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/fs.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/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52
  53 #include "nfs4_fs.h"
  54 #include "callback.h"
  55 #include "delegation.h"
  56 #include "internal.h"
  57 #include "pnfs.h"
  58
  59 #define OPENOWNER_POOL_SIZE     8
  60
  61 const nfs4_stateid zero_stateid;
  62
  63 static LIST_HEAD(nfs4_clientid_list);
  64
  65 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  66 {
  67         struct nfs4_setclientid_res clid;
  68         unsigned short port;
  69         int status;
  70
  71         port = nfs_callback_tcpport;
  72         if (clp->cl_addr.ss_family == AF_INET6)
  73                 port = nfs_callback_tcpport6;
  74
  75         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  76         if (status != 0)
  77                 goto out;
  78         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  79         if (status != 0)
  80                 goto out;
  81         clp->cl_clientid = clid.clientid;
  82         nfs4_schedule_state_renewal(clp);
  83 out:
  84         return status;
  85 }
  86
  87 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  88 {
  89         struct rpc_cred *cred = NULL;
  90
  91         if (clp->cl_machine_cred != NULL)
  92                 cred = get_rpccred(clp->cl_machine_cred);
  93         return cred;
  94 }
  95
  96 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  97 {
  98         struct rpc_cred *cred;
  99
 100         spin_lock(&clp->cl_lock);
 101         cred = clp->cl_machine_cred;
 102         clp->cl_machine_cred = NULL;
 103         spin_unlock(&clp->cl_lock);
 104         if (cred != NULL)
 105                 put_rpccred(cred);
 106 }
 107
 108 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 109 {
 110         struct nfs4_state_owner *sp;
 111         struct rb_node *pos;
 112         struct rpc_cred *cred = NULL;
 113
 114         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 115                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 116                 if (list_empty(&sp->so_states))
 117                         continue;
 118                 cred = get_rpccred(sp->so_cred);
 119                 break;
 120         }
 121         return cred;
 122 }
 123
 124 #if defined(CONFIG_NFS_V4_1)
 125
 126 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 127 {
 128         int status;
 129         struct nfs_fsinfo fsinfo;
 130
 131         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 132         if (status == 0) {
 133                 /* Update lease time and schedule renewal */
 134                 spin_lock(&clp->cl_lock);
 135                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 136                 clp->cl_last_renewal = jiffies;
 137                 spin_unlock(&clp->cl_lock);
 138
 139                 nfs4_schedule_state_renewal(clp);
 140         }
 141
 142         return status;
 143 }
 144
 145 static void nfs4_end_drain_session(struct nfs_client *clp)
 146 {
 147         struct nfs4_session *ses = clp->cl_session;
 148         int max_slots;
 149
 150         if (ses == NULL)
 151                 return;
 152         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 153                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 154                 max_slots = ses->fc_slot_table.max_slots;
 155                 while (max_slots--) {
 156                         struct rpc_task *task;
 157
 158                         task = rpc_wake_up_next(&ses->fc_slot_table.
 159                                                 slot_tbl_waitq);
 160                         if (!task)
 161                                 break;
 162                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 163                 }
 164                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 165         }
 166 }
 167
 168 static int nfs4_begin_drain_session(struct nfs_client *clp)
 169 {
 170         struct nfs4_session *ses = clp->cl_session;
 171         struct nfs4_slot_table *tbl = &ses->fc_slot_table;
 172
 173         spin_lock(&tbl->slot_tbl_lock);
 174         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 175         if (tbl->highest_used_slotid != -1) {
 176                 INIT_COMPLETION(ses->complete);
 177                 spin_unlock(&tbl->slot_tbl_lock);
 178                 return wait_for_completion_interruptible(&ses->complete);
 179         }
 180         spin_unlock(&tbl->slot_tbl_lock);
 181         return 0;
 182 }
 183
 184 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 185 {
 186         int status;
 187
 188         nfs4_begin_drain_session(clp);
 189         status = nfs4_proc_exchange_id(clp, cred);
 190         if (status != 0)
 191                 goto out;
 192         status = nfs4_proc_create_session(clp);
 193         if (status != 0)
 194                 goto out;
 195         nfs41_setup_state_renewal(clp);
 196         nfs_mark_client_ready(clp, NFS_CS_READY);
 197 out:
 198         return status;
 199 }
 200
 201 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 202 {
 203         struct rpc_cred *cred;
 204
 205         spin_lock(&clp->cl_lock);
 206         cred = nfs4_get_machine_cred_locked(clp);
 207         spin_unlock(&clp->cl_lock);
 208         return cred;
 209 }
 210
 211 #endif /* CONFIG_NFS_V4_1 */
 212
 213 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 214 {
 215         struct nfs4_state_owner *sp;
 216         struct rb_node *pos;
 217         struct rpc_cred *cred;
 218
 219         spin_lock(&clp->cl_lock);
 220         cred = nfs4_get_machine_cred_locked(clp);
 221         if (cred != NULL)
 222                 goto out;
 223         pos = rb_first(&clp->cl_state_owners);
 224         if (pos != NULL) {
 225                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 226                 cred = get_rpccred(sp->so_cred);
 227         }
 228 out:
 229         spin_unlock(&clp->cl_lock);
 230         return cred;
 231 }
 232
 233 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 234                 __u64 minval, int maxbits)
 235 {
 236         struct rb_node **p, *parent;
 237         struct nfs_unique_id *pos;
 238         __u64 mask = ~0ULL;
 239
 240         if (maxbits < 64)
 241                 mask = (1ULL << maxbits) - 1ULL;
 242
 243         /* Ensure distribution is more or less flat */
 244         get_random_bytes(&new->id, sizeof(new->id));
 245         new->id &= mask;
 246         if (new->id < minval)
 247                 new->id += minval;
 248 retry:
 249         p = &root->rb_node;
 250         parent = NULL;
 251
 252         while (*p != NULL) {
 253                 parent = *p;
 254                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 255
 256                 if (new->id < pos->id)
