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
  55
  56 STATIC void     xfs_qm_dqfree_one(struct xfs_dquot *dqp);
  57 /*
  58  * We use the batch lookup interface to iterate over the dquots as it
  59  * currently is the only interface into the radix tree code that allows
  60  * fuzzy lookups instead of exact matches.  Holding the lock over multiple
  61  * operations is fine as all callers are used either during mount/umount
  62  * or quotaoff.
  63  */
  64 #define XFS_DQ_LOOKUP_BATCH     32
  65
  66 STATIC int
  67 xfs_qm_dquot_walk(
  68         struct xfs_mount        *mp,
  69         int                     type,
  70         int                     (*execute)(struct xfs_dquot *dqp, void *data),
  71         void                    *data)
  72 {
  73         struct xfs_quotainfo    *qi = mp->m_quotainfo;
  74         struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
  75         uint32_t                next_index;
  76         int                     last_error = 0;
  77         int                     skipped;
  78         int                     nr_found;
  79
  80 restart:
  81         skipped = 0;
  82         next_index = 0;
  83         nr_found = 0;
  84
  85         while (1) {
  86                 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
  87                 int             error = 0;
  88                 int             i;
  89
  90                 mutex_lock(&qi->qi_tree_lock);
  91                 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
  92                                         next_index, XFS_DQ_LOOKUP_BATCH);
  93                 if (!nr_found) {
  94                         mutex_unlock(&qi->qi_tree_lock);
  95                         break;
  96                 }
  97
  98                 for (i = 0; i < nr_found; i++) {
  99                         struct xfs_dquot *dqp = batch[i];
 100
 101                         next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
 102
 103                         error = execute(batch[i], data);
 104                         if (error == EAGAIN) {
 105                                 skipped++;
 106                                 continue;
 107                         }
 108                         if (error && last_error != EFSCORRUPTED)
 109                                 last_error = error;
 110                 }
 111
 112                 mutex_unlock(&qi->qi_tree_lock);
 113
 114                 /* bail out if the filesystem is corrupted.  */
 115                 if (last_error == EFSCORRUPTED) {
 116                         skipped = 0;
 117                         break;
 118                 }
 119         }
 120
 121         if (skipped) {
 122                 delay(1);
 123                 goto restart;
 124         }
 125
 126         return last_error;
 127 }
 128
 129
 130 /*
 131  * Purge a dquot from all tracking data structures and free it.
 132  */
 133 STATIC int
 134 xfs_qm_dqpurge(
 135         struct xfs_dquot        *dqp,
 136         void                    *data)
 137 {
 138         struct xfs_mount        *mp = dqp->q_mount;
 139         struct xfs_quotainfo    *qi = mp->m_quotainfo;
 140         struct xfs_dquot        *gdqp = NULL;
 141         struct xfs_dquot        *pdqp = NULL;
 142
 143         xfs_dqlock(dqp);
 144         if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 145                 xfs_dqunlock(dqp);
 146                 return EAGAIN;
 147         }
 148
 149         /*
 150          * If this quota has a hint attached, prepare for releasing it now.
 151          */
 152         gdqp = dqp->q_gdquot;
 153         if (gdqp) {
 154                 xfs_dqlock(gdqp);
 155                 dqp->q_gdquot = NULL;
 156         }
 157
 158         pdqp = dqp->q_pdquot;
 159         if (pdqp) {
 160                 xfs_dqlock(pdqp);
 161                 dqp->q_pdquot = NULL;
 162         }
 163
 164         dqp->dq_flags |= XFS_DQ_FREEING;
 165
 166         xfs_dqflock(dqp);
 167
 168         /*
 169          * If we are turning this type of quotas off, we don't care
 170          * about the dirty metadata sitting in this dquot. OTOH, if
 171          * we're unmounting, we do care, so we flush it and wait.
 172          */
 173         if (XFS_DQ_IS_DIRTY(dqp)) {
 174                 struct xfs_buf  *bp = NULL;
 175                 int             error;
 176
 177                 /*
 178                  * We don't care about getting disk errors here. We need
 179                  * to purge this dquot anyway, so we go ahead regardless.
 180                  */
 181                 error = xfs_qm_dqflush(dqp, &bp);
 182                 if (error) {
 183                         xfs_warn(mp, "%s: dquot %p flush failed",
 184                                 __func__, dqp);
 185                 } else {
 186                         error = xfs_bwrite(bp);
 187                         xfs_buf_relse(bp);
 188                 }
 189                 xfs_dqflock(dqp);
 190         }
 191
 192         ASSERT(atomic_read(&dqp->q_pincount) == 0);
 193         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 194                !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 195
 196         xfs_dqfunlock(dqp);
 197         xfs_dqunlock(dqp);
 198
 199         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
 200                           be32_to_cpu(dqp->q_core.d_id));
 201         qi->qi_dquots--;
 202
 203         /*
 204          * We move dquots to the freelist as soon as their reference count
 205          * hits zero, so it really should be on the freelist here.
 206          */
 207         ASSERT(!list_empty(&dqp->q_lru));
 208         list_lru_del(&qi->qi_lru, &dqp->q_lru);
 209         XFS_STATS_DEC(xs_qm_dquot_unused);
 210
 211         xfs_qm_dqdestroy(dqp);
 212
 213         if (gdqp)
 214                 xfs_qm_dqput(gdqp);
 215         if (pdqp)
 216                 xfs_qm_dqput(pdqp);
 217         return 0;
 218 }
 219
 220 /*
 221  * Purge the dquot cache.
 222  */
 223 void
 224 xfs_qm_dqpurge_all(
 225         struct xfs_mount        *mp,
 226         uint                    flags)
 227 {
 228         if (flags & XFS_QMOPT_UQUOTA)
 229                 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
 230         if (flags & XFS_QMOPT_GQUOTA)
 231                 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 232         if (flags & XFS_QMOPT_PQUOTA)
 233                 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 234 }
 235
 236 /*
 237  * Just destroy the quotainfo structure.
 238  */
 239 void
 240 xfs_qm_unmount(
 241         struct xfs_mount        *mp)
 242 {
 243         if (mp->m_quotainfo) {
 244                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 245                 xfs_qm_destroy_quotainfo(mp);
 246         }
 247 }
 248
 249
 250 /*
 251  * This is called from xfs_mountfs to start quotas and initialize all
 252  * necessary data structures like quotainfo.  This is also responsible for
 253  * running a quotacheck as necessary.  We are guaranteed that the superblock
 254  * is consistently read in at this point.
 255  *
 256  * If we fail here, the mount will continue with quota turned off. We don't
 257  * need to inidicate success or failure at all.
 258  */
 259 void
 260 xfs_qm_mount_quotas(
 261         xfs_mount_t     *mp)
 262 {
 263         int             error = 0;
 264         uint            sbf;
 265
 266         /*
 267          * If quotas on realtime volumes is not supported, we disable
 268          * quotas immediately.
 269          */
 270         if (mp->m_sb.sb_rextents) {
 271                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 272                 mp->m_qflags = 0;
 273                 goto write_changes;
 274         }
 275
 276         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 277
 278         /*
 279          * Allocate the quotainfo structure inside the mount struct, and
 280          * create quotainode(s), and change/rev superblock if necessary.
 281          */
 282         error = xfs_qm_init_quotainfo(mp);
 283         if (error) {
 284                 /*
 285                  * We must turn off quotas.
 286                  */
 287                 ASSERT(mp->m_quotainfo == NULL);
 288                 mp->m_qflags = 0;
 289                 goto write_changes;
 290         }
 291         /*
 292          * If any of the quotas are not consistent, do a quotacheck.
 293          */
 294         if (XFS_QM_NEED_QUOTACHECK(mp)) {
 295                 error = xfs_qm_quotacheck(mp);
 296                 if (error) {
 297                         /* Quotacheck failed and disabled quotas. */
 298                         return;
 299                 }
 300         }
 301         /*
 302          * If one type of quotas is off, then it will lose its
 303          * quotachecked status, since we won't be doing accounting for
 304          * that type anymore.
 305          */
 306         if (!XFS_IS_UQUOTA_ON(mp))
 307                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 308         if (!XFS_IS_GQUOTA_ON(mp))
 309                 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
 310         if (!XFS_IS_PQUOTA_ON(mp))
 311                 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
 312
 313  write_changes:
 314         /*
 315          * We actually don't have to acquire the m_sb_lock at all.
