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 static void nfs_free_delegation(struct nfs_delegation *delegation)
  29 {
  30         put_cred(delegation->cred);
  31         delegation->cred = NULL;
  32         kfree_rcu(delegation, rcu);
  33 }
  34
  35 /**
  36  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
  37  * @delegation: delegation to process
  38  *
  39  */
  40 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
  41 {
  42         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
  43 }
  44
  45 static bool
  46 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
  47                 fmode_t flags)
  48 {
  49         if (delegation != NULL && (delegation->type & flags) == flags &&
  50             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
  51             !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
  52                 return true;
  53         return false;
  54 }
  55
  56 struct nfs_delegation *nfs4_get_valid_delegation(const struct inode *inode)
  57 {
  58         struct nfs_delegation *delegation;
  59
  60         delegation = rcu_dereference(NFS_I(inode)->delegation);
  61         if (nfs4_is_valid_delegation(delegation, 0))
  62                 return delegation;
  63         return NULL;
  64 }
  65
  66 static int
  67 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
  68 {
  69         struct nfs_delegation *delegation;
  70         int ret = 0;
  71
  72         flags &= FMODE_READ|FMODE_WRITE;
  73         rcu_read_lock();
  74         delegation = rcu_dereference(NFS_I(inode)->delegation);
  75         if (nfs4_is_valid_delegation(delegation, flags)) {
  76                 if (mark)
  77                         nfs_mark_delegation_referenced(delegation);
  78                 ret = 1;
  79         }
  80         rcu_read_unlock();
  81         return ret;
  82 }
  83 /**
  84  * nfs_have_delegation - check if inode has a delegation, mark it
  85  * NFS_DELEGATION_REFERENCED if there is one.
  86  * @inode: inode to check
  87  * @flags: delegation types to check for
  88  *
  89  * Returns one if inode has the indicated delegation, otherwise zero.
  90  */
  91 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
  92 {
  93         return nfs4_do_check_delegation(inode, flags, true);
  94 }
  95
  96 /*
  97  * nfs4_check_delegation - check if inode has a delegation, do not mark
  98  * NFS_DELEGATION_REFERENCED if it has one.
  99  */
 100 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
 101 {
 102         return nfs4_do_check_delegation(inode, flags, false);
 103 }
 104
 105 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
 106 {
 107         struct inode *inode = state->inode;
 108         struct file_lock *fl;
 109         struct file_lock_context *flctx = inode->i_flctx;
 110         struct list_head *list;
 111         int status = 0;
 112
 113         if (flctx == NULL)
 114                 goto out;
 115
 116         list = &flctx->flc_posix;
 117         spin_lock(&flctx->flc_lock);
 118 restart:
 119         list_for_each_entry(fl, list, fl_list) {
 120                 if (nfs_file_open_context(fl->fl_file)->state != state)
 121                         continue;
 122                 spin_unlock(&flctx->flc_lock);
 123                 status = nfs4_lock_delegation_recall(fl, state, stateid);
 124                 if (status < 0)
 125                         goto out;
 126                 spin_lock(&flctx->flc_lock);
 127         }
 128         if (list == &flctx->flc_posix) {
 129                 list = &flctx->flc_flock;
 130                 goto restart;
 131         }
 132         spin_unlock(&flctx->flc_lock);
 133 out:
 134         return status;
 135 }
 136
 137 static int nfs_delegation_claim_opens(struct inode *inode,
 138                 const nfs4_stateid *stateid, fmode_t type)
 139 {
 140         struct nfs_inode *nfsi = NFS_I(inode);
 141         struct nfs_open_context *ctx;
 142         struct nfs4_state_owner *sp;
 143         struct nfs4_state *state;
 144         unsigned int seq;
 145         int err;
 146
 147 again:
 148         rcu_read_lock();
 149         list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
 150                 state = ctx->state;
 151                 if (state == NULL)
 152                         continue;
 153                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
 154                         continue;
 155                 if (!nfs4_valid_open_stateid(state))
 156                         continue;
 157                 if (!nfs4_stateid_match(&state->stateid, stateid))
 158                         continue;
 159                 if (!get_nfs_open_context(ctx))
 160                         continue;
 161                 rcu_read_unlock();
 162                 sp = state->owner;
 163                 /* Block nfs4_proc_unlck */
 164                 mutex_lock(&sp->so_delegreturn_mutex);
 165                 seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
 166                 err = nfs4_open_delegation_recall(ctx, state, stateid);
 167                 if (!err)
 168                         err = nfs_delegation_claim_locks(state, stateid);
 169                 if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
 170                         err = -EAGAIN;
 171                 mutex_unlock(&sp->so_delegreturn_mutex);
 172                 put_nfs_open_context(ctx);
 173                 if (err != 0)
 174                         return err;
 175                 goto again;
 176         }
 177         rcu_read_unlock();
 178         return 0;
 179 }
 180
 181 /**
 182  * nfs_inode_reclaim_delegation - process a delegation reclaim request
 183  * @inode: inode to process
 184  * @cred: credential to use for request
 185  * @type: delegation type
 186  * @stateid: delegation stateid
 187  * @pagemod_limit: write delegation "space_limit"
 188  *
 189  */
 190 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
 191                                   fmode_t type,
 192                                   const nfs4_stateid *stateid,
