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         if (!delegation->inode) {
 302                 spin_unlock(&delegation->lock);
 303                 return NULL;
 304         }
 305         set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
 306         list_del_rcu(&delegation->super_list);
 307         delegation->inode = NULL;
 308         rcu_assign_pointer(nfsi->delegation, NULL);
 309         spin_unlock(&delegation->lock);
 310         return delegation;
 311 }
 312
 313 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
 314                 struct nfs_delegation *delegation,
 315                 struct nfs_server *server)
 316 {
 317         struct nfs_client *clp = server->nfs_client;
 318
 319         spin_lock(&clp->cl_lock);
 320         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
 321         spin_unlock(&clp->cl_lock);
 322         return delegation;
 323 }
 324
 325 static struct nfs_delegation *
 326 nfs_inode_detach_delegation(struct inode *inode)
 327 {
 328         struct nfs_inode *nfsi = NFS_I(inode);
 329         struct nfs_server *server = NFS_SERVER(inode);
 330         struct nfs_delegation *delegation;
 331
 332         delegation = nfs_start_delegation_return(nfsi);
 333         if (delegation == NULL)
 334                 return NULL;
 335         return nfs_detach_delegation(nfsi, delegation, server);
 336 }
 337
 338 static void
 339 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
 340                 const struct nfs_delegation *update)
 341 {
 342         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
 343                 delegation->stateid.seqid = update->stateid.seqid;
 344                 smp_wmb();
 345                 delegation->type = update->type;
 346                 clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 347         }
 348 }
 349
 350 /**
 351  * nfs_inode_set_delegation - set up a delegation on an inode
 352  * @inode: inode to which delegation applies
 353  * @cred: cred to use for subsequent delegation processing
 354  * @type: delegation type
 355  * @stateid: delegation stateid
 356  * @pagemod_limit: write delegation "space_limit"
 357  *
 358  * Returns zero on success, or a negative errno value.
 359  */
 360 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
 361                                   fmode_t type,
 362                                   const nfs4_stateid *stateid,
 363                                   unsigned long pagemod_limit)
 364 {
 365         struct nfs_server *server = NFS_SERVER(inode);
 366         struct nfs_client *clp = server->nfs_client;
 367         struct nfs_inode *nfsi = NFS_I(inode);
 368         struct nfs_delegation *delegation, *old_delegation;
 369         struct nfs_delegation *freeme = NULL;
 370         int status = 0;
 371
 372         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
 373         if (delegation == NULL)
 374                 return -ENOMEM;
 375         nfs4_stateid_copy(&delegation->stateid, stateid);
 376         delegation->type = type;
 377         delegation->pagemod_limit = pagemod_limit;
 378         delegation->change_attr = inode_peek_iversion_raw(inode);
 379         delegation->cred = get_cred(cred);
 380         delegation->inode = inode;
 381         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
 382         spin_lock_init(&delegation->lock);
 383
 384         spin_lock(&clp->cl_lock);
 385         old_delegation = rcu_dereference_protected(nfsi->delegation,
 386                                         lockdep_is_held(&clp->cl_lock));
 387         if (old_delegation != NULL) {
 388                 /* Is this an update of the existing delegation? */
 389                 if (nfs4_stateid_match_other(&old_delegation->stateid,
 390                                         &delegation->stateid)) {
 391                         spin_lock(&old_delegation->lock);
 392                         nfs_update_inplace_delegation(old_delegation,
 393                                         delegation);
 394                         spin_unlock(&old_delegation->lock);
 395                         goto out;
 396                 }
 397                 /*
 398                  * Deal with broken servers that hand out two
 399                  * delegations for the same file.
 400                  * Allow for upgrades to a WRITE delegation, but
 401                  * nothing else.