 316          * This can only be called from mount, and that's single threaded. XXX
 317          */
 318         spin_lock(&mp->m_sb_lock);
 319         sbf = mp->m_sb.sb_qflags;
 320         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 321         spin_unlock(&mp->m_sb_lock);
 322
 323         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 324                 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 325                         /*
 326                          * We could only have been turning quotas off.
 327                          * We aren't in very good shape actually because
 328                          * the incore structures are convinced that quotas are
 329                          * off, but the on disk superblock doesn't know that !
 330                          */
 331                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 332                         xfs_alert(mp, "%s: Superblock update failed!",
 333                                 __func__);
 334                 }
 335         }
 336
 337         if (error) {
 338                 xfs_warn(mp, "Failed to initialize disk quotas.");
 339                 return;
 340         }
 341 }
 342
 343 /*
 344  * Called from the vfsops layer.
 345  */
 346 void
 347 xfs_qm_unmount_quotas(
 348         xfs_mount_t     *mp)
 349 {
 350         /*
 351          * Release the dquots that root inode, et al might be holding,
 352          * before we flush quotas and blow away the quotainfo structure.
 353          */
 354         ASSERT(mp->m_rootip);
 355         xfs_qm_dqdetach(mp->m_rootip);
 356         if (mp->m_rbmip)
 357                 xfs_qm_dqdetach(mp->m_rbmip);
 358         if (mp->m_rsumip)
 359                 xfs_qm_dqdetach(mp->m_rsumip);
 360
 361         /*
 362          * Release the quota inodes.
 363          */
 364         if (mp->m_quotainfo) {
 365                 if (mp->m_quotainfo->qi_uquotaip) {
 366                         IRELE(mp->m_quotainfo->qi_uquotaip);
 367                         mp->m_quotainfo->qi_uquotaip = NULL;
 368                 }
 369                 if (mp->m_quotainfo->qi_gquotaip) {
 370                         IRELE(mp->m_quotainfo->qi_gquotaip);
 371                         mp->m_quotainfo->qi_gquotaip = NULL;
 372                 }
 373                 if (mp->m_quotainfo->qi_pquotaip) {
 374                         IRELE(mp->m_quotainfo->qi_pquotaip);
 375                         mp->m_quotainfo->qi_pquotaip = NULL;
 376                 }
 377         }
 378 }
 379
 380 STATIC int
 381 xfs_qm_dqattach_one(
 382         xfs_inode_t     *ip,
 383         xfs_dqid_t      id,
 384         uint            type,
 385         uint            doalloc,
 386         xfs_dquot_t     *udqhint, /* hint */
 387         xfs_dquot_t     **IO_idqpp)
 388 {
 389         xfs_dquot_t     *dqp;
 390         int             error;
 391
 392         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 393         error = 0;
 394
 395         /*
 396          * See if we already have it in the inode itself. IO_idqpp is
 397          * &i_udquot or &i_gdquot. This made the code look weird, but
 398          * made the logic a lot simpler.
 399          */
 400         dqp = *IO_idqpp;
 401         if (dqp) {
 402                 trace_xfs_dqattach_found(dqp);
 403                 return 0;
 404         }
 405
 406         /*
 407          * udqhint is the i_udquot field in inode, and is non-NULL only
 408          * when the type arg is group/project. Its purpose is to save a
 409          * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 410          * the user dquot.
 411          */
 412         if (udqhint) {
 413                 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 414                 xfs_dqlock(udqhint);
 415
 416                 /*
 417                  * No need to take dqlock to look at the id.
 418                  *
 419                  * The ID can't change until it gets reclaimed, and it won't
 420                  * be reclaimed as long as we have a ref from inode and we
 421                  * hold the ilock.
 422                  */
 423                 if (type == XFS_DQ_GROUP)
 424                         dqp = udqhint->q_gdquot;
 425                 else
 426                         dqp = udqhint->q_pdquot;
 427                 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 428                         ASSERT(*IO_idqpp == NULL);
 429
 430                         *IO_idqpp = xfs_qm_dqhold(dqp);
 431                         xfs_dqunlock(udqhint);
 432                         return 0;
 433                 }
 434
 435                 /*
 436                  * We can't hold a dquot lock when we call the dqget code.
 437                  * We'll deadlock in no time, because of (not conforming to)
 438                  * lock ordering - the inodelock comes before any dquot lock,
 439                  * and we may drop and reacquire the ilock in xfs_qm_dqget().
 440                  */
 441                 xfs_dqunlock(udqhint);
 442         }
 443
 444         /*
 445          * Find the dquot from somewhere. This bumps the
 446          * reference count of dquot and returns it locked.
 447          * This can return ENOENT if dquot didn't exist on
 448          * disk and we didn't ask it to allocate;
 449          * ESRCH if quotas got turned off suddenly.
 450          */
 451         error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 452                              doalloc | XFS_QMOPT_DOWARN, &dqp);
 453         if (error)
 454                 return error;
 455
 456         trace_xfs_dqattach_get(dqp);
 457
 458         /*
 459          * dqget may have dropped and re-acquired the ilock, but it guarantees
 460          * that the dquot returned is the one that should go in the inode.
 461          */
 462         *IO_idqpp = dqp;
 463         xfs_dqunlock(dqp);
 464         return 0;
 465 }
 466
 467
 468 /*
 469  * Given a udquot and group/project type, attach the group/project
 470  * dquot pointer to the udquot as a hint for future lookups.
 471  */
 472 STATIC void
 473 xfs_qm_dqattach_hint(
 474         struct xfs_inode        *ip,
 475         int                     type)
 476 {
 477         struct xfs_dquot **dqhintp;
 478         struct xfs_dquot *dqp;
 479         struct xfs_dquot *udq = ip->i_udquot;
 480
 481         ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 482
 483         xfs_dqlock(udq);
 484
 485         if (type == XFS_DQ_GROUP) {
 486                 dqp = ip->i_gdquot;
 487                 dqhintp = &udq->q_gdquot;
 488         } else {
 489                 dqp = ip->i_pdquot;
 490                 dqhintp = &udq->q_pdquot;
 491         }
 492
 493         if (*dqhintp) {
 494                 struct xfs_dquot *tmp;
 495
 496                 if (*dqhintp == dqp)
 497                         goto done;
 498
 499                 tmp = *dqhintp;
 500                 *dqhintp = NULL;
 501                 xfs_qm_dqrele(tmp);
 502         }
 503
 504         *dqhintp = xfs_qm_dqhold(dqp);
 505 done:
 506         xfs_dqunlock(udq);
 507 }
 508
 509 static bool
 510 xfs_qm_need_dqattach(
 511         struct xfs_inode        *ip)
 512 {
 513         struct xfs_mount        *mp = ip->i_mount;
 514
 515         if (!XFS_IS_QUOTA_RUNNING(mp))
 516                 return false;
 517         if (!XFS_IS_QUOTA_ON(mp))
 518                 return false;
 519         if (!XFS_NOT_DQATTACHED(mp, ip))
 520                 return false;
 521         if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
 522                 return false;
 523         return true;
 524 }
 525
 526 /*
 527  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 528  * into account.
 529  * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 530  * Inode may get unlocked and relocked in here, and the caller must deal with
 531  * the consequences.
 532  */
 533 int
 534 xfs_qm_dqattach_locked(
 535         xfs_inode_t     *ip,
 536         uint            flags)
 537 {
 538         xfs_mount_t     *mp = ip->i_mount;
 539         uint            nquotas = 0;
 540         int             error = 0;
 541
 542         if (!xfs_qm_need_dqattach(ip))
 543                 return 0;
 544
 545         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 546
 547         if (XFS_IS_UQUOTA_ON(mp)) {
 548                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 549                                                 flags & XFS_QMOPT_DQALLOC,
 550                                                 NULL, &ip->i_udquot);
 551                 if (error)
 552                         goto done;
 553                 nquotas++;
 554         }
 555
 556         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 557         if (XFS_IS_GQUOTA_ON(mp)) {
 558                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 559                                                 flags & XFS_QMOPT_DQALLOC,
 560                                                 ip->i_udquot, &ip->i_gdquot);
 561                 /*
 562                  * Don't worry about the udquot that we may have
 563                  * attached above. It'll get detached, if not already.
