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 static struct rpc_cred *
 109 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 110 {
 111         struct rpc_cred *cred = NULL;
 112         struct nfs4_state_owner *sp;
 113         struct rb_node *pos;
 114
 115         for (pos = rb_first(&server->state_owners);
 116              pos != NULL;
 117              pos = rb_next(pos)) {
 118                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 119                 if (list_empty(&sp->so_states))
 120                         continue;
 121                 cred = get_rpccred(sp->so_cred);
 122                 break;
 123         }
 124         return cred;
 125 }
 126
 127 /**
 128  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 129  * @clp: client state handle
 130  *
 131  * Returns an rpc_cred with reference count bumped, or NULL.
 132  * Caller must hold clp->cl_lock.
 133  */
 134 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 135 {
 136         struct rpc_cred *cred = NULL;
 137         struct nfs_server *server;
 138
 139         rcu_read_lock();
 140         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 141                 cred = nfs4_get_renew_cred_server_locked(server);
 142                 if (cred != NULL)
 143                         break;
 144         }
 145         rcu_read_unlock();
 146         return cred;
 147 }
 148
 149 #if defined(CONFIG_NFS_V4_1)
 150
 151 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 152 {
 153         int status;
 154         struct nfs_fsinfo fsinfo;
 155
 156         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 157                 nfs4_schedule_state_renewal(clp);
 158                 return 0;
 159         }
 160
 161         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 162         if (status == 0) {
 163                 /* Update lease time and schedule renewal */
 164                 spin_lock(&clp->cl_lock);
 165                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 166                 clp->cl_last_renewal = jiffies;
 167                 spin_unlock(&clp->cl_lock);
 168
 169                 nfs4_schedule_state_renewal(clp);
 170         }
 171
 172         return status;
 173 }
 174
 175 /*
 176  * Back channel returns NFS4ERR_DELAY for new requests when
 177  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 178  * is ended.
 179  */
 180 static void nfs4_end_drain_session(struct nfs_client *clp)
 181 {
 182         struct nfs4_session *ses = clp->cl_session;
 183         int max_slots;
 184
 185         if (ses == NULL)
 186                 return;
 187         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 188                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 189                 max_slots = ses->fc_slot_table.max_slots;
 190                 while (max_slots--) {
 191                         struct rpc_task *task;
 192
 193                         task = rpc_wake_up_next(&ses->fc_slot_table.
 194                                                 slot_tbl_waitq);
 195                         if (!task)
 196                                 break;
 197                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 198                 }
 199                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 200         }
 201 }
 202
 203 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 204 {
 205         spin_lock(&tbl->slot_tbl_lock);
 206         if (tbl->highest_used_slotid != -1) {
 207                 INIT_COMPLETION(tbl->complete);
 208                 spin_unlock(&tbl->slot_tbl_lock);
 209                 return wait_for_completion_interruptible(&tbl->complete);
 210         }
 211         spin_unlock(&tbl->slot_tbl_lock);
 212         return 0;
 213 }
 214
 215 static int nfs4_begin_drain_session(struct nfs_client *clp)
 216 {
 217         struct nfs4_session *ses = clp->cl_session;
 218         int ret = 0;
 219
 220         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 221         /* back channel */
 222         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 223         if (ret)
 224                 return ret;
 225         /* fore channel */
 226         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 227 }
 228
 229 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 230 {
 231         int status;
 232
 233         nfs4_begin_drain_session(clp);
 234         status = nfs4_proc_exchange_id(clp, cred);
 235         if (status != 0)
 236                 goto out;
 237         status = nfs4_proc_create_session(clp);
 238         if (status != 0)
 239                 goto out;
 240         nfs41_setup_state_renewal(clp);
 241         nfs_mark_client_ready(clp, NFS_CS_READY);
 242 out:
 243         return status;
 244 }
 245
 246 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 247 {
 248         struct rpc_cred *cred;
 249
 250         spin_lock(&clp->cl_lock);
 251         cred = nfs4_get_machine_cred_locked(clp);
 252         spin_unlock(&clp->cl_lock);
 253         return cred;
 254 }
 255
 256 #endif /* CONFIG_NFS_V4_1 */
 257
 258 static struct rpc_cred *
 259 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 260 {
 261         struct nfs_client *clp = server->nfs_client;
 262         struct rpc_cred *cred = NULL;
 263         struct nfs4_state_owner *sp;
 264         struct rb_node *pos;
 265
 266         spin_lock(&clp->cl_lock);
 267         pos = rb_first(&server->state_owners);
 268         if (pos != NULL) {
 269                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 270                 cred = get_rpccred(sp->so_cred);
 271         }
 272         spin_unlock(&clp->cl_lock);
 273         return cred;
 274 }
 275
 276 /**
 277  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 278  * @clp: client state handle
 279  *
 280  * Returns an rpc_cred with reference count bumped, or NULL.