 257                         p = &(*p)->rb_left;
 258                 else if (new->id > pos->id)
 259                         p = &(*p)->rb_right;
 260                 else
 261                         goto id_exists;
 262         }
 263         rb_link_node(&new->rb_node, parent, p);
 264         rb_insert_color(&new->rb_node, root);
 265         return;
 266 id_exists:
 267         for (;;) {
 268                 new->id++;
 269                 if (new->id < minval || (new->id & mask) != new->id) {
 270                         new->id = minval;
 271                         break;
 272                 }
 273                 parent = rb_next(parent);
 274                 if (parent == NULL)
 275                         break;
 276                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 277                 if (new->id < pos->id)
 278                         break;
 279         }
 280         goto retry;
 281 }
 282
 283 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 284 {
 285         rb_erase(&id->rb_node, root);
 286 }
 287
 288 static struct nfs4_state_owner *
 289 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 290 {
 291         struct nfs_client *clp = server->nfs_client;
 292         struct rb_node **p = &clp->cl_state_owners.rb_node,
 293                        *parent = NULL;
 294         struct nfs4_state_owner *sp, *res = NULL;
 295
 296         while (*p != NULL) {
 297                 parent = *p;
 298                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 299
 300                 if (server < sp->so_server) {
 301                         p = &parent->rb_left;
 302                         continue;
 303                 }
 304                 if (server > sp->so_server) {
 305                         p = &parent->rb_right;
 306                         continue;
 307                 }
 308                 if (cred < sp->so_cred)
 309                         p = &parent->rb_left;
 310                 else if (cred > sp->so_cred)
 311                         p = &parent->rb_right;
 312                 else {
 313                         atomic_inc(&sp->so_count);
 314                         res = sp;
 315                         break;
 316                 }
 317         }
 318         return res;
 319 }
 320
 321 static struct nfs4_state_owner *
 322 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 323 {
 324         struct rb_node **p = &clp->cl_state_owners.rb_node,
 325                        *parent = NULL;
 326         struct nfs4_state_owner *sp;
 327
 328         while (*p != NULL) {
 329                 parent = *p;
 330                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 331
 332                 if (new->so_server < sp->so_server) {
 333                         p = &parent->rb_left;
 334                         continue;
 335                 }
 336                 if (new->so_server > sp->so_server) {
 337                         p = &parent->rb_right;
 338                         continue;
 339                 }
 340                 if (new->so_cred < sp->so_cred)
 341                         p = &parent->rb_left;
 342                 else if (new->so_cred > sp->so_cred)
 343                         p = &parent->rb_right;
 344                 else {
 345                         atomic_inc(&sp->so_count);
 346                         return sp;
 347                 }
 348         }
 349         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 350         rb_link_node(&new->so_client_node, parent, p);
 351         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 352         return new;
 353 }
 354
 355 static void
 356 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 357 {
 358         if (!RB_EMPTY_NODE(&sp->so_client_node))
 359                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 360         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 361 }
 362
 363 /*
 364  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 365  * create a new state_owner.
 366  *
 367  */
 368 static struct nfs4_state_owner *
 369 nfs4_alloc_state_owner(void)
 370 {
 371         struct nfs4_state_owner *sp;
 372
 373         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 374         if (!sp)
 375                 return NULL;
 376         spin_lock_init(&sp->so_lock);
 377         INIT_LIST_HEAD(&sp->so_states);
 378         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 379         sp->so_seqid.sequence = &sp->so_sequence;
 380         spin_lock_init(&sp->so_sequence.lock);
 381         INIT_LIST_HEAD(&sp->so_sequence.list);
 382         atomic_set(&sp->so_count, 1);
 383         return sp;
 384 }
 385
 386 static void
 387 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 388 {
 389         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 390                 struct nfs_client *clp = sp->so_server->nfs_client;
 391
 392                 spin_lock(&clp->cl_lock);
 393                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 394                 RB_CLEAR_NODE(&sp->so_client_node);
 395                 spin_unlock(&clp->cl_lock);
 396         }
 397 }
 398
 399 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 400 {
 401         struct nfs_client *clp = server->nfs_client;
 402         struct nfs4_state_owner *sp, *new;
 403
 404         spin_lock(&clp->cl_lock);
 405         sp = nfs4_find_state_owner(server, cred);
 406         spin_unlock(&clp->cl_lock);
 407         if (sp != NULL)
 408                 return sp;
 409         new = nfs4_alloc_state_owner();
 410         if (new == NULL)
 411                 return NULL;
 412         new->so_server = server;
 413         new->so_cred = cred;
 414         spin_lock(&clp->cl_lock);
 415         sp = nfs4_insert_state_owner(clp, new);
 416         spin_unlock(&clp->cl_lock);
 417         if (sp == new)
 418                 get_rpccred(cred);
 419         else {
 420                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 421                 kfree(new);
 422         }
 423         return sp;
 424 }
 425
 426 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 427 {
 428         struct nfs_client *clp = sp->so_server->nfs_client;
 429         struct rpc_cred *cred = sp->so_cred;
 430
 431         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 432                 return;
 433         nfs4_remove_state_owner(clp, sp);
 434         spin_unlock(&clp->cl_lock);
 435         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 436         put_rpccred(cred);
 437         kfree(sp);
 438 }
 439
 440 static struct nfs4_state *
 441 nfs4_alloc_open_state(void)
 442 {
 443         struct nfs4_state *state;
 444
 445         state = kzalloc(sizeof(*state), GFP_NOFS);
 446         if (!state)
 447                 return NULL;
 448         atomic_set(&state->count, 1);