 564                  */
 565                 if (error)
 566                         goto done;
 567                 nquotas++;
 568         }
 569
 570         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 571         if (XFS_IS_PQUOTA_ON(mp)) {
 572                 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 573                                                 flags & XFS_QMOPT_DQALLOC,
 574                                                 ip->i_udquot, &ip->i_pdquot);
 575                 /*
 576                  * Don't worry about the udquot that we may have
 577                  * attached above. It'll get detached, if not already.
 578                  */
 579                 if (error)
 580                         goto done;
 581                 nquotas++;
 582         }
 583
 584         /*
 585          * Attach this group/project quota to the user quota as a hint.
 586          * This WON'T, in general, result in a thrash.
 587          */
 588         if (nquotas > 1 && ip->i_udquot) {
 589                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 590                 ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
 591                 ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
 592
 593                 /*
 594                  * We do not have i_udquot locked at this point, but this check
 595                  * is OK since we don't depend on the i_gdquot to be accurate
 596                  * 100% all the time. It is just a hint, and this will
 597                  * succeed in general.
 598                  */
 599                 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 600                         xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
 601
 602                 if (ip->i_udquot->q_pdquot != ip->i_pdquot)
 603                         xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
 604         }
 605
 606  done:
 607 #ifdef DEBUG
 608         if (!error) {
 609                 if (XFS_IS_UQUOTA_ON(mp))
 610                         ASSERT(ip->i_udquot);
 611                 if (XFS_IS_GQUOTA_ON(mp))
 612                         ASSERT(ip->i_gdquot);
 613                 if (XFS_IS_PQUOTA_ON(mp))
 614                         ASSERT(ip->i_pdquot);
 615         }
 616         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 617 #endif
 618         return error;
 619 }
 620
 621 int
 622 xfs_qm_dqattach(
 623         struct xfs_inode        *ip,
 624         uint                    flags)
 625 {
 626         int                     error;
 627
 628         if (!xfs_qm_need_dqattach(ip))
 629                 return 0;
 630
 631         xfs_ilock(ip, XFS_ILOCK_EXCL);
 632         error = xfs_qm_dqattach_locked(ip, flags);
 633         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 634
 635         return error;
 636 }
 637
 638 /*
 639  * Release dquots (and their references) if any.
 640  * The inode should be locked EXCL except when this's called by
 641  * xfs_ireclaim.
 642  */
 643 void
 644 xfs_qm_dqdetach(
 645         xfs_inode_t     *ip)
 646 {
 647         if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
 648                 return;
 649
 650         trace_xfs_dquot_dqdetach(ip);
 651
 652         ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
 653         if (ip->i_udquot) {
 654                 xfs_qm_dqrele(ip->i_udquot);
 655                 ip->i_udquot = NULL;
 656         }
 657         if (ip->i_gdquot) {
 658                 xfs_qm_dqrele(ip->i_gdquot);
 659                 ip->i_gdquot = NULL;
 660         }
 661         if (ip->i_pdquot) {
 662                 xfs_qm_dqrele(ip->i_pdquot);
 663                 ip->i_pdquot = NULL;
 664         }
 665 }
 666
 667 int
 668 xfs_qm_calc_dquots_per_chunk(
 669         struct xfs_mount        *mp,
 670         unsigned int            nbblks) /* basic block units */
 671 {
 672         unsigned int    ndquots;
 673
 674         ASSERT(nbblks > 0);
 675         ndquots = BBTOB(nbblks);
 676         do_div(ndquots, sizeof(xfs_dqblk_t));
 677
 678         return ndquots;
 679 }
 680
 681 struct xfs_qm_isolate {
 682         struct list_head        buffers;
 683         struct list_head        dispose;
 684 };
 685
 686 static enum lru_status
 687 xfs_qm_dquot_isolate(
 688         struct list_head        *item,
 689         spinlock_t              *lru_lock,
 690         void                    *arg)
 691 {
 692         struct xfs_dquot        *dqp = container_of(item,
 693                                                 struct xfs_dquot, q_lru);
 694         struct xfs_qm_isolate   *isol = arg;
 695
 696         if (!xfs_dqlock_nowait(dqp))
 697                 goto out_miss_busy;
 698
 699         /*
 700          * This dquot has acquired a reference in the meantime remove it from
 701          * the freelist and try again.
 702          */
 703         if (dqp->q_nrefs) {
 704                 xfs_dqunlock(dqp);
 705                 XFS_STATS_INC(xs_qm_dqwants);
 706
 707                 trace_xfs_dqreclaim_want(dqp);
 708                 list_del_init(&dqp->q_lru);
 709                 XFS_STATS_DEC(xs_qm_dquot_unused);
 710                 return 0;
 711         }
 712
 713         /*
 714          * If the dquot is dirty, flush it. If it's already being flushed, just
 715          * skip it so there is time for the IO to complete before we try to
 716          * reclaim it again on the next LRU pass.
 717          */
 718         if (!xfs_dqflock_nowait(dqp)) {
 719                 xfs_dqunlock(dqp);
 720                 goto out_miss_busy;
 721         }
 722
 723         if (XFS_DQ_IS_DIRTY(dqp)) {
 724                 struct xfs_buf  *bp = NULL;
 725                 int             error;
 726
 727                 trace_xfs_dqreclaim_dirty(dqp);
 728
 729                 /* we have to drop the LRU lock to flush the dquot */
 730                 spin_unlock(lru_lock);
 731
 732                 error = xfs_qm_dqflush(dqp, &bp);
 733                 if (error) {
 734                         xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
 735                                  __func__, dqp);
 736                         goto out_unlock_dirty;
 737                 }
 738
 739                 xfs_buf_delwri_queue(bp, &isol->buffers);
 740                 xfs_buf_relse(bp);
 741                 goto out_unlock_dirty;
 742         }
 743         xfs_dqfunlock(dqp);
 744
 745         /*
 746          * Prevent lookups now that we are past the point of no return.
 747          */
 748         dqp->dq_flags |= XFS_DQ_FREEING;
 749         xfs_dqunlock(dqp);
 750
 751         ASSERT(dqp->q_nrefs == 0);
 752         list_move_tail(&dqp->q_lru, &isol->dispose);
 753         XFS_STATS_DEC(xs_qm_dquot_unused);
 754         trace_xfs_dqreclaim_done(dqp);
 755         XFS_STATS_INC(xs_qm_dqreclaims);
 756         return 0;
 757
 758 out_miss_busy:
 759         trace_xfs_dqreclaim_busy(dqp);
 760         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 761         return 2;
 762
 763 out_unlock_dirty:
 764         trace_xfs_dqreclaim_busy(dqp);
 765         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 766         return 3;
 767 }
 768
 769 static long
 770 xfs_qm_shrink_scan(
 771         struct shrinker         *shrink,
 772         struct shrink_control   *sc)
 773 {
 774         struct xfs_quotainfo    *qi = container_of(shrink,
 775                                         struct xfs_quotainfo, qi_shrinker);
 776         struct xfs_qm_isolate   isol;
 777         long                    freed;
 778         int                     error;
 779         unsigned long           nr_to_scan = sc->nr_to_scan;
 780
 781         if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
 782                 return 0;
 783
 784         INIT_LIST_HEAD(&isol.buffers);
 785         INIT_LIST_HEAD(&isol.dispose);
 786
 787         freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
 788                                         &nr_to_scan);
 789
 790         error = xfs_buf_delwri_submit(&isol.buffers);
 791         if (error)
 792                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
 793
 794         while (!list_empty(&isol.dispose)) {
 795                 struct xfs_dquot        *dqp;
 796
 797                 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
 798                 list_del_init(&dqp->q_lru);
 799                 xfs_qm_dqfree_one(dqp);
 800         }
 801
 802         return freed;
 803 }
 804
 805 static long
 806 xfs_qm_shrink_count(
 807         struct shrinker         *shrink,
 808         struct shrink_control   *sc)
 809 {
 810         struct xfs_quotainfo    *qi = container_of(shrink,
 811                                         struct xfs_quotainfo, qi_shrinker);
 812
 813         return list_lru_count_node(&qi->qi_lru, sc->nid);
 814 }
 815
 816 /*
 817  * This initializes all the quota information that's kept in the
 818  * mount structure
 819  */
 820 STATIC int
 821 xfs_qm_init_quotainfo(
 822         xfs_mount_t     *mp)
 823 {
 824         xfs_quotainfo_t *qinf;
 825         int             error;
 826         xfs_dquot_t     *dqp;
 827
 828         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 829
 830         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 831
 832         /*
 833          * See if quotainodes are setup, and if not, allocate them,
 834          * and change the superblock accordingly.
