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         nfs41_setup_state_renewal(clp);
 236         nfs_mark_client_ready(clp, NFS_CS_READY);
 237 out:
 238         return status;
 239 }
 240
 241 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 242 {
 243         struct rpc_cred *cred;
 244
 245         spin_lock(&clp->cl_lock);
 246         cred = nfs4_get_machine_cred_locked(clp);
 247         spin_unlock(&clp->cl_lock);
 248         return cred;
 249 }
 250
 251 #endif /* CONFIG_NFS_V4_1 */
 252
 253 static struct rpc_cred *
 254 nfs4_get_setclientid_cred_server(struct nfs_server *server)
 255 {
 256         struct nfs_client *clp = server->nfs_client;
 257         struct rpc_cred *cred = NULL;
 258         struct nfs4_state_owner *sp;
 259         struct rb_node *pos;
 260
 261         spin_lock(&clp->cl_lock);
 262         pos = rb_first(&server->state_owners);
 263         if (pos != NULL) {
 264                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 265                 cred = get_rpccred(sp->so_cred);
 266         }
 267         spin_unlock(&clp->cl_lock);
 268         return cred;
 269 }
 270
 271 /**
 272  * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
 273  * @clp: client state handle
 274  *
 275  * Returns an rpc_cred with reference count bumped, or NULL.
 276  */
 277 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 278 {
 279         struct nfs_server *server;
 280         struct rpc_cred *cred;
 281
 282         spin_lock(&clp->cl_lock);
 283         cred = nfs4_get_machine_cred_locked(clp);
 284         spin_unlock(&clp->cl_lock);
 285         if (cred != NULL)
 286                 goto out;
 287
 288         rcu_read_lock();
 289         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 290                 cred = nfs4_get_setclientid_cred_server(server);
 291                 if (cred != NULL)
 292                         break;
 293         }
 294         rcu_read_unlock();
 295
 296 out:
 297         return cred;
 298 }
 299
 300 static void nfs_alloc_unique_id_locked(struct rb_root *root,
 301                                        struct nfs_unique_id *new,
 302                                        __u64 minval, int maxbits)
 303 {
 304         struct rb_node **p, *parent;
 305         struct nfs_unique_id *pos;
 306         __u64 mask = ~0ULL;
 307
 308         if (maxbits < 64)
 309                 mask = (1ULL << maxbits) - 1ULL;
 310
 311         /* Ensure distribution is more or less flat */
 312         get_random_bytes(&new->id, sizeof(new->id));
 313         new->id &= mask;
 314         if (new->id < minval)
 315                 new->id += minval;
 316 retry:
 317         p = &root->rb_node;
 318         parent = NULL;
 319
 320         while (*p != NULL) {
 321                 parent = *p;
 322                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 323
 324                 if (new->id < pos->id)
 325                         p = &(*p)->rb_left;
 326                 else if (new->id > pos->id)
 327                         p = &(*p)->rb_right;
 328                 else
 329                         goto id_exists;
 330         }
 331         rb_link_node(&new->rb_node, parent, p);
 332         rb_insert_color(&new->rb_node, root);
 333         return;
 334 id_exists:
 335         for (;;) {
 336                 new->id++;
 337                 if (new->id < minval || (new->id & mask) != new->id) {
 338                         new->id = minval;
 339                         break;
 340                 }
 341                 parent = rb_next(parent);
 342                 if (parent == NULL)
 343                         break;
 344                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 345                 if (new->id < pos->id)
 346                         break;
 347         }
 348         goto retry;
 349 }
 350
 351 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 352 {
 353         rb_erase(&id->rb_node, root);
 354 }
 355
 356 static struct nfs4_state_owner *
 357 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 358 {
 359         struct rb_node **p = &server->state_owners.rb_node,
 360                        *parent = NULL;
 361         struct nfs4_state_owner *sp, *res = NULL;
 362
 363         while (*p != NULL) {
 364                 parent = *p;
 365                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 366
 367                 if (server < sp->so_server) {
 368                         p = &parent->rb_left;
 369                         continue;
 370                 }
 371                 if (server > sp->so_server) {
 372                         p = &parent->rb_right;
 373                         continue;
 374                 }
 375                 if (cred < sp->so_cred)
 376                         p = &parent->rb_left;
 377                 else if (cred > sp->so_cred)
 378                         p = &parent->rb_right;
 379                 else {
 380                         atomic_inc(&sp->so_count);
 381                         res = sp;
 382                         break;
 383                 }
 384         }
 385         return res;
 386 }
 387
 388 static struct nfs4_state_owner *
 389 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 390 {
 391         struct nfs_server *server = new->so_server;
 392         struct rb_node **p = &server->state_owners.rb_node,
 393                        *parent = NULL;
 394         struct nfs4_state_owner *sp;
 395
 396         while (*p != NULL) {
 397                 parent = *p;
 398                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 399
 400                 if (new->so_cred < sp->so_cred)
 401                         p = &parent->rb_left;
 402                 else if (new->so_cred > sp->so_cred)
 403                         p = &parent->rb_right;
 404                 else {
 405                         atomic_inc(&sp->so_count);
 406                         return sp;
 407                 }
 408         }
 409         nfs_alloc_unique_id_locked(&server->openowner_id,
 410                                         &new->so_owner_id, 1, 64);
 411         rb_link_node(&new->so_server_node, parent, p);
 412         rb_insert_color(&new->so_server_node, &server->state_owners);
 413         return new;
 414 }
 415
 416 static void
 417 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 418 {
 419         struct nfs_server *server = sp->so_server;
 420
 421         if (!RB_EMPTY_NODE(&sp->so_server_node))
 422                 rb_erase(&sp->so_server_node, &server->state_owners);
 423         nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
 424 }
 425
 426 /*
 427  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 428  * create a new state_owner.