 449         INIT_LIST_HEAD(&state->lock_states);
 450         spin_lock_init(&state->state_lock);
 451         seqlock_init(&state->seqlock);
 452         return state;
 453 }
 454
 455 void
 456 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 457 {
 458         if (state->state == fmode)
 459                 return;
 460         /* NB! List reordering - see the reclaim code for why.  */
 461         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 462                 if (fmode & FMODE_WRITE)
 463                         list_move(&state->open_states, &state->owner->so_states);
 464                 else
 465                         list_move_tail(&state->open_states, &state->owner->so_states);
 466         }
 467         state->state = fmode;
 468 }
 469
 470 static struct nfs4_state *
 471 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 472 {
 473         struct nfs_inode *nfsi = NFS_I(inode);
 474         struct nfs4_state *state;
 475
 476         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 477                 if (state->owner != owner)
 478                         continue;
 479                 if (atomic_inc_not_zero(&state->count))
 480                         return state;
 481         }
 482         return NULL;
 483 }
 484
 485 static void
 486 nfs4_free_open_state(struct nfs4_state *state)
 487 {
 488         kfree(state);
 489 }
 490
 491 struct nfs4_state *
 492 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 493 {
 494         struct nfs4_state *state, *new;
 495         struct nfs_inode *nfsi = NFS_I(inode);
 496
 497         spin_lock(&inode->i_lock);
 498         state = __nfs4_find_state_byowner(inode, owner);
 499         spin_unlock(&inode->i_lock);
 500         if (state)
 501                 goto out;
 502         new = nfs4_alloc_open_state();
 503         spin_lock(&owner->so_lock);
 504         spin_lock(&inode->i_lock);
 505         state = __nfs4_find_state_byowner(inode, owner);
 506         if (state == NULL && new != NULL) {
 507                 state = new;
 508                 state->owner = owner;
 509                 atomic_inc(&owner->so_count);
 510                 list_add(&state->inode_states, &nfsi->open_states);
 511                 state->inode = igrab(inode);
 512                 spin_unlock(&inode->i_lock);
 513                 /* Note: The reclaim code dictates that we add stateless
 514                  * and read-only stateids to the end of the list */
 515                 list_add_tail(&state->open_states, &owner->so_states);
 516                 spin_unlock(&owner->so_lock);
 517         } else {
 518                 spin_unlock(&inode->i_lock);
 519                 spin_unlock(&owner->so_lock);
 520                 if (new)
 521                         nfs4_free_open_state(new);
 522         }
 523 out:
 524         return state;
 525 }
 526
 527 void nfs4_put_open_state(struct nfs4_state *state)
 528 {
 529         struct inode *inode = state->inode;
 530         struct nfs4_state_owner *owner = state->owner;
 531
 532         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 533                 return;
 534         spin_lock(&inode->i_lock);
 535         list_del(&state->inode_states);
 536         list_del(&state->open_states);
 537         spin_unlock(&inode->i_lock);
 538         spin_unlock(&owner->so_lock);
 539         iput(inode);
 540         nfs4_free_open_state(state);
 541         nfs4_put_state_owner(owner);
 542 }
 543
 544 /*
 545  * Close the current file.
 546  */
 547 static void __nfs4_close(struct path *path, struct nfs4_state *state,
 548                 fmode_t fmode, gfp_t gfp_mask, int wait)
 549 {
 550         struct nfs4_state_owner *owner = state->owner;
 551         int call_close = 0;
 552         fmode_t newstate;
 553
 554         atomic_inc(&owner->so_count);
 555         /* Protect against nfs4_find_state() */
 556         spin_lock(&owner->so_lock);
 557         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 558                 case FMODE_READ:
 559                         state->n_rdonly--;
 560                         break;
 561                 case FMODE_WRITE:
 562                         state->n_wronly--;
 563                         break;
 564                 case FMODE_READ|FMODE_WRITE:
 565                         state->n_rdwr--;
 566         }
 567         newstate = FMODE_READ|FMODE_WRITE;
 568         if (state->n_rdwr == 0) {
 569                 if (state->n_rdonly == 0) {
 570                         newstate &= ~FMODE_READ;
 571                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 572                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 573                 }
 574                 if (state->n_wronly == 0) {
 575                         newstate &= ~FMODE_WRITE;
 576                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 577                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 578                 }
 579                 if (newstate == 0)
 580                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 581         }
 582         nfs4_state_set_mode_locked(state, newstate);
 583         spin_unlock(&owner->so_lock);
 584
 585         if (!call_close) {
 586                 nfs4_put_open_state(state);
 587                 nfs4_put_state_owner(owner);
 588         } else
 589                 nfs4_do_close(path, state, gfp_mask, wait);
 590 }
 591
 592 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 593 {
 594         __nfs4_close(path, state, fmode, GFP_NOFS, 0);
 595 }
 596
 597 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 598 {
 599         __nfs4_close(path, state, fmode, GFP_KERNEL, 1);
 600 }
 601
 602 /*
 603  * Search the state->lock_states for an existing lock_owner
 604  * that is compatible with current->files
 605  */
 606 static struct nfs4_lock_state *
 607 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 608 {
 609         struct nfs4_lock_state *pos;
 610         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 611                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 612                         continue;
 613                 switch (pos->ls_owner.lo_type) {
 614                 case NFS4_POSIX_LOCK_TYPE:
 615                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 616                                 continue;
 617                         break;
 618                 case NFS4_FLOCK_LOCK_TYPE:
 619                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 620                                 continue;
 621                 }
 622                 atomic_inc(&pos->ls_count);
 623                 return pos;
 624         }
 625         return NULL;
 626 }
 627
 628 /*
 629  * Return a compatible lock_state. If no initialized lock_state structure
 630  * exists, return an uninitialized one.