 835          */
 836         if ((error = xfs_qm_init_quotainos(mp))) {
 837                 kmem_free(qinf);
 838                 mp->m_quotainfo = NULL;
 839                 return error;
 840         }
 841
 842         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 843         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 844         INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
 845         mutex_init(&qinf->qi_tree_lock);
 846
 847         list_lru_init(&qinf->qi_lru);
 848
 849         /* mutex used to serialize quotaoffs */
 850         mutex_init(&qinf->qi_quotaofflock);
 851
 852         /* Precalc some constants */
 853         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 854         qinf->qi_dqperchunk = xfs_qm_calc_dquots_per_chunk(mp,
 855                                                         qinf->qi_dqchunklen);
 856
 857         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 858
 859         /*
 860          * We try to get the limits from the superuser's limits fields.
 861          * This is quite hacky, but it is standard quota practice.
 862          *
 863          * We look at the USR dquot with id == 0 first, but if user quotas
 864          * are not enabled we goto the GRP dquot with id == 0.
 865          * We don't really care to keep separate default limits for user
 866          * and group quotas, at least not at this point.
 867          *
 868          * Since we may not have done a quotacheck by this point, just read
 869          * the dquot without attaching it to any hashtables or lists.
 870          */
 871         error = xfs_qm_dqread(mp, 0,
 872                         XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 873                          (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 874                           XFS_DQ_PROJ),
 875                         XFS_QMOPT_DOWARN, &dqp);
 876         if (!error) {
 877                 xfs_disk_dquot_t        *ddqp = &dqp->q_core;
 878
 879                 /*
 880                  * The warnings and timers set the grace period given to
 881                  * a user or group before he or she can not perform any
 882                  * more writing. If it is zero, a default is used.
 883                  */
 884                 qinf->qi_btimelimit = ddqp->d_btimer ?
 885                         be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 886                 qinf->qi_itimelimit = ddqp->d_itimer ?
 887                         be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 888                 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 889                         be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 890                 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 891                         be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 892                 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 893                         be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 894                 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 895                         be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 896                 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 897                 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 898                 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 899                 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 900                 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 901                 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 902
 903                 xfs_qm_dqdestroy(dqp);
 904         } else {
 905                 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 906                 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 907                 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 908                 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 909                 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 910                 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 911         }
 912
 913         qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
 914         qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
 915         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 916         qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
 917         register_shrinker(&qinf->qi_shrinker);
 918         return 0;
 919 }
 920
 921
 922 /*
 923  * Gets called when unmounting a filesystem or when all quotas get
 924  * turned off.
 925  * This purges the quota inodes, destroys locks and frees itself.
 926  */
 927 void
 928 xfs_qm_destroy_quotainfo(
 929         xfs_mount_t     *mp)
 930 {
 931         xfs_quotainfo_t *qi;
 932
 933         qi = mp->m_quotainfo;
 934         ASSERT(qi != NULL);
 935
 936         unregister_shrinker(&qi->qi_shrinker);
 937
 938         if (qi->qi_uquotaip) {
 939                 IRELE(qi->qi_uquotaip);
 940                 qi->qi_uquotaip = NULL; /* paranoia */
 941         }
 942         if (qi->qi_gquotaip) {
 943                 IRELE(qi->qi_gquotaip);
 944                 qi->qi_gquotaip = NULL;
 945         }
 946         if (qi->qi_pquotaip) {
 947                 IRELE(qi->qi_pquotaip);
 948                 qi->qi_pquotaip = NULL;
 949         }
 950         mutex_destroy(&qi->qi_quotaofflock);
 951         kmem_free(qi);
 952         mp->m_quotainfo = NULL;
 953 }
 954
 955 /*
 956  * Create an inode and return with a reference already taken, but unlocked
 957  * This is how we create quota inodes
 958  */
 959 STATIC int
 960 xfs_qm_qino_alloc(
 961         xfs_mount_t     *mp,
 962         xfs_inode_t     **ip,
 963         __int64_t       sbfields,
 964         uint            flags)
 965 {
 966         xfs_trans_t     *tp;
 967         int             error;
 968         int             committed;
 969
 970         *ip = NULL;
 971         /*
 972          * With superblock that doesn't have separate pquotino, we
 973          * share an inode between gquota and pquota. If the on-disk
 974          * superblock has GQUOTA and the filesystem is now mounted
 975          * with PQUOTA, just use sb_gquotino for sb_pquotino and
 976          * vice-versa.
 977          */
 978         if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
 979                         (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
 980                 xfs_ino_t ino = NULLFSINO;
 981
 982                 if ((flags & XFS_QMOPT_PQUOTA) &&
 983                              (mp->m_sb.sb_gquotino != NULLFSINO)) {
 984                         ino = mp->m_sb.sb_gquotino;
 985                         ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
 986                 } else if ((flags & XFS_QMOPT_GQUOTA) &&
 987                              (mp->m_sb.sb_pquotino != NULLFSINO)) {
 988                         ino = mp->m_sb.sb_pquotino;
 989                         ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
 990                 }
 991                 if (ino != NULLFSINO) {
 992                         error = xfs_iget(mp, NULL, ino, 0, 0, ip);
 993                         if (error)
 994                                 return error;
 995                         mp->m_sb.sb_gquotino = NULLFSINO;
 996                         mp->m_sb.sb_pquotino = NULLFSINO;
 997                 }
 998         }
 999
1000         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1001         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
1002                                   XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
1003         if (error) {
1004                 xfs_trans_cancel(tp, 0);
1005                 return error;
1006         }
1007
1008         if (!*ip) {
1009                 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
1010                                                                 &committed);
1011                 if (error) {
1012                         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1013                                          XFS_TRANS_ABORT);
1014                         return error;
1015                 }
1016         }
1017
1018         /*
1019          * Make the changes in the superblock, and log those too.
1020          * sbfields arg may contain fields other than *QUOTINO;
1021          * VERSIONNUM for example.
1022          */
1023         spin_lock(&mp->m_sb_lock);
1024         if (flags & XFS_QMOPT_SBVERSION) {
1025                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1026                 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1027                         XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
1028                                 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1029                                  XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1030                                  XFS_SB_QFLAGS));
1031
1032                 xfs_sb_version_addquota(&mp->m_sb);
1033                 mp->m_sb.sb_uquotino = NULLFSINO;
1034                 mp->m_sb.sb_gquotino = NULLFSINO;
1035                 mp->m_sb.sb_pquotino = NULLFSINO;
1036
1037                 /* qflags will get updated fully _after_ quotacheck */
1038                 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
1039         }
1040         if (flags & XFS_QMOPT_UQUOTA)
1041                 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1042         else if (flags & XFS_QMOPT_GQUOTA)
1043                 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1044         else
1045                 mp->m_sb.sb_pquotino = (*ip)->i_ino;
1046         spin_unlock(&mp->m_sb_lock);
1047         xfs_mod_sb(tp, sbfields);
1048
1049         if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1050                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1051                 return error;
1052         }
1053         return 0;
1054 }
1055
1056
1057 STATIC void
1058 xfs_qm_reset_dqcounts(
1059         xfs_mount_t     *mp,
1060         xfs_buf_t       *bp,
1061         xfs_dqid_t      id,
1062         uint            type)
1063 {
1064         struct xfs_dqblk        *dqb;
1065         int                     j;
1066
1067         trace_xfs_reset_dqcounts(bp, _RET_IP_);
1068
1069         /*
1070          * Reset all counters and timers. They'll be
1071          * started afresh by xfs_qm_quotacheck.
1072          */
1073 #ifdef DEBUG
1074         j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1075         do_div(j, sizeof(xfs_dqblk_t));
1076         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1077 #endif
1078         dqb = bp->b_addr;
1079         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1080                 struct xfs_disk_dquot   *ddq;
1081
1082                 ddq = (struct xfs_disk_dquot *)&dqb[j];
1083
1084                 /*
1085                  * Do a sanity check, and if needed, repair the dqblk. Don't
1086                  * output any warnings because it's perfectly possible to
1087                  * find uninitialised dquot blks. See comment in xfs_qm_dqcheck.
