fs/nfs/delegation.c

   1 /*
   2  * linux/fs/nfs/delegation.c
   3  *
   4  * Copyright (C) 2004 Trond Myklebust
   5  *
   6  * NFS file delegation management
   7  *
   8  */
   9 #include <linux/completion.h>
  10 #include <linux/kthread.h>
  11 #include <linux/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/iversion.h>
  16
  17 #include <linux/nfs4.h>
  18 #include <linux/nfs_fs.h>
  19 #include <linux/nfs_xdr.h>
  20
  21 #include "nfs4_fs.h"
  22 #include "nfs4session.h"
  23 #include "delegation.h"
  24 #include "internal.h"
  25 #include "nfs4trace.h"
  26
  27 static void nfs_free_delegation(struct nfs_delegation *delegation)
  28 {
  29         if (delegation->cred) {
  30                 put_rpccred(delegation->cred);
  31                 delegation->cred = NULL;
  32         }
  33         kfree_rcu(delegation, rcu);
  34 }
  35
  36 /**
  37  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
  38  * @delegation: delegation to process
  39  *
  40  */
  41 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
  42 {
  43         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
  44 }
  45
  46 static bool
  47 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
  48                 fmode_t flags)
  49 {
  50         if (delegation != NULL && (delegation->type & flags) == flags &&
  51             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
  52             !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
  53                 return true;
  54         return false;
  55 }
  56
  57 static int
  58 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
  59 {
  60         struct nfs_delegation *delegation;
  61         int ret = 0;
  62
  63         flags &= FMODE_READ|FMODE_WRITE;
  64         rcu_read_lock();
  65         delegation = rcu_dereference(NFS_I(inode)->delegation);
  66         if (nfs4_is_valid_delegation(delegation, flags)) {
  67                 if (mark)
  68                         nfs_mark_delegation_referenced(delegation);
  69                 ret = 1;
  70         }
  71         rcu_read_unlock();
  72         return ret;
  73 }
  74 /**
  75  * nfs_have_delegation - check if inode has a delegation, mark it
  76  * NFS_DELEGATION_REFERENCED if there is one.
  77  * @inode: inode to check
  78  * @flags: delegation types to check for
  79  *
  80  * Returns one if inode has the indicated delegation, otherwise zero.
  81  */
  82 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
  83 {
  84         return nfs4_do_check_delegation(inode, flags, true);
  85 }
  86
  87 /*
  88  * nfs4_check_delegation - check if inode has a delegation, do not mark
  89  * NFS_DELEGATION_REFERENCED if it has one.
  90  */
  91 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
  92 {
  93         return nfs4_do_check_delegation(inode, flags, false);
  94 }
  95
  96 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
  97 {
  98         struct inode *inode = state->inode;
  99         struct file_lock *fl;
 100         struct file_lock_context *flctx = inode->i_flctx;
 101         struct list_head *list;
 102         int status = 0;
 103
 104         if (flctx == NULL)
 105                 goto out;
 106
 107         list = &flctx->flc_posix;
 108         spin_lock(&flctx->flc_lock);
 109 restart:
 110         list_for_each_entry(fl, list, fl_list) {
 111                 if (nfs_file_open_context(fl->fl_file)->state != state)
 112                         continue;
 113                 spin_unlock(&flctx->flc_lock);
 114                 status = nfs4_lock_delegation_recall(fl, state, stateid);
 115                 if (status < 0)
 116                         goto out;
 117                 spin_lock(&flctx->flc_lock);
 118         }
 119         if (list == &flctx->flc_posix) {
 120                 list = &flctx->flc_flock;
 121                 goto restart;
 122         }
 123         spin_unlock(&flctx->flc_lock);
 124 out:
 125         return status;
 126 }
 127
 128 static int nfs_delegation_claim_opens(struct inode *inode,
 129                 const nfs4_stateid *stateid, fmode_t type)
 130 {
 131         struct nfs_inode *nfsi = NFS_I(inode);
 132         struct nfs_open_context *ctx;
 133         struct nfs4_state_owner *sp;
 134         struct nfs4_state *state;
 135         unsigned int seq;
 136         int err;
 137
 138 again:
 139         rcu_read_lock();
 140         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
 141                 state = ctx->state;
 142                 if (state == NULL)
 143                         continue;
 144                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 145                         continue;
 146                 if (!nfs4_valid_open_stateid(state))
 147                         continue;
 148                 if (!nfs4_stateid_match(&state->stateid, stateid))
 149                         continue;
 150                 if (!get_nfs_open_context(ctx))
 151                         continue;
 152                 rcu_read_unlock();
 153                 sp = state->owner;
 154                 /* Block nfs4_proc_unlck */
 155                 mutex_lock(&sp->so_delegreturn_mutex);
 156                 seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
 157                 err = nfs4_open_delegation_recall(ctx, state, stateid, type);
 158                 if (!err)
 159                         err = nfs_delegation_claim_locks(state, stateid);
 160                 if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
 161                         err = -EAGAIN;
 162                 mutex_unlock(&sp->so_delegreturn_mutex);
 163                 put_nfs_open_context(ctx);
 164                 if (err != 0)
 165                         return err;
 166                 goto again;
 167         }
 168         rcu_read_unlock();
 169         return 0;
 170 }
 171
 172 /**
 173  * nfs_inode_reclaim_delegation - process a delegation reclaim request
 174  * @inode: inode to process
 175  * @cred: credential to use for request
 176  * @type: delegation type
 177  * @stateid: delegation stateid
 178  * @pagemod_limit: write delegation "space_limit"
 179  *
 180  */
