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_format.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_alloc.h"
  27 #include "xfs_quota.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_ialloc_btree.h"
  31 #include "xfs_dinode.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_ialloc.h"
  34 #include "xfs_itable.h"
  35 #include "xfs_rtalloc.h"
  36 #include "xfs_error.h"
  37 #include "xfs_bmap.h"
  38 #include "xfs_attr.h"
  39 #include "xfs_buf_item.h"
  40 #include "xfs_trans_space.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         struct xfs_dquot        *pdqp = NULL;
 141
 142         xfs_dqlock(dqp);
 143         if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 144                 xfs_dqunlock(dqp);
 145                 return EAGAIN;
 146         }
 147
 148         /*
 149          * If this quota has a hint attached, prepare for releasing it now.
 150          */
 151         gdqp = dqp->q_gdquot;
 152         if (gdqp) {
 153                 xfs_dqlock(gdqp);
 154                 dqp->q_gdquot = NULL;
 155         }
 156
 157         pdqp = dqp->q_pdquot;
 158         if (pdqp) {
 159                 xfs_dqlock(pdqp);
 160                 dqp->q_pdquot = NULL;
 161         }
 162
 163         dqp->dq_flags |= XFS_DQ_FREEING;
 164
 165         xfs_dqflock(dqp);
 166
 167         /*
 168          * If we are turning this type of quotas off, we don't care
 169          * about the dirty metadata sitting in this dquot. OTOH, if
 170          * we're unmounting, we do care, so we flush it and wait.
 171          */
 172         if (XFS_DQ_IS_DIRTY(dqp)) {
 173                 struct xfs_buf  *bp = NULL;
 174                 int             error;
 175
 176                 /*
 177                  * We don't care about getting disk errors here. We need
 178                  * to purge this dquot anyway, so we go ahead regardless.
 179                  */
 180                 error = xfs_qm_dqflush(dqp, &bp);
 181                 if (error) {
 182                         xfs_warn(mp, "%s: dquot %p flush failed",
 183                                 __func__, dqp);
 184                 } else {
 185                         error = xfs_bwrite(bp);
 186                         xfs_buf_relse(bp);
 187                 }
 188                 xfs_dqflock(dqp);
 189         }
 190
 191         ASSERT(atomic_read(&dqp->q_pincount) == 0);
 192         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 193                !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 194
 195         xfs_dqfunlock(dqp);
 196         xfs_dqunlock(dqp);
 197
 198         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
 199                           be32_to_cpu(dqp->q_core.d_id));
 200         qi->qi_dquots--;
 201
 202         /*
 203          * We move dquots to the freelist as soon as their reference count
 204          * hits zero, so it really should be on the freelist here.
 205          */
 206         mutex_lock(&qi->qi_lru_lock);
 207         ASSERT(!list_empty(&dqp->q_lru));
 208         list_del_init(&dqp->q_lru);
 209         qi->qi_lru_count--;
 210         XFS_STATS_DEC(xs_qm_dquot_unused);
 211         mutex_unlock(&qi->qi_lru_lock);
 212
 213         xfs_qm_dqdestroy(dqp);
 214
 215         if (gdqp)
 216                 xfs_qm_dqput(gdqp);
 217         if (pdqp)
 218                 xfs_qm_dqput(pdqp);
 219         return 0;
 220 }
 221
 222 /*
 223  * Purge the dquot cache.
 224  */
 225 void
 226 xfs_qm_dqpurge_all(
 227         struct xfs_mount        *mp,
 228         uint                    flags)
 229 {
 230         if (flags & XFS_QMOPT_UQUOTA)
 231                 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
 232         if (flags & XFS_QMOPT_GQUOTA)
 233                 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 234         if (flags & XFS_QMOPT_PQUOTA)
 235                 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 236 }
 237
 238 /*
 239  * Just destroy the quotainfo structure.
 240  */
 241 void
 242 xfs_qm_unmount(
 243         struct xfs_mount        *mp)
 244 {
 245         if (mp->m_quotainfo) {
 246                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 247                 xfs_qm_destroy_quotainfo(mp);
 248         }
 249 }
 250
 251
 252 /*
 253  * This is called from xfs_mountfs to start quotas and initialize all
 254  * necessary data structures like quotainfo.  This is also responsible for
 255  * running a quotacheck as necessary.  We are guaranteed that the superblock
 256  * is consistently read in at this point.
 257  *
 258  * If we fail here, the mount will continue with quota turned off. We don't
 259  * need to inidicate success or failure at all.
 260  */
 261 void
 262 xfs_qm_mount_quotas(
 263         xfs_mount_t     *mp)
 264 {
 265         int             error = 0;
 266         uint            sbf;
 267
 268         /*
 269          * If quotas on realtime volumes is not supported, we disable
 270          * quotas immediately.
 271          */
 272         if (mp->m_sb.sb_rextents) {
 273                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 274                 mp->m_qflags = 0;
 275                 goto write_changes;
 276         }
 277
 278         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 279
 280         /*
 281          * Allocate the quotainfo structure inside the mount struct, and
 282          * create quotainode(s), and change/rev superblock if necessary.
 283          */
 284         error = xfs_qm_init_quotainfo(mp);
 285         if (error) {
 286                 /*
 287                  * We must turn off quotas.
 288                  */
 289                 ASSERT(mp->m_quotainfo == NULL);
 290                 mp->m_qflags = 0;
 291                 goto write_changes;
 292         }
 293         /*
 294          * If any of the quotas are not consistent, do a quotacheck.
 295          */
 296         if (XFS_QM_NEED_QUOTACHECK(mp)) {
 297                 error = xfs_qm_quotacheck(mp);
 298                 if (error) {
 299                         /* Quotacheck failed and disabled quotas. */
 300                         return;
 301                 }
 302         }
 303         /*
 304          * If one type of quotas is off, then it will lose its
 305          * quotachecked status, since we won't be doing accounting for
 306          * that type anymore.
 307          */
 308         if (!XFS_IS_UQUOTA_ON(mp))
 309                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 310         if (!XFS_IS_GQUOTA_ON(mp))
 311                 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
 312         if (!XFS_IS_PQUOTA_ON(mp))
 313                 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
 314
 315  write_changes:
 316         /*
 317          * We actually don't have to acquire the m_sb_lock at all.
 318          * This can only be called from mount, and that's single threaded. XXX
 319          */
 320         spin_lock(&mp->m_sb_lock);
 321         sbf = mp->m_sb.sb_qflags;
 322         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 323         spin_unlock(&mp->m_sb_lock);
 324
 325         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 326                 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 327                         /*
 328                          * We could only have been turning quotas off.
 329                          * We aren't in very good shape actually because
 330                          * the incore structures are convinced that quotas are
 331                          * off, but the on disk superblock doesn't know that !
 332                          */
 333                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 334                         xfs_alert(mp, "%s: Superblock update failed!",
 335                                 __func__);
 336                 }
 337         }
 338
 339         if (error) {
 340                 xfs_warn(mp, "Failed to initialize disk quotas.");
 341                 return;
 342         }
 343 }
 344
 345 /*
 346  * Called from the vfsops layer.
 347  */
 348 void
 349 xfs_qm_unmount_quotas(
 350         xfs_mount_t     *mp)
 351 {
 352         /*
 353          * Release the dquots that root inode, et al might be holding,
 354          * before we flush quotas and blow away the quotainfo structure.
 355          */
 356         ASSERT(mp->m_rootip);
 357         xfs_qm_dqdetach(mp->m_rootip);
 358         if (mp->m_rbmip)
 359                 xfs_qm_dqdetach(mp->m_rbmip);
 360         if (mp->m_rsumip)
 361                 xfs_qm_dqdetach(mp->m_rsumip);
 362
 363         /*
 364          * Release the quota inodes.
