fs/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 = locks_inode_context(inode);
 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         for_each_file_lock(fl, list) {
 160                 if (nfs_file_open_context(fl->c.flc_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         int err;
 185
 186 again:
 187         rcu_read_lock();
 188         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
 189                 state = ctx->state;
 190                 if (state == NULL)
 191                         continue;
 192                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 193                         continue;
 194                 if (!nfs4_valid_open_stateid(state))
 195                         continue;
 196                 if (!nfs4_stateid_match(&state->stateid, stateid))
 197                         continue;
 198                 if (!get_nfs_open_context(ctx))
 199                         continue;
 200                 rcu_read_unlock();
 201                 sp = state->owner;
 202                 /* Block nfs4_proc_unlck */
 203                 mutex_lock(&sp->so_delegreturn_mutex);
 204                 err = nfs4_open_delegation_recall(ctx, state, stateid);
 205                 if (!err)
 206                         err = nfs_delegation_claim_locks(state, stateid);
 207                 mutex_unlock(&sp->so_delegreturn_mutex);
 208                 put_nfs_open_context(ctx);
 209                 if (err != 0)
 210                         return err;
 211                 goto again;
 212         }
 213         rcu_read_unlock();
 214         return 0;
 215 }
 216
 217 /**
 218  * nfs_inode_reclaim_delegation - process a delegation reclaim request
 219  * @inode: inode to process
 220  * @cred: credential to use for request
 221  * @type: delegation type
 222  * @stateid: delegation stateid
 223  * @pagemod_limit: write delegation "space_limit"
 224  *
 225  */
 226 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
 227                                   fmode_t type, const nfs4_stateid *stateid,
 228                                   unsigned long pagemod_limit)
 229 {
 230         struct nfs_delegation *delegation;
 231         const struct cred *oldcred = NULL;
 232
 233         rcu_read_lock();
 234         delegation = rcu_dereference(NFS_I(inode)->delegation);
 235         if (delegation != NULL) {
 236                 spin_lock(&delegation->lock);
 237                 nfs4_stateid_copy(&delegation->stateid, stateid);
 238                 delegation->type = type;
 239                 delegation->pagemod_limit = pagemod_limit;
 240                 oldcred = delegation->cred;
 241                 delegation->cred = get_cred(cred);
 242                 clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 243                 if (test_and_clear_bit(NFS_DELEGATION_REVOKED,
 244                                        &delegation->flags))
 245                         atomic_long_inc(&nfs_active_delegations);
 246                 spin_unlock(&delegation->lock);
 247                 rcu_read_unlock();
 248                 put_cred(oldcred);
 249                 trace_nfs4_reclaim_delegation(inode, type);
 250         } else {
 251                 rcu_read_unlock();
 252                 nfs_inode_set_delegation(inode, cred, type, stateid,
 253                                          pagemod_limit);
 254         }
 255 }
 256
 257 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 258 {
 259         const struct cred *cred;
 260         int res = 0;
 261
 262         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
 263                 spin_lock(&delegation->lock);
 264                 cred = get_cred(delegation->cred);
 265                 spin_unlock(&delegation->lock);
 266                 res = nfs4_proc_delegreturn(inode, cred,
 267                                 &delegation->stateid,
 268                                 issync);
 269                 put_cred(cred);
 270         }
 271         return res;
 272 }
 273
 274 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 275 {
 276         struct inode *inode = NULL;
 277
 278         spin_lock(&delegation->lock);
 279         if (delegation->inode != NULL)
 280                 inode = igrab(delegation->inode);
 281         if (!inode)
 282                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 283         spin_unlock(&delegation->lock);
 284         return inode;
 285 }
 286
 287 static struct nfs_delegation *
 288 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
 289 {
 290         struct nfs_delegation *ret = NULL;
 291         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
 292
 293         if (delegation == NULL)
 294                 goto out;
 295         spin_lock(&delegation->lock);
 296         if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
 297                 clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
 298                 /* Refcount matched in nfs_end_delegation_return() */
 299                 ret = nfs_get_delegation(delegation);
 300         }
 301         spin_unlock(&delegation->lock);
 302         if (ret)
 303                 nfs_clear_verifier_delegated(&nfsi->vfs_inode);
 304 out:
 305         return ret;
 306 }
 307
 308 static struct nfs_delegation *
 309 nfs_start_delegation_return(struct nfs_inode *nfsi)
 310 {
 311         struct nfs_delegation *delegation;
 312
 313         rcu_read_lock();
 314         delegation = nfs_start_delegation_return_locked(nfsi);
 315         rcu_read_unlock();
 316         return delegation;
 317 }
 318
 319 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
 320                                         struct nfs_client *clp, int err)
 321 {
 322
 323         spin_lock(&delegation->lock);
 324         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 325         if (err == -EAGAIN) {
 326                 set_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
