fs/xfs/xfs_qm.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_alloc.h"
  26 #include "xfs_quota.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_ialloc_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_ialloc.h"
  33 #include "xfs_itable.h"
  34 #include "xfs_rtalloc.h"
  35 #include "xfs_error.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_attr.h"
  38 #include "xfs_buf_item.h"
  39 #include "xfs_trans_space.h"
  40 #include "xfs_utils.h"
  41 #include "xfs_qm.h"
  42 #include "xfs_trace.h"
  43 #include "xfs_icache.h"
  44 #include "xfs_cksum.h"
  45
  46 /*
  47  * The global quota manager. There is only one of these for the entire
  48  * system, _not_ one per file system. XQM keeps track of the overall
  49  * quota functionality, including maintaining the freelist and hash
  50  * tables of dquots.
  51  */
  52 STATIC int      xfs_qm_init_quotainos(xfs_mount_t *);
  53 STATIC int      xfs_qm_init_quotainfo(xfs_mount_t *);
  54 STATIC int      xfs_qm_shake(struct shrinker *, struct shrink_control *);
  55
  56 /*
  57  * We use the batch lookup interface to iterate over the dquots as it
  58  * currently is the only interface into the radix tree code that allows
  59  * fuzzy lookups instead of exact matches.  Holding the lock over multiple
  60  * operations is fine as all callers are used either during mount/umount
  61  * or quotaoff.
  62  */
  63 #define XFS_DQ_LOOKUP_BATCH     32
  64
  65 STATIC int
  66 xfs_qm_dquot_walk(
  67         struct xfs_mount        *mp,
  68         int                     type,
  69         int                     (*execute)(struct xfs_dquot *dqp, void *data),
  70         void                    *data)
  71 {
  72         struct xfs_quotainfo    *qi = mp->m_quotainfo;
  73         struct radix_tree_root  *tree = XFS_DQUOT_TREE(qi, type);
  74         uint32_t                next_index;
  75         int                     last_error = 0;
  76         int                     skipped;
  77         int                     nr_found;
  78
  79 restart:
  80         skipped = 0;
  81         next_index = 0;
  82         nr_found = 0;
  83
  84         while (1) {
  85                 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
  86                 int             error = 0;
  87                 int             i;
  88
  89                 mutex_lock(&qi->qi_tree_lock);
  90                 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
  91                                         next_index, XFS_DQ_LOOKUP_BATCH);
  92                 if (!nr_found) {
  93                         mutex_unlock(&qi->qi_tree_lock);
  94                         break;
  95                 }
  96
  97                 for (i = 0; i < nr_found; i++) {
  98                         struct xfs_dquot *dqp = batch[i];
  99
 100                         next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
 101
 102                         error = execute(batch[i], data);
 103                         if (error == EAGAIN) {
 104                                 skipped++;
 105                                 continue;
 106                         }
 107                         if (error && last_error != EFSCORRUPTED)
 108                                 last_error = error;
 109                 }
 110
 111                 mutex_unlock(&qi->qi_tree_lock);
 112
 113                 /* bail out if the filesystem is corrupted.  */
 114                 if (last_error == EFSCORRUPTED) {
 115                         skipped = 0;
 116                         break;
 117                 }
 118         }
 119
 120         if (skipped) {
 121                 delay(1);
 122                 goto restart;
 123         }
 124
 125         return last_error;
 126 }
 127
 128
 129 /*
 130  * Purge a dquot from all tracking data structures and free it.
 131  */
 132 STATIC int
 133 xfs_qm_dqpurge(
 134         struct xfs_dquot        *dqp,
 135         void                    *data)
 136 {
 137         struct xfs_mount        *mp = dqp->q_mount;
 138         struct xfs_quotainfo    *qi = mp->m_quotainfo;
 139         struct xfs_dquot        *gdqp = NULL;
 140
 141         xfs_dqlock(dqp);
 142         if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 143                 xfs_dqunlock(dqp);
 144                 return EAGAIN;
 145         }
 146
 147         /*
 148          * If this quota has a group hint attached, prepare for releasing it
 149          * now.
 150          */
 151         gdqp = dqp->q_gdquot;
 152         if (gdqp) {
 153                 xfs_dqlock(gdqp);
 154                 dqp->q_gdquot = NULL;
 155         }
 156
 157         dqp->dq_flags |= XFS_DQ_FREEING;
 158
 159         xfs_dqflock(dqp);
 160
 161         /*
 162          * If we are turning this type of quotas off, we don't care
 163          * about the dirty metadata sitting in this dquot. OTOH, if
 164          * we're unmounting, we do care, so we flush it and wait.
 165          */
 166         if (XFS_DQ_IS_DIRTY(dqp)) {
 167                 struct xfs_buf  *bp = NULL;
 168                 int             error;
 169
 170                 /*
 171                  * We don't care about getting disk errors here. We need
 172                  * to purge this dquot anyway, so we go ahead regardless.
 173                  */
 174                 error = xfs_qm_dqflush(dqp, &bp);
 175                 if (error) {
 176                         xfs_warn(mp, "%s: dquot %p flush failed",
 177                                 __func__, dqp);
 178                 } else {
 179                         error = xfs_bwrite(bp);
 180                         xfs_buf_relse(bp);
 181                 }
 182                 xfs_dqflock(dqp);
 183         }
 184
 185         ASSERT(atomic_read(&dqp->q_pincount) == 0);
 186         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 187                !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 188
 189         xfs_dqfunlock(dqp);
 190         xfs_dqunlock(dqp);
 191
 192         radix_tree_delete(XFS_DQUOT_TREE(qi, dqp->q_core.d_flags),
 193                           be32_to_cpu(dqp->q_core.d_id));
 194         qi->qi_dquots--;
 195
 196         /*
 197          * We move dquots to the freelist as soon as their reference count
 198          * hits zero, so it really should be on the freelist here.
 199          */
 200         mutex_lock(&qi->qi_lru_lock);
 201         ASSERT(!list_empty(&dqp->q_lru));
 202         list_del_init(&dqp->q_lru);
 203         qi->qi_lru_count--;
 204         XFS_STATS_DEC(xs_qm_dquot_unused);
 205         mutex_unlock(&qi->qi_lru_lock);
 206
 207         xfs_qm_dqdestroy(dqp);
 208
 209         if (gdqp)
 210                 xfs_qm_dqput(gdqp);
 211         return 0;
 212 }
 213
 214 /*
 215  * Purge the dquot cache.
 216  */
 217 void
 218 xfs_qm_dqpurge_all(
 219         struct xfs_mount        *mp,
 220         uint                    flags)
 221 {
 222         if (flags & XFS_QMOPT_UQUOTA)
 223                 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
 224         if (flags & XFS_QMOPT_GQUOTA)
 225                 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 226         if (flags & XFS_QMOPT_PQUOTA)
 227                 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 228 }
 229
 230 /*
 231  * Just destroy the quotainfo structure.
 232  */
 233 void
 234 xfs_qm_unmount(
 235         struct xfs_mount        *mp)
 236 {
 237         if (mp->m_quotainfo) {
 238                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 239                 xfs_qm_destroy_quotainfo(mp);
 240         }
 241 }
 242
 243
 244 /*
 245  * This is called from xfs_mountfs to start quotas and initialize all
 246  * necessary data structures like quotainfo.  This is also responsible for
 247  * running a quotacheck as necessary.  We are guaranteed that the superblock
 248  * is consistently read in at this point.
 249  *
 250  * If we fail here, the mount will continue with quota turned off. We don't
 251  * need to inidicate success or failure at all.
 252  */
 253 void
 254 xfs_qm_mount_quotas(
 255         xfs_mount_t     *mp)
 256 {
 257         int             error = 0;
 258         uint            sbf;
 259
 260         /*
 261          * If quotas on realtime volumes is not supported, we disable
 262          * quotas immediately.
 263          */
 264         if (mp->m_sb.sb_rextents) {
 265                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 266                 mp->m_qflags = 0;
 267                 goto write_changes;
 268         }
 269
 270         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 271
 272         /*
 273          * Allocate the quotainfo structure inside the mount struct, and
 274          * create quotainode(s), and change/rev superblock if necessary.
