git.oblomov.eu Git - linux-2.6/blob - fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/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 void
  62 init_nfsv4_state(struct nfs_server *server)
  63 {
  64         server->nfs_client = NULL;
  65         INIT_LIST_HEAD(&server->nfs4_siblings);
  66 }
  67
  68 void
  69 destroy_nfsv4_state(struct nfs_server *server)
  70 {
  71         kfree(server->mnt_path);
  72         server->mnt_path = NULL;
  73         if (server->nfs_client) {
  74                 nfs_put_client(server->nfs_client);
  75                 server->nfs_client = NULL;
  76         }
  77 }
  78
  79 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
  80 {
  81         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
  82                         nfs_callback_tcpport, cred);
  83         if (status == 0)
  84                 status = nfs4_proc_setclientid_confirm(clp, cred);
  85         if (status == 0)
  86                 nfs4_schedule_state_renewal(clp);
  87         return status;
  88 }
  89
  90 u32
  91 nfs4_alloc_lockowner_id(struct nfs_client *clp)
  92 {
  93         return clp->cl_lockowner_id ++;
  94 }
  95
  96 static struct nfs4_state_owner *
  97 nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred)
  98 {
  99         struct nfs4_state_owner *sp = NULL;
 100
 101         if (!list_empty(&clp->cl_unused)) {
 102                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 103                 atomic_inc(&sp->so_count);
 104                 sp->so_cred = cred;
 105                 list_move(&sp->so_list, &clp->cl_state_owners);
 106                 clp->cl_nunused--;
 107         }
 108         return sp;
 109 }
 110
 111 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
 112 {
 113         struct nfs4_state_owner *sp;
 114         struct rpc_cred *cred = NULL;
 115
 116         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 117                 if (list_empty(&sp->so_states))
 118                         continue;
 119                 cred = get_rpccred(sp->so_cred);
 120                 break;
 121         }
 122         return cred;
 123 }
 124
 125 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 126 {
 127         struct nfs4_state_owner *sp;
 128
 129         if (!list_empty(&clp->cl_state_owners)) {
 130                 sp = list_entry(clp->cl_state_owners.next,
 131                                 struct nfs4_state_owner, so_list);
 132                 return get_rpccred(sp->so_cred);
 133         }
 134         return NULL;
 135 }
 136
 137 static struct nfs4_state_owner *
 138 nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred)
 139 {
 140         struct nfs4_state_owner *sp, *res = NULL;
 141
 142         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 143                 if (sp->so_cred != cred)
 144                         continue;
 145                 atomic_inc(&sp->so_count);
 146                 /* Move to the head of the list */
 147                 list_move(&sp->so_list, &clp->cl_state_owners);
 148                 res = sp;
 149                 break;
 150         }
 151         return res;
 152 }
 153
 154 /*
 155  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 156  * create a new state_owner.
 157  *
 158  */
 159 static struct nfs4_state_owner *
 160 nfs4_alloc_state_owner(void)
 161 {
 162         struct nfs4_state_owner *sp;
 163
 164         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 165         if (!sp)
 166                 return NULL;
 167         spin_lock_init(&sp->so_lock);
 168         INIT_LIST_HEAD(&sp->so_states);
 169         INIT_LIST_HEAD(&sp->so_delegations);
 170         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 171         sp->so_seqid.sequence = &sp->so_sequence;
 172         spin_lock_init(&sp->so_sequence.lock);
 173         INIT_LIST_HEAD(&sp->so_sequence.list);
 174         atomic_set(&sp->so_count, 1);
 175         return sp;
 176 }
 177
 178 void
 179 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 180 {
 181         struct nfs_client *clp = sp->so_client;
 182         spin_lock(&clp->cl_lock);
 183         list_del_init(&sp->so_list);
 184         spin_unlock(&clp->cl_lock);
 185 }
 186
 187 /*
 188  * Note: must be called with clp->cl_sem held in order to prevent races
 189  *       with reboot recovery!