 281  */
 282 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 283 {
 284         struct nfs_server *server;
 285         struct rpc_cred *cred;
 286
 287         spin_lock(&clp->cl_lock);
 288         cred = nfs4_get_machine_cred_locked(clp);
 289         spin_unlock(&clp->cl_lock);
 290         if (cred != NULL)
 291                 goto out;
 292
 293         rcu_read_lock();
 294         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 295                 cred = nfs4_get_setclientid_cred_server(server);
 296                 if (cred != NULL)
 297                         break;
 298         }
 299         rcu_read_unlock();
 300
 301 out:
 302         return cred;
 303 }
 304
 305 static void nfs_alloc_unique_id_locked(struct rb_root *root,
 306                                        struct nfs_unique_id *new,
 307                                        __u64 minval, int maxbits)
 308 {
 309         struct rb_node **p, *parent;
 310         struct nfs_unique_id *pos;
 311         __u64 mask = ~0ULL;
 312
 313         if (maxbits < 64)
 314                 mask = (1ULL << maxbits) - 1ULL;
 315
 316         /* Ensure distribution is more or less flat */
 317         get_random_bytes(&new->id, sizeof(new->id));
 318         new->id &= mask;
 319         if (new->id < minval)
 320                 new->id += minval;
 321 retry:
 322         p = &root->rb_node;
 323         parent = NULL;
 324
 325         while (*p != NULL) {
 326                 parent = *p;
 327                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 328
 329                 if (new->id < pos->id)
 330                         p = &(*p)->rb_left;
 331                 else if (new->id > pos->id)
 332                         p = &(*p)->rb_right;
 333                 else
 334                         goto id_exists;
 335         }
 336         rb_link_node(&new->rb_node, parent, p);
 337         rb_insert_color(&new->rb_node, root);
 338         return;
 339 id_exists:
 340         for (;;) {
 341                 new->id++;
 342                 if (new->id < minval || (new->id & mask) != new->id) {
 343                         new->id = minval;
 344                         break;
 345                 }
 346                 parent = rb_next(parent);
 347                 if (parent == NULL)
 348                         break;
 349                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 350                 if (new->id < pos->id)
 351                         break;
 352         }
 353         goto retry;
 354 }
 355
 356 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 357 {
 358         rb_erase(&id->rb_node, root);
 359 }
 360
 361 static struct nfs4_state_owner *
 362 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 363 {
 364         struct rb_node **p = &server->state_owners.rb_node,
 365                        *parent = NULL;
 366         struct nfs4_state_owner *sp, *res = NULL;
 367
 368         while (*p != NULL) {
 369                 parent = *p;
 370                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 371
 372                 if (server < sp->so_server) {
 373                         p = &parent->rb_left;
 374                         continue;
 375                 }
 376                 if (server > sp->so_server) {
 377                         p = &parent->rb_right;
 378                         continue;
 379                 }
 380                 if (cred < sp->so_cred)
 381                         p = &parent->rb_left;
 382                 else if (cred > sp->so_cred)
 383                         p = &parent->rb_right;
 384                 else {
 385                         atomic_inc(&sp->so_count);
 386                         res = sp;
 387                         break;
 388                 }
 389         }
 390         return res;
 391 }
 392
 393 static struct nfs4_state_owner *
 394 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 395 {
 396         struct nfs_server *server = new->so_server;
 397         struct rb_node **p = &server->state_owners.rb_node,
 398                        *parent = NULL;
 399         struct nfs4_state_owner *sp;
 400
 401         while (*p != NULL) {
 402                 parent = *p;
 403                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 404
 405                 if (new->so_cred < sp->so_cred)
 406                         p = &parent->rb_left;
 407                 else if (new->so_cred > sp->so_cred)
 408                         p = &parent->rb_right;
 409                 else {
 410                         atomic_inc(&sp->so_count);
 411                         return sp;
 412                 }
 413         }
 414         nfs_alloc_unique_id_locked(&server->openowner_id,
 415                                         &new->so_owner_id, 1, 64);
 416         rb_link_node(&new->so_server_node, parent, p);
 417         rb_insert_color(&new->so_server_node, &server->state_owners);
 418         return new;
 419 }
 420
 421 static void
 422 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 423 {
 424         struct nfs_server *server = sp->so_server;
 425
 426         if (!RB_EMPTY_NODE(&sp->so_server_node))
 427                 rb_erase(&sp->so_server_node, &server->state_owners);
 428         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
 429 }
 430
 431 /*
 432  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 433  * create a new state_owner.
 434  *
 435  */
 436 static struct nfs4_state_owner *
 437 nfs4_alloc_state_owner(void)
 438 {
 439         struct nfs4_state_owner *sp;
 440
 441         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 442         if (!sp)
 443                 return NULL;
 444         spin_lock_init(&sp->so_lock);
 445         INIT_LIST_HEAD(&sp->so_states);
 446         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 447         sp->so_seqid.sequence = &sp->so_sequence;
 448         spin_lock_init(&sp->so_sequence.lock);
 449         INIT_LIST_HEAD(&sp->so_sequence.list);
 450         atomic_set(&sp->so_count, 1);
 451         return sp;
 452 }
 453
 454 static void
 455 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 456 {
 457         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 458                 struct nfs_server *server = sp->so_server;
 459                 struct nfs_client *clp = server->nfs_client;
 460
 461                 spin_lock(&clp->cl_lock);
 462                 rb_erase(&sp->so_server_node, &server->state_owners);
 463                 RB_CLEAR_NODE(&sp->so_server_node);
 464                 spin_unlock(&clp->cl_lock);
 465         }
 466 }
 467
 468 /**
 469  * nfs4_get_state_owner - Look up a state owner given a credential
 470  * @server: nfs_server to search
 471  * @cred: RPC credential to match
 472  *
 473  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 474  */
 475 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 476                                               struct rpc_cred *cred)
 477 {
 478         struct nfs_client *clp = server->nfs_client;
 479         struct nfs4_state_owner *sp, *new;
 480
 481         spin_lock(&clp->cl_lock);
 482         sp = nfs4_find_state_owner_locked(server, cred);
 483         spin_unlock(&clp->cl_lock);
 484         if (sp != NULL)
 485                 return sp;
 486         new = nfs4_alloc_state_owner();
 487         if (new == NULL)
 488                 return NULL;
 489         new->so_server = server;
 490         new->so_cred = cred;
 491         spin_lock(&clp->cl_lock);
 492         sp = nfs4_insert_state_owner_locked(new);
 493         spin_unlock(&clp->cl_lock);
 494         if (sp == new)
 495                 get_rpccred(cred);
 496         else {
 497                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 498                 kfree(new);
 499         }
 500         return sp;
 501 }
 502
 503 /**