 429  *
 430  */
 431 static struct nfs4_state_owner *
 432 nfs4_alloc_state_owner(void)
 433 {
 434         struct nfs4_state_owner *sp;
 435
 436         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 437         if (!sp)
 438                 return NULL;
 439         spin_lock_init(&sp->so_lock);
 440         INIT_LIST_HEAD(&sp->so_states);
 441         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 442         sp->so_seqid.sequence = &sp->so_sequence;
 443         spin_lock_init(&sp->so_sequence.lock);
 444         INIT_LIST_HEAD(&sp->so_sequence.list);
 445         atomic_set(&sp->so_count, 1);
 446         return sp;
 447 }
 448
 449 static void
 450 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 451 {
 452         if (!RB_EMPTY_NODE(&sp->so_server_node)) {
 453                 struct nfs_server *server = sp->so_server;
 454                 struct nfs_client *clp = server->nfs_client;
 455
 456                 spin_lock(&clp->cl_lock);
 457                 rb_erase(&sp->so_server_node, &server->state_owners);
 458                 RB_CLEAR_NODE(&sp->so_server_node);
 459                 spin_unlock(&clp->cl_lock);
 460         }
 461 }
 462
 463 /**
 464  * nfs4_get_state_owner - Look up a state owner given a credential
 465  * @server: nfs_server to search
 466  * @cred: RPC credential to match
 467  *
 468  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 469  */
 470 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 471                                               struct rpc_cred *cred)
 472 {
 473         struct nfs_client *clp = server->nfs_client;
 474         struct nfs4_state_owner *sp, *new;
 475
 476         spin_lock(&clp->cl_lock);
 477         sp = nfs4_find_state_owner_locked(server, cred);
 478         spin_unlock(&clp->cl_lock);
 479         if (sp != NULL)
 480                 return sp;
 481         new = nfs4_alloc_state_owner();
 482         if (new == NULL)
 483                 return NULL;
 484         new->so_server = server;
 485         new->so_cred = cred;
 486         spin_lock(&clp->cl_lock);
 487         sp = nfs4_insert_state_owner_locked(new);
 488         spin_unlock(&clp->cl_lock);
 489         if (sp == new)
 490                 get_rpccred(cred);
 491         else {
 492                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 493                 kfree(new);
 494         }
 495         return sp;
 496 }
 497
 498 /**
 499  * nfs4_put_state_owner - Release a nfs4_state_owner
 500  * @sp: state owner data to release
 501  *
 502  */
 503 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 504 {
 505         struct nfs_client *clp = sp->so_server->nfs_client;
 506         struct rpc_cred *cred = sp->so_cred;
 507
 508         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 509                 return;
 510         nfs4_remove_state_owner_locked(sp);
 511         spin_unlock(&clp->cl_lock);
 512         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 513         put_rpccred(cred);
 514         kfree(sp);
 515 }
 516
 517 static struct nfs4_state *
 518 nfs4_alloc_open_state(void)
 519 {
 520         struct nfs4_state *state;
 521
 522         state = kzalloc(sizeof(*state), GFP_NOFS);
 523         if (!state)
 524                 return NULL;
 525         atomic_set(&state->count, 1);
 526         INIT_LIST_HEAD(&state->lock_states);
 527         spin_lock_init(&state->state_lock);
 528         seqlock_init(&state->seqlock);
 529         return state;
 530 }
 531
 532 void
 533 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 534 {
 535         if (state->state == fmode)
 536                 return;
 537         /* NB! List reordering - see the reclaim code for why.  */
 538         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 539                 if (fmode & FMODE_WRITE)
 540                         list_move(&state->open_states, &state->owner->so_states);
 541                 else
 542                         list_move_tail(&state->open_states, &state->owner->so_states);
 543         }
 544         state->state = fmode;
 545 }
 546
 547 static struct nfs4_state *
 548 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 549 {
 550         struct nfs_inode *nfsi = NFS_I(inode);
 551         struct nfs4_state *state;
 552
 553         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 554                 if (state->owner != owner)
 555                         continue;
 556                 if (atomic_inc_not_zero(&state->count))
 557                         return state;
 558         }
 559         return NULL;
 560 }
 561
 562 static void
 563 nfs4_free_open_state(struct nfs4_state *state)
 564 {
 565         kfree(state);
 566 }
 567
 568 struct nfs4_state *
 569 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 570 {
 571         struct nfs4_state *state, *new;
 572         struct nfs_inode *nfsi = NFS_I(inode);
 573
 574         spin_lock(&inode->i_lock);
 575         state = __nfs4_find_state_byowner(inode, owner);
 576         spin_unlock(&inode->i_lock);
 577         if (state)
 578                 goto out;
 579         new = nfs4_alloc_open_state();
 580         spin_lock(&owner->so_lock);
 581         spin_lock(&inode->i_lock);
 582         state = __nfs4_find_state_byowner(inode, owner);
 583         if (state == NULL && new != NULL) {
 584                 state = new;
 585                 state->owner = owner;
 586                 atomic_inc(&owner->so_count);
 587                 list_add(&state->inode_states, &nfsi->open_states);
 588                 state->inode = igrab(inode);
 589                 spin_unlock(&inode->i_lock);
 590                 /* Note: The reclaim code dictates that we add stateless
 591                  * and read-only stateids to the end of the list */
 592                 list_add_tail(&state->open_states, &owner->so_states);
 593                 spin_unlock(&owner->so_lock);
 594         } else {
 595                 spin_unlock(&inode->i_lock);
 596                 spin_unlock(&owner->so_lock);
 597                 if (new)
 598                         nfs4_free_open_state(new);
 599         }
 600 out:
 601         return state;
 602 }
 603
 604 void nfs4_put_open_state(struct nfs4_state *state)
 605 {
 606         struct inode *inode = state->inode;
 607         struct nfs4_state_owner *owner = state->owner;
 608
 609         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 610                 return;
 611         spin_lock(&inode->i_lock);
 612         list_del(&state->inode_states);
 613         list_del(&state->open_states);
 614         spin_unlock(&inode->i_lock);
 615         spin_unlock(&owner->so_lock);
 616         iput(inode);
 617         nfs4_free_open_state(state);
 618         nfs4_put_state_owner(owner);
 619 }
 620
 621 /*
 622  * Close the current file.