 190  */
 191 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 192 {
 193         struct nfs_client *clp = server->nfs_client;
 194         struct nfs4_state_owner *sp, *new;
 195
 196         get_rpccred(cred);
 197         new = nfs4_alloc_state_owner();
 198         spin_lock(&clp->cl_lock);
 199         sp = nfs4_find_state_owner(clp, cred);
 200         if (sp == NULL)
 201                 sp = nfs4_client_grab_unused(clp, cred);
 202         if (sp == NULL && new != NULL) {
 203                 list_add(&new->so_list, &clp->cl_state_owners);
 204                 new->so_client = clp;
 205                 new->so_id = nfs4_alloc_lockowner_id(clp);
 206                 new->so_cred = cred;
 207                 sp = new;
 208                 new = NULL;
 209         }
 210         spin_unlock(&clp->cl_lock);
 211         kfree(new);
 212         if (sp != NULL)
 213                 return sp;
 214         put_rpccred(cred);
 215         return NULL;
 216 }
 217
 218 /*
 219  * Must be called with clp->cl_sem held in order to avoid races
 220  * with state recovery...
 221  */
 222 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 223 {
 224         struct nfs_client *clp = sp->so_client;
 225         struct rpc_cred *cred = sp->so_cred;
 226
 227         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 228                 return;
 229         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 230                 goto out_free;
 231         if (list_empty(&sp->so_list))
 232                 goto out_free;
 233         list_move(&sp->so_list, &clp->cl_unused);
 234         clp->cl_nunused++;
 235         spin_unlock(&clp->cl_lock);
 236         put_rpccred(cred);
 237         cred = NULL;
 238         return;
 239 out_free:
 240         list_del(&sp->so_list);
 241         spin_unlock(&clp->cl_lock);
 242         put_rpccred(cred);
 243         kfree(sp);
 244 }
 245
 246 static struct nfs4_state *
 247 nfs4_alloc_open_state(void)
 248 {
 249         struct nfs4_state *state;
 250
 251         state = kzalloc(sizeof(*state), GFP_KERNEL);
 252         if (!state)
 253                 return NULL;
 254         atomic_set(&state->count, 1);
 255         INIT_LIST_HEAD(&state->lock_states);
 256         spin_lock_init(&state->state_lock);
 257         return state;
 258 }
 259
 260 void
 261 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 262 {
 263         if (state->state == mode)
 264                 return;
 265         /* NB! List reordering - see the reclaim code for why.  */
 266         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 267                 if (mode & FMODE_WRITE)
 268                         list_move(&state->open_states, &state->owner->so_states);
 269                 else
 270                         list_move_tail(&state->open_states, &state->owner->so_states);
 271         }
 272         if (mode == 0)
 273                 list_del_init(&state->inode_states);
 274         state->state = mode;
 275 }
 276
 277 static struct nfs4_state *
 278 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 279 {
 280         struct nfs_inode *nfsi = NFS_I(inode);
 281         struct nfs4_state *state;
 282
 283         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 284                 /* Is this in the process of being freed? */
 285                 if (state->state == 0)
 286                         continue;
 287                 if (state->owner == owner) {
 288                         atomic_inc(&state->count);
 289                         return state;
 290                 }
 291         }
 292         return NULL;
 293 }
 294
 295 static void
 296 nfs4_free_open_state(struct nfs4_state *state)
 297 {
 298         kfree(state);
 299 }
 300
 301 struct nfs4_state *
 302 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 303 {
 304         struct nfs4_state *state, *new;
 305         struct nfs_inode *nfsi = NFS_I(inode);
 306
 307         spin_lock(&inode->i_lock);
 308         state = __nfs4_find_state_byowner(inode, owner);
 309         spin_unlock(&inode->i_lock);
 310         if (state)
 311                 goto out;
 312         new = nfs4_alloc_open_state();
 313         spin_lock(&owner->so_lock);
 314         spin_lock(&inode->i_lock);
 315         state = __nfs4_find_state_byowner(inode, owner);
 316         if (state == NULL && new != NULL) {
 317                 state = new;
 318                 state->owner = owner;
 319                 atomic_inc(&owner->so_count);
 320                 list_add(&state->inode_states, &nfsi->open_states);
 321                 state->inode = igrab(inode);
 322                 spin_unlock(&inode->i_lock);
 323                 /* Note: The reclaim code dictates that we add stateless
 324                  * and read-only stateids to the end of the list */
 325                 list_add_tail(&state->open_states, &owner->so_states);
 326                 spin_unlock(&owner->so_lock);
 327         } else {
 328                 spin_unlock(&inode->i_lock);
 329                 spin_unlock(&owner->so_lock);
 330                 if (new)
 331                         nfs4_free_open_state(new);
 332         }
 333 out:
 334         return state;
 335 }
 336
 337 /*
 338  * Beware! Caller must be holding exactly one
 339  * reference to clp->cl_sem!