 365          */
 366         if (mp->m_quotainfo) {
 367                 if (mp->m_quotainfo->qi_uquotaip) {
 368                         IRELE(mp->m_quotainfo->qi_uquotaip);
 369                         mp->m_quotainfo->qi_uquotaip = NULL;
 370                 }
 371                 if (mp->m_quotainfo->qi_gquotaip) {
 372                         IRELE(mp->m_quotainfo->qi_gquotaip);
 373                         mp->m_quotainfo->qi_gquotaip = NULL;
 374                 }
 375                 if (mp->m_quotainfo->qi_pquotaip) {
 376                         IRELE(mp->m_quotainfo->qi_pquotaip);
 377                         mp->m_quotainfo->qi_pquotaip = NULL;
 378                 }
 379         }
 380 }
 381
 382 STATIC int
 383 xfs_qm_dqattach_one(
 384         xfs_inode_t     *ip,
 385         xfs_dqid_t      id,
 386         uint            type,
 387         uint            doalloc,
 388         xfs_dquot_t     *udqhint, /* hint */
 389         xfs_dquot_t     **IO_idqpp)
 390 {
 391         xfs_dquot_t     *dqp;
 392         int             error;
 393
 394         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 395         error = 0;
 396
 397         /*
 398          * See if we already have it in the inode itself. IO_idqpp is
 399          * &i_udquot or &i_gdquot. This made the code look weird, but
 400          * made the logic a lot simpler.
 401          */
 402         dqp = *IO_idqpp;
 403         if (dqp) {
 404                 trace_xfs_dqattach_found(dqp);
 405                 return 0;
 406         }
 407
 408         /*
 409          * udqhint is the i_udquot field in inode, and is non-NULL only
 410          * when the type arg is group/project. Its purpose is to save a
 411          * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 412          * the user dquot.
 413          */
 414         if (udqhint) {
 415                 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 416                 xfs_dqlock(udqhint);
 417
 418                 /*
 419                  * No need to take dqlock to look at the id.
 420                  *
 421                  * The ID can't change until it gets reclaimed, and it won't
 422                  * be reclaimed as long as we have a ref from inode and we
 423                  * hold the ilock.
 424                  */
 425                 if (type == XFS_DQ_GROUP)
 426                         dqp = udqhint->q_gdquot;
 427                 else
 428                         dqp = udqhint->q_pdquot;
 429                 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 430                         ASSERT(*IO_idqpp == NULL);
 431
 432                         *IO_idqpp = xfs_qm_dqhold(dqp);
 433                         xfs_dqunlock(udqhint);
 434                         return 0;
 435                 }
 436
 437                 /*
 438                  * We can't hold a dquot lock when we call the dqget code.
 439                  * We'll deadlock in no time, because of (not conforming to)
 440                  * lock ordering - the inodelock comes before any dquot lock,
 441                  * and we may drop and reacquire the ilock in xfs_qm_dqget().
 442                  */
 443                 xfs_dqunlock(udqhint);
 444         }
 445
 446         /*
 447          * Find the dquot from somewhere. This bumps the
 448          * reference count of dquot and returns it locked.
 449          * This can return ENOENT if dquot didn't exist on
 450          * disk and we didn't ask it to allocate;
 451          * ESRCH if quotas got turned off suddenly.
 452          */
 453         error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 454                              doalloc | XFS_QMOPT_DOWARN, &dqp);
 455         if (error)
 456                 return error;
 457
 458         trace_xfs_dqattach_get(dqp);
 459
 460         /*
 461          * dqget may have dropped and re-acquired the ilock, but it guarantees
 462          * that the dquot returned is the one that should go in the inode.
 463          */
 464         *IO_idqpp = dqp;
 465         xfs_dqunlock(dqp);
 466         return 0;
 467 }
 468
 469
 470 /*
 471  * Given a udquot and group/project type, attach the group/project
 472  * dquot pointer to the udquot as a hint for future lookups.
 473  */
 474 STATIC void
 475 xfs_qm_dqattach_hint(
 476         struct xfs_inode        *ip,
 477         int                     type)
 478 {
 479         struct xfs_dquot **dqhintp;
 480         struct xfs_dquot *dqp;
 481         struct xfs_dquot *udq = ip->i_udquot;
 482
 483         ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 484
 485         xfs_dqlock(udq);
 486
 487         if (type == XFS_DQ_GROUP) {
 488                 dqp = ip->i_gdquot;
 489                 dqhintp = &udq->q_gdquot;
 490         } else {
 491                 dqp = ip->i_pdquot;
 492                 dqhintp = &udq->q_pdquot;
 493         }
 494
 495         if (*dqhintp) {
 496                 struct xfs_dquot *tmp;
 497
 498                 if (*dqhintp == dqp)
 499                         goto done;
 500
 501                 tmp = *dqhintp;
 502                 *dqhintp = NULL;
 503                 xfs_qm_dqrele(tmp);
 504         }
 505
 506         *dqhintp = xfs_qm_dqhold(dqp);
 507 done:
 508         xfs_dqunlock(udq);
 509 }
 510
 511 static bool
 512 xfs_qm_need_dqattach(
 513         struct xfs_inode        *ip)
 514 {
 515         struct xfs_mount        *mp = ip->i_mount;
 516
 517         if (!XFS_IS_QUOTA_RUNNING(mp))
 518                 return false;
 519         if (!XFS_IS_QUOTA_ON(mp))
 520                 return false;
 521         if (!XFS_NOT_DQATTACHED(mp, ip))
 522                 return false;
 523         if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
 524                 return false;
 525         return true;
 526 }
 527
 528 /*
 529  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 530  * into account.
 531  * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 532  * Inode may get unlocked and relocked in here, and the caller must deal with
 533  * the consequences.
 534  */
 535 int
 536 xfs_qm_dqattach_locked(
 537         xfs_inode_t     *ip,
 538         uint            flags)
 539 {
 540         xfs_mount_t     *mp = ip->i_mount;
 541         uint            nquotas = 0;
 542         int             error = 0;
 543
 544         if (!xfs_qm_need_dqattach(ip))
 545                 return 0;
 546
 547         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 548
 549         if (XFS_IS_UQUOTA_ON(mp)) {
 550                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 551                                                 flags & XFS_QMOPT_DQALLOC,
 552                                                 NULL, &ip->i_udquot);
 553                 if (error)
 554                         goto done;
 555                 nquotas++;
 556         }
 557
 558         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 559         if (XFS_IS_GQUOTA_ON(mp)) {
 560                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 561                                                 flags & XFS_QMOPT_DQALLOC,
 562                                                 ip->i_udquot, &ip->i_gdquot);
 563                 /*
 564                  * Don't worry about the udquot that we may have
 565                  * attached above. It'll get detached, if not already.
 566                  */
 567                 if (error)
 568                         goto done;
 569                 nquotas++;
 570         }
 571
 572         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 573         if (XFS_IS_PQUOTA_ON(mp)) {
 574                 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 575                                                 flags & XFS_QMOPT_DQALLOC,
 576                                                 ip->i_udquot, &ip->i_pdquot);
 577                 /*
 578                  * Don't worry about the udquot that we may have
 579                  * attached above. It'll get detached, if not already.
 580                  */
 581                 if (error)
 582                         goto done;
 583                 nquotas++;
 584         }
 585
 586         /*
 587          * Attach this group/project quota to the user quota as a hint.
 588          * This WON'T, in general, result in a thrash.
 589          */
 590         if (nquotas > 1 && ip->i_udquot) {
 591                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 592                 ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
 593                 ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
 594
 595                 /*
 596                  * We do not have i_udquot locked at this point, but this check
 597                  * is OK since we don't depend on the i_gdquot to be accurate
 598                  * 100% all the time. It is just a hint, and this will
 599                  * succeed in general.
 600                  */
 601                 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 602                         xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
 603
 604                 if (ip->i_udquot->q_pdquot != ip->i_pdquot)
 605                         xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
 606         }
 607
 608  done:
 609 #ifdef DEBUG
 610         if (!error) {
 611                 if (XFS_IS_UQUOTA_ON(mp))
 612                         ASSERT(ip->i_udquot);
 613                 if (XFS_IS_GQUOTA_ON(mp))
 614                         ASSERT(ip->i_gdquot);
 615                 if (XFS_IS_PQUOTA_ON(mp))
 616                         ASSERT(ip->i_pdquot);
 617         }
 618         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 619 #endif
 620         return error;
 621 }
 622
 623 int
 624 xfs_qm_dqattach(
 625         struct xfs_inode        *ip,
 626         uint                    flags)
 627 {
 628         int                     error;
 629
 630         if (!xfs_qm_need_dqattach(ip))
 631                 return 0;
 632
 633         xfs_ilock(ip, XFS_ILOCK_EXCL);
 634         error = xfs_qm_dqattach_locked(ip, flags);
 635         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 636
 637         return error;
 638 }
 639
 640 /*
 641  * Release dquots (and their references) if any.