 623  */
 624 static void __nfs4_close(struct path *path, struct nfs4_state *state,
 625                 fmode_t fmode, gfp_t gfp_mask, int wait)
 626 {
 627         struct nfs4_state_owner *owner = state->owner;
 628         int call_close = 0;
 629         fmode_t newstate;
 630
 631         atomic_inc(&owner->so_count);
 632         /* Protect against nfs4_find_state() */
 633         spin_lock(&owner->so_lock);
 634         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 635                 case FMODE_READ:
 636                         state->n_rdonly--;
 637                         break;
 638                 case FMODE_WRITE:
 639                         state->n_wronly--;
 640                         break;
 641                 case FMODE_READ|FMODE_WRITE:
 642                         state->n_rdwr--;
 643         }
 644         newstate = FMODE_READ|FMODE_WRITE;
 645         if (state->n_rdwr == 0) {
 646                 if (state->n_rdonly == 0) {
 647                         newstate &= ~FMODE_READ;
 648                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 649                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 650                 }
 651                 if (state->n_wronly == 0) {
 652                         newstate &= ~FMODE_WRITE;
 653                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 654                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 655                 }
 656                 if (newstate == 0)
 657                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 658         }
 659         nfs4_state_set_mode_locked(state, newstate);
 660         spin_unlock(&owner->so_lock);
 661
 662         if (!call_close) {
 663                 nfs4_put_open_state(state);
 664                 nfs4_put_state_owner(owner);
 665         } else {
 666                 bool roc = pnfs_roc(state->inode);
 667
 668                 nfs4_do_close(path, state, gfp_mask, wait, roc);
 669         }
 670 }
 671
 672 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 673 {
 674         __nfs4_close(path, state, fmode, GFP_NOFS, 0);
 675 }
 676
 677 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 678 {
 679         __nfs4_close(path, state, fmode, GFP_KERNEL, 1);
 680 }
 681
 682 /*
 683  * Search the state->lock_states for an existing lock_owner
 684  * that is compatible with current->files
 685  */
 686 static struct nfs4_lock_state *
 687 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 688 {
 689         struct nfs4_lock_state *pos;
 690         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 691                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 692                         continue;
 693                 switch (pos->ls_owner.lo_type) {
 694                 case NFS4_POSIX_LOCK_TYPE:
 695                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 696                                 continue;
 697                         break;
 698                 case NFS4_FLOCK_LOCK_TYPE:
 699                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 700                                 continue;
 701                 }
 702                 atomic_inc(&pos->ls_count);
 703                 return pos;
 704         }
 705         return NULL;
 706 }
 707
 708 /*
 709  * Return a compatible lock_state. If no initialized lock_state structure
 710  * exists, return an uninitialized one.
 711  *
 712  */
 713 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)
 714 {
 715         struct nfs4_lock_state *lsp;
 716         struct nfs_server *server = state->owner->so_server;
 717         struct nfs_client *clp = server->nfs_client;
 718
 719         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 720         if (lsp == NULL)
 721                 return NULL;
 722         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 723         spin_lock_init(&lsp->ls_sequence.lock);
 724         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 725         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 726         atomic_set(&lsp->ls_count, 1);
 727         lsp->ls_state = state;
 728         lsp->ls_owner.lo_type = type;
 729         switch (lsp->ls_owner.lo_type) {
 730         case NFS4_FLOCK_LOCK_TYPE:
 731                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 732                 break;
 733         case NFS4_POSIX_LOCK_TYPE:
 734                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 735                 break;
 736         default:
 737                 kfree(lsp);
 738                 return NULL;
 739         }
 740         spin_lock(&clp->cl_lock);
 741         nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
 742         spin_unlock(&clp->cl_lock);
 743         INIT_LIST_HEAD(&lsp->ls_locks);
 744         return lsp;
 745 }
 746
 747 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 748 {
 749         struct nfs_server *server = lsp->ls_state->owner->so_server;
 750         struct nfs_client *clp = server->nfs_client;
 751
 752         spin_lock(&clp->cl_lock);
 753         nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
 754         spin_unlock(&clp->cl_lock);
 755         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 756         kfree(lsp);
 757 }
 758
 759 /*
 760  * Return a compatible lock_state. If no initialized lock_state structure
 761  * exists, return an uninitialized one.
