nfs/delegation.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * linux/fs/nfs/delegation.c
   4  *
   5  * Copyright (C) 2004 Trond Myklebust
   6  *
   7  * NFS file delegation management
   8  *
   9  */
  10 #include <linux/completion.h>
  11 #include <linux/kthread.h>
  12 #include <linux/module.h>
  13 #include <linux/sched.h>
  14 #include <linux/slab.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/iversion.h>
  17
  18 #include <linux/nfs4.h>
  19 #include <linux/nfs_fs.h>
  20 #include <linux/nfs_xdr.h>
  21
  22 #include "nfs4_fs.h"
  23 #include "nfs4session.h"
  24 #include "delegation.h"
  25 #include "internal.h"
  26 #include "nfs4trace.h"
  27
  28 #define NFS_DEFAULT_DELEGATION_WATERMARK (5000U)
  29
  30 static atomic_long_t nfs_active_delegations;
  31 static unsigned nfs_delegation_watermark = NFS_DEFAULT_DELEGATION_WATERMARK;
  32
  33 static void __nfs_free_delegation(struct nfs_delegation *delegation)
  34 {
  35         put_cred(delegation->cred);
  36         delegation->cred = NULL;
  37         kfree_rcu(delegation, rcu);
  38 }
  39
  40 static void nfs_mark_delegation_revoked(struct nfs_delegation *delegation)
  41 {
  42         if (!test_and_set_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
  43                 delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
  44                 atomic_long_dec(&nfs_active_delegations);
  45                 if (!test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
  46                         nfs_clear_verifier_delegated(delegation->inode);
  47         }
  48 }
  49
  50 static struct nfs_delegation *nfs_get_delegation(struct nfs_delegation *delegation)
  51 {
  52         refcount_inc(&delegation->refcount);
  53         return delegation;
  54 }
  55
  56 static void nfs_put_delegation(struct nfs_delegation *delegation)
  57 {
  58         if (refcount_dec_and_test(&delegation->refcount))
  59                 __nfs_free_delegation(delegation);
  60 }
  61
  62 static void nfs_free_delegation(struct nfs_delegation *delegation)
  63 {
  64         nfs_mark_delegation_revoked(delegation);
  65         nfs_put_delegation(delegation);
  66 }
  67
  68 /**
  69  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
  70  * @delegation: delegation to process
  71  *
  72  */
  73 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
  74 {
  75         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
  76 }
  77
  78 static void nfs_mark_return_delegation(struct nfs_server *server,
  79                                        struct nfs_delegation *delegation)
  80 {
  81         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
  82         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
  83 }
  84
  85 static bool
  86 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
  87                 fmode_t flags)
  88 {
  89         if (delegation != NULL && (delegation->type & flags) == flags &&
  90             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
  91             !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
  92                 return true;
  93         return false;
  94 }
  95
  96 struct nfs_delegation *nfs4_get_valid_delegation(const struct inode *inode)
  97 {
  98         struct nfs_delegation *delegation;
  99
 100         delegation = rcu_dereference(NFS_I(inode)->delegation);
 101         if (nfs4_is_valid_delegation(delegation, 0))
 102                 return delegation;
 103         return NULL;
 104 }
 105
 106 static int
 107 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
 108 {
 109         struct nfs_delegation *delegation;
 110         int ret = 0;
 111
 112         flags &= FMODE_READ|FMODE_WRITE;
 113         rcu_read_lock();
 114         delegation = rcu_dereference(NFS_I(inode)->delegation);
 115         if (nfs4_is_valid_delegation(delegation, flags)) {
 116                 if (mark)
 117                         nfs_mark_delegation_referenced(delegation);
 118                 ret = 1;
 119         }
 120         rcu_read_unlock();
 121         return ret;
 122 }
 123 /**
 124  * nfs4_have_delegation - check if inode has a delegation, mark it
 125  * NFS_DELEGATION_REFERENCED if there is one.
 126  * @inode: inode to check
 127  * @flags: delegation types to check for
 128  *
 129  * Returns one if inode has the indicated delegation, otherwise zero.
 130  */
 131 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
 132 {
 133         return nfs4_do_check_delegation(inode, flags, true);
 134 }
 135
 136 /*
 137  * nfs4_check_delegation - check if inode has a delegation, do not mark
 138  * NFS_DELEGATION_REFERENCED if it has one.