 504  * nfs4_put_state_owner - Release a nfs4_state_owner
 505  * @sp: state owner data to release
 506  *
 507  */
 508 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 509 {
 510         struct nfs_client *clp = sp->so_server->nfs_client;
 511         struct rpc_cred *cred = sp->so_cred;
 512
 513         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 514                 return;
 515         nfs4_remove_state_owner_locked(sp);
 516         spin_unlock(&clp->cl_lock);
 517         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 518         put_rpccred(cred);
 519         kfree(sp);
 520 }
 521
 522 static struct nfs4_state *
 523 nfs4_alloc_open_state(void)
 524 {
 525         struct nfs4_state *state;
 526
 527         state = kzalloc(sizeof(*state), GFP_NOFS);
 528         if (!state)
 529                 return NULL;
 530         atomic_set(&state->count, 1);
 531         INIT_LIST_HEAD(&state->lock_states);
 532         spin_lock_init(&state->state_lock);
 533         seqlock_init(&state->seqlock);
 534         return state;
 535 }
 536
 537 void
 538 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 539 {
 540         if (state->state == fmode)
 541                 return;
 542         /* NB! List reordering - see the reclaim code for why.  */
 543         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 544                 if (fmode & FMODE_WRITE)
 545                         list_move(&state->open_states, &state->owner->so_states);
 546                 else
 547                         list_move_tail(&state->open_states, &state->owner->so_states);
 548         }
 549         state->state = fmode;
 550 }
 551
 552 static struct nfs4_state *
 553 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 554 {
 555         struct nfs_inode *nfsi = NFS_I(inode);
 556         struct nfs4_state *state;
 557
 558         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 559                 if (state->owner != owner)
 560                         continue;
 561                 if (atomic_inc_not_zero(&state->count))
 562                         return state;
 563         }
 564         return NULL;
 565 }
 566
 567 static void
 568 nfs4_free_open_state(struct nfs4_state *state)
 569 {
 570         kfree(state);
 571 }
 572
 573 struct nfs4_state *
 574 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 575 {
 576         struct nfs4_state *state, *new;
 577         struct nfs_inode *nfsi = NFS_I(inode);
 578
 579         spin_lock(&inode->i_lock);
 580         state = __nfs4_find_state_byowner(inode, owner);
 581         spin_unlock(&inode->i_lock);
 582         if (state)
 583                 goto out;
 584         new = nfs4_alloc_open_state();
 585         spin_lock(&owner->so_lock);
 586         spin_lock(&inode->i_lock);
 587         state = __nfs4_find_state_byowner(inode, owner);
 588         if (state == NULL && new != NULL) {
 589                 state = new;
 590                 state->owner = owner;
 591                 atomic_inc(&owner->so_count);
 592                 list_add(&state->inode_states, &nfsi->open_states);
 593                 ihold(inode);
 594                 state->inode = inode;
 595                 spin_unlock(&inode->i_lock);
 596                 /* Note: The reclaim code dictates that we add stateless
 597                  * and read-only stateids to the end of the list */
 598                 list_add_tail(&state->open_states, &owner->so_states);
 599                 spin_unlock(&owner->so_lock);
 600         } else {
 601                 spin_unlock(&inode->i_lock);
 602                 spin_unlock(&owner->so_lock);
 603                 if (new)
 604                         nfs4_free_open_state(new);
 605         }
 606 out:
 607         return state;
 608 }
 609
 610 void nfs4_put_open_state(struct nfs4_state *state)
 611 {
 612         struct inode *inode = state->inode;
 613         struct nfs4_state_owner *owner = state->owner;
 614
 615         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 616                 return;
 617         spin_lock(&inode->i_lock);
 618         list_del(&state->inode_states);
 619         list_del(&state->open_states);
 620         spin_unlock(&inode->i_lock);
 621         spin_unlock(&owner->so_lock);
 622         iput(inode);
 623         nfs4_free_open_state(state);
 624         nfs4_put_state_owner(owner);
 625 }
 626
 627 /*
 628  * Close the current file.
 629  */
 630 static void __nfs4_close(struct path *path, struct nfs4_state *state,
 631                 fmode_t fmode, gfp_t gfp_mask, int wait)
 632 {
 633         struct nfs4_state_owner *owner = state->owner;
 634         int call_close = 0;
 635         fmode_t newstate;
 636
 637         atomic_inc(&owner->so_count);
 638         /* Protect against nfs4_find_state() */
 639         spin_lock(&owner->so_lock);
 640         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 641                 case FMODE_READ:
 642                         state->n_rdonly--;
 643                         break;
 644                 case FMODE_WRITE:
 645                         state->n_wronly--;
 646                         break;
 647                 case FMODE_READ|FMODE_WRITE:
 648                         state->n_rdwr--;
 649         }
 650         newstate = FMODE_READ|FMODE_WRITE;
 651         if (state->n_rdwr == 0) {
 652                 if (state->n_rdonly == 0) {
 653                         newstate &= ~FMODE_READ;
 654                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 655                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 656                 }
 657                 if (state->n_wronly == 0) {
 658                         newstate &= ~FMODE_WRITE;
 659                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 660                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 661                 }
 662                 if (newstate == 0)
 663                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 664         }
 665         nfs4_state_set_mode_locked(state, newstate);
 666         spin_unlock(&owner->so_lock);
 667
 668         if (!call_close) {
 669                 nfs4_put_open_state(state);
 670                 nfs4_put_state_owner(owner);
 671         } else {
 672                 bool roc = pnfs_roc(state->inode);
 673
 674                 nfs4_do_close(path, state, gfp_mask, wait, roc);
 675         }
 676 }
 677
 678 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 679 {
 680         __nfs4_close(path, state, fmode, GFP_NOFS, 0);
 681 }
 682
 683 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 684 {
 685         __nfs4_close(path, state, fmode, GFP_KERNEL, 1);
 686 }
 687
 688 /*
 689  * Search the state->lock_states for an existing lock_owner
 690  * that is compatible with current->files
 691  */
 692 static struct nfs4_lock_state *
 693 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 694 {
 695         struct nfs4_lock_state *pos;
 696         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 697                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 698                         continue;
 699                 switch (pos->ls_owner.lo_type) {
 700                 case NFS4_POSIX_LOCK_TYPE:
 701                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 702                                 continue;
 703                         break;
 704                 case NFS4_FLOCK_LOCK_TYPE:
 705                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 706                                 continue;
 707                 }
 708                 atomic_inc(&pos->ls_count);
 709                 return pos;
 710         }
 711         return NULL;
 712 }
 713
 714 /*
 715  * Return a compatible lock_state. If no initialized lock_state structure
 716  * exists, return an uninitialized one.