1088                  */
1089                 (void) xfs_qm_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1090                                       "xfs_quotacheck");
1091                 ddq->d_bcount = 0;
1092                 ddq->d_icount = 0;
1093                 ddq->d_rtbcount = 0;
1094                 ddq->d_btimer = 0;
1095                 ddq->d_itimer = 0;
1096                 ddq->d_rtbtimer = 0;
1097                 ddq->d_bwarns = 0;
1098                 ddq->d_iwarns = 0;
1099                 ddq->d_rtbwarns = 0;
1100
1101                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1102                         xfs_update_cksum((char *)&dqb[j],
1103                                          sizeof(struct xfs_dqblk),
1104                                          XFS_DQUOT_CRC_OFF);
1105                 }
1106         }
1107 }
1108
1109 STATIC int
1110 xfs_qm_dqiter_bufs(
1111         struct xfs_mount        *mp,
1112         xfs_dqid_t              firstid,
1113         xfs_fsblock_t           bno,
1114         xfs_filblks_t           blkcnt,
1115         uint                    flags,
1116         struct list_head        *buffer_list)
1117 {
1118         struct xfs_buf          *bp;
1119         int                     error;
1120         int                     type;
1121
1122         ASSERT(blkcnt > 0);
1123         type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1124                 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1125         error = 0;
1126
1127         /*
1128          * Blkcnt arg can be a very big number, and might even be
1129          * larger than the log itself. So, we have to break it up into
1130          * manageable-sized transactions.
1131          * Note that we don't start a permanent transaction here; we might
1132          * not be able to get a log reservation for the whole thing up front,
1133          * and we don't really care to either, because we just discard
1134          * everything if we were to crash in the middle of this loop.
1135          */
1136         while (blkcnt--) {
1137                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1138                               XFS_FSB_TO_DADDR(mp, bno),
1139                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1140                               &xfs_dquot_buf_ops);
1141
1142                 /*
1143                  * CRC and validation errors will return a EFSCORRUPTED here. If
1144                  * this occurs, re-read without CRC validation so that we can
1145                  * repair the damage via xfs_qm_reset_dqcounts(). This process
1146                  * will leave a trace in the log indicating corruption has
1147                  * been detected.
1148                  */
1149                 if (error == EFSCORRUPTED) {
1150                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1151                                       XFS_FSB_TO_DADDR(mp, bno),
1152                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1153                                       NULL);
1154                 }
1155
1156                 if (error)
1157                         break;
1158
1159                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1160                 xfs_buf_delwri_queue(bp, buffer_list);
1161                 xfs_buf_relse(bp);
1162
1163                 /* goto the next block. */
1164                 bno++;
1165                 firstid += mp->m_quotainfo->qi_dqperchunk;
1166         }
1167
1168         return error;
1169 }
1170
1171 /*
1172  * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1173  * caller supplied function for every chunk of dquots that we find.
1174  */
1175 STATIC int
1176 xfs_qm_dqiterate(
1177         struct xfs_mount        *mp,
1178         struct xfs_inode        *qip,
1179         uint                    flags,
1180         struct list_head        *buffer_list)
1181 {
1182         struct xfs_bmbt_irec    *map;
1183         int                     i, nmaps;       /* number of map entries */
1184         int                     error;          /* return value */
1185         xfs_fileoff_t           lblkno;
1186         xfs_filblks_t           maxlblkcnt;
1187         xfs_dqid_t              firstid;
1188         xfs_fsblock_t           rablkno;
1189         xfs_filblks_t           rablkcnt;
1190
1191         error = 0;
1192         /*
1193          * This looks racy, but we can't keep an inode lock across a
1194          * trans_reserve. But, this gets called during quotacheck, and that
1195          * happens only at mount time which is single threaded.
1196          */
1197         if (qip->i_d.di_nblocks == 0)
1198                 return 0;
1199
1200         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1201
1202         lblkno = 0;
1203         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1204         do {
1205                 nmaps = XFS_DQITER_MAP_SIZE;
1206                 /*
1207                  * We aren't changing the inode itself. Just changing
1208                  * some of its data. No new blocks are added here, and
1209                  * the inode is never added to the transaction.
1210                  */
1211                 xfs_ilock(qip, XFS_ILOCK_SHARED);
1212                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1213                                        map, &nmaps, 0);
1214                 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1215                 if (error)
1216                         break;
1217
1218                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1219                 for (i = 0; i < nmaps; i++) {
1220                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1221                         ASSERT(map[i].br_blockcount);
1222
1223
1224                         lblkno += map[i].br_blockcount;
1225
1226                         if (map[i].br_startblock == HOLESTARTBLOCK)
1227                                 continue;
1228
1229                         firstid = (xfs_dqid_t) map[i].br_startoff *
1230                                 mp->m_quotainfo->qi_dqperchunk;
1231                         /*
1232                          * Do a read-ahead on the next extent.
1233                          */
1234                         if ((i+1 < nmaps) &&
1235                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1236                                 rablkcnt =  map[i+1].br_blockcount;
1237                                 rablkno = map[i+1].br_startblock;
1238                                 while (rablkcnt--) {
1239                                         xfs_buf_readahead(mp->m_ddev_targp,
1240                                                XFS_FSB_TO_DADDR(mp, rablkno),
1241                                                mp->m_quotainfo->qi_dqchunklen,
1242                                                NULL);
1243                                         rablkno++;
1244                                 }
1245                         }
1246                         /*
1247                          * Iterate thru all the blks in the extent and
1248                          * reset the counters of all the dquots inside them.
1249                          */
1250                         error = xfs_qm_dqiter_bufs(mp, firstid,
1251                                                    map[i].br_startblock,
1252                                                    map[i].br_blockcount,
1253                                                    flags, buffer_list);
1254                         if (error)
1255                                 goto out;
1256                 }
1257         } while (nmaps > 0);
1258
1259 out:
1260         kmem_free(map);
1261         return error;
1262 }
1263
1264 /*
1265  * Called by dqusage_adjust in doing a quotacheck.
1266  *
1267  * Given the inode, and a dquot id this updates both the incore dqout as well
1268  * as the buffer copy. This is so that once the quotacheck is done, we can
1269  * just log all the buffers, as opposed to logging numerous updates to
1270  * individual dquots.
1271  */
1272 STATIC int
1273 xfs_qm_quotacheck_dqadjust(
1274         struct xfs_inode        *ip,
1275         xfs_dqid_t              id,
1276         uint                    type,
1277         xfs_qcnt_t              nblks,
1278         xfs_qcnt_t              rtblks)
1279 {
1280         struct xfs_mount        *mp = ip->i_mount;
1281         struct xfs_dquot        *dqp;
1282         int                     error;
1283
1284         error = xfs_qm_dqget(mp, ip, id, type,
1285                              XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1286         if (error) {
1287                 /*
1288                  * Shouldn't be able to turn off quotas here.
1289                  */
1290                 ASSERT(error != ESRCH);
1291                 ASSERT(error != ENOENT);
1292                 return error;
1293         }
1294
1295         trace_xfs_dqadjust(dqp);
1296
1297         /*
1298          * Adjust the inode count and the block count to reflect this inode's
1299          * resource usage.
1300          */
1301         be64_add_cpu(&dqp->q_core.d_icount, 1);
1302         dqp->q_res_icount++;
1303         if (nblks) {
1304                 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1305                 dqp->q_res_bcount += nblks;
1306         }
1307         if (rtblks) {
1308                 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1309                 dqp->q_res_rtbcount += rtblks;
1310         }
1311
1312         /*
1313          * Set default limits, adjust timers (since we changed usages)
1314          *
1315          * There are no timers for the default values set in the root dquot.
1316          */
1317         if (dqp->q_core.d_id) {
1318                 xfs_qm_adjust_dqlimits(mp, dqp);
1319                 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1320         }
1321
1322         dqp->dq_flags |= XFS_DQ_DIRTY;
1323         xfs_qm_dqput(dqp);
1324         return 0;
1325 }
1326
1327 STATIC int
1328 xfs_qm_get_rtblks(
1329         xfs_inode_t     *ip,
1330         xfs_qcnt_t      *O_rtblks)
1331 {
1332         xfs_filblks_t   rtblks;                 /* total rt blks */
1333         xfs_extnum_t    idx;                    /* extent record index */
1334         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1335         xfs_extnum_t    nextents;               /* number of extent entries */
1336         int             error;
1337
1338         ASSERT(XFS_IS_REALTIME_INODE(ip));
1339         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1340         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1341                 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1342                         return error;
1343         }
1344         rtblks = 0;
1345         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1346         for (idx = 0; idx < nextents; idx++)
1347                 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1348         *O_rtblks = (xfs_qcnt_t)rtblks;
1349         return 0;
1350 }
1351
1352 /*
1353  * callback routine supplied to bulkstat(). Given an inumber, find its
1354  * dquots and update them to account for resources taken by that inode.
