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