 193                                   unsigned long pagemod_limit)
 194 {
 195         struct nfs_delegation *delegation;
 196         const struct cred *oldcred = NULL;
 197
 198         rcu_read_lock();
 199         delegation = rcu_dereference(NFS_I(inode)->delegation);
 200         if (delegation != NULL) {
 201                 spin_lock(&delegation->lock);
 202                 if (delegation->inode != NULL) {
 203                         nfs4_stateid_copy(&delegation->stateid, stateid);
 204                         delegation->type = type;
 205                         delegation->pagemod_limit = pagemod_limit;
 206                         oldcred = delegation->cred;
 207                         delegation->cred = get_cred(cred);
 208                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
 209                                   &delegation->flags);
 210                         spin_unlock(&delegation->lock);
 211                         rcu_read_unlock();
 212                         put_cred(oldcred);
 213                         trace_nfs4_reclaim_delegation(inode, type);
 214                         return;
 215                 }
 216                 /* We appear to have raced with a delegation return. */
 217                 spin_unlock(&delegation->lock);
 218         }
 219         rcu_read_unlock();
 220         nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
 221 }
 222
 223 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 224 {
 225         int res = 0;
 226
 227         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 228                 res = nfs4_proc_delegreturn(inode,
 229                                 delegation->cred,
 230                                 &delegation->stateid,
 231                                 issync);
 232         nfs_free_delegation(delegation);
 233         return res;
 234 }
 235
 236 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
 237 {
 238         struct inode *inode = NULL;
 239
 240         spin_lock(&delegation->lock);
 241         if (delegation->inode != NULL)
 242                 inode = igrab(delegation->inode);
 243         if (!inode)
 244                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 245         spin_unlock(&delegation->lock);
 246         return inode;
 247 }
 248
 249 static struct nfs_delegation *
 250 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
 251 {
 252         struct nfs_delegation *ret = NULL;
 253         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
 254
 255         if (delegation == NULL)
 256                 goto out;
 257         spin_lock(&delegation->lock);
 258         if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 259                 ret = delegation;
 260         spin_unlock(&delegation->lock);
 261 out:
 262         return ret;
 263 }
 264
 265 static struct nfs_delegation *
 266 nfs_start_delegation_return(struct nfs_inode *nfsi)
 267 {
 268         struct nfs_delegation *delegation;
 269
 270         rcu_read_lock();
 271         delegation = nfs_start_delegation_return_locked(nfsi);
 272         rcu_read_unlock();
 273         return delegation;
 274 }
 275
 276 static void
 277 nfs_abort_delegation_return(struct nfs_delegation *delegation,
 278                 struct nfs_client *clp)
 279 {
 280
 281         spin_lock(&delegation->lock);
 282         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 283         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 284         spin_unlock(&delegation->lock);
 285         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 286 }
 287
 288 static struct nfs_delegation *
 289 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
 290                 struct nfs_delegation *delegation,
 291                 struct nfs_client *clp)
 292 {
 293         struct nfs_delegation *deleg_cur =
 294                 rcu_dereference_protected(nfsi->delegation,
 295                                 lockdep_is_held(&clp->cl_lock));
 296
 297         if (deleg_cur == NULL || delegation != deleg_cur)
 298                 return NULL;
 299
 300         spin_lock(&delegation->lock);
 301         set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 302         list_del_rcu(&delegation->super_list);
 303         delegation->inode = NULL;
 304         rcu_assign_pointer(nfsi->delegation, NULL);
 305         spin_unlock(&delegation->lock);
 306         return delegation;
 307 }
 308
 309 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
 310                 struct nfs_delegation *delegation,
 311                 struct nfs_server *server)
 312 {
 313         struct nfs_client *clp = server->nfs_client;
 314
 315         spin_lock(&clp->cl_lock);
 316         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
 317         spin_unlock(&clp->cl_lock);
 318         return delegation;
 319 }
 320
 321 static struct nfs_delegation *
 322 nfs_inode_detach_delegation(struct inode *inode)
 323 {
 324         struct nfs_inode *nfsi = NFS_I(inode);
 325         struct nfs_server *server = NFS_SERVER(inode);
 326         struct nfs_delegation *delegation;
 327
 328         delegation = nfs_start_delegation_return(nfsi);
 329         if (delegation == NULL)
 330                 return NULL;
 331         return nfs_detach_delegation(nfsi, delegation, server);
 332 }
 333
 334 static void
 335 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
 336                 const struct nfs_delegation *update)
 337 {
 338         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
 339                 delegation->stateid.seqid = update->stateid.seqid;
 340                 smp_wmb();
 341                 delegation->type = update->type;
 342         }
 343 }
 344
 345 /**
 346  * nfs_inode_set_delegation - set up a delegation on an inode
 347  * @inode: inode to which delegation applies
 348  * @cred: cred to use for subsequent delegation processing
 349  * @type: delegation type
 350  * @stateid: delegation stateid
 351  * @pagemod_limit: write delegation "space_limit"
 352  *
 353  * Returns zero on success, or a negative errno value.
