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         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 157         if (status == 0) {
 158                 /* Update lease time and schedule renewal */
 159                 spin_lock(&clp->cl_lock);
 160                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 161                 clp->cl_last_renewal = jiffies;
 162                 spin_unlock(&clp->cl_lock);
 163
 164                 nfs4_schedule_state_renewal(clp);
 165         }
 166
 167         return status;
 168 }
 169
 170 /*
 171  * Back channel returns NFS4ERR_DELAY for new requests when
 172  * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
 173  * is ended.
 174  */
 175 static void nfs4_end_drain_session(struct nfs_client *clp)
 176 {
 177         struct nfs4_session *ses = clp->cl_session;
 178         int max_slots;
 179
 180         if (ses == NULL)
 181                 return;
 182         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 183                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 184                 max_slots = ses->fc_slot_table.max_slots;
 185                 while (max_slots--) {
 186                         struct rpc_task *task;
 187
 188                         task = rpc_wake_up_next(&ses->fc_slot_table.
 189                                                 slot_tbl_waitq);
 190                         if (!task)
 191                                 break;
 192                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 193                 }
 194                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 195         }
 196 }
 197
 198 static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
 199 {
 200         spin_lock(&tbl->slot_tbl_lock);
 201         if (tbl->highest_used_slotid != -1) {
 202                 INIT_COMPLETION(tbl->complete);
 203                 spin_unlock(&tbl->slot_tbl_lock);
 204                 return wait_for_completion_interruptible(&tbl->complete);
 205         }
 206         spin_unlock(&tbl->slot_tbl_lock);
 207         return 0;
 208 }
 209
 210 static int nfs4_begin_drain_session(struct nfs_client *clp)
 211 {
 212         struct nfs4_session *ses = clp->cl_session;
 213         int ret = 0;
 214
 215         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 216         /* back channel */
 217         ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
 218         if (ret)
 219                 return ret;
 220         /* fore channel */
 221         return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
 222 }
 223
 224 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 225 {
 226         int status;
 227
 228         nfs4_begin_drain_session(clp);
 229         status = nfs4_proc_exchange_id(clp, cred);
 230         if (status != 0)
 231                 goto out;
 232         status = nfs4_proc_create_session(clp);
 233         if (status != 0)
 234                 goto out;
 235         status = nfs4_set_callback_sessionid(clp);
 236         if (status != 0) {
 237                 printk(KERN_WARNING "Sessionid not set. No callback service\n");
 238                 nfs_callback_down(1);
 239                 status = 0;
 240         }
 241         nfs41_setup_state_renewal(clp);
 242         nfs_mark_client_ready(clp, NFS_CS_READY);
 243 out:
 244         return status;
 245 }
 246
 247 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 248 {
 249         struct rpc_cred *cred;
 250
 251         spin_lock(&clp->cl_lock);
 252         cred = nfs4_get_machine_cred_locked(clp);
 253         spin_unlock(&clp->cl_lock);
 254         return cred;
 255 }
 256
 257 #endif /* CONFIG_NFS_V4_1 */
 258
 259 static struct rpc_cred *
 260 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 261 {
 262         struct nfs_client *clp = server->nfs_client;
 263         struct rpc_cred *cred = NULL;
 264         struct nfs4_state_owner *sp;
 265         struct rb_node *pos;
 266
 267         spin_lock(&clp->cl_lock);
 268         pos = rb_first(&server->state_owners);
 269         if (pos != NULL) {
 270                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 271                 cred = get_rpccred(sp->so_cred);
 272         }
 273         spin_unlock(&clp->cl_lock);
 274         return cred;
 275 }
 276
 277 /**
 278  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 279  * @clp: client state handle
 280  *
 281  * Returns an rpc_cred with reference count bumped, or NULL.