 340  */
 341 void nfs4_put_open_state(struct nfs4_state *state)
 342 {
 343         struct inode *inode = state->inode;
 344         struct nfs4_state_owner *owner = state->owner;
 345
 346         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 347                 return;
 348         spin_lock(&inode->i_lock);
 349         if (!list_empty(&state->inode_states))
 350                 list_del(&state->inode_states);
 351         list_del(&state->open_states);
 352         spin_unlock(&inode->i_lock);
 353         spin_unlock(&owner->so_lock);
 354         iput(inode);
 355         nfs4_free_open_state(state);
 356         nfs4_put_state_owner(owner);
 357 }
 358
 359 /*
 360  * Close the current file.
 361  */
 362 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 363 {
 364         struct inode *inode = state->inode;
 365         struct nfs4_state_owner *owner = state->owner;
 366         int oldstate, newstate = 0;
 367
 368         atomic_inc(&owner->so_count);
 369         /* Protect against nfs4_find_state() */
 370         spin_lock(&owner->so_lock);
 371         spin_lock(&inode->i_lock);
 372         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 373                 case FMODE_READ:
 374                         state->n_rdonly--;
 375                         break;
 376                 case FMODE_WRITE:
 377                         state->n_wronly--;
 378                         break;
 379                 case FMODE_READ|FMODE_WRITE:
 380                         state->n_rdwr--;
 381         }
 382         oldstate = newstate = state->state;
 383         if (state->n_rdwr == 0) {
 384                 if (state->n_rdonly == 0)
 385                         newstate &= ~FMODE_READ;
 386                 if (state->n_wronly == 0)
 387                         newstate &= ~FMODE_WRITE;
 388         }
 389         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 390                 nfs4_state_set_mode_locked(state, newstate);
 391                 oldstate = newstate;
 392         }
 393         spin_unlock(&inode->i_lock);
 394         spin_unlock(&owner->so_lock);
 395
 396         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
 397                 return;
 398         nfs4_put_open_state(state);
 399         nfs4_put_state_owner(owner);
 400 }
 401
 402 /*
 403  * Search the state->lock_states for an existing lock_owner
 404  * that is compatible with current->files
 405  */
 406 static struct nfs4_lock_state *
 407 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 408 {
 409         struct nfs4_lock_state *pos;
 410         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 411                 if (pos->ls_owner != fl_owner)
 412                         continue;
 413                 atomic_inc(&pos->ls_count);
 414                 return pos;
 415         }
 416         return NULL;
 417 }
 418
 419 /*
 420  * Return a compatible lock_state. If no initialized lock_state structure
 421  * exists, return an uninitialized one.
 422  *
 423  */
 424 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 425 {
 426         struct nfs4_lock_state *lsp;
 427         struct nfs_client *clp = state->owner->so_client;
 428
 429         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 430         if (lsp == NULL)
 431                 return NULL;
 432         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 433         atomic_set(&lsp->ls_count, 1);
 434         lsp->ls_owner = fl_owner;
 435         spin_lock(&clp->cl_lock);
 436         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 437         spin_unlock(&clp->cl_lock);
 438         INIT_LIST_HEAD(&lsp->ls_locks);
 439         return lsp;
 440 }
 441
 442 /*
 443  * Return a compatible lock_state. If no initialized lock_state structure
 444  * exists, return an uninitialized one.
