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         put_cred(delegation->cred);
  30         delegation->cred = NULL;
  31         kfree_rcu(delegation, rcu);
  32 }
  33
  34 /**
  35  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
  36  * @delegation: delegation to process
  37  *
  38  */
  39 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
  40 {
  41         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
  42 }
  43
  44 static bool
  45 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
  46                 fmode_t flags)
  47 {
  48         if (delegation != NULL && (delegation->type & flags) == flags &&
  49             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
  50             !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
  51                 return true;
  52         return false;
  53 }
  54
  55 static int
  56 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
  57 {
  58         struct nfs_delegation *delegation;
  59         int ret = 0;
  60
  61         flags &= FMODE_READ|FMODE_WRITE;
  62         rcu_read_lock();
  63         delegation = rcu_dereference(NFS_I(inode)->delegation);
  64         if (nfs4_is_valid_delegation(delegation, flags)) {
  65                 if (mark)
  66                         nfs_mark_delegation_referenced(delegation);
  67                 ret = 1;
  68         }
  69         rcu_read_unlock();
  70         return ret;
  71 }
  72 /**
  73  * nfs_have_delegation - check if inode has a delegation, mark it
  74  * NFS_DELEGATION_REFERENCED if there is one.
  75  * @inode: inode to check
  76  * @flags: delegation types to check for
  77  *
  78  * Returns one if inode has the indicated delegation, otherwise zero.
  79  */
  80 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
  81 {
  82         return nfs4_do_check_delegation(inode, flags, true);
  83 }
  84
  85 /*
  86  * nfs4_check_delegation - check if inode has a delegation, do not mark
  87  * NFS_DELEGATION_REFERENCED if it has one.
  88  */
  89 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
  90 {
  91         return nfs4_do_check_delegation(inode, flags, false);
  92 }
  93
  94 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
  95 {
  96         struct inode *inode = state->inode;
  97         struct file_lock *fl;
  98         struct file_lock_context *flctx = inode->i_flctx;
  99         struct list_head *list;
 100         int status = 0;
 101
 102         if (flctx == NULL)
 103                 goto out;
 104
 105         list = &flctx->flc_posix;
 106         spin_lock(&flctx->flc_lock);
 107 restart:
 108         list_for_each_entry(fl, list, fl_list) {
 109                 if (nfs_file_open_context(fl->fl_file)->state != state)
 110                         continue;
 111                 spin_unlock(&flctx->flc_lock);
 112                 status = nfs4_lock_delegation_recall(fl, state, stateid);
 113                 if (status < 0)
 114                         goto out;
 115                 spin_lock(&flctx->flc_lock);
 116         }
 117         if (list == &flctx->flc_posix) {
 118                 list = &flctx->flc_flock;
 119                 goto restart;
 120         }
 121         spin_unlock(&flctx->flc_lock);
 122 out:
 123         return status;
 124 }
 125
 126 static int nfs_delegation_claim_opens(struct inode *inode,
 127                 const nfs4_stateid *stateid, fmode_t type)
 128 {
 129         struct nfs_inode *nfsi = NFS_I(inode);
 130         struct nfs_open_context *ctx;
 131         struct nfs4_state_owner *sp;
 132         struct nfs4_state *state;
 133         unsigned int seq;
 134         int err;
 135
 136 again:
 137         rcu_read_lock();
 138         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
 139                 state = ctx->state;
 140                 if (state == NULL)
 141                         continue;
 142                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 143                         continue;
 144                 if (!nfs4_valid_open_stateid(state))
 145                         continue;
 146                 if (!nfs4_stateid_match(&state->stateid, stateid))
 147                         continue;
 148                 if (!get_nfs_open_context(ctx))
 149                         continue;
 150                 rcu_read_unlock();
 151                 sp = state->owner;
 152                 /* Block nfs4_proc_unlck */
 153                 mutex_lock(&sp->so_delegreturn_mutex);
 154                 seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
 155                 err = nfs4_open_delegation_recall(ctx, state, stateid, type);
 156                 if (!err)
 157                         err = nfs_delegation_claim_locks(state, stateid);
 158                 if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
 159                         err = -EAGAIN;
 160                 mutex_unlock(&sp->so_delegreturn_mutex);
 161                 put_nfs_open_context(ctx);
 162                 if (err != 0)
 163                         return err;
 164                 goto again;
 165         }