 282  */
 283 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 284 {
 285         struct nfs_server *server;
 286         struct rpc_cred *cred;
 287
 288         spin_lock(&clp->cl_lock);
 289         cred = nfs4_get_machine_cred_locked(clp);
 290         spin_unlock(&clp->cl_lock);
 291         if (cred != NULL)
 292                 goto out;
 293
 294         rcu_read_lock();
 295         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 296                 cred = nfs4_get_setclientid_cred_server(server);
 297                 if (cred != NULL)
 298                         break;
 299         }
 300         rcu_read_unlock();
 301
 302 out:
 303         return cred;
 304 }
 305
 306 static void nfs_alloc_unique_id_locked(struct rb_root *root,
 307                                        struct nfs_unique_id *new,
 308                                        __u64 minval, int maxbits)
 309 {
 310         struct rb_node **p, *parent;
 311         struct nfs_unique_id *pos;
 312         __u64 mask = ~0ULL;
 313
 314         if (maxbits < 64)
 315                 mask = (1ULL << maxbits) - 1ULL;
 316
 317         /* Ensure distribution is more or less flat */
 318         get_random_bytes(&new->id, sizeof(new->id));
 319         new->id &= mask;
 320         if (new->id < minval)
 321                 new->id += minval;
 322 retry:
 323         p = &root->rb_node;
 324         parent = NULL;
 325
 326         while (*p != NULL) {
 327                 parent = *p;
 328                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 329
 330                 if (new->id < pos->id)
 331                         p = &(*p)->rb_left;
 332                 else if (new->id > pos->id)
 333                         p = &(*p)->rb_right;
 334                 else
 335                         goto id_exists;
 336         }
 337         rb_link_node(&new->rb_node, parent, p);
 338         rb_insert_color(&new->rb_node, root);
 339         return;
 340 id_exists:
 341         for (;;) {
 342                 new->id++;
 343                 if (new->id < minval || (new->id & mask) != new->id) {
 344                         new->id = minval;
 345                         break;
 346                 }
 347                 parent = rb_next(parent);
 348                 if (parent == NULL)
 349                         break;
 350                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 351                 if (new->id < pos->id)
 352                         break;
 353         }
 354         goto retry;
 355 }
 356
 357 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 358 {
 359         rb_erase(&id->rb_node, root);
 360 }
 361
 362 static struct nfs4_state_owner *
 363 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 364 {
 365         struct rb_node **p = &server->state_owners.rb_node,
 366                        *parent = NULL;
 367         struct nfs4_state_owner *sp, *res = NULL;
 368
 369         while (*p != NULL) {
 370                 parent = *p;
 371                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 372
 373                 if (server < sp->so_server) {
 374                         p = &parent->rb_left;
 375                         continue;
 376                 }
 377                 if (server > sp->so_server) {
 378                         p = &parent->rb_right;
 379                         continue;
 380                 }
 381                 if (cred < sp->so_cred)
 382                         p = &parent->rb_left;
 383                 else if (cred > sp->so_cred)
 384                         p = &parent->rb_right;
 385                 else {
 386                         atomic_inc(&sp->so_count);
 387                         res = sp;
 388                         break;
 389                 }
 390         }
 391         return res;
 392 }
 393
 394 static struct nfs4_state_owner *
 395 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 396 {
 397         struct nfs_server *server = new->so_server;
 398         struct rb_node **p = &server->state_owners.rb_node,
 399                        *parent = NULL;
 400         struct nfs4_state_owner *sp;
 401
 402         while (*p != NULL) {
 403                 parent = *p;
 404                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 405
 406                 if (new->so_cred < sp->so_cred)
 407                         p = &parent->rb_left;
 408                 else if (new->so_cred > sp->so_cred)
 409                         p = &parent->rb_right;
 410                 else {
 411                         atomic_inc(&sp->so_count);
 412                         return sp;
 413                 }
 414         }
 415         nfs_alloc_unique_id_locked(&server->openowner_id,
 416                                         &new->so_owner_id, 1, 64);
 417         rb_link_node(&new->so_server_node, parent, p);
 418         rb_insert_color(&new->so_server_node, &server->state_owners);
 419         return new;
 420 }
 421
 422 static void
 423 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 424 {
 425         struct nfs_server *server = sp->so_server;
 426
 427         if (!RB_EMPTY_NODE(&sp->so_server_node))
 428                 rb_erase(&sp->so_server_node, &server->state_owners);
 429         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
 430 }
 431
 432 /*
 433  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 434  * create a new state_owner.