 139  */
 140 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
 141 {
 142         return nfs4_do_check_delegation(inode, flags, false);
 143 }
 144
 145 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
 146 {
 147         struct inode *inode = state->inode;
 148         struct file_lock *fl;
 149         struct file_lock_context *flctx = inode->i_flctx;
 150         struct list_head *list;
 151         int status = 0;
 152
 153         if (flctx == NULL)
 154                 goto out;
 155
 156         list = &flctx->flc_posix;
 157         spin_lock(&flctx->flc_lock);
 158 restart:
 159         list_for_each_entry(fl, list, fl_list) {
 160                 if (nfs_file_open_context(fl->fl_file)->state != state)
 161                         continue;
 162                 spin_unlock(&flctx->flc_lock);
 163                 status = nfs4_lock_delegation_recall(fl, state, stateid);
 164                 if (status < 0)
 165                         goto out;
 166                 spin_lock(&flctx->flc_lock);
 167         }
 168         if (list == &flctx->flc_posix) {
 169                 list = &flctx->flc_flock;
 170                 goto restart;
 171         }
 172         spin_unlock(&flctx->flc_lock);
 173 out:
 174         return status;
 175 }
 176
 177 static int nfs_delegation_claim_opens(struct inode *inode,
 178                 const nfs4_stateid *stateid, fmode_t type)
 179 {
 180         struct nfs_inode *nfsi = NFS_I(inode);
 181         struct nfs_open_context *ctx;
 182         struct nfs4_state_owner *sp;
 183         struct nfs4_state *state;
 184         unsigned int seq;
 185         int err;
 186
 187 again:
 188         rcu_read_lock();
 189         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
 190                 state = ctx->state;
 191                 if (state == NULL)
 192                         continue;
 193                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 194                         continue;
 195                 if (!nfs4_valid_open_stateid(state))
 196                         continue;
 197                 if (!nfs4_stateid_match(&state->stateid, stateid))
 198                         continue;
 199                 if (!get_nfs_open_context(ctx))
 200                         continue;
 201                 rcu_read_unlock();
 202                 sp = state->owner;
 203                 /* Block nfs4_proc_unlck */
 204                 mutex_lock(&sp->so_delegreturn_mutex);
 205                 seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
 206                 err = nfs4_open_delegation_recall(ctx, state, stateid);
 207                 if (!err)
 208                         err = nfs_delegation_claim_locks(state, stateid);
 209                 if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
 210                         err = -EAGAIN;
 211                 mutex_unlock(&sp->so_delegreturn_mutex);
 212                 put_nfs_open_context(ctx);
 213                 if (err != 0)
 214                         return err;
 215                 goto again;
 216         }
 217         rcu_read_unlock();
 218         return 0;
 219 }
 220
 221 /**
 222  * nfs_inode_reclaim_delegation - process a delegation reclaim request
 223  * @inode: inode to process
 224  * @cred: credential to use for request
 225  * @type: delegation type
 226  * @stateid: delegation stateid
 227  * @pagemod_limit: write delegation "space_limit"
 228  *
 229  */
 230 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
 231                                   fmode_t type,
 232                                   const nfs4_stateid *stateid,
 233                                   unsigned long pagemod_limit)
 234 {
 235         struct nfs_delegation *delegation;
 236         const struct cred *oldcred = NULL;
 237
 238         rcu_read_lock();
 239         delegation = rcu_dereference(NFS_I(inode)->delegation);
 240         if (delegation != NULL) {
 241                 spin_lock(&delegation->lock);
 242                 if (nfs4_is_valid_delegation(delegation, 0)) {
 243                         nfs4_stateid_copy(&delegation->stateid, stateid);
 244                         delegation->type = type;
 245                         delegation->pagemod_limit = pagemod_limit;
 246                         oldcred = delegation->cred;
 247                         delegation->cred = get_cred(cred);
 248                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
 249                                   &delegation->flags);
 250                         spin_unlock(&delegation->lock);
 251                         rcu_read_unlock();
 252                         put_cred(oldcred);
 253                         trace_nfs4_reclaim_delegation(inode, type);
 254                         return;
 255                 }
 256                 /* We appear to have raced with a delegation return. */
 257                 spin_unlock(&delegation->lock);
 258         }
 259         rcu_read_unlock();
 260         nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
 261 }
 262
 263 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 264 {
 265         const struct cred *cred;
 266         int res = 0;
 267
 268         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
 269                 spin_lock(&delegation->lock);
 270                 cred = get_cred(delegation->cred);
 271                 spin_unlock(&delegation->lock);
 272                 res = nfs4_proc_delegreturn(inode, cred,
 273                                 &delegation->stateid,
 274                                 issync);
 275                 put_cred(cred);
 276         }
 277         return res;
 278 }
 279
 280 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 281 {
 282         struct inode *inode = NULL;
 283
 284         spin_lock(&delegation->lock);
 285         if (delegation->inode != NULL)
 286                 inode = igrab(delegation->inode);
 287         if (!inode)
 288                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 289         spin_unlock(&delegation->lock);
 290         return inode;
 291 }
 292
 293 static struct nfs_delegation *
 294 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
 295 {
 296         struct nfs_delegation *ret = NULL;
 297         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
 298
 299         if (delegation == NULL)
 300                 goto out;
 301         spin_lock(&delegation->lock);
 302         if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
 303                 clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
 304                 /* Refcount matched in nfs_end_delegation_return() */
 305                 ret = nfs_get_delegation(delegation);
 306         }
 307         spin_unlock(&delegation->lock);
 308         if (ret)
 309                 nfs_clear_verifier_delegated(&nfsi->vfs_inode);
 310 out:
 311         return ret;
 312 }
 313
 314 static struct nfs_delegation *
 315 nfs_start_delegation_return(struct nfs_inode *nfsi)
 316 {
 317         struct nfs_delegation *delegation;
 318
 319         rcu_read_lock();
 320         delegation = nfs_start_delegation_return_locked(nfsi);
 321         rcu_read_unlock();
 322         return delegation;
 323 }
 324
 325 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