 762  *
 763  */
 764 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 765 {
 766         struct nfs4_lock_state *lsp, *new = NULL;
 767
 768         for(;;) {
 769                 spin_lock(&state->state_lock);
 770                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 771                 if (lsp != NULL)
 772                         break;
 773                 if (new != NULL) {
 774                         list_add(&new->ls_locks, &state->lock_states);
 775                         set_bit(LK_STATE_IN_USE, &state->flags);
 776                         lsp = new;
 777                         new = NULL;
 778                         break;
 779                 }
 780                 spin_unlock(&state->state_lock);
 781                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 782                 if (new == NULL)
 783                         return NULL;
 784         }
 785         spin_unlock(&state->state_lock);
 786         if (new != NULL)
 787                 nfs4_free_lock_state(new);
 788         return lsp;
 789 }
 790
 791 /*
 792  * Release reference to lock_state, and free it if we see that
 793  * it is no longer in use
 794  */
 795 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 796 {
 797         struct nfs4_state *state;
 798
 799         if (lsp == NULL)
 800                 return;
 801         state = lsp->ls_state;
 802         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 803                 return;
 804         list_del(&lsp->ls_locks);
 805         if (list_empty(&state->lock_states))
 806                 clear_bit(LK_STATE_IN_USE, &state->flags);
 807         spin_unlock(&state->state_lock);
 808         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 809                 nfs4_release_lockowner(lsp);
 810         nfs4_free_lock_state(lsp);
 811 }
 812
 813 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 814 {
 815         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 816
 817         dst->fl_u.nfs4_fl.owner = lsp;
 818         atomic_inc(&lsp->ls_count);
 819 }
 820
 821 static void nfs4_fl_release_lock(struct file_lock *fl)
 822 {
 823         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 824 }
 825
 826 static const struct file_lock_operations nfs4_fl_lock_ops = {
 827         .fl_copy_lock = nfs4_fl_copy_lock,
 828         .fl_release_private = nfs4_fl_release_lock,
 829 };
 830
 831 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 832 {
 833         struct nfs4_lock_state *lsp;
 834
 835         if (fl->fl_ops != NULL)
 836                 return 0;
 837         if (fl->fl_flags & FL_POSIX)
 838                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 839         else if (fl->fl_flags & FL_FLOCK)
 840                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 841         else
 842                 return -EINVAL;
 843         if (lsp == NULL)
 844                 return -ENOMEM;
 845         fl->fl_u.nfs4_fl.owner = lsp;
 846         fl->fl_ops = &nfs4_fl_lock_ops;
 847         return 0;
 848 }
 849
 850 /*
 851  * Byte-range lock aware utility to initialize the stateid of read/write
 852  * requests.
 853  */
 854 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 855 {
 856         struct nfs4_lock_state *lsp;
 857         int seq;
 858
 859         do {
 860                 seq = read_seqbegin(&state->seqlock);
 861                 memcpy(dst, &state->stateid, sizeof(*dst));
 862         } while (read_seqretry(&state->seqlock, seq));
 863         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 864                 return;
 865
 866         spin_lock(&state->state_lock);
 867         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 868         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 869                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 870         spin_unlock(&state->state_lock);
 871         nfs4_put_lock_state(lsp);
 872 }
 873
 874 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 875 {
 876         struct nfs_seqid *new;
 877
 878         new = kmalloc(sizeof(*new), gfp_mask);
 879         if (new != NULL) {
 880                 new->sequence = counter;
 881                 INIT_LIST_HEAD(&new->list);
 882         }
 883         return new;
 884 }
 885
 886 void nfs_release_seqid(struct nfs_seqid *seqid)
 887 {
 888         if (!list_empty(&seqid->list)) {
 889                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 890
 891                 spin_lock(&sequence->lock);
 892                 list_del_init(&seqid->list);
 893                 spin_unlock(&sequence->lock);
 894                 rpc_wake_up(&sequence->wait);
 895         }
 896 }
 897
 898 void nfs_free_seqid(struct nfs_seqid *seqid)
 899 {
 900         nfs_release_seqid(seqid);
 901         kfree(seqid);
 902 }
 903
 904 /*
 905  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 906  * failed with a seqid incrementing error -
 907  * see comments nfs_fs.h:seqid_mutating_error()
 908  */
 909 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 910 {
 911         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 912         switch (status) {
 913                 case 0:
 914                         break;
 915                 case -NFS4ERR_BAD_SEQID:
 916                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 917                                 return;
 918                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 919                                         " sequence-id error on an"
 920                                         " unconfirmed sequence %p!\n",
 921                                         seqid->sequence);
 922                 case -NFS4ERR_STALE_CLIENTID:
 923                 case -NFS4ERR_STALE_STATEID:
 924                 case -NFS4ERR_BAD_STATEID:
 925                 case -NFS4ERR_BADXDR:
 926                 case -NFS4ERR_RESOURCE:
 927                 case -NFS4ERR_NOFILEHANDLE:
 928                         /* Non-seqid mutating errors */
 929                         return;
 930         };
 931         /*
 932          * Note: no locking needed as we are guaranteed to be first
 933          * on the sequence list
 934          */
 935         seqid->sequence->counter++;
 936 }
 937
 938 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 939 {
 940         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 941                                         struct nfs4_state_owner, so_seqid);
 942         struct nfs_server *server = sp->so_server;
 943
 944         if (status == -NFS4ERR_BAD_SEQID)
 945                 nfs4_drop_state_owner(sp);
 946         if (!nfs4_has_session(server->nfs_client))
 947                 nfs_increment_seqid(status, seqid);
 948 }
 949
 950 /*
 951  * Increment the seqid if the LOCK/LOCKU succeeded, or
 952  * failed with a seqid incrementing error -
 953  * see comments nfs_fs.h:seqid_mutating_error()
 954  */
 955 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 956 {
 957         nfs_increment_seqid(status, seqid);
 958 }
 959
 960 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 961 {
 962         struct rpc_sequence *sequence = seqid->sequence->sequence;
 963         int status = 0;
 964
 965         spin_lock(&sequence->lock);
 966         if (list_empty(&seqid->list))
 967                 list_add_tail(&seqid->list, &sequence->list);
 968         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 969                 goto unlock;
 970         rpc_sleep_on(&sequence->wait, task, NULL);
 971         status = -EAGAIN;
 972 unlock:
 973         spin_unlock(&sequence->lock);
 974         return status;
 975 }
 976
 977 static int nfs4_run_state_manager(void *);