 275          */
 276         error = xfs_qm_init_quotainfo(mp);
 277         if (error) {
 278                 /*
 279                  * We must turn off quotas.
 280                  */
 281                 ASSERT(mp->m_quotainfo == NULL);
 282                 mp->m_qflags = 0;
 283                 goto write_changes;
 284         }
 285         /*
 286          * If any of the quotas are not consistent, do a quotacheck.
 287          */
 288         if (XFS_QM_NEED_QUOTACHECK(mp)) {
 289                 error = xfs_qm_quotacheck(mp);
 290                 if (error) {
 291                         /* Quotacheck failed and disabled quotas. */
 292                         return;
 293                 }
 294         }
 295         /*
 296          * If one type of quotas is off, then it will lose its
 297          * quotachecked status, since we won't be doing accounting for
 298          * that type anymore.
 299          */
 300         if (!XFS_IS_UQUOTA_ON(mp))
 301                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 302         if (!(XFS_IS_GQUOTA_ON(mp) || XFS_IS_PQUOTA_ON(mp)))
 303                 mp->m_qflags &= ~XFS_OQUOTA_CHKD;
 304
 305  write_changes:
 306         /*
 307          * We actually don't have to acquire the m_sb_lock at all.
 308          * This can only be called from mount, and that's single threaded. XXX
 309          */
 310         spin_lock(&mp->m_sb_lock);
 311         sbf = mp->m_sb.sb_qflags;
 312         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 313         spin_unlock(&mp->m_sb_lock);
 314
 315         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 316                 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 317                         /*
 318                          * We could only have been turning quotas off.
 319                          * We aren't in very good shape actually because
 320                          * the incore structures are convinced that quotas are
 321                          * off, but the on disk superblock doesn't know that !
 322                          */
 323                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 324                         xfs_alert(mp, "%s: Superblock update failed!",
 325                                 __func__);
 326                 }
 327         }
 328
 329         if (error) {
 330                 xfs_warn(mp, "Failed to initialize disk quotas.");
 331                 return;
 332         }
 333 }
 334
 335 /*
 336  * Called from the vfsops layer.
 337  */
 338 void
 339 xfs_qm_unmount_quotas(
 340         xfs_mount_t     *mp)
 341 {
 342         /*
 343          * Release the dquots that root inode, et al might be holding,
 344          * before we flush quotas and blow away the quotainfo structure.
 345          */
 346         ASSERT(mp->m_rootip);
 347         xfs_qm_dqdetach(mp->m_rootip);
 348         if (mp->m_rbmip)
 349                 xfs_qm_dqdetach(mp->m_rbmip);
 350         if (mp->m_rsumip)
 351                 xfs_qm_dqdetach(mp->m_rsumip);
 352
 353         /*
 354          * Release the quota inodes.
 355          */
 356         if (mp->m_quotainfo) {
 357                 if (mp->m_quotainfo->qi_uquotaip) {
 358                         IRELE(mp->m_quotainfo->qi_uquotaip);
 359                         mp->m_quotainfo->qi_uquotaip = NULL;
 360                 }
 361                 if (mp->m_quotainfo->qi_gquotaip) {
 362                         IRELE(mp->m_quotainfo->qi_gquotaip);
 363                         mp->m_quotainfo->qi_gquotaip = NULL;
 364                 }
 365         }
 366 }
 367
 368 STATIC int
 369 xfs_qm_dqattach_one(
 370         xfs_inode_t     *ip,
 371         xfs_dqid_t      id,
 372         uint            type,
 373         uint            doalloc,
 374         xfs_dquot_t     *udqhint, /* hint */
 375         xfs_dquot_t     **IO_idqpp)
 376 {
 377         xfs_dquot_t     *dqp;
 378         int             error;
 379
 380         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 381         error = 0;
 382
 383         /*
 384          * See if we already have it in the inode itself. IO_idqpp is
 385          * &i_udquot or &i_gdquot. This made the code look weird, but
 386          * made the logic a lot simpler.
 387          */
 388         dqp = *IO_idqpp;
 389         if (dqp) {
 390                 trace_xfs_dqattach_found(dqp);
 391                 return 0;
 392         }
 393
 394         /*
 395          * udqhint is the i_udquot field in inode, and is non-NULL only
 396          * when the type arg is group/project. Its purpose is to save a
 397          * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 398          * the user dquot.
 399          */
 400         if (udqhint) {
 401                 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 402                 xfs_dqlock(udqhint);
 403
 404                 /*
 405                  * No need to take dqlock to look at the id.
 406                  *
 407                  * The ID can't change until it gets reclaimed, and it won't
 408                  * be reclaimed as long as we have a ref from inode and we
 409                  * hold the ilock.
 410                  */
 411                 dqp = udqhint->q_gdquot;
 412                 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 413                         ASSERT(*IO_idqpp == NULL);
 414
 415                         *IO_idqpp = xfs_qm_dqhold(dqp);
 416                         xfs_dqunlock(udqhint);
 417                         return 0;
 418                 }
 419
 420                 /*
 421                  * We can't hold a dquot lock when we call the dqget code.
 422                  * We'll deadlock in no time, because of (not conforming to)
 423                  * lock ordering - the inodelock comes before any dquot lock,
 424                  * and we may drop and reacquire the ilock in xfs_qm_dqget().
 425                  */
 426                 xfs_dqunlock(udqhint);
 427         }
 428
 429         /*
 430          * Find the dquot from somewhere. This bumps the
 431          * reference count of dquot and returns it locked.
 432          * This can return ENOENT if dquot didn't exist on
 433          * disk and we didn't ask it to allocate;
 434          * ESRCH if quotas got turned off suddenly.
 435          */
 436         error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 437                              doalloc | XFS_QMOPT_DOWARN, &dqp);
 438         if (error)
 439                 return error;
 440
 441         trace_xfs_dqattach_get(dqp);
 442
 443         /*
 444          * dqget may have dropped and re-acquired the ilock, but it guarantees
 445          * that the dquot returned is the one that should go in the inode.
 446          */
 447         *IO_idqpp = dqp;
 448         xfs_dqunlock(dqp);
 449         return 0;
 450 }
 451
 452
 453 /*
 454  * Given a udquot and gdquot, attach a ptr to the group dquot in the
 455  * udquot as a hint for future lookups.
 456  */
 457 STATIC void
 458 xfs_qm_dqattach_grouphint(
 459         xfs_dquot_t     *udq,
 460         xfs_dquot_t     *gdq)
 461 {
 462         xfs_dquot_t     *tmp;
 463
 464         xfs_dqlock(udq);
 465
 466         tmp = udq->q_gdquot;
 467         if (tmp) {
 468                 if (tmp == gdq)
 469                         goto done;
 470
 471                 udq->q_gdquot = NULL;
 472                 xfs_qm_dqrele(tmp);
 473         }
 474
 475         udq->q_gdquot = xfs_qm_dqhold(gdq);
 476 done:
 477         xfs_dqunlock(udq);
 478 }
 479
 480 static bool
 481 xfs_qm_need_dqattach(
 482         struct xfs_inode        *ip)
 483 {
 484         struct xfs_mount        *mp = ip->i_mount;
 485
 486         if (!XFS_IS_QUOTA_RUNNING(mp))
 487                 return false;
 488         if (!XFS_IS_QUOTA_ON(mp))
 489                 return false;
 490         if (!XFS_NOT_DQATTACHED(mp, ip))
 491                 return false;
 492         if (ip->i_ino == mp->m_sb.sb_uquotino ||
 493             ip->i_ino == mp->m_sb.sb_gquotino)
 494                 return false;
 495         return true;
 496 }
 497
 498 /*
 499  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 500  * into account.
 501  * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 502  * Inode may get unlocked and relocked in here, and the caller must deal with
 503  * the consequences.