 181 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred,
 182                                   fmode_t type,
 183                                   const nfs4_stateid *stateid,
 184                                   unsigned long pagemod_limit)
 185 {
 186         struct nfs_delegation *delegation;
 187         struct rpc_cred *oldcred = NULL;
 188
 189         rcu_read_lock();
 190         delegation = rcu_dereference(NFS_I(inode)->delegation);
 191         if (delegation != NULL) {
 192                 spin_lock(&delegation->lock);
 193                 if (delegation->inode != NULL) {
 194                         nfs4_stateid_copy(&delegation->stateid, stateid);
 195                         delegation->type = type;
 196                         delegation->pagemod_limit = pagemod_limit;
 197                         oldcred = delegation->cred;
 198                         delegation->cred = get_rpccred(cred);
 199                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
 200                                   &delegation->flags);
 201                         spin_unlock(&delegation->lock);
 202                         rcu_read_unlock();
 203                         put_rpccred(oldcred);
 204                         trace_nfs4_reclaim_delegation(inode, type);
 205                         return;
 206                 }
 207                 /* We appear to have raced with a delegation return. */
 208                 spin_unlock(&delegation->lock);
 209         }
 210         rcu_read_unlock();
 211         nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
 212 }
 213
 214 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 215 {
 216         int res = 0;
 217
 218         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 219                 res = nfs4_proc_delegreturn(inode,
 220                                 delegation->cred,
 221                                 &delegation->stateid,
 222                                 issync);
 223         nfs_free_delegation(delegation);
 224         return res;
 225 }
 226
 227 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 228 {
 229         struct inode *inode = NULL;
 230
 231         spin_lock(&delegation->lock);
 232         if (delegation->inode != NULL)
 233                 inode = igrab(delegation->inode);
 234         spin_unlock(&delegation->lock);
 235         return inode;
 236 }
 237
 238 static struct nfs_delegation *
 239 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
 240 {
 241         struct nfs_delegation *ret = NULL;
 242         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
 243
 244         if (delegation == NULL)
 245                 goto out;
 246         spin_lock(&delegation->lock);
 247         if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 248                 ret = delegation;
 249         spin_unlock(&delegation->lock);
 250 out:
 251         return ret;
 252 }
 253
 254 static struct nfs_delegation *
 255 nfs_start_delegation_return(struct nfs_inode *nfsi)
 256 {
 257         struct nfs_delegation *delegation;
 258
 259         rcu_read_lock();
 260         delegation = nfs_start_delegation_return_locked(nfsi);
 261         rcu_read_unlock();
 262         return delegation;
 263 }
 264
 265 static void
 266 nfs_abort_delegation_return(struct nfs_delegation *delegation,
 267                 struct nfs_client *clp)
 268 {
 269
 270         spin_lock(&delegation->lock);
 271         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 272         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 273         spin_unlock(&delegation->lock);
 274         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 275 }
 276
 277 static struct nfs_delegation *
 278 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
 279                 struct nfs_delegation *delegation,
 280                 struct nfs_client *clp)
 281 {
 282         struct nfs_delegation *deleg_cur =
 283                 rcu_dereference_protected(nfsi->delegation,
 284                                 lockdep_is_held(&clp->cl_lock));
 285
 286         if (deleg_cur == NULL || delegation != deleg_cur)
 287                 return NULL;
 288
 289         spin_lock(&delegation->lock);
 290         set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 291         list_del_rcu(&delegation->super_list);
 292         delegation->inode = NULL;
 293         rcu_assign_pointer(nfsi->delegation, NULL);
 294         spin_unlock(&delegation->lock);
 295         return delegation;
 296 }
 297
 298 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
 299                 struct nfs_delegation *delegation,
 300                 struct nfs_server *server)
 301 {
 302         struct nfs_client *clp = server->nfs_client;
 303
 304         spin_lock(&clp->cl_lock);
 305         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
 306         spin_unlock(&clp->cl_lock);
 307         return delegation;
 308 }
 309
 310 static struct nfs_delegation *
 311 nfs_inode_detach_delegation(struct inode *inode)
 312 {
 313         struct nfs_inode *nfsi = NFS_I(inode);
 314         struct nfs_server *server = NFS_SERVER(inode);
 315         struct nfs_delegation *delegation;
 316
 317         delegation = nfs_start_delegation_return(nfsi);
 318         if (delegation == NULL)
 319                 return NULL;
 320         return nfs_detach_delegation(nfsi, delegation, server);
 321 }
 322
 323 static void
 324 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
 325                 const struct nfs_delegation *update)
 326 {
 327         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
 328                 delegation->stateid.seqid = update->stateid.seqid;
 329                 smp_wmb();
 330                 delegation->type = update->type;
 331         }
 332 }
 333
 334 /**
 335  * nfs_inode_set_delegation - set up a delegation on an inode
 336  * @inode: inode to which delegation applies
 337  * @cred: cred to use for subsequent delegation processing
 338  * @type: delegation type
 339  * @stateid: delegation stateid
 340  * @pagemod_limit: write delegation "space_limit"
 341  *
 342  * Returns zero on success, or a negative errno value.