 354  */
 355 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
 356                                   fmode_t type,
 357                                   const nfs4_stateid *stateid,
 358                                   unsigned long pagemod_limit)
 359 {
 360         struct nfs_server *server = NFS_SERVER(inode);
 361         struct nfs_client *clp = server->nfs_client;
 362         struct nfs_inode *nfsi = NFS_I(inode);
 363         struct nfs_delegation *delegation, *old_delegation;
 364         struct nfs_delegation *freeme = NULL;
 365         int status = 0;
 366
 367         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
 368         if (delegation == NULL)
 369                 return -ENOMEM;
 370         nfs4_stateid_copy(&delegation->stateid, stateid);
 371         delegation->type = type;
 372         delegation->pagemod_limit = pagemod_limit;
 373         delegation->change_attr = inode_peek_iversion_raw(inode);
 374         delegation->cred = get_cred(cred);
 375         delegation->inode = inode;
 376         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 377         spin_lock_init(&delegation->lock);
 378
 379         spin_lock(&clp->cl_lock);
 380         old_delegation = rcu_dereference_protected(nfsi->delegation,
 381                                         lockdep_is_held(&clp->cl_lock));
 382         if (old_delegation != NULL) {
 383                 /* Is this an update of the existing delegation? */
 384                 if (nfs4_stateid_match_other(&old_delegation->stateid,
 385                                         &delegation->stateid)) {
 386                         nfs_update_inplace_delegation(old_delegation,
 387                                         delegation);
 388                         goto out;
 389                 }
 390                 /*
 391                  * Deal with broken servers that hand out two
 392                  * delegations for the same file.
 393                  * Allow for upgrades to a WRITE delegation, but
 394                  * nothing else.
 395                  */
 396                 dfprintk(FILE, "%s: server %s handed out "
 397                                 "a duplicate delegation!\n",
 398                                 __func__, clp->cl_hostname);
 399                 if (delegation->type == old_delegation->type ||
 400                     !(delegation->type & FMODE_WRITE)) {
 401                         freeme = delegation;
 402                         delegation = NULL;
 403                         goto out;
 404                 }
 405                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
 406                                         &old_delegation->flags))
 407                         goto out;
 408                 freeme = nfs_detach_delegation_locked(nfsi,
 409                                 old_delegation, clp);
 410                 if (freeme == NULL)
 411                         goto out;
 412         }
 413         list_add_tail_rcu(&delegation->super_list, &server->delegations);
 414         rcu_assign_pointer(nfsi->delegation, delegation);
 415         delegation = NULL;
 416
 417         trace_nfs4_set_delegation(inode, type);
 418
 419         spin_lock(&inode->i_lock);
 420         if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME))
 421                 NFS_I(inode)->cache_validity |= NFS_INO_REVAL_FORCED;
 422         spin_unlock(&inode->i_lock);
 423 out:
 424         spin_unlock(&clp->cl_lock);
 425         if (delegation != NULL)
 426                 nfs_free_delegation(delegation);
 427         if (freeme != NULL)
 428                 nfs_do_return_delegation(inode, freeme, 0);
 429         return status;
 430 }
 431
 432 /*
 433  * Basic procedure for returning a delegation to the server
 434  */
 435 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
 436 {
 437         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 438         struct nfs_inode *nfsi = NFS_I(inode);
 439         int err = 0;
 440
 441         if (delegation == NULL)
 442                 return 0;
 443         do {
 444                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 445                         break;
 446                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
 447                                 delegation->type);
 448                 if (!issync || err != -EAGAIN)
 449                         break;
 450                 /*
 451                  * Guard against state recovery
 452                  */
 453                 err = nfs4_wait_clnt_recover(clp);
 454         } while (err == 0);
 455
 456         if (err) {
 457                 nfs_abort_delegation_return(delegation, clp);
 458                 goto out;
 459         }
 460         if (!nfs_detach_delegation(nfsi, delegation, NFS_SERVER(inode)))
 461                 goto out;
 462
 463         err = nfs_do_return_delegation(inode, delegation, issync);
 464 out:
 465         return err;
 466 }
 467
 468 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
 469 {
 470         bool ret = false;
 471
 472         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 473                 goto out;
 474         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 475                 ret = true;
 476         if (test_and_clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) && !ret) {
 477                 struct inode *inode;
 478
 479                 spin_lock(&delegation->lock);
 480                 inode = delegation->inode;
 481                 if (inode && list_empty(&NFS_I(inode)->open_files))
 482                         ret = true;
 483                 spin_unlock(&delegation->lock);
 484         }
 485 out:
 486         return ret;
 487 }
 488
 489 /**
 490  * nfs_client_return_marked_delegations - return previously marked delegations
 491  * @clp: nfs_client to process
 492  *
 493  * Note that this function is designed to be called by the state
 494  * manager thread. For this reason, it cannot flush the dirty data,
 495  * since that could deadlock in case of a state recovery error.
