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 536                 if (pos->ls_owner != fl_owner)
 537                         continue;
 538                 atomic_inc(&pos->ls_count);
 539                 return pos;
 540         }
 541         return NULL;
 542 }
 543
 544 struct nfs4_lock_state *
 545 nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 546 {
 547         struct nfs4_lock_state *lsp;
 548         read_lock(&state->state_lock);
 549         lsp = __nfs4_find_lock_state(state, fl_owner);
 550         read_unlock(&state->state_lock);
 551         return lsp;
 552 }
 553
 554 /*
 555  * Return a compatible lock_state. If no initialized lock_state structure
 556  * exists, return an uninitialized one.
 557  *
 558  * The caller must be holding state->lock_sema
 559  */
 560 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 561 {
 562         struct nfs4_lock_state *lsp;
 563         struct nfs4_client *clp = state->owner->so_client;
 564
 565         lsp = kmalloc(sizeof(*lsp), GFP_KERNEL);
 566         if (lsp == NULL)
 567                 return NULL;
 568         lsp->ls_flags = 0;
 569         lsp->ls_seqid = 0;      /* arbitrary */
 570         lsp->ls_id = -1;
 571         memset(lsp->ls_stateid.data, 0, sizeof(lsp->ls_stateid.data));
 572         atomic_set(&lsp->ls_count, 1);
 573         lsp->ls_owner = fl_owner;
 574         INIT_LIST_HEAD(&lsp->ls_locks);
 575         spin_lock(&clp->cl_lock);
 576         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 577         spin_unlock(&clp->cl_lock);
 578         return lsp;
 579 }
 580
 581 /*
 582  * Return a compatible lock_state. If no initialized lock_state structure
 583  * exists, return an uninitialized one.
 584  *
 585  * The caller must be holding state->lock_sema and clp->cl_sem
 586  */
 587 struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 588 {
 589         struct nfs4_lock_state * lsp;
 590
 591         lsp = nfs4_find_lock_state(state, owner);
 592         if (lsp == NULL)
 593                 lsp = nfs4_alloc_lock_state(state, owner);
 594         return lsp;
 595 }
 596
 597 /*
 598  * Byte-range lock aware utility to initialize the stateid of read/write
 599  * requests.
 600  */
 601 void
 602 nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 603 {
 604         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
 605                 struct nfs4_lock_state *lsp;
 606
 607                 lsp = nfs4_find_lock_state(state, fl_owner);
 608                 if (lsp) {
 609                         memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 610                         nfs4_put_lock_state(lsp);
 611                         return;
 612                 }
 613         }
 614         memcpy(dst, &state->stateid, sizeof(*dst));
 615 }
 616
 617 /*
 618 * Called with state->lock_sema and clp->cl_sem held.
 619 */
 620 void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *lsp)
 621 {
 622         if (status == NFS_OK || seqid_mutating_err(-status))
 623                 lsp->ls_seqid++;
 624 }
 625
 626 /*
 627 * Check to see if the request lock (type FL_UNLK) effects the fl lock.
 628 *
 629 * fl and request must have the same posix owner
 630 *
 631 * return:
 632 * 0 -> fl not effected by request
 633 * 1 -> fl consumed by request
 634 */
 635
 636 static int
 637 nfs4_check_unlock(struct file_lock *fl, struct file_lock *request)
 638 {
 639         if (fl->fl_start >= request->fl_start && fl->fl_end <= request->fl_end)
 640                 return 1;
 641         return 0;
 642 }
 643
 644 /*
 645  * Post an initialized lock_state on the state->lock_states list.
 646  */
 647 void nfs4_notify_setlk(struct nfs4_state *state, struct file_lock *request, struct nfs4_lock_state *lsp)
 648 {
 649         if (!list_empty(&lsp->ls_locks))
 650                 return;
 651         atomic_inc(&lsp->ls_count);
 652         write_lock(&state->state_lock);
 653         list_add(&lsp->ls_locks, &state->lock_states);
 654         set_bit(LK_STATE_IN_USE, &state->flags);
 655         write_unlock(&state->state_lock);
 656 }
 657
 658 /*
 659  * to decide to 'reap' lock state:
 660  * 1) search i_flock for file_locks with fl.lock_state = to ls.
 661  * 2) determine if unlock will consume found lock.
 662  *      if so, reap
 663  *
 664  *      else, don't reap.