 504  */
 505 int
 506 xfs_qm_dqattach_locked(
 507         xfs_inode_t     *ip,
 508         uint            flags)
 509 {
 510         xfs_mount_t     *mp = ip->i_mount;
 511         uint            nquotas = 0;
 512         int             error = 0;
 513
 514         if (!xfs_qm_need_dqattach(ip))
 515                 return 0;
 516
 517         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 518
 519         if (XFS_IS_UQUOTA_ON(mp)) {
 520                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 521                                                 flags & XFS_QMOPT_DQALLOC,
 522                                                 NULL, &ip->i_udquot);
 523                 if (error)
 524                         goto done;
 525                 nquotas++;
 526         }
 527
 528         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 529         if (XFS_IS_OQUOTA_ON(mp)) {
 530                 error = XFS_IS_GQUOTA_ON(mp) ?
 531                         xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 532                                                 flags & XFS_QMOPT_DQALLOC,
 533                                                 ip->i_udquot, &ip->i_gdquot) :
 534                         xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 535                                                 flags & XFS_QMOPT_DQALLOC,
 536                                                 ip->i_udquot, &ip->i_gdquot);
 537                 /*
 538                  * Don't worry about the udquot that we may have
 539                  * attached above. It'll get detached, if not already.
 540                  */
 541                 if (error)
 542                         goto done;
 543                 nquotas++;
 544         }
 545
 546         /*
 547          * Attach this group quota to the user quota as a hint.
 548          * This WON'T, in general, result in a thrash.
 549          */
 550         if (nquotas == 2) {
 551                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 552                 ASSERT(ip->i_udquot);
 553                 ASSERT(ip->i_gdquot);
 554
 555                 /*
 556                  * We do not have i_udquot locked at this point, but this check
 557                  * is OK since we don't depend on the i_gdquot to be accurate
 558                  * 100% all the time. It is just a hint, and this will
 559                  * succeed in general.
 560                  */
 561                 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 562                         xfs_qm_dqattach_grouphint(ip->i_udquot, ip->i_gdquot);
 563         }
 564
 565  done:
 566 #ifdef DEBUG
 567         if (!error) {
 568                 if (XFS_IS_UQUOTA_ON(mp))
 569                         ASSERT(ip->i_udquot);
 570                 if (XFS_IS_OQUOTA_ON(mp))
 571                         ASSERT(ip->i_gdquot);
 572         }
 573         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 574 #endif
 575         return error;
 576 }
 577
 578 int
 579 xfs_qm_dqattach(
 580         struct xfs_inode        *ip,
 581         uint                    flags)
 582 {
 583         int                     error;
 584
 585         if (!xfs_qm_need_dqattach(ip))
 586                 return 0;
 587
 588         xfs_ilock(ip, XFS_ILOCK_EXCL);
 589         error = xfs_qm_dqattach_locked(ip, flags);
 590         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 591
 592         return error;
 593 }
 594
 595 /*
 596  * Release dquots (and their references) if any.
 597  * The inode should be locked EXCL except when this's called by
 598  * xfs_ireclaim.
 599  */
 600 void
 601 xfs_qm_dqdetach(
 602         xfs_inode_t     *ip)
 603 {
 604         if (!(ip->i_udquot || ip->i_gdquot))
 605                 return;
 606
 607         trace_xfs_dquot_dqdetach(ip);
 608
 609         ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_uquotino);
 610         ASSERT(ip->i_ino != ip->i_mount->m_sb.sb_gquotino);
 611         if (ip->i_udquot) {
 612                 xfs_qm_dqrele(ip->i_udquot);
 613                 ip->i_udquot = NULL;
 614         }
 615         if (ip->i_gdquot) {
 616                 xfs_qm_dqrele(ip->i_gdquot);
 617                 ip->i_gdquot = NULL;
 618         }
 619 }
 620
 621 int
 622 xfs_qm_calc_dquots_per_chunk(
 623         struct xfs_mount        *mp,
 624         unsigned int            nbblks) /* basic block units */
 625 {
 626         unsigned int    ndquots;
 627
 628         ASSERT(nbblks > 0);
 629         ndquots = BBTOB(nbblks);
 630         do_div(ndquots, sizeof(xfs_dqblk_t));
 631
 632         return ndquots;
 633 }
 634
 635 /*
 636  * This initializes all the quota information that's kept in the
 637  * mount structure
 638  */
 639 STATIC int
 640 xfs_qm_init_quotainfo(
 641         xfs_mount_t     *mp)
 642 {
 643         xfs_quotainfo_t *qinf;
 644         int             error;
 645         xfs_dquot_t     *dqp;
 646
 647         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 648
 649         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 650
 651         /*
 652          * See if quotainodes are setup, and if not, allocate them,
 653          * and change the superblock accordingly.
 654          */
 655         if ((error = xfs_qm_init_quotainos(mp))) {
 656                 kmem_free(qinf);
 657                 mp->m_quotainfo = NULL;
 658                 return error;
 659         }
 660
 661         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 662         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 663         mutex_init(&qinf->qi_tree_lock);
 664
 665         INIT_LIST_HEAD(&qinf->qi_lru_list);
 666         qinf->qi_lru_count = 0;
 667         mutex_init(&qinf->qi_lru_lock);
 668
 669         /* mutex used to serialize quotaoffs */
 670         mutex_init(&qinf->qi_quotaofflock);
 671
 672         /* Precalc some constants */
 673         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 674         qinf->qi_dqperchunk = xfs_qm_calc_dquots_per_chunk(mp,
 675                                                         qinf->qi_dqchunklen);
 676
 677         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 678
 679         /*
 680          * We try to get the limits from the superuser's limits fields.
 681          * This is quite hacky, but it is standard quota practice.
 682          *
 683          * We look at the USR dquot with id == 0 first, but if user quotas
 684          * are not enabled we goto the GRP dquot with id == 0.
 685          * We don't really care to keep separate default limits for user
 686          * and group quotas, at least not at this point.
 687          *
 688          * Since we may not have done a quotacheck by this point, just read
 689          * the dquot without attaching it to any hashtables or lists.
 690          */
 691         error = xfs_qm_dqread(mp, 0,
 692                         XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 693                          (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 694                           XFS_DQ_PROJ),
 695                         XFS_QMOPT_DOWARN, &dqp);
 696         if (!error) {
 697                 xfs_disk_dquot_t        *ddqp = &dqp->q_core;
 698
 699                 /*
 700                  * The warnings and timers set the grace period given to
 701                  * a user or group before he or she can not perform any
 702                  * more writing. If it is zero, a default is used.
 703                  */
 704                 qinf->qi_btimelimit = ddqp->d_btimer ?
 705                         be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 706                 qinf->qi_itimelimit = ddqp->d_itimer ?
 707                         be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 708                 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 709                         be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 710                 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 711                         be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 712                 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 713                         be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 714                 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 715                         be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 716                 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 717                 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 718                 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 719                 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 720                 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 721                 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 722
 723                 xfs_qm_dqdestroy(dqp);
 724         } else {
 725                 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 726                 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 727                 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 728                 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 729                 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 730                 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 731         }
 732
 733         qinf->qi_shrinker.shrink = xfs_qm_shake;
 734         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 735         register_shrinker(&qinf->qi_shrinker);
 736         return 0;
 737 }
 738
 739
 740 /*
 741  * Gets called when unmounting a filesystem or when all quotas get
 742  * turned off.
 743  * This purges the quota inodes, destroys locks and frees itself.
 744  */
 745 void
 746 xfs_qm_destroy_quotainfo(
 747         xfs_mount_t     *mp)
 748 {
 749         xfs_quotainfo_t *qi;
 750
 751         qi = mp->m_quotainfo;
 752         ASSERT(qi != NULL);
 753
 754         unregister_shrinker(&qi->qi_shrinker);
 755
 756         if (qi->qi_uquotaip) {
 757                 IRELE(qi->qi_uquotaip);
 758                 qi->qi_uquotaip = NULL; /* paranoia */
 759         }
 760         if (qi->qi_gquotaip) {
 761                 IRELE(qi->qi_gquotaip);
 762                 qi->qi_gquotaip = NULL;
 763         }
 764         mutex_destroy(&qi->qi_quotaofflock);
 765         kmem_free(qi);
 766         mp->m_quotainfo = NULL;
 767 }
 768
 769 /*
 770  * Create an inode and return with a reference already taken, but unlocked
 771  * This is how we create quota inodes
 772  */
 773 STATIC int
 774 xfs_qm_qino_alloc(
 775         xfs_mount_t     *mp,
 776         xfs_inode_t     **ip,
 777         __int64_t       sbfields,
 778         uint            flags)
 779 {
 780         xfs_trans_t     *tp;
 781         int             error;
 782         int             committed;
 783
 784         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
 785         if ((error = xfs_trans_reserve(tp,
 786                                       XFS_QM_QINOCREATE_SPACE_RES(mp),
 787                                       XFS_CREATE_LOG_RES(mp), 0,
 788                                       XFS_TRANS_PERM_LOG_RES,
 789                                       XFS_CREATE_LOG_COUNT))) {
 790                 xfs_trans_cancel(tp, 0);
 791                 return error;
 792         }
 793
 794         error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip, &committed);
 795         if (error) {
 796                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
 797                                  XFS_TRANS_ABORT);
 798                 return error;
 799         }
 800
 801         /*
 802          * Make the changes in the superblock, and log those too.