 978
 979 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 980 {
 981         smp_mb__before_clear_bit();
 982         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 983         smp_mb__after_clear_bit();
 984         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 985         rpc_wake_up(&clp->cl_rpcwaitq);
 986 }
 987
 988 /*
 989  * Schedule the nfs_client asynchronous state management routine
 990  */
 991 void nfs4_schedule_state_manager(struct nfs_client *clp)
 992 {
 993         struct task_struct *task;
 994
 995         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
 996                 return;
 997         __module_get(THIS_MODULE);
 998         atomic_inc(&clp->cl_count);
 999         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1000                                 rpc_peeraddr2str(clp->cl_rpcclient,
1001                                                         RPC_DISPLAY_ADDR));
1002         if (!IS_ERR(task))
1003                 return;
1004         nfs4_clear_state_manager_bit(clp);
1005         nfs_put_client(clp);
1006         module_put(THIS_MODULE);
1007 }
1008
1009 /*
1010  * Schedule a state recovery attempt
1011  */
1012 void nfs4_schedule_state_recovery(struct nfs_client *clp)
1013 {
1014         if (!clp)
1015                 return;
1016         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1017                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1018         nfs4_schedule_state_manager(clp);
1019 }
1020
1021 int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1022 {
1023
1024         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1025         /* Don't recover state that expired before the reboot */
1026         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1027                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1028                 return 0;
1029         }
1030         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1031         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1032         return 1;
1033 }
1034
1035 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1036 {
1037         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1038         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1039         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1040         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1041         return 1;
1042 }
1043
1044 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1045 {
1046         struct inode *inode = state->inode;
1047         struct nfs_inode *nfsi = NFS_I(inode);
1048         struct file_lock *fl;
1049         int status = 0;
1050
1051         if (inode->i_flock == NULL)
1052                 return 0;
1053
1054         /* Guard against delegation returns and new lock/unlock calls */
1055         down_write(&nfsi->rwsem);
1056         /* Protect inode->i_flock using the BKL */
1057         lock_flocks();
1058         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1059                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1060                         continue;
1061                 if (nfs_file_open_context(fl->fl_file)->state != state)
1062                         continue;
1063                 unlock_flocks();
1064                 status = ops->recover_lock(state, fl);
1065                 switch (status) {
1066                         case 0:
1067                                 break;
1068                         case -ESTALE:
1069                         case -NFS4ERR_ADMIN_REVOKED:
1070                         case -NFS4ERR_STALE_STATEID:
1071                         case -NFS4ERR_BAD_STATEID:
1072                         case -NFS4ERR_EXPIRED:
1073                         case -NFS4ERR_NO_GRACE:
1074                         case -NFS4ERR_STALE_CLIENTID:
1075                         case -NFS4ERR_BADSESSION:
1076                         case -NFS4ERR_BADSLOT:
1077                         case -NFS4ERR_BAD_HIGH_SLOT:
1078                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1079                                 goto out;
1080                         default:
1081                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1082                                                 __func__, status);
1083                         case -ENOMEM:
1084                         case -NFS4ERR_DENIED:
1085                         case -NFS4ERR_RECLAIM_BAD:
1086                         case -NFS4ERR_RECLAIM_CONFLICT:
1087                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1088                                 status = 0;
1089                 }
1090                 lock_flocks();
1091         }
1092         unlock_flocks();
1093 out:
1094         up_write(&nfsi->rwsem);
1095         return status;
1096 }
1097
1098 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1099 {
1100         struct nfs4_state *state;
1101         struct nfs4_lock_state *lock;
1102         int status = 0;
1103
1104         /* Note: we rely on the sp->so_states list being ordered
1105          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1106          * states first.
1107          * This is needed to ensure that the server won't give us any
1108          * read delegations that we have to return if, say, we are
1109          * recovering after a network partition or a reboot from a
1110          * server that doesn't support a grace period.
1111          */
1112 restart:
1113         spin_lock(&sp->so_lock);
1114         list_for_each_entry(state, &sp->so_states, open_states) {
1115                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1116                         continue;
1117                 if (state->state == 0)
1118                         continue;
1119                 atomic_inc(&state->count);
1120                 spin_unlock(&sp->so_lock);
1121                 status = ops->recover_open(sp, state);
1122                 if (status >= 0) {
1123                         status = nfs4_reclaim_locks(state, ops);
1124                         if (status >= 0) {
1125                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1126                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1127                                                 printk("%s: Lock reclaim failed!\n",
1128                                                         __func__);
1129                                 }
1130                                 nfs4_put_open_state(state);
1131                                 goto restart;
1132                         }
1133                 }
1134                 switch (status) {
1135                         default:
1136                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1137                                                 __func__, status);
1138                         case -ENOENT:
1139                         case -ENOMEM:
1140                         case -ESTALE:
1141                                 /*
1142                                  * Open state on this file cannot be recovered
1143                                  * All we can do is revert to using the zero stateid.
1144                                  */
1145                                 memset(state->stateid.data, 0,
1146                                         sizeof(state->stateid.data));
1147                                 /* Mark the file as being 'closed' */
1148                                 state->state = 0;
1149                                 break;
1150                         case -EKEYEXPIRED:
1151                                 /*
1152                                  * User RPCSEC_GSS context has expired.