 166         rcu_read_unlock();
 167         return 0;
 168 }
 169
 170 /**
 171  * nfs_inode_reclaim_delegation - process a delegation reclaim request
 172  * @inode: inode to process
 173  * @cred: credential to use for request
 174  * @type: delegation type
 175  * @stateid: delegation stateid
 176  * @pagemod_limit: write delegation "space_limit"
 177  *
 178  */
 179 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
 180                                   fmode_t type,
 181                                   const nfs4_stateid *stateid,
 182                                   unsigned long pagemod_limit)
 183 {
 184         struct nfs_delegation *delegation;
 185         const struct cred *oldcred = NULL;
 186
 187         rcu_read_lock();
 188         delegation = rcu_dereference(NFS_I(inode)->delegation);
 189         if (delegation != NULL) {
 190                 spin_lock(&delegation->lock);
 191                 if (delegation->inode != NULL) {
 192                         nfs4_stateid_copy(&delegation->stateid, stateid);
 193                         delegation->type = type;
 194                         delegation->pagemod_limit = pagemod_limit;
 195                         oldcred = delegation->cred;
 196                         delegation->cred = get_cred(cred);
 197                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
 198                                   &delegation->flags);
 199                         spin_unlock(&delegation->lock);
 200                         rcu_read_unlock();
 201                         put_cred(oldcred);
 202                         trace_nfs4_reclaim_delegation(inode, type);
 203                         return;
 204                 }
 205                 /* We appear to have raced with a delegation return. */
 206                 spin_unlock(&delegation->lock);
 207         }
 208         rcu_read_unlock();
 209         nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
 210 }
 211
 212 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 213 {
 214         int res = 0;
 215
 216         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 217                 res = nfs4_proc_delegreturn(inode,
 218                                 delegation->cred,
 219                                 &delegation->stateid,
 220                                 issync);
 221         nfs_free_delegation(delegation);
 222         return res;
 223 }
 224
 225 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 226 {
 227         struct inode *inode = NULL;
 228
 229         spin_lock(&delegation->lock);
 230         if (delegation->inode != NULL)
 231                 inode = igrab(delegation->inode);
 232         if (!inode)
 233                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 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, const struct 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_cred(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  * @server: pointer to nfs_server 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 *freeme, *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                         freeme = igrab(delegation->inode);
 860                         if (freeme && nfs_sb_active(freeme->i_sb))
 861                                 res = freeme;
 862                         spin_unlock(&delegation->lock);
 863                         if (res != NULL)
 864                                 return res;
 865                         if (freeme) {
 866                                 rcu_read_unlock();
 867                                 iput(freeme);
 868                                 rcu_read_lock();
 869                         }
 870                         return ERR_PTR(-EAGAIN);
 871                 }
 872                 spin_unlock(&delegation->lock);
 873         }
 874         return ERR_PTR(-ENOENT);
 875 }
 876
 877 /**
 878  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
 879  * @clp: client state handle
 880  * @fhandle: filehandle from a delegation recall
 881  *
 882  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
 883  * cannot be found.
 884  */
 885 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
 886                                         const struct nfs_fh *fhandle)
 887 {
 888         struct nfs_server *server;
 889         struct inode *res;
 890
 891         rcu_read_lock();
 892         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 893                 res = nfs_delegation_find_inode_server(server, fhandle);
 894                 if (res != ERR_PTR(-ENOENT)) {
 895                         rcu_read_unlock();
 896                         return res;
 897                 }
 898         }
 899         rcu_read_unlock();
 900         return ERR_PTR(-ENOENT);
 901 }
 902
 903 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
 904 {
 905         struct nfs_delegation *delegation;
 906
 907         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 908                 /*
 909                  * If the delegation may have been admin revoked, then we
 910                  * cannot reclaim it.
