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_shared.h"
  22 #include "xfs_bit.h"
  23 #include "xfs_log.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_quota.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dinode.h"
  33 #include "xfs_inode.h"
  34 #include "xfs_ialloc.h"
  35 #include "xfs_itable.h"
  36 #include "xfs_rtalloc.h"
  37 #include "xfs_error.h"
  38 #include "xfs_bmap.h"
  39 #include "xfs_attr.h"
  40 #include "xfs_buf_item.h"
  41 #include "xfs_trans_space.h"
  42 #include "xfs_qm.h"
  43 #include "xfs_trace.h"
  44 #include "xfs_icache.h"
  45 #include "xfs_cksum.h"
  46
  47 /*
  48  * The global quota manager. There is only one of these for the entire
  49  * system, _not_ one per file system. XQM keeps track of the overall
  50  * quota functionality, including maintaining the freelist and hash
  51  * tables of dquots.
  52  */
  53 STATIC int      xfs_qm_init_quotainos(xfs_mount_t *);
  54 STATIC int      xfs_qm_init_quotainfo(xfs_mount_t *);
  55
  56
  57 STATIC void     xfs_qm_dqfree_one(struct xfs_dquot *dqp);
  58 /*
  59  * We use the batch lookup interface to iterate over the dquots as it
  60  * currently is the only interface into the radix tree code that allows
  61  * fuzzy lookups instead of exact matches.  Holding the lock over multiple
  62  * operations is fine as all callers are used either during mount/umount
  63  * or quotaoff.
  64  */
  65 #define XFS_DQ_LOOKUP_BATCH     32
  66
  67 STATIC int
  68 xfs_qm_dquot_walk(
  69         struct xfs_mount        *mp,
  70         int                     type,
  71         int                     (*execute)(struct xfs_dquot *dqp, void *data),
  72         void                    *data)
  73 {
  74         struct xfs_quotainfo    *qi = mp->m_quotainfo;
  75         struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
  76         uint32_t                next_index;
  77         int                     last_error = 0;
  78         int                     skipped;
  79         int                     nr_found;
  80
  81 restart:
  82         skipped = 0;
  83         next_index = 0;
  84         nr_found = 0;
  85
  86         while (1) {
  87                 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
  88                 int             error = 0;
  89                 int             i;
  90
  91                 mutex_lock(&qi->qi_tree_lock);
  92                 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
  93                                         next_index, XFS_DQ_LOOKUP_BATCH);
  94                 if (!nr_found) {
  95                         mutex_unlock(&qi->qi_tree_lock);
  96                         break;
  97                 }
  98
  99                 for (i = 0; i < nr_found; i++) {
 100                         struct xfs_dquot *dqp = batch[i];
 101
 102                         next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
 103
 104                         error = execute(batch[i], data);
 105                         if (error == EAGAIN) {
 106                                 skipped++;
 107                                 continue;
 108                         }
 109                         if (error && last_error != EFSCORRUPTED)
 110                                 last_error = error;
 111                 }
 112
 113                 mutex_unlock(&qi->qi_tree_lock);
 114
 115                 /* bail out if the filesystem is corrupted.  */
 116                 if (last_error == EFSCORRUPTED) {
 117                         skipped = 0;
 118                         break;
 119                 }
 120         }
 121
 122         if (skipped) {
 123                 delay(1);
 124                 goto restart;
 125         }
 126
 127         return last_error;
 128 }
 129
 130
 131 /*
 132  * Purge a dquot from all tracking data structures and free it.
 133  */
 134 STATIC int
 135 xfs_qm_dqpurge(
 136         struct xfs_dquot        *dqp,
 137         void                    *data)
 138 {
 139         struct xfs_mount        *mp = dqp->q_mount;
 140         struct xfs_quotainfo    *qi = mp->m_quotainfo;
 141         struct xfs_dquot        *gdqp = NULL;
 142         struct xfs_dquot        *pdqp = NULL;
 143
 144         xfs_dqlock(dqp);
 145         if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 146                 xfs_dqunlock(dqp);
 147                 return EAGAIN;
 148         }
 149
 150         /*
 151          * If this quota has a hint attached, prepare for releasing it now.
 152          */
 153         gdqp = dqp->q_gdquot;
 154         if (gdqp) {
 155                 xfs_dqlock(gdqp);
 156                 dqp->q_gdquot = NULL;
 157         }
 158
 159         pdqp = dqp->q_pdquot;
 160         if (pdqp) {
 161                 xfs_dqlock(pdqp);
 162                 dqp->q_pdquot = NULL;
 163         }
 164
 165         dqp->dq_flags |= XFS_DQ_FREEING;
 166
 167         xfs_dqflock(dqp);
 168
 169         /*
 170          * If we are turning this type of quotas off, we don't care
 171          * about the dirty metadata sitting in this dquot. OTOH, if
 172          * we're unmounting, we do care, so we flush it and wait.
 173          */
 174         if (XFS_DQ_IS_DIRTY(dqp)) {
 175                 struct xfs_buf  *bp = NULL;
 176                 int             error;
 177
 178                 /*
 179                  * We don't care about getting disk errors here. We need
 180                  * to purge this dquot anyway, so we go ahead regardless.
 181                  */
 182                 error = xfs_qm_dqflush(dqp, &bp);
 183                 if (error) {
 184                         xfs_warn(mp, "%s: dquot %p flush failed",
 185                                 __func__, dqp);
 186                 } else {
 187                         error = xfs_bwrite(bp);
 188                         xfs_buf_relse(bp);
 189                 }
 190                 xfs_dqflock(dqp);
 191         }
 192
 193         ASSERT(atomic_read(&dqp->q_pincount) == 0);
 194         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 195                !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 196
 197         xfs_dqfunlock(dqp);
 198         xfs_dqunlock(dqp);
 199
 200         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
 201                           be32_to_cpu(dqp->q_core.d_id));
 202         qi->qi_dquots--;
 203
 204         /*
 205          * We move dquots to the freelist as soon as their reference count
 206          * hits zero, so it really should be on the freelist here.
 207          */
 208         ASSERT(!list_empty(&dqp->q_lru));
 209         list_lru_del(&qi->qi_lru, &dqp->q_lru);
 210         XFS_STATS_DEC(xs_qm_dquot_unused);
 211
 212         xfs_qm_dqdestroy(dqp);
 213
 214         if (gdqp)
 215                 xfs_qm_dqput(gdqp);
 216         if (pdqp)
 217                 xfs_qm_dqput(pdqp);
 218         return 0;
 219 }
 220
 221 /*
 222  * Purge the dquot cache.
 223  */
 224 void
 225 xfs_qm_dqpurge_all(
 226         struct xfs_mount        *mp,
 227         uint                    flags)
 228 {
 229         if (flags & XFS_QMOPT_UQUOTA)
 230                 xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge, NULL);
 231         if (flags & XFS_QMOPT_GQUOTA)
 232                 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 233         if (flags & XFS_QMOPT_PQUOTA)
 234                 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 235 }
 236
 237 /*
 238  * Just destroy the quotainfo structure.
 239  */
 240 void
 241 xfs_qm_unmount(
 242         struct xfs_mount        *mp)
 243 {
 244         if (mp->m_quotainfo) {
 245                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 246                 xfs_qm_destroy_quotainfo(mp);
 247         }
 248 }
 249
 250
 251 /*
 252  * This is called from xfs_mountfs to start quotas and initialize all
 253  * necessary data structures like quotainfo.  This is also responsible for
 254  * running a quotacheck as necessary.  We are guaranteed that the superblock
 255  * is consistently read in at this point.
 256  *
 257  * If we fail here, the mount will continue with quota turned off. We don't
 258  * need to inidicate success or failure at all.
 259  */
 260 void
 261 xfs_qm_mount_quotas(
 262         xfs_mount_t     *mp)
 263 {
 264         int             error = 0;
 265         uint            sbf;
 266
 267         /*
 268          * If quotas on realtime volumes is not supported, we disable
 269          * quotas immediately.
 270          */
 271         if (mp->m_sb.sb_rextents) {
 272                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 273                 mp->m_qflags = 0;
 274                 goto write_changes;
 275         }
 276
 277         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 278
 279         /*
 280          * Allocate the quotainfo structure inside the mount struct, and
 281          * create quotainode(s), and change/rev superblock if necessary.
 282          */
 283         error = xfs_qm_init_quotainfo(mp);
 284         if (error) {
 285                 /*
 286                  * We must turn off quotas.
