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/kernel.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/random.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51
  52 #include "nfs4_fs.h"
  53 #include "callback.h"
  54 #include "delegation.h"
  55 #include "internal.h"
  56
  57 #define OPENOWNER_POOL_SIZE     8
  58
  59 const nfs4_stateid zero_stateid;
  60
  61 static LIST_HEAD(nfs4_clientid_list);
  62
  63 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
  64 {
  65         int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
  66                         nfs_callback_tcpport, cred);
  67         if (status == 0)
  68                 status = nfs4_proc_setclientid_confirm(clp, cred);
  69         if (status == 0)
  70                 nfs4_schedule_state_renewal(clp);
  71         return status;
  72 }
  73
  74 static struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  75 {
  76         struct rpc_cred *cred = NULL;
  77
  78         if (clp->cl_machine_cred != NULL)
  79                 cred = get_rpccred(clp->cl_machine_cred);
  80         return cred;
  81 }
  82
  83 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  84 {
  85         struct rpc_cred *cred;
  86
  87         spin_lock(&clp->cl_lock);
  88         cred = clp->cl_machine_cred;
  89         clp->cl_machine_cred = NULL;
  90         spin_unlock(&clp->cl_lock);
  91         if (cred != NULL)
  92                 put_rpccred(cred);
  93 }
  94
  95 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
  96 {
  97         struct nfs4_state_owner *sp;
  98         struct rb_node *pos;
  99         struct rpc_cred *cred = NULL;
 100
 101         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 102                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 103                 if (list_empty(&sp->so_states))
 104                         continue;
 105                 cred = get_rpccred(sp->so_cred);
 106                 break;
 107         }
 108         return cred;
 109 }
 110
 111 static struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
 112 {
 113         struct rpc_cred *cred;
 114
 115         spin_lock(&clp->cl_lock);
 116         cred = nfs4_get_renew_cred_locked(clp);
 117         spin_unlock(&clp->cl_lock);
 118         return cred;
 119 }
 120
 121 static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 122 {
 123         struct nfs4_state_owner *sp;
 124         struct rb_node *pos;
 125         struct rpc_cred *cred;
 126
 127         spin_lock(&clp->cl_lock);
 128         cred = nfs4_get_machine_cred_locked(clp);
 129         if (cred != NULL)
 130                 goto out;
 131         pos = rb_first(&clp->cl_state_owners);
 132         if (pos != NULL) {
 133                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 134                 cred = get_rpccred(sp->so_cred);
 135         }
 136 out:
 137         spin_unlock(&clp->cl_lock);
 138         return cred;
 139 }
 140
 141 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 142                 __u64 minval, int maxbits)
 143 {
 144         struct rb_node **p, *parent;
 145         struct nfs_unique_id *pos;
 146         __u64 mask = ~0ULL;
 147
 148         if (maxbits < 64)
 149                 mask = (1ULL << maxbits) - 1ULL;
 150
 151         /* Ensure distribution is more or less flat */
 152         get_random_bytes(&new->id, sizeof(new->id));
 153         new->id &= mask;
 154         if (new->id < minval)
 155                 new->id += minval;
 156 retry:
 157         p = &root->rb_node;
 158         parent = NULL;
 159
 160         while (*p != NULL) {
 161                 parent = *p;
 162                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 163
 164                 if (new->id < pos->id)
 165                         p = &(*p)->rb_left;
 166                 else if (new->id > pos->id)
 167                         p = &(*p)->rb_right;
 168                 else
 169                         goto id_exists;
 170         }
 171         rb_link_node(&new->rb_node, parent, p);
 172         rb_insert_color(&new->rb_node, root);
 173         return;
 174 id_exists:
 175         for (;;) {
 176                 new->id++;
 177                 if (new->id < minval || (new->id & mask) != new->id) {
 178                         new->id = minval;
 179                         break;
 180                 }
 181                 parent = rb_next(parent);
 182                 if (parent == NULL)
 183                         break;
 184                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 185                 if (new->id < pos->id)
 186                         break;
 187         }
 188         goto retry;
 189 }
 190