 327                 set_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state);
 328         }
 329         spin_unlock(&delegation->lock);
 330 }
 331
 332 static struct nfs_delegation *
 333 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
 334                 struct nfs_delegation *delegation,
 335                 struct nfs_client *clp)
 336 {
 337         struct nfs_delegation *deleg_cur =
 338                 rcu_dereference_protected(nfsi->delegation,
 339                                 lockdep_is_held(&clp->cl_lock));
 340
 341         if (deleg_cur == NULL || delegation != deleg_cur)
 342                 return NULL;
 343
 344         spin_lock(&delegation->lock);
 345         if (!delegation->inode) {
 346                 spin_unlock(&delegation->lock);
 347                 return NULL;
 348         }
 349         list_del_rcu(&delegation->super_list);
 350         delegation->inode = NULL;
 351         rcu_assign_pointer(nfsi->delegation, NULL);
 352         spin_unlock(&delegation->lock);
 353         return delegation;
 354 }
 355
 356 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
 357                 struct nfs_delegation *delegation,
 358                 struct nfs_server *server)
 359 {
 360         struct nfs_client *clp = server->nfs_client;
 361
 362         spin_lock(&clp->cl_lock);
 363         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
 364         spin_unlock(&clp->cl_lock);
 365         return delegation;
 366 }
 367
 368 static struct nfs_delegation *
 369 nfs_inode_detach_delegation(struct inode *inode)
 370 {
 371         struct nfs_inode *nfsi = NFS_I(inode);
 372         struct nfs_server *server = NFS_SERVER(inode);
 373         struct nfs_delegation *delegation;
 374
 375         rcu_read_lock();
 376         delegation = rcu_dereference(nfsi->delegation);
 377         if (delegation != NULL)
 378                 delegation = nfs_detach_delegation(nfsi, delegation, server);
 379         rcu_read_unlock();
 380         return delegation;
 381 }
 382
 383 static void
 384 nfs_update_delegation_cred(struct nfs_delegation *delegation,
 385                 const struct cred *cred)
 386 {
 387         const struct cred *old;
 388
 389         if (cred_fscmp(delegation->cred, cred) != 0) {
 390                 old = xchg(&delegation->cred, get_cred(cred));
 391                 put_cred(old);
 392         }
 393 }
 394
 395 static void
 396 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
 397                 const struct nfs_delegation *update)
 398 {
 399         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
 400                 delegation->stateid.seqid = update->stateid.seqid;
 401                 smp_wmb();
 402                 delegation->type = update->type;
 403                 delegation->pagemod_limit = update->pagemod_limit;
 404                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
 405                         delegation->change_attr = update->change_attr;
 406                         nfs_update_delegation_cred(delegation, update->cred);
 407                         /* smp_mb__before_atomic() is implicit due to xchg() */
 408                         clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 409                         atomic_long_inc(&nfs_active_delegations);
 410                 }
 411         }
 412 }
 413
 414 /**
 415  * nfs_inode_set_delegation - set up a delegation on an inode
 416  * @inode: inode to which delegation applies
 417  * @cred: cred to use for subsequent delegation processing
 418  * @type: delegation type
 419  * @stateid: delegation stateid
 420  * @pagemod_limit: write delegation "space_limit"
 421  *
 422  * Returns zero on success, or a negative errno value.
 423  */
 424 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
 425                                   fmode_t type,
 426                                   const nfs4_stateid *stateid,
 427                                   unsigned long pagemod_limit)
 428 {
 429         struct nfs_server *server = NFS_SERVER(inode);
 430         struct nfs_client *clp = server->nfs_client;
 431         struct nfs_inode *nfsi = NFS_I(inode);
 432         struct nfs_delegation *delegation, *old_delegation;
 433         struct nfs_delegation *freeme = NULL;
 434         int status = 0;
 435
 436         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL_ACCOUNT);
 437         if (delegation == NULL)
 438                 return -ENOMEM;
 439         nfs4_stateid_copy(&delegation->stateid, stateid);
 440         refcount_set(&delegation->refcount, 1);
 441         delegation->type = type;
 442         delegation->pagemod_limit = pagemod_limit;
 443         delegation->change_attr = inode_peek_iversion_raw(inode);
 444         delegation->cred = get_cred(cred);
 445         delegation->inode = inode;
 446         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 447         delegation->test_gen = 0;
 448         spin_lock_init(&delegation->lock);
 449
 450         spin_lock(&clp->cl_lock);
 451         old_delegation = rcu_dereference_protected(nfsi->delegation,
 452                                         lockdep_is_held(&clp->cl_lock));
 453         if (old_delegation == NULL)
 454                 goto add_new;
 455         /* Is this an update of the existing delegation? */
 456         if (nfs4_stateid_match_other(&old_delegation->stateid,
 457                                 &delegation->stateid)) {
 458                 spin_lock(&old_delegation->lock);
 459                 nfs_update_inplace_delegation(old_delegation,
 460                                 delegation);
 461                 spin_unlock(&old_delegation->lock);
 462                 goto out;
 463         }
 464         if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
 465                 /*
 466                  * Deal with broken servers that hand out two
 467                  * delegations for the same file.