1153                                  * We cannot recover this stateid now, so
1154                                  * skip it and allow recovery thread to
1155                                  * proceed.
1156                                  */
1157                                 break;
1158                         case -NFS4ERR_ADMIN_REVOKED:
1159                         case -NFS4ERR_STALE_STATEID:
1160                         case -NFS4ERR_BAD_STATEID:
1161                         case -NFS4ERR_RECLAIM_BAD:
1162                         case -NFS4ERR_RECLAIM_CONFLICT:
1163                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1164                                 break;
1165                         case -NFS4ERR_EXPIRED:
1166                         case -NFS4ERR_NO_GRACE:
1167                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1168                         case -NFS4ERR_STALE_CLIENTID:
1169                         case -NFS4ERR_BADSESSION:
1170                         case -NFS4ERR_BADSLOT:
1171                         case -NFS4ERR_BAD_HIGH_SLOT:
1172                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1173                                 goto out_err;
1174                 }
1175                 nfs4_put_open_state(state);
1176                 goto restart;
1177         }
1178         spin_unlock(&sp->so_lock);
1179         return 0;
1180 out_err:
1181         nfs4_put_open_state(state);
1182         return status;
1183 }
1184
1185 static void nfs4_clear_open_state(struct nfs4_state *state)
1186 {
1187         struct nfs4_lock_state *lock;
1188
1189         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1190         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1191         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1192         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1193         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1194                 lock->ls_seqid.flags = 0;
1195                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1196         }
1197 }
1198
1199 static void nfs4_reset_seqids(struct nfs_server *server,
1200         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1201 {
1202         struct nfs_client *clp = server->nfs_client;
1203         struct nfs4_state_owner *sp;
1204         struct rb_node *pos;
1205         struct nfs4_state *state;
1206
1207         spin_lock(&clp->cl_lock);
1208         for (pos = rb_first(&server->state_owners);
1209              pos != NULL;
1210              pos = rb_next(pos)) {
1211                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1212                 sp->so_seqid.flags = 0;
1213                 spin_lock(&sp->so_lock);
1214                 list_for_each_entry(state, &sp->so_states, open_states) {
1215                         if (mark_reclaim(clp, state))
1216                                 nfs4_clear_open_state(state);
1217                 }
1218                 spin_unlock(&sp->so_lock);
1219         }
1220         spin_unlock(&clp->cl_lock);
1221 }
1222
1223 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1224         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1225 {
1226         struct nfs_server *server;
1227
1228         rcu_read_lock();
1229         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1230                 nfs4_reset_seqids(server, mark_reclaim);
1231         rcu_read_unlock();
1232 }
1233
1234 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1235 {
1236         /* Mark all delegations for reclaim */
1237         nfs_delegation_mark_reclaim(clp);
1238         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1239 }
1240
1241 static void nfs4_reclaim_complete(struct nfs_client *clp,
1242                                  const struct nfs4_state_recovery_ops *ops)
1243 {
1244         /* Notify the server we're done reclaiming our state */
1245         if (ops->reclaim_complete)
1246                 (void)ops->reclaim_complete(clp);
1247 }
1248
1249 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1250 {
1251         struct nfs_client *clp = server->nfs_client;
1252         struct nfs4_state_owner *sp;
1253         struct rb_node *pos;
1254         struct nfs4_state *state;
1255
1256         spin_lock(&clp->cl_lock);
1257         for (pos = rb_first(&server->state_owners);
1258              pos != NULL;
1259              pos = rb_next(pos)) {
1260                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1261                 spin_lock(&sp->so_lock);
1262                 list_for_each_entry(state, &sp->so_states, open_states) {
1263                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1264                                                 &state->flags))
1265                                 continue;
1266                         nfs4_state_mark_reclaim_nograce(clp, state);
1267                 }
1268                 spin_unlock(&sp->so_lock);
1269         }
1270         spin_unlock(&clp->cl_lock);
1271 }
1272
1273 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1274 {
1275         struct nfs_server *server;
1276
1277         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1278                 return 0;
1279
1280         rcu_read_lock();
1281         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1282                 nfs4_clear_reclaim_server(server);
1283         rcu_read_unlock();
1284
1285         nfs_delegation_reap_unclaimed(clp);
1286         return 1;
1287 }
1288
1289 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1290 {
1291         if (!nfs4_state_clear_reclaim_reboot(clp))
1292                 return;
1293         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1294 }
1295
1296 static void nfs_delegation_clear_all(struct nfs_client *clp)
1297 {
1298         nfs_delegation_mark_reclaim(clp);
1299         nfs_delegation_reap_unclaimed(clp);
1300 }
1301
1302 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1303 {
1304         nfs_delegation_clear_all(clp);
1305         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1306 }
1307
1308 static void nfs4_warn_keyexpired(const char *s)
1309 {
1310         printk_ratelimited(KERN_WARNING "Error: state manager"
1311                         " encountered RPCSEC_GSS session"
1312                         " expired against NFSv4 server %s.\n",
1313                         s);
1314 }
1315
1316 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1317 {
1318         switch (error) {
1319                 case -NFS4ERR_CB_PATH_DOWN:
1320                         nfs_handle_cb_pathdown(clp);
1321                         return 0;
1322                 case -NFS4ERR_NO_GRACE:
1323                         nfs4_state_end_reclaim_reboot(clp);
1324                         return 0;
1325                 case -NFS4ERR_STALE_CLIENTID:
1326                 case -NFS4ERR_LEASE_MOVED:
1327                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1328                         nfs4_state_clear_reclaim_reboot(clp);
1329                         nfs4_state_start_reclaim_reboot(clp);
1330                         break;
1331                 case -NFS4ERR_EXPIRED:
1332                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1333                         nfs4_state_start_reclaim_nograce(clp);
1334                         break;
1335                 case -NFS4ERR_BADSESSION:
1336                 case -NFS4ERR_BADSLOT:
1337                 case -NFS4ERR_BAD_HIGH_SLOT:
1338                 case -NFS4ERR_DEADSESSION:
1339                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1340                 case -NFS4ERR_SEQ_FALSE_RETRY:
1341                 case -NFS4ERR_SEQ_MISORDERED:
1342                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1343                         /* Zero session reset errors */
1344                         return 0;
1345                 case -EKEYEXPIRED:
1346                         /* Nothing we can do */
1347                         nfs4_warn_keyexpired(clp->cl_hostname);
1348                         return 0;
1349         }
1350         return error;
1351 }
1352
1353 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1354 {
1355         struct nfs4_state_owner *sp;
1356         struct nfs_server *server;
1357         struct rb_node *pos;
1358         int status = 0;
1359
1360 restart:
1361         rcu_read_lock();
1362         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1363                 spin_lock(&clp->cl_lock);
1364                 for (pos = rb_first(&server->state_owners);
1365                      pos != NULL;
1366                      pos = rb_next(pos)) {
1367                         sp = rb_entry(pos,
1368                                 struct nfs4_state_owner, so_server_node);
1369                         if (!test_and_clear_bit(ops->owner_flag_bit,
1370                                                         &sp->so_flags))
1371                                 continue;
1372                         atomic_inc(&sp->so_count);
1373                         spin_unlock(&clp->cl_lock);
1374                         rcu_read_unlock();
1375
1376                         status = nfs4_reclaim_open_state(sp, ops);
1377                         if (status < 0) {
1378                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1379                                 nfs4_put_state_owner(sp);
1380                                 return nfs4_recovery_handle_error(clp, status);
1381                         }
1382
1383                         nfs4_put_state_owner(sp);
1384                         goto restart;
1385                 }
1386                 spin_unlock(&clp->cl_lock);
1387         }
1388         rcu_read_unlock();
1389         return status;
1390 }
1391
1392 static int nfs4_check_lease(struct nfs_client *clp)
1393 {
1394         struct rpc_cred *cred;
1395         const struct nfs4_state_maintenance_ops *ops =
1396                 clp->cl_mvops->state_renewal_ops;
1397         int status = -NFS4ERR_EXPIRED;
1398
1399         /* Is the client already known to have an expired lease? */
1400         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1401                 return 0;
1402         spin_lock(&clp->cl_lock);
1403         cred = ops->get_state_renewal_cred_locked(clp);
1404         spin_unlock(&clp->cl_lock);
1405         if (cred == NULL) {
1406                 cred = nfs4_get_setclientid_cred(clp);
1407                 if (cred == NULL)
1408                         goto out;
1409         }
1410         status = ops->renew_lease(clp, cred);
1411         put_rpccred(cred);
1412 out:
1413         return nfs4_recovery_handle_error(clp, status);
1414 }
1415
1416 static int nfs4_reclaim_lease(struct nfs_client *clp)
1417 {
1418         struct rpc_cred *cred;
1419         const struct nfs4_state_recovery_ops *ops =
1420                 clp->cl_mvops->reboot_recovery_ops;
1421         int status = -ENOENT;
1422
1423         cred = ops->get_clid_cred(clp);
1424         if (cred != NULL) {
1425                 status = ops->establish_clid(clp, cred);
1426                 put_rpccred(cred);
1427                 /* Handle case where the user hasn't set up machine creds */
1428                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1429                         nfs4_clear_machine_cred(clp);
1430                         status = -EAGAIN;
1431                 }
1432                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1433                         status = -EPROTONOSUPPORT;
1434         }
1435         return status;
1436 }
1437
1438 #ifdef CONFIG_NFS_V4_1
1439 void nfs41_handle_recall_slot(struct nfs_client *clp)
1440 {
1441         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1442         nfs4_schedule_state_recovery(clp);
1443 }
1444
1445 static void nfs4_reset_all_state(struct nfs_client *clp)
1446 {
1447         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1448                 clp->cl_boot_time = CURRENT_TIME;
1449                 nfs4_state_start_reclaim_nograce(clp);
1450                 nfs4_schedule_state_recovery(clp);
1451         }
1452 }
1453
1454 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1455 {
1456         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1457                 nfs4_state_start_reclaim_reboot(clp);
1458                 nfs4_schedule_state_recovery(clp);
1459         }
1460 }
1461
1462 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1463 {
1464         /* Temporary */
1465         nfs4_reset_all_state(clp);
1466 }
1467
1468 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1469 {
1470         /* This will need to handle layouts too */
1471         nfs_expire_all_delegations(clp);
1472 }
1473
1474 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1475 {
1476         nfs_expire_all_delegations(clp);
1477         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1478                 nfs4_schedule_state_recovery(clp);
1479 }
1480
1481 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1482 {
1483         if (!flags)
1484                 return;
1485         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1486                 nfs41_handle_server_reboot(clp);
1487         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1488                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1489                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1490                             SEQ4_STATUS_LEASE_MOVED))
1491                 nfs41_handle_state_revoked(clp);
1492         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1493                 nfs41_handle_recallable_state_revoked(clp);
1494         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1495                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1496                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1497                 nfs41_handle_cb_path_down(clp);
1498 }
1499
1500 static int nfs4_reset_session(struct nfs_client *clp)
1501 {
1502         int status;
1503
1504         nfs4_begin_drain_session(clp);
1505         status = nfs4_proc_destroy_session(clp->cl_session);
1506         if (status && status != -NFS4ERR_BADSESSION &&
1507             status != -NFS4ERR_DEADSESSION) {
1508                 status = nfs4_recovery_handle_error(clp, status);
1509                 goto out;
1510         }
1511
1512         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1513         status = nfs4_proc_create_session(clp);
1514         if (status) {
1515                 status = nfs4_recovery_handle_error(clp, status);
1516                 goto out;
1517         }
1518         /* create_session negotiated new slot table */
1519         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1520
1521          /* Let the state manager reestablish state */