 435  *
 436  */
 437 static struct nfs4_state_owner *
 438 nfs4_alloc_state_owner(void)
 439 {
 440         struct nfs4_state_owner *sp;
 441
 442         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 443         if (!sp)
 444                 return NULL;
 445         spin_lock_init(&sp->so_lock);
 446         INIT_LIST_HEAD(&sp->so_states);
 447         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 448         sp->so_seqid.sequence = &sp->so_sequence;
 449         spin_lock_init(&sp->so_sequence.lock);
 450         INIT_LIST_HEAD(&sp->so_sequence.list);
 451         atomic_set(&sp->so_count, 1);
 452         return sp;
 453 }
 454
 455 static void
 456 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 457 {
 458         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 459                 struct nfs_server *server = sp->so_server;
 460                 struct nfs_client *clp = server->nfs_client;
 461
 462                 spin_lock(&clp->cl_lock);
 463                 rb_erase(&sp->so_server_node, &server->state_owners);
 464                 RB_CLEAR_NODE(&sp->so_server_node);
 465                 spin_unlock(&clp->cl_lock);
 466         }
 467 }
 468
 469 /**
 470  * nfs4_get_state_owner - Look up a state owner given a credential
 471  * @server: nfs_server to search
 472  * @cred: RPC credential to match
 473  *
 474  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 475  */
 476 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 477                                               struct rpc_cred *cred)
 478 {
 479         struct nfs_client *clp = server->nfs_client;
 480         struct nfs4_state_owner *sp, *new;
 481
 482         spin_lock(&clp->cl_lock);
 483         sp = nfs4_find_state_owner_locked(server, cred);
 484         spin_unlock(&clp->cl_lock);
 485         if (sp != NULL)
 486                 return sp;
 487         new = nfs4_alloc_state_owner();
 488         if (new == NULL)
 489                 return NULL;
 490         new->so_server = server;
 491         new->so_cred = cred;
 492         spin_lock(&clp->cl_lock);
 493         sp = nfs4_insert_state_owner_locked(new);
 494         spin_unlock(&clp->cl_lock);
 495         if (sp == new)
 496                 get_rpccred(cred);
 497         else {
 498                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 499                 kfree(new);
 500         }
 501         return sp;
 502 }
 503
 504 /**
 505  * nfs4_put_state_owner - Release a nfs4_state_owner
 506  * @sp: state owner data to release
 507  *
 508  */
 509 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 510 {
 511         struct nfs_client *clp = sp->so_server->nfs_client;
 512         struct rpc_cred *cred = sp->so_cred;
 513
 514         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 515                 return;
 516         nfs4_remove_state_owner_locked(sp);
 517         spin_unlock(&clp->cl_lock);
 518         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 519         put_rpccred(cred);
 520         kfree(sp);
 521 }
 522
 523 static struct nfs4_state *
 524 nfs4_alloc_open_state(void)
 525 {
 526         struct nfs4_state *state;
 527
 528         state = kzalloc(sizeof(*state), GFP_NOFS);
 529         if (!state)
 530                 return NULL;
 531         atomic_set(&state->count, 1);
 532         INIT_LIST_HEAD(&state->lock_states);
 533         spin_lock_init(&state->state_lock);
 534         seqlock_init(&state->seqlock);
 535         return state;
 536 }
 537
 538 void
 539 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 540 {
 541         if (state->state == fmode)
 542                 return;
 543         /* NB! List reordering - see the reclaim code for why.  */
 544         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 545                 if (fmode & FMODE_WRITE)
 546                         list_move(&state->open_states, &state->owner->so_states);
 547                 else
 548                         list_move_tail(&state->open_states, &state->owner->so_states);
 549         }
 550         state->state = fmode;
 551 }
 552
 553 static struct nfs4_state *
 554 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 555 {
 556         struct nfs_inode *nfsi = NFS_I(inode);
 557         struct nfs4_state *state;
 558
 559         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 560                 if (state->owner != owner)
 561                         continue;
 562                 if (atomic_inc_not_zero(&state->count))
 563                         return state;
 564         }
 565         return NULL;
 566 }
 567
 568 static void
 569 nfs4_free_open_state(struct nfs4_state *state)
 570 {
 571         kfree(state);
 572 }
 573
 574 struct nfs4_state *
 575 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 576 {
 577         struct nfs4_state *state, *new;
 578         struct nfs_inode *nfsi = NFS_I(inode);
 579
 580         spin_lock(&inode->i_lock);
 581         state = __nfs4_find_state_byowner(inode, owner);
 582         spin_unlock(&inode->i_lock);
 583         if (state)
 584                 goto out;
 585         new = nfs4_alloc_open_state();
 586         spin_lock(&owner->so_lock);
 587         spin_lock(&inode->i_lock);
 588         state = __nfs4_find_state_byowner(inode, owner);
 589         if (state == NULL && new != NULL) {
 590                 state = new;
 591                 state->owner = owner;
 592                 atomic_inc(&owner->so_count);
 593                 list_add(&state->inode_states, &nfsi->open_states);