 717  *
 718  */
 719 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)
 720 {
 721         struct nfs4_lock_state *lsp;
 722         struct nfs_server *server = state->owner->so_server;
 723         struct nfs_client *clp = server->nfs_client;
 724
 725         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 726         if (lsp == NULL)
 727                 return NULL;
 728         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 729         spin_lock_init(&lsp->ls_sequence.lock);
 730         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 731         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 732         atomic_set(&lsp->ls_count, 1);
 733         lsp->ls_state = state;
 734         lsp->ls_owner.lo_type = type;
 735         switch (lsp->ls_owner.lo_type) {
 736         case NFS4_FLOCK_LOCK_TYPE:
 737                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 738                 break;
 739         case NFS4_POSIX_LOCK_TYPE:
 740                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 741                 break;
 742         default:
 743                 kfree(lsp);
 744                 return NULL;
 745         }
 746         spin_lock(&clp->cl_lock);
 747         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
 748         spin_unlock(&clp->cl_lock);
 749         INIT_LIST_HEAD(&lsp->ls_locks);
 750         return lsp;
 751 }
 752
 753 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 754 {
 755         struct nfs_server *server = lsp->ls_state->owner->so_server;
 756         struct nfs_client *clp = server->nfs_client;
 757
 758         spin_lock(&clp->cl_lock);
 759         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
 760         spin_unlock(&clp->cl_lock);
 761         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 762         kfree(lsp);
 763 }
 764
 765 /*
 766  * Return a compatible lock_state. If no initialized lock_state structure
 767  * exists, return an uninitialized one.
 768  *
 769  */
 770 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 771 {
 772         struct nfs4_lock_state *lsp, *new = NULL;
 773
 774         for(;;) {
 775                 spin_lock(&state->state_lock);
 776                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 777                 if (lsp != NULL)
 778                         break;
 779                 if (new != NULL) {
 780                         list_add(&new->ls_locks, &state->lock_states);
 781                         set_bit(LK_STATE_IN_USE, &state->flags);
 782                         lsp = new;
 783                         new = NULL;
 784                         break;
 785                 }
 786                 spin_unlock(&state->state_lock);
 787                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 788                 if (new == NULL)
 789                         return NULL;
 790         }
 791         spin_unlock(&state->state_lock);
 792         if (new != NULL)
 793                 nfs4_free_lock_state(new);
 794         return lsp;
 795 }
 796
 797 /*
 798  * Release reference to lock_state, and free it if we see that
 799  * it is no longer in use
 800  */
 801 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 802 {
 803         struct nfs4_state *state;
 804
 805         if (lsp == NULL)
 806                 return;
 807         state = lsp->ls_state;
 808         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 809                 return;
 810         list_del(&lsp->ls_locks);
 811         if (list_empty(&state->lock_states))
 812                 clear_bit(LK_STATE_IN_USE, &state->flags);
 813         spin_unlock(&state->state_lock);
 814         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 815                 nfs4_release_lockowner(lsp);
 816         nfs4_free_lock_state(lsp);
 817 }
 818
 819 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 820 {
 821         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 822
 823         dst->fl_u.nfs4_fl.owner = lsp;
 824         atomic_inc(&lsp->ls_count);
 825 }
 826
 827 static void nfs4_fl_release_lock(struct file_lock *fl)
 828 {
 829         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 830 }
 831
 832 static const struct file_lock_operations nfs4_fl_lock_ops = {
 833         .fl_copy_lock = nfs4_fl_copy_lock,
 834         .fl_release_private = nfs4_fl_release_lock,
 835 };
 836
 837 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 838 {
 839         struct nfs4_lock_state *lsp;
 840
 841         if (fl->fl_ops != NULL)
 842                 return 0;
 843         if (fl->fl_flags & FL_POSIX)
 844                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 845         else if (fl->fl_flags & FL_FLOCK)
 846                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 847         else
 848                 return -EINVAL;
 849         if (lsp == NULL)
 850                 return -ENOMEM;
 851         fl->fl_u.nfs4_fl.owner = lsp;
 852         fl->fl_ops = &nfs4_fl_lock_ops;
 853         return 0;
 854 }
 855
 856 /*
 857  * Byte-range lock aware utility to initialize the stateid of read/write
 858  * requests.
 859  */
 860 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 861 {
 862         struct nfs4_lock_state *lsp;
 863         int seq;
 864
 865         do {
 866                 seq = read_seqbegin(&state->seqlock);
 867                 memcpy(dst, &state->stateid, sizeof(*dst));
 868         } while (read_seqretry(&state->seqlock, seq));
 869         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 870                 return;
 871
 872         spin_lock(&state->state_lock);
 873         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 874         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 875                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 876         spin_unlock(&state->state_lock);
 877         nfs4_put_lock_state(lsp);
 878 }
 879
 880 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 881 {
 882         struct nfs_seqid *new;
 883
 884         new = kmalloc(sizeof(*new), gfp_mask);
 885         if (new != NULL) {
 886                 new->sequence = counter;
 887                 INIT_LIST_HEAD(&new->list);
 888         }
 889         return new;
 890 }
 891
 892 void nfs_release_seqid(struct nfs_seqid *seqid)
 893 {
 894         if (!list_empty(&seqid->list)) {
 895                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 896
 897                 spin_lock(&sequence->lock);
 898                 list_del_init(&seqid->list);
 899                 spin_unlock(&sequence->lock);
 900                 rpc_wake_up(&sequence->wait);
 901         }
 902 }
 903
 904 void nfs_free_seqid(struct nfs_seqid *seqid)
 905 {
 906         nfs_release_seqid(seqid);
 907         kfree(seqid);
 908 }
 909
 910 /*
 911  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 912  * failed with a seqid incrementing error -
 913  * see comments nfs_fs.h:seqid_mutating_error()
 914  */
 915 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 916 {
 917         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 918         switch (status) {
 919                 case 0:
 920                         break;
 921                 case -NFS4ERR_BAD_SEQID:
 922                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 923                                 return;
 924                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 925                                         " sequence-id error on an"
 926                                         " unconfirmed sequence %p!\n",
 927                                         seqid->sequence);