 496  *
 497  * Returns zero on success, or a negative errno value.
 498  */
 499 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 500 {
 501         struct nfs_delegation *delegation;
 502         struct nfs_delegation *prev;
 503         struct nfs_server *server;
 504         struct inode *inode;
 505         struct inode *place_holder = NULL;
 506         struct nfs_delegation *place_holder_deleg = NULL;
 507         int err = 0;
 508
 509 restart:
 510         /*
 511          * To avoid quadratic looping we hold a reference
 512          * to an inode place_holder.  Each time we restart, we
 513          * list nfs_servers from the server of that inode, and
 514          * delegation in the server from the delegations of that
 515          * inode.
 516          * prev is an RCU-protected pointer to a delegation which
 517          * wasn't marked for return and might be a good choice for
 518          * the next place_holder.
 519          */
 520         rcu_read_lock();
 521         prev = NULL;
 522         if (place_holder)
 523                 server = NFS_SERVER(place_holder);
 524         else
 525                 server = list_entry_rcu(clp->cl_superblocks.next,
 526                                         struct nfs_server, client_link);
 527         list_for_each_entry_from_rcu(server, &clp->cl_superblocks, client_link) {
 528                 delegation = NULL;
 529                 if (place_holder && server == NFS_SERVER(place_holder))
 530                         delegation = rcu_dereference(NFS_I(place_holder)->delegation);
 531                 if (!delegation || delegation != place_holder_deleg)
 532                         delegation = list_entry_rcu(server->delegations.next,
 533                                                     struct nfs_delegation, super_list);
 534                 list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
 535                         struct inode *to_put = NULL;
 536
 537                         if (!nfs_delegation_need_return(delegation)) {
 538                                 prev = delegation;
 539                                 continue;
 540                         }
 541                         if (!nfs_sb_active(server->super))
 542                                 break; /* continue in outer loop */
 543
 544                         if (prev) {
 545                                 struct inode *tmp;
 546
 547                                 tmp = nfs_delegation_grab_inode(prev);
 548                                 if (tmp) {
 549                                         to_put = place_holder;
 550                                         place_holder = tmp;
 551                                         place_holder_deleg = prev;
 552                                 }
 553                         }
 554
 555                         inode = nfs_delegation_grab_inode(delegation);
 556                         if (inode == NULL) {
 557                                 rcu_read_unlock();
 558                                 if (to_put)
 559                                         iput(to_put);
 560                                 nfs_sb_deactive(server->super);
 561                                 goto restart;
 562                         }
 563                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 564                         rcu_read_unlock();
 565
 566                         if (to_put)
 567                                 iput(to_put);
 568
 569                         err = nfs_end_delegation_return(inode, delegation, 0);
 570                         iput(inode);
 571                         nfs_sb_deactive(server->super);
 572                         cond_resched();
 573                         if (!err)
 574                                 goto restart;
 575                         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 576                         if (place_holder)
 577                                 iput(place_holder);
 578                         return err;
 579                 }
 580         }
 581         rcu_read_unlock();
 582         if (place_holder)
 583                 iput(place_holder);
 584         return 0;
 585 }
 586
 587 /**
 588  * nfs_inode_return_delegation_noreclaim - return delegation, don't reclaim opens
 589  * @inode: inode to process
 590  *
 591  * Does not protect against delegation reclaims, therefore really only safe
 592  * to be called from nfs4_clear_inode().
 593  */
 594 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
 595 {
 596         struct nfs_delegation *delegation;
 597
 598         delegation = nfs_inode_detach_delegation(inode);
 599         if (delegation != NULL)
 600                 nfs_do_return_delegation(inode, delegation, 1);
 601 }
 602
 603 /**
 604  * nfs_inode_return_delegation - synchronously return a delegation
 605  * @inode: inode to process
 606  *
 607  * This routine will always flush any dirty data to disk on the
 608  * assumption that if we need to return the delegation, then
 609  * we should stop caching.