 594                 state->inode = igrab(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 state recovery attempt
1017  */
1018 void nfs4_schedule_state_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 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 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 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1051 {
1052         struct inode *inode = state->inode;
1053         struct nfs_inode *nfsi = NFS_I(inode);
1054         struct file_lock *fl;
1055         int status = 0;
1056
1057         if (inode->i_flock == NULL)
1058                 return 0;
1059
1060         /* Guard against delegation returns and new lock/unlock calls */
1061         down_write(&nfsi->rwsem);
1062         /* Protect inode->i_flock using the BKL */
1063         lock_flocks();
1064         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1065                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1066                         continue;
1067                 if (nfs_file_open_context(fl->fl_file)->state != state)
1068                         continue;
1069                 unlock_flocks();
1070                 status = ops->recover_lock(state, fl);
1071                 switch (status) {
1072                         case 0:
1073                                 break;
1074                         case -ESTALE:
1075                         case -NFS4ERR_ADMIN_REVOKED:
1076                         case -NFS4ERR_STALE_STATEID:
1077                         case -NFS4ERR_BAD_STATEID:
1078                         case -NFS4ERR_EXPIRED:
1079                         case -NFS4ERR_NO_GRACE:
1080                         case -NFS4ERR_STALE_CLIENTID:
1081                         case -NFS4ERR_BADSESSION:
1082                         case -NFS4ERR_BADSLOT:
1083                         case -NFS4ERR_BAD_HIGH_SLOT:
1084                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1085                                 goto out;
1086                         default:
1087                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1088                                                 __func__, status);
1089                         case -ENOMEM:
1090                         case -NFS4ERR_DENIED:
1091                         case -NFS4ERR_RECLAIM_BAD:
1092                         case -NFS4ERR_RECLAIM_CONFLICT:
1093                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1094                                 status = 0;
1095                 }
1096                 lock_flocks();
1097         }
1098         unlock_flocks();
1099 out:
1100         up_write(&nfsi->rwsem);
1101         return status;
1102 }
1103
1104 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1105 {
1106         struct nfs4_state *state;
1107         struct nfs4_lock_state *lock;
1108         int status = 0;
1109
1110         /* Note: we rely on the sp->so_states list being ordered
1111          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1112          * states first.
1113          * This is needed to ensure that the server won't give us any
1114          * read delegations that we have to return if, say, we are
1115          * recovering after a network partition or a reboot from a
1116          * server that doesn't support a grace period.
1117          */
1118 restart:
1119         spin_lock(&sp->so_lock);
1120         list_for_each_entry(state, &sp->so_states, open_states) {
1121                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1122                         continue;
1123                 if (state->state == 0)
1124                         continue;
1125                 atomic_inc(&state->count);
1126                 spin_unlock(&sp->so_lock);
1127                 status = ops->recover_open(sp, state);
1128                 if (status >= 0) {
1129                         status = nfs4_reclaim_locks(state, ops);
1130                         if (status >= 0) {
1131                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1132                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1133                                                 printk("%s: Lock reclaim failed!\n",
1134                                                         __func__);
1135                                 }
1136                                 nfs4_put_open_state(state);
1137                                 goto restart;
1138                         }
1139                 }
1140                 switch (status) {
1141                         default:
1142                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1143                                                 __func__, status);
1144                         case -ENOENT:
1145                         case -ENOMEM:
1146                         case -ESTALE:
1147                                 /*
1148                                  * Open state on this file cannot be recovered
1149                                  * All we can do is revert to using the zero stateid.
1150                                  */
1151                                 memset(state->stateid.data, 0,
1152                                         sizeof(state->stateid.data));
1153                                 /* Mark the file as being 'closed' */
1154                                 state->state = 0;
1155                                 break;
1156                         case -EKEYEXPIRED:
1157                                 /*
1158                                  * User RPCSEC_GSS context has expired.
