fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/config.h>
  42 #include <linux/slab.h>
  43 #include <linux/smp_lock.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/bitops.h>
  50
  51 #include "nfs4_fs.h"
  52 #include "callback.h"
  53 #include "delegation.h"
  54
  55 #define OPENOWNER_POOL_SIZE     8
  56
  57 const nfs4_stateid zero_stateid;
  58
  59 static DEFINE_SPINLOCK(state_spinlock);
  60 static LIST_HEAD(nfs4_clientid_list);
  61
  62 void
  63 init_nfsv4_state(struct nfs_server *server)
  64 {
  65         server->nfs4_state = NULL;
  66         INIT_LIST_HEAD(&server->nfs4_siblings);
  67 }
  68
  69 void
  70 destroy_nfsv4_state(struct nfs_server *server)
  71 {
  72         kfree(server->mnt_path);
  73         server->mnt_path = NULL;
  74         if (server->nfs4_state) {
  75                 nfs4_put_client(server->nfs4_state);
  76                 server->nfs4_state = NULL;
  77         }
  78 }
  79
  80 /*
  81  * nfs4_get_client(): returns an empty client structure
  82  * nfs4_put_client(): drops reference to client structure
  83  *
  84  * Since these are allocated/deallocated very rarely, we don't
  85  * bother putting them in a slab cache...
  86  */
  87 static struct nfs4_client *
  88 nfs4_alloc_client(struct in_addr *addr)
  89 {
  90         struct nfs4_client *clp;
  91
  92         if (nfs_callback_up() < 0)
  93                 return NULL;
  94         if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) {
  95                 nfs_callback_down();
  96                 return NULL;
  97         }
  98         memset(clp, 0, sizeof(*clp));
  99         memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr));
 100         init_rwsem(&clp->cl_sem);
 101         INIT_LIST_HEAD(&clp->cl_delegations);
 102         INIT_LIST_HEAD(&clp->cl_state_owners);
 103         INIT_LIST_HEAD(&clp->cl_unused);
 104         spin_lock_init(&clp->cl_lock);
 105         atomic_set(&clp->cl_count, 1);
 106         INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp);
 107         INIT_LIST_HEAD(&clp->cl_superblocks);
 108         rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client");
 109         clp->cl_rpcclient = ERR_PTR(-EINVAL);
 110         clp->cl_boot_time = CURRENT_TIME;
 111         clp->cl_state = 0;
 112         return clp;
 113 }
 114
 115 static void
 116 nfs4_free_client(struct nfs4_client *clp)
 117 {
 118         struct nfs4_state_owner *sp;
 119
 120         while (!list_empty(&clp->cl_unused)) {
 121                 sp = list_entry(clp->cl_unused.next,
 122                                 struct nfs4_state_owner,
 123                                 so_list);
 124                 list_del(&sp->so_list);
 125                 kfree(sp);
 126         }
 127         BUG_ON(!list_empty(&clp->cl_state_owners));
 128         if (clp->cl_cred)
 129                 put_rpccred(clp->cl_cred);
 130         nfs_idmap_delete(clp);
 131         if (!IS_ERR(clp->cl_rpcclient))
 132                 rpc_shutdown_client(clp->cl_rpcclient);
 133         kfree(clp);
 134         nfs_callback_down();
 135 }
 136
 137 static struct nfs4_client *__nfs4_find_client(struct in_addr *addr)
 138 {
 139         struct nfs4_client *clp;
 140         list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) {
 141                 if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) {
 142                         atomic_inc(&clp->cl_count);
 143                         return clp;
 144                 }
 145         }
 146         return NULL;
 147 }
 148
 149 struct nfs4_client *nfs4_find_client(struct in_addr *addr)
 150 {
 151         struct nfs4_client *clp;
 152         spin_lock(&state_spinlock);
 153         clp = __nfs4_find_client(addr);
 154         spin_unlock(&state_spinlock);
 155         return clp;
 156 }
 157
 158 struct nfs4_client *
 159 nfs4_get_client(struct in_addr *addr)
 160 {
 161         struct nfs4_client *clp, *new = NULL;
 162
 163         spin_lock(&state_spinlock);
 164         for (;;) {
 165                 clp = __nfs4_find_client(addr);