 343  */
 344 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred,
 345                                   fmode_t type,
 346                                   const nfs4_stateid *stateid,
 347                                   unsigned long pagemod_limit)
 348 {
 349         struct nfs_server *server = NFS_SERVER(inode);
 350         struct nfs_client *clp = server->nfs_client;
 351         struct nfs_inode *nfsi = NFS_I(inode);
 352         struct nfs_delegation *delegation, *old_delegation;
 353         struct nfs_delegation *freeme = NULL;
 354         int status = 0;
 355
 356         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
 357         if (delegation == NULL)
 358                 return -ENOMEM;
 359         nfs4_stateid_copy(&delegation->stateid, stateid);
 360         delegation->type = type;
 361         delegation->pagemod_limit = pagemod_limit;
 362         delegation->change_attr = inode_peek_iversion_raw(inode);
 363         delegation->cred = get_rpccred(cred);
 364         delegation->inode = inode;
 365         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 366         spin_lock_init(&delegation->lock);
 367
 368         spin_lock(&clp->cl_lock);
 369         old_delegation = rcu_dereference_protected(nfsi->delegation,
 370                                         lockdep_is_held(&clp->cl_lock));
 371         if (old_delegation != NULL) {
 372                 /* Is this an update of the existing delegation? */
 373                 if (nfs4_stateid_match_other(&old_delegation->stateid,
 374                                         &delegation->stateid)) {
 375                         nfs_update_inplace_delegation(old_delegation,
 376                                         delegation);
 377                         goto out;
 378                 }
 379                 /*
 380                  * Deal with broken servers that hand out two
 381                  * delegations for the same file.
 382                  * Allow for upgrades to a WRITE delegation, but
 383                  * nothing else.
 384                  */
 385                 dfprintk(FILE, "%s: server %s handed out "
 386                                 "a duplicate delegation!\n",
 387                                 __func__, clp->cl_hostname);
 388                 if (delegation->type == old_delegation->type ||
 389                     !(delegation->type & FMODE_WRITE)) {
 390                         freeme = delegation;
 391                         delegation = NULL;
 392                         goto out;
 393                 }
 394                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
 395                                         &old_delegation->flags))
 396                         goto out;
 397                 freeme = nfs_detach_delegation_locked(nfsi,
 398                                 old_delegation, clp);
 399                 if (freeme == NULL)
 400                         goto out;
 401         }
 402         list_add_tail_rcu(&delegation->super_list, &server->delegations);
 403         rcu_assign_pointer(nfsi->delegation, delegation);
 404         delegation = NULL;
 405
 406         trace_nfs4_set_delegation(inode, type);
 407
 408         spin_lock(&inode->i_lock);
 409         if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME))
 410                 NFS_I(inode)->cache_validity |= NFS_INO_REVAL_FORCED;
 411         spin_unlock(&inode->i_lock);
 412 out:
 413         spin_unlock(&clp->cl_lock);
 414         if (delegation != NULL)
 415                 nfs_free_delegation(delegation);
 416         if (freeme != NULL)
 417                 nfs_do_return_delegation(inode, freeme, 0);
 418         return status;
 419 }
 420
 421 /*
 422  * Basic procedure for returning a delegation to the server
 423  */
 424 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
 425 {
 426         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 427         struct nfs_inode *nfsi = NFS_I(inode);
 428         int err = 0;
 429
 430         if (delegation == NULL)
 431                 return 0;
 432         do {
 433                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 434                         break;
 435                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
 436                                 delegation->type);
 437                 if (!issync || err != -EAGAIN)
 438                         break;
 439                 /*
 440                  * Guard against state recovery
 441                  */
 442                 err = nfs4_wait_clnt_recover(clp);
 443         } while (err == 0);
 444
 445         if (err) {
 446                 nfs_abort_delegation_return(delegation, clp);
 447                 goto out;
 448         }
 449         if (!nfs_detach_delegation(nfsi, delegation, NFS_SERVER(inode)))
 450                 goto out;
 451
 452         err = nfs_do_return_delegation(inode, delegation, issync);
 453 out:
 454         return err;
 455 }
 456
 457 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
 458 {
 459         bool ret = false;
 460
 461         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 462                 goto out;
 463         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 464                 ret = true;