 191 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 192 {
 193         rb_erase(&id->rb_node, root);
 194 }
 195
 196 static struct nfs4_state_owner *
 197 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 198 {
 199         struct nfs_client *clp = server->nfs_client;
 200         struct rb_node **p = &clp->cl_state_owners.rb_node,
 201                        *parent = NULL;
 202         struct nfs4_state_owner *sp, *res = NULL;
 203
 204         while (*p != NULL) {
 205                 parent = *p;
 206                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 207
 208                 if (server < sp->so_server) {
 209                         p = &parent->rb_left;
 210                         continue;
 211                 }
 212                 if (server > sp->so_server) {
 213                         p = &parent->rb_right;
 214                         continue;
 215                 }
 216                 if (cred < sp->so_cred)
 217                         p = &parent->rb_left;
 218                 else if (cred > sp->so_cred)
 219                         p = &parent->rb_right;
 220                 else {
 221                         atomic_inc(&sp->so_count);
 222                         res = sp;
 223                         break;
 224                 }
 225         }
 226         return res;
 227 }
 228
 229 static struct nfs4_state_owner *
 230 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 231 {
 232         struct rb_node **p = &clp->cl_state_owners.rb_node,
 233                        *parent = NULL;
 234         struct nfs4_state_owner *sp;
 235
 236         while (*p != NULL) {
 237                 parent = *p;
 238                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 239
 240                 if (new->so_server < sp->so_server) {
 241                         p = &parent->rb_left;
 242                         continue;
 243                 }
 244                 if (new->so_server > sp->so_server) {
 245                         p = &parent->rb_right;
 246                         continue;
 247                 }
 248                 if (new->so_cred < sp->so_cred)
 249                         p = &parent->rb_left;
 250                 else if (new->so_cred > sp->so_cred)
 251                         p = &parent->rb_right;
 252                 else {
 253                         atomic_inc(&sp->so_count);
 254                         return sp;
 255                 }
 256         }
 257         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 258         rb_link_node(&new->so_client_node, parent, p);
 259         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 260         return new;
 261 }
 262
 263 static void
 264 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 265 {
 266         if (!RB_EMPTY_NODE(&sp->so_client_node))
 267                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 268         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 269 }
 270
 271 /*
 272  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 273  * create a new state_owner.
 274  *
 275  */
 276 static struct nfs4_state_owner *
 277 nfs4_alloc_state_owner(void)
 278 {
 279         struct nfs4_state_owner *sp;
 280
 281         sp = kzalloc(sizeof(*sp),GFP_KERNEL);
 282         if (!sp)
 283                 return NULL;
 284         spin_lock_init(&sp->so_lock);
 285         INIT_LIST_HEAD(&sp->so_states);
 286         INIT_LIST_HEAD(&sp->so_delegations);
 287         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 288         sp->so_seqid.sequence = &sp->so_sequence;
 289         spin_lock_init(&sp->so_sequence.lock);
 290         INIT_LIST_HEAD(&sp->so_sequence.list);
 291         atomic_set(&sp->so_count, 1);
 292         return sp;
 293 }
 294
 295 static void
 296 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 297 {
 298         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 299                 struct nfs_client *clp = sp->so_client;
 300
 301                 spin_lock(&clp->cl_lock);
 302                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 303                 RB_CLEAR_NODE(&sp->so_client_node);
 304                 spin_unlock(&clp->cl_lock);
 305         }
 306 }
 307
 308 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 309 {
 310         struct nfs_client *clp = server->nfs_client;
 311         struct nfs4_state_owner *sp, *new;
 312
 313         spin_lock(&clp->cl_lock);
 314         sp = nfs4_find_state_owner(server, cred);
 315         spin_unlock(&clp->cl_lock);
 316         if (sp != NULL)
 317                 return sp;
 318         new = nfs4_alloc_state_owner();
 319         if (new == NULL)
 320                 return NULL;
 321         new->so_client = clp;