 631  *
 632  */
 633 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 634 {
 635         struct nfs4_lock_state *lsp;
 636         struct nfs_client *clp = state->owner->so_server->nfs_client;
 637
 638         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 639         if (lsp == NULL)
 640                 return NULL;
 641         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 642         spin_lock_init(&lsp->ls_sequence.lock);
 643         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 644         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 645         atomic_set(&lsp->ls_count, 1);
 646         lsp->ls_state = state;
 647         lsp->ls_owner.lo_type = type;
 648         switch (lsp->ls_owner.lo_type) {
 649         case NFS4_FLOCK_LOCK_TYPE:
 650                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 651                 break;
 652         case NFS4_POSIX_LOCK_TYPE:
 653                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 654                 break;
 655         default:
 656                 kfree(lsp);
 657                 return NULL;
 658         }
 659         spin_lock(&clp->cl_lock);
 660         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 661         spin_unlock(&clp->cl_lock);
 662         INIT_LIST_HEAD(&lsp->ls_locks);
 663         return lsp;
 664 }
 665
 666 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 667 {
 668         struct nfs_client *clp = lsp->ls_state->owner->so_server->nfs_client;
 669
 670         spin_lock(&clp->cl_lock);
 671         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 672         spin_unlock(&clp->cl_lock);
 673         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 674         kfree(lsp);
 675 }
 676
 677 /*
 678  * Return a compatible lock_state. If no initialized lock_state structure
 679  * exists, return an uninitialized one.
 680  *
 681  */
 682 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 683 {
 684         struct nfs4_lock_state *lsp, *new = NULL;
 685
 686         for(;;) {
 687                 spin_lock(&state->state_lock);
 688                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 689                 if (lsp != NULL)
 690                         break;
 691                 if (new != NULL) {
 692                         list_add(&new->ls_locks, &state->lock_states);
 693                         set_bit(LK_STATE_IN_USE, &state->flags);
 694                         lsp = new;
 695                         new = NULL;
 696                         break;
 697                 }
 698                 spin_unlock(&state->state_lock);
 699                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 700                 if (new == NULL)
 701                         return NULL;
 702         }
 703         spin_unlock(&state->state_lock);
 704         if (new != NULL)
 705                 nfs4_free_lock_state(new);
 706         return lsp;
 707 }
 708
 709 /*
 710  * Release reference to lock_state, and free it if we see that
 711  * it is no longer in use
 712  */
 713 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 714 {
 715         struct nfs4_state *state;
 716
 717         if (lsp == NULL)
 718                 return;
 719         state = lsp->ls_state;
 720         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 721                 return;
 722         list_del(&lsp->ls_locks);
 723         if (list_empty(&state->lock_states))
 724                 clear_bit(LK_STATE_IN_USE, &state->flags);
 725         spin_unlock(&state->state_lock);
 726         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 727                 nfs4_release_lockowner(lsp);
 728         nfs4_free_lock_state(lsp);
 729 }
 730
 731 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 732 {
 733         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 734
 735         dst->fl_u.nfs4_fl.owner = lsp;
 736         atomic_inc(&lsp->ls_count);
 737 }
 738
 739 static void nfs4_fl_release_lock(struct file_lock *fl)
 740 {
 741         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 742 }
 743
 744 static const struct file_lock_operations nfs4_fl_lock_ops = {
 745         .fl_copy_lock = nfs4_fl_copy_lock,
 746         .fl_release_private = nfs4_fl_release_lock,
 747 };
 748
 749 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 750 {
 751         struct nfs4_lock_state *lsp;
 752
 753         if (fl->fl_ops != NULL)
 754                 return 0;
 755         if (fl->fl_flags & FL_POSIX)
 756                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 757         else if (fl->fl_flags & FL_FLOCK)
 758                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 759         else
 760                 return -EINVAL;
 761         if (lsp == NULL)
 762                 return -ENOMEM;
 763         fl->fl_u.nfs4_fl.owner = lsp;
 764         fl->fl_ops = &nfs4_fl_lock_ops;
 765         return 0;
 766 }
 767
 768 /*
 769  * Byte-range lock aware utility to initialize the stateid of read/write
 770  * requests.