 465         if (test_and_clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) && !ret) {
 466                 struct inode *inode;
 467
 468                 spin_lock(&delegation->lock);
 469                 inode = delegation->inode;
 470                 if (inode && list_empty(&NFS_I(inode)->open_files))
 471                         ret = true;
 472                 spin_unlock(&delegation->lock);
 473         }
 474 out:
 475         return ret;
 476 }
 477
 478 /**
 479  * nfs_client_return_marked_delegations - return previously marked delegations
 480  * @clp: nfs_client to process
 481  *
 482  * Note that this function is designed to be called by the state
 483  * manager thread. For this reason, it cannot flush the dirty data,
 484  * since that could deadlock in case of a state recovery error.
 485  *
 486  * Returns zero on success, or a negative errno value.
 487  */
 488 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 489 {
 490         struct nfs_delegation *delegation;
 491         struct nfs_delegation *prev;
 492         struct nfs_server *server;
 493         struct inode *inode;
 494         struct inode *place_holder = NULL;
 495         struct nfs_delegation *place_holder_deleg = NULL;
 496         int err = 0;
 497
 498 restart:
 499         /*
 500          * To avoid quadratic looping we hold a reference
 501          * to an inode place_holder.  Each time we restart, we
 502          * list nfs_servers from the server of that inode, and
 503          * delegation in the server from the delegations of that
 504          * inode.
 505          * prev is an RCU-protected pointer to a delegation which
 506          * wasn't marked for return and might be a good choice for
 507          * the next place_holder.
 508          */
 509         rcu_read_lock();
 510         prev = NULL;
 511         if (place_holder)
 512                 server = NFS_SERVER(place_holder);
 513         else
 514                 server = list_entry_rcu(clp->cl_superblocks.next,
 515                                         struct nfs_server, client_link);
 516         list_for_each_entry_from_rcu(server, &clp->cl_superblocks, client_link) {
 517                 delegation = NULL;
 518                 if (place_holder && server == NFS_SERVER(place_holder))
 519                         delegation = rcu_dereference(NFS_I(place_holder)->delegation);
 520                 if (!delegation || delegation != place_holder_deleg)
 521                         delegation = list_entry_rcu(server->delegations.next,
 522                                                     struct nfs_delegation, super_list);
 523                 list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
 524                         struct inode *to_put = NULL;
 525
 526                         if (!nfs_delegation_need_return(delegation)) {
 527                                 prev = delegation;
 528                                 continue;
 529                         }
 530                         if (!nfs_sb_active(server->super))
 531                                 break; /* continue in outer loop */
 532
 533                         if (prev) {
 534                                 struct inode *tmp;
 535
 536                                 tmp = nfs_delegation_grab_inode(prev);
 537                                 if (tmp) {
 538                                         to_put = place_holder;
 539                                         place_holder = tmp;
 540                                         place_holder_deleg = prev;
 541                                 }
 542                         }
 543
 544                         inode = nfs_delegation_grab_inode(delegation);
 545                         if (inode == NULL) {
 546                                 rcu_read_unlock();
 547                                 if (to_put)
 548                                         iput(to_put);
 549                                 nfs_sb_deactive(server->super);
 550                                 goto restart;
 551                         }
 552                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 553                         rcu_read_unlock();
 554
 555                         if (to_put)
 556                                 iput(to_put);
 557
 558                         err = nfs_end_delegation_return(inode, delegation, 0);
 559                         iput(inode);
 560                         nfs_sb_deactive(server->super);
 561                         cond_resched();
 562                         if (!err)
 563                                 goto restart;
 564                         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 565                         if (place_holder)
 566                                 iput(place_holder);
 567                         return err;
 568                 }
 569         }
 570         rcu_read_unlock();
 571         if (place_holder)
 572                 iput(place_holder);
 573         return 0;
 574 }
 575
 576 /**
 577  * nfs_inode_return_delegation_noreclaim - return delegation, don't reclaim opens
 578  * @inode: inode to process
 579  *
 580  * Does not protect against delegation reclaims, therefore really only safe
 581  * to be called from nfs4_clear_inode().
