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 if (!nfs4_stateid_match(stateid, &delegation->stateid))
 775                 goto out;
 776         nfs_mark_delegation_revoked(NFS_SERVER(inode), delegation);
 777         ret = true;
 778 out:
 779         rcu_read_unlock();
 780         if (ret)
 781                 nfs_inode_find_state_and_recover(inode, stateid);
 782         return ret;
 783 }
 784
 785 void nfs_remove_bad_delegation(struct inode *inode,
 786                 const nfs4_stateid *stateid)
 787 {
 788         struct nfs_delegation *delegation;
 789
 790         if (!nfs_revoke_delegation(inode, stateid))
 791                 return;
 792         delegation = nfs_inode_detach_delegation(inode);
 793         if (delegation)
 794                 nfs_free_delegation(delegation);
 795 }
 796 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
 797
 798 /**
 799  * nfs_expire_unused_delegation_types
 800  * @clp: client to process
 801  * @flags: delegation types to expire
 802  *
 803  */
 804 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
 805 {
 806         nfs_client_mark_return_unused_delegation_types(clp, flags);
 807         nfs_delegation_run_state_manager(clp);
 808 }
 809
 810 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
 811 {
 812         struct nfs_delegation *delegation;
 813
 814         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 815                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
 816                         continue;
 817                 nfs_mark_return_if_closed_delegation(server, delegation);
 818         }
 819 }
 820
 821 /**
 822  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
 823  * @clp: nfs_client to process
 824  *
 825  */
 826 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
 827 {
 828         struct nfs_server *server;
 829
 830         rcu_read_lock();
 831         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 832                 nfs_mark_return_unreferenced_delegations(server);
 833         rcu_read_unlock();
 834
 835         nfs_delegation_run_state_manager(clp);
 836 }
 837
 838 /**
 839  * nfs_async_inode_return_delegation - asynchronously return a delegation
 840  * @inode: inode to process
 841  * @stateid: state ID information
 842  *
 843  * Returns zero on success, or a negative errno value.
 844  */
 845 int nfs_async_inode_return_delegation(struct inode *inode,
 846                                       const nfs4_stateid *stateid)
 847 {
 848         struct nfs_server *server = NFS_SERVER(inode);
 849         struct nfs_client *clp = server->nfs_client;
 850         struct nfs_delegation *delegation;
 851
 852         rcu_read_lock();
 853         delegation = nfs4_get_valid_delegation(inode);
 854         if (delegation == NULL)
 855                 goto out_enoent;
 856         if (stateid != NULL &&
 857             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
 858                 goto out_enoent;
 859         nfs_mark_return_delegation(server, delegation);
 860         rcu_read_unlock();
 861
 862         nfs_delegation_run_state_manager(clp);
 863         return 0;
 864 out_enoent:
 865         rcu_read_unlock();
 866         return -ENOENT;
 867 }
 868
 869 static struct inode *
 870 nfs_delegation_find_inode_server(struct nfs_server *server,
 871                                  const struct nfs_fh *fhandle)
 872 {
 873         struct nfs_delegation *delegation;
 874         struct inode *freeme, *res = NULL;
 875
 876         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 877                 spin_lock(&delegation->lock);
 878                 if (delegation->inode != NULL &&
 879                     !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
 880                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
 881                         freeme = igrab(delegation->inode);
 882                         if (freeme && nfs_sb_active(freeme->i_sb))
 883                                 res = freeme;
 884                         spin_unlock(&delegation->lock);
 885                         if (res != NULL)
 886                                 return res;
 887                         if (freeme) {
 888                                 rcu_read_unlock();
 889                                 iput(freeme);
 890                                 rcu_read_lock();
 891                         }
 892                         return ERR_PTR(-EAGAIN);
 893                 }
 894                 spin_unlock(&delegation->lock);
 895         }
 896         return ERR_PTR(-ENOENT);
 897 }
 898
 899 /**
 900  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
 901  * @clp: client state handle
 902  * @fhandle: filehandle from a delegation recall
 903  *
 904  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
 905  * cannot be found.
 906  */
 907 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
 908                                         const struct nfs_fh *fhandle)
 909 {
 910         struct nfs_server *server;
 911         struct inode *res;
 912
 913         rcu_read_lock();
 914         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 915                 res = nfs_delegation_find_inode_server(server, fhandle);
 916                 if (res != ERR_PTR(-ENOENT)) {
 917                         rcu_read_unlock();
 918                         return res;
 919                 }
 920         }
 921         rcu_read_unlock();
 922         return ERR_PTR(-ENOENT);
 923 }
 924
 925 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
 926 {
 927         struct nfs_delegation *delegation;
 928
 929         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
 930                 /*
 931                  * If the delegation may have been admin revoked, then we
 932                  * cannot reclaim it.
