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