 803          * sbfields arg may contain fields other than *QUOTINO;
 804          * VERSIONNUM for example.
 805          */
 806         spin_lock(&mp->m_sb_lock);
 807         if (flags & XFS_QMOPT_SBVERSION) {
 808                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
 809                 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
 810                                    XFS_SB_GQUOTINO | XFS_SB_QFLAGS)) ==
 811                        (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
 812                         XFS_SB_GQUOTINO | XFS_SB_QFLAGS));
 813
 814                 xfs_sb_version_addquota(&mp->m_sb);
 815                 mp->m_sb.sb_uquotino = NULLFSINO;
 816                 mp->m_sb.sb_gquotino = NULLFSINO;
 817
 818                 /* qflags will get updated _after_ quotacheck */
 819                 mp->m_sb.sb_qflags = 0;
 820         }
 821         if (flags & XFS_QMOPT_UQUOTA)
 822                 mp->m_sb.sb_uquotino = (*ip)->i_ino;
 823         else
 824                 mp->m_sb.sb_gquotino = (*ip)->i_ino;
 825         spin_unlock(&mp->m_sb_lock);
 826         xfs_mod_sb(tp, sbfields);
 827
 828         if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
 829                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
 830                 return error;
 831         }
 832         return 0;
 833 }
 834
 835
 836 STATIC void
 837 xfs_qm_reset_dqcounts(
 838         xfs_mount_t     *mp,
 839         xfs_buf_t       *bp,
 840         xfs_dqid_t      id,
 841         uint            type)
 842 {
 843         struct xfs_dqblk        *dqb;
 844         int                     j;
 845
 846         trace_xfs_reset_dqcounts(bp, _RET_IP_);
 847
 848         /*
 849          * Reset all counters and timers. They'll be
 850          * started afresh by xfs_qm_quotacheck.
 851          */
 852 #ifdef DEBUG
 853         j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 854         do_div(j, sizeof(xfs_dqblk_t));
 855         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
 856 #endif
 857         dqb = bp->b_addr;
 858         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
 859                 struct xfs_disk_dquot   *ddq;
 860
 861                 ddq = (struct xfs_disk_dquot *)&dqb[j];
 862
 863                 /*
 864                  * Do a sanity check, and if needed, repair the dqblk. Don't
 865                  * output any warnings because it's perfectly possible to
 866                  * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
 867                  */
 868                 (void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
 869                                       "xfs_quotacheck");
 870                 ddq->d_bcount = 0;
 871                 ddq->d_icount = 0;
 872                 ddq->d_rtbcount = 0;
 873                 ddq->d_btimer = 0;
 874                 ddq->d_itimer = 0;
 875                 ddq->d_rtbtimer = 0;
 876                 ddq->d_bwarns = 0;
 877                 ddq->d_iwarns = 0;
 878                 ddq->d_rtbwarns = 0;
 879
 880                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 881                         xfs_update_cksum((char *)&dqb[j],
 882                                          sizeof(struct xfs_dqblk),
 883                                          XFS_DQUOT_CRC_OFF);
 884                 }
 885         }
 886 }
 887
 888 STATIC int
 889 xfs_qm_dqiter_bufs(
 890         struct xfs_mount        *mp,
 891         xfs_dqid_t              firstid,
 892         xfs_fsblock_t           bno,
 893         xfs_filblks_t           blkcnt,
 894         uint                    flags,
 895         struct list_head        *buffer_list)
 896 {
 897         struct xfs_buf          *bp;
 898         int                     error;
 899         int                     type;
 900
 901         ASSERT(blkcnt > 0);
 902         type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
 903                 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
 904         error = 0;
 905
 906         /*
 907          * Blkcnt arg can be a very big number, and might even be
 908          * larger than the log itself. So, we have to break it up into
 909          * manageable-sized transactions.
 910          * Note that we don't start a permanent transaction here; we might
 911          * not be able to get a log reservation for the whole thing up front,
 912          * and we don't really care to either, because we just discard
 913          * everything if we were to crash in the middle of this loop.
 914          */
 915         while (blkcnt--) {
 916                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 917                               XFS_FSB_TO_DADDR(mp, bno),
 918                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 919                               &xfs_dquot_buf_ops);
 920
 921                 /*
 922                  * CRC and validation errors will return a EFSCORRUPTED here. If
 923                  * this occurs, re-read without CRC validation so that we can
 924                  * repair the damage via xfs_qm_reset_dqcounts(). This process
 925                  * will leave a trace in the log indicating corruption has
 926                  * been detected.
 927                  */
 928                 if (error == EFSCORRUPTED) {
 929                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 930                                       XFS_FSB_TO_DADDR(mp, bno),
 931                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
 932                                       NULL);
 933                 }
 934
 935                 if (error)
 936                         break;
 937
 938                 /*
 939                  * A corrupt buffer might not have a verifier attached, so
 940                  * make sure we have the correct one attached before writeback
 941                  * occurs.
 942                  */
 943                 bp->b_ops = &xfs_dquot_buf_ops;
 944                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
 945                 xfs_buf_delwri_queue(bp, buffer_list);
 946                 xfs_buf_relse(bp);
 947
 948                 /* goto the next block. */
 949                 bno++;
 950                 firstid += mp->m_quotainfo->qi_dqperchunk;
 951         }
 952
 953         return error;
 954 }
 955
 956 /*
 957  * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
 958  * caller supplied function for every chunk of dquots that we find.
 959  */
 960 STATIC int
 961 xfs_qm_dqiterate(
 962         struct xfs_mount        *mp,
 963         struct xfs_inode        *qip,
 964         uint                    flags,
 965         struct list_head        *buffer_list)
 966 {
 967         struct xfs_bmbt_irec    *map;
 968         int                     i, nmaps;       /* number of map entries */
 969         int                     error;          /* return value */
 970         xfs_fileoff_t           lblkno;
 971         xfs_filblks_t           maxlblkcnt;
 972         xfs_dqid_t              firstid;
 973         xfs_fsblock_t           rablkno;
 974         xfs_filblks_t           rablkcnt;
 975
 976         error = 0;
 977         /*
 978          * This looks racy, but we can't keep an inode lock across a
 979          * trans_reserve. But, this gets called during quotacheck, and that
 980          * happens only at mount time which is single threaded.
 981          */
 982         if (qip->i_d.di_nblocks == 0)
 983                 return 0;
 984
 985         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
 986
 987         lblkno = 0;
 988         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
 989         do {
 990                 nmaps = XFS_DQITER_MAP_SIZE;
 991                 /*
 992                  * We aren't changing the inode itself. Just changing
 993                  * some of its data. No new blocks are added here, and
 994                  * the inode is never added to the transaction.
 995                  */
 996                 xfs_ilock(qip, XFS_ILOCK_SHARED);
 997                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
 998                                        map, &nmaps, 0);
 999                 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1000                 if (error)
1001                         break;
1002
1003                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1004                 for (i = 0; i < nmaps; i++) {
1005                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1006                         ASSERT(map[i].br_blockcount);
1007
1008
1009                         lblkno += map[i].br_blockcount;
1010
1011                         if (map[i].br_startblock == HOLESTARTBLOCK)
1012                                 continue;
1013
1014                         firstid = (xfs_dqid_t) map[i].br_startoff *
1015                                 mp->m_quotainfo->qi_dqperchunk;
1016                         /*
1017                          * Do a read-ahead on the next extent.