 402                  */
 403                 dfprintk(FILE, "%s: server %s handed out "
 404                                 "a duplicate delegation!\n",
 405                                 __func__, clp->cl_hostname);
 406                 if (delegation->type == old_delegation->type ||
 407                     !(delegation->type & FMODE_WRITE)) {
 408                         freeme = delegation;
 409                         delegation = NULL;
 410                         goto out;
 411                 }
 412                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
 413                                         &old_delegation->flags))
 414                         goto out;
 415                 freeme = nfs_detach_delegation_locked(nfsi,
 416                                 old_delegation, clp);
 417                 if (freeme == NULL)
 418                         goto out;
 419         }
 420         list_add_tail_rcu(&delegation->super_list, &server->delegations);
 421         rcu_assign_pointer(nfsi->delegation, delegation);
 422         delegation = NULL;
 423
 424         trace_nfs4_set_delegation(inode, type);
 425
 426         spin_lock(&inode->i_lock);
 427         if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME))
 428                 NFS_I(inode)->cache_validity |= NFS_INO_REVAL_FORCED;
 429         spin_unlock(&inode->i_lock);
 430 out:
 431         spin_unlock(&clp->cl_lock);
 432         if (delegation != NULL)
 433                 nfs_free_delegation(delegation);
 434         if (freeme != NULL)
 435                 nfs_do_return_delegation(inode, freeme, 0);
 436         return status;
 437 }
 438
 439 /*
 440  * Basic procedure for returning a delegation to the server
 441  */
 442 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
 443 {
 444         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
 445         struct nfs_inode *nfsi = NFS_I(inode);
 446         int err = 0;
 447
 448         if (delegation == NULL)
 449                 return 0;
 450         do {
 451                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
 452                         break;
 453                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
 454                                 delegation->type);
 455                 if (!issync || err != -EAGAIN)
 456                         break;
 457                 /*
 458                  * Guard against state recovery
 459                  */
 460                 err = nfs4_wait_clnt_recover(clp);
 461         } while (err == 0);
 462
 463         if (err) {
 464                 nfs_abort_delegation_return(delegation, clp);
 465                 goto out;
 466         }
 467         if (!nfs_detach_delegation(nfsi, delegation, NFS_SERVER(inode)))
 468                 goto out;
 469
 470         err = nfs_do_return_delegation(inode, delegation, issync);
 471 out:
 472         return err;
 473 }
 474
 475 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
 476 {
 477         bool ret = false;
 478
 479         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 480                 goto out;
 481         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
 482                 ret = true;
 483         if (test_and_clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) && !ret) {
 484                 struct inode *inode;
 485
 486                 spin_lock(&delegation->lock);
 487                 inode = delegation->inode;
 488                 if (inode && list_empty(&NFS_I(inode)->open_files))
 489                         ret = true;
 490                 spin_unlock(&delegation->lock);
 491         }
 492 out:
 493         return ret;
 494 }
 495
 496 /**
 497  * nfs_client_return_marked_delegations - return previously marked delegations
 498  * @clp: nfs_client to process
 499  *
 500  * Note that this function is designed to be called by the state
 501  * manager thread. For this reason, it cannot flush the dirty data,
 502  * since that could deadlock in case of a state recovery error.
 503  *
 504  * Returns zero on success, or a negative errno value.
 505  */
 506 int nfs_client_return_marked_delegations(struct nfs_client *clp)
 507 {
 508         struct nfs_delegation *delegation;
 509         struct nfs_delegation *prev;
 510         struct nfs_server *server;
 511         struct inode *inode;
 512         struct inode *place_holder = NULL;
 513         struct nfs_delegation *place_holder_deleg = NULL;
 514         int err = 0;
 515
 516 restart:
 517         /*
 518          * To avoid quadratic looping we hold a reference
 519          * to an inode place_holder.  Each time we restart, we
 520          * list nfs_servers from the server of that inode, and
 521          * delegation in the server from the delegations of that
 522          * inode.
 523          * prev is an RCU-protected pointer to a delegation which
 524          * wasn't marked for return and might be a good choice for
 525          * the next place_holder.
