Pileus Git - ~andy/linux/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/slab.h>
  42 #include <linux/smp_lock.h>
  43 #include <linux/nfs_fs.h>
  44 #include <linux/nfs_idmap.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/workqueue.h>
  48 #include <linux/bitops.h>
  49
  50 #include "nfs4_fs.h"
  51 #include "callback.h"
  52 #include "delegation.h"
  53 #include "internal.h"
  54
  55 #define OPENOWNER_POOL_SIZE     8
  56
  57 const nfs4_stateid zero_stateid;
  58
  59 static LIST_HEAD(nfs4_clientid_list);
  60
  61 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
  62 {
  63         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
  64                         nfs_callback_tcpport, cred);
  65         if (status == 0)
  66                 status = nfs4_proc_setclientid_confirm(clp, cred);
  67         if (status == 0)
  68                 nfs4_schedule_state_renewal(clp);
  69         return status;
  70 }
  71
  72 u32
  73 nfs4_alloc_lockowner_id(struct nfs_client *clp)
  74 {
  75         return clp->cl_lockowner_id ++;
  76 }
  77
  78 static struct nfs4_state_owner *
  79 nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred)
  80 {
  81         struct nfs4_state_owner *sp = NULL;
  82
  83         if (!list_empty(&clp->cl_unused)) {
  84                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
  85                 atomic_inc(&sp->so_count);
  86                 sp->so_cred = cred;
  87                 list_move(&sp->so_list, &clp->cl_state_owners);
  88                 clp->cl_nunused--;
  89         }
  90         return sp;
  91 }
  92
  93 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
  94 {
  95         struct nfs4_state_owner *sp;
  96         struct rpc_cred *cred = NULL;
  97
  98         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
  99                 if (list_empty(&sp->so_states))
 100                         continue;
 101                 cred = get_rpccred(sp->so_cred);
 102                 break;
 103         }
 104         return cred;
 105 }
 106
 107 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 108 {
 109         struct nfs4_state_owner *sp;
 110
 111         if (!list_empty(&clp->cl_state_owners)) {
 112                 sp = list_entry(clp->cl_state_owners.next,
 113                                 struct nfs4_state_owner, so_list);
 114                 return get_rpccred(sp->so_cred);
 115         }
 116         return NULL;
 117 }
 118
 119 static struct nfs4_state_owner *
 120 nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred)
 121 {
 122         struct nfs4_state_owner *sp, *res = NULL;
 123
 124         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 125                 if (sp->so_cred != cred)
 126                         continue;
 127                 atomic_inc(&sp->so_count);
 128                 /* Move to the head of the list */
 129                 list_move(&sp->so_list, &clp->cl_state_owners);
 130                 res = sp;
 131                 break;
 132         }
 133         return res;
 134 }
 135
 136 /*
 137  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 138  * create a new state_owner.
 139  *
 140  */
 141 static struct nfs4_state_owner *
 142 nfs4_alloc_state_owner(void)
 143 {
 144         struct nfs4_state_owner *sp;
 145
 146         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 147         if (!sp)
 148                 return NULL;
 149         spin_lock_init(&sp->so_lock);
 150         INIT_LIST_HEAD(&sp->so_states);
 151         INIT_LIST_HEAD(&sp->so_delegations);
 152         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 153         sp->so_seqid.sequence = &sp->so_sequence;
 154         spin_lock_init(&sp->so_sequence.lock);
 155         INIT_LIST_HEAD(&sp->so_sequence.list);
 156         atomic_set(&sp->so_count, 1);
 157         return sp;
 158 }
 159
 160 void
 161 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 162 {
 163         struct nfs_client *clp = sp->so_client;
 164         spin_lock(&clp->cl_lock);
 165         list_del_init(&sp->so_list);
 166         spin_unlock(&clp->cl_lock);
 167 }
 168
 169 /*
 170  * Note: must be called with clp->cl_sem held in order to prevent races
 171  *       with reboot recovery!