 468                  * Allow for upgrades to a WRITE delegation, but
 469                  * nothing else.
 470                  */
 471                 dfprintk(FILE, "%s: server %s handed out "
 472                                 "a duplicate delegation!\n",
 473                                 __func__, clp->cl_hostname);
 474                 if (delegation->type == old_delegation->type ||
 475                     !(delegation->type & FMODE_WRITE)) {
 476                         freeme = delegation;
 477                         delegation = NULL;
 478                         goto out;
 479                 }
 480                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
 481                                         &old_delegation->flags))
 482                         goto out;
 483         }
 484         freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
 485         if (freeme == NULL)
 486                 goto out;
 487 add_new:
 488         /*
 489          * If we didn't revalidate the change attribute before setting
 490          * the delegation, then pre-emptively ask for a full attribute
 491          * cache revalidation.
 492          */
 493         spin_lock(&inode->i_lock);
 494         if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
 495                 nfs_set_cache_invalid(inode,
 496                         NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
 497                         NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
 498                         NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
 499                         NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
 500                         NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
 501                         NFS_INO_INVALID_XATTR);
 502         spin_unlock(&inode->i_lock);
 503
 504         list_add_tail_rcu(&delegation->super_list, &server->delegations);
 505         rcu_assign_pointer(nfsi->delegation, delegation);
 506         delegation = NULL;
 507
 508         atomic_long_inc(&nfs_active_delegations);
 509
 510         trace_nfs4_set_delegation(inode, type);
 511 out:
 512         spin_unlock(&clp->cl_lock);
 513         if (delegation != NULL)
 514                 __nfs_free_delegation(delegation);
 515         if (freeme != NULL) {
 516                 nfs_do_return_delegation(inode, freeme, 0);
 517                 nfs_free_delegation(freeme);
 518         }
 519         return status;
 520 }
 521
 522 /*
 523  * Basic procedure for returning a delegation to the server
 524  */
 525 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
 526 {
 527         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 528         unsigned int mode = O_WRONLY | O_RDWR;
 529         int err = 0;
 530
 531         if (delegation == NULL)
 532                 return 0;
 533
 534         if (!issync)
 535                 mode |= O_NONBLOCK;
 536         /* Recall of any remaining application leases */
 537         err = break_lease(inode, mode);
 538
 539         while (err == 0) {
 540                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 541                         break;
 542                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
 543                                 delegation->type);
 544                 if (!issync || err != -EAGAIN)
 545                         break;
 546                 /*
 547                  * Guard against state recovery
 548                  */
 549                 err = nfs4_wait_clnt_recover(clp);
 550         }
 551
 552         if (err) {
 553                 nfs_abort_delegation_return(delegation, clp, err);
 554                 goto out;
 555         }
 556
 557         err = nfs_do_return_delegation(inode, delegation, issync);
 558 out:
 559         /* Refcount matched in nfs_start_delegation_return_locked() */
 560         nfs_put_delegation(delegation);
 561         return err;
 562 }
 563
 564 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
 565 {
 566         bool ret = false;
 567
 568         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 569                 ret = true;
 570         else if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags)) {
 571                 struct inode *inode;
 572
 573                 spin_lock(&delegation->lock);
 574                 inode = delegation->inode;
 575                 if (inode && list_empty(&NFS_I(inode)->open_files))
 576                         ret = true;
 577                 spin_unlock(&delegation->lock);
 578         }
 579         if (ret)
 580                 clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 581         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
 582             test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
 583             test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 584                 ret = false;
 585
 586         return ret;
 587 }
 588
 589 static int nfs_server_return_marked_delegations(struct nfs_server *server,
 590                 void __always_unused *data)
 591 {
 592         struct nfs_delegation *delegation;
 593         struct nfs_delegation *prev;
 594         struct inode *inode;
 595         struct inode *place_holder = NULL;
 596         struct nfs_delegation *place_holder_deleg = NULL;
 597         int err = 0;
 598
 599 restart:
 600         /*
 601          * To avoid quadratic looping we hold a reference
 602          * to an inode place_holder.  Each time we restart, we
 603          * list delegation in the server from the delegations
 604          * of that inode.
 605          * prev is an RCU-protected pointer to a delegation which
 606          * wasn't marked for return and might be a good choice for
 607          * the next place_holder.