 322         new->so_server = server;
 323         new->so_cred = cred;
 324         spin_lock(&clp->cl_lock);
 325         sp = nfs4_insert_state_owner(clp, new);
 326         spin_unlock(&clp->cl_lock);
 327         if (sp == new)
 328                 get_rpccred(cred);
 329         else {
 330                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 331                 kfree(new);
 332         }
 333         return sp;
 334 }
 335
 336 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 337 {
 338         struct nfs_client *clp = sp->so_client;
 339         struct rpc_cred *cred = sp->so_cred;
 340
 341         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 342                 return;
 343         nfs4_remove_state_owner(clp, sp);
 344         spin_unlock(&clp->cl_lock);
 345         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 346         put_rpccred(cred);
 347         kfree(sp);
 348 }
 349
 350 static struct nfs4_state *
 351 nfs4_alloc_open_state(void)
 352 {
 353         struct nfs4_state *state;
 354
 355         state = kzalloc(sizeof(*state), GFP_KERNEL);
 356         if (!state)
 357                 return NULL;
 358         atomic_set(&state->count, 1);
 359         INIT_LIST_HEAD(&state->lock_states);
 360         spin_lock_init(&state->state_lock);
 361         seqlock_init(&state->seqlock);
 362         return state;
 363 }
 364
 365 void
 366 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 367 {
 368         if (state->state == fmode)
 369                 return;
 370         /* NB! List reordering - see the reclaim code for why.  */
 371         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 372                 if (fmode & FMODE_WRITE)
 373                         list_move(&state->open_states, &state->owner->so_states);
 374                 else
 375                         list_move_tail(&state->open_states, &state->owner->so_states);
 376         }
 377         state->state = fmode;
 378 }
 379
 380 static struct nfs4_state *
 381 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 382 {
 383         struct nfs_inode *nfsi = NFS_I(inode);
 384         struct nfs4_state *state;
 385
 386         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 387                 if (state->owner != owner)
 388                         continue;
 389                 if (atomic_inc_not_zero(&state->count))
 390                         return state;
 391         }
 392         return NULL;
 393 }
 394
 395 static void
 396 nfs4_free_open_state(struct nfs4_state *state)
 397 {
 398         kfree(state);
 399 }
 400
 401 struct nfs4_state *
 402 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 403 {
 404         struct nfs4_state *state, *new;
 405         struct nfs_inode *nfsi = NFS_I(inode);
 406
 407         spin_lock(&inode->i_lock);
 408         state = __nfs4_find_state_byowner(inode, owner);
 409         spin_unlock(&inode->i_lock);
 410         if (state)
 411                 goto out;
 412         new = nfs4_alloc_open_state();
 413         spin_lock(&owner->so_lock);
 414         spin_lock(&inode->i_lock);
 415         state = __nfs4_find_state_byowner(inode, owner);
 416         if (state == NULL && new != NULL) {
 417                 state = new;
 418                 state->owner = owner;
 419                 atomic_inc(&owner->so_count);
 420                 list_add(&state->inode_states, &nfsi->open_states);
 421                 state->inode = igrab(inode);
 422                 spin_unlock(&inode->i_lock);
 423                 /* Note: The reclaim code dictates that we add stateless
 424                  * and read-only stateids to the end of the list */
 425                 list_add_tail(&state->open_states, &owner->so_states);
 426                 spin_unlock(&owner->so_lock);
 427         } else {
 428                 spin_unlock(&inode->i_lock);
 429                 spin_unlock(&owner->so_lock);
 430                 if (new)
 431                         nfs4_free_open_state(new);
 432         }
 433 out:
 434         return state;
 435 }
 436
 437 void nfs4_put_open_state(struct nfs4_state *state)
 438 {
 439         struct inode *inode = state->inode;
 440         struct nfs4_state_owner *owner = state->owner;
 441
 442         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 443                 return;
 444         spin_lock(&inode->i_lock);
 445         list_del(&state->inode_states);
 446         list_del(&state->open_states);
 447         spin_unlock(&inode->i_lock);
 448         spin_unlock(&owner->so_lock);
 449         iput(inode);
 450         nfs4_free_open_state(state);
 451         nfs4_put_state_owner(owner);
 452 }
 453
 454 /*
 455  * Close the current file.