 642  * The inode should be locked EXCL except when this's called by
 643  * xfs_ireclaim.
 644  */
 645 void
 646 xfs_qm_dqdetach(
 647         xfs_inode_t     *ip)
 648 {
 649         if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
 650                 return;
 651
 652         trace_xfs_dquot_dqdetach(ip);
 653
 654         ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
 655         if (ip->i_udquot) {
 656                 xfs_qm_dqrele(ip->i_udquot);
 657                 ip->i_udquot = NULL;
 658         }
 659         if (ip->i_gdquot) {
 660                 xfs_qm_dqrele(ip->i_gdquot);
 661                 ip->i_gdquot = NULL;
 662         }
 663         if (ip->i_pdquot) {
 664                 xfs_qm_dqrele(ip->i_pdquot);
 665                 ip->i_pdquot = NULL;
 666         }
 667 }
 668
 669 int
 670 xfs_qm_calc_dquots_per_chunk(
 671         struct xfs_mount        *mp,
 672         unsigned int            nbblks) /* basic block units */
 673 {
 674         unsigned int    ndquots;
 675
 676         ASSERT(nbblks > 0);
 677         ndquots = BBTOB(nbblks);
 678         do_div(ndquots, sizeof(xfs_dqblk_t));
 679
 680         return ndquots;
 681 }
 682
 683 /*
 684  * This initializes all the quota information that's kept in the
 685  * mount structure
 686  */
 687 STATIC int
 688 xfs_qm_init_quotainfo(
 689         xfs_mount_t     *mp)
 690 {
 691         xfs_quotainfo_t *qinf;
 692         int             error;
 693         xfs_dquot_t     *dqp;
 694
 695         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 696
 697         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 698
 699         /*
 700          * See if quotainodes are setup, and if not, allocate them,
 701          * and change the superblock accordingly.
 702          */
 703         if ((error = xfs_qm_init_quotainos(mp))) {
 704                 kmem_free(qinf);
 705                 mp->m_quotainfo = NULL;
 706                 return error;
 707         }
 708
 709         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 710         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 711         INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
 712         mutex_init(&qinf->qi_tree_lock);
 713
 714         INIT_LIST_HEAD(&qinf->qi_lru_list);
 715         qinf->qi_lru_count = 0;
 716         mutex_init(&qinf->qi_lru_lock);
 717
 718         /* mutex used to serialize quotaoffs */
 719         mutex_init(&qinf->qi_quotaofflock);
 720
 721         /* Precalc some constants */
 722         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 723         qinf->qi_dqperchunk = xfs_qm_calc_dquots_per_chunk(mp,
 724                                                         qinf->qi_dqchunklen);
 725
 726         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 727
 728         /*
 729          * We try to get the limits from the superuser's limits fields.
 730          * This is quite hacky, but it is standard quota practice.
 731          *
 732          * We look at the USR dquot with id == 0 first, but if user quotas
 733          * are not enabled we goto the GRP dquot with id == 0.
 734          * We don't really care to keep separate default limits for user
 735          * and group quotas, at least not at this point.
 736          *
 737          * Since we may not have done a quotacheck by this point, just read
 738          * the dquot without attaching it to any hashtables or lists.
 739          */
 740         error = xfs_qm_dqread(mp, 0,
 741                         XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 742                          (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 743                           XFS_DQ_PROJ),
 744                         XFS_QMOPT_DOWARN, &dqp);
 745         if (!error) {
 746                 xfs_disk_dquot_t        *ddqp = &dqp->q_core;
 747
 748                 /*
 749                  * The warnings and timers set the grace period given to
 750                  * a user or group before he or she can not perform any
 751                  * more writing. If it is zero, a default is used.
 752                  */
 753                 qinf->qi_btimelimit = ddqp->d_btimer ?
 754                         be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 755                 qinf->qi_itimelimit = ddqp->d_itimer ?
 756                         be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 757                 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 758                         be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 759                 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 760                         be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 761                 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 762                         be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 763                 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 764                         be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 765                 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 766                 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 767                 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 768                 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 769                 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 770                 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 771
 772                 xfs_qm_dqdestroy(dqp);
 773         } else {
 774                 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 775                 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 776                 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 777                 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 778                 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 779                 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 780         }
 781
 782         qinf->qi_shrinker.shrink = xfs_qm_shake;
 783         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 784         register_shrinker(&qinf->qi_shrinker);
 785         return 0;
 786 }
 787
 788
 789 /*
 790  * Gets called when unmounting a filesystem or when all quotas get
 791  * turned off.
 792  * This purges the quota inodes, destroys locks and frees itself.
 793  */
 794 void
 795 xfs_qm_destroy_quotainfo(
 796         xfs_mount_t     *mp)
 797 {
 798         xfs_quotainfo_t *qi;
 799
 800         qi = mp->m_quotainfo;
 801         ASSERT(qi != NULL);
 802
 803         unregister_shrinker(&qi->qi_shrinker);
 804
 805         if (qi->qi_uquotaip) {
 806                 IRELE(qi->qi_uquotaip);
 807                 qi->qi_uquotaip = NULL; /* paranoia */
 808         }
 809         if (qi->qi_gquotaip) {
 810                 IRELE(qi->qi_gquotaip);
 811                 qi->qi_gquotaip = NULL;
 812         }
 813         if (qi->qi_pquotaip) {
 814                 IRELE(qi->qi_pquotaip);
 815                 qi->qi_pquotaip = NULL;
 816         }
 817         mutex_destroy(&qi->qi_quotaofflock);
 818         kmem_free(qi);
 819         mp->m_quotainfo = NULL;
 820 }
 821
 822 /*
 823  * Create an inode and return with a reference already taken, but unlocked
 824  * This is how we create quota inodes
 825  */
 826 STATIC int
 827 xfs_qm_qino_alloc(
 828         xfs_mount_t     *mp,
 829         xfs_inode_t     **ip,
 830         __int64_t       sbfields,
 831         uint            flags)
 832 {
 833         xfs_trans_t     *tp;
 834         int             error;
 835         int             committed;
 836
 837         *ip = NULL;
 838         /*
 839          * With superblock that doesn't have separate pquotino, we
 840          * share an inode between gquota and pquota. If the on-disk
 841          * superblock has GQUOTA and the filesystem is now mounted
 842          * with PQUOTA, just use sb_gquotino for sb_pquotino and
 843          * vice-versa.
 844          */
 845         if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
 846                         (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
 847                 xfs_ino_t ino = NULLFSINO;
 848
 849                 if ((flags & XFS_QMOPT_PQUOTA) &&
 850                              (mp->m_sb.sb_gquotino != NULLFSINO)) {
 851                         ino = mp->m_sb.sb_gquotino;
 852                         ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
 853                 } else if ((flags & XFS_QMOPT_GQUOTA) &&
 854                              (mp->m_sb.sb_pquotino != NULLFSINO)) {
 855                         ino = mp->m_sb.sb_pquotino;
 856                         ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
 857                 }
 858                 if (ino != NULLFSINO) {
 859                         error = xfs_iget(mp, NULL, ino, 0, 0, ip);
 860                         if (error)
 861                                 return error;
 862                         mp->m_sb.sb_gquotino = NULLFSINO;
 863                         mp->m_sb.sb_pquotino = NULLFSINO;
 864                 }
 865         }
 866
 867         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
 868         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
 869                                   XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
 870         if (error) {
 871                 xfs_trans_cancel(tp, 0);
 872                 return error;
 873         }
 874
 875         if (!*ip) {
 876                 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
 877                                                                 &committed);
 878                 if (error) {
 879                         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
 880                                          XFS_TRANS_ABORT);
 881                         return error;
 882                 }
 883         }
 884
 885         /*
 886          * Make the changes in the superblock, and log those too.
 887          * sbfields arg may contain fields other than *QUOTINO;
 888          * VERSIONNUM for example.