1018                          */
1019                         if ((i+1 < nmaps) &&
1020                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1021                                 rablkcnt =  map[i+1].br_blockcount;
1022                                 rablkno = map[i+1].br_startblock;
1023                                 while (rablkcnt--) {
1024                                         xfs_buf_readahead(mp->m_ddev_targp,
1025                                                XFS_FSB_TO_DADDR(mp, rablkno),
1026                                                mp->m_quotainfo->qi_dqchunklen,
1027                                                &xfs_dquot_buf_ops);
1028                                         rablkno++;
1029                                 }
1030                         }
1031                         /*
1032                          * Iterate thru all the blks in the extent and
1033                          * reset the counters of all the dquots inside them.
1034                          */
1035                         error = xfs_qm_dqiter_bufs(mp, firstid,
1036                                                    map[i].br_startblock,
1037                                                    map[i].br_blockcount,
1038                                                    flags, buffer_list);
1039                         if (error)
1040                                 goto out;
1041                 }
1042         } while (nmaps > 0);
1043
1044 out:
1045         kmem_free(map);
1046         return error;
1047 }
1048
1049 /*
1050  * Called by dqusage_adjust in doing a quotacheck.
1051  *
1052  * Given the inode, and a dquot id this updates both the incore dqout as well
1053  * as the buffer copy. This is so that once the quotacheck is done, we can
1054  * just log all the buffers, as opposed to logging numerous updates to
1055  * individual dquots.
1056  */
1057 STATIC int
1058 xfs_qm_quotacheck_dqadjust(
1059         struct xfs_inode        *ip,
1060         xfs_dqid_t              id,
1061         uint                    type,
1062         xfs_qcnt_t              nblks,
1063         xfs_qcnt_t              rtblks)
1064 {
1065         struct xfs_mount        *mp = ip->i_mount;
1066         struct xfs_dquot        *dqp;
1067         int                     error;
1068
1069         error = xfs_qm_dqget(mp, ip, id, type,
1070                              XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1071         if (error) {
1072                 /*
1073                  * Shouldn't be able to turn off quotas here.
1074                  */
1075                 ASSERT(error != ESRCH);
1076                 ASSERT(error != ENOENT);
1077                 return error;
1078         }
1079
1080         trace_xfs_dqadjust(dqp);
1081
1082         /*
1083          * Adjust the inode count and the block count to reflect this inode's
1084          * resource usage.
1085          */
1086         be64_add_cpu(&dqp->q_core.d_icount, 1);
1087         dqp->q_res_icount++;
1088         if (nblks) {
1089                 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1090                 dqp->q_res_bcount += nblks;
1091         }
1092         if (rtblks) {
1093                 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1094                 dqp->q_res_rtbcount += rtblks;
1095         }
1096
1097         /*
1098          * Set default limits, adjust timers (since we changed usages)
1099          *
1100          * There are no timers for the default values set in the root dquot.
1101          */
1102         if (dqp->q_core.d_id) {
1103                 xfs_qm_adjust_dqlimits(mp, dqp);
1104                 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1105         }
1106
1107         dqp->dq_flags |= XFS_DQ_DIRTY;
1108         xfs_qm_dqput(dqp);
1109         return 0;
1110 }
1111
1112 STATIC int
1113 xfs_qm_get_rtblks(
1114         xfs_inode_t     *ip,
1115         xfs_qcnt_t      *O_rtblks)
1116 {
1117         xfs_filblks_t   rtblks;                 /* total rt blks */
1118         xfs_extnum_t    idx;                    /* extent record index */
1119         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1120         xfs_extnum_t    nextents;               /* number of extent entries */
1121         int             error;
1122
1123         ASSERT(XFS_IS_REALTIME_INODE(ip));
1124         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1125         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1126                 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1127                         return error;
1128         }
1129         rtblks = 0;
1130         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1131         for (idx = 0; idx < nextents; idx++)
1132                 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1133         *O_rtblks = (xfs_qcnt_t)rtblks;
1134         return 0;
1135 }
1136
1137 /*
1138  * callback routine supplied to bulkstat(). Given an inumber, find its
1139  * dquots and update them to account for resources taken by that inode.
1140  */
1141 /* ARGSUSED */
1142 STATIC int
1143 xfs_qm_dqusage_adjust(
1144         xfs_mount_t     *mp,            /* mount point for filesystem */
1145         xfs_ino_t       ino,            /* inode number to get data for */
1146         void            __user *buffer, /* not used */
1147         int             ubsize,         /* not used */
1148         int             *ubused,        /* not used */
1149         int             *res)           /* result code value */
1150 {
1151         xfs_inode_t     *ip;
1152         xfs_qcnt_t      nblks, rtblks = 0;
1153         int             error;
1154
1155         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1156
1157         /*
1158          * rootino must have its resources accounted for, not so with the quota
1159          * inodes.
1160          */
1161         if (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino) {
1162                 *res = BULKSTAT_RV_NOTHING;
1163                 return XFS_ERROR(EINVAL);
1164         }
1165
1166         /*
1167          * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1168          * interface expects the inode to be exclusively locked because that's
1169          * the case in all other instances. It's OK that we do this because
1170          * quotacheck is done only at mount time.
1171          */
1172         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1173         if (error) {
1174                 *res = BULKSTAT_RV_NOTHING;
1175                 return error;
1176         }
1177
1178         ASSERT(ip->i_delayed_blks == 0);
1179
1180         if (XFS_IS_REALTIME_INODE(ip)) {
1181                 /*
1182                  * Walk thru the extent list and count the realtime blocks.
1183                  */
1184                 error = xfs_qm_get_rtblks(ip, &rtblks);
1185                 if (error)
1186                         goto error0;
1187         }
1188
1189         nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1190
1191         /*
1192          * Add the (disk blocks and inode) resources occupied by this
1193          * inode to its dquots. We do this adjustment in the incore dquot,
1194          * and also copy the changes to its buffer.
1195          * We don't care about putting these changes in a transaction
1196          * envelope because if we crash in the middle of a 'quotacheck'
1197          * we have to start from the beginning anyway.
1198          * Once we're done, we'll log all the dquot bufs.
1199          *
1200          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1201          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1202          */
1203         if (XFS_IS_UQUOTA_ON(mp)) {
1204                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1205                                                    XFS_DQ_USER, nblks, rtblks);
1206                 if (error)
1207                         goto error0;
1208         }
1209
1210         if (XFS_IS_GQUOTA_ON(mp)) {
1211                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1212                                                    XFS_DQ_GROUP, nblks, rtblks);
1213                 if (error)
1214                         goto error0;
1215         }
1216
1217         if (XFS_IS_PQUOTA_ON(mp)) {
1218                 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1219                                                    XFS_DQ_PROJ, nblks, rtblks);
1220                 if (error)
1221                         goto error0;
1222         }
1223
1224         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1225         IRELE(ip);
1226         *res = BULKSTAT_RV_DIDONE;
1227         return 0;
1228
1229 error0:
1230         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1231         IRELE(ip);
1232         *res = BULKSTAT_RV_GIVEUP;
1233         return error;
1234 }
1235
1236 STATIC int
1237 xfs_qm_flush_one(
1238         struct xfs_dquot        *dqp,
1239         void                    *data)
1240 {
1241         struct list_head        *buffer_list = data;
1242         struct xfs_buf          *bp = NULL;
1243         int                     error = 0;
1244
1245         xfs_dqlock(dqp);
1246         if (dqp->dq_flags & XFS_DQ_FREEING)
1247                 goto out_unlock;
1248         if (!XFS_DQ_IS_DIRTY(dqp))
1249                 goto out_unlock;
1250
1251         xfs_dqflock(dqp);
1252         error = xfs_qm_dqflush(dqp, &bp);
1253         if (error)
1254                 goto out_unlock;
1255
1256         xfs_buf_delwri_queue(bp, buffer_list);
1257         xfs_buf_relse(bp);
1258 out_unlock:
1259         xfs_dqunlock(dqp);
1260         return error;
1261 }
1262
1263 /*
1264  * Walk thru all the filesystem inodes and construct a consistent view
1265  * of the disk quota world. If the quotacheck fails, disable quotas.