 610  *
 611  * Returns zero on success, or a negative errno value.
 612  */
 613 int nfs4_inode_return_delegation(struct inode *inode)
 614 {
 615         struct nfs_inode *nfsi = NFS_I(inode);
 616         struct nfs_delegation *delegation;
 617         int err = 0;
 618
 619         nfs_wb_all(inode);
 620         delegation = nfs_start_delegation_return(nfsi);
 621         if (delegation != NULL)
 622                 err = nfs_end_delegation_return(inode, delegation, 1);
 623         return err;
 624 }
 625
 626 /**
 627  * nfs4_inode_make_writeable
 628  * @inode: pointer to inode
 629  *
 630  * Make the inode writeable by returning the delegation if necessary
 631  *
 632  * Returns zero on success, or a negative errno value.
 633  */
 634 int nfs4_inode_make_writeable(struct inode *inode)
 635 {
 636         if (!nfs4_has_session(NFS_SERVER(inode)->nfs_client) ||
 637             !nfs4_check_delegation(inode, FMODE_WRITE))
 638                 return nfs4_inode_return_delegation(inode);
 639         return 0;
 640 }
 641
 642 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
 643                 struct nfs_delegation *delegation)
 644 {
 645         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 646         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 647 }
 648
 649 static void nfs_mark_return_delegation(struct nfs_server *server,
 650                 struct nfs_delegation *delegation)
 651 {
 652         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 653         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 654 }
 655
 656 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
 657 {
 658         struct nfs_delegation *delegation;
 659         bool ret = false;
 660
 661         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 662                 nfs_mark_return_delegation(server, delegation);
 663                 ret = true;
 664         }
 665         return ret;
 666 }
 667
 668 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 669 {
 670         struct nfs_server *server;
 671
 672         rcu_read_lock();
 673         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 674                 nfs_server_mark_return_all_delegations(server);
 675         rcu_read_unlock();
 676 }
 677
 678 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 679 {
 680         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 681                 nfs4_schedule_state_manager(clp);
 682 }
 683
 684 /**
 685  * nfs_expire_all_delegations
 686  * @clp: client to process
 687  *
 688  */
 689 void nfs_expire_all_delegations(struct nfs_client *clp)
 690 {
 691         nfs_client_mark_return_all_delegations(clp);
 692         nfs_delegation_run_state_manager(clp);
 693 }
 694
 695 /**
 696  * nfs_super_return_all_delegations - return delegations for one superblock
 697  * @server: pointer to nfs_server to process
 698  *
 699  */
 700 void nfs_server_return_all_delegations(struct nfs_server *server)
 701 {
 702         struct nfs_client *clp = server->nfs_client;
 703         bool need_wait;
 704
 705         if (clp == NULL)
 706                 return;
 707
 708         rcu_read_lock();
 709         need_wait = nfs_server_mark_return_all_delegations(server);
 710         rcu_read_unlock();
 711
 712         if (need_wait) {
 713                 nfs4_schedule_state_manager(clp);
 714                 nfs4_wait_clnt_recover(clp);
 715         }
 716 }
 717
 718 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
 719                                                  fmode_t flags)
 720 {
 721         struct nfs_delegation *delegation;
 722
 723         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 724                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 725                         continue;
 726                 if (delegation->type & flags)
 727                         nfs_mark_return_if_closed_delegation(server, delegation);
 728         }
 729 }
 730
 731 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
 732                                                         fmode_t flags)
 733 {
 734         struct nfs_server *server;
 735
 736         rcu_read_lock();
 737         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 738                 nfs_mark_return_unused_delegation_types(server, flags);
 739         rcu_read_unlock();
 740 }
 741
 742 static void nfs_mark_delegation_revoked(struct nfs_server *server,
 743                 struct nfs_delegation *delegation)
 744 {
 745         set_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 746         delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
 747         nfs_mark_return_delegation(server, delegation);
 748 }
 749
 750 static bool nfs_revoke_delegation(struct inode *inode,
 751                 const nfs4_stateid *stateid)
 752 {
 753         struct nfs_delegation *delegation;
 754         nfs4_stateid tmp;
 755         bool ret = false;
 756
 757         rcu_read_lock();
 758         delegation = rcu_dereference(NFS_I(inode)->delegation);
 759         if (delegation == NULL)
 760                 goto out;
 761         if (stateid == NULL) {
 762                 nfs4_stateid_copy(&tmp, &delegation->stateid);
 763                 stateid = &tmp;
 764         } else if (!nfs4_stateid_match(stateid, &delegation->stateid))
 765                 goto out;
 766         nfs_mark_delegation_revoked(NFS_SERVER(inode), delegation);
 767         ret = true;
 768 out:
 769         rcu_read_unlock();