 445  *
 446  * The caller must be holding clp->cl_sem
 447  */
 448 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 449 {
 450         struct nfs4_lock_state *lsp, *new = NULL;
 451
 452         for(;;) {
 453                 spin_lock(&state->state_lock);
 454                 lsp = __nfs4_find_lock_state(state, owner);
 455                 if (lsp != NULL)
 456                         break;
 457                 if (new != NULL) {
 458                         new->ls_state = state;
 459                         list_add(&new->ls_locks, &state->lock_states);
 460                         set_bit(LK_STATE_IN_USE, &state->flags);
 461                         lsp = new;
 462                         new = NULL;
 463                         break;
 464                 }
 465                 spin_unlock(&state->state_lock);
 466                 new = nfs4_alloc_lock_state(state, owner);
 467                 if (new == NULL)
 468                         return NULL;
 469         }
 470         spin_unlock(&state->state_lock);
 471         kfree(new);
 472         return lsp;
 473 }
 474
 475 /*
 476  * Release reference to lock_state, and free it if we see that
 477  * it is no longer in use
 478  */
 479 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 480 {
 481         struct nfs4_state *state;
 482
 483         if (lsp == NULL)
 484                 return;
 485         state = lsp->ls_state;
 486         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 487                 return;
 488         list_del(&lsp->ls_locks);
 489         if (list_empty(&state->lock_states))
 490                 clear_bit(LK_STATE_IN_USE, &state->flags);
 491         spin_unlock(&state->state_lock);
 492         kfree(lsp);
 493 }
 494
 495 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 496 {
 497         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 498
 499         dst->fl_u.nfs4_fl.owner = lsp;
 500         atomic_inc(&lsp->ls_count);
 501 }
 502
 503 static void nfs4_fl_release_lock(struct file_lock *fl)
 504 {
 505         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 506 }
 507
 508 static struct file_lock_operations nfs4_fl_lock_ops = {
 509         .fl_copy_lock = nfs4_fl_copy_lock,
 510         .fl_release_private = nfs4_fl_release_lock,
 511 };
 512
 513 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 514 {
 515         struct nfs4_lock_state *lsp;
 516
 517         if (fl->fl_ops != NULL)
 518                 return 0;
 519         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 520         if (lsp == NULL)
 521                 return -ENOMEM;
 522         fl->fl_u.nfs4_fl.owner = lsp;
 523         fl->fl_ops = &nfs4_fl_lock_ops;
 524         return 0;
 525 }
 526
 527 /*
 528  * Byte-range lock aware utility to initialize the stateid of read/write
 529  * requests.
 530  */
 531 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 532 {
 533         struct nfs4_lock_state *lsp;
 534
 535         memcpy(dst, &state->stateid, sizeof(*dst));
 536         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 537                 return;
 538
 539         spin_lock(&state->state_lock);
 540         lsp = __nfs4_find_lock_state(state, fl_owner);
 541         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 542                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 543         spin_unlock(&state->state_lock);
 544         nfs4_put_lock_state(lsp);
 545 }
 546
 547 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 548 {
 549         struct rpc_sequence *sequence = counter->sequence;
 550         struct nfs_seqid *new;
 551
 552         new = kmalloc(sizeof(*new), GFP_KERNEL);
 553         if (new != NULL) {
 554                 new->sequence = counter;
 555                 spin_lock(&sequence->lock);
 556                 list_add_tail(&new->list, &sequence->list);
 557                 spin_unlock(&sequence->lock);
 558         }
 559         return new;
 560 }
 561
 562 void nfs_free_seqid(struct nfs_seqid *seqid)
 563 {
 564         struct rpc_sequence *sequence = seqid->sequence->sequence;
 565
 566         spin_lock(&sequence->lock);
 567         list_del(&seqid->list);
 568         spin_unlock(&sequence->lock);
 569         rpc_wake_up(&sequence->wait);
 570         kfree(seqid);
 571 }
 572
 573 /*
 574  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 575  * failed with a seqid incrementing error -
 576  * see comments nfs_fs.h:seqid_mutating_error()
 577  */
 578 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 579 {
 580         switch (status) {
 581                 case 0:
 582                         break;
 583                 case -NFS4ERR_BAD_SEQID:
 584                 case -NFS4ERR_STALE_CLIENTID:
 585                 case -NFS4ERR_STALE_STATEID:
 586                 case -NFS4ERR_BAD_STATEID:
 587                 case -NFS4ERR_BADXDR:
 588                 case -NFS4ERR_RESOURCE:
 589                 case -NFS4ERR_NOFILEHANDLE:
 590                         /* Non-seqid mutating errors */
 591                         return;
 592         };
 593         /*
 594          * Note: no locking needed as we are guaranteed to be first
 595          * on the sequence list
 596          */
 597         seqid->sequence->counter++;
 598 }
 599
 600 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 601 {
 602         if (status == -NFS4ERR_BAD_SEQID) {
 603                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 604                                 struct nfs4_state_owner, so_seqid);
 605                 nfs4_drop_state_owner(sp);
 606         }