 526          */
 527         rcu_read_lock();
 528         prev = NULL;
 529         if (place_holder)
 530                 server = NFS_SERVER(place_holder);
 531         else
 532                 server = list_entry_rcu(clp->cl_superblocks.next,
 533                                         struct nfs_server, client_link);
 534         list_for_each_entry_from_rcu(server, &clp->cl_superblocks, client_link) {
 535                 delegation = NULL;
 536                 if (place_holder && server == NFS_SERVER(place_holder))
 537                         delegation = rcu_dereference(NFS_I(place_holder)->delegation);
 538                 if (!delegation || delegation != place_holder_deleg)
 539                         delegation = list_entry_rcu(server->delegations.next,
 540                                                     struct nfs_delegation, super_list);
 541                 list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
 542                         struct inode *to_put = NULL;
 543
 544                         if (!nfs_delegation_need_return(delegation)) {
 545                                 prev = delegation;
 546                                 continue;
 547                         }
 548                         if (!nfs_sb_active(server->super))
 549                                 break; /* continue in outer loop */
 550
 551                         if (prev) {
 552                                 struct inode *tmp;
 553
 554                                 tmp = nfs_delegation_grab_inode(prev);
 555                                 if (tmp) {
 556                                         to_put = place_holder;
 557                                         place_holder = tmp;
 558                                         place_holder_deleg = prev;
 559                                 }
 560                         }
 561
 562                         inode = nfs_delegation_grab_inode(delegation);
 563                         if (inode == NULL) {
 564                                 rcu_read_unlock();
 565                                 if (to_put)
 566                                         iput(to_put);
 567                                 nfs_sb_deactive(server->super);
 568                                 goto restart;
 569                         }
 570                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 571                         rcu_read_unlock();
 572
 573                         if (to_put)
 574                                 iput(to_put);
 575
 576                         err = nfs_end_delegation_return(inode, delegation, 0);
 577                         iput(inode);
 578                         nfs_sb_deactive(server->super);
 579                         cond_resched();
 580                         if (!err)
 581                                 goto restart;
 582                         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
 583                         if (place_holder)
 584                                 iput(place_holder);
 585                         return err;
 586                 }
 587         }
 588         rcu_read_unlock();
 589         if (place_holder)
 590                 iput(place_holder);
 591         return 0;
 592 }
 593
 594 /**
 595  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
 596  * @inode: inode to process
 597  *
 598  * Does not protect against delegation reclaims, therefore really only safe
 599  * to be called from nfs4_clear_inode(). Guaranteed to always free
 600  * the delegation structure.
 601  */
 602 void nfs_inode_evict_delegation(struct inode *inode)
 603 {
 604         struct nfs_delegation *delegation;
 605
 606         delegation = nfs_inode_detach_delegation(inode);
 607         if (delegation != NULL) {
 608                 set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
 609                 nfs_do_return_delegation(inode, delegation, 1);
 610         }
 611 }
 612
 613 /**
 614  * nfs_inode_return_delegation - synchronously return a delegation
 615  * @inode: inode to process
 616  *
 617  * This routine will always flush any dirty data to disk on the
 618  * assumption that if we need to return the delegation, then
 619  * we should stop caching.
 620  *
 621  * Returns zero on success, or a negative errno value.
 622  */
 623 int nfs4_inode_return_delegation(struct inode *inode)
 624 {
 625         struct nfs_inode *nfsi = NFS_I(inode);
 626         struct nfs_delegation *delegation;
 627         int err = 0;
 628
 629         nfs_wb_all(inode);
 630         delegation = nfs_start_delegation_return(nfsi);
 631         if (delegation != NULL)
 632                 err = nfs_end_delegation_return(inode, delegation, 1);
 633         return err;
 634 }
 635
 636 /**
 637  * nfs4_inode_make_writeable
 638  * @inode: pointer to inode
 639  *
 640  * Make the inode writeable by returning the delegation if necessary
 641  *
 642  * Returns zero on success, or a negative errno value.