 287                  */
 288                 ASSERT(mp->m_quotainfo == NULL);
 289                 mp->m_qflags = 0;
 290                 goto write_changes;
 291         }
 292         /*
 293          * If any of the quotas are not consistent, do a quotacheck.
 294          */
 295         if (XFS_QM_NEED_QUOTACHECK(mp)) {
 296                 error = xfs_qm_quotacheck(mp);
 297                 if (error) {
 298                         /* Quotacheck failed and disabled quotas. */
 299                         return;
 300                 }
 301         }
 302         /*
 303          * If one type of quotas is off, then it will lose its
 304          * quotachecked status, since we won't be doing accounting for
 305          * that type anymore.
 306          */
 307         if (!XFS_IS_UQUOTA_ON(mp))
 308                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 309         if (!XFS_IS_GQUOTA_ON(mp))
 310                 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
 311         if (!XFS_IS_PQUOTA_ON(mp))
 312                 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
 313
 314  write_changes:
 315         /*
 316          * We actually don't have to acquire the m_sb_lock at all.
 317          * This can only be called from mount, and that's single threaded. XXX
 318          */
 319         spin_lock(&mp->m_sb_lock);
 320         sbf = mp->m_sb.sb_qflags;
 321         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 322         spin_unlock(&mp->m_sb_lock);
 323
 324         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 325                 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 326                         /*
 327                          * We could only have been turning quotas off.
 328                          * We aren't in very good shape actually because
 329                          * the incore structures are convinced that quotas are
 330                          * off, but the on disk superblock doesn't know that !
 331                          */
 332                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 333                         xfs_alert(mp, "%s: Superblock update failed!",
 334                                 __func__);
 335                 }
 336         }
 337
 338         if (error) {
 339                 xfs_warn(mp, "Failed to initialize disk quotas.");
 340                 return;
 341         }
 342 }
 343
 344 /*
 345  * Called from the vfsops layer.
 346  */
 347 void
 348 xfs_qm_unmount_quotas(
 349         xfs_mount_t     *mp)
 350 {
 351         /*
 352          * Release the dquots that root inode, et al might be holding,
 353          * before we flush quotas and blow away the quotainfo structure.
 354          */
 355         ASSERT(mp->m_rootip);
 356         xfs_qm_dqdetach(mp->m_rootip);
 357         if (mp->m_rbmip)
 358                 xfs_qm_dqdetach(mp->m_rbmip);
 359         if (mp->m_rsumip)
 360                 xfs_qm_dqdetach(mp->m_rsumip);
 361
 362         /*
 363          * Release the quota inodes.
 364          */
 365         if (mp->m_quotainfo) {
 366                 if (mp->m_quotainfo->qi_uquotaip) {
 367                         IRELE(mp->m_quotainfo->qi_uquotaip);
 368                         mp->m_quotainfo->qi_uquotaip = NULL;
 369                 }
 370                 if (mp->m_quotainfo->qi_gquotaip) {
 371                         IRELE(mp->m_quotainfo->qi_gquotaip);
 372                         mp->m_quotainfo->qi_gquotaip = NULL;
 373                 }
 374                 if (mp->m_quotainfo->qi_pquotaip) {
 375                         IRELE(mp->m_quotainfo->qi_pquotaip);
 376                         mp->m_quotainfo->qi_pquotaip = NULL;
 377                 }
 378         }
 379 }
 380
 381 STATIC int
 382 xfs_qm_dqattach_one(
 383         xfs_inode_t     *ip,
 384         xfs_dqid_t      id,
 385         uint            type,
 386         uint            doalloc,
 387         xfs_dquot_t     *udqhint, /* hint */
 388         xfs_dquot_t     **IO_idqpp)
 389 {
 390         xfs_dquot_t     *dqp;
 391         int             error;
 392
 393         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 394         error = 0;
 395
 396         /*
 397          * See if we already have it in the inode itself. IO_idqpp is
 398          * &i_udquot or &i_gdquot. This made the code look weird, but
 399          * made the logic a lot simpler.
 400          */
 401         dqp = *IO_idqpp;
 402         if (dqp) {
 403                 trace_xfs_dqattach_found(dqp);
 404                 return 0;
 405         }
 406
 407         /*
 408          * udqhint is the i_udquot field in inode, and is non-NULL only
 409          * when the type arg is group/project. Its purpose is to save a
 410          * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 411          * the user dquot.
 412          */
 413         if (udqhint) {
 414                 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 415                 xfs_dqlock(udqhint);
 416
 417                 /*
 418                  * No need to take dqlock to look at the id.
 419                  *
 420                  * The ID can't change until it gets reclaimed, and it won't
 421                  * be reclaimed as long as we have a ref from inode and we
 422                  * hold the ilock.
 423                  */
 424                 if (type == XFS_DQ_GROUP)
 425                         dqp = udqhint->q_gdquot;
 426                 else
 427                         dqp = udqhint->q_pdquot;
 428                 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 429                         ASSERT(*IO_idqpp == NULL);
 430
 431                         *IO_idqpp = xfs_qm_dqhold(dqp);
 432                         xfs_dqunlock(udqhint);
 433                         return 0;
 434                 }
 435
 436                 /*
 437                  * We can't hold a dquot lock when we call the dqget code.
 438                  * We'll deadlock in no time, because of (not conforming to)
 439                  * lock ordering - the inodelock comes before any dquot lock,
 440                  * and we may drop and reacquire the ilock in xfs_qm_dqget().
 441                  */
 442                 xfs_dqunlock(udqhint);
 443         }
 444
 445         /*
 446          * Find the dquot from somewhere. This bumps the
 447          * reference count of dquot and returns it locked.
 448          * This can return ENOENT if dquot didn't exist on
 449          * disk and we didn't ask it to allocate;
 450          * ESRCH if quotas got turned off suddenly.
 451          */
 452         error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 453                              doalloc | XFS_QMOPT_DOWARN, &dqp);
 454         if (error)
 455                 return error;
 456
 457         trace_xfs_dqattach_get(dqp);
 458
 459         /*
 460          * dqget may have dropped and re-acquired the ilock, but it guarantees
 461          * that the dquot returned is the one that should go in the inode.
 462          */
 463         *IO_idqpp = dqp;
 464         xfs_dqunlock(dqp);
 465         return 0;
 466 }
 467
 468
 469 /*
 470  * Given a udquot and group/project type, attach the group/project
 471  * dquot pointer to the udquot as a hint for future lookups.
 472  */
 473 STATIC void
 474 xfs_qm_dqattach_hint(
 475         struct xfs_inode        *ip,
 476         int                     type)
 477 {
 478         struct xfs_dquot **dqhintp;
 479         struct xfs_dquot *dqp;
 480         struct xfs_dquot *udq = ip->i_udquot;
 481
 482         ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 483
 484         xfs_dqlock(udq);
 485
 486         if (type == XFS_DQ_GROUP) {
 487                 dqp = ip->i_gdquot;
 488                 dqhintp = &udq->q_gdquot;
 489         } else {
 490                 dqp = ip->i_pdquot;
 491                 dqhintp = &udq->q_pdquot;
 492         }
 493
 494         if (*dqhintp) {
 495                 struct xfs_dquot *tmp;
 496
 497                 if (*dqhintp == dqp)
 498                         goto done;
 499
 500                 tmp = *dqhintp;
 501                 *dqhintp = NULL;
 502                 xfs_qm_dqrele(tmp);
 503         }
 504
 505         *dqhintp = xfs_qm_dqhold(dqp);
 506 done:
 507         xfs_dqunlock(udq);
 508 }
 509
 510 static bool
 511 xfs_qm_need_dqattach(
 512         struct xfs_inode        *ip)
 513 {
 514         struct xfs_mount        *mp = ip->i_mount;
 515
 516         if (!XFS_IS_QUOTA_RUNNING(mp))
 517                 return false;
 518         if (!XFS_IS_QUOTA_ON(mp))
 519                 return false;
 520         if (!XFS_NOT_DQATTACHED(mp, ip))
 521                 return false;
 522         if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
 523                 return false;
 524         return true;
 525 }
 526
 527 /*
 528  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 529  * into account.
 530  * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 531  * Inode may get unlocked and relocked in here, and the caller must deal with
 532  * the consequences.