1266  */
1267 int
1268 xfs_qm_quotacheck(
1269         xfs_mount_t     *mp)
1270 {
1271         int             done, count, error, error2;
1272         xfs_ino_t       lastino;
1273         size_t          structsz;
1274         xfs_inode_t     *uip, *gip;
1275         uint            flags;
1276         LIST_HEAD       (buffer_list);
1277
1278         count = INT_MAX;
1279         structsz = 1;
1280         lastino = 0;
1281         flags = 0;
1282
1283         ASSERT(mp->m_quotainfo->qi_uquotaip || mp->m_quotainfo->qi_gquotaip);
1284         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1285
1286         xfs_notice(mp, "Quotacheck needed: Please wait.");
1287
1288         /*
1289          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1290          * their counters to zero. We need a clean slate.
1291          * We don't log our changes till later.
1292          */
1293         uip = mp->m_quotainfo->qi_uquotaip;
1294         if (uip) {
1295                 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1296                                          &buffer_list);
1297                 if (error)
1298                         goto error_return;
1299                 flags |= XFS_UQUOTA_CHKD;
1300         }
1301
1302         gip = mp->m_quotainfo->qi_gquotaip;
1303         if (gip) {
1304                 error = xfs_qm_dqiterate(mp, gip, XFS_IS_GQUOTA_ON(mp) ?
1305                                          XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA,
1306                                          &buffer_list);
1307                 if (error)
1308                         goto error_return;
1309                 flags |= XFS_OQUOTA_CHKD;
1310         }
1311
1312         do {
1313                 /*
1314                  * Iterate thru all the inodes in the file system,
1315                  * adjusting the corresponding dquot counters in core.
1316                  */
1317                 error = xfs_bulkstat(mp, &lastino, &count,
1318                                      xfs_qm_dqusage_adjust,
1319                                      structsz, NULL, &done);
1320                 if (error)
1321                         break;
1322
1323         } while (!done);
1324
1325         /*
1326          * We've made all the changes that we need to make incore.  Flush them
1327          * down to disk buffers if everything was updated successfully.
1328          */
1329         if (XFS_IS_UQUOTA_ON(mp)) {
1330                 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1331                                           &buffer_list);
1332         }
1333         if (XFS_IS_GQUOTA_ON(mp)) {
1334                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1335                                            &buffer_list);
1336                 if (!error)
1337                         error = error2;
1338         }
1339         if (XFS_IS_PQUOTA_ON(mp)) {
1340                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1341                                            &buffer_list);
1342                 if (!error)
1343                         error = error2;
1344         }
1345
1346         error2 = xfs_buf_delwri_submit(&buffer_list);
1347         if (!error)
1348                 error = error2;
1349
1350         /*
1351          * We can get this error if we couldn't do a dquot allocation inside
1352          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1353          * dirty dquots that might be cached, we just want to get rid of them
1354          * and turn quotaoff. The dquots won't be attached to any of the inodes
1355          * at this point (because we intentionally didn't in dqget_noattach).
1356          */
1357         if (error) {
1358                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1359                 goto error_return;
1360         }
1361
1362         /*
1363          * If one type of quotas is off, then it will lose its
1364          * quotachecked status, since we won't be doing accounting for
1365          * that type anymore.
1366          */
1367         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1368         mp->m_qflags |= flags;
1369
1370  error_return:
1371         while (!list_empty(&buffer_list)) {
1372                 struct xfs_buf *bp =
1373                         list_first_entry(&buffer_list, struct xfs_buf, b_list);
1374                 list_del_init(&bp->b_list);
1375                 xfs_buf_relse(bp);
1376         }
1377
1378         if (error) {
1379                 xfs_warn(mp,
1380         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1381                         error);
1382                 /*
1383                  * We must turn off quotas.
1384                  */
1385                 ASSERT(mp->m_quotainfo != NULL);
1386                 xfs_qm_destroy_quotainfo(mp);
1387                 if (xfs_mount_reset_sbqflags(mp)) {
1388                         xfs_warn(mp,
1389                                 "Quotacheck: Failed to reset quota flags.");
1390                 }
1391         } else
1392                 xfs_notice(mp, "Quotacheck: Done.");
1393         return (error);
1394 }
1395
1396 /*
1397  * This is called after the superblock has been read in and we're ready to
1398  * iget the quota inodes.
1399  */
1400 STATIC int
1401 xfs_qm_init_quotainos(
1402         xfs_mount_t     *mp)
1403 {
1404         xfs_inode_t     *uip, *gip;
1405         int             error;
1406         __int64_t       sbflags;
1407         uint            flags;
1408
1409         ASSERT(mp->m_quotainfo);
1410         uip = gip = NULL;
1411         sbflags = 0;
1412         flags = 0;
1413
1414         /*
1415          * Get the uquota and gquota inodes
1416          */
1417         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1418                 if (XFS_IS_UQUOTA_ON(mp) &&
1419                     mp->m_sb.sb_uquotino != NULLFSINO) {
1420                         ASSERT(mp->m_sb.sb_uquotino > 0);
1421                         if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1422                                              0, 0, &uip)))
1423                                 return XFS_ERROR(error);
1424                 }
1425                 if (XFS_IS_OQUOTA_ON(mp) &&
1426                     mp->m_sb.sb_gquotino != NULLFSINO) {
1427                         ASSERT(mp->m_sb.sb_gquotino > 0);
1428                         if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1429                                              0, 0, &gip))) {
1430                                 if (uip)
1431                                         IRELE(uip);
1432                                 return XFS_ERROR(error);
1433                         }
1434                 }
1435         } else {
1436                 flags |= XFS_QMOPT_SBVERSION;
1437                 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1438                             XFS_SB_GQUOTINO | XFS_SB_QFLAGS);
1439         }
1440
1441         /*
1442          * Create the two inodes, if they don't exist already. The changes
1443          * made above will get added to a transaction and logged in one of
1444          * the qino_alloc calls below.  If the device is readonly,
1445          * temporarily switch to read-write to do this.
1446          */
1447         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1448                 if ((error = xfs_qm_qino_alloc(mp, &uip,
1449                                               sbflags | XFS_SB_UQUOTINO,
1450                                               flags | XFS_QMOPT_UQUOTA)))
1451                         return XFS_ERROR(error);
1452
1453                 flags &= ~XFS_QMOPT_SBVERSION;
1454         }
1455         if (XFS_IS_OQUOTA_ON(mp) && gip == NULL) {
1456                 flags |= (XFS_IS_GQUOTA_ON(mp) ?
1457                                 XFS_QMOPT_GQUOTA : XFS_QMOPT_PQUOTA);
1458                 error = xfs_qm_qino_alloc(mp, &gip,
1459                                           sbflags | XFS_SB_GQUOTINO, flags);
1460                 if (error) {
1461                         if (uip)
1462                                 IRELE(uip);
1463
1464                         return XFS_ERROR(error);
1465                 }
1466         }
1467
1468         mp->m_quotainfo->qi_uquotaip = uip;
1469         mp->m_quotainfo->qi_gquotaip = gip;
1470
1471         return 0;
1472 }
1473
1474 STATIC void
1475 xfs_qm_dqfree_one(
1476         struct xfs_dquot        *dqp)
1477 {
1478         struct xfs_mount        *mp = dqp->q_mount;
1479         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1480
1481         mutex_lock(&qi->qi_tree_lock);
1482         radix_tree_delete(XFS_DQUOT_TREE(qi, dqp->q_core.d_flags),
1483                           be32_to_cpu(dqp->q_core.d_id));
1484
1485         qi->qi_dquots--;
1486         mutex_unlock(&qi->qi_tree_lock);
1487
1488         xfs_qm_dqdestroy(dqp);
1489 }
1490
1491 STATIC void
1492 xfs_qm_dqreclaim_one(
1493         struct xfs_dquot        *dqp,
1494         struct list_head        *buffer_list,
1495         struct list_head        *dispose_list)
1496 {
1497         struct xfs_mount        *mp = dqp->q_mount;
1498         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1499         int                     error;
1500
1501         if (!xfs_dqlock_nowait(dqp))
1502                 goto out_move_tail;
1503
1504         /*
1505          * This dquot has acquired a reference in the meantime remove it from
1506          * the freelist and try again.