 172  */
 173 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 174 {
 175         struct nfs_client *clp = server->nfs_client;
 176         struct nfs4_state_owner *sp, *new;
 177
 178         get_rpccred(cred);
 179         new = nfs4_alloc_state_owner();
 180         spin_lock(&clp->cl_lock);
 181         sp = nfs4_find_state_owner(clp, cred);
 182         if (sp == NULL)
 183                 sp = nfs4_client_grab_unused(clp, cred);
 184         if (sp == NULL && new != NULL) {
 185                 list_add(&new->so_list, &clp->cl_state_owners);
 186                 new->so_client = clp;
 187                 new->so_id = nfs4_alloc_lockowner_id(clp);
 188                 new->so_cred = cred;
 189                 sp = new;
 190                 new = NULL;
 191         }
 192         spin_unlock(&clp->cl_lock);
 193         kfree(new);
 194         if (sp != NULL)
 195                 return sp;
 196         put_rpccred(cred);
 197         return NULL;
 198 }
 199
 200 /*
 201  * Must be called with clp->cl_sem held in order to avoid races
 202  * with state recovery...
 203  */
 204 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 205 {
 206         struct nfs_client *clp = sp->so_client;
 207         struct rpc_cred *cred = sp->so_cred;
 208
 209         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 210                 return;
 211         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 212                 goto out_free;
 213         if (list_empty(&sp->so_list))
 214                 goto out_free;
 215         list_move(&sp->so_list, &clp->cl_unused);
 216         clp->cl_nunused++;
 217         spin_unlock(&clp->cl_lock);
 218         put_rpccred(cred);
 219         cred = NULL;
 220         return;
 221 out_free:
 222         list_del(&sp->so_list);
 223         spin_unlock(&clp->cl_lock);
 224         put_rpccred(cred);
 225         kfree(sp);
 226 }
 227
 228 static struct nfs4_state *
 229 nfs4_alloc_open_state(void)
 230 {
 231         struct nfs4_state *state;
 232
 233         state = kzalloc(sizeof(*state), GFP_KERNEL);
 234         if (!state)
 235                 return NULL;
 236         atomic_set(&state->count, 1);
 237         INIT_LIST_HEAD(&state->lock_states);
 238         spin_lock_init(&state->state_lock);
 239         return state;
 240 }
 241
 242 void
 243 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 244 {
 245         if (state->state == mode)
 246                 return;
 247         /* NB! List reordering - see the reclaim code for why.  */
 248         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 249                 if (mode & FMODE_WRITE)
 250                         list_move(&state->open_states, &state->owner->so_states);
 251                 else
 252                         list_move_tail(&state->open_states, &state->owner->so_states);
 253         }
 254         if (mode == 0)
 255                 list_del_init(&state->inode_states);
 256         state->state = mode;
 257 }
 258
 259 static struct nfs4_state *
 260 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 261 {
 262         struct nfs_inode *nfsi = NFS_I(inode);
 263         struct nfs4_state *state;
 264
 265         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 266                 /* Is this in the process of being freed? */
 267                 if (state->state == 0)
 268                         continue;
 269                 if (state->owner == owner) {
 270                         atomic_inc(&state->count);
 271                         return state;
 272                 }
 273         }
 274         return NULL;
 275 }
 276
 277 static void
 278 nfs4_free_open_state(struct nfs4_state *state)
 279 {
 280         kfree(state);
 281 }
 282
 283 struct nfs4_state *
 284 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 285 {
 286         struct nfs4_state *state, *new;
 287         struct nfs_inode *nfsi = NFS_I(inode);
 288
 289         spin_lock(&inode->i_lock);
 290         state = __nfs4_find_state_byowner(inode, owner);
 291         spin_unlock(&inode->i_lock);
 292         if (state)
 293                 goto out;
 294         new = nfs4_alloc_open_state();
 295         spin_lock(&owner->so_lock);
 296         spin_lock(&inode->i_lock);
 297         state = __nfs4_find_state_byowner(inode, owner);
 298         if (state == NULL && new != NULL) {
 299                 state = new;
 300                 state->owner = owner;
 301                 atomic_inc(&owner->so_count);
 302                 list_add(&state->inode_states, &nfsi->open_states);
 303                 state->inode = igrab(inode);
 304                 spin_unlock(&inode->i_lock);
 305                 /* Note: The reclaim code dictates that we add stateless
 306                  * and read-only stateids to the end of the list */
 307                 list_add_tail(&state->open_states, &owner->so_states);
 308                 spin_unlock(&owner->so_lock);
 309         } else {
 310                 spin_unlock(&inode->i_lock);
 311                 spin_unlock(&owner->so_lock);
 312                 if (new)
 313                         nfs4_free_open_state(new);
 314         }
 315 out:
 316         return state;
 317 }
 318
 319 /*
 320  * Beware! Caller must be holding exactly one
 321  * reference to clp->cl_sem!