 665  *
 666  */
 667 void
 668 nfs4_notify_unlck(struct nfs4_state *state, struct file_lock *request, struct nfs4_lock_state *lsp)
 669 {
 670         struct inode *inode = state->inode;
 671         struct file_lock *fl;
 672
 673         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 674                 if (!(fl->fl_flags & FL_POSIX))
 675                         continue;
 676                 if (fl->fl_owner != lsp->ls_owner)
 677                         continue;
 678                 /* Exit if we find at least one lock which is not consumed */
 679                 if (nfs4_check_unlock(fl,request) == 0)
 680                         return;
 681         }
 682
 683         write_lock(&state->state_lock);
 684         list_del_init(&lsp->ls_locks);
 685         if (list_empty(&state->lock_states))
 686                 clear_bit(LK_STATE_IN_USE, &state->flags);
 687         write_unlock(&state->state_lock);
 688         nfs4_put_lock_state(lsp);
 689 }
 690
 691 /*
 692  * Release reference to lock_state, and free it if we see that
 693  * it is no longer in use
 694  */
 695 void
 696 nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 697 {
 698         if (!atomic_dec_and_test(&lsp->ls_count))
 699                 return;
 700         BUG_ON (!list_empty(&lsp->ls_locks));
 701         kfree(lsp);
 702 }
 703
 704 /*
 705 * Called with sp->so_sema and clp->cl_sem held.
 706 *
 707 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 708 * failed with a seqid incrementing error -
 709 * see comments nfs_fs.h:seqid_mutating_error()
 710 */
 711 void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp)
 712 {
 713         if (status == NFS_OK || seqid_mutating_err(-status))
 714                 sp->so_seqid++;
 715         /* If the server returns BAD_SEQID, unhash state_owner here */
 716         if (status == -NFS4ERR_BAD_SEQID)
 717                 nfs4_drop_state_owner(sp);
 718 }
 719
 720 static int reclaimer(void *);
 721 struct reclaimer_args {
 722         struct nfs4_client *clp;
 723         struct completion complete;
 724 };
 725
 726 /*
 727  * State recovery routine
 728  */
 729 void
 730 nfs4_recover_state(void *data)
 731 {
 732         struct nfs4_client *clp = (struct nfs4_client *)data;
 733         struct reclaimer_args args = {
 734                 .clp = clp,
 735         };
 736         might_sleep();
 737
 738         init_completion(&args.complete);
 739
 740         if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0)
 741                 goto out_failed_clear;
 742         wait_for_completion(&args.complete);
 743         return;
 744 out_failed_clear:
 745         set_bit(NFS4CLNT_OK, &clp->cl_state);
 746         wake_up_all(&clp->cl_waitq);
 747         rpc_wake_up(&clp->cl_rpcwaitq);
 748 }
 749
 750 /*
 751  * Schedule a state recovery attempt
 752  */
 753 void
 754 nfs4_schedule_state_recovery(struct nfs4_client *clp)
 755 {
 756         if (!clp)
 757                 return;
 758         if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state))
 759                 schedule_work(&clp->cl_recoverd);
 760 }
 761
 762 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 763 {
 764         struct inode *inode = state->inode;
 765         struct file_lock *fl;
 766         int status = 0;
 767
 768         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 769                 if (!(fl->fl_flags & FL_POSIX))
 770                         continue;
 771                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 772                         continue;
 773                 status = ops->recover_lock(state, fl);
 774                 if (status >= 0)
 775                         continue;
 776                 switch (status) {
 777                         default:
 778                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 779                                                 __FUNCTION__, status);
 780                         case -NFS4ERR_EXPIRED:
 781                         case -NFS4ERR_NO_GRACE:
 782                         case -NFS4ERR_RECLAIM_BAD:
 783                         case -NFS4ERR_RECLAIM_CONFLICT:
 784                                 /* kill_proc(fl->fl_owner, SIGLOST, 1); */
 785                                 break;
 786                         case -NFS4ERR_STALE_CLIENTID:
 787                                 goto out_err;
 788                 }
 789         }
 790         return 0;
 791 out_err:
 792         return status;
 793 }
 794
 795 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 796 {
 797         struct nfs4_state *state;
 798         struct nfs4_lock_state *lock;
 799         int status = 0;
 800
 801         /* Note: we rely on the sp->so_states list being ordered
 802          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 803          * states first.
 804          * This is needed to ensure that the server won't give us any
 805          * read delegations that we have to return if, say, we are
 806          * recovering after a network partition or a reboot from a
 807          * server that doesn't support a grace period.