1507          */
1508         if (dqp->q_nrefs) {
1509                 xfs_dqunlock(dqp);
1510
1511                 trace_xfs_dqreclaim_want(dqp);
1512                 XFS_STATS_INC(xs_qm_dqwants);
1513
1514                 list_del_init(&dqp->q_lru);
1515                 qi->qi_lru_count--;
1516                 XFS_STATS_DEC(xs_qm_dquot_unused);
1517                 return;
1518         }
1519
1520         /*
1521          * Try to grab the flush lock. If this dquot is in the process of
1522          * getting flushed to disk, we don't want to reclaim it.
1523          */
1524         if (!xfs_dqflock_nowait(dqp))
1525                 goto out_unlock_move_tail;
1526
1527         if (XFS_DQ_IS_DIRTY(dqp)) {
1528                 struct xfs_buf  *bp = NULL;
1529
1530                 trace_xfs_dqreclaim_dirty(dqp);
1531
1532                 error = xfs_qm_dqflush(dqp, &bp);
1533                 if (error) {
1534                         xfs_warn(mp, "%s: dquot %p flush failed",
1535                                  __func__, dqp);
1536                         goto out_unlock_move_tail;
1537                 }
1538
1539                 xfs_buf_delwri_queue(bp, buffer_list);
1540                 xfs_buf_relse(bp);
1541                 /*
1542                  * Give the dquot another try on the freelist, as the
1543                  * flushing will take some time.
1544                  */
1545                 goto out_unlock_move_tail;
1546         }
1547         xfs_dqfunlock(dqp);
1548
1549         /*
1550          * Prevent lookups now that we are past the point of no return.
1551          */
1552         dqp->dq_flags |= XFS_DQ_FREEING;
1553         xfs_dqunlock(dqp);
1554
1555         ASSERT(dqp->q_nrefs == 0);
1556         list_move_tail(&dqp->q_lru, dispose_list);
1557         qi->qi_lru_count--;
1558         XFS_STATS_DEC(xs_qm_dquot_unused);
1559
1560         trace_xfs_dqreclaim_done(dqp);
1561         XFS_STATS_INC(xs_qm_dqreclaims);
1562         return;
1563
1564         /*
1565          * Move the dquot to the tail of the list so that we don't spin on it.
1566          */
1567 out_unlock_move_tail:
1568         xfs_dqunlock(dqp);
1569 out_move_tail:
1570         list_move_tail(&dqp->q_lru, &qi->qi_lru_list);
1571         trace_xfs_dqreclaim_busy(dqp);
1572         XFS_STATS_INC(xs_qm_dqreclaim_misses);
1573 }
1574
1575 STATIC int
1576 xfs_qm_shake(
1577         struct shrinker         *shrink,
1578         struct shrink_control   *sc)
1579 {
1580         struct xfs_quotainfo    *qi =
1581                 container_of(shrink, struct xfs_quotainfo, qi_shrinker);
1582         int                     nr_to_scan = sc->nr_to_scan;
1583         LIST_HEAD               (buffer_list);
1584         LIST_HEAD               (dispose_list);
1585         struct xfs_dquot        *dqp;
1586         int                     error;
1587
1588         if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
1589                 return 0;
1590         if (!nr_to_scan)
1591                 goto out;
1592
1593         mutex_lock(&qi->qi_lru_lock);
1594         while (!list_empty(&qi->qi_lru_list)) {
1595                 if (nr_to_scan-- <= 0)
1596                         break;
1597                 dqp = list_first_entry(&qi->qi_lru_list, struct xfs_dquot,
1598                                        q_lru);
1599                 xfs_qm_dqreclaim_one(dqp, &buffer_list, &dispose_list);
1600         }
1601         mutex_unlock(&qi->qi_lru_lock);
1602
1603         error = xfs_buf_delwri_submit(&buffer_list);
1604         if (error)
1605                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
1606
1607         while (!list_empty(&dispose_list)) {
1608                 dqp = list_first_entry(&dispose_list, struct xfs_dquot, q_lru);
1609                 list_del_init(&dqp->q_lru);
1610                 xfs_qm_dqfree_one(dqp);
1611         }
1612
1613 out:
1614         return (qi->qi_lru_count / 100) * sysctl_vfs_cache_pressure;
1615 }
1616
1617 /*
1618  * Start a transaction and write the incore superblock changes to
1619  * disk. flags parameter indicates which fields have changed.
1620  */
1621 int
1622 xfs_qm_write_sb_changes(
1623         xfs_mount_t     *mp,
1624         __int64_t       flags)
1625 {
1626         xfs_trans_t     *tp;
1627         int             error;
1628
1629         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1630         error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp),
1631                                   0, 0, XFS_DEFAULT_LOG_COUNT);
1632         if (error) {
1633                 xfs_trans_cancel(tp, 0);
1634                 return error;
1635         }
1636
1637         xfs_mod_sb(tp, flags);
1638         error = xfs_trans_commit(tp, 0);
1639
1640         return error;
1641 }
1642
1643
1644 /* --------------- utility functions for vnodeops ---------------- */
1645
1646
1647 /*
1648  * Given an inode, a uid, gid and prid make sure that we have
1649  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1650  * quotas by creating this file.
1651  * This also attaches dquot(s) to the given inode after locking it,
1652  * and returns the dquots corresponding to the uid and/or gid.
1653  *
1654  * in   : inode (unlocked)
1655  * out  : udquot, gdquot with references taken and unlocked
1656  */
1657 int
1658 xfs_qm_vop_dqalloc(
1659         struct xfs_inode        *ip,
1660         uid_t                   uid,
1661         gid_t                   gid,
1662         prid_t                  prid,
1663         uint                    flags,
1664         struct xfs_dquot        **O_udqpp,
1665         struct xfs_dquot        **O_gdqpp)
1666 {
1667         struct xfs_mount        *mp = ip->i_mount;
1668         struct xfs_dquot        *uq, *gq;
1669         int                     error;
1670         uint                    lockflags;
1671
1672         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1673                 return 0;
1674
1675         lockflags = XFS_ILOCK_EXCL;
1676         xfs_ilock(ip, lockflags);
1677
1678         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1679                 gid = ip->i_d.di_gid;
1680
1681         /*
1682          * Attach the dquot(s) to this inode, doing a dquot allocation
1683          * if necessary. The dquot(s) will not be locked.
1684          */
1685         if (XFS_NOT_DQATTACHED(mp, ip)) {
1686                 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1687                 if (error) {
1688                         xfs_iunlock(ip, lockflags);
1689                         return error;
1690                 }
1691         }
1692
1693         uq = gq = NULL;
1694         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1695                 if (ip->i_d.di_uid != uid) {
1696                         /*
1697                          * What we need is the dquot that has this uid, and
1698                          * if we send the inode to dqget, the uid of the inode
1699                          * takes priority over what's sent in the uid argument.
1700                          * We must unlock inode here before calling dqget if
1701                          * we're not sending the inode, because otherwise
1702                          * we'll deadlock by doing trans_reserve while
1703                          * holding ilock.
1704                          */
1705                         xfs_iunlock(ip, lockflags);
1706                         if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t) uid,
1707                                                  XFS_DQ_USER,
1708                                                  XFS_QMOPT_DQALLOC |
1709                                                  XFS_QMOPT_DOWARN,
1710                                                  &uq))) {
1711                                 ASSERT(error != ENOENT);
1712                                 return error;
1713                         }
1714                         /*
1715                          * Get the ilock in the right order.