 643  */
 644 int nfs4_inode_make_writeable(struct inode *inode)
 645 {
 646         if (!nfs4_has_session(NFS_SERVER(inode)->nfs_client) ||
 647             !nfs4_check_delegation(inode, FMODE_WRITE))
 648                 return nfs4_inode_return_delegation(inode);
 649         return 0;
 650 }
 651
 652 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
 653                 struct nfs_delegation *delegation)
 654 {
 655         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
 656         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 657 }
 658
 659 static void nfs_mark_return_delegation(struct nfs_server *server,
 660                 struct nfs_delegation *delegation)
 661 {
 662         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
 663         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
 664 }
 665
 666 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
 667 {
 668         struct nfs_delegation *delegation;
 669         bool ret = false;
 670
 671         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 672                 nfs_mark_return_delegation(server, delegation);
 673                 ret = true;
 674         }
 675         return ret;
 676 }
 677
 678 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
 679 {
 680         struct nfs_server *server;
 681
 682         rcu_read_lock();
 683         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 684                 nfs_server_mark_return_all_delegations(server);
 685         rcu_read_unlock();
 686 }
 687
 688 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
 689 {
 690         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
 691                 nfs4_schedule_state_manager(clp);
 692 }
 693
 694 /**
 695  * nfs_expire_all_delegations
 696  * @clp: client to process
 697  *
 698  */
 699 void nfs_expire_all_delegations(struct nfs_client *clp)
 700 {
 701         nfs_client_mark_return_all_delegations(clp);
 702         nfs_delegation_run_state_manager(clp);
 703 }
 704
 705 /**
 706  * nfs_super_return_all_delegations - return delegations for one superblock
 707  * @server: pointer to nfs_server to process
 708  *
 709  */
 710 void nfs_server_return_all_delegations(struct nfs_server *server)
 711 {
 712         struct nfs_client *clp = server->nfs_client;
 713         bool need_wait;
 714
 715         if (clp == NULL)
 716                 return;
 717
 718         rcu_read_lock();
 719         need_wait = nfs_server_mark_return_all_delegations(server);
 720         rcu_read_unlock();
 721
 722         if (need_wait) {
 723                 nfs4_schedule_state_manager(clp);
 724                 nfs4_wait_clnt_recover(clp);
 725         }
 726 }
 727
 728 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
 729                                                  fmode_t flags)
 730 {
 731         struct nfs_delegation *delegation;
 732
 733         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 734                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
 735                         continue;
 736                 if (delegation->type & flags)
 737                         nfs_mark_return_if_closed_delegation(server, delegation);
 738         }
 739 }
 740
 741 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
 742                                                         fmode_t flags)
 743 {
 744         struct nfs_server *server;
 745
 746         rcu_read_lock();
 747         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 748                 nfs_mark_return_unused_delegation_types(server, flags);
 749         rcu_read_unlock();
 750 }
 751
 752 static void nfs_mark_delegation_revoked(struct nfs_server *server,
 753                 struct nfs_delegation *delegation)
 754 {
 755         set_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
 756         delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
 757         nfs_mark_return_delegation(server, delegation);
 758 }
 759
 760 static bool nfs_revoke_delegation(struct inode *inode,
 761                 const nfs4_stateid *stateid)
 762 {
 763         struct nfs_delegation *delegation;
 764         nfs4_stateid tmp;
 765         bool ret = false;
 766
 767         rcu_read_lock();
 768         delegation = rcu_dereference(NFS_I(inode)->delegation);
 769         if (delegation == NULL)
 770                 goto out;
 771         if (stateid == NULL) {
 772                 nfs4_stateid_copy(&tmp, &delegation->stateid);
 773                 stateid = &tmp;
 774         } else {
 775                 if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
 776                         goto out;
 777                 spin_lock(&delegation->lock);
 778                 if (stateid->seqid) {
 779                         if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
 780                                 spin_unlock(&delegation->lock);
 781                                 goto out;
 782                         }
 783                         delegation->stateid.seqid = stateid->seqid;
 784                 }
 785                 spin_unlock(&delegation->lock);
 786         }
 787         nfs_mark_delegation_revoked(NFS_SERVER(inode), delegation);
 788         ret = true;
 789 out:
 790         rcu_read_unlock();
 791         if (ret)
 792                 nfs_inode_find_state_and_recover(inode, stateid);
 793         return ret;
 794 }
 795
 796 void nfs_remove_bad_delegation(struct inode *inode,
 797                 const nfs4_stateid *stateid)
 798 {
 799         struct nfs_delegation *delegation;
 800
 801         if (!nfs_revoke_delegation(inode, stateid))
 802                 return;
 803         delegation = nfs_inode_detach_delegation(inode);
 804         if (delegation)
 805                 nfs_free_delegation(delegation);
 806 }
 807 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 808
 809 /**
 810  * nfs_expire_unused_delegation_types
 811  * @clp: client to process
 812  * @flags: delegation types to expire
 813  *
 814  */
 815 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
 816 {
 817         nfs_client_mark_return_unused_delegation_types(clp, flags);
 818         nfs_delegation_run_state_manager(clp);
 819 }
 820
 821 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
 822 {
 823         struct nfs_delegation *delegation;
 824
 825         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 826                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
 827                         continue;
 828                 nfs_mark_return_if_closed_delegation(server, delegation);
 829         }
 830 }
 831
 832 /**
 833  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
 834  * @clp: nfs_client to process
 835  *
 836  */
 837 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
 838 {
 839         struct nfs_server *server;
 840
 841         rcu_read_lock();
 842         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 843                 nfs_mark_return_unreferenced_delegations(server);
 844         rcu_read_unlock();
 845
 846         nfs_delegation_run_state_manager(clp);
 847 }
 848
 849 /**
 850  * nfs_async_inode_return_delegation - asynchronously return a delegation
 851  * @inode: inode to process
 852  * @stateid: state ID information
 853  *
 854  * Returns zero on success, or a negative errno value.