 608          */
 609         prev = NULL;
 610         delegation = NULL;
 611         rcu_read_lock();
 612         if (place_holder)
 613                 delegation = rcu_dereference(NFS_I(place_holder)->delegation);
 614         if (!delegation || delegation != place_holder_deleg)
 615                 delegation = list_entry_rcu(server->delegations.next,
 616                                             struct nfs_delegation, super_list);
 617         list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
 618                 struct inode *to_put = NULL;
 619
 620                 if (test_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags))
 621                         continue;
 622                 if (!nfs_delegation_need_return(delegation)) {
 623                         if (nfs4_is_valid_delegation(delegation, 0))
 624                                 prev = delegation;
 625                         continue;
 626                 }
 627
 628                 if (prev) {
 629                         struct inode *tmp = nfs_delegation_grab_inode(prev);
 630                         if (tmp) {
 631                                 to_put = place_holder;
 632                                 place_holder = tmp;
 633                                 place_holder_deleg = prev;
 634                         }
 635                 }
 636
 637                 inode = nfs_delegation_grab_inode(delegation);
 638                 if (inode == NULL) {
 639                         rcu_read_unlock();
 640                         iput(to_put);
 641                         goto restart;
 642                 }
 643                 delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 644                 rcu_read_unlock();
 645
 646                 iput(to_put);
 647
 648                 err = nfs_end_delegation_return(inode, delegation, 0);
 649                 iput(inode);
 650                 cond_resched();
 651                 if (!err)
 652                         goto restart;
 653                 set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 654                 goto out;
 655         }
 656         rcu_read_unlock();
 657 out:
 658         iput(place_holder);
 659         return err;
 660 }
 661
 662 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
 663 {
 664         struct nfs_delegation *d;
 665         bool ret = false;
 666
 667         list_for_each_entry_rcu (d, &server->delegations, super_list) {
 668                 if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
 669                         continue;
 670                 nfs_mark_return_delegation(server, d);
 671                 clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
 672                 ret = true;
 673         }
 674         return ret;
 675 }
 676
 677 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
 678 {
 679         struct nfs_server *server;
 680         bool ret = false;
 681
 682         if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
 683                 goto out;
 684         rcu_read_lock();
 685         list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
 686                 if (nfs_server_clear_delayed_delegations(server))
 687                         ret = true;
 688         }
 689         rcu_read_unlock();
 690 out:
 691         return ret;
 692 }
 693
 694 /**
 695  * nfs_client_return_marked_delegations - return previously marked delegations
 696  * @clp: nfs_client to process
 697  *
 698  * Note that this function is designed to be called by the state
 699  * manager thread. For this reason, it cannot flush the dirty data,
 700  * since that could deadlock in case of a state recovery error.
 701  *
 702  * Returns zero on success, or a negative errno value.
 703  */
 704 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 705 {
 706         int err = nfs_client_for_each_server(
 707                 clp, nfs_server_return_marked_delegations, NULL);
 708         if (err)
 709                 return err;
 710         /* If a return was delayed, sleep to prevent hard looping */
 711         if (nfs_client_clear_delayed_delegations(clp))
 712                 ssleep(1);
 713         return 0;
 714 }
 715
 716 /**
 717  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
 718  * @inode: inode to process
 719  *
 720  * Does not protect against delegation reclaims, therefore really only safe
 721  * to be called from nfs4_clear_inode(). Guaranteed to always free
 722  * the delegation structure.
 723  */
 724 void nfs_inode_evict_delegation(struct inode *inode)
 725 {
 726         struct nfs_delegation *delegation;
 727
 728         delegation = nfs_inode_detach_delegation(inode);
 729         if (delegation != NULL) {
 730                 set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 731                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 732                 nfs_do_return_delegation(inode, delegation, 1);
 733                 nfs_free_delegation(delegation);
 734         }
 735 }
 736
 737 /**
 738  * nfs4_inode_return_delegation - synchronously return a delegation
 739  * @inode: inode to process
 740  *
 741  * This routine will always flush any dirty data to disk on the
 742  * assumption that if we need to return the delegation, then
 743  * we should stop caching.
 744  *
 745  * Returns zero on success, or a negative errno value.
 746  */
 747 int nfs4_inode_return_delegation(struct inode *inode)
 748 {
 749         struct nfs_inode *nfsi = NFS_I(inode);
 750         struct nfs_delegation *delegation;
 751
 752         delegation = nfs_start_delegation_return(nfsi);
 753         if (delegation != NULL) {
 754                 /* Synchronous recall of any application leases */
 755                 break_lease(inode, O_WRONLY | O_RDWR);
 756                 if (S_ISREG(inode->i_mode))
 757                         nfs_wb_all(inode);
 758                 return nfs_end_delegation_return(inode, delegation, 1);
 759         }
 760         return 0;
 761 }
 762
 763 /**
 764  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
 765  * @inode: inode to process
 766  *
 767  * This routine is called on file close in order to determine if the
 768  * inode delegation needs to be returned immediately.
 769  */
 770 void nfs4_inode_return_delegation_on_close(struct inode *inode)
 771 {
 772         struct nfs_delegation *delegation;
 773         struct nfs_delegation *ret = NULL;
 774
 775         if (!inode)
 776                 return;
 777         rcu_read_lock();
 778         delegation = nfs4_get_valid_delegation(inode);
 779         if (!delegation)
 780                 goto out;
 781         if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
 782             atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
 783                 spin_lock(&delegation->lock);
 784                 if (delegation->inode &&
 785                     list_empty(&NFS_I(inode)->open_files) &&
 786                     !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
 787                         clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 788                         /* Refcount matched in nfs_end_delegation_return() */
 789                         ret = nfs_get_delegation(delegation);
 790                 }
 791                 spin_unlock(&delegation->lock);
 792                 if (ret)
 793                         nfs_clear_verifier_delegated(inode);
 794         }
 795 out:
 796         rcu_read_unlock();
 797         nfs_end_delegation_return(inode, ret, 0);
 798 }
 799
 800 /**
 801  * nfs4_inode_make_writeable
 802  * @inode: pointer to inode
 803  *
 804  * Make the inode writeable by returning the delegation if necessary
 805  *
 806  * Returns zero on success, or a negative errno value.