 771  */
 772 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 773 {
 774         struct nfs4_lock_state *lsp;
 775         int seq;
 776
 777         do {
 778                 seq = read_seqbegin(&state->seqlock);
 779                 memcpy(dst, &state->stateid, sizeof(*dst));
 780         } while (read_seqretry(&state->seqlock, seq));
 781         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 782                 return;
 783
 784         spin_lock(&state->state_lock);
 785         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 786         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 787                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 788         spin_unlock(&state->state_lock);
 789         nfs4_put_lock_state(lsp);
 790 }
 791
 792 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 793 {
 794         struct nfs_seqid *new;
 795
 796         new = kmalloc(sizeof(*new), gfp_mask);
 797         if (new != NULL) {
 798                 new->sequence = counter;
 799                 INIT_LIST_HEAD(&new->list);
 800         }
 801         return new;
 802 }
 803
 804 void nfs_release_seqid(struct nfs_seqid *seqid)
 805 {
 806         if (!list_empty(&seqid->list)) {
 807                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 808
 809                 spin_lock(&sequence->lock);
 810                 list_del_init(&seqid->list);
 811                 spin_unlock(&sequence->lock);
 812                 rpc_wake_up(&sequence->wait);
 813         }
 814 }
 815
 816 void nfs_free_seqid(struct nfs_seqid *seqid)
 817 {
 818         nfs_release_seqid(seqid);
 819         kfree(seqid);
 820 }
 821
 822 /*
 823  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 824  * failed with a seqid incrementing error -
 825  * see comments nfs_fs.h:seqid_mutating_error()
 826  */
 827 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 828 {
 829         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 830         switch (status) {
 831                 case 0:
 832                         break;
 833                 case -NFS4ERR_BAD_SEQID:
 834                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 835                                 return;
 836                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 837                                         " sequence-id error on an"
 838                                         " unconfirmed sequence %p!\n",
 839                                         seqid->sequence);
 840                 case -NFS4ERR_STALE_CLIENTID:
 841                 case -NFS4ERR_STALE_STATEID:
 842                 case -NFS4ERR_BAD_STATEID:
 843                 case -NFS4ERR_BADXDR:
 844                 case -NFS4ERR_RESOURCE:
 845                 case -NFS4ERR_NOFILEHANDLE:
 846                         /* Non-seqid mutating errors */
 847                         return;
 848         };
 849         /*
 850          * Note: no locking needed as we are guaranteed to be first
 851          * on the sequence list
 852          */
 853         seqid->sequence->counter++;
 854 }
 855
 856 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 857 {
 858         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 859                                         struct nfs4_state_owner, so_seqid);
 860         struct nfs_server *server = sp->so_server;
 861
 862         if (status == -NFS4ERR_BAD_SEQID)
 863                 nfs4_drop_state_owner(sp);
 864         if (!nfs4_has_session(server->nfs_client))
 865                 nfs_increment_seqid(status, seqid);
 866 }
 867
 868 /*
 869  * Increment the seqid if the LOCK/LOCKU succeeded, or
 870  * failed with a seqid incrementing error -
 871  * see comments nfs_fs.h:seqid_mutating_error()
 872  */
 873 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 874 {
 875         nfs_increment_seqid(status, seqid);
 876 }
 877
 878 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 879 {
 880         struct rpc_sequence *sequence = seqid->sequence->sequence;
 881         int status = 0;
 882
 883         spin_lock(&sequence->lock);
 884         if (list_empty(&seqid->list))
 885                 list_add_tail(&seqid->list, &sequence->list);
 886         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 887                 goto unlock;
 888         rpc_sleep_on(&sequence->wait, task, NULL);
 889         status = -EAGAIN;
 890 unlock:
 891         spin_unlock(&sequence->lock);
 892         return status;
 893 }
 894
 895 static int nfs4_run_state_manager(void *);
 896
 897 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 898 {
 899         smp_mb__before_clear_bit();
 900         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 901         smp_mb__after_clear_bit();
 902         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 903         rpc_wake_up(&clp->cl_rpcwaitq);
 904 }
 905
 906 /*
 907  * Schedule the nfs_client asynchronous state management routine
 908  */
 909 void nfs4_schedule_state_manager(struct nfs_client *clp)
 910 {
 911         struct task_struct *task;
 912
 913         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
 914                 return;
 915         __module_get(THIS_MODULE);
 916         atomic_inc(&clp->cl_count);
 917         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
 918                                 rpc_peeraddr2str(clp->cl_rpcclient,
 919                                                         RPC_DISPLAY_ADDR));
 920         if (!IS_ERR(task))
 921                 return;
 922         nfs4_clear_state_manager_bit(clp);
 923         nfs_put_client(clp);
 924         module_put(THIS_MODULE);
 925 }
 926
 927 /*
 928  * Schedule a state recovery attempt
 929  */
 930 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 931 {
 932         if (!clp)
 933                 return;
 934         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
 935                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 936         nfs4_schedule_state_manager(clp);
 937 }
 938
 939 int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
 940 {
 941
 942         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 943         /* Don't recover state that expired before the reboot */
 944         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
 945                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 946                 return 0;
 947         }
 948         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
 949         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
 950         return 1;
 951 }
 952
 953 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
 954 {
 955         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
 956         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 957         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
 958         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
 959         return 1;
 960 }
 961
 962 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 963 {
 964         struct inode *inode = state->inode;
 965         struct nfs_inode *nfsi = NFS_I(inode);
 966         struct file_lock *fl;
 967         int status = 0;
 968
 969         if (inode->i_flock == NULL)
 970                 return 0;
 971
 972         /* Guard against delegation returns and new lock/unlock calls */
 973         down_write(&nfsi->rwsem);
 974         /* Protect inode->i_flock using the BKL */
 975         lock_flocks();
 976         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 977                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 978                         continue;
 979                 if (nfs_file_open_context(fl->fl_file)->state != state)
 980                         continue;
 981                 unlock_flocks();
 982                 status = ops->recover_lock(state, fl);
 983                 switch (status) {
 984                         case 0:
 985                                 break;
 986                         case -ESTALE:
 987                         case -NFS4ERR_ADMIN_REVOKED:
 988                         case -NFS4ERR_STALE_STATEID:
 989                         case -NFS4ERR_BAD_STATEID:
 990                         case -NFS4ERR_EXPIRED:
 991                         case -NFS4ERR_NO_GRACE:
 992                         case -NFS4ERR_STALE_CLIENTID:
 993                         case -NFS4ERR_BADSESSION:
 994                         case -NFS4ERR_BADSLOT:
 995                         case -NFS4ERR_BAD_HIGH_SLOT:
 996                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
 997                                 goto out;
 998                         default:
 999                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1000                                                 __func__, status);
1001                         case -ENOMEM:
1002                         case -NFS4ERR_DENIED:
1003                         case -NFS4ERR_RECLAIM_BAD:
1004                         case -NFS4ERR_RECLAIM_CONFLICT:
1005                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1006                                 status = 0;
1007                 }
1008                 lock_flocks();
1009         }
1010         unlock_flocks();
1011 out:
1012         up_write(&nfsi->rwsem);
1013         return status;
1014 }
1015
1016 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1017 {
1018         struct nfs4_state *state;
1019         struct nfs4_lock_state *lock;
1020         int status = 0;
1021
1022         /* Note: we rely on the sp->so_states list being ordered
1023          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1024          * states first.