 166                 if (clp != NULL)
 167                         break;
 168                 clp = new;
 169                 if (clp != NULL) {
 170                         list_add(&clp->cl_servers, &nfs4_clientid_list);
 171                         new = NULL;
 172                         break;
 173                 }
 174                 spin_unlock(&state_spinlock);
 175                 new = nfs4_alloc_client(addr);
 176                 spin_lock(&state_spinlock);
 177                 if (new == NULL)
 178                         break;
 179         }
 180         spin_unlock(&state_spinlock);
 181         if (new)
 182                 nfs4_free_client(new);
 183         return clp;
 184 }
 185
 186 void
 187 nfs4_put_client(struct nfs4_client *clp)
 188 {
 189         if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock))
 190                 return;
 191         list_del(&clp->cl_servers);
 192         spin_unlock(&state_spinlock);
 193         BUG_ON(!list_empty(&clp->cl_superblocks));
 194         rpc_wake_up(&clp->cl_rpcwaitq);
 195         nfs4_kill_renewd(clp);
 196         nfs4_free_client(clp);
 197 }
 198
 199 static int __nfs4_init_client(struct nfs4_client *clp)
 200 {
 201         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport);
 202         if (status == 0)
 203                 status = nfs4_proc_setclientid_confirm(clp);
 204         if (status == 0)
 205                 nfs4_schedule_state_renewal(clp);
 206         return status;
 207 }
 208
 209 int nfs4_init_client(struct nfs4_client *clp)
 210 {
 211         return nfs4_map_errors(__nfs4_init_client(clp));
 212 }
 213
 214 u32
 215 nfs4_alloc_lockowner_id(struct nfs4_client *clp)
 216 {
 217         return clp->cl_lockowner_id ++;
 218 }
 219
 220 static struct nfs4_state_owner *
 221 nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred)
 222 {
 223         struct nfs4_state_owner *sp = NULL;
 224
 225         if (!list_empty(&clp->cl_unused)) {
 226                 sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
 227                 atomic_inc(&sp->so_count);
 228                 sp->so_cred = cred;
 229                 list_move(&sp->so_list, &clp->cl_state_owners);
 230                 clp->cl_nunused--;
 231         }
 232         return sp;
 233 }
 234
 235 static struct nfs4_state_owner *
 236 nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred)
 237 {
 238         struct nfs4_state_owner *sp, *res = NULL;
 239
 240         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 241                 if (sp->so_cred != cred)
 242                         continue;
 243                 atomic_inc(&sp->so_count);
 244                 /* Move to the head of the list */
 245                 list_move(&sp->so_list, &clp->cl_state_owners);
 246                 res = sp;
 247                 break;
 248         }
 249         return res;
 250 }
 251
 252 /*
 253  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 254  * create a new state_owner.
 255  *
 256  */
 257 static struct nfs4_state_owner *
 258 nfs4_alloc_state_owner(void)
 259 {
 260         struct nfs4_state_owner *sp;
 261
 262         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 263         if (!sp)
 264                 return NULL;
 265         spin_lock_init(&sp->so_lock);
 266         INIT_LIST_HEAD(&sp->so_states);
 267         INIT_LIST_HEAD(&sp->so_delegations);
 268         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 269         sp->so_seqid.sequence = &sp->so_sequence;
 270         spin_lock_init(&sp->so_sequence.lock);
 271         INIT_LIST_HEAD(&sp->so_sequence.list);
 272         atomic_set(&sp->so_count, 1);
 273         return sp;
 274 }
 275
 276 void
 277 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 278 {
 279         struct nfs4_client *clp = sp->so_client;
 280         spin_lock(&clp->cl_lock);
 281         list_del_init(&sp->so_list);
 282         spin_unlock(&clp->cl_lock);
 283 }
 284
 285 /*
 286  * Note: must be called with clp->cl_sem held in order to prevent races
 287  *       with reboot recovery!