 807  */
 808 int nfs4_inode_make_writeable(struct inode *inode)
 809 {
 810         struct nfs_delegation *delegation;
 811
 812         rcu_read_lock();
 813         delegation = nfs4_get_valid_delegation(inode);
 814         if (delegation == NULL ||
 815             (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
 816              (delegation->type & FMODE_WRITE))) {
 817                 rcu_read_unlock();
 818                 return 0;
 819         }
 820         rcu_read_unlock();
 821         return nfs4_inode_return_delegation(inode);
 822 }
 823
 824 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
 825                 struct nfs_delegation *delegation)
 826 {
 827         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 828         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 829 }
 830
 831 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
 832 {
 833         struct nfs_delegation *delegation;
 834         bool ret = false;
 835
 836         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 837                 nfs_mark_return_delegation(server, delegation);
 838                 ret = true;
 839         }
 840         return ret;
 841 }
 842
 843 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 844 {
 845         struct nfs_server *server;
 846
 847         rcu_read_lock();
 848         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 849                 nfs_server_mark_return_all_delegations(server);
 850         rcu_read_unlock();
 851 }
 852
 853 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 854 {
 855         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 856                 nfs4_schedule_state_manager(clp);
 857 }
 858
 859 /**
 860  * nfs_expire_all_delegations
 861  * @clp: client to process
 862  *
 863  */
 864 void nfs_expire_all_delegations(struct nfs_client *clp)
 865 {
 866         nfs_client_mark_return_all_delegations(clp);
 867         nfs_delegation_run_state_manager(clp);
 868 }
 869
 870 /**
 871  * nfs_server_return_all_delegations - return delegations for one superblock
 872  * @server: pointer to nfs_server to process
 873  *
 874  */
 875 void nfs_server_return_all_delegations(struct nfs_server *server)
 876 {
 877         struct nfs_client *clp = server->nfs_client;
 878         bool need_wait;
 879
 880         if (clp == NULL)
 881                 return;
 882
 883         rcu_read_lock();
 884         need_wait = nfs_server_mark_return_all_delegations(server);
 885         rcu_read_unlock();
 886
 887         if (need_wait) {
 888                 nfs4_schedule_state_manager(clp);
 889                 nfs4_wait_clnt_recover(clp);
 890         }
 891 }
 892
 893 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
 894                                                  fmode_t flags)
 895 {
 896         struct nfs_delegation *delegation;
 897
 898         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 899                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 900                         continue;
 901                 if (delegation->type & flags)
 902                         nfs_mark_return_if_closed_delegation(server, delegation);
 903         }
 904 }
 905
 906 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
 907                                                         fmode_t flags)
 908 {
 909         struct nfs_server *server;
 910
 911         rcu_read_lock();
 912         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 913                 nfs_mark_return_unused_delegation_types(server, flags);
 914         rcu_read_unlock();
 915 }
 916
 917 static void nfs_revoke_delegation(struct inode *inode,
 918                 const nfs4_stateid *stateid)
 919 {
 920         struct nfs_delegation *delegation;
 921         nfs4_stateid tmp;
 922         bool ret = false;
 923
 924         rcu_read_lock();
 925         delegation = rcu_dereference(NFS_I(inode)->delegation);
 926         if (delegation == NULL)
 927                 goto out;
 928         if (stateid == NULL) {
 929                 nfs4_stateid_copy(&tmp, &delegation->stateid);
 930                 stateid = &tmp;
 931         } else {
 932                 if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
 933                         goto out;
 934                 spin_lock(&delegation->lock);
 935                 if (stateid->seqid) {
 936                         if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
 937                                 spin_unlock(&delegation->lock);
 938                                 goto out;
 939                         }
 940                         delegation->stateid.seqid = stateid->seqid;
 941                 }
 942                 spin_unlock(&delegation->lock);
 943         }
 944         nfs_mark_delegation_revoked(delegation);
 945         ret = true;
 946 out:
 947         rcu_read_unlock();
 948         if (ret)
 949                 nfs_inode_find_state_and_recover(inode, stateid);
 950 }
 951
 952 void nfs_remove_bad_delegation(struct inode *inode,
 953                 const nfs4_stateid *stateid)
 954 {
 955         nfs_revoke_delegation(inode, stateid);
 956 }
 957 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 958
 959 void nfs_delegation_mark_returned(struct inode *inode,
 960                 const nfs4_stateid *stateid)
 961 {
 962         struct nfs_delegation *delegation;
 963
 964         if (!inode)
 965                 return;
 966
 967         rcu_read_lock();
 968         delegation = rcu_dereference(NFS_I(inode)->delegation);
 969         if (!delegation)
 970                 goto out_rcu_unlock;
 971
 972         spin_lock(&delegation->lock);
 973         if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
 974                 goto out_spin_unlock;
 975         if (stateid->seqid) {
 976                 /* If delegation->stateid is newer, dont mark as returned */