 326                                         struct nfs_client *clp, int err)
 327 {
 328
 329         spin_lock(&delegation->lock);
 330         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 331         if (err == -EAGAIN) {
 332                 set_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
 333                 set_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state);
 334         }
 335         spin_unlock(&delegation->lock);
 336 }
 337
 338 static struct nfs_delegation *
 339 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
 340                 struct nfs_delegation *delegation,
 341                 struct nfs_client *clp)
 342 {
 343         struct nfs_delegation *deleg_cur =
 344                 rcu_dereference_protected(nfsi->delegation,
 345                                 lockdep_is_held(&clp->cl_lock));
 346
 347         if (deleg_cur == NULL || delegation != deleg_cur)
 348                 return NULL;
 349
 350         spin_lock(&delegation->lock);
 351         if (!delegation->inode) {
 352                 spin_unlock(&delegation->lock);
 353                 return NULL;
 354         }
 355         list_del_rcu(&delegation->super_list);
 356         delegation->inode = NULL;
 357         rcu_assign_pointer(nfsi->delegation, NULL);
 358         spin_unlock(&delegation->lock);
 359         return delegation;
 360 }
 361
 362 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
 363                 struct nfs_delegation *delegation,
 364                 struct nfs_server *server)
 365 {
 366         struct nfs_client *clp = server->nfs_client;
 367
 368         spin_lock(&clp->cl_lock);
 369         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
 370         spin_unlock(&clp->cl_lock);
 371         return delegation;
 372 }
 373
 374 static struct nfs_delegation *
 375 nfs_inode_detach_delegation(struct inode *inode)
 376 {
 377         struct nfs_inode *nfsi = NFS_I(inode);
 378         struct nfs_server *server = NFS_SERVER(inode);
 379         struct nfs_delegation *delegation;
 380
 381         rcu_read_lock();
 382         delegation = rcu_dereference(nfsi->delegation);
 383         if (delegation != NULL)
 384                 delegation = nfs_detach_delegation(nfsi, delegation, server);
 385         rcu_read_unlock();
 386         return delegation;
 387 }
 388
 389 static void
 390 nfs_update_delegation_cred(struct nfs_delegation *delegation,
 391                 const struct cred *cred)
 392 {
 393         const struct cred *old;
 394
 395         if (cred_fscmp(delegation->cred, cred) != 0) {
 396                 old = xchg(&delegation->cred, get_cred(cred));
 397                 put_cred(old);
 398         }
 399 }
 400
 401 static void
 402 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
 403                 const struct nfs_delegation *update)
 404 {
 405         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
 406                 delegation->stateid.seqid = update->stateid.seqid;
 407                 smp_wmb();
 408                 delegation->type = update->type;
 409                 delegation->pagemod_limit = update->pagemod_limit;
 410                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
 411                         delegation->change_attr = update->change_attr;
 412                         nfs_update_delegation_cred(delegation, update->cred);
 413                         /* smp_mb__before_atomic() is implicit due to xchg() */
 414                         clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 415                         atomic_long_inc(&nfs_active_delegations);
 416                 }
 417         }
 418 }
 419
 420 /**
 421  * nfs_inode_set_delegation - set up a delegation on an inode
 422  * @inode: inode to which delegation applies
 423  * @cred: cred to use for subsequent delegation processing
 424  * @type: delegation type
 425  * @stateid: delegation stateid
 426  * @pagemod_limit: write delegation "space_limit"
 427  *
 428  * Returns zero on success, or a negative errno value.
 429  */
 430 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
 431                                   fmode_t type,
 432                                   const nfs4_stateid *stateid,
 433                                   unsigned long pagemod_limit)
 434 {
 435         struct nfs_server *server = NFS_SERVER(inode);
 436         struct nfs_client *clp = server->nfs_client;
 437         struct nfs_inode *nfsi = NFS_I(inode);
 438         struct nfs_delegation *delegation, *old_delegation;
 439         struct nfs_delegation *freeme = NULL;
 440         int status = 0;
 441
 442         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
 443         if (delegation == NULL)
 444                 return -ENOMEM;
 445         nfs4_stateid_copy(&delegation->stateid, stateid);
 446         refcount_set(&delegation->refcount, 1);
 447         delegation->type = type;
 448         delegation->pagemod_limit = pagemod_limit;
 449         delegation->change_attr = inode_peek_iversion_raw(inode);
 450         delegation->cred = get_cred(cred);
 451         delegation->inode = inode;
 452         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 453         spin_lock_init(&delegation->lock);
 454
 455         spin_lock(&clp->cl_lock);
 456         old_delegation = rcu_dereference_protected(nfsi->delegation,
 457                                         lockdep_is_held(&clp->cl_lock));
 458         if (old_delegation == NULL)
 459                 goto add_new;
 460         /* Is this an update of the existing delegation? */
 461         if (nfs4_stateid_match_other(&old_delegation->stateid,
 462                                 &delegation->stateid)) {
 463                 spin_lock(&old_delegation->lock);
 464                 nfs_update_inplace_delegation(old_delegation,
 465                                 delegation);
 466                 spin_unlock(&old_delegation->lock);
 467                 goto out;
 468         }
 469         if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
 470                 /*
 471                  * Deal with broken servers that hand out two
 472                  * delegations for the same file.
 473                  * Allow for upgrades to a WRITE delegation, but
 474                  * nothing else.
 475                  */
 476                 dfprintk(FILE, "%s: server %s handed out "
 477                                 "a duplicate delegation!\n",
 478                                 __func__, clp->cl_hostname);
 479                 if (delegation->type == old_delegation->type ||
 480                     !(delegation->type & FMODE_WRITE)) {
 481                         freeme = delegation;
 482                         delegation = NULL;
 483                         goto out;
 484                 }
 485                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
 486                                         &old_delegation->flags))
 487                         goto out;
 488         }
 489         freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
 490         if (freeme == NULL)
 491                 goto out;
 492 add_new:
 493         /*
 494          * If we didn't revalidate the change attribute before setting
 495          * the delegation, then pre-emptively ask for a full attribute
 496          * cache revalidation.