 928                 case -NFS4ERR_STALE_CLIENTID:
 929                 case -NFS4ERR_STALE_STATEID:
 930                 case -NFS4ERR_BAD_STATEID:
 931                 case -NFS4ERR_BADXDR:
 932                 case -NFS4ERR_RESOURCE:
 933                 case -NFS4ERR_NOFILEHANDLE:
 934                         /* Non-seqid mutating errors */
 935                         return;
 936         };
 937         /*
 938          * Note: no locking needed as we are guaranteed to be first
 939          * on the sequence list
 940          */
 941         seqid->sequence->counter++;
 942 }
 943
 944 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 945 {
 946         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 947                                         struct nfs4_state_owner, so_seqid);
 948         struct nfs_server *server = sp->so_server;
 949
 950         if (status == -NFS4ERR_BAD_SEQID)
 951                 nfs4_drop_state_owner(sp);
 952         if (!nfs4_has_session(server->nfs_client))
 953                 nfs_increment_seqid(status, seqid);
 954 }
 955
 956 /*
 957  * Increment the seqid if the LOCK/LOCKU succeeded, or
 958  * failed with a seqid incrementing error -
 959  * see comments nfs_fs.h:seqid_mutating_error()
 960  */
 961 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 962 {
 963         nfs_increment_seqid(status, seqid);
 964 }
 965
 966 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 967 {
 968         struct rpc_sequence *sequence = seqid->sequence->sequence;
 969         int status = 0;
 970
 971         spin_lock(&sequence->lock);
 972         if (list_empty(&seqid->list))
 973                 list_add_tail(&seqid->list, &sequence->list);
 974         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 975                 goto unlock;
 976         rpc_sleep_on(&sequence->wait, task, NULL);
 977         status = -EAGAIN;
 978 unlock:
 979         spin_unlock(&sequence->lock);
 980         return status;
 981 }
 982
 983 static int nfs4_run_state_manager(void *);
 984
 985 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 986 {
 987         smp_mb__before_clear_bit();
 988         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 989         smp_mb__after_clear_bit();
 990         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 991         rpc_wake_up(&clp->cl_rpcwaitq);
 992 }
 993
 994 /*
 995  * Schedule the nfs_client asynchronous state management routine
 996  */
 997 void nfs4_schedule_state_manager(struct nfs_client *clp)
 998 {
 999         struct task_struct *task;
1000
1001         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1002                 return;
1003         __module_get(THIS_MODULE);
1004         atomic_inc(&clp->cl_count);
1005         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1006                                 rpc_peeraddr2str(clp->cl_rpcclient,
1007                                                         RPC_DISPLAY_ADDR));
1008         if (!IS_ERR(task))
1009                 return;
1010         nfs4_clear_state_manager_bit(clp);
1011         nfs_put_client(clp);
1012         module_put(THIS_MODULE);
1013 }
1014
1015 /*
1016  * Schedule a lease recovery attempt
1017  */
1018 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1019 {
1020         if (!clp)
1021                 return;
1022         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1023                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1024         nfs4_schedule_state_manager(clp);
1025 }
1026
1027 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1028 {
1029
1030         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1031         /* Don't recover state that expired before the reboot */
1032         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1033                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1034                 return 0;
1035         }
1036         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1037         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1038         return 1;
1039 }
1040
1041 static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1042 {
1043         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1044         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1045         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1046         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1047         return 1;
1048 }
1049
1050 void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1051 {
1052         struct nfs_client *clp = server->nfs_client;
1053
1054         nfs4_state_mark_reclaim_nograce(clp, state);
1055         nfs4_schedule_state_manager(clp);
1056 }
1057
1058 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1059 {
1060         struct inode *inode = state->inode;
1061         struct nfs_inode *nfsi = NFS_I(inode);
1062         struct file_lock *fl;
1063         int status = 0;
1064
1065         if (inode->i_flock == NULL)
1066                 return 0;
1067
1068         /* Guard against delegation returns and new lock/unlock calls */
1069         down_write(&nfsi->rwsem);
1070         /* Protect inode->i_flock using the BKL */
1071         lock_flocks();
1072         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1073                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1074                         continue;
1075                 if (nfs_file_open_context(fl->fl_file)->state != state)
1076                         continue;
1077                 unlock_flocks();
1078                 status = ops->recover_lock(state, fl);
1079                 switch (status) {
1080                         case 0:
1081                                 break;
1082                         case -ESTALE:
1083                         case -NFS4ERR_ADMIN_REVOKED:
1084                         case -NFS4ERR_STALE_STATEID:
1085                         case -NFS4ERR_BAD_STATEID:
1086                         case -NFS4ERR_EXPIRED:
1087                         case -NFS4ERR_NO_GRACE:
1088                         case -NFS4ERR_STALE_CLIENTID:
1089                         case -NFS4ERR_BADSESSION:
1090                         case -NFS4ERR_BADSLOT:
1091                         case -NFS4ERR_BAD_HIGH_SLOT:
1092                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1093                                 goto out;
1094                         default:
1095                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1096                                                 __func__, status);
1097                         case -ENOMEM:
1098                         case -NFS4ERR_DENIED:
1099                         case -NFS4ERR_RECLAIM_BAD:
1100                         case -NFS4ERR_RECLAIM_CONFLICT:
1101                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1102                                 status = 0;
1103                 }
1104                 lock_flocks();
1105         }
1106         unlock_flocks();
1107 out:
1108         up_write(&nfsi->rwsem);
1109         return status;
1110 }
1111
1112 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1113 {
1114         struct nfs4_state *state;
1115         struct nfs4_lock_state *lock;
1116         int status = 0;
1117
1118         /* Note: we rely on the sp->so_states list being ordered
1119          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1120          * states first.
1121          * This is needed to ensure that the server won't give us any
1122          * read delegations that we have to return if, say, we are
1123          * recovering after a network partition or a reboot from a
1124          * server that doesn't support a grace period.