 808          */
 809         list_for_each_entry(state, &sp->so_states, open_states) {
 810                 if (state->state == 0)
 811                         continue;
 812                 status = ops->recover_open(sp, state);
 813                 list_for_each_entry(lock, &state->lock_states, ls_locks)
 814                         lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 815                 if (status >= 0) {
 816                         status = nfs4_reclaim_locks(ops, state);
 817                         if (status < 0)
 818                                 goto out_err;
 819                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 820                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 821                                         printk("%s: Lock reclaim failed!\n",
 822                                                         __FUNCTION__);
 823                         }
 824                         continue;
 825                 }
 826                 switch (status) {
 827                         default:
 828                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 829                                                 __FUNCTION__, status);
 830                         case -ENOENT:
 831                         case -NFS4ERR_RECLAIM_BAD:
 832                         case -NFS4ERR_RECLAIM_CONFLICT:
 833                                 /*
 834                                  * Open state on this file cannot be recovered
 835                                  * All we can do is revert to using the zero stateid.
 836                                  */
 837                                 memset(state->stateid.data, 0,
 838                                         sizeof(state->stateid.data));
 839                                 /* Mark the file as being 'closed' */
 840                                 state->state = 0;
 841                                 break;
 842                         case -NFS4ERR_EXPIRED:
 843                         case -NFS4ERR_NO_GRACE:
 844                         case -NFS4ERR_STALE_CLIENTID:
 845                                 goto out_err;
 846                 }
 847         }
 848         return 0;
 849 out_err:
 850         return status;
 851 }
 852
 853 static int reclaimer(void *ptr)
 854 {
 855         struct reclaimer_args *args = (struct reclaimer_args *)ptr;
 856         struct nfs4_client *clp = args->clp;
 857         struct nfs4_state_owner *sp;
 858         struct nfs4_state_recovery_ops *ops;
 859         int status = 0;
 860
 861         daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr));
 862         allow_signal(SIGKILL);
 863
 864         atomic_inc(&clp->cl_count);
 865         complete(&args->complete);
 866
 867         /* Ensure exclusive access to NFSv4 state */
 868         lock_kernel();
 869         down_write(&clp->cl_sem);
 870         /* Are there any NFS mounts out there? */
 871         if (list_empty(&clp->cl_superblocks))
 872                 goto out;
 873 restart_loop:
 874         status = nfs4_proc_renew(clp);
 875         switch (status) {
 876                 case 0:
 877                 case -NFS4ERR_CB_PATH_DOWN:
 878                         goto out;
 879                 case -NFS4ERR_STALE_CLIENTID:
 880                 case -NFS4ERR_LEASE_MOVED:
 881                         ops = &nfs4_reboot_recovery_ops;
 882                         break;
 883                 default:
 884                         ops = &nfs4_network_partition_recovery_ops;
 885         };
 886         status = __nfs4_init_client(clp);
 887         if (status)
 888                 goto out_error;
 889         /* Mark all delegations for reclaim */
 890         nfs_delegation_mark_reclaim(clp);
 891         /* Note: list is protected by exclusive lock on cl->cl_sem */
 892         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 893                 status = nfs4_reclaim_open_state(ops, sp);
 894                 if (status < 0) {
 895                         if (status == -NFS4ERR_NO_GRACE) {
 896                                 ops = &nfs4_network_partition_recovery_ops;
 897                                 status = nfs4_reclaim_open_state(ops, sp);
 898                         }
 899                         if (status == -NFS4ERR_STALE_CLIENTID)
 900                                 goto restart_loop;
 901                         if (status == -NFS4ERR_EXPIRED)
 902                                 goto restart_loop;
 903                 }
 904         }
 905         nfs_delegation_reap_unclaimed(clp);
 906 out:
 907         set_bit(NFS4CLNT_OK, &clp->cl_state);
 908         up_write(&clp->cl_sem);
 909         unlock_kernel();
 910         wake_up_all(&clp->cl_waitq);
 911         rpc_wake_up(&clp->cl_rpcwaitq);
 912         if (status == -NFS4ERR_CB_PATH_DOWN)
 913                 nfs_handle_cb_pathdown(clp);
 914         nfs4_put_client(clp);
 915         return 0;
 916 out_error:
 917         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 918                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 919         goto out;
 920 }
 921
 922 /*
 923  * Local variables:
 924  *  c-basic-offset: 8
 925  * End:
 926  */