 889          */
 890         spin_lock(&mp->m_sb_lock);
 891         if (flags & XFS_QMOPT_SBVERSION) {
 892                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
 893                 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
 894                         XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
 895                                 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
 896                                  XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
 897                                  XFS_SB_QFLAGS));
 898
 899                 xfs_sb_version_addquota(&mp->m_sb);
 900                 mp->m_sb.sb_uquotino = NULLFSINO;
 901                 mp->m_sb.sb_gquotino = NULLFSINO;
 902                 mp->m_sb.sb_pquotino = NULLFSINO;
 903
 904                 /* qflags will get updated fully _after_ quotacheck */
 905                 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
 906         }
 907         if (flags & XFS_QMOPT_UQUOTA)
 908                 mp->m_sb.sb_uquotino = (*ip)->i_ino;
 909         else if (flags & XFS_QMOPT_GQUOTA)
 910                 mp->m_sb.sb_gquotino = (*ip)->i_ino;
 911         else
 912                 mp->m_sb.sb_pquotino = (*ip)->i_ino;
 913         spin_unlock(&mp->m_sb_lock);
 914         xfs_mod_sb(tp, sbfields);
 915
 916         if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
 917                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
 918                 return error;
 919         }
 920         return 0;
 921 }
 922
 923
 924 STATIC void
 925 xfs_qm_reset_dqcounts(
 926         xfs_mount_t     *mp,
 927         xfs_buf_t       *bp,
 928         xfs_dqid_t      id,
 929         uint            type)
 930 {
 931         struct xfs_dqblk        *dqb;
 932         int                     j;
 933
 934         trace_xfs_reset_dqcounts(bp, _RET_IP_);
 935
 936         /*
 937          * Reset all counters and timers. They'll be
 938          * started afresh by xfs_qm_quotacheck.
 939          */
 940 #ifdef DEBUG
 941         j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 942         do_div(j, sizeof(xfs_dqblk_t));
 943         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
 944 #endif
 945         dqb = bp->b_addr;
 946         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
 947                 struct xfs_disk_dquot   *ddq;
 948
 949                 ddq = (struct xfs_disk_dquot *)&dqb[j];
 950
 951                 /*
 952                  * Do a sanity check, and if needed, repair the dqblk. Don't
 953                  * output any warnings because it's perfectly possible to
 954                  * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
 955                  */
 956                 (void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
 957                                       "xfs_quotacheck");
 958                 ddq->d_bcount = 0;
 959                 ddq->d_icount = 0;
 960                 ddq->d_rtbcount = 0;
 961                 ddq->d_btimer = 0;
 962                 ddq->d_itimer = 0;
 963                 ddq->d_rtbtimer = 0;
 964                 ddq->d_bwarns = 0;
 965                 ddq->d_iwarns = 0;
 966                 ddq->d_rtbwarns = 0;
 967
 968                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 969                         xfs_update_cksum((char *)&dqb[j],
 970                                          sizeof(struct xfs_dqblk),
 971                                          XFS_DQUOT_CRC_OFF);
 972                 }
 973         }
 974 }
 975
 976 STATIC int
 977 xfs_qm_dqiter_bufs(
 978         struct xfs_mount        *mp,
 979         xfs_dqid_t              firstid,
 980         xfs_fsblock_t           bno,
 981         xfs_filblks_t           blkcnt,
 982         uint                    flags,
 983         struct list_head        *buffer_list)
 984 {
 985         struct xfs_buf          *bp;
 986         int                     error;
 987         int                     type;
 988
 989         ASSERT(blkcnt > 0);
 990         type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
 991                 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
 992         error = 0;
 993
 994         /*
 995          * Blkcnt arg can be a very big number, and might even be
 996          * larger than the log itself. So, we have to break it up into
 997          * manageable-sized transactions.
 998          * Note that we don't start a permanent transaction here; we might
 999          * not be able to get a log reservation for the whole thing up front,
1000          * and we don't really care to either, because we just discard
1001          * everything if we were to crash in the middle of this loop.
1002          */
1003         while (blkcnt--) {
1004                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1005                               XFS_FSB_TO_DADDR(mp, bno),
1006                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1007                               &xfs_dquot_buf_ops);
1008
1009                 /*
1010                  * CRC and validation errors will return a EFSCORRUPTED here. If
1011                  * this occurs, re-read without CRC validation so that we can
1012                  * repair the damage via xfs_qm_reset_dqcounts(). This process
1013                  * will leave a trace in the log indicating corruption has
1014                  * been detected.
1015                  */
1016                 if (error == EFSCORRUPTED) {
1017                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1018                                       XFS_FSB_TO_DADDR(mp, bno),
1019                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1020                                       NULL);
1021                 }
1022
1023                 if (error)
1024                         break;
1025
1026                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1027                 xfs_buf_delwri_queue(bp, buffer_list);
1028                 xfs_buf_relse(bp);
1029
1030                 /* goto the next block. */
1031                 bno++;
1032                 firstid += mp->m_quotainfo->qi_dqperchunk;
1033         }
1034
1035         return error;
1036 }
1037
1038 /*
1039  * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1040  * caller supplied function for every chunk of dquots that we find.
1041  */
1042 STATIC int
1043 xfs_qm_dqiterate(
1044         struct xfs_mount        *mp,
1045         struct xfs_inode        *qip,
1046         uint                    flags,
1047         struct list_head        *buffer_list)
1048 {
1049         struct xfs_bmbt_irec    *map;
1050         int                     i, nmaps;       /* number of map entries */
1051         int                     error;          /* return value */
1052         xfs_fileoff_t           lblkno;
1053         xfs_filblks_t           maxlblkcnt;
1054         xfs_dqid_t              firstid;
1055         xfs_fsblock_t           rablkno;
1056         xfs_filblks_t           rablkcnt;
1057
1058         error = 0;
1059         /*
1060          * This looks racy, but we can't keep an inode lock across a
1061          * trans_reserve. But, this gets called during quotacheck, and that
1062          * happens only at mount time which is single threaded.
1063          */
1064         if (qip->i_d.di_nblocks == 0)
1065                 return 0;
1066
1067         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1068
1069         lblkno = 0;
1070         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1071         do {
1072                 nmaps = XFS_DQITER_MAP_SIZE;
1073                 /*
1074                  * We aren't changing the inode itself. Just changing
1075                  * some of its data. No new blocks are added here, and
1076                  * the inode is never added to the transaction.
1077                  */
1078                 xfs_ilock(qip, XFS_ILOCK_SHARED);
1079                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1080                                        map, &nmaps, 0);
1081                 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1082                 if (error)
1083                         break;
1084
1085                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1086                 for (i = 0; i < nmaps; i++) {
1087                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1088                         ASSERT(map[i].br_blockcount);
1089
1090
1091                         lblkno += map[i].br_blockcount;
1092
1093                         if (map[i].br_startblock == HOLESTARTBLOCK)
1094                                 continue;
1095
1096                         firstid = (xfs_dqid_t) map[i].br_startoff *
1097                                 mp->m_quotainfo->qi_dqperchunk;
1098                         /*
1099                          * Do a read-ahead on the next extent.
1100                          */
1101                         if ((i+1 < nmaps) &&
1102                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1103                                 rablkcnt =  map[i+1].br_blockcount;
1104                                 rablkno = map[i+1].br_startblock;
1105                                 while (rablkcnt--) {
1106                                         xfs_buf_readahead(mp->m_ddev_targp,
1107                                                XFS_FSB_TO_DADDR(mp, rablkno),
1108                                                mp->m_quotainfo->qi_dqchunklen,
1109                                                NULL);
1110                                         rablkno++;
1111                                 }
1112                         }
1113                         /*
1114                          * Iterate thru all the blks in the extent and
1115                          * reset the counters of all the dquots inside them.
1116                          */
1117                         error = xfs_qm_dqiter_bufs(mp, firstid,
1118                                                    map[i].br_startblock,
1119                                                    map[i].br_blockcount,
1120                                                    flags, buffer_list);
1121                         if (error)
1122                                 goto out;
1123                 }
1124         } while (nmaps > 0);
1125
1126 out:
1127         kmem_free(map);
1128         return error;
1129 }
1130
1131 /*
1132  * Called by dqusage_adjust in doing a quotacheck.