 977                 if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
 978                         goto out_clear_returning;
 979                 if (delegation->stateid.seqid != stateid->seqid)
 980                         delegation->stateid.seqid = stateid->seqid;
 981         }
 982
 983         nfs_mark_delegation_revoked(delegation);
 984
 985 out_clear_returning:
 986         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 987 out_spin_unlock:
 988         spin_unlock(&delegation->lock);
 989 out_rcu_unlock:
 990         rcu_read_unlock();
 991
 992         nfs_inode_find_state_and_recover(inode, stateid);
 993 }
 994
 995 /**
 996  * nfs_expire_unused_delegation_types
 997  * @clp: client to process
 998  * @flags: delegation types to expire
 999  *
1000  */
1001 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1002 {
1003         nfs_client_mark_return_unused_delegation_types(clp, flags);
1004         nfs_delegation_run_state_manager(clp);
1005 }
1006
1007 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1008 {
1009         struct nfs_delegation *delegation;
1010
1011         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1012                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1013                         continue;
1014                 nfs_mark_return_if_closed_delegation(server, delegation);
1015         }
1016 }
1017
1018 /**
1019  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1020  * @clp: nfs_client to process
1021  *
1022  */
1023 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1024 {
1025         struct nfs_server *server;
1026
1027         rcu_read_lock();
1028         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1029                 nfs_mark_return_unreferenced_delegations(server);
1030         rcu_read_unlock();
1031
1032         nfs_delegation_run_state_manager(clp);
1033 }
1034
1035 /**
1036  * nfs_async_inode_return_delegation - asynchronously return a delegation
1037  * @inode: inode to process
1038  * @stateid: state ID information
1039  *
1040  * Returns zero on success, or a negative errno value.
1041  */
1042 int nfs_async_inode_return_delegation(struct inode *inode,
1043                                       const nfs4_stateid *stateid)
1044 {
1045         struct nfs_server *server = NFS_SERVER(inode);
1046         struct nfs_client *clp = server->nfs_client;
1047         struct nfs_delegation *delegation;
1048
1049         rcu_read_lock();
1050         delegation = nfs4_get_valid_delegation(inode);
1051         if (delegation == NULL)
1052                 goto out_enoent;
1053         if (stateid != NULL &&
1054             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1055                 goto out_enoent;
1056         nfs_mark_return_delegation(server, delegation);
1057         rcu_read_unlock();
1058
1059         /* If there are any application leases or delegations, recall them */
1060         break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1061
1062         nfs_delegation_run_state_manager(clp);
1063         return 0;
1064 out_enoent:
1065         rcu_read_unlock();
1066         return -ENOENT;
1067 }
1068
1069 static struct inode *
1070 nfs_delegation_find_inode_server(struct nfs_server *server,
1071                                  const struct nfs_fh *fhandle)
1072 {
1073         struct nfs_delegation *delegation;
1074         struct super_block *freeme = NULL;
1075         struct inode *res = NULL;
1076
1077         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1078                 spin_lock(&delegation->lock);
1079                 if (delegation->inode != NULL &&
1080                     !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1081                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1082                         if (nfs_sb_active(server->super)) {
1083                                 freeme = server->super;
1084                                 res = igrab(delegation->inode);
1085                         }
1086                         spin_unlock(&delegation->lock);
1087                         if (res != NULL)
1088                                 return res;
1089                         if (freeme) {
1090                                 rcu_read_unlock();
1091                                 nfs_sb_deactive(freeme);
1092                                 rcu_read_lock();
1093                         }
1094                         return ERR_PTR(-EAGAIN);
1095                 }
1096                 spin_unlock(&delegation->lock);
1097         }
1098         return ERR_PTR(-ENOENT);
1099 }
1100
1101 /**
1102  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1103  * @clp: client state handle
1104  * @fhandle: filehandle from a delegation recall
1105  *
1106  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1107  * cannot be found.
1108  */
1109 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1110                                         const struct nfs_fh *fhandle)
1111 {
1112         struct nfs_server *server;
1113         struct inode *res;
1114
1115         rcu_read_lock();
1116         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1117                 res = nfs_delegation_find_inode_server(server, fhandle);
1118                 if (res != ERR_PTR(-ENOENT)) {
1119                         rcu_read_unlock();
1120                         return res;
1121                 }
1122         }
1123         rcu_read_unlock();
1124         return ERR_PTR(-ENOENT);
1125 }
1126
1127 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1128 {
1129         struct nfs_delegation *delegation;
1130
1131         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1132                 /*
1133                  * If the delegation may have been admin revoked, then we
1134                  * cannot reclaim it.