 497          */
 498         spin_lock(&inode->i_lock);
 499         if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
 500                 nfs_set_cache_invalid(inode,
 501                         NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
 502                         NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
 503                         NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
 504                         NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
 505                         NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
 506                         NFS_INO_INVALID_XATTR);
 507         spin_unlock(&inode->i_lock);
 508
 509         list_add_tail_rcu(&delegation->super_list, &server->delegations);
 510         rcu_assign_pointer(nfsi->delegation, delegation);
 511         delegation = NULL;
 512
 513         atomic_long_inc(&nfs_active_delegations);
 514
 515         trace_nfs4_set_delegation(inode, type);
 516 out:
 517         spin_unlock(&clp->cl_lock);
 518         if (delegation != NULL)
 519                 __nfs_free_delegation(delegation);
 520         if (freeme != NULL) {
 521                 nfs_do_return_delegation(inode, freeme, 0);
 522                 nfs_free_delegation(freeme);
 523         }
 524         return status;
 525 }
 526
 527 /*
 528  * Basic procedure for returning a delegation to the server
 529  */
 530 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
 531 {
 532         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 533         int err = 0;
 534
 535         if (delegation == NULL)
 536                 return 0;
 537         do {
 538                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 539                         break;
 540                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
 541                                 delegation->type);
 542                 if (!issync || err != -EAGAIN)
 543                         break;
 544                 /*
 545                  * Guard against state recovery
 546                  */
 547                 err = nfs4_wait_clnt_recover(clp);
 548         } while (err == 0);
 549
 550         if (err) {
 551                 nfs_abort_delegation_return(delegation, clp, err);
 552                 goto out;
 553         }
 554
 555         err = nfs_do_return_delegation(inode, delegation, issync);
 556 out:
 557         /* Refcount matched in nfs_start_delegation_return_locked() */
 558         nfs_put_delegation(delegation);
 559         return err;
 560 }
 561
 562 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
 563 {
 564         bool ret = false;
 565
 566         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 567                 ret = true;
 568         else if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags)) {
 569                 struct inode *inode;
 570
 571                 spin_lock(&delegation->lock);
 572                 inode = delegation->inode;
 573                 if (inode && list_empty(&NFS_I(inode)->open_files))
 574                         ret = true;
 575                 spin_unlock(&delegation->lock);
 576         }
 577         if (ret)
 578                 clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 579         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
 580             test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
 581             test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 582                 ret = false;
 583
 584         return ret;
 585 }
 586
 587 static int nfs_server_return_marked_delegations(struct nfs_server *server,
 588                 void __always_unused *data)
 589 {
 590         struct nfs_delegation *delegation;
 591         struct nfs_delegation *prev;
 592         struct inode *inode;
 593         struct inode *place_holder = NULL;
 594         struct nfs_delegation *place_holder_deleg = NULL;
 595         int err = 0;
 596
 597 restart:
 598         /*
 599          * To avoid quadratic looping we hold a reference
 600          * to an inode place_holder.  Each time we restart, we
 601          * list delegation in the server from the delegations
 602          * of that inode.
 603          * prev is an RCU-protected pointer to a delegation which
 604          * wasn't marked for return and might be a good choice for
 605          * the next place_holder.
 606          */
 607         prev = NULL;
 608         delegation = NULL;
 609         rcu_read_lock();
 610         if (place_holder)
 611                 delegation = rcu_dereference(NFS_I(place_holder)->delegation);
 612         if (!delegation || delegation != place_holder_deleg)
 613                 delegation = list_entry_rcu(server->delegations.next,
 614                                             struct nfs_delegation, super_list);
 615         list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
 616                 struct inode *to_put = NULL;
 617
 618                 if (test_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags))
 619                         continue;
 620                 if (!nfs_delegation_need_return(delegation)) {
 621                         if (nfs4_is_valid_delegation(delegation, 0))
 622                                 prev = delegation;
 623                         continue;
 624                 }
 625
 626                 if (prev) {
 627                         struct inode *tmp = nfs_delegation_grab_inode(prev);
 628                         if (tmp) {
 629                                 to_put = place_holder;
 630                                 place_holder = tmp;
 631                                 place_holder_deleg = prev;
 632                         }
 633                 }
 634
 635                 inode = nfs_delegation_grab_inode(delegation);
 636                 if (inode == NULL) {
 637                         rcu_read_unlock();
 638                         iput(to_put);
 639                         goto restart;
 640                 }
 641                 delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 642                 rcu_read_unlock();
 643
 644                 iput(to_put);
 645
 646                 err = nfs_end_delegation_return(inode, delegation, 0);
 647                 iput(inode);
 648                 cond_resched();
 649                 if (!err)
 650                         goto restart;
 651                 set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 652                 goto out;
 653         }
 654         rcu_read_unlock();
 655 out:
 656         iput(place_holder);
 657         return err;
 658 }
 659
 660 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
 661 {
 662         struct nfs_delegation *d;
 663         bool ret = false;
 664
 665         list_for_each_entry_rcu (d, &server->delegations, super_list) {
 666                 if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
 667                         continue;
 668                 nfs_mark_return_delegation(server, d);
 669                 clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
 670                 ret = true;
 671         }
 672         return ret;
 673 }
 674
 675 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
 676 {
 677         struct nfs_server *server;
 678         bool ret = false;
 679
 680         if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
 681                 goto out;
 682         rcu_read_lock();
 683         list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
 684                 if (nfs_server_clear_delayed_delegations(server))
 685                         ret = true;
 686         }
 687         rcu_read_unlock();
 688 out:
 689         return ret;
 690 }
 691
 692 /**
 693  * nfs_client_return_marked_delegations - return previously marked delegations
 694  * @clp: nfs_client to process
 695  *
 696  * Note that this function is designed to be called by the state
 697  * manager thread. For this reason, it cannot flush the dirty data,
 698  * since that could deadlock in case of a state recovery error.