1355  */
1356 /* ARGSUSED */
1357 STATIC int
1358 xfs_qm_dqusage_adjust(
1359         xfs_mount_t     *mp,            /* mount point for filesystem */
1360         xfs_ino_t       ino,            /* inode number to get data for */
1361         void            __user *buffer, /* not used */
1362         int             ubsize,         /* not used */
1363         int             *ubused,        /* not used */
1364         int             *res)           /* result code value */
1365 {
1366         xfs_inode_t     *ip;
1367         xfs_qcnt_t      nblks, rtblks = 0;
1368         int             error;
1369
1370         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1371
1372         /*
1373          * rootino must have its resources accounted for, not so with the quota
1374          * inodes.
1375          */
1376         if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1377                 *res = BULKSTAT_RV_NOTHING;
1378                 return XFS_ERROR(EINVAL);
1379         }
1380
1381         /*
1382          * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1383          * interface expects the inode to be exclusively locked because that's
1384          * the case in all other instances. It's OK that we do this because
1385          * quotacheck is done only at mount time.
1386          */
1387         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1388         if (error) {
1389                 *res = BULKSTAT_RV_NOTHING;
1390                 return error;
1391         }
1392
1393         ASSERT(ip->i_delayed_blks == 0);
1394
1395         if (XFS_IS_REALTIME_INODE(ip)) {
1396                 /*
1397                  * Walk thru the extent list and count the realtime blocks.
1398                  */
1399                 error = xfs_qm_get_rtblks(ip, &rtblks);
1400                 if (error)
1401                         goto error0;
1402         }
1403
1404         nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1405
1406         /*
1407          * Add the (disk blocks and inode) resources occupied by this
1408          * inode to its dquots. We do this adjustment in the incore dquot,
1409          * and also copy the changes to its buffer.
1410          * We don't care about putting these changes in a transaction
1411          * envelope because if we crash in the middle of a 'quotacheck'
1412          * we have to start from the beginning anyway.
1413          * Once we're done, we'll log all the dquot bufs.
1414          *
1415          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1416          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1417          */
1418         if (XFS_IS_UQUOTA_ON(mp)) {
1419                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1420                                                    XFS_DQ_USER, nblks, rtblks);
1421                 if (error)
1422                         goto error0;
1423         }
1424
1425         if (XFS_IS_GQUOTA_ON(mp)) {
1426                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1427                                                    XFS_DQ_GROUP, nblks, rtblks);
1428                 if (error)
1429                         goto error0;
1430         }
1431
1432         if (XFS_IS_PQUOTA_ON(mp)) {
1433                 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1434                                                    XFS_DQ_PROJ, nblks, rtblks);
1435                 if (error)
1436                         goto error0;
1437         }
1438
1439         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1440         IRELE(ip);
1441         *res = BULKSTAT_RV_DIDONE;
1442         return 0;
1443
1444 error0:
1445         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1446         IRELE(ip);
1447         *res = BULKSTAT_RV_GIVEUP;
1448         return error;
1449 }
1450
1451 STATIC int
1452 xfs_qm_flush_one(
1453         struct xfs_dquot        *dqp,
1454         void                    *data)
1455 {
1456         struct list_head        *buffer_list = data;
1457         struct xfs_buf          *bp = NULL;
1458         int                     error = 0;
1459
1460         xfs_dqlock(dqp);
1461         if (dqp->dq_flags & XFS_DQ_FREEING)
1462                 goto out_unlock;
1463         if (!XFS_DQ_IS_DIRTY(dqp))
1464                 goto out_unlock;
1465
1466         xfs_dqflock(dqp);
1467         error = xfs_qm_dqflush(dqp, &bp);
1468         if (error)
1469                 goto out_unlock;
1470
1471         xfs_buf_delwri_queue(bp, buffer_list);
1472         xfs_buf_relse(bp);
1473 out_unlock:
1474         xfs_dqunlock(dqp);
1475         return error;
1476 }
1477
1478 /*
1479  * Walk thru all the filesystem inodes and construct a consistent view
1480  * of the disk quota world. If the quotacheck fails, disable quotas.
1481  */
1482 int
1483 xfs_qm_quotacheck(
1484         xfs_mount_t     *mp)
1485 {
1486         int                     done, count, error, error2;
1487         xfs_ino_t               lastino;
1488         size_t                  structsz;
1489         uint                    flags;
1490         LIST_HEAD               (buffer_list);
1491         struct xfs_inode        *uip = mp->m_quotainfo->qi_uquotaip;
1492         struct xfs_inode        *gip = mp->m_quotainfo->qi_gquotaip;
1493         struct xfs_inode        *pip = mp->m_quotainfo->qi_pquotaip;
1494
1495         count = INT_MAX;
1496         structsz = 1;
1497         lastino = 0;
1498         flags = 0;
1499
1500         ASSERT(uip || gip || pip);
1501         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1502
1503         xfs_notice(mp, "Quotacheck needed: Please wait.");
1504
1505         /*
1506          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1507          * their counters to zero. We need a clean slate.
1508          * We don't log our changes till later.
1509          */
1510         if (uip) {
1511                 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1512                                          &buffer_list);
1513                 if (error)
1514                         goto error_return;
1515                 flags |= XFS_UQUOTA_CHKD;
1516         }
1517
1518         if (gip) {
1519                 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1520                                          &buffer_list);
1521                 if (error)
1522                         goto error_return;
1523                 flags |= XFS_GQUOTA_CHKD;
1524         }
1525
1526         if (pip) {
1527                 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1528                                          &buffer_list);
1529                 if (error)
1530                         goto error_return;
1531                 flags |= XFS_PQUOTA_CHKD;
1532         }
1533
1534         do {
1535                 /*
1536                  * Iterate thru all the inodes in the file system,
1537                  * adjusting the corresponding dquot counters in core.
1538                  */
1539                 error = xfs_bulkstat(mp, &lastino, &count,
1540                                      xfs_qm_dqusage_adjust,
1541                                      structsz, NULL, &done);
1542                 if (error)
1543                         break;
1544
1545         } while (!done);
1546
1547         /*
1548          * We've made all the changes that we need to make incore.  Flush them
1549          * down to disk buffers if everything was updated successfully.
1550          */
1551         if (XFS_IS_UQUOTA_ON(mp)) {
1552                 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1553                                           &buffer_list);
1554         }
1555         if (XFS_IS_GQUOTA_ON(mp)) {
1556                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1557                                            &buffer_list);
1558                 if (!error)
1559                         error = error2;
1560         }
1561         if (XFS_IS_PQUOTA_ON(mp)) {
1562                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1563                                            &buffer_list);
1564                 if (!error)
1565                         error = error2;
1566         }
1567
1568         error2 = xfs_buf_delwri_submit(&buffer_list);
1569         if (!error)
1570                 error = error2;
1571
1572         /*
1573          * We can get this error if we couldn't do a dquot allocation inside
1574          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1575          * dirty dquots that might be cached, we just want to get rid of them
1576          * and turn quotaoff. The dquots won't be attached to any of the inodes
1577          * at this point (because we intentionally didn't in dqget_noattach).
1578          */
1579         if (error) {
1580                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1581                 goto error_return;
1582         }
1583
1584         /*
1585          * If one type of quotas is off, then it will lose its
1586          * quotachecked status, since we won't be doing accounting for
1587          * that type anymore.
1588          */
1589         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1590         mp->m_qflags |= flags;
1591
1592  error_return:
1593         while (!list_empty(&buffer_list)) {
1594                 struct xfs_buf *bp =
1595                         list_first_entry(&buffer_list, struct xfs_buf, b_list);
1596                 list_del_init(&bp->b_list);
1597                 xfs_buf_relse(bp);
1598         }
1599
1600         if (error) {
1601                 xfs_warn(mp,
1602         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1603                         error);
1604                 /*
1605                  * We must turn off quotas.
1606                  */
1607                 ASSERT(mp->m_quotainfo != NULL);
1608                 xfs_qm_destroy_quotainfo(mp);
1609                 if (xfs_mount_reset_sbqflags(mp)) {
1610                         xfs_warn(mp,
1611                                 "Quotacheck: Failed to reset quota flags.");
1612                 }
1613         } else
1614                 xfs_notice(mp, "Quotacheck: Done.");
1615         return (error);
1616 }
1617
1618 /*
1619  * This is called after the superblock has been read in and we're ready to
1620  * iget the quota inodes.