 288  */
 289 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 290 {
 291         struct nfs4_client *clp = server->nfs4_state;
 292         struct nfs4_state_owner *sp, *new;
 293
 294         get_rpccred(cred);
 295         new = nfs4_alloc_state_owner();
 296         spin_lock(&clp->cl_lock);
 297         sp = nfs4_find_state_owner(clp, cred);
 298         if (sp == NULL)
 299                 sp = nfs4_client_grab_unused(clp, cred);
 300         if (sp == NULL && new != NULL) {
 301                 list_add(&new->so_list, &clp->cl_state_owners);
 302                 new->so_client = clp;
 303                 new->so_id = nfs4_alloc_lockowner_id(clp);
 304                 new->so_cred = cred;
 305                 sp = new;
 306                 new = NULL;
 307         }
 308         spin_unlock(&clp->cl_lock);
 309         kfree(new);
 310         if (sp != NULL)
 311                 return sp;
 312         put_rpccred(cred);
 313         return NULL;
 314 }
 315
 316 /*
 317  * Must be called with clp->cl_sem held in order to avoid races
 318  * with state recovery...
 319  */
 320 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 321 {
 322         struct nfs4_client *clp = sp->so_client;
 323         struct rpc_cred *cred = sp->so_cred;
 324
 325         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 326                 return;
 327         if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
 328                 goto out_free;
 329         if (list_empty(&sp->so_list))
 330                 goto out_free;
 331         list_move(&sp->so_list, &clp->cl_unused);
 332         clp->cl_nunused++;
 333         spin_unlock(&clp->cl_lock);
 334         put_rpccred(cred);
 335         cred = NULL;
 336         return;
 337 out_free:
 338         list_del(&sp->so_list);
 339         spin_unlock(&clp->cl_lock);
 340         put_rpccred(cred);
 341         kfree(sp);
 342 }
 343
 344 static struct nfs4_state *
 345 nfs4_alloc_open_state(void)
 346 {
 347         struct nfs4_state *state;
 348
 349         state = kzalloc(sizeof(*state), GFP_KERNEL);
 350         if (!state)
 351                 return NULL;
 352         atomic_set(&state->count, 1);
 353         INIT_LIST_HEAD(&state->lock_states);
 354         spin_lock_init(&state->state_lock);
 355         return state;
 356 }
 357
 358 void
 359 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
 360 {
 361         if (state->state == mode)
 362                 return;
 363         /* NB! List reordering - see the reclaim code for why.  */
 364         if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 365                 if (mode & FMODE_WRITE)
 366                         list_move(&state->open_states, &state->owner->so_states);
 367                 else
 368                         list_move_tail(&state->open_states, &state->owner->so_states);
 369         }
 370         if (mode == 0)
 371                 list_del_init(&state->inode_states);
 372         state->state = mode;
 373 }
 374
 375 static struct nfs4_state *
 376 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 377 {
 378         struct nfs_inode *nfsi = NFS_I(inode);
 379         struct nfs4_state *state;
 380
 381         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 382                 /* Is this in the process of being freed? */
 383                 if (state->state == 0)
 384                         continue;
 385                 if (state->owner == owner) {
 386                         atomic_inc(&state->count);
 387                         return state;
 388                 }
 389         }
 390         return NULL;
 391 }
 392
 393 static void
 394 nfs4_free_open_state(struct nfs4_state *state)
 395 {
 396         kfree(state);
 397 }
 398
 399 struct nfs4_state *
 400 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 401 {
 402         struct nfs4_state *state, *new;
 403         struct nfs_inode *nfsi = NFS_I(inode);
 404
 405         spin_lock(&inode->i_lock);
 406         state = __nfs4_find_state_byowner(inode, owner);
 407         spin_unlock(&inode->i_lock);
 408         if (state)
 409                 goto out;
 410         new = nfs4_alloc_open_state();
 411         spin_lock(&owner->so_lock);
 412         spin_lock(&inode->i_lock);
 413         state = __nfs4_find_state_byowner(inode, owner);
 414         if (state == NULL && new != NULL) {
 415                 state = new;
 416                 state->owner = owner;
 417                 atomic_inc(&owner->so_count);
 418                 list_add(&state->inode_states, &nfsi->open_states);
 419                 state->inode = igrab(inode);
 420                 spin_unlock(&inode->i_lock);
 421                 /* Note: The reclaim code dictates that we add stateless
 422                  * and read-only stateids to the end of the list */
 423                 list_add_tail(&state->open_states, &owner->so_states);
 424                 spin_unlock(&owner->so_lock);
 425         } else {
 426                 spin_unlock(&inode->i_lock);
 427                 spin_unlock(&owner->so_lock);
 428                 if (new)
 429                         nfs4_free_open_state(new);
 430         }
 431 out:
 432         return state;
 433 }
 434
 435 /*
 436  * Beware! Caller must be holding exactly one
 437  * reference to clp->cl_sem!