 699  *
 700  * Returns zero on success, or a negative errno value.
 701  */
 702 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 703 {
 704         int err = nfs_client_for_each_server(
 705                 clp, nfs_server_return_marked_delegations, NULL);
 706         if (err)
 707                 return err;
 708         /* If a return was delayed, sleep to prevent hard looping */
 709         if (nfs_client_clear_delayed_delegations(clp))
 710                 ssleep(1);
 711         return 0;
 712 }
 713
 714 /**
 715  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
 716  * @inode: inode to process
 717  *
 718  * Does not protect against delegation reclaims, therefore really only safe
 719  * to be called from nfs4_clear_inode(). Guaranteed to always free
 720  * the delegation structure.
 721  */
 722 void nfs_inode_evict_delegation(struct inode *inode)
 723 {
 724         struct nfs_delegation *delegation;
 725
 726         delegation = nfs_inode_detach_delegation(inode);
 727         if (delegation != NULL) {
 728                 set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 729                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 730                 nfs_do_return_delegation(inode, delegation, 1);
 731                 nfs_free_delegation(delegation);
 732         }
 733 }
 734
 735 /**
 736  * nfs4_inode_return_delegation - synchronously return a delegation
 737  * @inode: inode to process
 738  *
 739  * This routine will always flush any dirty data to disk on the
 740  * assumption that if we need to return the delegation, then
 741  * we should stop caching.
 742  *
 743  * Returns zero on success, or a negative errno value.
 744  */
 745 int nfs4_inode_return_delegation(struct inode *inode)
 746 {
 747         struct nfs_inode *nfsi = NFS_I(inode);
 748         struct nfs_delegation *delegation;
 749         int err = 0;
 750
 751         nfs_wb_all(inode);
 752         delegation = nfs_start_delegation_return(nfsi);
 753         if (delegation != NULL)
 754                 err = nfs_end_delegation_return(inode, delegation, 1);
 755         return err;
 756 }
 757
 758 /**
 759  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
 760  * @inode: inode to process
 761  *
 762  * This routine is called on file close in order to determine if the
 763  * inode delegation needs to be returned immediately.
 764  */
 765 void nfs4_inode_return_delegation_on_close(struct inode *inode)
 766 {
 767         struct nfs_delegation *delegation;
 768         struct nfs_delegation *ret = NULL;
 769
 770         if (!inode)
 771                 return;
 772         rcu_read_lock();
 773         delegation = nfs4_get_valid_delegation(inode);
 774         if (!delegation)
 775                 goto out;
 776         if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
 777             atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
 778                 spin_lock(&delegation->lock);
 779                 if (delegation->inode &&
 780                     list_empty(&NFS_I(inode)->open_files) &&
 781                     !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
 782                         clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 783                         /* Refcount matched in nfs_end_delegation_return() */
 784                         ret = nfs_get_delegation(delegation);
 785                 }
 786                 spin_unlock(&delegation->lock);
 787                 if (ret)
 788                         nfs_clear_verifier_delegated(inode);
 789         }
 790 out:
 791         rcu_read_unlock();
 792         nfs_end_delegation_return(inode, ret, 0);
 793 }
 794
 795 /**
 796  * nfs4_inode_make_writeable
 797  * @inode: pointer to inode
 798  *
 799  * Make the inode writeable by returning the delegation if necessary
 800  *
 801  * Returns zero on success, or a negative errno value.
 802  */
 803 int nfs4_inode_make_writeable(struct inode *inode)
 804 {
 805         struct nfs_delegation *delegation;
 806
 807         rcu_read_lock();
 808         delegation = nfs4_get_valid_delegation(inode);
 809         if (delegation == NULL ||
 810             (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
 811              (delegation->type & FMODE_WRITE))) {
 812                 rcu_read_unlock();
 813                 return 0;
 814         }
 815         rcu_read_unlock();
 816         return nfs4_inode_return_delegation(inode);
 817 }
 818
 819 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
 820                 struct nfs_delegation *delegation)
 821 {
 822         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 823         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 824 }
 825
 826 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
 827 {
 828         struct nfs_delegation *delegation;
 829         bool ret = false;
 830
 831         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 832                 nfs_mark_return_delegation(server, delegation);
 833                 ret = true;
 834         }
 835         return ret;
 836 }
 837
 838 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 839 {
 840         struct nfs_server *server;
 841
 842         rcu_read_lock();
 843         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 844                 nfs_server_mark_return_all_delegations(server);
 845         rcu_read_unlock();
 846 }
 847
 848 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 849 {
 850         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 851                 nfs4_schedule_state_manager(clp);
 852 }
 853
 854 /**
 855  * nfs_expire_all_delegations
 856  * @clp: client to process
 857  *
 858  */
 859 void nfs_expire_all_delegations(struct nfs_client *clp)
 860 {
 861         nfs_client_mark_return_all_delegations(clp);
 862         nfs_delegation_run_state_manager(clp);
 863 }
 864
 865 /**
 866  * nfs_server_return_all_delegations - return delegations for one superblock
 867  * @server: pointer to nfs_server to process
 868  *
 869  */
 870 void nfs_server_return_all_delegations(struct nfs_server *server)
 871 {
 872         struct nfs_client *clp = server->nfs_client;
 873         bool need_wait;
 874
 875         if (clp == NULL)
 876                 return;
 877
 878         rcu_read_lock();
 879         need_wait = nfs_server_mark_return_all_delegations(server);
 880         rcu_read_unlock();
 881
 882         if (need_wait) {
 883                 nfs4_schedule_state_manager(clp);
 884                 nfs4_wait_clnt_recover(clp);
 885         }
 886 }
 887
 888 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
 889                                                  fmode_t flags)
 890 {
 891         struct nfs_delegation *delegation;
 892
 893         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 894                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 895                         continue;
 896                 if (delegation->type & flags)
 897                         nfs_mark_return_if_closed_delegation(server, delegation);
 898         }
 899 }
 900
 901 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