1133  *
1134  * Given the inode, and a dquot id this updates both the incore dqout as well
1135  * as the buffer copy. This is so that once the quotacheck is done, we can
1136  * just log all the buffers, as opposed to logging numerous updates to
1137  * individual dquots.
1138  */
1139 STATIC int
1140 xfs_qm_quotacheck_dqadjust(
1141         struct xfs_inode        *ip,
1142         xfs_dqid_t              id,
1143         uint                    type,
1144         xfs_qcnt_t              nblks,
1145         xfs_qcnt_t              rtblks)
1146 {
1147         struct xfs_mount        *mp = ip->i_mount;
1148         struct xfs_dquot        *dqp;
1149         int                     error;
1150
1151         error = xfs_qm_dqget(mp, ip, id, type,
1152                              XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1153         if (error) {
1154                 /*
1155                  * Shouldn't be able to turn off quotas here.
1156                  */
1157                 ASSERT(error != ESRCH);
1158                 ASSERT(error != ENOENT);
1159                 return error;
1160         }
1161
1162         trace_xfs_dqadjust(dqp);
1163
1164         /*
1165          * Adjust the inode count and the block count to reflect this inode's
1166          * resource usage.
1167          */
1168         be64_add_cpu(&dqp->q_core.d_icount, 1);
1169         dqp->q_res_icount++;
1170         if (nblks) {
1171                 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1172                 dqp->q_res_bcount += nblks;
1173         }
1174         if (rtblks) {
1175                 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1176                 dqp->q_res_rtbcount += rtblks;
1177         }
1178
1179         /*
1180          * Set default limits, adjust timers (since we changed usages)
1181          *
1182          * There are no timers for the default values set in the root dquot.
1183          */
1184         if (dqp->q_core.d_id) {
1185                 xfs_qm_adjust_dqlimits(mp, dqp);
1186                 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1187         }
1188
1189         dqp->dq_flags |= XFS_DQ_DIRTY;
1190         xfs_qm_dqput(dqp);
1191         return 0;
1192 }
1193
1194 STATIC int
1195 xfs_qm_get_rtblks(
1196         xfs_inode_t     *ip,
1197         xfs_qcnt_t      *O_rtblks)
1198 {
1199         xfs_filblks_t   rtblks;                 /* total rt blks */
1200         xfs_extnum_t    idx;                    /* extent record index */
1201         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1202         xfs_extnum_t    nextents;               /* number of extent entries */
1203         int             error;
1204
1205         ASSERT(XFS_IS_REALTIME_INODE(ip));
1206         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1207         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1208                 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1209                         return error;
1210         }
1211         rtblks = 0;
1212         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1213         for (idx = 0; idx < nextents; idx++)
1214                 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1215         *O_rtblks = (xfs_qcnt_t)rtblks;
1216         return 0;
1217 }
1218
1219 /*
1220  * callback routine supplied to bulkstat(). Given an inumber, find its
1221  * dquots and update them to account for resources taken by that inode.
1222  */
1223 /* ARGSUSED */
1224 STATIC int
1225 xfs_qm_dqusage_adjust(
1226         xfs_mount_t     *mp,            /* mount point for filesystem */
1227         xfs_ino_t       ino,            /* inode number to get data for */
1228         void            __user *buffer, /* not used */
1229         int             ubsize,         /* not used */
1230         int             *ubused,        /* not used */
1231         int             *res)           /* result code value */
1232 {
1233         xfs_inode_t     *ip;
1234         xfs_qcnt_t      nblks, rtblks = 0;
1235         int             error;
1236
1237         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1238
1239         /*
1240          * rootino must have its resources accounted for, not so with the quota
1241          * inodes.
1242          */
1243         if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1244                 *res = BULKSTAT_RV_NOTHING;
1245                 return XFS_ERROR(EINVAL);
1246         }
1247
1248         /*
1249          * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1250          * interface expects the inode to be exclusively locked because that's
1251          * the case in all other instances. It's OK that we do this because
1252          * quotacheck is done only at mount time.
1253          */
1254         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1255         if (error) {
1256                 *res = BULKSTAT_RV_NOTHING;
1257                 return error;
1258         }
1259
1260         ASSERT(ip->i_delayed_blks == 0);
1261
1262         if (XFS_IS_REALTIME_INODE(ip)) {
1263                 /*
1264                  * Walk thru the extent list and count the realtime blocks.
1265                  */
1266                 error = xfs_qm_get_rtblks(ip, &rtblks);
1267                 if (error)
1268                         goto error0;
1269         }
1270
1271         nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1272
1273         /*
1274          * Add the (disk blocks and inode) resources occupied by this
1275          * inode to its dquots. We do this adjustment in the incore dquot,
1276          * and also copy the changes to its buffer.
1277          * We don't care about putting these changes in a transaction
1278          * envelope because if we crash in the middle of a 'quotacheck'
1279          * we have to start from the beginning anyway.
1280          * Once we're done, we'll log all the dquot bufs.
1281          *
1282          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1283          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1284          */
1285         if (XFS_IS_UQUOTA_ON(mp)) {
1286                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1287                                                    XFS_DQ_USER, nblks, rtblks);
1288                 if (error)
1289                         goto error0;
1290         }
1291
1292         if (XFS_IS_GQUOTA_ON(mp)) {
1293                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1294                                                    XFS_DQ_GROUP, nblks, rtblks);
1295                 if (error)
1296                         goto error0;
1297         }
1298
1299         if (XFS_IS_PQUOTA_ON(mp)) {
1300                 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1301                                                    XFS_DQ_PROJ, nblks, rtblks);
1302                 if (error)
1303                         goto error0;
1304         }
1305
1306         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1307         IRELE(ip);
1308         *res = BULKSTAT_RV_DIDONE;
1309         return 0;
1310
1311 error0:
1312         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1313         IRELE(ip);
1314         *res = BULKSTAT_RV_GIVEUP;
1315         return error;
1316 }
1317
1318 STATIC int
1319 xfs_qm_flush_one(
1320         struct xfs_dquot        *dqp,
1321         void                    *data)
1322 {
1323         struct list_head        *buffer_list = data;
1324         struct xfs_buf          *bp = NULL;
1325         int                     error = 0;
1326
1327         xfs_dqlock(dqp);
1328         if (dqp->dq_flags & XFS_DQ_FREEING)
1329                 goto out_unlock;
1330         if (!XFS_DQ_IS_DIRTY(dqp))
1331                 goto out_unlock;
1332
1333         xfs_dqflock(dqp);
1334         error = xfs_qm_dqflush(dqp, &bp);
1335         if (error)
1336                 goto out_unlock;
1337
1338         xfs_buf_delwri_queue(bp, buffer_list);
1339         xfs_buf_relse(bp);
1340 out_unlock:
1341         xfs_dqunlock(dqp);
1342         return error;
1343 }
1344
1345 /*
1346  * Walk thru all the filesystem inodes and construct a consistent view
1347  * of the disk quota world. If the quotacheck fails, disable quotas.
1348  */
1349 int
1350 xfs_qm_quotacheck(
1351         xfs_mount_t     *mp)
1352 {
1353         int                     done, count, error, error2;
1354         xfs_ino_t               lastino;
1355         size_t                  structsz;
1356         uint                    flags;
1357         LIST_HEAD               (buffer_list);
1358         struct xfs_inode        *uip = mp->m_quotainfo->qi_uquotaip;
1359         struct xfs_inode        *gip = mp->m_quotainfo->qi_gquotaip;
1360         struct xfs_inode        *pip = mp->m_quotainfo->qi_pquotaip;
1361
1362         count = INT_MAX;
1363         structsz = 1;
1364         lastino = 0;
1365         flags = 0;
1366
1367         ASSERT(uip || gip || pip);
1368         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1369
1370         xfs_notice(mp, "Quotacheck needed: Please wait.");
1371
1372         /*
1373          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1374          * their counters to zero. We need a clean slate.
1375          * We don't log our changes till later.