1125          */
1126 restart:
1127         spin_lock(&sp->so_lock);
1128         list_for_each_entry(state, &sp->so_states, open_states) {
1129                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1130                         continue;
1131                 if (state->state == 0)
1132                         continue;
1133                 atomic_inc(&state->count);
1134                 spin_unlock(&sp->so_lock);
1135                 status = ops->recover_open(sp, state);
1136                 if (status >= 0) {
1137                         status = nfs4_reclaim_locks(state, ops);
1138                         if (status >= 0) {
1139                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1140                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1141                                                 printk("%s: Lock reclaim failed!\n",
1142                                                         __func__);
1143                                 }
1144                                 nfs4_put_open_state(state);
1145                                 goto restart;
1146                         }
1147                 }
1148                 switch (status) {
1149                         default:
1150                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1151                                                 __func__, status);
1152                         case -ENOENT:
1153                         case -ENOMEM:
1154                         case -ESTALE:
1155                                 /*
1156                                  * Open state on this file cannot be recovered
1157                                  * All we can do is revert to using the zero stateid.
1158                                  */
1159                                 memset(state->stateid.data, 0,
1160                                         sizeof(state->stateid.data));
1161                                 /* Mark the file as being 'closed' */
1162                                 state->state = 0;
1163                                 break;
1164                         case -EKEYEXPIRED:
1165                                 /*
1166                                  * User RPCSEC_GSS context has expired.
1167                                  * We cannot recover this stateid now, so
1168                                  * skip it and allow recovery thread to
1169                                  * proceed.
1170                                  */
1171                                 break;
1172                         case -NFS4ERR_ADMIN_REVOKED:
1173                         case -NFS4ERR_STALE_STATEID:
1174                         case -NFS4ERR_BAD_STATEID:
1175                         case -NFS4ERR_RECLAIM_BAD:
1176                         case -NFS4ERR_RECLAIM_CONFLICT:
1177                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1178                                 break;
1179                         case -NFS4ERR_EXPIRED:
1180                         case -NFS4ERR_NO_GRACE:
1181                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1182                         case -NFS4ERR_STALE_CLIENTID:
1183                         case -NFS4ERR_BADSESSION:
1184                         case -NFS4ERR_BADSLOT:
1185                         case -NFS4ERR_BAD_HIGH_SLOT:
1186                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1187                                 goto out_err;
1188                 }
1189                 nfs4_put_open_state(state);
1190                 goto restart;
1191         }
1192         spin_unlock(&sp->so_lock);
1193         return 0;
1194 out_err:
1195         nfs4_put_open_state(state);
1196         return status;
1197 }
1198
1199 static void nfs4_clear_open_state(struct nfs4_state *state)
1200 {
1201         struct nfs4_lock_state *lock;
1202
1203         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1204         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1205         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1206         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1207         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1208                 lock->ls_seqid.flags = 0;
1209                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1210         }
1211 }
1212
1213 static void nfs4_reset_seqids(struct nfs_server *server,
1214         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1215 {
1216         struct nfs_client *clp = server->nfs_client;
1217         struct nfs4_state_owner *sp;
1218         struct rb_node *pos;
1219         struct nfs4_state *state;
1220
1221         spin_lock(&clp->cl_lock);
1222         for (pos = rb_first(&server->state_owners);
1223              pos != NULL;
1224              pos = rb_next(pos)) {
1225                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1226                 sp->so_seqid.flags = 0;
1227                 spin_lock(&sp->so_lock);
1228                 list_for_each_entry(state, &sp->so_states, open_states) {
1229                         if (mark_reclaim(clp, state))
1230                                 nfs4_clear_open_state(state);
1231                 }
1232                 spin_unlock(&sp->so_lock);
1233         }
1234         spin_unlock(&clp->cl_lock);
1235 }
1236
1237 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1238         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1239 {
1240         struct nfs_server *server;
1241
1242         rcu_read_lock();
1243         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1244                 nfs4_reset_seqids(server, mark_reclaim);
1245         rcu_read_unlock();
1246 }
1247
1248 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1249 {
1250         /* Mark all delegations for reclaim */
1251         nfs_delegation_mark_reclaim(clp);
1252         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1253 }
1254
1255 static void nfs4_reclaim_complete(struct nfs_client *clp,
1256                                  const struct nfs4_state_recovery_ops *ops)
1257 {
1258         /* Notify the server we're done reclaiming our state */
1259         if (ops->reclaim_complete)
1260                 (void)ops->reclaim_complete(clp);
1261 }
1262
1263 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1264 {
1265         struct nfs_client *clp = server->nfs_client;
1266         struct nfs4_state_owner *sp;
1267         struct rb_node *pos;
1268         struct nfs4_state *state;
1269
1270         spin_lock(&clp->cl_lock);
1271         for (pos = rb_first(&server->state_owners);
1272              pos != NULL;
1273              pos = rb_next(pos)) {
1274                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1275                 spin_lock(&sp->so_lock);
1276                 list_for_each_entry(state, &sp->so_states, open_states) {
1277                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1278                                                 &state->flags))
1279                                 continue;
1280                         nfs4_state_mark_reclaim_nograce(clp, state);
1281                 }
1282                 spin_unlock(&sp->so_lock);
1283         }
1284         spin_unlock(&clp->cl_lock);
1285 }
1286
1287 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1288 {
1289         struct nfs_server *server;
1290
1291         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1292                 return 0;
1293
1294         rcu_read_lock();
1295         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1296                 nfs4_clear_reclaim_server(server);
1297         rcu_read_unlock();
1298
1299         nfs_delegation_reap_unclaimed(clp);
1300         return 1;
1301 }
1302
1303 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1304 {
1305         if (!nfs4_state_clear_reclaim_reboot(clp))
1306                 return;
1307         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1308 }
1309
1310 static void nfs_delegation_clear_all(struct nfs_client *clp)
1311 {
1312         nfs_delegation_mark_reclaim(clp);
1313         nfs_delegation_reap_unclaimed(clp);
1314 }
1315
1316 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1317 {
1318         nfs_delegation_clear_all(clp);