 855  */
 856 int nfs_async_inode_return_delegation(struct inode *inode,
 857                                       const nfs4_stateid *stateid)
 858 {
 859         struct nfs_server *server = NFS_SERVER(inode);
 860         struct nfs_client *clp = server->nfs_client;
 861         struct nfs_delegation *delegation;
 862
 863         rcu_read_lock();
 864         delegation = nfs4_get_valid_delegation(inode);
 865         if (delegation == NULL)
 866                 goto out_enoent;
 867         if (stateid != NULL &&
 868             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
 869                 goto out_enoent;
 870         nfs_mark_return_delegation(server, delegation);
 871         rcu_read_unlock();
 872
 873         nfs_delegation_run_state_manager(clp);
 874         return 0;
 875 out_enoent:
 876         rcu_read_unlock();
 877         return -ENOENT;
 878 }
 879
 880 static struct inode *
 881 nfs_delegation_find_inode_server(struct nfs_server *server,
 882                                  const struct nfs_fh *fhandle)
 883 {
 884         struct nfs_delegation *delegation;
 885         struct inode *freeme, *res = NULL;
 886
 887         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 888                 spin_lock(&delegation->lock);
 889                 if (delegation->inode != NULL &&
 890                     !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
 891                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
 892                         freeme = igrab(delegation->inode);
 893                         if (freeme && nfs_sb_active(freeme->i_sb))
 894                                 res = freeme;
 895                         spin_unlock(&delegation->lock);
 896                         if (res != NULL)
 897                                 return res;
 898                         if (freeme) {
 899                                 rcu_read_unlock();
 900                                 iput(freeme);
 901                                 rcu_read_lock();
 902                         }
 903                         return ERR_PTR(-EAGAIN);
 904                 }
 905                 spin_unlock(&delegation->lock);
 906         }
 907         return ERR_PTR(-ENOENT);
 908 }
 909
 910 /**
 911  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
 912  * @clp: client state handle
 913  * @fhandle: filehandle from a delegation recall
 914  *
 915  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
 916  * cannot be found.
 917  */
 918 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
 919                                         const struct nfs_fh *fhandle)
 920 {
 921         struct nfs_server *server;
 922         struct inode *res;
 923
 924         rcu_read_lock();
 925         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 926                 res = nfs_delegation_find_inode_server(server, fhandle);
 927                 if (res != ERR_PTR(-ENOENT)) {
 928                         rcu_read_unlock();
 929                         return res;
 930                 }
 931         }
 932         rcu_read_unlock();
 933         return ERR_PTR(-ENOENT);
 934 }
 935
 936 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
 937 {
 938         struct nfs_delegation *delegation;
 939
 940         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 941                 /*
 942                  * If the delegation may have been admin revoked, then we
 943                  * cannot reclaim it.