1376          */
1377         if (uip) {
1378                 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1379                                          &buffer_list);
1380                 if (error)
1381                         goto error_return;
1382                 flags |= XFS_UQUOTA_CHKD;
1383         }
1384
1385         if (gip) {
1386                 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1387                                          &buffer_list);
1388                 if (error)
1389                         goto error_return;
1390                 flags |= XFS_GQUOTA_CHKD;
1391         }
1392
1393         if (pip) {
1394                 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1395                                          &buffer_list);
1396                 if (error)
1397                         goto error_return;
1398                 flags |= XFS_PQUOTA_CHKD;
1399         }
1400
1401         do {
1402                 /*
1403                  * Iterate thru all the inodes in the file system,
1404                  * adjusting the corresponding dquot counters in core.
1405                  */
1406                 error = xfs_bulkstat(mp, &lastino, &count,
1407                                      xfs_qm_dqusage_adjust,
1408                                      structsz, NULL, &done);
1409                 if (error)
1410                         break;
1411
1412         } while (!done);
1413
1414         /*
1415          * We've made all the changes that we need to make incore.  Flush them
1416          * down to disk buffers if everything was updated successfully.
1417          */
1418         if (XFS_IS_UQUOTA_ON(mp)) {
1419                 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1420                                           &buffer_list);
1421         }
1422         if (XFS_IS_GQUOTA_ON(mp)) {
1423                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1424                                            &buffer_list);
1425                 if (!error)
1426                         error = error2;
1427         }
1428         if (XFS_IS_PQUOTA_ON(mp)) {
1429                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1430                                            &buffer_list);
1431                 if (!error)
1432                         error = error2;
1433         }
1434
1435         error2 = xfs_buf_delwri_submit(&buffer_list);
1436         if (!error)
1437                 error = error2;
1438
1439         /*
1440          * We can get this error if we couldn't do a dquot allocation inside
1441          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1442          * dirty dquots that might be cached, we just want to get rid of them
1443          * and turn quotaoff. The dquots won't be attached to any of the inodes
1444          * at this point (because we intentionally didn't in dqget_noattach).
1445          */
1446         if (error) {
1447                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1448                 goto error_return;
1449         }
1450
1451         /*
1452          * If one type of quotas is off, then it will lose its
1453          * quotachecked status, since we won't be doing accounting for
1454          * that type anymore.
1455          */
1456         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1457         mp->m_qflags |= flags;
1458
1459  error_return:
1460         while (!list_empty(&buffer_list)) {
1461                 struct xfs_buf *bp =
1462                         list_first_entry(&buffer_list, struct xfs_buf, b_list);
1463                 list_del_init(&bp->b_list);
1464                 xfs_buf_relse(bp);
1465         }
1466
1467         if (error) {
1468                 xfs_warn(mp,
1469         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1470                         error);
1471                 /*
1472                  * We must turn off quotas.
1473                  */
1474                 ASSERT(mp->m_quotainfo != NULL);
1475                 xfs_qm_destroy_quotainfo(mp);
1476                 if (xfs_mount_reset_sbqflags(mp)) {
1477                         xfs_warn(mp,
1478                                 "Quotacheck: Failed to reset quota flags.");
1479                 }
1480         } else
1481                 xfs_notice(mp, "Quotacheck: Done.");
1482         return (error);
1483 }
1484
1485 /*
1486  * This is called after the superblock has been read in and we're ready to
1487  * iget the quota inodes.
1488  */
1489 STATIC int
1490 xfs_qm_init_quotainos(
1491         xfs_mount_t     *mp)
1492 {
1493         struct xfs_inode        *uip = NULL;
1494         struct xfs_inode        *gip = NULL;
1495         struct xfs_inode        *pip = NULL;
1496         int                     error;
1497         __int64_t               sbflags = 0;
1498         uint                    flags = 0;
1499
1500         ASSERT(mp->m_quotainfo);
1501
1502         /*
1503          * Get the uquota and gquota inodes
1504          */
1505         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1506                 if (XFS_IS_UQUOTA_ON(mp) &&
1507                     mp->m_sb.sb_uquotino != NULLFSINO) {
1508                         ASSERT(mp->m_sb.sb_uquotino > 0);
1509                         error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1510                                              0, 0, &uip);
1511                         if (error)
1512                                 return XFS_ERROR(error);
1513                 }
1514                 if (XFS_IS_GQUOTA_ON(mp) &&
1515                     mp->m_sb.sb_gquotino != NULLFSINO) {
1516                         ASSERT(mp->m_sb.sb_gquotino > 0);
1517                         error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1518                                              0, 0, &gip);
1519                         if (error)
1520                                 goto error_rele;
1521                 }
1522                 if (XFS_IS_PQUOTA_ON(mp) &&
1523                     mp->m_sb.sb_pquotino != NULLFSINO) {
1524                         ASSERT(mp->m_sb.sb_pquotino > 0);
1525                         error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1526                                              0, 0, &pip);
1527                         if (error)
1528                                 goto error_rele;
1529                 }
1530         } else {
1531                 flags |= XFS_QMOPT_SBVERSION;
1532                 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1533                             XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1534                             XFS_SB_QFLAGS);
1535         }
1536
1537         /*
1538          * Create the three inodes, if they don't exist already. The changes
1539          * made above will get added to a transaction and logged in one of
1540          * the qino_alloc calls below.  If the device is readonly,
1541          * temporarily switch to read-write to do this.
1542          */
1543         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1544                 error = xfs_qm_qino_alloc(mp, &uip,
1545                                               sbflags | XFS_SB_UQUOTINO,
1546                                               flags | XFS_QMOPT_UQUOTA);
1547                 if (error)
1548                         goto error_rele;
1549
1550                 flags &= ~XFS_QMOPT_SBVERSION;
1551         }
1552         if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1553                 error = xfs_qm_qino_alloc(mp, &gip,
1554                                           sbflags | XFS_SB_GQUOTINO,
1555                                           flags | XFS_QMOPT_GQUOTA);
1556                 if (error)
1557                         goto error_rele;
1558
1559                 flags &= ~XFS_QMOPT_SBVERSION;
1560         }
1561         if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1562                 error = xfs_qm_qino_alloc(mp, &pip,
1563                                           sbflags | XFS_SB_PQUOTINO,
1564                                           flags | XFS_QMOPT_PQUOTA);
1565                 if (error)
1566                         goto error_rele;
1567         }
1568
1569         mp->m_quotainfo->qi_uquotaip = uip;
1570         mp->m_quotainfo->qi_gquotaip = gip;
1571         mp->m_quotainfo->qi_pquotaip = pip;
1572
1573         return 0;
1574
1575 error_rele:
1576         if (uip)
1577                 IRELE(uip);
1578         if (gip)
1579                 IRELE(gip);
1580         if (pip)
1581                 IRELE(pip);
1582         return XFS_ERROR(error);
1583 }
1584
1585 STATIC void
1586 xfs_qm_dqfree_one(
1587         struct xfs_dquot        *dqp)
1588 {
1589         struct xfs_mount        *mp = dqp->q_mount;
1590         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1591
1592         mutex_lock(&qi->qi_tree_lock);
1593         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1594                           be32_to_cpu(dqp->q_core.d_id));
1595
1596         qi->qi_dquots--;
1597         mutex_unlock(&qi->qi_tree_lock);
1598
1599         xfs_qm_dqdestroy(dqp);
1600 }
1601
1602 STATIC void
1603 xfs_qm_dqreclaim_one(
1604         struct xfs_dquot        *dqp,
1605         struct list_head        *buffer_list,
1606         struct list_head        *dispose_list)
1607 {
1608         struct xfs_mount        *mp = dqp->q_mount;
1609         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1610         int                     error;
1611
1612         if (!xfs_dqlock_nowait(dqp))
1613                 goto out_move_tail;
1614
1615         /*
1616          * This dquot has acquired a reference in the meantime remove it from
1617          * the freelist and try again.
1618          */
1619         if (dqp->q_nrefs) {
1620                 xfs_dqunlock(dqp);
1621
1622                 trace_xfs_dqreclaim_want(dqp);
1623                 XFS_STATS_INC(xs_qm_dqwants);
1624
1625                 list_del_init(&dqp->q_lru);
1626                 qi->qi_lru_count--;
1627                 XFS_STATS_DEC(xs_qm_dquot_unused);
1628                 return;
1629         }
1630
1631         /*
1632          * Try to grab the flush lock. If this dquot is in the process of
1633          * getting flushed to disk, we don't want to reclaim it.