 533  */
 534 int
 535 xfs_qm_dqattach_locked(
 536         xfs_inode_t     *ip,
 537         uint            flags)
 538 {
 539         xfs_mount_t     *mp = ip->i_mount;
 540         uint            nquotas = 0;
 541         int             error = 0;
 542
 543         if (!xfs_qm_need_dqattach(ip))
 544                 return 0;
 545
 546         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 547
 548         if (XFS_IS_UQUOTA_ON(mp)) {
 549                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 550                                                 flags & XFS_QMOPT_DQALLOC,
 551                                                 NULL, &ip->i_udquot);
 552                 if (error)
 553                         goto done;
 554                 nquotas++;
 555         }
 556
 557         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 558         if (XFS_IS_GQUOTA_ON(mp)) {
 559                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 560                                                 flags & XFS_QMOPT_DQALLOC,
 561                                                 ip->i_udquot, &ip->i_gdquot);
 562                 /*
 563                  * Don't worry about the udquot that we may have
 564                  * attached above. It'll get detached, if not already.
 565                  */
 566                 if (error)
 567                         goto done;
 568                 nquotas++;
 569         }
 570
 571         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 572         if (XFS_IS_PQUOTA_ON(mp)) {
 573                 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 574                                                 flags & XFS_QMOPT_DQALLOC,
 575                                                 ip->i_udquot, &ip->i_pdquot);
 576                 /*
 577                  * Don't worry about the udquot that we may have
 578                  * attached above. It'll get detached, if not already.
 579                  */
 580                 if (error)
 581                         goto done;
 582                 nquotas++;
 583         }
 584
 585         /*
 586          * Attach this group/project quota to the user quota as a hint.
 587          * This WON'T, in general, result in a thrash.
 588          */
 589         if (nquotas > 1 && ip->i_udquot) {
 590                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 591                 ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
 592                 ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
 593
 594                 /*
 595                  * We do not have i_udquot locked at this point, but this check
 596                  * is OK since we don't depend on the i_gdquot to be accurate
 597                  * 100% all the time. It is just a hint, and this will
 598                  * succeed in general.
 599                  */
 600                 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 601                         xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
 602
 603                 if (ip->i_udquot->q_pdquot != ip->i_pdquot)
 604                         xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
 605         }
 606
 607  done:
 608 #ifdef DEBUG
 609         if (!error) {
 610                 if (XFS_IS_UQUOTA_ON(mp))
 611                         ASSERT(ip->i_udquot);
 612                 if (XFS_IS_GQUOTA_ON(mp))
 613                         ASSERT(ip->i_gdquot);
 614                 if (XFS_IS_PQUOTA_ON(mp))
 615                         ASSERT(ip->i_pdquot);
 616         }
 617         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 618 #endif
 619         return error;
 620 }
 621
 622 int
 623 xfs_qm_dqattach(
 624         struct xfs_inode        *ip,
 625         uint                    flags)
 626 {
 627         int                     error;
 628
 629         if (!xfs_qm_need_dqattach(ip))
 630                 return 0;
 631
 632         xfs_ilock(ip, XFS_ILOCK_EXCL);
 633         error = xfs_qm_dqattach_locked(ip, flags);
 634         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 635
 636         return error;
 637 }
 638
 639 /*
 640  * Release dquots (and their references) if any.
 641  * The inode should be locked EXCL except when this's called by
 642  * xfs_ireclaim.
 643  */
 644 void
 645 xfs_qm_dqdetach(
 646         xfs_inode_t     *ip)
 647 {
 648         if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
 649                 return;
 650
 651         trace_xfs_dquot_dqdetach(ip);
 652
 653         ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
 654         if (ip->i_udquot) {
 655                 xfs_qm_dqrele(ip->i_udquot);
 656                 ip->i_udquot = NULL;
 657         }
 658         if (ip->i_gdquot) {
 659                 xfs_qm_dqrele(ip->i_gdquot);
 660                 ip->i_gdquot = NULL;
 661         }
 662         if (ip->i_pdquot) {
 663                 xfs_qm_dqrele(ip->i_pdquot);
 664                 ip->i_pdquot = NULL;
 665         }
 666 }
 667
 668 struct xfs_qm_isolate {
 669         struct list_head        buffers;
 670         struct list_head        dispose;
 671 };
 672
 673 static enum lru_status
 674 xfs_qm_dquot_isolate(
 675         struct list_head        *item,
 676         spinlock_t              *lru_lock,
 677         void                    *arg)
 678 {
 679         struct xfs_dquot        *dqp = container_of(item,
 680                                                 struct xfs_dquot, q_lru);
 681         struct xfs_qm_isolate   *isol = arg;
 682
 683         if (!xfs_dqlock_nowait(dqp))
 684                 goto out_miss_busy;
 685
 686         /*
 687          * This dquot has acquired a reference in the meantime remove it from
 688          * the freelist and try again.
 689          */
 690         if (dqp->q_nrefs) {
 691                 xfs_dqunlock(dqp);
 692                 XFS_STATS_INC(xs_qm_dqwants);
 693
 694                 trace_xfs_dqreclaim_want(dqp);
 695                 list_del_init(&dqp->q_lru);
 696                 XFS_STATS_DEC(xs_qm_dquot_unused);
 697                 return LRU_REMOVED;
 698         }
 699
 700         /*
 701          * If the dquot is dirty, flush it. If it's already being flushed, just
 702          * skip it so there is time for the IO to complete before we try to
 703          * reclaim it again on the next LRU pass.
 704          */
 705         if (!xfs_dqflock_nowait(dqp)) {
 706                 xfs_dqunlock(dqp);
 707                 goto out_miss_busy;
 708         }
 709
 710         if (XFS_DQ_IS_DIRTY(dqp)) {
 711                 struct xfs_buf  *bp = NULL;
 712                 int             error;
 713
 714                 trace_xfs_dqreclaim_dirty(dqp);
 715
 716                 /* we have to drop the LRU lock to flush the dquot */
 717                 spin_unlock(lru_lock);
 718
 719                 error = xfs_qm_dqflush(dqp, &bp);
 720                 if (error) {
 721                         xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
 722                                  __func__, dqp);
 723                         goto out_unlock_dirty;
 724                 }
 725
 726                 xfs_buf_delwri_queue(bp, &isol->buffers);
 727                 xfs_buf_relse(bp);
 728                 goto out_unlock_dirty;
 729         }
 730         xfs_dqfunlock(dqp);
 731
 732         /*
 733          * Prevent lookups now that we are past the point of no return.
 734          */
 735         dqp->dq_flags |= XFS_DQ_FREEING;
 736         xfs_dqunlock(dqp);
 737
 738         ASSERT(dqp->q_nrefs == 0);
 739         list_move_tail(&dqp->q_lru, &isol->dispose);
 740         XFS_STATS_DEC(xs_qm_dquot_unused);
 741         trace_xfs_dqreclaim_done(dqp);
 742         XFS_STATS_INC(xs_qm_dqreclaims);
 743         return LRU_REMOVED;
 744
 745 out_miss_busy:
 746         trace_xfs_dqreclaim_busy(dqp);
 747         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 748         return LRU_SKIP;
 749
 750 out_unlock_dirty:
 751         trace_xfs_dqreclaim_busy(dqp);
 752         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 753         xfs_dqunlock(dqp);
 754         spin_lock(lru_lock);
 755         return LRU_RETRY;
 756 }
 757
 758 static unsigned long
 759 xfs_qm_shrink_scan(
 760         struct shrinker         *shrink,
 761         struct shrink_control   *sc)
 762 {
 763         struct xfs_quotainfo    *qi = container_of(shrink,
 764                                         struct xfs_quotainfo, qi_shrinker);
 765         struct xfs_qm_isolate   isol;
 766         unsigned long           freed;
 767         int                     error;
 768         unsigned long           nr_to_scan = sc->nr_to_scan;
 769
 770         if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
 771                 return 0;
 772
 773         INIT_LIST_HEAD(&isol.buffers);
 774         INIT_LIST_HEAD(&isol.dispose);
 775
 776         freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
 777                                         &nr_to_scan);
 778
 779         error = xfs_buf_delwri_submit(&isol.buffers);
 780         if (error)
 781                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
 782
 783         while (!list_empty(&isol.dispose)) {
 784                 struct xfs_dquot        *dqp;
 785
 786                 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
 787                 list_del_init(&dqp->q_lru);
 788                 xfs_qm_dqfree_one(dqp);
 789         }
 790
 791         return freed;
 792 }
 793
 794 static unsigned long
 795 xfs_qm_shrink_count(
 796         struct shrinker         *shrink,
 797         struct shrink_control   *sc)
 798 {
 799         struct xfs_quotainfo    *qi = container_of(shrink,
 800                                         struct xfs_quotainfo, qi_shrinker);
 801
 802         return list_lru_count_node(&qi->qi_lru, sc->nid);
 803 }
 804
 805 /*
 806  * This initializes all the quota information that's kept in the
 807  * mount structure
 808  */
 809 STATIC int
 810 xfs_qm_init_quotainfo(
 811         xfs_mount_t     *mp)
 812 {
 813         xfs_quotainfo_t *qinf;
 814         int             error;
 815         xfs_dquot_t     *dqp;
 816
 817         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 818
 819         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 820
 821         if ((error = list_lru_init(&qinf->qi_lru))) {
 822                 kmem_free(qinf);
 823                 mp->m_quotainfo = NULL;
 824                 return error;
 825         }
 826
 827         /*
 828          * See if quotainodes are setup, and if not, allocate them,
 829          * and change the superblock accordingly.