 582  */
 583 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
 584 {
 585         struct nfs_delegation *delegation;
 586
 587         delegation = nfs_inode_detach_delegation(inode);
 588         if (delegation != NULL)
 589                 nfs_do_return_delegation(inode, delegation, 1);
 590 }
 591
 592 /**
 593  * nfs_inode_return_delegation - synchronously return a delegation
 594  * @inode: inode to process
 595  *
 596  * This routine will always flush any dirty data to disk on the
 597  * assumption that if we need to return the delegation, then
 598  * we should stop caching.
 599  *
 600  * Returns zero on success, or a negative errno value.
 601  */
 602 int nfs4_inode_return_delegation(struct inode *inode)
 603 {
 604         struct nfs_inode *nfsi = NFS_I(inode);
 605         struct nfs_delegation *delegation;
 606         int err = 0;
 607
 608         nfs_wb_all(inode);
 609         delegation = nfs_start_delegation_return(nfsi);
 610         if (delegation != NULL)
 611                 err = nfs_end_delegation_return(inode, delegation, 1);
 612         return err;
 613 }
 614
 615 /**
 616  * nfs4_inode_make_writeable
 617  * @inode: pointer to inode
 618  *
 619  * Make the inode writeable by returning the delegation if necessary
 620  *
 621  * Returns zero on success, or a negative errno value.
 622  */
 623 int nfs4_inode_make_writeable(struct inode *inode)
 624 {
 625         if (!nfs4_has_session(NFS_SERVER(inode)->nfs_client) ||
 626             !nfs4_check_delegation(inode, FMODE_WRITE))
 627                 return nfs4_inode_return_delegation(inode);
 628         return 0;
 629 }
 630
 631 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
 632                 struct nfs_delegation *delegation)
 633 {
 634         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 635         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 636 }
 637
 638 static void nfs_mark_return_delegation(struct nfs_server *server,
 639                 struct nfs_delegation *delegation)
 640 {
 641         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 642         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 643 }
 644
 645 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
 646 {
 647         struct nfs_delegation *delegation;
 648         bool ret = false;
 649
 650         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 651                 nfs_mark_return_delegation(server, delegation);
 652                 ret = true;
 653         }
 654         return ret;
 655 }
 656
 657 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 658 {
 659         struct nfs_server *server;
 660
 661         rcu_read_lock();
 662         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 663                 nfs_server_mark_return_all_delegations(server);
 664         rcu_read_unlock();
 665 }
 666
 667 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 668 {
 669         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 670                 nfs4_schedule_state_manager(clp);
 671 }
 672
 673 /**
 674  * nfs_expire_all_delegations
 675  * @clp: client to process
 676  *
 677  */
 678 void nfs_expire_all_delegations(struct nfs_client *clp)
 679 {
 680         nfs_client_mark_return_all_delegations(clp);
 681         nfs_delegation_run_state_manager(clp);
 682 }
 683
 684 /**
 685  * nfs_super_return_all_delegations - return delegations for one superblock
 686  * @sb: sb to process
 687  *
 688  */
 689 void nfs_server_return_all_delegations(struct nfs_server *server)
 690 {
 691         struct nfs_client *clp = server->nfs_client;
 692         bool need_wait;
 693
 694         if (clp == NULL)
 695                 return;
 696
 697         rcu_read_lock();
 698         need_wait = nfs_server_mark_return_all_delegations(server);
 699         rcu_read_unlock();
 700
 701         if (need_wait) {
 702                 nfs4_schedule_state_manager(clp);
 703                 nfs4_wait_clnt_recover(clp);
 704         }
 705 }
 706
 707 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
 708                                                  fmode_t flags)
 709 {
 710         struct nfs_delegation *delegation;
 711
 712         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 713                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 714                         continue;
 715                 if (delegation->type & flags)
 716                         nfs_mark_return_if_closed_delegation(server, delegation);
 717         }
 718 }
 719
 720 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
 721                                                         fmode_t flags)
 722 {
 723         struct nfs_server *server;
 724
 725         rcu_read_lock();
 726         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 727                 nfs_mark_return_unused_delegation_types(server, flags);
 728         rcu_read_unlock();
 729 }
 730
 731 static void nfs_mark_delegation_revoked(struct nfs_server *server,
 732                 struct nfs_delegation *delegation)
 733 {
 734         set_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 735         delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
 736         nfs_mark_return_delegation(server, delegation);
 737 }
 738
 739 static bool nfs_revoke_delegation(struct inode *inode,
 740                 const nfs4_stateid *stateid)
 741 {
 742         struct nfs_delegation *delegation;
 743         nfs4_stateid tmp;
 744         bool ret = false;
 745
 746         rcu_read_lock();
 747         delegation = rcu_dereference(NFS_I(inode)->delegation);