 456  */
 457 static void __nfs4_close(struct path *path, struct nfs4_state *state, fmode_t fmode, int wait)
 458 {
 459         struct nfs4_state_owner *owner = state->owner;
 460         int call_close = 0;
 461         fmode_t newstate;
 462
 463         atomic_inc(&owner->so_count);
 464         /* Protect against nfs4_find_state() */
 465         spin_lock(&owner->so_lock);
 466         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 467                 case FMODE_READ:
 468                         state->n_rdonly--;
 469                         break;
 470                 case FMODE_WRITE:
 471                         state->n_wronly--;
 472                         break;
 473                 case FMODE_READ|FMODE_WRITE:
 474                         state->n_rdwr--;
 475         }
 476         newstate = FMODE_READ|FMODE_WRITE;
 477         if (state->n_rdwr == 0) {
 478                 if (state->n_rdonly == 0) {
 479                         newstate &= ~FMODE_READ;
 480                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 481                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 482                 }
 483                 if (state->n_wronly == 0) {
 484                         newstate &= ~FMODE_WRITE;
 485                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 486                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 487                 }
 488                 if (newstate == 0)
 489                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 490         }
 491         nfs4_state_set_mode_locked(state, newstate);
 492         spin_unlock(&owner->so_lock);
 493
 494         if (!call_close) {
 495                 nfs4_put_open_state(state);
 496                 nfs4_put_state_owner(owner);
 497         } else
 498                 nfs4_do_close(path, state, wait);
 499 }
 500
 501 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 502 {
 503         __nfs4_close(path, state, fmode, 0);
 504 }
 505
 506 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 507 {
 508         __nfs4_close(path, state, fmode, 1);
 509 }
 510
 511 /*
 512  * Search the state->lock_states for an existing lock_owner
 513  * that is compatible with current->files
 514  */
 515 static struct nfs4_lock_state *
 516 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 517 {
 518         struct nfs4_lock_state *pos;
 519         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 520                 if (pos->ls_owner != fl_owner)
 521                         continue;
 522                 atomic_inc(&pos->ls_count);
 523                 return pos;
 524         }
 525         return NULL;
 526 }
 527
 528 /*
 529  * Return a compatible lock_state. If no initialized lock_state structure
 530  * exists, return an uninitialized one.
 531  *
 532  */
 533 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 534 {
 535         struct nfs4_lock_state *lsp;
 536         struct nfs_client *clp = state->owner->so_client;
 537
 538         lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
 539         if (lsp == NULL)
 540                 return NULL;
 541         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 542         spin_lock_init(&lsp->ls_sequence.lock);
 543         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 544         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 545         atomic_set(&lsp->ls_count, 1);
 546         lsp->ls_owner = fl_owner;
 547         spin_lock(&clp->cl_lock);
 548         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 549         spin_unlock(&clp->cl_lock);
 550         INIT_LIST_HEAD(&lsp->ls_locks);
 551         return lsp;
 552 }
 553
 554 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 555 {
 556         struct nfs_client *clp = lsp->ls_state->owner->so_client;
 557
 558         spin_lock(&clp->cl_lock);
 559         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 560         spin_unlock(&clp->cl_lock);
 561         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 562         kfree(lsp);
 563 }
 564
 565 /*
 566  * Return a compatible lock_state. If no initialized lock_state structure
 567  * exists, return an uninitialized one.
 568  *
 569  */
 570 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 571 {
 572         struct nfs4_lock_state *lsp, *new = NULL;
 573
 574         for(;;) {
 575                 spin_lock(&state->state_lock);
 576                 lsp = __nfs4_find_lock_state(state, owner);
 577                 if (lsp != NULL)
 578                         break;
 579                 if (new != NULL) {
 580                         new->ls_state = state;
 581                         list_add(&new->ls_locks, &state->lock_states);
 582                         set_bit(LK_STATE_IN_USE, &state->flags);
 583                         lsp = new;
 584                         new = NULL;
 585                         break;
 586                 }
 587                 spin_unlock(&state->state_lock);
 588                 new = nfs4_alloc_lock_state(state, owner);
 589                 if (new == NULL)
 590                         return NULL;
 591         }
 592         spin_unlock(&state->state_lock);
 593         if (new != NULL)
 594                 nfs4_free_lock_state(new);
 595         return lsp;
 596 }
 597
 598 /*
 599  * Release reference to lock_state, and free it if we see that
 600  * it is no longer in use
 601  */
 602 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 603 {
 604         struct nfs4_state *state;
 605
 606         if (lsp == NULL)
 607                 return;
 608         state = lsp->ls_state;
 609         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 610                 return;
 611         list_del(&lsp->ls_locks);
 612         if (list_empty(&state->lock_states))
 613                 clear_bit(LK_STATE_IN_USE, &state->flags);
 614         spin_unlock(&state->state_lock);
 615         nfs4_free_lock_state(lsp);
 616 }
 617
 618 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 619 {
 620         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 621
 622         dst->fl_u.nfs4_fl.owner = lsp;
 623         atomic_inc(&lsp->ls_count);
 624 }
 625
 626 static void nfs4_fl_release_lock(struct file_lock *fl)
 627 {
 628         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 629 }
 630
 631 static struct file_lock_operations nfs4_fl_lock_ops = {
 632         .fl_copy_lock = nfs4_fl_copy_lock,
 633         .fl_release_private = nfs4_fl_release_lock,
 634 };
 635
 636 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 637 {
 638         struct nfs4_lock_state *lsp;
 639
 640         if (fl->fl_ops != NULL)
 641                 return 0;
 642         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 643         if (lsp == NULL)
 644                 return -ENOMEM;
 645         fl->fl_u.nfs4_fl.owner = lsp;
 646         fl->fl_ops = &nfs4_fl_lock_ops;
 647         return 0;
 648 }
 649
 650 /*
 651  * Byte-range lock aware utility to initialize the stateid of read/write
 652  * requests.