 902                                                         fmode_t flags)
 903 {
 904         struct nfs_server *server;
 905
 906         rcu_read_lock();
 907         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 908                 nfs_mark_return_unused_delegation_types(server, flags);
 909         rcu_read_unlock();
 910 }
 911
 912 static void nfs_revoke_delegation(struct inode *inode,
 913                 const nfs4_stateid *stateid)
 914 {
 915         struct nfs_delegation *delegation;
 916         nfs4_stateid tmp;
 917         bool ret = false;
 918
 919         rcu_read_lock();
 920         delegation = rcu_dereference(NFS_I(inode)->delegation);
 921         if (delegation == NULL)
 922                 goto out;
 923         if (stateid == NULL) {
 924                 nfs4_stateid_copy(&tmp, &delegation->stateid);
 925                 stateid = &tmp;
 926         } else {
 927                 if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
 928                         goto out;
 929                 spin_lock(&delegation->lock);
 930                 if (stateid->seqid) {
 931                         if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
 932                                 spin_unlock(&delegation->lock);
 933                                 goto out;
 934                         }
 935                         delegation->stateid.seqid = stateid->seqid;
 936                 }
 937                 spin_unlock(&delegation->lock);
 938         }
 939         nfs_mark_delegation_revoked(delegation);
 940         ret = true;
 941 out:
 942         rcu_read_unlock();
 943         if (ret)
 944                 nfs_inode_find_state_and_recover(inode, stateid);
 945 }
 946
 947 void nfs_remove_bad_delegation(struct inode *inode,
 948                 const nfs4_stateid *stateid)
 949 {
 950         nfs_revoke_delegation(inode, stateid);
 951 }
 952 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 953
 954 void nfs_delegation_mark_returned(struct inode *inode,
 955                 const nfs4_stateid *stateid)
 956 {
 957         struct nfs_delegation *delegation;
 958
 959         if (!inode)
 960                 return;
 961
 962         rcu_read_lock();
 963         delegation = rcu_dereference(NFS_I(inode)->delegation);
 964         if (!delegation)
 965                 goto out_rcu_unlock;
 966
 967         spin_lock(&delegation->lock);
 968         if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
 969                 goto out_spin_unlock;
 970         if (stateid->seqid) {
 971                 /* If delegation->stateid is newer, dont mark as returned */
 972                 if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
 973                         goto out_clear_returning;
 974                 if (delegation->stateid.seqid != stateid->seqid)
 975                         delegation->stateid.seqid = stateid->seqid;
 976         }
 977
 978         nfs_mark_delegation_revoked(delegation);
 979
 980 out_clear_returning:
 981         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 982 out_spin_unlock:
 983         spin_unlock(&delegation->lock);
 984 out_rcu_unlock:
 985         rcu_read_unlock();
 986
 987         nfs_inode_find_state_and_recover(inode, stateid);
 988 }
 989
 990 /**
 991  * nfs_expire_unused_delegation_types
 992  * @clp: client to process
 993  * @flags: delegation types to expire
 994  *
 995  */
 996 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
 997 {
 998         nfs_client_mark_return_unused_delegation_types(clp, flags);
 999         nfs_delegation_run_state_manager(clp);
1000 }
1001
1002 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1003 {
1004         struct nfs_delegation *delegation;
1005
1006         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1007                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1008                         continue;
1009                 nfs_mark_return_if_closed_delegation(server, delegation);
1010         }
1011 }
1012
1013 /**
1014  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1015  * @clp: nfs_client to process
1016  *
1017  */
1018 void nfs_expire_unreferenced_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_mark_return_unreferenced_delegations(server);
1025         rcu_read_unlock();
1026
1027         nfs_delegation_run_state_manager(clp);
1028 }
1029
1030 /**
1031  * nfs_async_inode_return_delegation - asynchronously return a delegation
1032  * @inode: inode to process
1033  * @stateid: state ID information
1034  *
1035  * Returns zero on success, or a negative errno value.
1036  */
1037 int nfs_async_inode_return_delegation(struct inode *inode,
1038                                       const nfs4_stateid *stateid)
1039 {
1040         struct nfs_server *server = NFS_SERVER(inode);
1041         struct nfs_client *clp = server->nfs_client;
1042         struct nfs_delegation *delegation;
1043
1044         rcu_read_lock();
1045         delegation = nfs4_get_valid_delegation(inode);
1046         if (delegation == NULL)
1047                 goto out_enoent;
1048         if (stateid != NULL &&
1049             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1050                 goto out_enoent;
1051         nfs_mark_return_delegation(server, delegation);
1052         rcu_read_unlock();
1053
1054         nfs_delegation_run_state_manager(clp);
1055         return 0;
1056 out_enoent:
1057         rcu_read_unlock();
1058         return -ENOENT;
1059 }
1060
1061 static struct inode *
1062 nfs_delegation_find_inode_server(struct nfs_server *server,
1063                                  const struct nfs_fh *fhandle)
1064 {
1065         struct nfs_delegation *delegation;
1066         struct super_block *freeme = NULL;
1067         struct inode *res = NULL;
1068
1069         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1070                 spin_lock(&delegation->lock);
1071                 if (delegation->inode != NULL &&
1072                     !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1073                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1074                         if (nfs_sb_active(server->super)) {
1075                                 freeme = server->super;
1076                                 res = igrab(delegation->inode);
1077                         }
1078                         spin_unlock(&delegation->lock);
1079                         if (res != NULL)
1080                                 return res;
1081                         if (freeme) {
1082                                 rcu_read_unlock();
1083                                 nfs_sb_deactive(freeme);
1084                                 rcu_read_lock();
1085                         }
1086                         return ERR_PTR(-EAGAIN);
1087                 }
1088                 spin_unlock(&delegation->lock);
1089         }
1090         return ERR_PTR(-ENOENT);
1091 }
1092
1093 /**
1094  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1095  * @clp: client state handle
1096  * @fhandle: filehandle from a delegation recall
1097  *
1098  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1099  * cannot be found.