 322  */
 323 void nfs4_put_open_state(struct nfs4_state *state)
 324 {
 325         struct inode *inode = state->inode;
 326         struct nfs4_state_owner *owner = state->owner;
 327
 328         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 329                 return;
 330         spin_lock(&inode->i_lock);
 331         if (!list_empty(&state->inode_states))
 332                 list_del(&state->inode_states);
 333         list_del(&state->open_states);
 334         spin_unlock(&inode->i_lock);
 335         spin_unlock(&owner->so_lock);
 336         iput(inode);
 337         nfs4_free_open_state(state);
 338         nfs4_put_state_owner(owner);
 339 }
 340
 341 /*
 342  * Close the current file.
 343  */
 344 void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
 345 {
 346         struct inode *inode = state->inode;
 347         struct nfs4_state_owner *owner = state->owner;
 348         int oldstate, newstate = 0;
 349
 350         atomic_inc(&owner->so_count);
 351         /* Protect against nfs4_find_state() */
 352         spin_lock(&owner->so_lock);
 353         spin_lock(&inode->i_lock);
 354         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 355                 case FMODE_READ:
 356                         state->n_rdonly--;
 357                         break;
 358                 case FMODE_WRITE:
 359                         state->n_wronly--;
 360                         break;
 361                 case FMODE_READ|FMODE_WRITE:
 362                         state->n_rdwr--;
 363         }
 364         oldstate = newstate = state->state;
 365         if (state->n_rdwr == 0) {
 366                 if (state->n_rdonly == 0)
 367                         newstate &= ~FMODE_READ;
 368                 if (state->n_wronly == 0)
 369                         newstate &= ~FMODE_WRITE;
 370         }
 371         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 372                 nfs4_state_set_mode_locked(state, newstate);
 373                 oldstate = newstate;
 374         }
 375         spin_unlock(&inode->i_lock);
 376         spin_unlock(&owner->so_lock);
 377
 378         if (oldstate == newstate) {
 379                 nfs4_put_open_state(state);
 380                 nfs4_put_state_owner(owner);
 381         } else
 382                 nfs4_do_close(path, state);
 383 }
 384
 385 /*
 386  * Search the state->lock_states for an existing lock_owner
 387  * that is compatible with current->files
 388  */
 389 static struct nfs4_lock_state *
 390 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 391 {
 392         struct nfs4_lock_state *pos;
 393         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 394                 if (pos->ls_owner != fl_owner)
 395                         continue;
 396                 atomic_inc(&pos->ls_count);
 397                 return pos;
 398         }
 399         return NULL;
 400 }
 401
 402 /*
 403  * Return a compatible lock_state. If no initialized lock_state structure
 404  * exists, return an uninitialized one.
 405  *
 406  */
 407 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 408 {
 409         struct nfs4_lock_state *lsp;
 410         struct nfs_client *clp = state->owner->so_client;
 411
 412         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 413         if (lsp == NULL)
 414                 return NULL;
 415         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 416         atomic_set(&lsp->ls_count, 1);
 417         lsp->ls_owner = fl_owner;
 418         spin_lock(&clp->cl_lock);
 419         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 420         spin_unlock(&clp->cl_lock);
 421         INIT_LIST_HEAD(&lsp->ls_locks);
 422         return lsp;
 423 }
 424
 425 /*
 426  * Return a compatible lock_state. If no initialized lock_state structure
 427  * exists, return an uninitialized one.