 653  */
 654 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
 655 {
 656         struct nfs4_lock_state *lsp;
 657         int seq;
 658
 659         do {
 660                 seq = read_seqbegin(&state->seqlock);
 661                 memcpy(dst, &state->stateid, sizeof(*dst));
 662         } while (read_seqretry(&state->seqlock, seq));
 663         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 664                 return;
 665
 666         spin_lock(&state->state_lock);
 667         lsp = __nfs4_find_lock_state(state, fl_owner);
 668         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 669                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 670         spin_unlock(&state->state_lock);
 671         nfs4_put_lock_state(lsp);
 672 }
 673
 674 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
 675 {
 676         struct nfs_seqid *new;
 677
 678         new = kmalloc(sizeof(*new), GFP_KERNEL);
 679         if (new != NULL) {
 680                 new->sequence = counter;
 681                 INIT_LIST_HEAD(&new->list);
 682         }
 683         return new;
 684 }
 685
 686 void nfs_free_seqid(struct nfs_seqid *seqid)
 687 {
 688         if (!list_empty(&seqid->list)) {
 689                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 690
 691                 spin_lock(&sequence->lock);
 692                 list_del(&seqid->list);
 693                 spin_unlock(&sequence->lock);
 694                 rpc_wake_up(&sequence->wait);
 695         }
 696         kfree(seqid);
 697 }
 698
 699 /*
 700  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 701  * failed with a seqid incrementing error -
 702  * see comments nfs_fs.h:seqid_mutating_error()
 703  */
 704 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 705 {
 706         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 707         switch (status) {
 708                 case 0:
 709                         break;
 710                 case -NFS4ERR_BAD_SEQID:
 711                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 712                                 return;
 713                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 714                                         " sequence-id error on an"
 715                                         " unconfirmed sequence %p!\n",
 716                                         seqid->sequence);
 717                 case -NFS4ERR_STALE_CLIENTID:
 718                 case -NFS4ERR_STALE_STATEID:
 719                 case -NFS4ERR_BAD_STATEID:
 720                 case -NFS4ERR_BADXDR:
 721                 case -NFS4ERR_RESOURCE:
 722                 case -NFS4ERR_NOFILEHANDLE:
 723                         /* Non-seqid mutating errors */
 724                         return;
 725         };
 726         /*
 727          * Note: no locking needed as we are guaranteed to be first
 728          * on the sequence list
 729          */
 730         seqid->sequence->counter++;
 731 }
 732
 733 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 734 {
 735         if (status == -NFS4ERR_BAD_SEQID) {
 736                 struct nfs4_state_owner *sp = container_of(seqid->sequence,
 737                                 struct nfs4_state_owner, so_seqid);
 738                 nfs4_drop_state_owner(sp);
 739         }
 740         nfs_increment_seqid(status, seqid);
 741 }
 742
 743 /*
 744  * Increment the seqid if the LOCK/LOCKU succeeded, or
 745  * failed with a seqid incrementing error -
 746  * see comments nfs_fs.h:seqid_mutating_error()
 747  */
 748 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 749 {
 750         nfs_increment_seqid(status, seqid);
 751 }
 752
 753 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 754 {
 755         struct rpc_sequence *sequence = seqid->sequence->sequence;
 756         int status = 0;
 757
 758         spin_lock(&sequence->lock);
 759         if (list_empty(&seqid->list))
 760                 list_add_tail(&seqid->list, &sequence->list);
 761         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 762                 goto unlock;
 763         rpc_sleep_on(&sequence->wait, task, NULL);
 764         status = -EAGAIN;
 765 unlock:
 766         spin_unlock(&sequence->lock);
 767         return status;
 768 }
 769
 770 static int nfs4_run_state_manager(void *);
 771
 772 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 773 {
 774         smp_mb__before_clear_bit();
 775         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 776         smp_mb__after_clear_bit();
 777         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 778         rpc_wake_up(&clp->cl_rpcwaitq);
 779 }
 780
 781 /*
 782  * Schedule the nfs_client asynchronous state management routine
 783  */
 784 void nfs4_schedule_state_manager(struct nfs_client *clp)
 785 {
 786         struct task_struct *task;
 787
 788         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
 789                 return;
 790         __module_get(THIS_MODULE);
 791         atomic_inc(&clp->cl_count);
 792         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
 793                                 rpc_peeraddr2str(clp->cl_rpcclient,
 794                                                         RPC_DISPLAY_ADDR));
 795         if (!IS_ERR(task))