 438  */
 439 void nfs4_put_open_state(struct nfs4_state *state)
 440 {
 441         struct inode *inode = state->inode;
 442         struct nfs4_state_owner *owner = state->owner;
 443
 444         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 445                 return;
 446         spin_lock(&inode->i_lock);
 447         if (!list_empty(&state->inode_states))
 448                 list_del(&state->inode_states);
 449         list_del(&state->open_states);
 450         spin_unlock(&inode->i_lock);
 451         spin_unlock(&owner->so_lock);
 452         iput(inode);
 453         nfs4_free_open_state(state);
 454         nfs4_put_state_owner(owner);
 455 }
 456
 457 /*
 458  * Close the current file.
 459  */
 460 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
 461 {
 462         struct inode *inode = state->inode;
 463         struct nfs4_state_owner *owner = state->owner;
 464         int oldstate, newstate = 0;
 465
 466         atomic_inc(&owner->so_count);
 467         /* Protect against nfs4_find_state() */
 468         spin_lock(&owner->so_lock);
 469         spin_lock(&inode->i_lock);
 470         switch (mode & (FMODE_READ | FMODE_WRITE)) {
 471                 case FMODE_READ:
 472                         state->n_rdonly--;
 473                         break;
 474                 case FMODE_WRITE:
 475                         state->n_wronly--;
 476                         break;
 477                 case FMODE_READ|FMODE_WRITE:
 478                         state->n_rdwr--;
 479         }
 480         oldstate = newstate = state->state;
 481         if (state->n_rdwr == 0) {
 482                 if (state->n_rdonly == 0)
 483                         newstate &= ~FMODE_READ;
 484                 if (state->n_wronly == 0)
 485                         newstate &= ~FMODE_WRITE;
 486         }
 487         if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
 488                 nfs4_state_set_mode_locked(state, newstate);
 489                 oldstate = newstate;
 490         }
 491         spin_unlock(&inode->i_lock);
 492         spin_unlock(&owner->so_lock);
 493
 494         if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
 495                 return;
 496         nfs4_put_open_state(state);
 497         nfs4_put_state_owner(owner);
 498 }
 499
 500 /*
 501  * Search the state->lock_states for an existing lock_owner
 502  * that is compatible with current->files
 503  */
 504 static struct nfs4_lock_state *
 505 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 506 {
 507         struct nfs4_lock_state *pos;
 508         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 509                 if (pos->ls_owner != fl_owner)
 510                         continue;
 511                 atomic_inc(&pos->ls_count);
 512                 return pos;
 513         }
 514         return NULL;
 515 }
 516
 517 /*
 518  * Return a compatible lock_state. If no initialized lock_state structure
 519  * exists, return an uninitialized one.
 520  *
 521  */
 522 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 523 {
 524         struct nfs4_lock_state *lsp;
 525         struct nfs4_client *clp = state->owner->so_client;
 526
 527         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 528         if (lsp == NULL)
 529                 return NULL;
 530         lsp->ls_seqid.sequence = &state->owner->so_sequence;
 531         atomic_set(&lsp->ls_count, 1);
 532         lsp->ls_owner = fl_owner;
 533         spin_lock(&clp->cl_lock);
 534         lsp->ls_id = nfs4_alloc_lockowner_id(clp);
 535         spin_unlock(&clp->cl_lock);
 536         INIT_LIST_HEAD(&lsp->ls_locks);
 537         return lsp;
 538 }
 539
 540 /*
 541  * Return a compatible lock_state. If no initialized lock_state structure
 542  * exists, return an uninitialized one.