1025          * This is needed to ensure that the server won't give us any
1026          * read delegations that we have to return if, say, we are
1027          * recovering after a network partition or a reboot from a
1028          * server that doesn't support a grace period.
1029          */
1030 restart:
1031         spin_lock(&sp->so_lock);
1032         list_for_each_entry(state, &sp->so_states, open_states) {
1033                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1034                         continue;
1035                 if (state->state == 0)
1036                         continue;
1037                 atomic_inc(&state->count);
1038                 spin_unlock(&sp->so_lock);
1039                 status = ops->recover_open(sp, state);
1040                 if (status >= 0) {
1041                         status = nfs4_reclaim_locks(state, ops);
1042                         if (status >= 0) {
1043                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1044                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1045                                                 printk("%s: Lock reclaim failed!\n",
1046                                                         __func__);
1047                                 }
1048                                 nfs4_put_open_state(state);
1049                                 goto restart;
1050                         }
1051                 }
1052                 switch (status) {
1053                         default:
1054                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1055                                                 __func__, status);
1056                         case -ENOENT:
1057                         case -ENOMEM:
1058                         case -ESTALE:
1059                                 /*
1060                                  * Open state on this file cannot be recovered
1061                                  * All we can do is revert to using the zero stateid.
1062                                  */
1063                                 memset(state->stateid.data, 0,
1064                                         sizeof(state->stateid.data));
1065                                 /* Mark the file as being 'closed' */
1066                                 state->state = 0;
1067                                 break;
1068                         case -EKEYEXPIRED:
1069                                 /*
1070                                  * User RPCSEC_GSS context has expired.
1071                                  * We cannot recover this stateid now, so
1072                                  * skip it and allow recovery thread to
1073                                  * proceed.
1074                                  */
1075                                 break;
1076                         case -NFS4ERR_ADMIN_REVOKED:
1077                         case -NFS4ERR_STALE_STATEID:
1078                         case -NFS4ERR_BAD_STATEID:
1079                         case -NFS4ERR_RECLAIM_BAD:
1080                         case -NFS4ERR_RECLAIM_CONFLICT:
1081                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1082                                 break;
1083                         case -NFS4ERR_EXPIRED:
1084                         case -NFS4ERR_NO_GRACE:
1085                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1086                         case -NFS4ERR_STALE_CLIENTID:
1087                         case -NFS4ERR_BADSESSION:
1088                         case -NFS4ERR_BADSLOT:
1089                         case -NFS4ERR_BAD_HIGH_SLOT:
1090                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1091                                 goto out_err;
1092                 }
1093                 nfs4_put_open_state(state);
1094                 goto restart;
1095         }
1096         spin_unlock(&sp->so_lock);
1097         return 0;
1098 out_err:
1099         nfs4_put_open_state(state);
1100         return status;
1101 }
1102
1103 static void nfs4_clear_open_state(struct nfs4_state *state)
1104 {
1105         struct nfs4_lock_state *lock;
1106
1107         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1108         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1109         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1110         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1111         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1112                 lock->ls_seqid.flags = 0;
1113                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1114         }
1115 }
1116
1117 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1118 {
1119         struct nfs4_state_owner *sp;
1120         struct rb_node *pos;
1121         struct nfs4_state *state;
1122
1123         /* Reset all sequence ids to zero */
1124         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1125                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1126                 sp->so_seqid.flags = 0;
1127                 spin_lock(&sp->so_lock);
1128                 list_for_each_entry(state, &sp->so_states, open_states) {
1129                         if (mark_reclaim(clp, state))
1130                                 nfs4_clear_open_state(state);
1131                 }
1132                 spin_unlock(&sp->so_lock);
1133         }
1134 }
1135
1136 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1137 {
1138         /* Mark all delegations for reclaim */
1139         nfs_delegation_mark_reclaim(clp);
1140         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1141 }
1142
1143 static void nfs4_reclaim_complete(struct nfs_client *clp,
1144                                  const struct nfs4_state_recovery_ops *ops)
1145 {
1146         /* Notify the server we're done reclaiming our state */
1147         if (ops->reclaim_complete)
1148                 (void)ops->reclaim_complete(clp);
1149 }
1150
1151 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1152 {
1153         struct nfs4_state_owner *sp;
1154         struct rb_node *pos;
1155         struct nfs4_state *state;
1156
1157         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1158                 return 0;
1159
1160         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1161                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1162                 spin_lock(&sp->so_lock);
1163                 list_for_each_entry(state, &sp->so_states, open_states) {
1164                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1165                                 continue;
1166                         nfs4_state_mark_reclaim_nograce(clp, state);
1167                 }
1168                 spin_unlock(&sp->so_lock);
1169         }
1170
1171         nfs_delegation_reap_unclaimed(clp);
1172         return 1;
1173 }
1174
1175 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1176 {
1177         if (!nfs4_state_clear_reclaim_reboot(clp))
1178                 return;
1179         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1180 }
1181
1182 static void nfs_delegation_clear_all(struct nfs_client *clp)
1183 {
1184         nfs_delegation_mark_reclaim(clp);
1185         nfs_delegation_reap_unclaimed(clp);
1186 }
1187
1188 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1189 {
1190         nfs_delegation_clear_all(clp);
1191         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1192 }
1193
1194 static void nfs4_warn_keyexpired(const char *s)
1195 {
1196         printk_ratelimited(KERN_WARNING "Error: state manager"
1197                         " encountered RPCSEC_GSS session"
1198                         " expired against NFSv4 server %s.\n",