 770         if (ret)
 771                 nfs_inode_find_state_and_recover(inode, stateid);
 772         return ret;
 773 }
 774
 775 void nfs_remove_bad_delegation(struct inode *inode,
 776                 const nfs4_stateid *stateid)
 777 {
 778         struct nfs_delegation *delegation;
 779
 780         if (!nfs_revoke_delegation(inode, stateid))
 781                 return;
 782         delegation = nfs_inode_detach_delegation(inode);
 783         if (delegation)
 784                 nfs_free_delegation(delegation);
 785 }
 786 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 787
 788 /**
 789  * nfs_expire_unused_delegation_types
 790  * @clp: client to process
 791  * @flags: delegation types to expire
 792  *
 793  */
 794 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
 795 {
 796         nfs_client_mark_return_unused_delegation_types(clp, flags);
 797         nfs_delegation_run_state_manager(clp);
 798 }
 799
 800 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
 801 {
 802         struct nfs_delegation *delegation;
 803
 804         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 805                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
 806                         continue;
 807                 nfs_mark_return_if_closed_delegation(server, delegation);
 808         }
 809 }
 810
 811 /**
 812  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
 813  * @clp: nfs_client to process
 814  *
 815  */
 816 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
 817 {
 818         struct nfs_server *server;
 819
 820         rcu_read_lock();
 821         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 822                 nfs_mark_return_unreferenced_delegations(server);
 823         rcu_read_unlock();
 824
 825         nfs_delegation_run_state_manager(clp);
 826 }
 827
 828 /**
 829  * nfs_async_inode_return_delegation - asynchronously return a delegation
 830  * @inode: inode to process
 831  * @stateid: state ID information
 832  *
 833  * Returns zero on success, or a negative errno value.
 834  */
 835 int nfs_async_inode_return_delegation(struct inode *inode,
 836                                       const nfs4_stateid *stateid)
 837 {
 838         struct nfs_server *server = NFS_SERVER(inode);
 839         struct nfs_client *clp = server->nfs_client;
 840         struct nfs_delegation *delegation;
 841
 842         rcu_read_lock();
 843         delegation = rcu_dereference(NFS_I(inode)->delegation);
 844         if (delegation == NULL)
 845                 goto out_enoent;
 846         if (stateid != NULL &&
 847             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
 848                 goto out_enoent;
 849         nfs_mark_return_delegation(server, delegation);
 850         rcu_read_unlock();
 851
 852         nfs_delegation_run_state_manager(clp);
 853         return 0;
 854 out_enoent:
 855         rcu_read_unlock();
 856         return -ENOENT;
 857 }
 858
 859 static struct inode *
 860 nfs_delegation_find_inode_server(struct nfs_server *server,
 861                                  const struct nfs_fh *fhandle)
 862 {
 863         struct nfs_delegation *delegation;
 864         struct inode *freeme, *res = NULL;
 865
 866         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 867                 spin_lock(&delegation->lock);
 868                 if (delegation->inode != NULL &&
 869                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
 870                         freeme = igrab(delegation->inode);
 871                         if (freeme && nfs_sb_active(freeme->i_sb))
 872                                 res = freeme;
 873                         spin_unlock(&delegation->lock);
 874                         if (res != NULL)
 875                                 return res;
 876                         if (freeme) {
 877                                 rcu_read_unlock();
 878                                 iput(freeme);
 879                                 rcu_read_lock();
 880                         }
 881                         return ERR_PTR(-EAGAIN);
 882                 }
 883                 spin_unlock(&delegation->lock);
 884         }
 885         return ERR_PTR(-ENOENT);
 886 }
 887
 888 /**
 889  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
 890  * @clp: client state handle
 891  * @fhandle: filehandle from a delegation recall
 892  *
 893  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
 894  * cannot be found.
 895  */
 896 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
 897                                         const struct nfs_fh *fhandle)
 898 {
 899         struct nfs_server *server;
 900         struct inode *res;
 901
 902         rcu_read_lock();
 903         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 904                 res = nfs_delegation_find_inode_server(server, fhandle);
 905                 if (res != ERR_PTR(-ENOENT)) {
 906                         rcu_read_unlock();
 907                         return res;
 908                 }
 909         }
 910         rcu_read_unlock();
 911         return ERR_PTR(-ENOENT);
 912 }
 913
 914 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
 915 {
 916         struct nfs_delegation *delegation;
 917
 918         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 919                 /*
 920                  * If the delegation may have been admin revoked, then we
 921                  * cannot reclaim it.