1159                                  * We cannot recover this stateid now, so
1160                                  * skip it and allow recovery thread to
1161                                  * proceed.
1162                                  */
1163                                 break;
1164                         case -NFS4ERR_ADMIN_REVOKED:
1165                         case -NFS4ERR_STALE_STATEID:
1166                         case -NFS4ERR_BAD_STATEID:
1167                         case -NFS4ERR_RECLAIM_BAD:
1168                         case -NFS4ERR_RECLAIM_CONFLICT:
1169                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1170                                 break;
1171                         case -NFS4ERR_EXPIRED:
1172                         case -NFS4ERR_NO_GRACE:
1173                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1174                         case -NFS4ERR_STALE_CLIENTID:
1175                         case -NFS4ERR_BADSESSION:
1176                         case -NFS4ERR_BADSLOT:
1177                         case -NFS4ERR_BAD_HIGH_SLOT:
1178                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1179                                 goto out_err;
1180                 }
1181                 nfs4_put_open_state(state);
1182                 goto restart;
1183         }
1184         spin_unlock(&sp->so_lock);
1185         return 0;
1186 out_err:
1187         nfs4_put_open_state(state);
1188         return status;
1189 }
1190
1191 static void nfs4_clear_open_state(struct nfs4_state *state)
1192 {
1193         struct nfs4_lock_state *lock;
1194
1195         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1196         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1197         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1198         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1199         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1200                 lock->ls_seqid.flags = 0;
1201                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1202         }
1203 }
1204
1205 static void nfs4_reset_seqids(struct nfs_server *server,
1206         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1207 {
1208         struct nfs_client *clp = server->nfs_client;
1209         struct nfs4_state_owner *sp;
1210         struct rb_node *pos;
1211         struct nfs4_state *state;
1212
1213         spin_lock(&clp->cl_lock);
1214         for (pos = rb_first(&server->state_owners);
1215              pos != NULL;
1216              pos = rb_next(pos)) {
1217                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1218                 sp->so_seqid.flags = 0;
1219                 spin_lock(&sp->so_lock);
1220                 list_for_each_entry(state, &sp->so_states, open_states) {
1221                         if (mark_reclaim(clp, state))
1222                                 nfs4_clear_open_state(state);
1223                 }
1224                 spin_unlock(&sp->so_lock);
1225         }
1226         spin_unlock(&clp->cl_lock);
1227 }
1228
1229 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1230         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1231 {
1232         struct nfs_server *server;
1233
1234         rcu_read_lock();
1235         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1236                 nfs4_reset_seqids(server, mark_reclaim);
1237         rcu_read_unlock();
1238 }
1239
1240 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1241 {
1242         /* Mark all delegations for reclaim */
1243         nfs_delegation_mark_reclaim(clp);
1244         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1245 }
1246
1247 static void nfs4_reclaim_complete(struct nfs_client *clp,
1248                                  const struct nfs4_state_recovery_ops *ops)
1249 {
1250         /* Notify the server we're done reclaiming our state */
1251         if (ops->reclaim_complete)
1252                 (void)ops->reclaim_complete(clp);
1253 }
1254
1255 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1256 {
1257         struct nfs_client *clp = server->nfs_client;
1258         struct nfs4_state_owner *sp;
1259         struct rb_node *pos;
1260         struct nfs4_state *state;
1261
1262         spin_lock(&clp->cl_lock);
1263         for (pos = rb_first(&server->state_owners);
1264              pos != NULL;
1265              pos = rb_next(pos)) {
1266                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1267                 spin_lock(&sp->so_lock);
1268                 list_for_each_entry(state, &sp->so_states, open_states) {
1269                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1270                                                 &state->flags))
1271                                 continue;
1272                         nfs4_state_mark_reclaim_nograce(clp, state);
1273                 }
1274                 spin_unlock(&sp->so_lock);
1275         }
1276         spin_unlock(&clp->cl_lock);
1277 }
1278
1279 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1280 {
1281         struct nfs_server *server;
1282
1283         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1284                 return 0;
1285
1286         rcu_read_lock();
1287         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1288                 nfs4_clear_reclaim_server(server);
1289         rcu_read_unlock();
1290
1291         nfs_delegation_reap_unclaimed(clp);
1292         return 1;
1293 }
1294
1295 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1296 {