1135                  */
1136                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1137                         continue;
1138                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1139         }
1140 }
1141
1142 /**
1143  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1144  * @clp: nfs_client to process
1145  *
1146  */
1147 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1148 {
1149         struct nfs_server *server;
1150
1151         rcu_read_lock();
1152         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1153                 nfs_delegation_mark_reclaim_server(server);
1154         rcu_read_unlock();
1155 }
1156
1157 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1158                 void __always_unused *data)
1159 {
1160         struct nfs_delegation *delegation;
1161         struct inode *inode;
1162 restart:
1163         rcu_read_lock();
1164 restart_locked:
1165         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1166                 if (test_bit(NFS_DELEGATION_INODE_FREEING,
1167                                         &delegation->flags) ||
1168                     test_bit(NFS_DELEGATION_RETURNING,
1169                                         &delegation->flags) ||
1170                     test_bit(NFS_DELEGATION_NEED_RECLAIM,
1171                                         &delegation->flags) == 0)
1172                         continue;
1173                 inode = nfs_delegation_grab_inode(delegation);
1174                 if (inode == NULL)
1175                         goto restart_locked;
1176                 delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1177                 rcu_read_unlock();
1178                 if (delegation != NULL) {
1179                         if (nfs_detach_delegation(NFS_I(inode), delegation,
1180                                                 server) != NULL)
1181                                 nfs_free_delegation(delegation);
1182                         /* Match nfs_start_delegation_return_locked */
1183                         nfs_put_delegation(delegation);
1184                 }
1185                 iput(inode);
1186                 cond_resched();
1187                 goto restart;
1188         }
1189         rcu_read_unlock();
1190         return 0;
1191 }
1192
1193 /**
1194  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1195  * @clp: nfs_client to process
1196  *
1197  */
1198 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1199 {
1200         nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1201                         NULL);
1202 }
1203
1204 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1205 {
1206         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1207                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
1208                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
1209 }
1210
1211 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1212             struct nfs_delegation *delegation)
1213 {
1214         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1215                 return;
1216         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1217         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1218         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1219 }
1220
1221 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1222                 struct inode *inode)
1223 {
1224         struct nfs_delegation *delegation;
1225
1226         rcu_read_lock();
1227         delegation = rcu_dereference(NFS_I(inode)->delegation);
1228         if (delegation)
1229                 nfs_mark_test_expired_delegation(server, delegation);
1230         rcu_read_unlock();
1231
1232 }
1233
1234 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1235 {
1236         struct nfs_delegation *delegation;
1237
1238         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1239                 nfs_mark_test_expired_delegation(server, delegation);
1240 }
1241
1242 /**
1243  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1244  * @clp: nfs_client to process
1245  *
1246  * Iterates through all the delegations associated with this server and
1247  * marks them as needing to be checked for validity.
1248  */
1249 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1250 {
1251         struct nfs_server *server;
1252
1253         rcu_read_lock();
1254         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1255                 nfs_delegation_mark_test_expired_server(server);
1256         rcu_read_unlock();
1257 }
1258
1259 /**
1260  * nfs_test_expired_all_delegations - test all delegations for a client
1261  * @clp: nfs_client to process
1262  *
1263  * Helper for handling "recallable state revoked" status from server.
1264  */
1265 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1266 {
1267         nfs_mark_test_expired_all_delegations(clp);
1268         nfs4_schedule_state_manager(clp);
1269 }
1270
1271 static void
1272 nfs_delegation_test_free_expired(struct inode *inode,
1273                 nfs4_stateid *stateid,
1274                 const struct cred *cred)
1275 {
1276         struct nfs_server *server = NFS_SERVER(inode);
1277         const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1278         int status;
1279
1280         if (!cred)
1281                 return;
1282         status = ops->test_and_free_expired(server, stateid, cred);
1283         if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1284                 nfs_remove_bad_delegation(inode, stateid);
1285 }
1286
1287 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1288                 void __always_unused *data)
1289 {
1290         struct nfs_delegation *delegation;
1291         struct inode *inode;
1292         const struct cred *cred;
1293         nfs4_stateid stateid;
1294         unsigned long gen = ++server->delegation_gen;
1295
1296 restart:
1297         rcu_read_lock();
1298 restart_locked:
1299         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1300                 if (test_bit(NFS_DELEGATION_INODE_FREEING,
1301                                         &delegation->flags) ||
1302                     test_bit(NFS_DELEGATION_RETURNING,
1303                                         &delegation->flags) ||
1304                     test_bit(NFS_DELEGATION_TEST_EXPIRED,
1305                                         &delegation->flags) == 0 ||
1306                         delegation->test_gen == gen)
1307                         continue;
1308                 inode = nfs_delegation_grab_inode(delegation);
1309                 if (inode == NULL)
1310                         goto restart_locked;
1311                 spin_lock(&delegation->lock);
1312                 cred = get_cred_rcu(delegation->cred);
1313                 nfs4_stateid_copy(&stateid, &delegation->stateid);
1314                 spin_unlock(&delegation->lock);
1315                 delegation->test_gen = gen;
1316                 clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1317                 rcu_read_unlock();
1318                 nfs_delegation_test_free_expired(inode, &stateid, cred);
1319                 put_cred(cred);
1320                 if (!nfs4_server_rebooted(server->nfs_client)) {
1321                         iput(inode);
1322                         cond_resched();
1323                         goto restart;
1324                 }
1325                 nfs_inode_mark_test_expired_delegation(server,inode);
1326                 iput(inode);
1327                 return -EAGAIN;
1328         }
1329         rcu_read_unlock();
1330         return 0;
1331 }
1332
1333 /**
1334  * nfs_reap_expired_delegations - reap expired delegations
1335  * @clp: nfs_client to process
1336  *
1337  * Iterates through all the delegations associated with this server and
1338  * checks if they have may have been revoked. This function is usually
1339  * expected to be called in cases where the server may have lost its
1340  * lease.