 830          */
 831         if ((error = xfs_qm_init_quotainos(mp))) {
 832                 list_lru_destroy(&qinf->qi_lru);
 833                 kmem_free(qinf);
 834                 mp->m_quotainfo = NULL;
 835                 return error;
 836         }
 837
 838         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 839         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 840         INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
 841         mutex_init(&qinf->qi_tree_lock);
 842
 843         /* mutex used to serialize quotaoffs */
 844         mutex_init(&qinf->qi_quotaofflock);
 845
 846         /* Precalc some constants */
 847         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 848         qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(mp,
 849                                                         qinf->qi_dqchunklen);
 850
 851         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 852
 853         /*
 854          * We try to get the limits from the superuser's limits fields.
 855          * This is quite hacky, but it is standard quota practice.
 856          *
 857          * We look at the USR dquot with id == 0 first, but if user quotas
 858          * are not enabled we goto the GRP dquot with id == 0.
 859          * We don't really care to keep separate default limits for user
 860          * and group quotas, at least not at this point.
 861          *
 862          * Since we may not have done a quotacheck by this point, just read
 863          * the dquot without attaching it to any hashtables or lists.
 864          */
 865         error = xfs_qm_dqread(mp, 0,
 866                         XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 867                          (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 868                           XFS_DQ_PROJ),
 869                         XFS_QMOPT_DOWARN, &dqp);
 870         if (!error) {
 871                 xfs_disk_dquot_t        *ddqp = &dqp->q_core;
 872
 873                 /*
 874                  * The warnings and timers set the grace period given to
 875                  * a user or group before he or she can not perform any
 876                  * more writing. If it is zero, a default is used.
 877                  */
 878                 qinf->qi_btimelimit = ddqp->d_btimer ?
 879                         be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 880                 qinf->qi_itimelimit = ddqp->d_itimer ?
 881                         be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 882                 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 883                         be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 884                 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 885                         be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 886                 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 887                         be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 888                 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 889                         be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 890                 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 891                 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 892                 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 893                 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 894                 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 895                 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 896
 897                 xfs_qm_dqdestroy(dqp);
 898         } else {
 899                 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 900                 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 901                 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 902                 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 903                 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 904                 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 905         }
 906
 907         qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
 908         qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
 909         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 910         qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
 911         register_shrinker(&qinf->qi_shrinker);
 912         return 0;
 913 }
 914
 915
 916 /*
 917  * Gets called when unmounting a filesystem or when all quotas get
 918  * turned off.
 919  * This purges the quota inodes, destroys locks and frees itself.
 920  */
 921 void
 922 xfs_qm_destroy_quotainfo(
 923         xfs_mount_t     *mp)
 924 {
 925         xfs_quotainfo_t *qi;
 926
 927         qi = mp->m_quotainfo;
 928         ASSERT(qi != NULL);
 929
 930         unregister_shrinker(&qi->qi_shrinker);
 931         list_lru_destroy(&qi->qi_lru);
 932
 933         if (qi->qi_uquotaip) {
 934                 IRELE(qi->qi_uquotaip);
 935                 qi->qi_uquotaip = NULL; /* paranoia */
 936         }
 937         if (qi->qi_gquotaip) {
 938                 IRELE(qi->qi_gquotaip);
 939                 qi->qi_gquotaip = NULL;
 940         }
 941         if (qi->qi_pquotaip) {
 942                 IRELE(qi->qi_pquotaip);
 943                 qi->qi_pquotaip = NULL;
 944         }
 945         mutex_destroy(&qi->qi_quotaofflock);
 946         kmem_free(qi);
 947         mp->m_quotainfo = NULL;
 948 }
 949
 950 /*
 951  * Create an inode and return with a reference already taken, but unlocked
 952  * This is how we create quota inodes
 953  */
 954 STATIC int
 955 xfs_qm_qino_alloc(
 956         xfs_mount_t     *mp,
 957         xfs_inode_t     **ip,
 958         __int64_t       sbfields,
 959         uint            flags)
 960 {
 961         xfs_trans_t     *tp;
 962         int             error;
 963         int             committed;
 964
 965         *ip = NULL;
 966         /*
 967          * With superblock that doesn't have separate pquotino, we
 968          * share an inode between gquota and pquota. If the on-disk
 969          * superblock has GQUOTA and the filesystem is now mounted
 970          * with PQUOTA, just use sb_gquotino for sb_pquotino and
 971          * vice-versa.
 972          */
 973         if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
 974                         (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
 975                 xfs_ino_t ino = NULLFSINO;
 976
 977                 if ((flags & XFS_QMOPT_PQUOTA) &&
 978                              (mp->m_sb.sb_gquotino != NULLFSINO)) {
 979                         ino = mp->m_sb.sb_gquotino;
 980                         ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
 981                 } else if ((flags & XFS_QMOPT_GQUOTA) &&
 982                              (mp->m_sb.sb_pquotino != NULLFSINO)) {
 983                         ino = mp->m_sb.sb_pquotino;
 984                         ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
 985                 }
 986                 if (ino != NULLFSINO) {
 987                         error = xfs_iget(mp, NULL, ino, 0, 0, ip);
 988                         if (error)
 989                                 return error;
 990                         mp->m_sb.sb_gquotino = NULLFSINO;
 991                         mp->m_sb.sb_pquotino = NULLFSINO;
 992                 }
 993         }
 994
 995         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
 996         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
 997                                   XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
 998         if (error) {
 999                 xfs_trans_cancel(tp, 0);
1000                 return error;
1001         }
1002
1003         if (!*ip) {
1004                 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
1005                                                                 &committed);
1006                 if (error) {
1007                         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1008                                          XFS_TRANS_ABORT);
1009                         return error;
1010                 }
1011         }
1012
1013         /*
1014          * Make the changes in the superblock, and log those too.
1015          * sbfields arg may contain fields other than *QUOTINO;
1016          * VERSIONNUM for example.
1017          */
1018         spin_lock(&mp->m_sb_lock);
1019         if (flags & XFS_QMOPT_SBVERSION) {
1020                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1021                 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1022                         XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
1023                                 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1024                                  XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1025                                  XFS_SB_QFLAGS));
1026
1027                 xfs_sb_version_addquota(&mp->m_sb);
1028                 mp->m_sb.sb_uquotino = NULLFSINO;
1029                 mp->m_sb.sb_gquotino = NULLFSINO;
1030                 mp->m_sb.sb_pquotino = NULLFSINO;
1031
1032                 /* qflags will get updated fully _after_ quotacheck */
1033                 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
1034         }
1035         if (flags & XFS_QMOPT_UQUOTA)
1036                 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1037         else if (flags & XFS_QMOPT_GQUOTA)
1038                 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1039         else
1040                 mp->m_sb.sb_pquotino = (*ip)->i_ino;
1041         spin_unlock(&mp->m_sb_lock);
1042         xfs_mod_sb(tp, sbfields);
1043
1044         if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1045                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1046                 return error;
1047         }
1048         return 0;
1049 }
1050
1051
1052 STATIC void
1053 xfs_qm_reset_dqcounts(
1054         xfs_mount_t     *mp,
1055         xfs_buf_t       *bp,
1056         xfs_dqid_t      id,
1057         uint            type)
1058 {
1059         struct xfs_dqblk        *dqb;
1060         int                     j;
1061
1062         trace_xfs_reset_dqcounts(bp, _RET_IP_);
1063
1064         /*
1065          * Reset all counters and timers. They'll be
1066          * started afresh by xfs_qm_quotacheck.