 428  *
 429  * The caller must be holding clp->cl_sem
 430  */
 431 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 432 {
 433         struct nfs4_lock_state *lsp, *new = NULL;
 434
 435         for(;;) {
 436                 spin_lock(&state->state_lock);
 437                 lsp = __nfs4_find_lock_state(state, owner);
 438                 if (lsp != NULL)
 439                         break;
 440                 if (new != NULL) {
 441                         new->ls_state = state;
 442                         list_add(&new->ls_locks, &state->lock_states);
 443                         set_bit(LK_STATE_IN_USE, &state->flags);
 444                         lsp = new;
 445                         new = NULL;
 446                         break;
 447                 }
 448                 spin_unlock(&state->state_lock);
 449                 new = nfs4_alloc_lock_state(state, owner);
 450                 if (new == NULL)
 451                         return NULL;
 452         }
 453         spin_unlock(&state->state_lock);
 454         kfree(new);
 455         return lsp;
 456 }
 457
 458 /*
 459  * Release reference to lock_state, and free it if we see that
 460  * it is no longer in use
 461  */
 462 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 463 {
 464         struct nfs4_state *state;
 465
 466         if (lsp == NULL)
 467                 return;
 468         state = lsp->ls_state;
 469         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 470                 return;
 471         list_del(&lsp->ls_locks);
 472         if (list_empty(&state->lock_states))
 473                 clear_bit(LK_STATE_IN_USE, &state->flags);
 474         spin_unlock(&state->state_lock);
 475         kfree(lsp);
 476 }
 477
 478 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 479 {
 480         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 481
 482         dst->fl_u.nfs4_fl.owner = lsp;
 483         atomic_inc(&lsp->ls_count);
 484 }
 485
 486 static void nfs4_fl_release_lock(struct file_lock *fl)
 487 {
 488         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 489 }
 490
 491 static struct file_lock_operations nfs4_fl_lock_ops = {
 492         .fl_copy_lock = nfs4_fl_copy_lock,
 493         .fl_release_private = nfs4_fl_release_lock,
 494 };
 495
 496 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 497 {
 498         struct nfs4_lock_state *lsp;
 499
 500         if (fl->fl_ops != NULL)
 501                 return 0;
 502         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 503         if (lsp == NULL)
 504                 return -ENOMEM;
 505         fl->fl_u.nfs4_fl.owner = lsp;
 506         fl->fl_ops = &nfs4_fl_lock_ops;
 507         return 0;
 508 }
 509
 510 /*
 511  * Byte-range lock aware utility to initialize the stateid of read/write
 512  * requests.
 513  */
 514 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 515 {
 516         struct nfs4_lock_state *lsp;
 517
 518         memcpy(dst, &state->stateid, sizeof(*dst));
 519         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 520                 return;
 521
 522         spin_lock(&state->state_lock);
 523         lsp = __nfs4_find_lock_state(state, fl_owner);
 524         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 525                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 526         spin_unlock(&state->state_lock);
 527         nfs4_put_lock_state(lsp);
 528 }
 529
 530 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 531 {
 532         struct rpc_sequence *sequence = counter->sequence;
 533         struct nfs_seqid *new;
 534
 535         new = kmalloc(sizeof(*new), GFP_KERNEL);
 536         if (new != NULL) {
 537                 new->sequence = counter;
 538                 spin_lock(&sequence->lock);
 539                 list_add_tail(&new->list, &sequence->list);
 540                 spin_unlock(&sequence->lock);
 541         }
 542         return new;
 543 }
 544
 545 void nfs_free_seqid(struct nfs_seqid *seqid)
 546 {
 547         struct rpc_sequence *sequence = seqid->sequence->sequence;
 548
 549         spin_lock(&sequence->lock);
 550         list_del(&seqid->list);
 551         spin_unlock(&sequence->lock);
 552         rpc_wake_up(&sequence->wait);
 553         kfree(seqid);
 554 }
 555
 556 /*
 557  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 558  * failed with a seqid incrementing error -
 559  * see comments nfs_fs.h:seqid_mutating_error()
 560  */
 561 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 562 {
 563         switch (status) {
 564                 case 0:
 565                         break;
 566                 case -NFS4ERR_BAD_SEQID:
 567                 case -NFS4ERR_STALE_CLIENTID:
 568                 case -NFS4ERR_STALE_STATEID:
 569                 case -NFS4ERR_BAD_STATEID:
 570                 case -NFS4ERR_BADXDR:
 571                 case -NFS4ERR_RESOURCE:
 572                 case -NFS4ERR_NOFILEHANDLE:
 573                         /* Non-seqid mutating errors */
 574                         return;
 575         };
 576         /*
 577          * Note: no locking needed as we are guaranteed to be first
 578          * on the sequence list
 579          */
 580         seqid->sequence->counter++;
 581 }
 582
 583 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 584 {
 585         if (status == -NFS4ERR_BAD_SEQID) {
 586                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 587                                 struct nfs4_state_owner, so_seqid);
 588                 nfs4_drop_state_owner(sp);
 589         }