 944                  */
 945                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
 946                         continue;
 947                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 948         }
 949 }
 950
 951 /**
 952  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
 953  * @clp: nfs_client to process
 954  *
 955  */
 956 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 957 {
 958         struct nfs_server *server;
 959
 960         rcu_read_lock();
 961         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 962                 nfs_delegation_mark_reclaim_server(server);
 963         rcu_read_unlock();
 964 }
 965
 966 /**
 967  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
 968  * @clp: nfs_client to process
 969  *
 970  */
 971 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 972 {
 973         struct nfs_delegation *delegation;
 974         struct nfs_server *server;
 975         struct inode *inode;
 976
 977 restart:
 978         rcu_read_lock();
 979         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 980                 list_for_each_entry_rcu(delegation, &server->delegations,
 981                                                                 super_list) {
 982                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
 983                                                 &delegation->flags) ||
 984                             test_bit(NFS_DELEGATION_RETURNING,
 985                                                 &delegation->flags) ||
 986                             test_bit(NFS_DELEGATION_NEED_RECLAIM,
 987                                                 &delegation->flags) == 0)
 988                                 continue;
 989                         if (!nfs_sb_active(server->super))
 990                                 break; /* continue in outer loop */
 991                         inode = nfs_delegation_grab_inode(delegation);
 992                         if (inode == NULL) {
 993                                 rcu_read_unlock();
 994                                 nfs_sb_deactive(server->super);
 995                                 goto restart;
 996                         }
 997                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 998                         rcu_read_unlock();
 999                         if (delegation != NULL) {
1000                                 delegation = nfs_detach_delegation(NFS_I(inode),
1001                                         delegation, server);
1002                                 if (delegation != NULL)
1003                                         nfs_free_delegation(delegation);
1004                         }
1005                         iput(inode);
1006                         nfs_sb_deactive(server->super);
1007                         cond_resched();
1008                         goto restart;
1009                 }
1010         }
1011         rcu_read_unlock();
1012 }
1013
1014 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1015 {
1016         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1017                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
1018                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
1019 }
1020
1021 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1022             struct nfs_delegation *delegation)
1023 {
1024         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1025                 return;
1026         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1027         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1028         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1029 }
1030
1031 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1032                 struct inode *inode)
1033 {
1034         struct nfs_delegation *delegation;
1035
1036         rcu_read_lock();
1037         delegation = rcu_dereference(NFS_I(inode)->delegation);
1038         if (delegation)
1039                 nfs_mark_test_expired_delegation(server, delegation);
1040         rcu_read_unlock();
1041
1042 }
1043
1044 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1045 {
1046         struct nfs_delegation *delegation;
1047
1048         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1049                 nfs_mark_test_expired_delegation(server, delegation);
1050 }
1051
1052 /**
1053  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1054  * @clp: nfs_client to process
1055  *
1056  * Iterates through all the delegations associated with this server and
1057  * marks them as needing to be checked for validity.
1058  */
1059 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1060 {
1061         struct nfs_server *server;
1062
1063         rcu_read_lock();
1064         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1065                 nfs_delegation_mark_test_expired_server(server);
1066         rcu_read_unlock();
1067 }
1068
1069 /**
1070  * nfs_test_expired_all_delegations - test all delegations for a client
1071  * @clp: nfs_client to process
1072  *
1073  * Helper for handling "recallable state revoked" status from server.
1074  */
1075 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1076 {
1077         nfs_mark_test_expired_all_delegations(clp);
1078         nfs4_schedule_state_manager(clp);
1079 }
1080
1081 static void
1082 nfs_delegation_test_free_expired(struct inode *inode,
1083                 nfs4_stateid *stateid,
1084                 const struct cred *cred)
1085 {
1086         struct nfs_server *server = NFS_SERVER(inode);
1087         const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1088         int status;
1089
1090         if (!cred)
1091                 return;
1092         status = ops->test_and_free_expired(server, stateid, cred);
1093         if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1094                 nfs_remove_bad_delegation(inode, stateid);
1095 }
1096
1097 /**
1098  * nfs_reap_expired_delegations - reap expired delegations
1099  * @clp: nfs_client to process
1100  *
1101  * Iterates through all the delegations associated with this server and
1102  * checks if they have may have been revoked. This function is usually
1103  * expected to be called in cases where the server may have lost its
1104  * lease.