1621  */
1622 STATIC int
1623 xfs_qm_init_quotainos(
1624         xfs_mount_t     *mp)
1625 {
1626         struct xfs_inode        *uip = NULL;
1627         struct xfs_inode        *gip = NULL;
1628         struct xfs_inode        *pip = NULL;
1629         int                     error;
1630         __int64_t               sbflags = 0;
1631         uint                    flags = 0;
1632
1633         ASSERT(mp->m_quotainfo);
1634
1635         /*
1636          * Get the uquota and gquota inodes
1637          */
1638         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1639                 if (XFS_IS_UQUOTA_ON(mp) &&
1640                     mp->m_sb.sb_uquotino != NULLFSINO) {
1641                         ASSERT(mp->m_sb.sb_uquotino > 0);
1642                         error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1643                                              0, 0, &uip);
1644                         if (error)
1645                                 return XFS_ERROR(error);
1646                 }
1647                 if (XFS_IS_GQUOTA_ON(mp) &&
1648                     mp->m_sb.sb_gquotino != NULLFSINO) {
1649                         ASSERT(mp->m_sb.sb_gquotino > 0);
1650                         error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1651                                              0, 0, &gip);
1652                         if (error)
1653                                 goto error_rele;
1654                 }
1655                 if (XFS_IS_PQUOTA_ON(mp) &&
1656                     mp->m_sb.sb_pquotino != NULLFSINO) {
1657                         ASSERT(mp->m_sb.sb_pquotino > 0);
1658                         error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1659                                              0, 0, &pip);
1660                         if (error)
1661                                 goto error_rele;
1662                 }
1663         } else {
1664                 flags |= XFS_QMOPT_SBVERSION;
1665                 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1666                             XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1667                             XFS_SB_QFLAGS);
1668         }
1669
1670         /*
1671          * Create the three inodes, if they don't exist already. The changes
1672          * made above will get added to a transaction and logged in one of
1673          * the qino_alloc calls below.  If the device is readonly,
1674          * temporarily switch to read-write to do this.
1675          */
1676         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1677                 error = xfs_qm_qino_alloc(mp, &uip,
1678                                               sbflags | XFS_SB_UQUOTINO,
1679                                               flags | XFS_QMOPT_UQUOTA);
1680                 if (error)
1681                         goto error_rele;
1682
1683                 flags &= ~XFS_QMOPT_SBVERSION;
1684         }
1685         if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1686                 error = xfs_qm_qino_alloc(mp, &gip,
1687                                           sbflags | XFS_SB_GQUOTINO,
1688                                           flags | XFS_QMOPT_GQUOTA);
1689                 if (error)
1690                         goto error_rele;
1691
1692                 flags &= ~XFS_QMOPT_SBVERSION;
1693         }
1694         if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1695                 error = xfs_qm_qino_alloc(mp, &pip,
1696                                           sbflags | XFS_SB_PQUOTINO,
1697                                           flags | XFS_QMOPT_PQUOTA);
1698                 if (error)
1699                         goto error_rele;
1700         }
1701
1702         mp->m_quotainfo->qi_uquotaip = uip;
1703         mp->m_quotainfo->qi_gquotaip = gip;
1704         mp->m_quotainfo->qi_pquotaip = pip;
1705
1706         return 0;
1707
1708 error_rele:
1709         if (uip)
1710                 IRELE(uip);
1711         if (gip)
1712                 IRELE(gip);
1713         if (pip)
1714                 IRELE(pip);
1715         return XFS_ERROR(error);
1716 }
1717
1718 STATIC void
1719 xfs_qm_dqfree_one(
1720         struct xfs_dquot        *dqp)
1721 {
1722         struct xfs_mount        *mp = dqp->q_mount;
1723         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1724
1725         mutex_lock(&qi->qi_tree_lock);
1726         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1727                           be32_to_cpu(dqp->q_core.d_id));
1728
1729         qi->qi_dquots--;
1730         mutex_unlock(&qi->qi_tree_lock);
1731
1732         xfs_qm_dqdestroy(dqp);
1733 }
1734
1735 /*
1736  * Start a transaction and write the incore superblock changes to
1737  * disk. flags parameter indicates which fields have changed.
1738  */
1739 int
1740 xfs_qm_write_sb_changes(
1741         xfs_mount_t     *mp,
1742         __int64_t       flags)
1743 {
1744         xfs_trans_t     *tp;
1745         int             error;
1746
1747         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1748         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1749         if (error) {
1750                 xfs_trans_cancel(tp, 0);
1751                 return error;
1752         }
1753
1754         xfs_mod_sb(tp, flags);
1755         error = xfs_trans_commit(tp, 0);
1756
1757         return error;
1758 }
1759
1760
1761 /* --------------- utility functions for vnodeops ---------------- */
1762
1763
1764 /*
1765  * Given an inode, a uid, gid and prid make sure that we have
1766  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1767  * quotas by creating this file.
1768  * This also attaches dquot(s) to the given inode after locking it,
1769  * and returns the dquots corresponding to the uid and/or gid.
1770  *
1771  * in   : inode (unlocked)
1772  * out  : udquot, gdquot with references taken and unlocked
1773  */
1774 int
1775 xfs_qm_vop_dqalloc(
1776         struct xfs_inode        *ip,
1777         xfs_dqid_t              uid,
1778         xfs_dqid_t              gid,
1779         prid_t                  prid,
1780         uint                    flags,
1781         struct xfs_dquot        **O_udqpp,
1782         struct xfs_dquot        **O_gdqpp,
1783         struct xfs_dquot        **O_pdqpp)
1784 {
1785         struct xfs_mount        *mp = ip->i_mount;
1786         struct xfs_dquot        *uq = NULL;
1787         struct xfs_dquot        *gq = NULL;
1788         struct xfs_dquot        *pq = NULL;
1789         int                     error;
1790         uint                    lockflags;
1791
1792         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1793                 return 0;
1794
1795         lockflags = XFS_ILOCK_EXCL;
1796         xfs_ilock(ip, lockflags);
1797
1798         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1799                 gid = ip->i_d.di_gid;
1800
1801         /*
1802          * Attach the dquot(s) to this inode, doing a dquot allocation
1803          * if necessary. The dquot(s) will not be locked.
1804          */
1805         if (XFS_NOT_DQATTACHED(mp, ip)) {
1806                 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1807                 if (error) {
1808                         xfs_iunlock(ip, lockflags);
1809                         return error;
1810                 }
1811         }
1812
1813         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1814                 if (ip->i_d.di_uid != uid) {
1815                         /*
1816                          * What we need is the dquot that has this uid, and
1817                          * if we send the inode to dqget, the uid of the inode
1818                          * takes priority over what's sent in the uid argument.
1819                          * We must unlock inode here before calling dqget if
1820                          * we're not sending the inode, because otherwise
1821                          * we'll deadlock by doing trans_reserve while
1822                          * holding ilock.
1823                          */
1824                         xfs_iunlock(ip, lockflags);
1825                         error = xfs_qm_dqget(mp, NULL, uid,
1826                                                  XFS_DQ_USER,
1827                                                  XFS_QMOPT_DQALLOC |
1828                                                  XFS_QMOPT_DOWARN,
1829                                                  &uq);
1830                         if (error) {
1831                                 ASSERT(error != ENOENT);
1832                                 return error;
1833                         }
1834                         /*
1835                          * Get the ilock in the right order.