1199                         s);
1200 }
1201
1202 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1203 {
1204         switch (error) {
1205                 case -NFS4ERR_CB_PATH_DOWN:
1206                         nfs_handle_cb_pathdown(clp);
1207                         return 0;
1208                 case -NFS4ERR_NO_GRACE:
1209                         nfs4_state_end_reclaim_reboot(clp);
1210                         return 0;
1211                 case -NFS4ERR_STALE_CLIENTID:
1212                 case -NFS4ERR_LEASE_MOVED:
1213                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1214                         nfs4_state_clear_reclaim_reboot(clp);
1215                         nfs4_state_start_reclaim_reboot(clp);
1216                         break;
1217                 case -NFS4ERR_EXPIRED:
1218                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1219                         nfs4_state_start_reclaim_nograce(clp);
1220                         break;
1221                 case -NFS4ERR_BADSESSION:
1222                 case -NFS4ERR_BADSLOT:
1223                 case -NFS4ERR_BAD_HIGH_SLOT:
1224                 case -NFS4ERR_DEADSESSION:
1225                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1226                 case -NFS4ERR_SEQ_FALSE_RETRY:
1227                 case -NFS4ERR_SEQ_MISORDERED:
1228                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1229                         /* Zero session reset errors */
1230                         return 0;
1231                 case -EKEYEXPIRED:
1232                         /* Nothing we can do */
1233                         nfs4_warn_keyexpired(clp->cl_hostname);
1234                         return 0;
1235         }
1236         return error;
1237 }
1238
1239 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1240 {
1241         struct rb_node *pos;
1242         int status = 0;
1243
1244 restart:
1245         spin_lock(&clp->cl_lock);
1246         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1247                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1248                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1249                         continue;
1250                 atomic_inc(&sp->so_count);
1251                 spin_unlock(&clp->cl_lock);
1252                 status = nfs4_reclaim_open_state(sp, ops);
1253                 if (status < 0) {
1254                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1255                         nfs4_put_state_owner(sp);
1256                         return nfs4_recovery_handle_error(clp, status);
1257                 }
1258                 nfs4_put_state_owner(sp);
1259                 goto restart;
1260         }
1261         spin_unlock(&clp->cl_lock);
1262         return status;
1263 }
1264
1265 static int nfs4_check_lease(struct nfs_client *clp)
1266 {
1267         struct rpc_cred *cred;
1268         const struct nfs4_state_maintenance_ops *ops =
1269                 clp->cl_mvops->state_renewal_ops;
1270         int status = -NFS4ERR_EXPIRED;
1271
1272         /* Is the client already known to have an expired lease? */
1273         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1274                 return 0;
1275         spin_lock(&clp->cl_lock);
1276         cred = ops->get_state_renewal_cred_locked(clp);
1277         spin_unlock(&clp->cl_lock);
1278         if (cred == NULL) {
1279                 cred = nfs4_get_setclientid_cred(clp);
1280                 if (cred == NULL)
1281                         goto out;
1282         }
1283         status = ops->renew_lease(clp, cred);
1284         put_rpccred(cred);
1285 out:
1286         return nfs4_recovery_handle_error(clp, status);
1287 }
1288
1289 static int nfs4_reclaim_lease(struct nfs_client *clp)
1290 {
1291         struct rpc_cred *cred;
1292         const struct nfs4_state_recovery_ops *ops =
1293                 clp->cl_mvops->reboot_recovery_ops;
1294         int status = -ENOENT;
1295
1296         cred = ops->get_clid_cred(clp);
1297         if (cred != NULL) {
1298                 status = ops->establish_clid(clp, cred);
1299                 put_rpccred(cred);
1300                 /* Handle case where the user hasn't set up machine creds */
1301                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1302                         nfs4_clear_machine_cred(clp);
1303                         status = -EAGAIN;
1304                 }
1305                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1306                         status = -EPROTONOSUPPORT;
1307         }
1308         return status;
1309 }
1310
1311 #ifdef CONFIG_NFS_V4_1
1312 void nfs41_handle_recall_slot(struct nfs_client *clp)
1313 {
1314         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1315         nfs4_schedule_state_recovery(clp);
1316 }
1317
1318 static void nfs4_reset_all_state(struct nfs_client *clp)
1319 {
1320         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1321                 clp->cl_boot_time = CURRENT_TIME;
1322                 nfs4_state_start_reclaim_nograce(clp);
1323                 nfs4_schedule_state_recovery(clp);
1324         }
1325 }
1326
1327 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1328 {
1329         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1330                 nfs4_state_start_reclaim_reboot(clp);
1331                 nfs4_schedule_state_recovery(clp);
1332         }
1333 }
1334
1335 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1336 {
1337         /* Temporary */
1338         nfs4_reset_all_state(clp);
1339 }
1340
1341 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1342 {
1343         /* This will need to handle layouts too */
1344         nfs_expire_all_delegations(clp);
1345 }
1346
1347 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1348 {
1349         nfs_expire_all_delegations(clp);
1350         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1351                 nfs4_schedule_state_recovery(clp);
1352 }
1353
1354 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1355 {
1356         if (!flags)
1357                 return;
1358         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1359                 nfs41_handle_server_reboot(clp);
1360         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1361                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1362                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1363                             SEQ4_STATUS_LEASE_MOVED))
1364                 nfs41_handle_state_revoked(clp);
1365         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1366                 nfs41_handle_recallable_state_revoked(clp);
1367         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1368                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1369                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1370                 nfs41_handle_cb_path_down(clp);
1371 }
1372
1373 static int nfs4_reset_session(struct nfs_client *clp)
1374 {
1375         int status;
1376
1377         nfs4_begin_drain_session(clp);
1378         status = nfs4_proc_destroy_session(clp->cl_session);
1379         if (status && status != -NFS4ERR_BADSESSION &&
1380             status != -NFS4ERR_DEADSESSION) {
1381                 status = nfs4_recovery_handle_error(clp, status);
1382                 goto out;
1383         }
1384
1385         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1386         status = nfs4_proc_create_session(clp);
1387         if (status) {
1388                 status = nfs4_recovery_handle_error(clp, status);
1389                 goto out;
1390         }