1319         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1320 }
1321
1322 static void nfs4_warn_keyexpired(const char *s)
1323 {
1324         printk_ratelimited(KERN_WARNING "Error: state manager"
1325                         " encountered RPCSEC_GSS session"
1326                         " expired against NFSv4 server %s.\n",
1327                         s);
1328 }
1329
1330 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1331 {
1332         switch (error) {
1333                 case -NFS4ERR_CB_PATH_DOWN:
1334                         nfs_handle_cb_pathdown(clp);
1335                         return 0;
1336                 case -NFS4ERR_NO_GRACE:
1337                         nfs4_state_end_reclaim_reboot(clp);
1338                         return 0;
1339                 case -NFS4ERR_STALE_CLIENTID:
1340                 case -NFS4ERR_LEASE_MOVED:
1341                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1342                         nfs4_state_clear_reclaim_reboot(clp);
1343                         nfs4_state_start_reclaim_reboot(clp);
1344                         break;
1345                 case -NFS4ERR_EXPIRED:
1346                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1347                         nfs4_state_start_reclaim_nograce(clp);
1348                         break;
1349                 case -NFS4ERR_BADSESSION:
1350                 case -NFS4ERR_BADSLOT:
1351                 case -NFS4ERR_BAD_HIGH_SLOT:
1352                 case -NFS4ERR_DEADSESSION:
1353                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1354                 case -NFS4ERR_SEQ_FALSE_RETRY:
1355                 case -NFS4ERR_SEQ_MISORDERED:
1356                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1357                         /* Zero session reset errors */
1358                         return 0;
1359                 case -EKEYEXPIRED:
1360                         /* Nothing we can do */
1361                         nfs4_warn_keyexpired(clp->cl_hostname);
1362                         return 0;
1363         }
1364         return error;
1365 }
1366
1367 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1368 {
1369         struct nfs4_state_owner *sp;
1370         struct nfs_server *server;
1371         struct rb_node *pos;
1372         int status = 0;
1373
1374 restart:
1375         rcu_read_lock();
1376         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1377                 spin_lock(&clp->cl_lock);
1378                 for (pos = rb_first(&server->state_owners);
1379                      pos != NULL;
1380                      pos = rb_next(pos)) {
1381                         sp = rb_entry(pos,
1382                                 struct nfs4_state_owner, so_server_node);
1383                         if (!test_and_clear_bit(ops->owner_flag_bit,
1384                                                         &sp->so_flags))
1385                                 continue;
1386                         atomic_inc(&sp->so_count);
1387                         spin_unlock(&clp->cl_lock);
1388                         rcu_read_unlock();
1389
1390                         status = nfs4_reclaim_open_state(sp, ops);
1391                         if (status < 0) {
1392                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1393                                 nfs4_put_state_owner(sp);
1394                                 return nfs4_recovery_handle_error(clp, status);
1395                         }
1396
1397                         nfs4_put_state_owner(sp);
1398                         goto restart;
1399                 }
1400                 spin_unlock(&clp->cl_lock);
1401         }
1402         rcu_read_unlock();
1403         return status;
1404 }
1405
1406 static int nfs4_check_lease(struct nfs_client *clp)
1407 {
1408         struct rpc_cred *cred;
1409         const struct nfs4_state_maintenance_ops *ops =
1410                 clp->cl_mvops->state_renewal_ops;
1411         int status = -NFS4ERR_EXPIRED;
1412
1413         /* Is the client already known to have an expired lease? */
1414         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1415                 return 0;
1416         spin_lock(&clp->cl_lock);
1417         cred = ops->get_state_renewal_cred_locked(clp);
1418         spin_unlock(&clp->cl_lock);
1419         if (cred == NULL) {
1420                 cred = nfs4_get_setclientid_cred(clp);
1421                 if (cred == NULL)
1422                         goto out;
1423         }
1424         status = ops->renew_lease(clp, cred);
1425         put_rpccred(cred);
1426 out:
1427         return nfs4_recovery_handle_error(clp, status);
1428 }
1429
1430 static int nfs4_reclaim_lease(struct nfs_client *clp)
1431 {
1432         struct rpc_cred *cred;
1433         const struct nfs4_state_recovery_ops *ops =
1434                 clp->cl_mvops->reboot_recovery_ops;
1435         int status = -ENOENT;
1436
1437         cred = ops->get_clid_cred(clp);
1438         if (cred != NULL) {
1439                 status = ops->establish_clid(clp, cred);
1440                 put_rpccred(cred);
1441                 /* Handle case where the user hasn't set up machine creds */
1442                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1443                         nfs4_clear_machine_cred(clp);
1444                         status = -EAGAIN;
1445                 }
1446                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1447                         status = -EPROTONOSUPPORT;
1448         }
1449         return status;
1450 }
1451
1452 #ifdef CONFIG_NFS_V4_1
1453 void nfs4_schedule_session_recovery(struct nfs4_session *session)
1454 {
1455         nfs4_schedule_lease_recovery(session->clp);
1456 }
1457 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1458
1459 void nfs41_handle_recall_slot(struct nfs_client *clp)
1460 {
1461         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1462         nfs4_schedule_state_manager(clp);
1463 }
1464
1465 static void nfs4_reset_all_state(struct nfs_client *clp)
1466 {
1467         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1468                 clp->cl_boot_time = CURRENT_TIME;
1469                 nfs4_state_start_reclaim_nograce(clp);
1470                 nfs4_schedule_state_manager(clp);
1471         }
1472 }
1473
1474 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1475 {
1476         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1477                 nfs4_state_start_reclaim_reboot(clp);
1478                 nfs4_schedule_state_manager(clp);
1479         }
1480 }
1481
1482 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1483 {
1484         /* Temporary */
1485         nfs4_reset_all_state(clp);
1486 }
1487
1488 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1489 {
1490         /* This will need to handle layouts too */
1491         nfs_expire_all_delegations(clp);
1492 }
1493
1494 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1495 {
1496         nfs_expire_all_delegations(clp);
1497         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1498                 nfs4_schedule_state_manager(clp);
1499 }
1500
1501 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1502 {
1503         if (!flags)
1504                 return;
1505         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1506                 nfs41_handle_server_reboot(clp);
1507         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1508                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1509                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1510                             SEQ4_STATUS_LEASE_MOVED))
1511                 nfs41_handle_state_revoked(clp);
1512         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1513                 nfs41_handle_recallable_state_revoked(clp);
1514         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1515                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1516                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1517                 nfs41_handle_cb_path_down(clp);
1518 }
1519
1520 static int nfs4_reset_session(struct nfs_client *clp)
1521 {
1522         int status;
1523
1524         nfs4_begin_drain_session(clp);