1297         if (!nfs4_state_clear_reclaim_reboot(clp))
1298                 return;
1299         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1300 }
1301
1302 static void nfs_delegation_clear_all(struct nfs_client *clp)
1303 {
1304         nfs_delegation_mark_reclaim(clp);
1305         nfs_delegation_reap_unclaimed(clp);
1306 }
1307
1308 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1309 {
1310         nfs_delegation_clear_all(clp);
1311         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1312 }
1313
1314 static void nfs4_warn_keyexpired(const char *s)
1315 {
1316         printk_ratelimited(KERN_WARNING "Error: state manager"
1317                         " encountered RPCSEC_GSS session"
1318                         " expired against NFSv4 server %s.\n",
1319                         s);
1320 }
1321
1322 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1323 {
1324         switch (error) {
1325                 case -NFS4ERR_CB_PATH_DOWN:
1326                         nfs_handle_cb_pathdown(clp);
1327                         return 0;
1328                 case -NFS4ERR_NO_GRACE:
1329                         nfs4_state_end_reclaim_reboot(clp);
1330                         return 0;
1331                 case -NFS4ERR_STALE_CLIENTID:
1332                 case -NFS4ERR_LEASE_MOVED:
1333                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1334                         nfs4_state_clear_reclaim_reboot(clp);
1335                         nfs4_state_start_reclaim_reboot(clp);
1336                         break;
1337                 case -NFS4ERR_EXPIRED:
1338                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1339                         nfs4_state_start_reclaim_nograce(clp);
1340                         break;
1341                 case -NFS4ERR_BADSESSION:
1342                 case -NFS4ERR_BADSLOT:
1343                 case -NFS4ERR_BAD_HIGH_SLOT:
1344                 case -NFS4ERR_DEADSESSION:
1345                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1346                 case -NFS4ERR_SEQ_FALSE_RETRY:
1347                 case -NFS4ERR_SEQ_MISORDERED:
1348                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1349                         /* Zero session reset errors */
1350                         return 0;
1351                 case -EKEYEXPIRED:
1352                         /* Nothing we can do */
1353                         nfs4_warn_keyexpired(clp->cl_hostname);
1354                         return 0;
1355         }
1356         return error;
1357 }
1358
1359 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1360 {
1361         struct nfs4_state_owner *sp;
1362         struct nfs_server *server;
1363         struct rb_node *pos;
1364         int status = 0;
1365
1366 restart:
1367         rcu_read_lock();
1368         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1369                 spin_lock(&clp->cl_lock);
1370                 for (pos = rb_first(&server->state_owners);
1371                      pos != NULL;
1372                      pos = rb_next(pos)) {
1373                         sp = rb_entry(pos,
1374                                 struct nfs4_state_owner, so_server_node);
1375                         if (!test_and_clear_bit(ops->owner_flag_bit,
1376                                                         &sp->so_flags))
1377                                 continue;
1378                         atomic_inc(&sp->so_count);
1379                         spin_unlock(&clp->cl_lock);
1380                         rcu_read_unlock();
1381
1382                         status = nfs4_reclaim_open_state(sp, ops);
1383                         if (status < 0) {
1384                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1385                                 nfs4_put_state_owner(sp);
1386                                 return nfs4_recovery_handle_error(clp, status);
1387                         }
1388
1389                         nfs4_put_state_owner(sp);
1390                         goto restart;
1391                 }
1392                 spin_unlock(&clp->cl_lock);
1393         }
1394         rcu_read_unlock();
1395         return status;
1396 }
1397
1398 static int nfs4_check_lease(struct nfs_client *clp)
1399 {
1400         struct rpc_cred *cred;
1401         const struct nfs4_state_maintenance_ops *ops =
1402                 clp->cl_mvops->state_renewal_ops;
1403         int status = -NFS4ERR_EXPIRED;
1404
1405         /* Is the client already known to have an expired lease? */
1406         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1407                 return 0;
1408         spin_lock(&clp->cl_lock);
1409         cred = ops->get_state_renewal_cred_locked(clp);
1410         spin_unlock(&clp->cl_lock);
1411         if (cred == NULL) {
1412                 cred = nfs4_get_setclientid_cred(clp);
1413                 if (cred == NULL)
1414                         goto out;
1415         }
1416         status = ops->renew_lease(clp, cred);
1417         put_rpccred(cred);
1418 out:
1419         return nfs4_recovery_handle_error(clp, status);
1420 }
1421
1422 static int nfs4_reclaim_lease(struct nfs_client *clp)
1423 {
1424         struct rpc_cred *cred;
1425         const struct nfs4_state_recovery_ops *ops =
1426                 clp->cl_mvops->reboot_recovery_ops;
1427         int status = -ENOENT;
1428
1429         cred = ops->get_clid_cred(clp);
1430         if (cred != NULL) {
1431                 status = ops->establish_clid(clp, cred);
1432                 put_rpccred(cred);