 590         return nfs_increment_seqid(status, seqid);
 591 }
 592
 593 /*
 594  * Increment the seqid if the LOCK/LOCKU succeeded, or
 595  * failed with a seqid incrementing error -
 596  * see comments nfs_fs.h:seqid_mutating_error()
 597  */
 598 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 599 {
 600         return nfs_increment_seqid(status, seqid);
 601 }
 602
 603 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 604 {
 605         struct rpc_sequence *sequence = seqid->sequence->sequence;
 606         int status = 0;
 607
 608         if (sequence->list.next == &seqid->list)
 609                 goto out;
 610         spin_lock(&sequence->lock);
 611         if (sequence->list.next != &seqid->list) {
 612                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 613                 status = -EAGAIN;
 614         }
 615         spin_unlock(&sequence->lock);
 616 out:
 617         return status;
 618 }
 619
 620 static int reclaimer(void *);
 621
 622 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 623 {
 624         smp_mb__before_clear_bit();
 625         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 626         smp_mb__after_clear_bit();
 627         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 628         rpc_wake_up(&clp->cl_rpcwaitq);
 629 }
 630
 631 /*
 632  * State recovery routine
 633  */
 634 static void nfs4_recover_state(struct nfs_client *clp)
 635 {
 636         struct task_struct *task;
 637
 638         __module_get(THIS_MODULE);
 639         atomic_inc(&clp->cl_count);
 640         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 641                         NIPQUAD(clp->cl_addr.sin_addr));
 642         if (!IS_ERR(task))
 643                 return;
 644         nfs4_clear_recover_bit(clp);
 645         nfs_put_client(clp);
 646         module_put(THIS_MODULE);
 647 }
 648
 649 /*
 650  * Schedule a state recovery attempt
 651  */
 652 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 653 {
 654         if (!clp)
 655                 return;
 656         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 657                 nfs4_recover_state(clp);
 658 }
 659
 660 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 661 {
 662         struct inode *inode = state->inode;
 663         struct file_lock *fl;
 664         int status = 0;
 665
 666         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 667                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 668                         continue;
 669                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 670                         continue;
 671                 status = ops->recover_lock(state, fl);
 672                 if (status >= 0)
 673                         continue;
 674                 switch (status) {
 675                         default:
 676                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 677                                                 __FUNCTION__, status);
 678                         case -NFS4ERR_EXPIRED:
 679                         case -NFS4ERR_NO_GRACE:
 680                         case -NFS4ERR_RECLAIM_BAD:
 681                         case -NFS4ERR_RECLAIM_CONFLICT:
 682                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 683                                 break;
 684                         case -NFS4ERR_STALE_CLIENTID:
 685                                 goto out_err;
 686                 }
 687         }
 688         return 0;
 689 out_err:
 690         return status;
 691 }
 692
 693 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 694 {
 695         struct nfs4_state *state;
 696         struct nfs4_lock_state *lock;
 697         int status = 0;
 698
 699         /* Note: we rely on the sp->so_states list being ordered
 700          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 701          * states first.
 702          * This is needed to ensure that the server won't give us any
 703          * read delegations that we have to return if, say, we are
 704          * recovering after a network partition or a reboot from a
 705          * server that doesn't support a grace period.
 706          */
 707         list_for_each_entry(state, &sp->so_states, open_states) {
 708                 if (state->state == 0)
 709                         continue;
 710                 status = ops->recover_open(sp, state);
 711                 if (status >= 0) {
 712                         status = nfs4_reclaim_locks(ops, state);
 713                         if (status < 0)
 714                                 goto out_err;
 715                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 716                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 717                                         printk("%s: Lock reclaim failed!\n",
 718                                                         __FUNCTION__);
 719                         }
 720                         continue;
 721                 }
 722                 switch (status) {
 723                         default:
 724                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 725                                                 __FUNCTION__, status);
 726                         case -ENOENT:
 727                         case -NFS4ERR_RECLAIM_BAD:
 728                         case -NFS4ERR_RECLAIM_CONFLICT:
 729                                 /*
 730                                  * Open state on this file cannot be recovered
 731                                  * All we can do is revert to using the zero stateid.