 922                  */
 923                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
 924                         continue;
 925                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 926         }
 927 }
 928
 929 /**
 930  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
 931  * @clp: nfs_client to process
 932  *
 933  */
 934 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 935 {
 936         struct nfs_server *server;
 937
 938         rcu_read_lock();
 939         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 940                 nfs_delegation_mark_reclaim_server(server);
 941         rcu_read_unlock();
 942 }
 943
 944 /**
 945  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
 946  * @clp: nfs_client to process
 947  *
 948  */
 949 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 950 {
 951         struct nfs_delegation *delegation;
 952         struct nfs_server *server;
 953         struct inode *inode;
 954
 955 restart:
 956         rcu_read_lock();
 957         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 958                 list_for_each_entry_rcu(delegation, &server->delegations,
 959                                                                 super_list) {
 960                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
 961                                                 &delegation->flags) ||
 962                             test_bit(NFS_DELEGATION_RETURNING,
 963                                                 &delegation->flags) ||
 964                             test_bit(NFS_DELEGATION_NEED_RECLAIM,
 965                                                 &delegation->flags) == 0)
 966                                 continue;
 967                         if (!nfs_sb_active(server->super))
 968                                 break; /* continue in outer loop */
 969                         inode = nfs_delegation_grab_inode(delegation);
 970                         if (inode == NULL) {
 971                                 rcu_read_unlock();
 972                                 nfs_sb_deactive(server->super);
 973                                 goto restart;
 974                         }
 975                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 976                         rcu_read_unlock();
 977                         if (delegation != NULL) {
 978                                 delegation = nfs_detach_delegation(NFS_I(inode),
 979                                         delegation, server);
 980                                 if (delegation != NULL)
 981                                         nfs_free_delegation(delegation);
 982                         }
 983                         iput(inode);
 984                         nfs_sb_deactive(server->super);
 985                         cond_resched();
 986                         goto restart;
 987                 }
 988         }
 989         rcu_read_unlock();
 990 }
 991
 992 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
 993 {
 994         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
 995                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
 996                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
 997 }
 998
 999 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1000             struct nfs_delegation *delegation)
1001 {
1002         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1003                 return;
1004         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1005         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1006         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1007 }
1008
1009 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1010                 struct inode *inode)
1011 {
1012         struct nfs_delegation *delegation;
1013
1014         rcu_read_lock();
1015         delegation = rcu_dereference(NFS_I(inode)->delegation);
1016         if (delegation)
1017                 nfs_mark_test_expired_delegation(server, delegation);
1018         rcu_read_unlock();
1019
1020 }
1021
1022 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1023 {
1024         struct nfs_delegation *delegation;
1025
1026         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1027                 nfs_mark_test_expired_delegation(server, delegation);
1028 }
1029
1030 /**
1031  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1032  * @clp: nfs_client to process
1033  *
1034  * Iterates through all the delegations associated with this server and
1035  * marks them as needing to be checked for validity.
1036  */
1037 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1038 {
1039         struct nfs_server *server;
1040
1041         rcu_read_lock();
1042         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1043                 nfs_delegation_mark_test_expired_server(server);
1044         rcu_read_unlock();
1045 }
1046
1047 /**
1048  * nfs_test_expired_all_delegations - test all delegations for a client
1049  * @clp: nfs_client to process
1050  *
1051  * Helper for handling "recallable state revoked" status from server.
1052  */
1053 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1054 {
1055         nfs_mark_test_expired_all_delegations(clp);
1056         nfs4_schedule_state_manager(clp);
1057 }
1058
1059 static void
1060 nfs_delegation_test_free_expired(struct inode *inode,
1061                 nfs4_stateid *stateid,
1062                 const struct cred *cred)
1063 {
1064         struct nfs_server *server = NFS_SERVER(inode);
1065         const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1066         int status;
1067
1068         if (!cred)
1069                 return;
1070         status = ops->test_and_free_expired(server, stateid, cred);
1071         if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1072                 nfs_remove_bad_delegation(inode, stateid);
1073 }
1074
1075 /**
1076  * nfs_reap_expired_delegations - reap expired delegations
1077  * @clp: nfs_client to process
1078  *
1079  * Iterates through all the delegations associated with this server and
1080  * checks if they have may have been revoked. This function is usually
1081  * expected to be called in cases where the server may have lost its
1082  * lease.