 911                  */
 912                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
 913                         continue;
 914                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 915         }
 916 }
 917
 918 /**
 919  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
 920  * @clp: nfs_client to process
 921  *
 922  */
 923 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 924 {
 925         struct nfs_server *server;
 926
 927         rcu_read_lock();
 928         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 929                 nfs_delegation_mark_reclaim_server(server);
 930         rcu_read_unlock();
 931 }
 932
 933 /**
 934  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
 935  * @clp: nfs_client to process
 936  *
 937  */
 938 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 939 {
 940         struct nfs_delegation *delegation;
 941         struct nfs_server *server;
 942         struct inode *inode;
 943
 944 restart:
 945         rcu_read_lock();
 946         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 947                 list_for_each_entry_rcu(delegation, &server->delegations,
 948                                                                 super_list) {
 949                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
 950                                                 &delegation->flags) ||
 951                             test_bit(NFS_DELEGATION_RETURNING,
 952                                                 &delegation->flags) ||
 953                             test_bit(NFS_DELEGATION_NEED_RECLAIM,
 954                                                 &delegation->flags) == 0)
 955                                 continue;
 956                         if (!nfs_sb_active(server->super))
 957                                 break; /* continue in outer loop */
 958                         inode = nfs_delegation_grab_inode(delegation);
 959                         if (inode == NULL) {
 960                                 rcu_read_unlock();
 961                                 nfs_sb_deactive(server->super);
 962                                 goto restart;
 963                         }
 964                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 965                         rcu_read_unlock();
 966                         if (delegation != NULL) {
 967                                 delegation = nfs_detach_delegation(NFS_I(inode),
 968                                         delegation, server);
 969                                 if (delegation != NULL)
 970                                         nfs_free_delegation(delegation);
 971                         }
 972                         iput(inode);
 973                         nfs_sb_deactive(server->super);
 974                         cond_resched();
 975                         goto restart;
 976                 }
 977         }
 978         rcu_read_unlock();
 979 }
 980
 981 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
 982 {
 983         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
 984                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
 985                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
 986 }
 987
 988 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
 989             struct nfs_delegation *delegation)
 990 {
 991         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
 992                 return;
 993         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 994         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
 995         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
 996 }
 997
 998 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
 999                 struct inode *inode)
1000 {
1001         struct nfs_delegation *delegation;
1002
1003         rcu_read_lock();
1004         delegation = rcu_dereference(NFS_I(inode)->delegation);
1005         if (delegation)
1006                 nfs_mark_test_expired_delegation(server, delegation);
1007         rcu_read_unlock();
1008
1009 }
1010
1011 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1012 {
1013         struct nfs_delegation *delegation;
1014
1015         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1016                 nfs_mark_test_expired_delegation(server, delegation);
1017 }
1018
1019 /**
1020  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1021  * @clp: nfs_client to process
1022  *
1023  * Iterates through all the delegations associated with this server and
1024  * marks them as needing to be checked for validity.
1025  */
1026 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1027 {
1028         struct nfs_server *server;
1029
1030         rcu_read_lock();
1031         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1032                 nfs_delegation_mark_test_expired_server(server);
1033         rcu_read_unlock();
1034 }
1035
1036 /**
1037  * nfs_test_expired_all_delegations - test all delegations for a client
1038  * @clp: nfs_client to process
1039  *
1040  * Helper for handling "recallable state revoked" status from server.
1041  */
1042 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1043 {
1044         nfs_mark_test_expired_all_delegations(clp);
1045         nfs4_schedule_state_manager(clp);
1046 }
1047
1048 /**
1049  * nfs_reap_expired_delegations - reap expired delegations
1050  * @clp: nfs_client to process
1051  *
1052  * Iterates through all the delegations associated with this server and
1053  * checks if they have may have been revoked. This function is usually
1054  * expected to be called in cases where the server may have lost its
1055  * lease.