1836                          */
1837                         xfs_dqunlock(uq);
1838                         lockflags = XFS_ILOCK_SHARED;
1839                         xfs_ilock(ip, lockflags);
1840                 } else {
1841                         /*
1842                          * Take an extra reference, because we'll return
1843                          * this to caller
1844                          */
1845                         ASSERT(ip->i_udquot);
1846                         uq = xfs_qm_dqhold(ip->i_udquot);
1847                 }
1848         }
1849         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1850                 if (ip->i_d.di_gid != gid) {
1851                         xfs_iunlock(ip, lockflags);
1852                         error = xfs_qm_dqget(mp, NULL, gid,
1853                                                  XFS_DQ_GROUP,
1854                                                  XFS_QMOPT_DQALLOC |
1855                                                  XFS_QMOPT_DOWARN,
1856                                                  &gq);
1857                         if (error) {
1858                                 ASSERT(error != ENOENT);
1859                                 goto error_rele;
1860                         }
1861                         xfs_dqunlock(gq);
1862                         lockflags = XFS_ILOCK_SHARED;
1863                         xfs_ilock(ip, lockflags);
1864                 } else {
1865                         ASSERT(ip->i_gdquot);
1866                         gq = xfs_qm_dqhold(ip->i_gdquot);
1867                 }
1868         }
1869         if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1870                 if (xfs_get_projid(ip) != prid) {
1871                         xfs_iunlock(ip, lockflags);
1872                         error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1873                                                  XFS_DQ_PROJ,
1874                                                  XFS_QMOPT_DQALLOC |
1875                                                  XFS_QMOPT_DOWARN,
1876                                                  &pq);
1877                         if (error) {
1878                                 ASSERT(error != ENOENT);
1879                                 goto error_rele;
1880                         }
1881                         xfs_dqunlock(pq);
1882                         lockflags = XFS_ILOCK_SHARED;
1883                         xfs_ilock(ip, lockflags);
1884                 } else {
1885                         ASSERT(ip->i_pdquot);
1886                         pq = xfs_qm_dqhold(ip->i_pdquot);
1887                 }
1888         }
1889         if (uq)
1890                 trace_xfs_dquot_dqalloc(ip);
1891
1892         xfs_iunlock(ip, lockflags);
1893         if (O_udqpp)
1894                 *O_udqpp = uq;
1895         else if (uq)
1896                 xfs_qm_dqrele(uq);
1897         if (O_gdqpp)
1898                 *O_gdqpp = gq;
1899         else if (gq)
1900                 xfs_qm_dqrele(gq);
1901         if (O_pdqpp)
1902                 *O_pdqpp = pq;
1903         else if (pq)
1904                 xfs_qm_dqrele(pq);
1905         return 0;
1906
1907 error_rele:
1908         if (gq)
1909                 xfs_qm_dqrele(gq);
1910         if (uq)
1911                 xfs_qm_dqrele(uq);
1912         return error;
1913 }
1914
1915 /*
1916  * Actually transfer ownership, and do dquot modifications.
1917  * These were already reserved.
1918  */
1919 xfs_dquot_t *
1920 xfs_qm_vop_chown(
1921         xfs_trans_t     *tp,
1922         xfs_inode_t     *ip,
1923         xfs_dquot_t     **IO_olddq,
1924         xfs_dquot_t     *newdq)
1925 {
1926         xfs_dquot_t     *prevdq;
1927         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1928                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1929
1930
1931         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1932         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1933
1934         /* old dquot */
1935         prevdq = *IO_olddq;
1936         ASSERT(prevdq);
1937         ASSERT(prevdq != newdq);
1938
1939         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1940         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1941
1942         /* the sparkling new dquot */
1943         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1944         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1945
1946         /*
1947          * Take an extra reference, because the inode is going to keep
1948          * this dquot pointer even after the trans_commit.
1949          */
1950         *IO_olddq = xfs_qm_dqhold(newdq);
1951
1952         return prevdq;
1953 }
1954
1955 /*
1956  * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1957  */
1958 int
1959 xfs_qm_vop_chown_reserve(
1960         struct xfs_trans        *tp,
1961         struct xfs_inode        *ip,
1962         struct xfs_dquot        *udqp,
1963         struct xfs_dquot        *gdqp,
1964         struct xfs_dquot        *pdqp,
1965         uint                    flags)
1966 {
1967         struct xfs_mount        *mp = ip->i_mount;
1968         uint                    delblks, blkflags, prjflags = 0;
1969         struct xfs_dquot        *udq_unres = NULL;
1970         struct xfs_dquot        *gdq_unres = NULL;
1971         struct xfs_dquot        *pdq_unres = NULL;
1972         struct xfs_dquot        *udq_delblks = NULL;
1973         struct xfs_dquot        *gdq_delblks = NULL;
1974         struct xfs_dquot        *pdq_delblks = NULL;
1975         int                     error;
1976
1977
1978         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1979         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1980
1981         delblks = ip->i_delayed_blks;
1982         blkflags = XFS_IS_REALTIME_INODE(ip) ?
1983                         XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1984
1985         if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1986             ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1987                 udq_delblks = udqp;
1988                 /*
1989                  * If there are delayed allocation blocks, then we have to
1990                  * unreserve those from the old dquot, and add them to the
1991                  * new dquot.
1992                  */
1993                 if (delblks) {
1994                         ASSERT(ip->i_udquot);
1995                         udq_unres = ip->i_udquot;
1996                 }
1997         }
1998         if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1999             ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
2000                 gdq_delblks = gdqp;
2001                 if (delblks) {
2002                         ASSERT(ip->i_gdquot);
2003                         gdq_unres = ip->i_gdquot;
2004                 }
2005         }
2006
2007         if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2008             xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2009                 prjflags = XFS_QMOPT_ENOSPC;
2010                 pdq_delblks = pdqp;
2011                 if (delblks) {
2012                         ASSERT(ip->i_pdquot);
2013                         pdq_unres = ip->i_pdquot;
2014                 }
2015         }
2016
2017         error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2018                                 udq_delblks, gdq_delblks, pdq_delblks,
2019                                 ip->i_d.di_nblocks, 1,
2020                                 flags | blkflags | prjflags);
2021         if (error)
2022                 return error;
2023
2024         /*
2025          * Do the delayed blks reservations/unreservations now. Since, these
2026          * are done without the help of a transaction, if a reservation fails
2027          * its previous reservations won't be automatically undone by trans
2028          * code. So, we have to do it manually here.
2029          */
2030         if (delblks) {
2031                 /*
2032                  * Do the reservations first. Unreservation can't fail.
2033                  */
2034                 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2035                 ASSERT(udq_unres || gdq_unres || pdq_unres);
2036                 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2037                             udq_delblks, gdq_delblks, pdq_delblks,
2038                             (xfs_qcnt_t)delblks, 0,
2039                             flags | blkflags | prjflags);
2040                 if (error)
2041                         return error;
2042                 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2043                                 udq_unres, gdq_unres, pdq_unres,
2044                                 -((xfs_qcnt_t)delblks), 0, blkflags);
2045         }
2046
2047         return (0);
2048 }
2049
2050 int
2051 xfs_qm_vop_rename_dqattach(
2052         struct xfs_inode        **i_tab)
2053 {
2054         struct xfs_mount        *mp = i_tab[0]->i_mount;
2055         int                     i;
2056
2057         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2058                 return 0;
2059
2060         for (i = 0; (i < 4 && i_tab[i]); i++) {
2061                 struct xfs_inode        *ip = i_tab[i];
2062                 int                     error;
2063
2064                 /*
2065                  * Watch out for duplicate entries in the table.
2066                  */
2067                 if (i == 0 || ip != i_tab[i-1]) {
2068                         if (XFS_NOT_DQATTACHED(mp, ip)) {
2069                                 error = xfs_qm_dqattach(ip, 0);
2070                                 if (error)
2071                                         return error;
2072                         }
2073                 }
2074         }
2075         return 0;
2076 }
2077
2078 void
2079 xfs_qm_vop_create_dqattach(
2080         struct xfs_trans        *tp,
2081         struct xfs_inode        *ip,
2082         struct xfs_dquot        *udqp,
2083         struct xfs_dquot        *gdqp,
2084         struct xfs_dquot        *pdqp)
2085 {
2086         struct xfs_mount        *mp = tp->t_mountp;
2087
2088         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2089                 return;
2090
2091         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2092         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2093
2094         if (udqp) {
2095                 ASSERT(ip->i_udquot == NULL);
2096                 ASSERT(XFS_IS_UQUOTA_ON(mp));
2097                 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2098
2099                 ip->i_udquot = xfs_qm_dqhold(udqp);
2100                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2101         }
2102         if (gdqp) {
2103                 ASSERT(ip->i_gdquot == NULL);
2104                 ASSERT(XFS_IS_GQUOTA_ON(mp));
2105                 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
2106                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
2107                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2108         }
2109         if (pdqp) {
2110                 ASSERT(ip->i_pdquot == NULL);
2111                 ASSERT(XFS_IS_PQUOTA_ON(mp));
2112                 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2113
2114                 ip->i_pdquot = xfs_qm_dqhold(pdqp);
2115                 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2116         }
2117 }
2118