1341  */
1342 void nfs_reap_expired_delegations(struct nfs_client *clp)
1343 {
1344         nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1345                         NULL);
1346 }
1347
1348 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1349                 const nfs4_stateid *stateid)
1350 {
1351         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1352         struct nfs_delegation *delegation;
1353         bool found = false;
1354
1355         rcu_read_lock();
1356         delegation = rcu_dereference(NFS_I(inode)->delegation);
1357         if (delegation &&
1358             nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1359             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1360                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1361                 found = true;
1362         }
1363         rcu_read_unlock();
1364         if (found)
1365                 nfs4_schedule_state_manager(clp);
1366 }
1367
1368 /**
1369  * nfs_delegations_present - check for existence of delegations
1370  * @clp: client state handle
1371  *
1372  * Returns one if there are any nfs_delegation structures attached
1373  * to this nfs_client.
1374  */
1375 int nfs_delegations_present(struct nfs_client *clp)
1376 {
1377         struct nfs_server *server;
1378         int ret = 0;
1379
1380         rcu_read_lock();
1381         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1382                 if (!list_empty(&server->delegations)) {
1383                         ret = 1;
1384                         break;
1385                 }
1386         rcu_read_unlock();
1387         return ret;
1388 }
1389
1390 /**
1391  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1392  * @dst: stateid to refresh
1393  * @inode: inode to check
1394  *
1395  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1396  * that matches our delegation stateid. Otherwise "false" is returned.
1397  */
1398 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1399 {
1400         struct nfs_delegation *delegation;
1401         bool ret = false;
1402         if (!inode)
1403                 goto out;
1404
1405         rcu_read_lock();
1406         delegation = rcu_dereference(NFS_I(inode)->delegation);
1407         if (delegation != NULL &&
1408             nfs4_stateid_match_other(dst, &delegation->stateid) &&
1409             nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1410             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1411                 dst->seqid = delegation->stateid.seqid;
1412                 ret = true;
1413         }
1414         rcu_read_unlock();
1415 out:
1416         return ret;
1417 }
1418
1419 /**
1420  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1421  * @inode: inode to check
1422  * @flags: delegation type requirement
1423  * @dst: stateid data structure to fill in
1424  * @cred: optional argument to retrieve credential
1425  *
1426  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1427  * otherwise "false" is returned.
1428  */
1429 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1430                 nfs4_stateid *dst, const struct cred **cred)
1431 {
1432         struct nfs_inode *nfsi = NFS_I(inode);
1433         struct nfs_delegation *delegation;
1434         bool ret = false;
1435
1436         flags &= FMODE_READ|FMODE_WRITE;
1437         rcu_read_lock();
1438         delegation = rcu_dereference(nfsi->delegation);
1439         if (!delegation)
1440                 goto out;
1441         spin_lock(&delegation->lock);
1442         ret = nfs4_is_valid_delegation(delegation, flags);
1443         if (ret) {
1444                 nfs4_stateid_copy(dst, &delegation->stateid);
1445                 nfs_mark_delegation_referenced(delegation);
1446                 if (cred)
1447                         *cred = get_cred(delegation->cred);
1448         }
1449         spin_unlock(&delegation->lock);
1450 out:
1451         rcu_read_unlock();
1452         return ret;
1453 }
1454
1455 /**
1456  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1457  * @inode: inode to check
1458  *
1459  * This function checks the number of outstanding writes to the file
1460  * against the delegation 'space_limit' field to see if
1461  * the spec requires us to flush the file on close.
1462  */
1463 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1464 {
1465         struct nfs_inode *nfsi = NFS_I(inode);
1466         struct nfs_delegation *delegation;
1467         bool ret = true;
1468
1469         rcu_read_lock();
1470         delegation = rcu_dereference(nfsi->delegation);
1471         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1472                 goto out;
1473         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1474                 ret = false;
1475 out:
1476         rcu_read_unlock();
1477         return ret;
1478 }
1479
1480 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);