 933                  */
 934                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
 935                         continue;
 936                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
 937         }
 938 }
 939
 940 /**
 941  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
 942  * @clp: nfs_client to process
 943  *
 944  */
 945 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
 946 {
 947         struct nfs_server *server;
 948
 949         rcu_read_lock();
 950         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
 951                 nfs_delegation_mark_reclaim_server(server);
 952         rcu_read_unlock();
 953 }
 954
 955 /**
 956  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
 957  * @clp: nfs_client to process
 958  *
 959  */
 960 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
 961 {
 962         struct nfs_delegation *delegation;
 963         struct nfs_server *server;
 964         struct inode *inode;
 965
 966 restart:
 967         rcu_read_lock();
 968         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 969                 list_for_each_entry_rcu(delegation, &server->delegations,
 970                                                                 super_list) {
 971                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
 972                                                 &delegation->flags) ||
 973                             test_bit(NFS_DELEGATION_RETURNING,
 974                                                 &delegation->flags) ||
 975                             test_bit(NFS_DELEGATION_NEED_RECLAIM,
 976                                                 &delegation->flags) == 0)
 977                                 continue;
 978                         if (!nfs_sb_active(server->super))
 979                                 break; /* continue in outer loop */
 980                         inode = nfs_delegation_grab_inode(delegation);
 981                         if (inode == NULL) {
 982                                 rcu_read_unlock();
 983                                 nfs_sb_deactive(server->super);
 984                                 goto restart;
 985                         }
 986                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
 987                         rcu_read_unlock();
 988                         if (delegation != NULL) {
 989                                 delegation = nfs_detach_delegation(NFS_I(inode),
 990                                         delegation, server);
 991                                 if (delegation != NULL)
 992                                         nfs_free_delegation(delegation);
 993                         }
 994                         iput(inode);
 995                         nfs_sb_deactive(server->super);
 996                         cond_resched();
 997                         goto restart;
 998                 }
 999         }
1000         rcu_read_unlock();
1001 }
1002
1003 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1004 {
1005         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1006                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
1007                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
1008 }
1009
1010 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1011             struct nfs_delegation *delegation)
1012 {
1013         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1014                 return;
1015         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1016         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1017         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1018 }
1019
1020 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1021                 struct inode *inode)
1022 {
1023         struct nfs_delegation *delegation;
1024
1025         rcu_read_lock();
1026         delegation = rcu_dereference(NFS_I(inode)->delegation);
1027         if (delegation)
1028                 nfs_mark_test_expired_delegation(server, delegation);
1029         rcu_read_unlock();
1030
1031 }
1032
1033 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1034 {
1035         struct nfs_delegation *delegation;
1036
1037         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1038                 nfs_mark_test_expired_delegation(server, delegation);
1039 }
1040
1041 /**
1042  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1043  * @clp: nfs_client to process
1044  *
1045  * Iterates through all the delegations associated with this server and
1046  * marks them as needing to be checked for validity.
1047  */
1048 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1049 {
1050         struct nfs_server *server;
1051
1052         rcu_read_lock();
1053         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1054                 nfs_delegation_mark_test_expired_server(server);
1055         rcu_read_unlock();
1056 }
1057
1058 /**
1059  * nfs_test_expired_all_delegations - test all delegations for a client
1060  * @clp: nfs_client to process
1061  *
1062  * Helper for handling "recallable state revoked" status from server.
1063  */
1064 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1065 {
1066         nfs_mark_test_expired_all_delegations(clp);
1067         nfs4_schedule_state_manager(clp);
1068 }
1069
1070 static void
1071 nfs_delegation_test_free_expired(struct inode *inode,
1072                 nfs4_stateid *stateid,
1073                 const struct cred *cred)
1074 {
1075         struct nfs_server *server = NFS_SERVER(inode);
1076         const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1077         int status;
1078
1079         if (!cred)
1080                 return;
1081         status = ops->test_and_free_expired(server, stateid, cred);
1082         if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1083                 nfs_remove_bad_delegation(inode, stateid);
1084 }
1085
1086 /**
1087  * nfs_reap_expired_delegations - reap expired delegations
1088  * @clp: nfs_client to process
1089  *
1090  * Iterates through all the delegations associated with this server and
1091  * checks if they have may have been revoked. This function is usually
1092  * expected to be called in cases where the server may have lost its
1093  * lease.