 748         if (delegation == NULL)
 749                 goto out;
 750         if (stateid == NULL) {
 751                 nfs4_stateid_copy(&tmp, &delegation->stateid);
 752                 stateid = &tmp;
 753         } else if (!nfs4_stateid_match(stateid, &delegation->stateid))
 754                 goto out;
 755         nfs_mark_delegation_revoked(NFS_SERVER(inode), delegation);
 756         ret = true;
 757 out:
 758         rcu_read_unlock();
 759         if (ret)
 760                 nfs_inode_find_state_and_recover(inode, stateid);
 761         return ret;
 762 }
 763
 764 void nfs_remove_bad_delegation(struct inode *inode,
 765                 const nfs4_stateid *stateid)
 766 {
 767         struct nfs_delegation *delegation;
 768
 769         if (!nfs_revoke_delegation(inode, stateid))
 770                 return;
 771         delegation = nfs_inode_detach_delegation(inode);
 772         if (delegation)
 773                 nfs_free_delegation(delegation);
 774 }
 775 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 776
 777 /**
 778  * nfs_expire_unused_delegation_types
 779  * @clp: client to process
 780  * @flags: delegation types to expire
 781  *
 782  */
 783 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
 784 {
 785         nfs_client_mark_return_unused_delegation_types(clp, flags);
 786         nfs_delegation_run_state_manager(clp);
 787 }
 788
 789 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
 790 {
 791         struct nfs_delegation *delegation;
 792
 793         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 794                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
 795                         continue;
 796                 nfs_mark_return_if_closed_delegation(server, delegation);
 797         }
 798 }
 799
 800 /**
 801  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
 802  * @clp: nfs_client to process
 803  *
 804  */
 805 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
 806 {
 807         struct nfs_server *server;
 808
 809         rcu_read_lock();
 810         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 811                 nfs_mark_return_unreferenced_delegations(server);
 812         rcu_read_unlock();
 813
 814         nfs_delegation_run_state_manager(clp);
 815 }
 816
 817 /**
 818  * nfs_async_inode_return_delegation - asynchronously return a delegation
 819  * @inode: inode to process
 820  * @stateid: state ID information
 821  *
 822  * Returns zero on success, or a negative errno value.
 823  */
 824 int nfs_async_inode_return_delegation(struct inode *inode,
 825                                       const nfs4_stateid *stateid)
 826 {
 827         struct nfs_server *server = NFS_SERVER(inode);
 828         struct nfs_client *clp = server->nfs_client;
 829         struct nfs_delegation *delegation;
 830
 831         rcu_read_lock();
 832         delegation = rcu_dereference(NFS_I(inode)->delegation);
 833         if (delegation == NULL)
 834                 goto out_enoent;
 835         if (stateid != NULL &&
 836             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
 837                 goto out_enoent;
 838         nfs_mark_return_delegation(server, delegation);
 839         rcu_read_unlock();
 840
 841         nfs_delegation_run_state_manager(clp);
 842         return 0;
 843 out_enoent:
 844         rcu_read_unlock();
 845         return -ENOENT;
 846 }
 847
 848 static struct inode *
 849 nfs_delegation_find_inode_server(struct nfs_server *server,
 850                                  const struct nfs_fh *fhandle)
 851 {
 852         struct nfs_delegation *delegation;
 853         struct inode *res = NULL;
 854
 855         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 856                 spin_lock(&delegation->lock);
 857                 if (delegation->inode != NULL &&
 858                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
 859                         res = igrab(delegation->inode);
 860                         spin_unlock(&delegation->lock);
 861                         if (res != NULL)
 862                                 return res;
 863                         return ERR_PTR(-EAGAIN);
 864                 }
 865                 spin_unlock(&delegation->lock);
 866         }
 867         return ERR_PTR(-ENOENT);
 868 }
 869
 870 /**
 871  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
 872  * @clp: client state handle
 873  * @fhandle: filehandle from a delegation recall
 874  *
 875  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
 876  * cannot be found.
 877  */
 878 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
 879                                         const struct nfs_fh *fhandle)
 880 {
 881         struct nfs_server *server;
 882         struct inode *res;
 883
 884         rcu_read_lock();
 885         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 886                 res = nfs_delegation_find_inode_server(server, fhandle);
 887                 if (res != ERR_PTR(-ENOENT)) {
 888                         rcu_read_unlock();
 889                         return res;
 890                 }
 891         }
 892         rcu_read_unlock();
 893         return ERR_PTR(-ENOENT);
 894 }
 895
 896 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
 897 {
 898         struct nfs_delegation *delegation;
 899
 900         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 901                 /*
 902                  * If the delegation may have been admin revoked, then we
 903                  * cannot reclaim it.