1522         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1523                 nfs41_setup_state_renewal(clp);
1524 out:
1525         return status;
1526 }
1527
1528 static int nfs4_recall_slot(struct nfs_client *clp)
1529 {
1530         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1531         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1532         struct nfs4_slot *new, *old;
1533         int i;
1534
1535         nfs4_begin_drain_session(clp);
1536         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1537                       GFP_NOFS);
1538         if (!new)
1539                 return -ENOMEM;
1540
1541         spin_lock(&fc_tbl->slot_tbl_lock);
1542         for (i = 0; i < fc_tbl->target_max_slots; i++)
1543                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1544         old = fc_tbl->slots;
1545         fc_tbl->slots = new;
1546         fc_tbl->max_slots = fc_tbl->target_max_slots;
1547         fc_tbl->target_max_slots = 0;
1548         fc_attrs->max_reqs = fc_tbl->max_slots;
1549         spin_unlock(&fc_tbl->slot_tbl_lock);
1550
1551         kfree(old);
1552         nfs4_end_drain_session(clp);
1553         return 0;
1554 }
1555
1556 #else /* CONFIG_NFS_V4_1 */
1557 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1558 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1559 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1560 #endif /* CONFIG_NFS_V4_1 */
1561
1562 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1563  * on EXCHANGE_ID for v4.1
1564  */
1565 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1566 {
1567         if (nfs4_has_session(clp)) {
1568                 switch (status) {
1569                 case -NFS4ERR_DELAY:
1570                 case -NFS4ERR_CLID_INUSE:
1571                 case -EAGAIN:
1572                         break;
1573
1574                 case -EKEYEXPIRED:
1575                         nfs4_warn_keyexpired(clp->cl_hostname);
1576                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1577                                          * in nfs4_exchange_id */
1578                 default:
1579                         return;
1580                 }
1581         }
1582         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1583 }
1584
1585 static void nfs4_state_manager(struct nfs_client *clp)
1586 {
1587         int status = 0;
1588
1589         /* Ensure exclusive access to NFSv4 state */
1590         for(;;) {
1591                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1592                         /* We're going to have to re-establish a clientid */
1593                         status = nfs4_reclaim_lease(clp);
1594                         if (status) {
1595                                 nfs4_set_lease_expired(clp, status);
1596                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1597                                                         &clp->cl_state))
1598                                         continue;
1599                                 if (clp->cl_cons_state ==
1600                                                         NFS_CS_SESSION_INITING)
1601                                         nfs_mark_client_ready(clp, status);
1602                                 goto out_error;
1603                         }
1604                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1605                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1606                         pnfs_destroy_all_layouts(clp);
1607                 }
1608
1609                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1610                         status = nfs4_check_lease(clp);
1611                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1612                                 continue;
1613                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1614                                 goto out_error;
1615                 }
1616
1617                 /* Initialize or reset the session */
1618                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1619                    && nfs4_has_session(clp)) {
1620                         status = nfs4_reset_session(clp);
1621                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1622                                 continue;
1623                         if (status < 0)
1624                                 goto out_error;
1625                 }
1626
1627                 /* First recover reboot state... */
1628                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1629                         status = nfs4_do_reclaim(clp,
1630                                 clp->cl_mvops->reboot_recovery_ops);
1631                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1632                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1633                                 continue;
1634                         nfs4_state_end_reclaim_reboot(clp);
1635                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1636                                 continue;
1637                         if (status < 0)
1638                                 goto out_error;
1639                 }
1640
1641                 /* Now recover expired state... */
1642                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1643                         status = nfs4_do_reclaim(clp,
1644                                 clp->cl_mvops->nograce_recovery_ops);
1645                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1646                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1647                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1648                                 continue;
1649                         if (status < 0)
1650                                 goto out_error;
1651                 }
1652
1653                 nfs4_end_drain_session(clp);
1654                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1655                         nfs_client_return_marked_delegations(clp);
1656                         continue;
1657                 }
1658                 /* Recall session slots */
1659                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1660                    && nfs4_has_session(clp)) {
1661                         status = nfs4_recall_slot(clp);
1662                         if (status < 0)
1663                                 goto out_error;
1664                         continue;
1665                 }
1666
1667
1668                 nfs4_clear_state_manager_bit(clp);
1669                 /* Did we race with an attempt to give us more work? */
1670                 if (clp->cl_state == 0)
1671                         break;
1672                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1673                         break;
1674         }
1675         return;
1676 out_error:
1677         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1678                         " with error %d\n", clp->cl_hostname, -status);
1679         nfs4_end_drain_session(clp);
1680         nfs4_clear_state_manager_bit(clp);
1681 }
1682
1683 static int nfs4_run_state_manager(void *ptr)
1684 {
1685         struct nfs_client *clp = ptr;
1686
1687         allow_signal(SIGKILL);
1688         nfs4_state_manager(clp);
1689         nfs_put_client(clp);
1690         module_put_and_exit(0);
1691         return 0;
1692 }
1693
1694 /*
1695  * Local variables:
1696  *  c-basic-offset: 8
1697  * End:
1698  */