 796                 return;
 797         nfs4_clear_state_manager_bit(clp);
 798         nfs_put_client(clp);
 799         module_put(THIS_MODULE);
 800 }
 801
 802 /*
 803  * Schedule a state recovery attempt
 804  */
 805 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 806 {
 807         if (!clp)
 808                 return;
 809         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
 810                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 811         nfs4_schedule_state_manager(clp);
 812 }
 813
 814 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
 815 {
 816
 817         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 818         /* Don't recover state that expired before the reboot */
 819         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
 820                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 821                 return 0;
 822         }
 823         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
 824         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
 825         return 1;
 826 }
 827
 828 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
 829 {
 830         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
 831         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 832         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
 833         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
 834         return 1;
 835 }
 836
 837 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 838 {
 839         struct inode *inode = state->inode;
 840         struct nfs_inode *nfsi = NFS_I(inode);
 841         struct file_lock *fl;
 842         int status = 0;
 843
 844         down_write(&nfsi->rwsem);
 845         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 846                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 847                         continue;
 848                 if (nfs_file_open_context(fl->fl_file)->state != state)
 849                         continue;
 850                 status = ops->recover_lock(state, fl);
 851                 if (status >= 0)
 852                         continue;
 853                 switch (status) {
 854                         default:
 855                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 856                                                 __func__, status);
 857                         case -NFS4ERR_EXPIRED:
 858                         case -NFS4ERR_NO_GRACE:
 859                         case -NFS4ERR_RECLAIM_BAD:
 860                         case -NFS4ERR_RECLAIM_CONFLICT:
 861                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
 862                                 break;
 863                         case -NFS4ERR_STALE_CLIENTID:
 864                                 goto out_err;
 865                 }
 866         }
 867         up_write(&nfsi->rwsem);
 868         return 0;
 869 out_err:
 870         up_write(&nfsi->rwsem);
 871         return status;
 872 }
 873
 874 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
 875 {
 876         struct nfs4_state *state;
 877         struct nfs4_lock_state *lock;
 878         int status = 0;
 879
 880         /* Note: we rely on the sp->so_states list being ordered
 881          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
 882          * states first.
 883          * This is needed to ensure that the server won't give us any
 884          * read delegations that we have to return if, say, we are
 885          * recovering after a network partition or a reboot from a
 886          * server that doesn't support a grace period.
 887          */
 888 restart:
 889         spin_lock(&sp->so_lock);
 890         list_for_each_entry(state, &sp->so_states, open_states) {
 891                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
 892                         continue;
 893                 if (state->state == 0)
 894                         continue;
 895                 atomic_inc(&state->count);
 896                 spin_unlock(&sp->so_lock);
 897                 status = ops->recover_open(sp, state);
 898                 if (status >= 0) {
 899                         status = nfs4_reclaim_locks(state, ops);
 900                         if (status >= 0) {
 901                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
 902                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
 903                                                 printk("%s: Lock reclaim failed!\n",
 904                                                         __func__);
 905                                 }
 906                                 nfs4_put_open_state(state);
 907                                 goto restart;
 908                         }
 909                 }
 910                 switch (status) {
 911                         default:
 912                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
 913                                                 __func__, status);
 914                         case -ENOENT:
 915                         case -ESTALE:
 916                                 /*
 917                                  * Open state on this file cannot be recovered
 918                                  * All we can do is revert to using the zero stateid.