 904                  */
 905                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
 906                         continue;
 907                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 908         }
 909 }
 910
 911 /**
 912  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
 913  * @clp: nfs_client to process
 914  *
 915  */
 916 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 917 {
 918         struct nfs_server *server;
 919
 920         rcu_read_lock();
 921         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 922                 nfs_delegation_mark_reclaim_server(server);
 923         rcu_read_unlock();
 924 }
 925
 926 /**
 927  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
 928  * @clp: nfs_client to process
 929  *
 930  */
 931 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 932 {
 933         struct nfs_delegation *delegation;
 934         struct nfs_server *server;
 935         struct inode *inode;
 936
 937 restart:
 938         rcu_read_lock();
 939         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 940                 list_for_each_entry_rcu(delegation, &server->delegations,
 941                                                                 super_list) {
 942                         if (test_bit(NFS_DELEGATION_RETURNING,
 943                                                 &delegation->flags))
 944                                 continue;
 945                         if (test_bit(NFS_DELEGATION_NEED_RECLAIM,
 946                                                 &delegation->flags) == 0)
 947                                 continue;
 948                         if (!nfs_sb_active(server->super))
 949                                 break; /* continue in outer loop */
 950                         inode = nfs_delegation_grab_inode(delegation);
 951                         if (inode == NULL) {
 952                                 rcu_read_unlock();
 953                                 nfs_sb_deactive(server->super);
 954                                 goto restart;
 955                         }
 956                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 957                         rcu_read_unlock();
 958                         if (delegation != NULL) {
 959                                 delegation = nfs_detach_delegation(NFS_I(inode),
 960                                         delegation, server);
 961                                 if (delegation != NULL)
 962                                         nfs_free_delegation(delegation);
 963                         }
 964                         iput(inode);
 965                         nfs_sb_deactive(server->super);
 966                         cond_resched();
 967                         goto restart;
 968                 }
 969         }
 970         rcu_read_unlock();
 971 }
 972
 973 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
 974 {
 975         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
 976                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
 977                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
 978 }
 979
 980 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
 981             struct nfs_delegation *delegation)
 982 {
 983         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
 984                 return;
 985         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 986         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
 987         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
 988 }
 989
 990 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
 991                 struct inode *inode)
 992 {
 993         struct nfs_delegation *delegation;
 994
 995         rcu_read_lock();
 996         delegation = rcu_dereference(NFS_I(inode)->delegation);
 997         if (delegation)
 998                 nfs_mark_test_expired_delegation(server, delegation);
 999         rcu_read_unlock();
1000
1001 }
1002
1003 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1004 {
1005         struct nfs_delegation *delegation;
1006
1007         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1008                 nfs_mark_test_expired_delegation(server, delegation);
1009 }
1010
1011 /**
1012  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1013  * @clp: nfs_client to process
1014  *
1015  * Iterates through all the delegations associated with this server and
1016  * marks them as needing to be checked for validity.
1017  */
1018 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1019 {
1020         struct nfs_server *server;
1021
1022         rcu_read_lock();
1023         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1024                 nfs_delegation_mark_test_expired_server(server);
1025         rcu_read_unlock();
1026 }
1027
1028 /**
1029  * nfs_reap_expired_delegations - reap expired delegations
1030  * @clp: nfs_client to process
1031  *
1032  * Iterates through all the delegations associated with this server and
1033  * checks if they have may have been revoked. This function is usually
1034  * expected to be called in cases where the server may have lost its
1035  * lease.