 543  *
 544  * The caller must be holding clp->cl_sem
 545  */
 546 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 547 {
 548         struct nfs4_lock_state *lsp, *new = NULL;
 549
 550         for(;;) {
 551                 spin_lock(&state->state_lock);
 552                 lsp = __nfs4_find_lock_state(state, owner);
 553                 if (lsp != NULL)
 554                         break;
 555                 if (new != NULL) {
 556                         new->ls_state = state;
 557                         list_add(&new->ls_locks, &state->lock_states);
 558                         set_bit(LK_STATE_IN_USE, &state->flags);
 559                         lsp = new;
 560                         new = NULL;
 561                         break;
 562                 }
 563                 spin_unlock(&state->state_lock);
 564                 new = nfs4_alloc_lock_state(state, owner);
 565                 if (new == NULL)
 566                         return NULL;
 567         }
 568         spin_unlock(&state->state_lock);
 569         kfree(new);
 570         return lsp;
 571 }
 572
 573 /*
 574  * Release reference to lock_state, and free it if we see that
 575  * it is no longer in use
 576  */
 577 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 578 {
 579         struct nfs4_state *state;
 580
 581         if (lsp == NULL)
 582                 return;
 583         state = lsp->ls_state;
 584         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 585                 return;
 586         list_del(&lsp->ls_locks);
 587         if (list_empty(&state->lock_states))
 588                 clear_bit(LK_STATE_IN_USE, &state->flags);
 589         spin_unlock(&state->state_lock);
 590         kfree(lsp);
 591 }
 592
 593 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 594 {
 595         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 596
 597         dst->fl_u.nfs4_fl.owner = lsp;
 598         atomic_inc(&lsp->ls_count);
 599 }
 600
 601 static void nfs4_fl_release_lock(struct file_lock *fl)
 602 {
 603         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 604 }
 605
 606 static struct file_lock_operations nfs4_fl_lock_ops = {
 607         .fl_copy_lock = nfs4_fl_copy_lock,
 608         .fl_release_private = nfs4_fl_release_lock,
 609 };
 610
 611 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 612 {
 613         struct nfs4_lock_state *lsp;
 614
 615         if (fl->fl_ops != NULL)
 616                 return 0;
 617         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 618         if (lsp == NULL)
 619                 return -ENOMEM;
 620         fl->fl_u.nfs4_fl.owner = lsp;
 621         fl->fl_ops = &nfs4_fl_lock_ops;
 622         return 0;
 623 }
 624
 625 /*
 626  * Byte-range lock aware utility to initialize the stateid of read/write
 627  * requests.
 628  */
 629 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 630 {
 631         struct nfs4_lock_state *lsp;
 632
 633         memcpy(dst, &state->stateid, sizeof(*dst));
 634         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 635                 return;
 636
 637         spin_lock(&state->state_lock);
 638         lsp = __nfs4_find_lock_state(state, fl_owner);
 639         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 640                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 641         spin_unlock(&state->state_lock);
 642         nfs4_put_lock_state(lsp);
 643 }
 644
 645 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 646 {
 647         struct rpc_sequence *sequence = counter->sequence;
 648         struct nfs_seqid *new;
 649
 650         new = kmalloc(sizeof(*new), GFP_KERNEL);
 651         if (new != NULL) {
 652                 new->sequence = counter;
 653                 spin_lock(&sequence->lock);
 654                 list_add_tail(&new->list, &sequence->list);
 655                 spin_unlock(&sequence->lock);
 656         }
 657         return new;
 658 }
 659
 660 void nfs_free_seqid(struct nfs_seqid *seqid)
 661 {
 662         struct rpc_sequence *sequence = seqid->sequence->sequence;
 663
 664         spin_lock(&sequence->lock);
 665         list_del(&seqid->list);
 666         spin_unlock(&sequence->lock);
 667         rpc_wake_up(&sequence->wait);
 668         kfree(seqid);
 669 }
 670
 671 /*
 672  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 673  * failed with a seqid incrementing error -
 674  * see comments nfs_fs.h:seqid_mutating_error()
 675  */
 676 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 677 {
 678         switch (status) {
 679                 case 0:
 680                         break;
 681                 case -NFS4ERR_BAD_SEQID:
 682                 case -NFS4ERR_STALE_CLIENTID:
 683                 case -NFS4ERR_STALE_STATEID:
 684                 case -NFS4ERR_BAD_STATEID:
 685                 case -NFS4ERR_BADXDR:
 686                 case -NFS4ERR_RESOURCE:
 687                 case -NFS4ERR_NOFILEHANDLE:
 688                         /* Non-seqid mutating errors */
 689                         return;
 690         };
 691         /*
 692          * Note: no locking needed as we are guaranteed to be first
 693          * on the sequence list
 694          */
 695         seqid->sequence->counter++;
 696 }
 697
 698 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 699 {
 700         if (status == -NFS4ERR_BAD_SEQID) {
 701                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 702                                 struct nfs4_state_owner, so_seqid);
 703                 nfs4_drop_state_owner(sp);
 704         }
 705         return nfs_increment_seqid(status, seqid);