 919                                  */
 920                                 memset(state->stateid.data, 0,
 921                                         sizeof(state->stateid.data));
 922                                 /* Mark the file as being 'closed' */
 923                                 state->state = 0;
 924                                 break;
 925                         case -NFS4ERR_RECLAIM_BAD:
 926                         case -NFS4ERR_RECLAIM_CONFLICT:
 927                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
 928                                 break;
 929                         case -NFS4ERR_EXPIRED:
 930                         case -NFS4ERR_NO_GRACE:
 931                                 nfs4_state_mark_reclaim_nograce(sp->so_client, state);
 932                         case -NFS4ERR_STALE_CLIENTID:
 933                                 goto out_err;
 934                 }
 935                 nfs4_put_open_state(state);
 936                 goto restart;
 937         }
 938         spin_unlock(&sp->so_lock);
 939         return 0;
 940 out_err:
 941         nfs4_put_open_state(state);
 942         return status;
 943 }
 944
 945 static void nfs4_clear_open_state(struct nfs4_state *state)
 946 {
 947         struct nfs4_lock_state *lock;
 948
 949         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 950         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
 951         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
 952         clear_bit(NFS_O_RDWR_STATE, &state->flags);
 953         list_for_each_entry(lock, &state->lock_states, ls_locks) {
 954                 lock->ls_seqid.flags = 0;
 955                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
 956         }
 957 }
 958
 959 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
 960 {
 961         struct nfs4_state_owner *sp;
 962         struct rb_node *pos;
 963         struct nfs4_state *state;
 964
 965         /* Reset all sequence ids to zero */
 966         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 967                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 968                 sp->so_seqid.flags = 0;
 969                 spin_lock(&sp->so_lock);
 970                 list_for_each_entry(state, &sp->so_states, open_states) {
 971                         if (mark_reclaim(clp, state))
 972                                 nfs4_clear_open_state(state);
 973                 }
 974                 spin_unlock(&sp->so_lock);
 975         }
 976 }
 977
 978 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
 979 {
 980         /* Mark all delegations for reclaim */
 981         nfs_delegation_mark_reclaim(clp);
 982         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
 983 }
 984
 985 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
 986 {
 987         struct nfs4_state_owner *sp;
 988         struct rb_node *pos;
 989         struct nfs4_state *state;
 990
 991         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
 992                 return;
 993
 994         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 995                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 996                 spin_lock(&sp->so_lock);
 997                 list_for_each_entry(state, &sp->so_states, open_states) {
 998                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
 999                                 continue;
1000                         nfs4_state_mark_reclaim_nograce(clp, state);
1001                 }
1002                 spin_unlock(&sp->so_lock);
1003         }
1004
1005         nfs_delegation_reap_unclaimed(clp);
1006 }
1007
1008 static void nfs_delegation_clear_all(struct nfs_client *clp)
1009 {
1010         nfs_delegation_mark_reclaim(clp);
1011         nfs_delegation_reap_unclaimed(clp);
1012 }
1013
1014 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1015 {
1016         nfs_delegation_clear_all(clp);
1017         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1018 }
1019
1020 static void nfs4_state_end_reclaim_nograce(struct nfs_client *clp)
1021 {
1022         clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1023 }
1024
1025 static void nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1026 {
1027         switch (error) {
1028                 case -NFS4ERR_CB_PATH_DOWN:
1029                         nfs_handle_cb_pathdown(clp);
1030                         break;
1031                 case -NFS4ERR_STALE_CLIENTID:
1032                 case -NFS4ERR_LEASE_MOVED:
1033                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1034                         nfs4_state_start_reclaim_reboot(clp);
1035                         break;
1036                 case -NFS4ERR_EXPIRED:
1037                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1038                         nfs4_state_start_reclaim_nograce(clp);
1039         }
1040 }
1041
1042 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1043 {
1044         struct rb_node *pos;
1045         int status = 0;
1046
1047 restart:
1048         spin_lock(&clp->cl_lock);
1049         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1050                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1051                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1052                         continue;
1053                 atomic_inc(&sp->so_count);
1054                 spin_unlock(&clp->cl_lock);
1055                 status = nfs4_reclaim_open_state(sp, ops);