1067          */
1068 #ifdef DEBUG
1069         j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1070         do_div(j, sizeof(xfs_dqblk_t));
1071         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1072 #endif
1073         dqb = bp->b_addr;
1074         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1075                 struct xfs_disk_dquot   *ddq;
1076
1077                 ddq = (struct xfs_disk_dquot *)&dqb[j];
1078
1079                 /*
1080                  * Do a sanity check, and if needed, repair the dqblk. Don't
1081                  * output any warnings because it's perfectly possible to
1082                  * find uninitialised dquot blks. See comment in xfs_dqcheck.
1083                  */
1084                 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1085                             "xfs_quotacheck");
1086                 ddq->d_bcount = 0;
1087                 ddq->d_icount = 0;
1088                 ddq->d_rtbcount = 0;
1089                 ddq->d_btimer = 0;
1090                 ddq->d_itimer = 0;
1091                 ddq->d_rtbtimer = 0;
1092                 ddq->d_bwarns = 0;
1093                 ddq->d_iwarns = 0;
1094                 ddq->d_rtbwarns = 0;
1095
1096                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1097                         xfs_update_cksum((char *)&dqb[j],
1098                                          sizeof(struct xfs_dqblk),
1099                                          XFS_DQUOT_CRC_OFF);
1100                 }
1101         }
1102 }
1103
1104 STATIC int
1105 xfs_qm_dqiter_bufs(
1106         struct xfs_mount        *mp,
1107         xfs_dqid_t              firstid,
1108         xfs_fsblock_t           bno,
1109         xfs_filblks_t           blkcnt,
1110         uint                    flags,
1111         struct list_head        *buffer_list)
1112 {
1113         struct xfs_buf          *bp;
1114         int                     error;
1115         int                     type;
1116
1117         ASSERT(blkcnt > 0);
1118         type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1119                 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1120         error = 0;
1121
1122         /*
1123          * Blkcnt arg can be a very big number, and might even be
1124          * larger than the log itself. So, we have to break it up into
1125          * manageable-sized transactions.
1126          * Note that we don't start a permanent transaction here; we might
1127          * not be able to get a log reservation for the whole thing up front,
1128          * and we don't really care to either, because we just discard
1129          * everything if we were to crash in the middle of this loop.
1130          */
1131         while (blkcnt--) {
1132                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1133                               XFS_FSB_TO_DADDR(mp, bno),
1134                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1135                               &xfs_dquot_buf_ops);
1136
1137                 /*
1138                  * CRC and validation errors will return a EFSCORRUPTED here. If
1139                  * this occurs, re-read without CRC validation so that we can
1140                  * repair the damage via xfs_qm_reset_dqcounts(). This process
1141                  * will leave a trace in the log indicating corruption has
1142                  * been detected.
1143                  */
1144                 if (error == EFSCORRUPTED) {
1145                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1146                                       XFS_FSB_TO_DADDR(mp, bno),
1147                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1148                                       NULL);
1149                 }
1150
1151                 if (error)
1152                         break;
1153
1154                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1155                 xfs_buf_delwri_queue(bp, buffer_list);
1156                 xfs_buf_relse(bp);
1157
1158                 /* goto the next block. */
1159                 bno++;
1160                 firstid += mp->m_quotainfo->qi_dqperchunk;
1161         }
1162
1163         return error;
1164 }
1165
1166 /*
1167  * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1168  * caller supplied function for every chunk of dquots that we find.
1169  */
1170 STATIC int
1171 xfs_qm_dqiterate(
1172         struct xfs_mount        *mp,
1173         struct xfs_inode        *qip,
1174         uint                    flags,
1175         struct list_head        *buffer_list)
1176 {
1177         struct xfs_bmbt_irec    *map;
1178         int                     i, nmaps;       /* number of map entries */
1179         int                     error;          /* return value */
1180         xfs_fileoff_t           lblkno;
1181         xfs_filblks_t           maxlblkcnt;
1182         xfs_dqid_t              firstid;
1183         xfs_fsblock_t           rablkno;
1184         xfs_filblks_t           rablkcnt;
1185
1186         error = 0;
1187         /*
1188          * This looks racy, but we can't keep an inode lock across a
1189          * trans_reserve. But, this gets called during quotacheck, and that
1190          * happens only at mount time which is single threaded.
1191          */
1192         if (qip->i_d.di_nblocks == 0)
1193                 return 0;
1194
1195         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1196
1197         lblkno = 0;
1198         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1199         do {
1200                 nmaps = XFS_DQITER_MAP_SIZE;
1201                 /*
1202                  * We aren't changing the inode itself. Just changing
1203                  * some of its data. No new blocks are added here, and
1204                  * the inode is never added to the transaction.
1205                  */
1206                 xfs_ilock(qip, XFS_ILOCK_SHARED);
1207                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1208                                        map, &nmaps, 0);
1209                 xfs_iunlock(qip, XFS_ILOCK_SHARED);
1210                 if (error)
1211                         break;
1212
1213                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1214                 for (i = 0; i < nmaps; i++) {
1215                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1216                         ASSERT(map[i].br_blockcount);
1217
1218
1219                         lblkno += map[i].br_blockcount;
1220
1221                         if (map[i].br_startblock == HOLESTARTBLOCK)
1222                                 continue;
1223
1224                         firstid = (xfs_dqid_t) map[i].br_startoff *
1225                                 mp->m_quotainfo->qi_dqperchunk;
1226                         /*
1227                          * Do a read-ahead on the next extent.
1228                          */
1229                         if ((i+1 < nmaps) &&
1230                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1231                                 rablkcnt =  map[i+1].br_blockcount;
1232                                 rablkno = map[i+1].br_startblock;
1233                                 while (rablkcnt--) {
1234                                         xfs_buf_readahead(mp->m_ddev_targp,
1235                                                XFS_FSB_TO_DADDR(mp, rablkno),
1236                                                mp->m_quotainfo->qi_dqchunklen,
1237                                                NULL);
1238                                         rablkno++;
1239                                 }
1240                         }
1241                         /*
1242                          * Iterate thru all the blks in the extent and
1243                          * reset the counters of all the dquots inside them.
1244                          */
1245                         error = xfs_qm_dqiter_bufs(mp, firstid,
1246                                                    map[i].br_startblock,
1247                                                    map[i].br_blockcount,
1248                                                    flags, buffer_list);
1249                         if (error)
1250                                 goto out;
1251                 }
1252         } while (nmaps > 0);
1253
1254 out:
1255         kmem_free(map);
1256         return error;
1257 }
1258
1259 /*
1260  * Called by dqusage_adjust in doing a quotacheck.
1261  *
1262  * Given the inode, and a dquot id this updates both the incore dqout as well
1263  * as the buffer copy. This is so that once the quotacheck is done, we can
1264  * just log all the buffers, as opposed to logging numerous updates to
1265  * individual dquots.
1266  */
1267 STATIC int
1268 xfs_qm_quotacheck_dqadjust(
1269         struct xfs_inode        *ip,
1270         xfs_dqid_t              id,
1271         uint                    type,
1272         xfs_qcnt_t              nblks,
1273         xfs_qcnt_t              rtblks)
1274 {
1275         struct xfs_mount        *mp = ip->i_mount;
1276         struct xfs_dquot        *dqp;
1277         int                     error;
1278
1279         error = xfs_qm_dqget(mp, ip, id, type,
1280                              XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1281         if (error) {
1282                 /*
1283                  * Shouldn't be able to turn off quotas here.
1284                  */
1285                 ASSERT(error != ESRCH);
1286                 ASSERT(error != ENOENT);
1287                 return error;
1288         }
1289
1290         trace_xfs_dqadjust(dqp);
1291
1292         /*
1293          * Adjust the inode count and the block count to reflect this inode's
1294          * resource usage.
1295          */
1296         be64_add_cpu(&dqp->q_core.d_icount, 1);
1297         dqp->q_res_icount++;
1298         if (nblks) {
1299                 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1300                 dqp->q_res_bcount += nblks;
1301         }
1302         if (rtblks) {
1303                 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1304                 dqp->q_res_rtbcount += rtblks;
1305         }
1306
1307         /*
1308          * Set default limits, adjust timers (since we changed usages)
1309          *
1310          * There are no timers for the default values set in the root dquot.