1716                          */
1717                         xfs_dqunlock(uq);
1718                         lockflags = XFS_ILOCK_SHARED;
1719                         xfs_ilock(ip, lockflags);
1720                 } else {
1721                         /*
1722                          * Take an extra reference, because we'll return
1723                          * this to caller
1724                          */
1725                         ASSERT(ip->i_udquot);
1726                         uq = xfs_qm_dqhold(ip->i_udquot);
1727                 }
1728         }
1729         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1730                 if (ip->i_d.di_gid != gid) {
1731                         xfs_iunlock(ip, lockflags);
1732                         if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)gid,
1733                                                  XFS_DQ_GROUP,
1734                                                  XFS_QMOPT_DQALLOC |
1735                                                  XFS_QMOPT_DOWARN,
1736                                                  &gq))) {
1737                                 if (uq)
1738                                         xfs_qm_dqrele(uq);
1739                                 ASSERT(error != ENOENT);
1740                                 return error;
1741                         }
1742                         xfs_dqunlock(gq);
1743                         lockflags = XFS_ILOCK_SHARED;
1744                         xfs_ilock(ip, lockflags);
1745                 } else {
1746                         ASSERT(ip->i_gdquot);
1747                         gq = xfs_qm_dqhold(ip->i_gdquot);
1748                 }
1749         } else if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1750                 if (xfs_get_projid(ip) != prid) {
1751                         xfs_iunlock(ip, lockflags);
1752                         if ((error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1753                                                  XFS_DQ_PROJ,
1754                                                  XFS_QMOPT_DQALLOC |
1755                                                  XFS_QMOPT_DOWARN,
1756                                                  &gq))) {
1757                                 if (uq)
1758                                         xfs_qm_dqrele(uq);
1759                                 ASSERT(error != ENOENT);
1760                                 return (error);
1761                         }
1762                         xfs_dqunlock(gq);
1763                         lockflags = XFS_ILOCK_SHARED;
1764                         xfs_ilock(ip, lockflags);
1765                 } else {
1766                         ASSERT(ip->i_gdquot);
1767                         gq = xfs_qm_dqhold(ip->i_gdquot);
1768                 }
1769         }
1770         if (uq)
1771                 trace_xfs_dquot_dqalloc(ip);
1772
1773         xfs_iunlock(ip, lockflags);
1774         if (O_udqpp)
1775                 *O_udqpp = uq;
1776         else if (uq)
1777                 xfs_qm_dqrele(uq);
1778         if (O_gdqpp)
1779                 *O_gdqpp = gq;
1780         else if (gq)
1781                 xfs_qm_dqrele(gq);
1782         return 0;
1783 }
1784
1785 /*
1786  * Actually transfer ownership, and do dquot modifications.
1787  * These were already reserved.
1788  */
1789 xfs_dquot_t *
1790 xfs_qm_vop_chown(
1791         xfs_trans_t     *tp,
1792         xfs_inode_t     *ip,
1793         xfs_dquot_t     **IO_olddq,
1794         xfs_dquot_t     *newdq)
1795 {
1796         xfs_dquot_t     *prevdq;
1797         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1798                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1799
1800
1801         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1802         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1803
1804         /* old dquot */
1805         prevdq = *IO_olddq;
1806         ASSERT(prevdq);
1807         ASSERT(prevdq != newdq);
1808
1809         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1810         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1811
1812         /* the sparkling new dquot */
1813         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1814         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1815
1816         /*
1817          * Take an extra reference, because the inode is going to keep
1818          * this dquot pointer even after the trans_commit.
1819          */
1820         *IO_olddq = xfs_qm_dqhold(newdq);
1821
1822         return prevdq;
1823 }
1824
1825 /*
1826  * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1827  */
1828 int
1829 xfs_qm_vop_chown_reserve(
1830         xfs_trans_t     *tp,
1831         xfs_inode_t     *ip,
1832         xfs_dquot_t     *udqp,
1833         xfs_dquot_t     *gdqp,
1834         uint            flags)
1835 {
1836         xfs_mount_t     *mp = ip->i_mount;
1837         uint            delblks, blkflags, prjflags = 0;
1838         xfs_dquot_t     *unresudq, *unresgdq, *delblksudq, *delblksgdq;
1839         int             error;
1840
1841
1842         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1843         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1844
1845         delblks = ip->i_delayed_blks;
1846         delblksudq = delblksgdq = unresudq = unresgdq = NULL;
1847         blkflags = XFS_IS_REALTIME_INODE(ip) ?
1848                         XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1849
1850         if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1851             ip->i_d.di_uid != (uid_t)be32_to_cpu(udqp->q_core.d_id)) {
1852                 delblksudq = udqp;
1853                 /*
1854                  * If there are delayed allocation blocks, then we have to
1855                  * unreserve those from the old dquot, and add them to the
1856                  * new dquot.
1857                  */
1858                 if (delblks) {
1859                         ASSERT(ip->i_udquot);
1860                         unresudq = ip->i_udquot;
1861                 }
1862         }
1863         if (XFS_IS_OQUOTA_ON(ip->i_mount) && gdqp) {
1864                 if (XFS_IS_PQUOTA_ON(ip->i_mount) &&
1865                      xfs_get_projid(ip) != be32_to_cpu(gdqp->q_core.d_id))
1866                         prjflags = XFS_QMOPT_ENOSPC;
1867
1868                 if (prjflags ||
1869                     (XFS_IS_GQUOTA_ON(ip->i_mount) &&
1870                      ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id))) {
1871                         delblksgdq = gdqp;
1872                         if (delblks) {
1873                                 ASSERT(ip->i_gdquot);
1874                                 unresgdq = ip->i_gdquot;
1875                         }
1876                 }
1877         }
1878
1879         if ((error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
1880                                 delblksudq, delblksgdq, ip->i_d.di_nblocks, 1,
1881                                 flags | blkflags | prjflags)))
1882                 return (error);
1883
1884         /*
1885          * Do the delayed blks reservations/unreservations now. Since, these
1886          * are done without the help of a transaction, if a reservation fails
1887          * its previous reservations won't be automatically undone by trans
1888          * code. So, we have to do it manually here.
1889          */
1890         if (delblks) {
1891                 /*
1892                  * Do the reservations first. Unreservation can't fail.
1893                  */
1894                 ASSERT(delblksudq || delblksgdq);
1895                 ASSERT(unresudq || unresgdq);
1896                 if ((error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1897                                 delblksudq, delblksgdq, (xfs_qcnt_t)delblks, 0,
1898                                 flags | blkflags | prjflags)))
1899                         return (error);
1900                 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
1901                                 unresudq, unresgdq, -((xfs_qcnt_t)delblks), 0,
1902                                 blkflags);
1903         }
1904
1905         return (0);
1906 }
1907
1908 int
1909 xfs_qm_vop_rename_dqattach(
1910         struct xfs_inode        **i_tab)
1911 {
1912         struct xfs_mount        *mp = i_tab[0]->i_mount;
1913         int                     i;
1914
1915         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1916                 return 0;
1917
1918         for (i = 0; (i < 4 && i_tab[i]); i++) {
1919                 struct xfs_inode        *ip = i_tab[i];
1920                 int                     error;
1921
1922                 /*
1923                  * Watch out for duplicate entries in the table.
1924                  */
1925                 if (i == 0 || ip != i_tab[i-1]) {
1926                         if (XFS_NOT_DQATTACHED(mp, ip)) {
1927                                 error = xfs_qm_dqattach(ip, 0);
1928                                 if (error)
1929                                         return error;
1930                         }
1931                 }
1932         }
1933         return 0;
1934 }
1935
1936 void
1937 xfs_qm_vop_create_dqattach(
1938         struct xfs_trans        *tp,
1939         struct xfs_inode        *ip,
1940         struct xfs_dquot        *udqp,
1941         struct xfs_dquot        *gdqp)
1942 {
1943         struct xfs_mount        *mp = tp->t_mountp;
1944
1945         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1946                 return;
1947
1948         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1949         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1950
1951         if (udqp) {
1952                 ASSERT(ip->i_udquot == NULL);
1953                 ASSERT(XFS_IS_UQUOTA_ON(mp));
1954                 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
1955
1956                 ip->i_udquot = xfs_qm_dqhold(udqp);
1957                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
1958         }
1959         if (gdqp) {
1960                 ASSERT(ip->i_gdquot == NULL);
1961                 ASSERT(XFS_IS_OQUOTA_ON(mp));
1962                 ASSERT((XFS_IS_GQUOTA_ON(mp) ?
1963                         ip->i_d.di_gid : xfs_get_projid(ip)) ==
1964                                 be32_to_cpu(gdqp->q_core.d_id));
1965
1966                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
1967                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
1968         }
1969 }
1970