1056                 if (status < 0) {
1057                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1058                         nfs4_put_state_owner(sp);
1059                         nfs4_recovery_handle_error(clp, status);
1060                         return status;
1061                 }
1062                 nfs4_put_state_owner(sp);
1063                 goto restart;
1064         }
1065         spin_unlock(&clp->cl_lock);
1066         return status;
1067 }
1068
1069 static int nfs4_check_lease(struct nfs_client *clp)
1070 {
1071         struct rpc_cred *cred;
1072         int status = -NFS4ERR_EXPIRED;
1073
1074         /* Is the client already known to have an expired lease? */
1075         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1076                 return 0;
1077         cred = nfs4_get_renew_cred(clp);
1078         if (cred == NULL) {
1079                 cred = nfs4_get_setclientid_cred(clp);
1080                 if (cred == NULL)
1081                         goto out;
1082         }
1083         status = nfs4_proc_renew(clp, cred);
1084         put_rpccred(cred);
1085 out:
1086         nfs4_recovery_handle_error(clp, status);
1087         return status;
1088 }
1089
1090 static int nfs4_reclaim_lease(struct nfs_client *clp)
1091 {
1092         struct rpc_cred *cred;
1093         int status = -ENOENT;
1094
1095         cred = nfs4_get_setclientid_cred(clp);
1096         if (cred != NULL) {
1097                 status = nfs4_init_client(clp, cred);
1098                 put_rpccred(cred);
1099                 /* Handle case where the user hasn't set up machine creds */
1100                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1101                         nfs4_clear_machine_cred(clp);
1102                         status = -EAGAIN;
1103                 }
1104         }
1105         return status;
1106 }
1107
1108 static void nfs4_state_manager(struct nfs_client *clp)
1109 {
1110         int status = 0;
1111
1112         /* Ensure exclusive access to NFSv4 state */
1113         for(;;) {
1114                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1115                         /* We're going to have to re-establish a clientid */
1116                         status = nfs4_reclaim_lease(clp);
1117                         if (status) {
1118                                 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1119                                 if (status == -EAGAIN)
1120                                         continue;
1121                                 goto out_error;
1122                         }
1123                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1124                 }
1125
1126                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1127                         status = nfs4_check_lease(clp);
1128                         if (status != 0)
1129                                 continue;
1130                 }
1131
1132                 /* First recover reboot state... */
1133                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1134                         status = nfs4_do_reclaim(clp, &nfs4_reboot_recovery_ops);
1135                         if (status == -NFS4ERR_STALE_CLIENTID)
1136                                 continue;
1137                         nfs4_state_end_reclaim_reboot(clp);
1138                         continue;
1139                 }
1140
1141                 /* Now recover expired state... */
1142                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1143                         status = nfs4_do_reclaim(clp, &nfs4_nograce_recovery_ops);
1144                         if (status < 0) {
1145                                 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1146                                 if (status == -NFS4ERR_STALE_CLIENTID)
1147                                         continue;
1148                                 if (status == -NFS4ERR_EXPIRED)
1149                                         continue;
1150                                 goto out_error;
1151                         } else
1152                                 nfs4_state_end_reclaim_nograce(clp);
1153                         continue;
1154                 }
1155
1156                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1157                         nfs_client_return_marked_delegations(clp);
1158                         continue;
1159                 }
1160
1161                 nfs4_clear_state_manager_bit(clp);
1162                 /* Did we race with an attempt to give us more work? */
1163                 if (clp->cl_state == 0)
1164                         break;
1165                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1166                         break;
1167         }
1168         return;
1169 out_error:
1170         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1171                         " with error %d\n", clp->cl_hostname, -status);
1172         if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1173                 nfs4_state_end_reclaim_reboot(clp);
1174         nfs4_clear_state_manager_bit(clp);
1175 }
1176
1177 static int nfs4_run_state_manager(void *ptr)
1178 {
1179         struct nfs_client *clp = ptr;
1180
1181         allow_signal(SIGKILL);
1182         nfs4_state_manager(clp);
1183         nfs_put_client(clp);
1184         module_put_and_exit(0);
1185         return 0;
1186 }
1187
1188 /*
1189  * Local variables:
1190  *  c-basic-offset: 8
1191  * End:
1192  */