 706 }
 707
 708 /*
 709  * Increment the seqid if the LOCK/LOCKU succeeded, or
 710  * failed with a seqid incrementing error -
 711  * see comments nfs_fs.h:seqid_mutating_error()
 712  */
 713 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 714 {
 715         return nfs_increment_seqid(status, seqid);
 716 }
 717
 718 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 719 {
 720         struct rpc_sequence *sequence = seqid->sequence->sequence;
 721         int status = 0;
 722
 723         if (sequence->list.next == &seqid->list)
 724                 goto out;
 725         spin_lock(&sequence->lock);
 726         if (sequence->list.next != &seqid->list) {
 727                 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
 728                 status = -EAGAIN;
 729         }
 730         spin_unlock(&sequence->lock);
 731 out:
 732         return status;
 733 }
 734
 735 static int reclaimer(void *);
 736
 737 static inline void nfs4_clear_recover_bit(struct nfs4_client *clp)
 738 {
 739         smp_mb__before_clear_bit();
 740         clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
 741         smp_mb__after_clear_bit();
 742         wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
 743         rpc_wake_up(&clp->cl_rpcwaitq);
 744 }
 745
 746 /*
 747  * State recovery routine
 748  */
 749 static void nfs4_recover_state(struct nfs4_client *clp)
 750 {
 751         struct task_struct *task;
 752
 753         __module_get(THIS_MODULE);
 754         atomic_inc(&clp->cl_count);
 755         task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
 756                         NIPQUAD(clp->cl_addr));
 757         if (!IS_ERR(task))
 758                 return;
 759         nfs4_clear_recover_bit(clp);
 760         nfs4_put_client(clp);
 761         module_put(THIS_MODULE);
 762 }
 763
 764 /*
 765  * Schedule a state recovery attempt
 766  */
 767 void nfs4_schedule_state_recovery(struct nfs4_client *clp)
 768 {
 769         if (!clp)
 770                 return;
 771         if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
 772                 nfs4_recover_state(clp);
 773 }
 774
 775 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
 776 {
 777         struct inode *inode = state->inode;
 778         struct file_lock *fl;
 779         int status = 0;
 780
 781         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
 782                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 783                         continue;
 784                 if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
 785                         continue;
 786                 status = ops->recover_lock(state, fl);
 787                 if (status >= 0)
 788                         continue;
 789                 switch (status) {
 790                         default:
 791                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 792                                                 __FUNCTION__, status);
 793                         case -NFS4ERR_EXPIRED:
 794                         case -NFS4ERR_NO_GRACE:
 795                         case -NFS4ERR_RECLAIM_BAD:
 796                         case -NFS4ERR_RECLAIM_CONFLICT:
 797                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 798                                 break;
 799                         case -NFS4ERR_STALE_CLIENTID:
 800                                 goto out_err;
 801                 }
 802         }
 803         return 0;
 804 out_err:
 805         return status;
 806 }
 807
 808 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
 809 {
 810         struct nfs4_state *state;
 811         struct nfs4_lock_state *lock;
 812         int status = 0;
 813
 814         /* Note: we rely on the sp->so_states list being ordered
 815          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 816          * states first.
 817          * This is needed to ensure that the server won't give us any
 818          * read delegations that we have to return if, say, we are
 819          * recovering after a network partition or a reboot from a
 820          * server that doesn't support a grace period.
 821          */
 822         list_for_each_entry(state, &sp->so_states, open_states) {
 823                 if (state->state == 0)
 824                         continue;
 825                 status = ops->recover_open(sp, state);
 826                 if (status >= 0) {
 827                         status = nfs4_reclaim_locks(ops, state);
 828                         if (status < 0)
 829                                 goto out_err;
 830                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 831                                 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 832                                         printk("%s: Lock reclaim failed!\n",
 833                                                         __FUNCTION__);
 834                         }
 835                         continue;
 836                 }
 837                 switch (status) {
 838                         default:
 839                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 840                                                 __FUNCTION__, status);
 841                         case -ENOENT:
 842                         case -NFS4ERR_RECLAIM_BAD:
 843                         case -NFS4ERR_RECLAIM_CONFLICT:
 844                                 /*
 845                                  * Open state on this file cannot be recovered
 846                                  * All we can do is revert to using the zero stateid.