 607         return nfs_increment_seqid(status, seqid);
 608 }
 609
 610 /*
 611  * Increment the seqid if the LOCK/LOCKU succeeded, or
 612  * failed with a seqid incrementing error -
 613  * see comments nfs_fs.h:seqid_mutating_error()
 614  */
 615 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 616 {
 617         return nfs_increment_seqid(status, seqid);
 618 }
 619
 620 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 621 {
 622         struct rpc_sequence *sequence = seqid->sequence->sequence;
 623         int status = 0;
 624
 625         if (sequence->list.next == &seqid->list)
 626                 goto out;
 627         spin_lock(&sequence->lock);
 628         if (sequence->list.next != &seqid->list) {
 629                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 630                 status = -EAGAIN;
 631         }
 632         spin_unlock(&sequence->lock);
 633 out:
 634         return status;
 635 }
 636
 637 static int reclaimer(void *);
 638
 639 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
 640 {
 641         smp_mb__before_clear_bit();
 642         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 643         smp_mb__after_clear_bit();
 644         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 645         rpc_wake_up(&clp->cl_rpcwaitq);
 646 }
 647
 648 /*
 649  * State recovery routine
 650  */
 651 static void nfs4_recover_state(struct nfs_client *clp)
 652 {
 653         struct task_struct *task;
 654
 655         __module_get(THIS_MODULE);
 656         atomic_inc(&clp->cl_count);
 657         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 658                         NIPQUAD(clp->cl_addr.sin_addr));
 659         if (!IS_ERR(task))
 660                 return;
 661         nfs4_clear_recover_bit(clp);
 662         nfs_put_client(clp);
 663         module_put(THIS_MODULE);
 664 }
 665
 666 /*
 667  * Schedule a state recovery attempt
 668  */
 669 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 670 {
 671         if (!clp)
 672                 return;
 673         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 674                 nfs4_recover_state(clp);
 675 }
 676
 677 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 678 {
 679         struct inode *inode = state->inode;
 680         struct file_lock *fl;
 681         int status = 0;
 682
 683         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 684                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 685                         continue;
 686                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 687                         continue;
 688                 status = ops->recover_lock(state, fl);
 689                 if (status >= 0)
 690                         continue;
 691                 switch (status) {
 692                         default:
 693                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 694                                                 __FUNCTION__, status);
 695                         case -NFS4ERR_EXPIRED:
 696                         case -NFS4ERR_NO_GRACE:
 697                         case -NFS4ERR_RECLAIM_BAD:
 698                         case -NFS4ERR_RECLAIM_CONFLICT:
 699                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 700                                 break;
 701                         case -NFS4ERR_STALE_CLIENTID:
 702                                 goto out_err;
 703                 }
 704         }
 705         return 0;
 706 out_err:
 707         return status;
 708 }
 709
 710 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 711 {
 712         struct nfs4_state *state;
 713         struct nfs4_lock_state *lock;
 714         int status = 0;
 715
 716         /* Note: we rely on the sp->so_states list being ordered
 717          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 718          * states first.
 719          * This is needed to ensure that the server won't give us any
 720          * read delegations that we have to return if, say, we are
 721          * recovering after a network partition or a reboot from a
 722          * server that doesn't support a grace period.
 723          */
 724         list_for_each_entry(state, &sp->so_states, open_states) {
 725                 if (state->state == 0)
 726                         continue;
 727                 status = ops->recover_open(sp, state);
 728                 if (status >= 0) {
 729                         status = nfs4_reclaim_locks(ops, state);
 730                         if (status < 0)
 731                                 goto out_err;
 732                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 733                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 734                                         printk("%s: Lock reclaim failed!\n",
 735                                                         __FUNCTION__);
 736                         }
 737                         continue;
 738                 }
 739                 switch (status) {
 740                         default:
 741                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 742                                                 __FUNCTION__, status);
 743                         case -ENOENT:
 744                         case -NFS4ERR_RECLAIM_BAD:
 745                         case -NFS4ERR_RECLAIM_CONFLICT:
 746                                 /*
 747                                  * Open state on this file cannot be recovered
 748                                  * All we can do is revert to using the zero stateid.