1083  */
1084 void nfs_reap_expired_delegations(struct nfs_client *clp)
1085 {
1086         struct nfs_delegation *delegation;
1087         struct nfs_server *server;
1088         struct inode *inode;
1089         const struct cred *cred;
1090         nfs4_stateid stateid;
1091
1092 restart:
1093         rcu_read_lock();
1094         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1095                 list_for_each_entry_rcu(delegation, &server->delegations,
1096                                                                 super_list) {
1097                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
1098                                                 &delegation->flags) ||
1099                             test_bit(NFS_DELEGATION_RETURNING,
1100                                                 &delegation->flags) ||
1101                             test_bit(NFS_DELEGATION_TEST_EXPIRED,
1102                                                 &delegation->flags) == 0)
1103                                 continue;
1104                         if (!nfs_sb_active(server->super))
1105                                 break; /* continue in outer loop */
1106                         inode = nfs_delegation_grab_inode(delegation);
1107                         if (inode == NULL) {
1108                                 rcu_read_unlock();
1109                                 nfs_sb_deactive(server->super);
1110                                 goto restart;
1111                         }
1112                         cred = get_cred_rcu(delegation->cred);
1113                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1114                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1115                         rcu_read_unlock();
1116                         nfs_delegation_test_free_expired(inode, &stateid, cred);
1117                         put_cred(cred);
1118                         if (nfs4_server_rebooted(clp)) {
1119                                 nfs_inode_mark_test_expired_delegation(server,inode);
1120                                 iput(inode);
1121                                 nfs_sb_deactive(server->super);
1122                                 return;
1123                         }
1124                         iput(inode);
1125                         nfs_sb_deactive(server->super);
1126                         cond_resched();
1127                         goto restart;
1128                 }
1129         }
1130         rcu_read_unlock();
1131 }
1132
1133 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1134                 const nfs4_stateid *stateid)
1135 {
1136         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1137         struct nfs_delegation *delegation;
1138         bool found = false;
1139
1140         rcu_read_lock();
1141         delegation = rcu_dereference(NFS_I(inode)->delegation);
1142         if (delegation &&
1143             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1144                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1145                 found = true;
1146         }
1147         rcu_read_unlock();
1148         if (found)
1149                 nfs4_schedule_state_manager(clp);
1150 }
1151
1152 /**
1153  * nfs_delegations_present - check for existence of delegations
1154  * @clp: client state handle
1155  *
1156  * Returns one if there are any nfs_delegation structures attached
1157  * to this nfs_client.
1158  */
1159 int nfs_delegations_present(struct nfs_client *clp)
1160 {
1161         struct nfs_server *server;
1162         int ret = 0;
1163
1164         rcu_read_lock();
1165         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1166                 if (!list_empty(&server->delegations)) {
1167                         ret = 1;
1168                         break;
1169                 }
1170         rcu_read_unlock();
1171         return ret;
1172 }
1173
1174 /**
1175  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1176  * @dst: stateid to refresh
1177  * @inode: inode to check
1178  *
1179  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1180  * that matches our delegation stateid. Otherwise "false" is returned.
1181  */
1182 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1183 {
1184         struct nfs_delegation *delegation;
1185         bool ret = false;
1186         if (!inode)
1187                 goto out;
1188
1189         rcu_read_lock();
1190         delegation = rcu_dereference(NFS_I(inode)->delegation);
1191         if (delegation != NULL &&
1192             nfs4_stateid_match_other(dst, &delegation->stateid)) {
1193                 dst->seqid = delegation->stateid.seqid;
1194                 ret = true;
1195         }
1196         rcu_read_unlock();
1197 out:
1198         return ret;
1199 }
1200
1201 /**
1202  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1203  * @inode: inode to check
1204  * @flags: delegation type requirement
1205  * @dst: stateid data structure to fill in
1206  * @cred: optional argument to retrieve credential
1207  *
1208  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1209  * otherwise "false" is returned.
1210  */
1211 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1212                 nfs4_stateid *dst, const struct cred **cred)
1213 {
1214         struct nfs_inode *nfsi = NFS_I(inode);
1215         struct nfs_delegation *delegation;
1216         bool ret;
1217
1218         flags &= FMODE_READ|FMODE_WRITE;
1219         rcu_read_lock();
1220         delegation = rcu_dereference(nfsi->delegation);
1221         ret = nfs4_is_valid_delegation(delegation, flags);
1222         if (ret) {
1223                 nfs4_stateid_copy(dst, &delegation->stateid);
1224                 nfs_mark_delegation_referenced(delegation);
1225                 if (cred)
1226                         *cred = get_cred(delegation->cred);
1227         }
1228         rcu_read_unlock();
1229         return ret;
1230 }
1231
1232 /**
1233  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1234  * @inode: inode to check
1235  *
1236  * This function checks the number of outstanding writes to the file
1237  * against the delegation 'space_limit' field to see if
1238  * the spec requires us to flush the file on close.
1239  */
1240 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1241 {
1242         struct nfs_inode *nfsi = NFS_I(inode);
1243         struct nfs_delegation *delegation;
1244         bool ret = true;
1245
1246         rcu_read_lock();
1247         delegation = rcu_dereference(nfsi->delegation);
1248         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1249                 goto out;
1250         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1251                 ret = false;
1252 out:
1253         rcu_read_unlock();
1254         return ret;
1255 }