 732                                  */
 733                                 memset(state->stateid.data, 0,
 734                                         sizeof(state->stateid.data));
 735                                 /* Mark the file as being 'closed' */
 736                                 state->state = 0;
 737                                 break;
 738                         case -NFS4ERR_EXPIRED:
 739                         case -NFS4ERR_NO_GRACE:
 740                         case -NFS4ERR_STALE_CLIENTID:
 741                                 goto out_err;
 742                 }
 743         }
 744         return 0;
 745 out_err:
 746         return status;
 747 }
 748
 749 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 750 {
 751         struct nfs4_state_owner *sp;
 752         struct nfs4_state *state;
 753         struct nfs4_lock_state *lock;
 754
 755         /* Reset all sequence ids to zero */
 756         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 757                 sp->so_seqid.counter = 0;
 758                 sp->so_seqid.flags = 0;
 759                 spin_lock(&sp->so_lock);
 760                 list_for_each_entry(state, &sp->so_states, open_states) {
 761                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 762                                 lock->ls_seqid.counter = 0;
 763                                 lock->ls_seqid.flags = 0;
 764                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 765                         }
 766                 }
 767                 spin_unlock(&sp->so_lock);
 768         }
 769 }
 770
 771 static int reclaimer(void *ptr)
 772 {
 773         struct nfs_client *clp = ptr;
 774         struct nfs4_state_owner *sp;
 775         struct nfs4_state_recovery_ops *ops;
 776         struct rpc_cred *cred;
 777         int status = 0;
 778
 779         allow_signal(SIGKILL);
 780
 781         /* Ensure exclusive access to NFSv4 state */
 782         lock_kernel();
 783         down_write(&clp->cl_sem);
 784         /* Are there any NFS mounts out there? */
 785         if (list_empty(&clp->cl_superblocks))
 786                 goto out;
 787 restart_loop:
 788         ops = &nfs4_network_partition_recovery_ops;
 789         /* Are there any open files on this volume? */
 790         cred = nfs4_get_renew_cred(clp);
 791         if (cred != NULL) {
 792                 /* Yes there are: try to renew the old lease */
 793                 status = nfs4_proc_renew(clp, cred);
 794                 switch (status) {
 795                         case 0:
 796                         case -NFS4ERR_CB_PATH_DOWN:
 797                                 put_rpccred(cred);
 798                                 goto out;
 799                         case -NFS4ERR_STALE_CLIENTID:
 800                         case -NFS4ERR_LEASE_MOVED:
 801                                 ops = &nfs4_reboot_recovery_ops;
 802                 }
 803         } else {
 804                 /* "reboot" to ensure we clear all state on the server */
 805                 clp->cl_boot_time = CURRENT_TIME;
 806                 cred = nfs4_get_setclientid_cred(clp);
 807         }
 808         /* We're going to have to re-establish a clientid */
 809         nfs4_state_mark_reclaim(clp);
 810         status = -ENOENT;
 811         if (cred != NULL) {
 812                 status = nfs4_init_client(clp, cred);
 813                 put_rpccred(cred);
 814         }
 815         if (status)
 816                 goto out_error;
 817         /* Mark all delegations for reclaim */
 818         nfs_delegation_mark_reclaim(clp);
 819         /* Note: list is protected by exclusive lock on cl->cl_sem */
 820         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 821                 status = nfs4_reclaim_open_state(ops, sp);
 822                 if (status < 0) {
 823                         if (status == -NFS4ERR_NO_GRACE) {
 824                                 ops = &nfs4_network_partition_recovery_ops;
 825                                 status = nfs4_reclaim_open_state(ops, sp);
 826                         }
 827                         if (status == -NFS4ERR_STALE_CLIENTID)
 828                                 goto restart_loop;
 829                         if (status == -NFS4ERR_EXPIRED)
 830                                 goto restart_loop;
 831                 }
 832         }
 833         nfs_delegation_reap_unclaimed(clp);
 834 out:
 835         up_write(&clp->cl_sem);
 836         unlock_kernel();
 837         if (status == -NFS4ERR_CB_PATH_DOWN)
 838                 nfs_handle_cb_pathdown(clp);
 839         nfs4_clear_recover_bit(clp);
 840         nfs_put_client(clp);
 841         module_put_and_exit(0);
 842         return 0;
 843 out_error:
 844         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 845                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
 846         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 847         goto out;
 848 }
 849
 850 /*
 851  * Local variables:
 852  *  c-basic-offset: 8
 853  * End:
 854  */