 749                                  */
 750                                 memset(state->stateid.data, 0,
 751                                         sizeof(state->stateid.data));
 752                                 /* Mark the file as being 'closed' */
 753                                 state->state = 0;
 754                                 break;
 755                         case -NFS4ERR_EXPIRED:
 756                         case -NFS4ERR_NO_GRACE:
 757                         case -NFS4ERR_STALE_CLIENTID:
 758                                 goto out_err;
 759                 }
 760         }
 761         return 0;
 762 out_err:
 763         return status;
 764 }
 765
 766 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
 767 {
 768         struct nfs4_state_owner *sp;
 769         struct nfs4_state *state;
 770         struct nfs4_lock_state *lock;
 771
 772         /* Reset all sequence ids to zero */
 773         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 774                 sp->so_seqid.counter = 0;
 775                 sp->so_seqid.flags = 0;
 776                 spin_lock(&sp->so_lock);
 777                 list_for_each_entry(state, &sp->so_states, open_states) {
 778                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 779                                 lock->ls_seqid.counter = 0;
 780                                 lock->ls_seqid.flags = 0;
 781                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 782                         }
 783                 }
 784                 spin_unlock(&sp->so_lock);
 785         }
 786 }
 787
 788 static int reclaimer(void *ptr)
 789 {
 790         struct nfs_client *clp = ptr;
 791         struct nfs4_state_owner *sp;
 792         struct nfs4_state_recovery_ops *ops;
 793         struct rpc_cred *cred;
 794         int status = 0;
 795
 796         allow_signal(SIGKILL);
 797
 798         /* Ensure exclusive access to NFSv4 state */
 799         lock_kernel();
 800         down_write(&clp->cl_sem);
 801         /* Are there any NFS mounts out there? */
 802         if (list_empty(&clp->cl_superblocks))
 803                 goto out;
 804 restart_loop:
 805         ops = &nfs4_network_partition_recovery_ops;
 806         /* Are there any open files on this volume? */
 807         cred = nfs4_get_renew_cred(clp);
 808         if (cred != NULL) {
 809                 /* Yes there are: try to renew the old lease */
 810                 status = nfs4_proc_renew(clp, cred);
 811                 switch (status) {
 812                         case 0:
 813                         case -NFS4ERR_CB_PATH_DOWN:
 814                                 put_rpccred(cred);
 815                                 goto out;
 816                         case -NFS4ERR_STALE_CLIENTID:
 817                         case -NFS4ERR_LEASE_MOVED:
 818                                 ops = &nfs4_reboot_recovery_ops;
 819                 }
 820         } else {
 821                 /* "reboot" to ensure we clear all state on the server */
 822                 clp->cl_boot_time = CURRENT_TIME;
 823                 cred = nfs4_get_setclientid_cred(clp);
 824         }
 825         /* We're going to have to re-establish a clientid */
 826         nfs4_state_mark_reclaim(clp);
 827         status = -ENOENT;
 828         if (cred != NULL) {
 829                 status = nfs4_init_client(clp, cred);
 830                 put_rpccred(cred);
 831         }
 832         if (status)
 833                 goto out_error;
 834         /* Mark all delegations for reclaim */
 835         nfs_delegation_mark_reclaim(clp);
 836         /* Note: list is protected by exclusive lock on cl->cl_sem */
 837         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 838                 status = nfs4_reclaim_open_state(ops, sp);
 839                 if (status < 0) {
 840                         if (status == -NFS4ERR_NO_GRACE) {
 841                                 ops = &nfs4_network_partition_recovery_ops;
 842                                 status = nfs4_reclaim_open_state(ops, sp);
 843                         }
 844                         if (status == -NFS4ERR_STALE_CLIENTID)
 845                                 goto restart_loop;
 846                         if (status == -NFS4ERR_EXPIRED)
 847                                 goto restart_loop;
 848                 }
 849         }
 850         nfs_delegation_reap_unclaimed(clp);
 851 out:
 852         up_write(&clp->cl_sem);
 853         unlock_kernel();
 854         if (status == -NFS4ERR_CB_PATH_DOWN)
 855                 nfs_handle_cb_pathdown(clp);
 856         nfs4_clear_recover_bit(clp);
 857         nfs_put_client(clp);
 858         module_put_and_exit(0);
 859         return 0;
 860 out_error:
 861         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 862                                 NIPQUAD(clp->cl_addr.sin_addr), -status);
 863         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
 864         goto out;
 865 }
 866
 867 /*
 868  * Local variables:
 869  *  c-basic-offset: 8
 870  * End:
 871  */