1634          */
1635         if (!xfs_dqflock_nowait(dqp))
1636                 goto out_unlock_move_tail;
1637
1638         if (XFS_DQ_IS_DIRTY(dqp)) {
1639                 struct xfs_buf  *bp = NULL;
1640
1641                 trace_xfs_dqreclaim_dirty(dqp);
1642
1643                 error = xfs_qm_dqflush(dqp, &bp);
1644                 if (error) {
1645                         xfs_warn(mp, "%s: dquot %p flush failed",
1646                                  __func__, dqp);
1647                         goto out_unlock_move_tail;
1648                 }
1649
1650                 xfs_buf_delwri_queue(bp, buffer_list);
1651                 xfs_buf_relse(bp);
1652                 /*
1653                  * Give the dquot another try on the freelist, as the
1654                  * flushing will take some time.
1655                  */
1656                 goto out_unlock_move_tail;
1657         }
1658         xfs_dqfunlock(dqp);
1659
1660         /*
1661          * Prevent lookups now that we are past the point of no return.
1662          */
1663         dqp->dq_flags |= XFS_DQ_FREEING;
1664         xfs_dqunlock(dqp);
1665
1666         ASSERT(dqp->q_nrefs == 0);
1667         list_move_tail(&dqp->q_lru, dispose_list);
1668         qi->qi_lru_count--;
1669         XFS_STATS_DEC(xs_qm_dquot_unused);
1670
1671         trace_xfs_dqreclaim_done(dqp);
1672         XFS_STATS_INC(xs_qm_dqreclaims);
1673         return;
1674
1675         /*
1676          * Move the dquot to the tail of the list so that we don't spin on it.
1677          */
1678 out_unlock_move_tail:
1679         xfs_dqunlock(dqp);
1680 out_move_tail:
1681         list_move_tail(&dqp->q_lru, &qi->qi_lru_list);
1682         trace_xfs_dqreclaim_busy(dqp);
1683         XFS_STATS_INC(xs_qm_dqreclaim_misses);
1684 }
1685
1686 STATIC int
1687 xfs_qm_shake(
1688         struct shrinker         *shrink,
1689         struct shrink_control   *sc)
1690 {
1691         struct xfs_quotainfo    *qi =
1692                 container_of(shrink, struct xfs_quotainfo, qi_shrinker);
1693         int                     nr_to_scan = sc->nr_to_scan;
1694         LIST_HEAD               (buffer_list);
1695         LIST_HEAD               (dispose_list);
1696         struct xfs_dquot        *dqp;
1697         int                     error;
1698
1699         if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
1700                 return 0;
1701         if (!nr_to_scan)
1702                 goto out;
1703
1704         mutex_lock(&qi->qi_lru_lock);
1705         while (!list_empty(&qi->qi_lru_list)) {
1706                 if (nr_to_scan-- <= 0)
1707                         break;
1708                 dqp = list_first_entry(&qi->qi_lru_list, struct xfs_dquot,
1709                                        q_lru);
1710                 xfs_qm_dqreclaim_one(dqp, &buffer_list, &dispose_list);
1711         }
1712         mutex_unlock(&qi->qi_lru_lock);
1713
1714         error = xfs_buf_delwri_submit(&buffer_list);
1715         if (error)
1716                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
1717
1718         while (!list_empty(&dispose_list)) {
1719                 dqp = list_first_entry(&dispose_list, struct xfs_dquot, q_lru);
1720                 list_del_init(&dqp->q_lru);
1721                 xfs_qm_dqfree_one(dqp);
1722         }
1723
1724 out:
1725         return (qi->qi_lru_count / 100) * sysctl_vfs_cache_pressure;
1726 }
1727
1728 /*
1729  * Start a transaction and write the incore superblock changes to
1730  * disk. flags parameter indicates which fields have changed.
1731  */
1732 int
1733 xfs_qm_write_sb_changes(
1734         xfs_mount_t     *mp,
1735         __int64_t       flags)
1736 {
1737         xfs_trans_t     *tp;
1738         int             error;
1739
1740         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1741         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1742         if (error) {
1743                 xfs_trans_cancel(tp, 0);
1744                 return error;
1745         }
1746
1747         xfs_mod_sb(tp, flags);
1748         error = xfs_trans_commit(tp, 0);
1749
1750         return error;
1751 }
1752
1753
1754 /* --------------- utility functions for vnodeops ---------------- */
1755
1756
1757 /*
1758  * Given an inode, a uid, gid and prid make sure that we have
1759  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1760  * quotas by creating this file.
1761  * This also attaches dquot(s) to the given inode after locking it,
1762  * and returns the dquots corresponding to the uid and/or gid.
1763  *
1764  * in   : inode (unlocked)
1765  * out  : udquot, gdquot with references taken and unlocked
1766  */
1767 int
1768 xfs_qm_vop_dqalloc(
1769         struct xfs_inode        *ip,
1770         xfs_dqid_t              uid,
1771         xfs_dqid_t              gid,
1772         prid_t                  prid,
1773         uint                    flags,
1774         struct xfs_dquot        **O_udqpp,
1775         struct xfs_dquot        **O_gdqpp,
1776         struct xfs_dquot        **O_pdqpp)
1777 {
1778         struct xfs_mount        *mp = ip->i_mount;
1779         struct xfs_dquot        *uq = NULL;
1780         struct xfs_dquot        *gq = NULL;
1781         struct xfs_dquot        *pq = NULL;
1782         int                     error;
1783         uint                    lockflags;
1784
1785         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1786                 return 0;
1787
1788         lockflags = XFS_ILOCK_EXCL;
1789         xfs_ilock(ip, lockflags);
1790
1791         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1792                 gid = ip->i_d.di_gid;
1793
1794         /*
1795          * Attach the dquot(s) to this inode, doing a dquot allocation
1796          * if necessary. The dquot(s) will not be locked.
1797          */
1798         if (XFS_NOT_DQATTACHED(mp, ip)) {
1799                 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1800                 if (error) {
1801                         xfs_iunlock(ip, lockflags);
1802                         return error;
1803                 }
1804         }
1805
1806         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1807                 if (ip->i_d.di_uid != uid) {
1808                         /*
1809                          * What we need is the dquot that has this uid, and
1810                          * if we send the inode to dqget, the uid of the inode
1811                          * takes priority over what's sent in the uid argument.
1812                          * We must unlock inode here before calling dqget if
1813                          * we're not sending the inode, because otherwise
1814                          * we'll deadlock by doing trans_reserve while
1815                          * holding ilock.
1816                          */
1817                         xfs_iunlock(ip, lockflags);
1818                         error = xfs_qm_dqget(mp, NULL, uid,
1819                                                  XFS_DQ_USER,
1820                                                  XFS_QMOPT_DQALLOC |
1821                                                  XFS_QMOPT_DOWARN,
1822                                                  &uq);
1823                         if (error) {
1824                                 ASSERT(error != ENOENT);
1825                                 return error;
1826                         }
1827                         /*
1828                          * Get the ilock in the right order.