1100  */
1101 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1102                                         const struct nfs_fh *fhandle)
1103 {
1104         struct nfs_server *server;
1105         struct inode *res;
1106
1107         rcu_read_lock();
1108         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1109                 res = nfs_delegation_find_inode_server(server, fhandle);
1110                 if (res != ERR_PTR(-ENOENT)) {
1111                         rcu_read_unlock();
1112                         return res;
1113                 }
1114         }
1115         rcu_read_unlock();
1116         return ERR_PTR(-ENOENT);
1117 }
1118
1119 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1120 {
1121         struct nfs_delegation *delegation;
1122
1123         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1124                 /*
1125                  * If the delegation may have been admin revoked, then we
1126                  * cannot reclaim it.
1127                  */
1128                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1129                         continue;
1130                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1131         }
1132 }
1133
1134 /**
1135  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1136  * @clp: nfs_client to process
1137  *
1138  */
1139 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1140 {
1141         struct nfs_server *server;
1142
1143         rcu_read_lock();
1144         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1145                 nfs_delegation_mark_reclaim_server(server);
1146         rcu_read_unlock();
1147 }
1148
1149 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1150                 void __always_unused *data)
1151 {
1152         struct nfs_delegation *delegation;
1153         struct inode *inode;
1154 restart:
1155         rcu_read_lock();
1156 restart_locked:
1157         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1158                 if (test_bit(NFS_DELEGATION_INODE_FREEING,
1159                                         &delegation->flags) ||
1160                     test_bit(NFS_DELEGATION_RETURNING,
1161                                         &delegation->flags) ||
1162                     test_bit(NFS_DELEGATION_NEED_RECLAIM,
1163                                         &delegation->flags) == 0)
1164                         continue;
1165                 inode = nfs_delegation_grab_inode(delegation);
1166                 if (inode == NULL)
1167                         goto restart_locked;
1168                 delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1169                 rcu_read_unlock();
1170                 if (delegation != NULL) {
1171                         if (nfs_detach_delegation(NFS_I(inode), delegation,
1172                                                 server) != NULL)
1173                                 nfs_free_delegation(delegation);
1174                         /* Match nfs_start_delegation_return_locked */
1175                         nfs_put_delegation(delegation);
1176                 }
1177                 iput(inode);
1178                 cond_resched();
1179                 goto restart;
1180         }
1181         rcu_read_unlock();
1182         return 0;
1183 }
1184
1185 /**
1186  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1187  * @clp: nfs_client to process
1188  *
1189  */
1190 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1191 {
1192         nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1193                         NULL);
1194 }
1195
1196 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1197 {
1198         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1199                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
1200                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
1201 }
1202
1203 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1204             struct nfs_delegation *delegation)
1205 {
1206         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1207                 return;
1208         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1209         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1210         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1211 }
1212
1213 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1214                 struct inode *inode)
1215 {
1216         struct nfs_delegation *delegation;
1217
1218         rcu_read_lock();
1219         delegation = rcu_dereference(NFS_I(inode)->delegation);
1220         if (delegation)
1221                 nfs_mark_test_expired_delegation(server, delegation);
1222         rcu_read_unlock();
1223
1224 }
1225
1226 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1227 {
1228         struct nfs_delegation *delegation;
1229
1230         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1231                 nfs_mark_test_expired_delegation(server, delegation);
1232 }
1233
1234 /**
1235  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1236  * @clp: nfs_client to process
1237  *
1238  * Iterates through all the delegations associated with this server and
1239  * marks them as needing to be checked for validity.
1240  */
1241 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1242 {
1243         struct nfs_server *server;
1244
1245         rcu_read_lock();
1246         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1247                 nfs_delegation_mark_test_expired_server(server);
1248         rcu_read_unlock();
1249 }
1250
1251 /**
1252  * nfs_test_expired_all_delegations - test all delegations for a client
1253  * @clp: nfs_client to process
1254  *
1255  * Helper for handling "recallable state revoked" status from server.