1311          */
1312         if (dqp->q_core.d_id) {
1313                 xfs_qm_adjust_dqlimits(mp, dqp);
1314                 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1315         }
1316
1317         dqp->dq_flags |= XFS_DQ_DIRTY;
1318         xfs_qm_dqput(dqp);
1319         return 0;
1320 }
1321
1322 STATIC int
1323 xfs_qm_get_rtblks(
1324         xfs_inode_t     *ip,
1325         xfs_qcnt_t      *O_rtblks)
1326 {
1327         xfs_filblks_t   rtblks;                 /* total rt blks */
1328         xfs_extnum_t    idx;                    /* extent record index */
1329         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1330         xfs_extnum_t    nextents;               /* number of extent entries */
1331         int             error;
1332
1333         ASSERT(XFS_IS_REALTIME_INODE(ip));
1334         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1335         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1336                 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1337                         return error;
1338         }
1339         rtblks = 0;
1340         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1341         for (idx = 0; idx < nextents; idx++)
1342                 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1343         *O_rtblks = (xfs_qcnt_t)rtblks;
1344         return 0;
1345 }
1346
1347 /*
1348  * callback routine supplied to bulkstat(). Given an inumber, find its
1349  * dquots and update them to account for resources taken by that inode.
1350  */
1351 /* ARGSUSED */
1352 STATIC int
1353 xfs_qm_dqusage_adjust(
1354         xfs_mount_t     *mp,            /* mount point for filesystem */
1355         xfs_ino_t       ino,            /* inode number to get data for */
1356         void            __user *buffer, /* not used */
1357         int             ubsize,         /* not used */
1358         int             *ubused,        /* not used */
1359         int             *res)           /* result code value */
1360 {
1361         xfs_inode_t     *ip;
1362         xfs_qcnt_t      nblks, rtblks = 0;
1363         int             error;
1364
1365         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1366
1367         /*
1368          * rootino must have its resources accounted for, not so with the quota
1369          * inodes.
1370          */
1371         if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1372                 *res = BULKSTAT_RV_NOTHING;
1373                 return XFS_ERROR(EINVAL);
1374         }
1375
1376         /*
1377          * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1378          * interface expects the inode to be exclusively locked because that's
1379          * the case in all other instances. It's OK that we do this because
1380          * quotacheck is done only at mount time.
1381          */
1382         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1383         if (error) {
1384                 *res = BULKSTAT_RV_NOTHING;
1385                 return error;
1386         }
1387
1388         ASSERT(ip->i_delayed_blks == 0);
1389
1390         if (XFS_IS_REALTIME_INODE(ip)) {
1391                 /*
1392                  * Walk thru the extent list and count the realtime blocks.
1393                  */
1394                 error = xfs_qm_get_rtblks(ip, &rtblks);
1395                 if (error)
1396                         goto error0;
1397         }
1398
1399         nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1400
1401         /*
1402          * Add the (disk blocks and inode) resources occupied by this
1403          * inode to its dquots. We do this adjustment in the incore dquot,
1404          * and also copy the changes to its buffer.
1405          * We don't care about putting these changes in a transaction
1406          * envelope because if we crash in the middle of a 'quotacheck'
1407          * we have to start from the beginning anyway.
1408          * Once we're done, we'll log all the dquot bufs.
1409          *
1410          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1411          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1412          */
1413         if (XFS_IS_UQUOTA_ON(mp)) {
1414                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1415                                                    XFS_DQ_USER, nblks, rtblks);
1416                 if (error)
1417                         goto error0;
1418         }
1419
1420         if (XFS_IS_GQUOTA_ON(mp)) {
1421                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1422                                                    XFS_DQ_GROUP, nblks, rtblks);
1423                 if (error)
1424                         goto error0;
1425         }
1426
1427         if (XFS_IS_PQUOTA_ON(mp)) {
1428                 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1429                                                    XFS_DQ_PROJ, nblks, rtblks);
1430                 if (error)
1431                         goto error0;
1432         }
1433
1434         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1435         IRELE(ip);
1436         *res = BULKSTAT_RV_DIDONE;
1437         return 0;
1438
1439 error0:
1440         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1441         IRELE(ip);
1442         *res = BULKSTAT_RV_GIVEUP;
1443         return error;
1444 }
1445
1446 STATIC int
1447 xfs_qm_flush_one(
1448         struct xfs_dquot        *dqp,
1449         void                    *data)
1450 {
1451         struct list_head        *buffer_list = data;
1452         struct xfs_buf          *bp = NULL;
1453         int                     error = 0;
1454
1455         xfs_dqlock(dqp);
1456         if (dqp->dq_flags & XFS_DQ_FREEING)
1457                 goto out_unlock;
1458         if (!XFS_DQ_IS_DIRTY(dqp))
1459                 goto out_unlock;
1460
1461         xfs_dqflock(dqp);
1462         error = xfs_qm_dqflush(dqp, &bp);
1463         if (error)
1464                 goto out_unlock;
1465
1466         xfs_buf_delwri_queue(bp, buffer_list);
1467         xfs_buf_relse(bp);
1468 out_unlock:
1469         xfs_dqunlock(dqp);
1470         return error;
1471 }
1472
1473 /*
1474  * Walk thru all the filesystem inodes and construct a consistent view
1475  * of the disk quota world. If the quotacheck fails, disable quotas.
1476  */
1477 int
1478 xfs_qm_quotacheck(
1479         xfs_mount_t     *mp)
1480 {
1481         int                     done, count, error, error2;
1482         xfs_ino_t               lastino;
1483         size_t                  structsz;
1484         uint                    flags;
1485         LIST_HEAD               (buffer_list);
1486         struct xfs_inode        *uip = mp->m_quotainfo->qi_uquotaip;
1487         struct xfs_inode        *gip = mp->m_quotainfo->qi_gquotaip;
1488         struct xfs_inode        *pip = mp->m_quotainfo->qi_pquotaip;
1489
1490         count = INT_MAX;
1491         structsz = 1;
1492         lastino = 0;
1493         flags = 0;
1494
1495         ASSERT(uip || gip || pip);
1496         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1497
1498         xfs_notice(mp, "Quotacheck needed: Please wait.");
1499
1500         /*
1501          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1502          * their counters to zero. We need a clean slate.
1503          * We don't log our changes till later.
1504          */
1505         if (uip) {
1506                 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1507                                          &buffer_list);
1508                 if (error)
1509                         goto error_return;
1510                 flags |= XFS_UQUOTA_CHKD;
1511         }
1512
1513         if (gip) {
1514                 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1515                                          &buffer_list);
1516                 if (error)
1517                         goto error_return;
1518                 flags |= XFS_GQUOTA_CHKD;
1519         }
1520
1521         if (pip) {
1522                 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1523                                          &buffer_list);
1524                 if (error)
1525                         goto error_return;
1526                 flags |= XFS_PQUOTA_CHKD;
1527         }
1528
1529         do {
1530                 /*
1531                  * Iterate thru all the inodes in the file system,
1532                  * adjusting the corresponding dquot counters in core.
1533                  */
1534                 error = xfs_bulkstat(mp, &lastino, &count,
1535                                      xfs_qm_dqusage_adjust,
1536                                      structsz, NULL, &done);
1537                 if (error)
1538                         break;
1539
1540         } while (!done);
1541
1542         /*
1543          * We've made all the changes that we need to make incore.  Flush them
1544          * down to disk buffers if everything was updated successfully.
1545          */
1546         if (XFS_IS_UQUOTA_ON(mp)) {
1547                 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1548                                           &buffer_list);
1549         }
1550         if (XFS_IS_GQUOTA_ON(mp)) {
1551                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1552                                            &buffer_list);
1553                 if (!error)
1554                         error = error2;
1555         }
1556         if (XFS_IS_PQUOTA_ON(mp)) {
1557                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1558                                            &buffer_list);
1559                 if (!error)
1560                         error = error2;
1561         }
1562
1563         error2 = xfs_buf_delwri_submit(&buffer_list);
1564         if (!error)
1565                 error = error2;
1566
1567         /*
1568          * We can get this error if we couldn't do a dquot allocation inside
1569          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1570          * dirty dquots that might be cached, we just want to get rid of them
1571          * and turn quotaoff. The dquots won't be attached to any of the inodes
1572          * at this point (because we intentionally didn't in dqget_noattach).
1573          */
1574         if (error) {
1575                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1576                 goto error_return;
1577         }
1578
1579         /*
1580          * If one type of quotas is off, then it will lose its
1581          * quotachecked status, since we won't be doing accounting for
1582          * that type anymore.