1829                          */
1830                         xfs_dqunlock(uq);
1831                         lockflags = XFS_ILOCK_SHARED;
1832                         xfs_ilock(ip, lockflags);
1833                 } else {
1834                         /*
1835                          * Take an extra reference, because we'll return
1836                          * this to caller
1837                          */
1838                         ASSERT(ip->i_udquot);
1839                         uq = xfs_qm_dqhold(ip->i_udquot);
1840                 }
1841         }
1842         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1843                 if (ip->i_d.di_gid != gid) {
1844                         xfs_iunlock(ip, lockflags);
1845                         error = xfs_qm_dqget(mp, NULL, gid,
1846                                                  XFS_DQ_GROUP,
1847                                                  XFS_QMOPT_DQALLOC |
1848                                                  XFS_QMOPT_DOWARN,
1849                                                  &gq);
1850                         if (error) {
1851                                 ASSERT(error != ENOENT);
1852                                 goto error_rele;
1853                         }
1854                         xfs_dqunlock(gq);
1855                         lockflags = XFS_ILOCK_SHARED;
1856                         xfs_ilock(ip, lockflags);
1857                 } else {
1858                         ASSERT(ip->i_gdquot);
1859                         gq = xfs_qm_dqhold(ip->i_gdquot);
1860                 }
1861         }
1862         if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1863                 if (xfs_get_projid(ip) != prid) {
1864                         xfs_iunlock(ip, lockflags);
1865                         error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1866                                                  XFS_DQ_PROJ,
1867                                                  XFS_QMOPT_DQALLOC |
1868                                                  XFS_QMOPT_DOWARN,
1869                                                  &pq);
1870                         if (error) {
1871                                 ASSERT(error != ENOENT);
1872                                 goto error_rele;
1873                         }
1874                         xfs_dqunlock(pq);
1875                         lockflags = XFS_ILOCK_SHARED;
1876                         xfs_ilock(ip, lockflags);
1877                 } else {
1878                         ASSERT(ip->i_pdquot);
1879                         pq = xfs_qm_dqhold(ip->i_pdquot);
1880                 }
1881         }
1882         if (uq)
1883                 trace_xfs_dquot_dqalloc(ip);
1884
1885         xfs_iunlock(ip, lockflags);
1886         if (O_udqpp)
1887                 *O_udqpp = uq;
1888         else if (uq)
1889                 xfs_qm_dqrele(uq);
1890         if (O_gdqpp)
1891                 *O_gdqpp = gq;
1892         else if (gq)
1893                 xfs_qm_dqrele(gq);
1894         if (O_pdqpp)
1895                 *O_pdqpp = pq;
1896         else if (pq)
1897                 xfs_qm_dqrele(pq);
1898         return 0;
1899
1900 error_rele:
1901         if (gq)
1902                 xfs_qm_dqrele(gq);
1903         if (uq)
1904                 xfs_qm_dqrele(uq);
1905         return error;
1906 }
1907
1908 /*
1909  * Actually transfer ownership, and do dquot modifications.
1910  * These were already reserved.
1911  */
1912 xfs_dquot_t *
1913 xfs_qm_vop_chown(
1914         xfs_trans_t     *tp,
1915         xfs_inode_t     *ip,
1916         xfs_dquot_t     **IO_olddq,
1917         xfs_dquot_t     *newdq)
1918 {
1919         xfs_dquot_t     *prevdq;
1920         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1921                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1922
1923
1924         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1925         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1926
1927         /* old dquot */
1928         prevdq = *IO_olddq;
1929         ASSERT(prevdq);
1930         ASSERT(prevdq != newdq);
1931
1932         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1933         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1934
1935         /* the sparkling new dquot */
1936         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1937         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1938
1939         /*
1940          * Take an extra reference, because the inode is going to keep
1941          * this dquot pointer even after the trans_commit.
1942          */
1943         *IO_olddq = xfs_qm_dqhold(newdq);
1944
1945         return prevdq;
1946 }
1947
1948 /*
1949  * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1950  */
1951 int
1952 xfs_qm_vop_chown_reserve(
1953         struct xfs_trans        *tp,
1954         struct xfs_inode        *ip,
1955         struct xfs_dquot        *udqp,
1956         struct xfs_dquot        *gdqp,
1957         struct xfs_dquot        *pdqp,
1958         uint                    flags)
1959 {
1960         struct xfs_mount        *mp = ip->i_mount;
1961         uint                    delblks, blkflags, prjflags = 0;
1962         struct xfs_dquot        *udq_unres = NULL;
1963         struct xfs_dquot        *gdq_unres = NULL;
1964         struct xfs_dquot        *pdq_unres = NULL;
1965         struct xfs_dquot        *udq_delblks = NULL;
1966         struct xfs_dquot        *gdq_delblks = NULL;
1967         struct xfs_dquot        *pdq_delblks = NULL;
1968         int                     error;
1969
1970
1971         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1972         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1973
1974         delblks = ip->i_delayed_blks;
1975         blkflags = XFS_IS_REALTIME_INODE(ip) ?
1976                         XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1977
1978         if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1979             ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1980                 udq_delblks = udqp;
1981                 /*
1982                  * If there are delayed allocation blocks, then we have to
1983                  * unreserve those from the old dquot, and add them to the
1984                  * new dquot.
1985                  */
1986                 if (delblks) {
1987                         ASSERT(ip->i_udquot);
1988                         udq_unres = ip->i_udquot;
1989                 }
1990         }
1991         if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1992             ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1993                 gdq_delblks = gdqp;
1994                 if (delblks) {
1995                         ASSERT(ip->i_gdquot);
1996                         gdq_unres = ip->i_gdquot;
1997                 }
1998         }
1999
2000         if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2001             xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2002                 prjflags = XFS_QMOPT_ENOSPC;
2003                 pdq_delblks = pdqp;
2004                 if (delblks) {
2005                         ASSERT(ip->i_pdquot);
2006                         pdq_unres = ip->i_pdquot;
2007                 }
2008         }
2009
2010         error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2011                                 udq_delblks, gdq_delblks, pdq_delblks,
2012                                 ip->i_d.di_nblocks, 1,
2013                                 flags | blkflags | prjflags);
2014         if (error)
2015                 return error;
2016
2017         /*
2018          * Do the delayed blks reservations/unreservations now. Since, these
2019          * are done without the help of a transaction, if a reservation fails
2020          * its previous reservations won't be automatically undone by trans
2021          * code. So, we have to do it manually here.
2022          */
2023         if (delblks) {
2024                 /*
2025                  * Do the reservations first. Unreservation can't fail.
2026                  */
2027                 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2028                 ASSERT(udq_unres || gdq_unres || pdq_unres);
2029                 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2030                             udq_delblks, gdq_delblks, pdq_delblks,
2031                             (xfs_qcnt_t)delblks, 0,
2032                             flags | blkflags | prjflags);
2033                 if (error)
2034                         return error;
2035                 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2036                                 udq_unres, gdq_unres, pdq_unres,
2037                                 -((xfs_qcnt_t)delblks), 0, blkflags);
2038         }
2039
2040         return (0);
2041 }
2042
2043 int
2044 xfs_qm_vop_rename_dqattach(
2045         struct xfs_inode        **i_tab)
2046 {
2047         struct xfs_mount        *mp = i_tab[0]->i_mount;
2048         int                     i;
2049
2050         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2051                 return 0;
2052
2053         for (i = 0; (i < 4 && i_tab[i]); i++) {
2054                 struct xfs_inode        *ip = i_tab[i];
2055                 int                     error;
2056
2057                 /*
2058                  * Watch out for duplicate entries in the table.
2059                  */
2060                 if (i == 0 || ip != i_tab[i-1]) {
2061                         if (XFS_NOT_DQATTACHED(mp, ip)) {
2062                                 error = xfs_qm_dqattach(ip, 0);
2063                                 if (error)
2064                                         return error;
2065                         }
2066                 }
2067         }
2068         return 0;
2069 }
2070
2071 void
2072 xfs_qm_vop_create_dqattach(
2073         struct xfs_trans        *tp,
2074         struct xfs_inode        *ip,
2075         struct xfs_dquot        *udqp,
2076         struct xfs_dquot        *gdqp,
2077         struct xfs_dquot        *pdqp)
2078 {
2079         struct xfs_mount        *mp = tp->t_mountp;
2080
2081         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2082                 return;
2083
2084         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2085         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2086
2087         if (udqp) {
2088                 ASSERT(ip->i_udquot == NULL);
2089                 ASSERT(XFS_IS_UQUOTA_ON(mp));
2090                 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2091
2092                 ip->i_udquot = xfs_qm_dqhold(udqp);
2093                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2094         }
2095         if (gdqp) {
2096                 ASSERT(ip->i_gdquot == NULL);
2097                 ASSERT(XFS_IS_GQUOTA_ON(mp));
2098                 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
2099                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
2100                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2101         }
2102         if (pdqp) {
2103                 ASSERT(ip->i_pdquot == NULL);
2104                 ASSERT(XFS_IS_PQUOTA_ON(mp));
2105                 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2106
2107                 ip->i_pdquot = xfs_qm_dqhold(pdqp);
2108                 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2109         }
2110 }
2111