1036  */
1037 void nfs_reap_expired_delegations(struct nfs_client *clp)
1038 {
1039         const struct nfs4_minor_version_ops *ops = clp->cl_mvops;
1040         struct nfs_delegation *delegation;
1041         struct nfs_server *server;
1042         struct inode *inode;
1043         struct rpc_cred *cred;
1044         nfs4_stateid stateid;
1045
1046 restart:
1047         rcu_read_lock();
1048         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1049                 list_for_each_entry_rcu(delegation, &server->delegations,
1050                                                                 super_list) {
1051                         if (test_bit(NFS_DELEGATION_RETURNING,
1052                                                 &delegation->flags))
1053                                 continue;
1054                         if (test_bit(NFS_DELEGATION_TEST_EXPIRED,
1055                                                 &delegation->flags) == 0)
1056                                 continue;
1057                         if (!nfs_sb_active(server->super))
1058                                 break; /* continue in outer loop */
1059                         inode = nfs_delegation_grab_inode(delegation);
1060                         if (inode == NULL) {
1061                                 rcu_read_unlock();
1062                                 nfs_sb_deactive(server->super);
1063                                 goto restart;
1064                         }
1065                         cred = get_rpccred_rcu(delegation->cred);
1066                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1067                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1068                         rcu_read_unlock();
1069                         if (cred != NULL &&
1070                             ops->test_and_free_expired(server, &stateid, cred) < 0) {
1071                                 nfs_revoke_delegation(inode, &stateid);
1072                                 nfs_inode_find_state_and_recover(inode, &stateid);
1073                         }
1074                         put_rpccred(cred);
1075                         if (nfs4_server_rebooted(clp)) {
1076                                 nfs_inode_mark_test_expired_delegation(server,inode);
1077                                 iput(inode);
1078                                 nfs_sb_deactive(server->super);
1079                                 return;
1080                         }
1081                         iput(inode);
1082                         nfs_sb_deactive(server->super);
1083                         cond_resched();
1084                         goto restart;
1085                 }
1086         }
1087         rcu_read_unlock();
1088 }
1089
1090 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1091                 const nfs4_stateid *stateid)
1092 {
1093         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1094         struct nfs_delegation *delegation;
1095         bool found = false;
1096
1097         rcu_read_lock();
1098         delegation = rcu_dereference(NFS_I(inode)->delegation);
1099         if (delegation &&
1100             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1101                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1102                 found = true;
1103         }
1104         rcu_read_unlock();
1105         if (found)
1106                 nfs4_schedule_state_manager(clp);
1107 }
1108
1109 /**
1110  * nfs_delegations_present - check for existence of delegations
1111  * @clp: client state handle
1112  *
1113  * Returns one if there are any nfs_delegation structures attached
1114  * to this nfs_client.
1115  */
1116 int nfs_delegations_present(struct nfs_client *clp)
1117 {
1118         struct nfs_server *server;
1119         int ret = 0;
1120
1121         rcu_read_lock();
1122         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1123                 if (!list_empty(&server->delegations)) {
1124                         ret = 1;
1125                         break;
1126                 }
1127         rcu_read_unlock();
1128         return ret;
1129 }
1130
1131 /**
1132  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1133  * @dst: stateid to refresh
1134  * @inode: inode to check
1135  *
1136  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1137  * that matches our delegation stateid. Otherwise "false" is returned.
1138  */
1139 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1140 {
1141         struct nfs_delegation *delegation;
1142         bool ret = false;
1143         if (!inode)
1144                 goto out;
1145
1146         rcu_read_lock();
1147         delegation = rcu_dereference(NFS_I(inode)->delegation);
1148         if (delegation != NULL &&
1149             nfs4_stateid_match_other(dst, &delegation->stateid)) {
1150                 dst->seqid = delegation->stateid.seqid;
1151                 return ret;
1152         }
1153         rcu_read_unlock();
1154 out:
1155         return ret;
1156 }
1157
1158 /**
1159  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1160  * @inode: inode to check
1161  * @flags: delegation type requirement
1162  * @dst: stateid data structure to fill in
1163  * @cred: optional argument to retrieve credential
1164  *
1165  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1166  * otherwise "false" is returned.
1167  */
1168 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1169                 nfs4_stateid *dst, struct rpc_cred **cred)
1170 {
1171         struct nfs_inode *nfsi = NFS_I(inode);
1172         struct nfs_delegation *delegation;
1173         bool ret;
1174
1175         flags &= FMODE_READ|FMODE_WRITE;
1176         rcu_read_lock();
1177         delegation = rcu_dereference(nfsi->delegation);
1178         ret = nfs4_is_valid_delegation(delegation, flags);
1179         if (ret) {
1180                 nfs4_stateid_copy(dst, &delegation->stateid);
1181                 nfs_mark_delegation_referenced(delegation);
1182                 if (cred)
1183                         *cred = get_rpccred(delegation->cred);
1184         }
1185         rcu_read_unlock();
1186         return ret;
1187 }
1188
1189 /**
1190  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1191  * @inode: inode to check
1192  *
1193  * This function checks the number of outstanding writes to the file
1194  * against the delegation 'space_limit' field to see if
1195  * the spec requires us to flush the file on close.
1196  */
1197 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1198 {
1199         struct nfs_inode *nfsi = NFS_I(inode);
1200         struct nfs_delegation *delegation;
1201         bool ret = true;
1202
1203         rcu_read_lock();
1204         delegation = rcu_dereference(nfsi->delegation);
1205         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1206                 goto out;
1207         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1208                 ret = false;
1209 out:
1210         rcu_read_unlock();
1211         return ret;
1212 }