1583          */
1584         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1585         mp->m_qflags |= flags;
1586
1587  error_return:
1588         while (!list_empty(&buffer_list)) {
1589                 struct xfs_buf *bp =
1590                         list_first_entry(&buffer_list, struct xfs_buf, b_list);
1591                 list_del_init(&bp->b_list);
1592                 xfs_buf_relse(bp);
1593         }
1594
1595         if (error) {
1596                 xfs_warn(mp,
1597         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1598                         error);
1599                 /*
1600                  * We must turn off quotas.
1601                  */
1602                 ASSERT(mp->m_quotainfo != NULL);
1603                 xfs_qm_destroy_quotainfo(mp);
1604                 if (xfs_mount_reset_sbqflags(mp)) {
1605                         xfs_warn(mp,
1606                                 "Quotacheck: Failed to reset quota flags.");
1607                 }
1608         } else
1609                 xfs_notice(mp, "Quotacheck: Done.");
1610         return (error);
1611 }
1612
1613 /*
1614  * This is called after the superblock has been read in and we're ready to
1615  * iget the quota inodes.
1616  */
1617 STATIC int
1618 xfs_qm_init_quotainos(
1619         xfs_mount_t     *mp)
1620 {
1621         struct xfs_inode        *uip = NULL;
1622         struct xfs_inode        *gip = NULL;
1623         struct xfs_inode        *pip = NULL;
1624         int                     error;
1625         __int64_t               sbflags = 0;
1626         uint                    flags = 0;
1627
1628         ASSERT(mp->m_quotainfo);
1629
1630         /*
1631          * Get the uquota and gquota inodes
1632          */
1633         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1634                 if (XFS_IS_UQUOTA_ON(mp) &&
1635                     mp->m_sb.sb_uquotino != NULLFSINO) {
1636                         ASSERT(mp->m_sb.sb_uquotino > 0);
1637                         error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1638                                              0, 0, &uip);
1639                         if (error)
1640                                 return XFS_ERROR(error);
1641                 }
1642                 if (XFS_IS_GQUOTA_ON(mp) &&
1643                     mp->m_sb.sb_gquotino != NULLFSINO) {
1644                         ASSERT(mp->m_sb.sb_gquotino > 0);
1645                         error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1646                                              0, 0, &gip);
1647                         if (error)
1648                                 goto error_rele;
1649                 }
1650                 if (XFS_IS_PQUOTA_ON(mp) &&
1651                     mp->m_sb.sb_pquotino != NULLFSINO) {
1652                         ASSERT(mp->m_sb.sb_pquotino > 0);
1653                         error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1654                                              0, 0, &pip);
1655                         if (error)
1656                                 goto error_rele;
1657                 }
1658         } else {
1659                 flags |= XFS_QMOPT_SBVERSION;
1660                 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1661                             XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1662                             XFS_SB_QFLAGS);
1663         }
1664
1665         /*
1666          * Create the three inodes, if they don't exist already. The changes
1667          * made above will get added to a transaction and logged in one of
1668          * the qino_alloc calls below.  If the device is readonly,
1669          * temporarily switch to read-write to do this.
1670          */
1671         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1672                 error = xfs_qm_qino_alloc(mp, &uip,
1673                                               sbflags | XFS_SB_UQUOTINO,
1674                                               flags | XFS_QMOPT_UQUOTA);
1675                 if (error)
1676                         goto error_rele;
1677
1678                 flags &= ~XFS_QMOPT_SBVERSION;
1679         }
1680         if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1681                 error = xfs_qm_qino_alloc(mp, &gip,
1682                                           sbflags | XFS_SB_GQUOTINO,
1683                                           flags | XFS_QMOPT_GQUOTA);
1684                 if (error)
1685                         goto error_rele;
1686
1687                 flags &= ~XFS_QMOPT_SBVERSION;
1688         }
1689         if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1690                 error = xfs_qm_qino_alloc(mp, &pip,
1691                                           sbflags | XFS_SB_PQUOTINO,
1692                                           flags | XFS_QMOPT_PQUOTA);
1693                 if (error)
1694                         goto error_rele;
1695         }
1696
1697         mp->m_quotainfo->qi_uquotaip = uip;
1698         mp->m_quotainfo->qi_gquotaip = gip;
1699         mp->m_quotainfo->qi_pquotaip = pip;
1700
1701         return 0;
1702
1703 error_rele:
1704         if (uip)
1705                 IRELE(uip);
1706         if (gip)
1707                 IRELE(gip);
1708         if (pip)
1709                 IRELE(pip);
1710         return XFS_ERROR(error);
1711 }
1712
1713 STATIC void
1714 xfs_qm_dqfree_one(
1715         struct xfs_dquot        *dqp)
1716 {
1717         struct xfs_mount        *mp = dqp->q_mount;
1718         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1719
1720         mutex_lock(&qi->qi_tree_lock);
1721         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1722                           be32_to_cpu(dqp->q_core.d_id));
1723
1724         qi->qi_dquots--;
1725         mutex_unlock(&qi->qi_tree_lock);
1726
1727         xfs_qm_dqdestroy(dqp);
1728 }
1729
1730 /*
1731  * Start a transaction and write the incore superblock changes to
1732  * disk. flags parameter indicates which fields have changed.
1733  */
1734 int
1735 xfs_qm_write_sb_changes(
1736         xfs_mount_t     *mp,
1737         __int64_t       flags)
1738 {
1739         xfs_trans_t     *tp;
1740         int             error;
1741
1742         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1743         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1744         if (error) {
1745                 xfs_trans_cancel(tp, 0);
1746                 return error;
1747         }
1748
1749         xfs_mod_sb(tp, flags);
1750         error = xfs_trans_commit(tp, 0);
1751
1752         return error;
1753 }
1754
1755
1756 /* --------------- utility functions for vnodeops ---------------- */
1757
1758
1759 /*
1760  * Given an inode, a uid, gid and prid make sure that we have
1761  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1762  * quotas by creating this file.
1763  * This also attaches dquot(s) to the given inode after locking it,
1764  * and returns the dquots corresponding to the uid and/or gid.
1765  *
1766  * in   : inode (unlocked)
1767  * out  : udquot, gdquot with references taken and unlocked
1768  */
1769 int
1770 xfs_qm_vop_dqalloc(
1771         struct xfs_inode        *ip,
1772         xfs_dqid_t              uid,
1773         xfs_dqid_t              gid,
1774         prid_t                  prid,
1775         uint                    flags,
1776         struct xfs_dquot        **O_udqpp,
1777         struct xfs_dquot        **O_gdqpp,
1778         struct xfs_dquot        **O_pdqpp)
1779 {
1780         struct xfs_mount        *mp = ip->i_mount;
1781         struct xfs_dquot        *uq = NULL;
1782         struct xfs_dquot        *gq = NULL;
1783         struct xfs_dquot        *pq = NULL;
1784         int                     error;
1785         uint                    lockflags;
1786
1787         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1788                 return 0;
1789
1790         lockflags = XFS_ILOCK_EXCL;
1791         xfs_ilock(ip, lockflags);
1792
1793         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1794                 gid = ip->i_d.di_gid;
1795
1796         /*
1797          * Attach the dquot(s) to this inode, doing a dquot allocation
1798          * if necessary. The dquot(s) will not be locked.
1799          */
1800         if (XFS_NOT_DQATTACHED(mp, ip)) {
1801                 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1802                 if (error) {
1803                         xfs_iunlock(ip, lockflags);
1804                         return error;
1805                 }
1806         }
1807
1808         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1809                 if (ip->i_d.di_uid != uid) {
1810                         /*
1811                          * What we need is the dquot that has this uid, and
1812                          * if we send the inode to dqget, the uid of the inode
1813                          * takes priority over what's sent in the uid argument.
1814                          * We must unlock inode here before calling dqget if
1815                          * we're not sending the inode, because otherwise
1816                          * we'll deadlock by doing trans_reserve while
1817                          * holding ilock.
1818                          */
1819                         xfs_iunlock(ip, lockflags);
1820                         error = xfs_qm_dqget(mp, NULL, uid,
1821                                                  XFS_DQ_USER,
1822                                                  XFS_QMOPT_DQALLOC |
1823                                                  XFS_QMOPT_DOWARN,
1824                                                  &uq);
1825                         if (error) {
1826                                 ASSERT(error != ENOENT);
1827                                 return error;
1828                         }
1829                         /*
1830                          * Get the ilock in the right order.