1525         status = nfs4_proc_destroy_session(clp->cl_session);
1526         if (status && status != -NFS4ERR_BADSESSION &&
1527             status != -NFS4ERR_DEADSESSION) {
1528                 status = nfs4_recovery_handle_error(clp, status);
1529                 goto out;
1530         }
1531
1532         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1533         status = nfs4_proc_create_session(clp);
1534         if (status) {
1535                 status = nfs4_recovery_handle_error(clp, status);
1536                 goto out;
1537         }
1538         /* create_session negotiated new slot table */
1539         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1540
1541          /* Let the state manager reestablish state */
1542         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1543                 nfs41_setup_state_renewal(clp);
1544 out:
1545         return status;
1546 }
1547
1548 static int nfs4_recall_slot(struct nfs_client *clp)
1549 {
1550         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1551         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1552         struct nfs4_slot *new, *old;
1553         int i;
1554
1555         nfs4_begin_drain_session(clp);
1556         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1557                       GFP_NOFS);
1558         if (!new)
1559                 return -ENOMEM;
1560
1561         spin_lock(&fc_tbl->slot_tbl_lock);
1562         for (i = 0; i < fc_tbl->target_max_slots; i++)
1563                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1564         old = fc_tbl->slots;
1565         fc_tbl->slots = new;
1566         fc_tbl->max_slots = fc_tbl->target_max_slots;
1567         fc_tbl->target_max_slots = 0;
1568         fc_attrs->max_reqs = fc_tbl->max_slots;
1569         spin_unlock(&fc_tbl->slot_tbl_lock);
1570
1571         kfree(old);
1572         nfs4_end_drain_session(clp);
1573         return 0;
1574 }
1575
1576 #else /* CONFIG_NFS_V4_1 */
1577 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1578 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1579 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1580 #endif /* CONFIG_NFS_V4_1 */
1581
1582 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1583  * on EXCHANGE_ID for v4.1
1584  */
1585 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1586 {
1587         if (nfs4_has_session(clp)) {
1588                 switch (status) {
1589                 case -NFS4ERR_DELAY:
1590                 case -NFS4ERR_CLID_INUSE:
1591                 case -EAGAIN:
1592                         break;
1593
1594                 case -EKEYEXPIRED:
1595                         nfs4_warn_keyexpired(clp->cl_hostname);
1596                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1597                                          * in nfs4_exchange_id */
1598                 default:
1599                         return;
1600                 }
1601         }
1602         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1603 }
1604
1605 static void nfs4_state_manager(struct nfs_client *clp)
1606 {
1607         int status = 0;
1608
1609         /* Ensure exclusive access to NFSv4 state */
1610         for(;;) {
1611                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1612                         /* We're going to have to re-establish a clientid */
1613                         status = nfs4_reclaim_lease(clp);
1614                         if (status) {
1615                                 nfs4_set_lease_expired(clp, status);
1616                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1617                                                         &clp->cl_state))
1618                                         continue;
1619                                 if (clp->cl_cons_state ==
1620                                                         NFS_CS_SESSION_INITING)
1621                                         nfs_mark_client_ready(clp, status);
1622                                 goto out_error;
1623                         }
1624                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1625                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1626                         pnfs_destroy_all_layouts(clp);
1627                 }
1628
1629                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1630                         status = nfs4_check_lease(clp);
1631                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1632                                 continue;
1633                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1634                                 goto out_error;
1635                 }
1636
1637                 /* Initialize or reset the session */
1638                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1639                    && nfs4_has_session(clp)) {
1640                         status = nfs4_reset_session(clp);
1641                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1642                                 continue;
1643                         if (status < 0)
1644                                 goto out_error;
1645                 }
1646
1647                 /* First recover reboot state... */
1648                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1649                         status = nfs4_do_reclaim(clp,
1650                                 clp->cl_mvops->reboot_recovery_ops);
1651                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1652                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1653                                 continue;
1654                         nfs4_state_end_reclaim_reboot(clp);
1655                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1656                                 continue;
1657                         if (status < 0)
1658                                 goto out_error;
1659                 }
1660
1661                 /* Now recover expired state... */
1662                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1663                         status = nfs4_do_reclaim(clp,
1664                                 clp->cl_mvops->nograce_recovery_ops);
1665                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1666                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1667                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1668                                 continue;
1669                         if (status < 0)
1670                                 goto out_error;
1671                 }
1672
1673                 nfs4_end_drain_session(clp);
1674                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1675                         nfs_client_return_marked_delegations(clp);
1676                         continue;
1677                 }
1678                 /* Recall session slots */
1679                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1680                    && nfs4_has_session(clp)) {
1681                         status = nfs4_recall_slot(clp);
1682                         if (status < 0)
1683                                 goto out_error;
1684                         continue;
1685                 }
1686
1687
1688                 nfs4_clear_state_manager_bit(clp);
1689                 /* Did we race with an attempt to give us more work? */
1690                 if (clp->cl_state == 0)
1691                         break;
1692                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1693                         break;
1694         }
1695         return;
1696 out_error:
1697         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1698                         " with error %d\n", clp->cl_hostname, -status);
1699         nfs4_end_drain_session(clp);
1700         nfs4_clear_state_manager_bit(clp);
1701 }
1702
1703 static int nfs4_run_state_manager(void *ptr)
1704 {
1705         struct nfs_client *clp = ptr;
1706
1707         allow_signal(SIGKILL);
1708         nfs4_state_manager(clp);
1709         nfs_put_client(clp);
1710         module_put_and_exit(0);
1711         return 0;
1712 }
1713
1714 /*
1715  * Local variables:
1716  *  c-basic-offset: 8
1717  * End:
1718  */