1391         /* create_session negotiated new slot table */
1392         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1393
1394          /* Let the state manager reestablish state */
1395         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1396                 nfs41_setup_state_renewal(clp);
1397 out:
1398         return status;
1399 }
1400
1401 static int nfs4_recall_slot(struct nfs_client *clp)
1402 {
1403         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1404         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1405         struct nfs4_slot *new, *old;
1406         int i;
1407
1408         nfs4_begin_drain_session(clp);
1409         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1410                       GFP_NOFS);
1411         if (!new)
1412                 return -ENOMEM;
1413
1414         spin_lock(&fc_tbl->slot_tbl_lock);
1415         for (i = 0; i < fc_tbl->target_max_slots; i++)
1416                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1417         old = fc_tbl->slots;
1418         fc_tbl->slots = new;
1419         fc_tbl->max_slots = fc_tbl->target_max_slots;
1420         fc_tbl->target_max_slots = 0;
1421         fc_attrs->max_reqs = fc_tbl->max_slots;
1422         spin_unlock(&fc_tbl->slot_tbl_lock);
1423
1424         kfree(old);
1425         nfs4_end_drain_session(clp);
1426         return 0;
1427 }
1428
1429 #else /* CONFIG_NFS_V4_1 */
1430 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1431 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1432 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1433 #endif /* CONFIG_NFS_V4_1 */
1434
1435 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1436  * on EXCHANGE_ID for v4.1
1437  */
1438 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1439 {
1440         if (nfs4_has_session(clp)) {
1441                 switch (status) {
1442                 case -NFS4ERR_DELAY:
1443                 case -NFS4ERR_CLID_INUSE:
1444                 case -EAGAIN:
1445                         break;
1446
1447                 case -EKEYEXPIRED:
1448                         nfs4_warn_keyexpired(clp->cl_hostname);
1449                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1450                                          * in nfs4_exchange_id */
1451                 default:
1452                         return;
1453                 }
1454         }
1455         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1456 }
1457
1458 static void nfs4_state_manager(struct nfs_client *clp)
1459 {
1460         int status = 0;
1461
1462         /* Ensure exclusive access to NFSv4 state */
1463         for(;;) {
1464                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1465                         /* We're going to have to re-establish a clientid */
1466                         status = nfs4_reclaim_lease(clp);
1467                         if (status) {
1468                                 nfs4_set_lease_expired(clp, status);
1469                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1470                                                         &clp->cl_state))
1471                                         continue;
1472                                 if (clp->cl_cons_state ==
1473                                                         NFS_CS_SESSION_INITING)
1474                                         nfs_mark_client_ready(clp, status);
1475                                 goto out_error;
1476                         }
1477                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1478                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1479                         pnfs_destroy_all_layouts(clp);
1480                 }
1481
1482                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1483                         status = nfs4_check_lease(clp);
1484                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1485                                 continue;
1486                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1487                                 goto out_error;
1488                 }
1489
1490                 /* Initialize or reset the session */
1491                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1492                    && nfs4_has_session(clp)) {
1493                         status = nfs4_reset_session(clp);
1494                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1495                                 continue;
1496                         if (status < 0)
1497                                 goto out_error;
1498                 }
1499
1500                 /* First recover reboot state... */
1501                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1502                         status = nfs4_do_reclaim(clp,
1503                                 clp->cl_mvops->reboot_recovery_ops);
1504                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1505                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1506                                 continue;
1507                         nfs4_state_end_reclaim_reboot(clp);
1508                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1509                                 continue;
1510                         if (status < 0)
1511                                 goto out_error;
1512                 }
1513
1514                 /* Now recover expired state... */
1515                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1516                         status = nfs4_do_reclaim(clp,
1517                                 clp->cl_mvops->nograce_recovery_ops);
1518                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1519                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1520                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1521                                 continue;
1522                         if (status < 0)
1523                                 goto out_error;
1524                 }
1525
1526                 nfs4_end_drain_session(clp);
1527                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1528                         nfs_client_return_marked_delegations(clp);
1529                         continue;
1530                 }
1531                 /* Recall session slots */
1532                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1533                    && nfs4_has_session(clp)) {
1534                         status = nfs4_recall_slot(clp);
1535                         if (status < 0)
1536                                 goto out_error;
1537                         continue;
1538                 }
1539
1540
1541                 nfs4_clear_state_manager_bit(clp);
1542                 /* Did we race with an attempt to give us more work? */
1543                 if (clp->cl_state == 0)
1544                         break;
1545                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1546                         break;
1547         }
1548         return;
1549 out_error:
1550         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1551                         " with error %d\n", clp->cl_hostname, -status);
1552         nfs4_end_drain_session(clp);
1553         nfs4_clear_state_manager_bit(clp);
1554 }
1555
1556 static int nfs4_run_state_manager(void *ptr)
1557 {
1558         struct nfs_client *clp = ptr;
1559
1560         allow_signal(SIGKILL);
1561         nfs4_state_manager(clp);
1562         nfs_put_client(clp);
1563         module_put_and_exit(0);
1564         return 0;
1565 }
1566
1567 /*
1568  * Local variables:
1569  *  c-basic-offset: 8
1570  * End:
1571  */