1094  */
1095 void nfs_reap_expired_delegations(struct nfs_client *clp)
1096 {
1097         struct nfs_delegation *delegation;
1098         struct nfs_server *server;
1099         struct inode *inode;
1100         const struct cred *cred;
1101         nfs4_stateid stateid;
1102
1103 restart:
1104         rcu_read_lock();
1105         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1106                 list_for_each_entry_rcu(delegation, &server->delegations,
1107                                                                 super_list) {
1108                         if (test_bit(NFS_DELEGATION_INODE_FREEING,
1109                                                 &delegation->flags) ||
1110                             test_bit(NFS_DELEGATION_RETURNING,
1111                                                 &delegation->flags) ||
1112                             test_bit(NFS_DELEGATION_TEST_EXPIRED,
1113                                                 &delegation->flags) == 0)
1114                                 continue;
1115                         if (!nfs_sb_active(server->super))
1116                                 break; /* continue in outer loop */
1117                         inode = nfs_delegation_grab_inode(delegation);
1118                         if (inode == NULL) {
1119                                 rcu_read_unlock();
1120                                 nfs_sb_deactive(server->super);
1121                                 goto restart;
1122                         }
1123                         cred = get_cred_rcu(delegation->cred);
1124                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1125                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1126                         rcu_read_unlock();
1127                         nfs_delegation_test_free_expired(inode, &stateid, cred);
1128                         put_cred(cred);
1129                         if (nfs4_server_rebooted(clp)) {
1130                                 nfs_inode_mark_test_expired_delegation(server,inode);
1131                                 iput(inode);
1132                                 nfs_sb_deactive(server->super);
1133                                 return;
1134                         }
1135                         iput(inode);
1136                         nfs_sb_deactive(server->super);
1137                         cond_resched();
1138                         goto restart;
1139                 }
1140         }
1141         rcu_read_unlock();
1142 }
1143
1144 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1145                 const nfs4_stateid *stateid)
1146 {
1147         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1148         struct nfs_delegation *delegation;
1149         bool found = false;
1150
1151         rcu_read_lock();
1152         delegation = rcu_dereference(NFS_I(inode)->delegation);
1153         if (delegation &&
1154             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1155                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1156                 found = true;
1157         }
1158         rcu_read_unlock();
1159         if (found)
1160                 nfs4_schedule_state_manager(clp);
1161 }
1162
1163 /**
1164  * nfs_delegations_present - check for existence of delegations
1165  * @clp: client state handle
1166  *
1167  * Returns one if there are any nfs_delegation structures attached
1168  * to this nfs_client.
1169  */
1170 int nfs_delegations_present(struct nfs_client *clp)
1171 {
1172         struct nfs_server *server;
1173         int ret = 0;
1174
1175         rcu_read_lock();
1176         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1177                 if (!list_empty(&server->delegations)) {
1178                         ret = 1;
1179                         break;
1180                 }
1181         rcu_read_unlock();
1182         return ret;
1183 }
1184
1185 /**
1186  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1187  * @dst: stateid to refresh
1188  * @inode: inode to check
1189  *
1190  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1191  * that matches our delegation stateid. Otherwise "false" is returned.
1192  */
1193 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1194 {
1195         struct nfs_delegation *delegation;
1196         bool ret = false;
1197         if (!inode)
1198                 goto out;
1199
1200         rcu_read_lock();
1201         delegation = rcu_dereference(NFS_I(inode)->delegation);
1202         if (delegation != NULL &&
1203             nfs4_stateid_match_other(dst, &delegation->stateid) &&
1204             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1205                 dst->seqid = delegation->stateid.seqid;
1206                 ret = true;
1207         }
1208         rcu_read_unlock();
1209 out:
1210         return ret;
1211 }
1212
1213 /**
1214  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1215  * @inode: inode to check
1216  * @flags: delegation type requirement
1217  * @dst: stateid data structure to fill in
1218  * @cred: optional argument to retrieve credential
1219  *
1220  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1221  * otherwise "false" is returned.
1222  */
1223 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1224                 nfs4_stateid *dst, const struct cred **cred)
1225 {
1226         struct nfs_inode *nfsi = NFS_I(inode);
1227         struct nfs_delegation *delegation;
1228         bool ret;
1229
1230         flags &= FMODE_READ|FMODE_WRITE;
1231         rcu_read_lock();
1232         delegation = rcu_dereference(nfsi->delegation);
1233         ret = nfs4_is_valid_delegation(delegation, flags);
1234         if (ret) {
1235                 nfs4_stateid_copy(dst, &delegation->stateid);
1236                 nfs_mark_delegation_referenced(delegation);
1237                 if (cred)
1238                         *cred = get_cred(delegation->cred);
1239         }
1240         rcu_read_unlock();
1241         return ret;
1242 }
1243
1244 /**
1245  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1246  * @inode: inode to check
1247  *
1248  * This function checks the number of outstanding writes to the file
1249  * against the delegation 'space_limit' field to see if
1250  * the spec requires us to flush the file on close.
1251  */
1252 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1253 {
1254         struct nfs_inode *nfsi = NFS_I(inode);
1255         struct nfs_delegation *delegation;
1256         bool ret = true;
1257
1258         rcu_read_lock();
1259         delegation = rcu_dereference(nfsi->delegation);
1260         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1261                 goto out;
1262         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1263                 ret = false;
1264 out:
1265         rcu_read_unlock();
1266         return ret;
1267 }