1433                 /* Handle case where the user hasn't set up machine creds */
1434                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1435                         nfs4_clear_machine_cred(clp);
1436                         status = -EAGAIN;
1437                 }
1438                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1439                         status = -EPROTONOSUPPORT;
1440         }
1441         return status;
1442 }
1443
1444 #ifdef CONFIG_NFS_V4_1
1445 void nfs41_handle_recall_slot(struct nfs_client *clp)
1446 {
1447         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1448         nfs4_schedule_state_recovery(clp);
1449 }
1450
1451 static void nfs4_reset_all_state(struct nfs_client *clp)
1452 {
1453         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1454                 clp->cl_boot_time = CURRENT_TIME;
1455                 nfs4_state_start_reclaim_nograce(clp);
1456                 nfs4_schedule_state_recovery(clp);
1457         }
1458 }
1459
1460 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1461 {
1462         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1463                 nfs4_state_start_reclaim_reboot(clp);
1464                 nfs4_schedule_state_recovery(clp);
1465         }
1466 }
1467
1468 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1469 {
1470         /* Temporary */
1471         nfs4_reset_all_state(clp);
1472 }
1473
1474 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1475 {
1476         /* This will need to handle layouts too */
1477         nfs_expire_all_delegations(clp);
1478 }
1479
1480 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1481 {
1482         nfs_expire_all_delegations(clp);
1483         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1484                 nfs4_schedule_state_recovery(clp);
1485 }
1486
1487 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1488 {
1489         if (!flags)
1490                 return;
1491         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1492                 nfs41_handle_server_reboot(clp);
1493         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1494                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1495                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1496                             SEQ4_STATUS_LEASE_MOVED))
1497                 nfs41_handle_state_revoked(clp);
1498         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1499                 nfs41_handle_recallable_state_revoked(clp);
1500         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1501                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1502                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1503                 nfs41_handle_cb_path_down(clp);
1504 }
1505
1506 static int nfs4_reset_session(struct nfs_client *clp)
1507 {
1508         int status;
1509
1510         nfs4_begin_drain_session(clp);
1511         status = nfs4_proc_destroy_session(clp->cl_session);
1512         if (status && status != -NFS4ERR_BADSESSION &&
1513             status != -NFS4ERR_DEADSESSION) {
1514                 status = nfs4_recovery_handle_error(clp, status);
1515                 goto out;
1516         }
1517
1518         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1519         status = nfs4_proc_create_session(clp);
1520         if (status) {
1521                 status = nfs4_recovery_handle_error(clp, status);
1522                 goto out;
1523         }
1524         /* create_session negotiated new slot table */
1525         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1526
1527          /* Let the state manager reestablish state */
1528         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1529                 nfs41_setup_state_renewal(clp);
1530 out:
1531         return status;
1532 }
1533
1534 static int nfs4_recall_slot(struct nfs_client *clp)
1535 {
1536         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1537         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1538         struct nfs4_slot *new, *old;
1539         int i;
1540
1541         nfs4_begin_drain_session(clp);
1542         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1543                       GFP_NOFS);
1544         if (!new)
1545                 return -ENOMEM;
1546
1547         spin_lock(&fc_tbl->slot_tbl_lock);
1548         for (i = 0; i < fc_tbl->target_max_slots; i++)
1549                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1550         old = fc_tbl->slots;
1551         fc_tbl->slots = new;
1552         fc_tbl->max_slots = fc_tbl->target_max_slots;
1553         fc_tbl->target_max_slots = 0;
1554         fc_attrs->max_reqs = fc_tbl->max_slots;
1555         spin_unlock(&fc_tbl->slot_tbl_lock);
1556
1557         kfree(old);
1558         nfs4_end_drain_session(clp);
1559         return 0;
1560 }
1561
1562 #else /* CONFIG_NFS_V4_1 */
1563 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1564 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1565 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1566 #endif /* CONFIG_NFS_V4_1 */
1567
1568 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1569  * on EXCHANGE_ID for v4.1
1570  */
1571 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1572 {
1573         if (nfs4_has_session(clp)) {
1574                 switch (status) {
1575                 case -NFS4ERR_DELAY:
1576                 case -NFS4ERR_CLID_INUSE:
1577                 case -EAGAIN:
1578                         break;
1579
1580                 case -EKEYEXPIRED:
1581                         nfs4_warn_keyexpired(clp->cl_hostname);
1582                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1583                                          * in nfs4_exchange_id */
1584                 default:
1585                         return;
1586                 }
1587         }
1588         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1589 }
1590
1591 static void nfs4_state_manager(struct nfs_client *clp)
1592 {
1593         int status = 0;
1594
1595         /* Ensure exclusive access to NFSv4 state */
1596         for(;;) {
1597                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1598                         /* We're going to have to re-establish a clientid */
1599                         status = nfs4_reclaim_lease(clp);
1600                         if (status) {
1601                                 nfs4_set_lease_expired(clp, status);
1602                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1603                                                         &clp->cl_state))
1604                                         continue;
1605                                 if (clp->cl_cons_state ==
1606                                                         NFS_CS_SESSION_INITING)
1607                                         nfs_mark_client_ready(clp, status);
1608                                 goto out_error;
1609                         }
1610                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1611                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1612                         pnfs_destroy_all_layouts(clp);
1613                 }
1614
1615                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1616                         status = nfs4_check_lease(clp);
1617                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1618                                 continue;
1619                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1620                                 goto out_error;
1621                 }
1622
1623                 /* Initialize or reset the session */
1624                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1625                    && nfs4_has_session(clp)) {
1626                         status = nfs4_reset_session(clp);
1627                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1628                                 continue;
1629                         if (status < 0)
1630                                 goto out_error;
1631                 }
1632
1633                 /* First recover reboot state... */
1634                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1635                         status = nfs4_do_reclaim(clp,
1636                                 clp->cl_mvops->reboot_recovery_ops);
1637                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1638                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1639                                 continue;
1640                         nfs4_state_end_reclaim_reboot(clp);
1641                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1642                                 continue;
1643                         if (status < 0)
1644                                 goto out_error;
1645                 }
1646
1647                 /* Now recover expired state... */
1648                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1649                         status = nfs4_do_reclaim(clp,
1650                                 clp->cl_mvops->nograce_recovery_ops);
1651                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1652                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1653                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1654                                 continue;
1655                         if (status < 0)
1656                                 goto out_error;
1657                 }
1658
1659                 nfs4_end_drain_session(clp);
1660                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1661                         nfs_client_return_marked_delegations(clp);
1662                         continue;
1663                 }
1664                 /* Recall session slots */
1665                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1666                    && nfs4_has_session(clp)) {
1667                         status = nfs4_recall_slot(clp);
1668                         if (status < 0)
1669                                 goto out_error;
1670                         continue;
1671                 }
1672
1673
1674                 nfs4_clear_state_manager_bit(clp);
1675                 /* Did we race with an attempt to give us more work? */
1676                 if (clp->cl_state == 0)
1677                         break;
1678                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1679                         break;
1680         }
1681         return;
1682 out_error:
1683         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1684                         " with error %d\n", clp->cl_hostname, -status);
1685         nfs4_end_drain_session(clp);
1686         nfs4_clear_state_manager_bit(clp);
1687 }
1688
1689 static int nfs4_run_state_manager(void *ptr)
1690 {
1691         struct nfs_client *clp = ptr;
1692
1693         allow_signal(SIGKILL);
1694         nfs4_state_manager(clp);
1695         nfs_put_client(clp);
1696         module_put_and_exit(0);
1697         return 0;
1698 }
1699
1700 /*
1701  * Local variables:
1702  *  c-basic-offset: 8
1703  * End:
1704  */