 847                                  */
 848                                 memset(state->stateid.data, 0,
 849                                         sizeof(state->stateid.data));
 850                                 /* Mark the file as being 'closed' */
 851                                 state->state = 0;
 852                                 break;
 853                         case -NFS4ERR_EXPIRED:
 854                         case -NFS4ERR_NO_GRACE:
 855                         case -NFS4ERR_STALE_CLIENTID:
 856                                 goto out_err;
 857                 }
 858         }
 859         return 0;
 860 out_err:
 861         return status;
 862 }
 863
 864 static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
 865 {
 866         struct nfs4_state_owner *sp;
 867         struct nfs4_state *state;
 868         struct nfs4_lock_state *lock;
 869
 870         /* Reset all sequence ids to zero */
 871         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 872                 sp->so_seqid.counter = 0;
 873                 sp->so_seqid.flags = 0;
 874                 spin_lock(&sp->so_lock);
 875                 list_for_each_entry(state, &sp->so_states, open_states) {
 876                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 877                                 lock->ls_seqid.counter = 0;
 878                                 lock->ls_seqid.flags = 0;
 879                                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 880                         }
 881                 }
 882                 spin_unlock(&sp->so_lock);
 883         }
 884 }
 885
 886 static int reclaimer(void *ptr)
 887 {
 888         struct nfs4_client *clp = ptr;
 889         struct nfs4_state_owner *sp;
 890         struct nfs4_state_recovery_ops *ops;
 891         int status = 0;
 892
 893         allow_signal(SIGKILL);
 894
 895         /* Ensure exclusive access to NFSv4 state */
 896         lock_kernel();
 897         down_write(&clp->cl_sem);
 898         /* Are there any NFS mounts out there? */
 899         if (list_empty(&clp->cl_superblocks))
 900                 goto out;
 901 restart_loop:
 902         status = nfs4_proc_renew(clp);
 903         switch (status) {
 904                 case 0:
 905                 case -NFS4ERR_CB_PATH_DOWN:
 906                         goto out;
 907                 case -NFS4ERR_STALE_CLIENTID:
 908                 case -NFS4ERR_LEASE_MOVED:
 909                         ops = &nfs4_reboot_recovery_ops;
 910                         break;
 911                 default:
 912                         ops = &nfs4_network_partition_recovery_ops;
 913         };
 914         nfs4_state_mark_reclaim(clp);
 915         status = __nfs4_init_client(clp);
 916         if (status)
 917                 goto out_error;
 918         /* Mark all delegations for reclaim */
 919         nfs_delegation_mark_reclaim(clp);
 920         /* Note: list is protected by exclusive lock on cl->cl_sem */
 921         list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
 922                 status = nfs4_reclaim_open_state(ops, sp);
 923                 if (status < 0) {
 924                         if (status == -NFS4ERR_NO_GRACE) {
 925                                 ops = &nfs4_network_partition_recovery_ops;
 926                                 status = nfs4_reclaim_open_state(ops, sp);
 927                         }
 928                         if (status == -NFS4ERR_STALE_CLIENTID)
 929                                 goto restart_loop;
 930                         if (status == -NFS4ERR_EXPIRED)
 931                                 goto restart_loop;
 932                 }
 933         }
 934         nfs_delegation_reap_unclaimed(clp);
 935 out:
 936         up_write(&clp->cl_sem);
 937         unlock_kernel();
 938         if (status == -NFS4ERR_CB_PATH_DOWN)
 939                 nfs_handle_cb_pathdown(clp);
 940         nfs4_clear_recover_bit(clp);
 941         nfs4_put_client(clp);
 942         module_put_and_exit(0);
 943         return 0;
 944 out_error:
 945         printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
 946                                 NIPQUAD(clp->cl_addr.s_addr), -status);
 947         goto out;
 948 }
 949
 950 /*
 951  * Local variables:
 952  *  c-basic-offset: 8
 953  * End:
 954  */