1105  */
1106 void nfs_reap_expired_delegations(struct nfs_client *clp)
1107 {
1108         struct nfs_delegation *delegation;
1109         struct nfs_server *server;
1110         struct inode *inode;
1111         const struct cred *cred;
1112         nfs4_stateid stateid;
1113
1114 restart:
1115         rcu_read_lock();
1116         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1117                 list_for_each_entry_rcu(delegation, &server->delegations,
1118                                                                 super_list) {
1119                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
1120                                                 &delegation->flags) ||
1121                             test_bit(NFS_DELEGATION_RETURNING,
1122                                                 &delegation->flags) ||
1123                             test_bit(NFS_DELEGATION_TEST_EXPIRED,
1124                                                 &delegation->flags) == 0)
1125                                 continue;
1126                         if (!nfs_sb_active(server->super))
1127                                 break; /* continue in outer loop */
1128                         inode = nfs_delegation_grab_inode(delegation);
1129                         if (inode == NULL) {
1130                                 rcu_read_unlock();
1131                                 nfs_sb_deactive(server->super);
1132                                 goto restart;
1133                         }
1134                         cred = get_cred_rcu(delegation->cred);
1135                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1136                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1137                         rcu_read_unlock();
1138                         nfs_delegation_test_free_expired(inode, &stateid, cred);
1139                         put_cred(cred);
1140                         if (nfs4_server_rebooted(clp)) {
1141                                 nfs_inode_mark_test_expired_delegation(server,inode);
1142                                 iput(inode);
1143                                 nfs_sb_deactive(server->super);
1144                                 return;
1145                         }
1146                         iput(inode);
1147                         nfs_sb_deactive(server->super);
1148                         cond_resched();
1149                         goto restart;
1150                 }
1151         }
1152         rcu_read_unlock();
1153 }
1154
1155 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1156                 const nfs4_stateid *stateid)
1157 {
1158         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1159         struct nfs_delegation *delegation;
1160         bool found = false;
1161
1162         rcu_read_lock();
1163         delegation = rcu_dereference(NFS_I(inode)->delegation);
1164         if (delegation &&
1165             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1166                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1167                 found = true;
1168         }
1169         rcu_read_unlock();
1170         if (found)
1171                 nfs4_schedule_state_manager(clp);
1172 }
1173
1174 /**
1175  * nfs_delegations_present - check for existence of delegations
1176  * @clp: client state handle
1177  *
1178  * Returns one if there are any nfs_delegation structures attached
1179  * to this nfs_client.
1180  */
1181 int nfs_delegations_present(struct nfs_client *clp)
1182 {
1183         struct nfs_server *server;
1184         int ret = 0;
1185
1186         rcu_read_lock();
1187         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1188                 if (!list_empty(&server->delegations)) {
1189                         ret = 1;
1190                         break;
1191                 }
1192         rcu_read_unlock();
1193         return ret;
1194 }
1195
1196 /**
1197  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1198  * @dst: stateid to refresh
1199  * @inode: inode to check
1200  *
1201  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1202  * that matches our delegation stateid. Otherwise "false" is returned.
1203  */
1204 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1205 {
1206         struct nfs_delegation *delegation;
1207         bool ret = false;
1208         if (!inode)
1209                 goto out;
1210
1211         rcu_read_lock();
1212         delegation = rcu_dereference(NFS_I(inode)->delegation);
1213         if (delegation != NULL &&
1214             nfs4_stateid_match_other(dst, &delegation->stateid) &&
1215             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1216                 dst->seqid = delegation->stateid.seqid;
1217                 ret = true;
1218         }
1219         rcu_read_unlock();
1220 out:
1221         return ret;
1222 }
1223
1224 /**
1225  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1226  * @inode: inode to check
1227  * @flags: delegation type requirement
1228  * @dst: stateid data structure to fill in
1229  * @cred: optional argument to retrieve credential
1230  *
1231  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1232  * otherwise "false" is returned.
1233  */
1234 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1235                 nfs4_stateid *dst, const struct cred **cred)
1236 {
1237         struct nfs_inode *nfsi = NFS_I(inode);
1238         struct nfs_delegation *delegation;
1239         bool ret;
1240
1241         flags &= FMODE_READ|FMODE_WRITE;
1242         rcu_read_lock();
1243         delegation = rcu_dereference(nfsi->delegation);
1244         ret = nfs4_is_valid_delegation(delegation, flags);
1245         if (ret) {
1246                 nfs4_stateid_copy(dst, &delegation->stateid);
1247                 nfs_mark_delegation_referenced(delegation);
1248                 if (cred)
1249                         *cred = get_cred(delegation->cred);
1250         }
1251         rcu_read_unlock();
1252         return ret;
1253 }
1254
1255 /**
1256  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1257  * @inode: inode to check
1258  *
1259  * This function checks the number of outstanding writes to the file
1260  * against the delegation 'space_limit' field to see if
1261  * the spec requires us to flush the file on close.
1262  */
1263 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1264 {
1265         struct nfs_inode *nfsi = NFS_I(inode);
1266         struct nfs_delegation *delegation;
1267         bool ret = true;
1268
1269         rcu_read_lock();
1270         delegation = rcu_dereference(nfsi->delegation);
1271         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1272                 goto out;
1273         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1274                 ret = false;
1275 out:
1276         rcu_read_unlock();
1277         return ret;
1278 }