1056  */
1057 void nfs_reap_expired_delegations(struct nfs_client *clp)
1058 {
1059         const struct nfs4_minor_version_ops *ops = clp->cl_mvops;
1060         struct nfs_delegation *delegation;
1061         struct nfs_server *server;
1062         struct inode *inode;
1063         const struct cred *cred;
1064         nfs4_stateid stateid;
1065
1066 restart:
1067         rcu_read_lock();
1068         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1069                 list_for_each_entry_rcu(delegation, &server->delegations,
1070                                                                 super_list) {
1071                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
1072                                                 &delegation->flags) ||
1073                             test_bit(NFS_DELEGATION_RETURNING,
1074                                                 &delegation->flags) ||
1075                             test_bit(NFS_DELEGATION_TEST_EXPIRED,
1076                                                 &delegation->flags) == 0)
1077                                 continue;
1078                         if (!nfs_sb_active(server->super))
1079                                 break; /* continue in outer loop */
1080                         inode = nfs_delegation_grab_inode(delegation);
1081                         if (inode == NULL) {
1082                                 rcu_read_unlock();
1083                                 nfs_sb_deactive(server->super);
1084                                 goto restart;
1085                         }
1086                         cred = get_cred_rcu(delegation->cred);
1087                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1088                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1089                         rcu_read_unlock();
1090                         if (cred != NULL &&
1091                             ops->test_and_free_expired(server, &stateid, cred) < 0) {
1092                                 nfs_revoke_delegation(inode, &stateid);
1093                                 nfs_inode_find_state_and_recover(inode, &stateid);
1094                         }
1095                         put_cred(cred);
1096                         if (nfs4_server_rebooted(clp)) {
1097                                 nfs_inode_mark_test_expired_delegation(server,inode);
1098                                 iput(inode);
1099                                 nfs_sb_deactive(server->super);
1100                                 return;
1101                         }
1102                         iput(inode);
1103                         nfs_sb_deactive(server->super);
1104                         cond_resched();
1105                         goto restart;
1106                 }
1107         }
1108         rcu_read_unlock();
1109 }
1110
1111 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1112                 const nfs4_stateid *stateid)
1113 {
1114         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1115         struct nfs_delegation *delegation;
1116         bool found = false;
1117
1118         rcu_read_lock();
1119         delegation = rcu_dereference(NFS_I(inode)->delegation);
1120         if (delegation &&
1121             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1122                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1123                 found = true;
1124         }
1125         rcu_read_unlock();
1126         if (found)
1127                 nfs4_schedule_state_manager(clp);
1128 }
1129
1130 /**
1131  * nfs_delegations_present - check for existence of delegations
1132  * @clp: client state handle
1133  *
1134  * Returns one if there are any nfs_delegation structures attached
1135  * to this nfs_client.
1136  */
1137 int nfs_delegations_present(struct nfs_client *clp)
1138 {
1139         struct nfs_server *server;
1140         int ret = 0;
1141
1142         rcu_read_lock();
1143         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1144                 if (!list_empty(&server->delegations)) {
1145                         ret = 1;
1146                         break;
1147                 }
1148         rcu_read_unlock();
1149         return ret;
1150 }
1151
1152 /**
1153  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1154  * @dst: stateid to refresh
1155  * @inode: inode to check
1156  *
1157  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1158  * that matches our delegation stateid. Otherwise "false" is returned.
1159  */
1160 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1161 {
1162         struct nfs_delegation *delegation;
1163         bool ret = false;
1164         if (!inode)
1165                 goto out;
1166
1167         rcu_read_lock();
1168         delegation = rcu_dereference(NFS_I(inode)->delegation);
1169         if (delegation != NULL &&
1170             nfs4_stateid_match_other(dst, &delegation->stateid)) {
1171                 dst->seqid = delegation->stateid.seqid;
1172                 return ret;
1173         }
1174         rcu_read_unlock();
1175 out:
1176         return ret;
1177 }
1178
1179 /**
1180  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1181  * @inode: inode to check
1182  * @flags: delegation type requirement
1183  * @dst: stateid data structure to fill in
1184  * @cred: optional argument to retrieve credential
1185  *
1186  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1187  * otherwise "false" is returned.
1188  */
1189 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1190                 nfs4_stateid *dst, const struct cred **cred)
1191 {
1192         struct nfs_inode *nfsi = NFS_I(inode);
1193         struct nfs_delegation *delegation;
1194         bool ret;
1195
1196         flags &= FMODE_READ|FMODE_WRITE;
1197         rcu_read_lock();
1198         delegation = rcu_dereference(nfsi->delegation);
1199         ret = nfs4_is_valid_delegation(delegation, flags);
1200         if (ret) {
1201                 nfs4_stateid_copy(dst, &delegation->stateid);
1202                 nfs_mark_delegation_referenced(delegation);
1203                 if (cred)
1204                         *cred = get_cred(delegation->cred);
1205         }
1206         rcu_read_unlock();
1207         return ret;
1208 }
1209
1210 /**
1211  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1212  * @inode: inode to check
1213  *
1214  * This function checks the number of outstanding writes to the file
1215  * against the delegation 'space_limit' field to see if
1216  * the spec requires us to flush the file on close.
1217  */
1218 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1219 {
1220         struct nfs_inode *nfsi = NFS_I(inode);
1221         struct nfs_delegation *delegation;
1222         bool ret = true;
1223
1224         rcu_read_lock();
1225         delegation = rcu_dereference(nfsi->delegation);
1226         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1227                 goto out;
1228         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1229                 ret = false;
1230 out:
1231         rcu_read_unlock();
1232         return ret;
1233 }