1256  */
1257 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1258 {
1259         nfs_mark_test_expired_all_delegations(clp);
1260         nfs4_schedule_state_manager(clp);
1261 }
1262
1263 static void
1264 nfs_delegation_test_free_expired(struct inode *inode,
1265                 nfs4_stateid *stateid,
1266                 const struct cred *cred)
1267 {
1268         struct nfs_server *server = NFS_SERVER(inode);
1269         const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1270         int status;
1271
1272         if (!cred)
1273                 return;
1274         status = ops->test_and_free_expired(server, stateid, cred);
1275         if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1276                 nfs_remove_bad_delegation(inode, stateid);
1277 }
1278
1279 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1280                 void __always_unused *data)
1281 {
1282         struct nfs_delegation *delegation;
1283         struct inode *inode;
1284         const struct cred *cred;
1285         nfs4_stateid stateid;
1286 restart:
1287         rcu_read_lock();
1288 restart_locked:
1289         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1290                 if (test_bit(NFS_DELEGATION_INODE_FREEING,
1291                                         &delegation->flags) ||
1292                     test_bit(NFS_DELEGATION_RETURNING,
1293                                         &delegation->flags) ||
1294                     test_bit(NFS_DELEGATION_TEST_EXPIRED,
1295                                         &delegation->flags) == 0)
1296                         continue;
1297                 inode = nfs_delegation_grab_inode(delegation);
1298                 if (inode == NULL)
1299                         goto restart_locked;
1300                 spin_lock(&delegation->lock);
1301                 cred = get_cred_rcu(delegation->cred);
1302                 nfs4_stateid_copy(&stateid, &delegation->stateid);
1303                 spin_unlock(&delegation->lock);
1304                 clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1305                 rcu_read_unlock();
1306                 nfs_delegation_test_free_expired(inode, &stateid, cred);
1307                 put_cred(cred);
1308                 if (!nfs4_server_rebooted(server->nfs_client)) {
1309                         iput(inode);
1310                         cond_resched();
1311                         goto restart;
1312                 }
1313                 nfs_inode_mark_test_expired_delegation(server,inode);
1314                 iput(inode);
1315                 return -EAGAIN;
1316         }
1317         rcu_read_unlock();
1318         return 0;
1319 }
1320
1321 /**
1322  * nfs_reap_expired_delegations - reap expired delegations
1323  * @clp: nfs_client to process
1324  *
1325  * Iterates through all the delegations associated with this server and
1326  * checks if they have may have been revoked. This function is usually
1327  * expected to be called in cases where the server may have lost its
1328  * lease.
1329  */
1330 void nfs_reap_expired_delegations(struct nfs_client *clp)
1331 {
1332         nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1333                         NULL);
1334 }
1335
1336 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1337                 const nfs4_stateid *stateid)
1338 {
1339         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1340         struct nfs_delegation *delegation;
1341         bool found = false;
1342
1343         rcu_read_lock();
1344         delegation = rcu_dereference(NFS_I(inode)->delegation);
1345         if (delegation &&
1346             nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1347             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1348                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1349                 found = true;
1350         }
1351         rcu_read_unlock();
1352         if (found)
1353                 nfs4_schedule_state_manager(clp);
1354 }
1355
1356 /**
1357  * nfs_delegations_present - check for existence of delegations
1358  * @clp: client state handle
1359  *
1360  * Returns one if there are any nfs_delegation structures attached
1361  * to this nfs_client.
1362  */
1363 int nfs_delegations_present(struct nfs_client *clp)
1364 {
1365         struct nfs_server *server;
1366         int ret = 0;
1367
1368         rcu_read_lock();
1369         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1370                 if (!list_empty(&server->delegations)) {
1371                         ret = 1;
1372                         break;
1373                 }
1374         rcu_read_unlock();
1375         return ret;
1376 }
1377
1378 /**
1379  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1380  * @dst: stateid to refresh
1381  * @inode: inode to check
1382  *
1383  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1384  * that matches our delegation stateid. Otherwise "false" is returned.
1385  */
1386 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1387 {
1388         struct nfs_delegation *delegation;
1389         bool ret = false;
1390         if (!inode)
1391                 goto out;
1392
1393         rcu_read_lock();
1394         delegation = rcu_dereference(NFS_I(inode)->delegation);
1395         if (delegation != NULL &&
1396             nfs4_stateid_match_other(dst, &delegation->stateid) &&
1397             nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1398             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1399                 dst->seqid = delegation->stateid.seqid;
1400                 ret = true;
1401         }
1402         rcu_read_unlock();
1403 out:
1404         return ret;
1405 }
1406
1407 /**
1408  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1409  * @inode: inode to check
1410  * @flags: delegation type requirement
1411  * @dst: stateid data structure to fill in
1412  * @cred: optional argument to retrieve credential
1413  *
1414  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1415  * otherwise "false" is returned.
1416  */
1417 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1418                 nfs4_stateid *dst, const struct cred **cred)
1419 {
1420         struct nfs_inode *nfsi = NFS_I(inode);
1421         struct nfs_delegation *delegation;
1422         bool ret = false;
1423
1424         flags &= FMODE_READ|FMODE_WRITE;
1425         rcu_read_lock();
1426         delegation = rcu_dereference(nfsi->delegation);
1427         if (!delegation)
1428                 goto out;
1429         spin_lock(&delegation->lock);
1430         ret = nfs4_is_valid_delegation(delegation, flags);
1431         if (ret) {
1432                 nfs4_stateid_copy(dst, &delegation->stateid);
1433                 nfs_mark_delegation_referenced(delegation);
1434                 if (cred)
1435                         *cred = get_cred(delegation->cred);
1436         }
1437         spin_unlock(&delegation->lock);
1438 out:
1439         rcu_read_unlock();
1440         return ret;
1441 }
1442
1443 /**
1444  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1445  * @inode: inode to check
1446  *
1447  * This function checks the number of outstanding writes to the file
1448  * against the delegation 'space_limit' field to see if
1449  * the spec requires us to flush the file on close.
1450  */
1451 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1452 {
1453         struct nfs_inode *nfsi = NFS_I(inode);
1454         struct nfs_delegation *delegation;
1455         bool ret = true;
1456
1457         rcu_read_lock();
1458         delegation = rcu_dereference(nfsi->delegation);
1459         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1460                 goto out;
1461         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1462                 ret = false;
1463 out:
1464         rcu_read_unlock();
1465         return ret;
1466 }
1467
1468 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);