1831                          */
1832                         xfs_dqunlock(uq);
1833                         lockflags = XFS_ILOCK_SHARED;
1834                         xfs_ilock(ip, lockflags);
1835                 } else {
1836                         /*
1837                          * Take an extra reference, because we'll return
1838                          * this to caller
1839                          */
1840                         ASSERT(ip->i_udquot);
1841                         uq = xfs_qm_dqhold(ip->i_udquot);
1842                 }
1843         }
1844         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1845                 if (ip->i_d.di_gid != gid) {
1846                         xfs_iunlock(ip, lockflags);
1847                         error = xfs_qm_dqget(mp, NULL, gid,
1848                                                  XFS_DQ_GROUP,
1849                                                  XFS_QMOPT_DQALLOC |
1850                                                  XFS_QMOPT_DOWARN,
1851                                                  &gq);
1852                         if (error) {
1853                                 ASSERT(error != ENOENT);
1854                                 goto error_rele;
1855                         }
1856                         xfs_dqunlock(gq);
1857                         lockflags = XFS_ILOCK_SHARED;
1858                         xfs_ilock(ip, lockflags);
1859                 } else {
1860                         ASSERT(ip->i_gdquot);
1861                         gq = xfs_qm_dqhold(ip->i_gdquot);
1862                 }
1863         }
1864         if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1865                 if (xfs_get_projid(ip) != prid) {
1866                         xfs_iunlock(ip, lockflags);
1867                         error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1868                                                  XFS_DQ_PROJ,
1869                                                  XFS_QMOPT_DQALLOC |
1870                                                  XFS_QMOPT_DOWARN,
1871                                                  &pq);
1872                         if (error) {
1873                                 ASSERT(error != ENOENT);
1874                                 goto error_rele;
1875                         }
1876                         xfs_dqunlock(pq);
1877                         lockflags = XFS_ILOCK_SHARED;
1878                         xfs_ilock(ip, lockflags);
1879                 } else {
1880                         ASSERT(ip->i_pdquot);
1881                         pq = xfs_qm_dqhold(ip->i_pdquot);
1882                 }
1883         }
1884         if (uq)
1885                 trace_xfs_dquot_dqalloc(ip);
1886
1887         xfs_iunlock(ip, lockflags);
1888         if (O_udqpp)
1889                 *O_udqpp = uq;
1890         else if (uq)
1891                 xfs_qm_dqrele(uq);
1892         if (O_gdqpp)
1893                 *O_gdqpp = gq;
1894         else if (gq)
1895                 xfs_qm_dqrele(gq);
1896         if (O_pdqpp)
1897                 *O_pdqpp = pq;
1898         else if (pq)
1899                 xfs_qm_dqrele(pq);
1900         return 0;
1901
1902 error_rele:
1903         if (gq)
1904                 xfs_qm_dqrele(gq);
1905         if (uq)
1906                 xfs_qm_dqrele(uq);
1907         return error;
1908 }
1909
1910 /*
1911  * Actually transfer ownership, and do dquot modifications.
1912  * These were already reserved.
1913  */
1914 xfs_dquot_t *
1915 xfs_qm_vop_chown(
1916         xfs_trans_t     *tp,
1917         xfs_inode_t     *ip,
1918         xfs_dquot_t     **IO_olddq,
1919         xfs_dquot_t     *newdq)
1920 {
1921         xfs_dquot_t     *prevdq;
1922         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1923                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1924
1925
1926         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1927         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1928
1929         /* old dquot */
1930         prevdq = *IO_olddq;
1931         ASSERT(prevdq);
1932         ASSERT(prevdq != newdq);
1933
1934         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1935         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1936
1937         /* the sparkling new dquot */
1938         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1939         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1940
1941         /*
1942          * Take an extra reference, because the inode is going to keep
1943          * this dquot pointer even after the trans_commit.
1944          */
1945         *IO_olddq = xfs_qm_dqhold(newdq);
1946
1947         return prevdq;
1948 }
1949
1950 /*
1951  * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1952  */
1953 int
1954 xfs_qm_vop_chown_reserve(
1955         struct xfs_trans        *tp,
1956         struct xfs_inode        *ip,
1957         struct xfs_dquot        *udqp,
1958         struct xfs_dquot        *gdqp,
1959         struct xfs_dquot        *pdqp,
1960         uint                    flags)
1961 {
1962         struct xfs_mount        *mp = ip->i_mount;
1963         uint                    delblks, blkflags, prjflags = 0;
1964         struct xfs_dquot        *udq_unres = NULL;
1965         struct xfs_dquot        *gdq_unres = NULL;
1966         struct xfs_dquot        *pdq_unres = NULL;
1967         struct xfs_dquot        *udq_delblks = NULL;
1968         struct xfs_dquot        *gdq_delblks = NULL;
1969         struct xfs_dquot        *pdq_delblks = NULL;
1970         int                     error;
1971
1972
1973         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
1974         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1975
1976         delblks = ip->i_delayed_blks;
1977         blkflags = XFS_IS_REALTIME_INODE(ip) ?
1978                         XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
1979
1980         if (XFS_IS_UQUOTA_ON(mp) && udqp &&
1981             ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
1982                 udq_delblks = udqp;
1983                 /*
1984                  * If there are delayed allocation blocks, then we have to
1985                  * unreserve those from the old dquot, and add them to the
1986                  * new dquot.
1987                  */
1988                 if (delblks) {
1989                         ASSERT(ip->i_udquot);
1990                         udq_unres = ip->i_udquot;
1991                 }
1992         }
1993         if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
1994             ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
1995                 gdq_delblks = gdqp;
1996                 if (delblks) {
1997                         ASSERT(ip->i_gdquot);
1998                         gdq_unres = ip->i_gdquot;
1999                 }
2000         }
2001
2002         if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2003             xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2004                 prjflags = XFS_QMOPT_ENOSPC;
2005                 pdq_delblks = pdqp;
2006                 if (delblks) {
2007                         ASSERT(ip->i_pdquot);
2008                         pdq_unres = ip->i_pdquot;
2009                 }
2010         }
2011
2012         error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2013                                 udq_delblks, gdq_delblks, pdq_delblks,
2014                                 ip->i_d.di_nblocks, 1,
2015                                 flags | blkflags | prjflags);
2016         if (error)
2017                 return error;
2018
2019         /*
2020          * Do the delayed blks reservations/unreservations now. Since, these
2021          * are done without the help of a transaction, if a reservation fails
2022          * its previous reservations won't be automatically undone by trans
2023          * code. So, we have to do it manually here.
2024          */
2025         if (delblks) {
2026                 /*
2027                  * Do the reservations first. Unreservation can't fail.
2028                  */
2029                 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2030                 ASSERT(udq_unres || gdq_unres || pdq_unres);
2031                 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2032                             udq_delblks, gdq_delblks, pdq_delblks,
2033                             (xfs_qcnt_t)delblks, 0,
2034                             flags | blkflags | prjflags);
2035                 if (error)
2036                         return error;
2037                 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2038                                 udq_unres, gdq_unres, pdq_unres,
2039                                 -((xfs_qcnt_t)delblks), 0, blkflags);
2040         }
2041
2042         return (0);
2043 }
2044
2045 int
2046 xfs_qm_vop_rename_dqattach(
2047         struct xfs_inode        **i_tab)
2048 {
2049         struct xfs_mount        *mp = i_tab[0]->i_mount;
2050         int                     i;
2051
2052         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2053                 return 0;
2054
2055         for (i = 0; (i < 4 && i_tab[i]); i++) {
2056                 struct xfs_inode        *ip = i_tab[i];
2057                 int                     error;
2058
2059                 /*
2060                  * Watch out for duplicate entries in the table.
2061                  */
2062                 if (i == 0 || ip != i_tab[i-1]) {
2063                         if (XFS_NOT_DQATTACHED(mp, ip)) {
2064                                 error = xfs_qm_dqattach(ip, 0);
2065                                 if (error)
2066                                         return error;
2067                         }
2068                 }
2069         }
2070         return 0;
2071 }
2072
2073 void
2074 xfs_qm_vop_create_dqattach(
2075         struct xfs_trans        *tp,
2076         struct xfs_inode        *ip,
2077         struct xfs_dquot        *udqp,
2078         struct xfs_dquot        *gdqp,
2079         struct xfs_dquot        *pdqp)
2080 {
2081         struct xfs_mount        *mp = tp->t_mountp;
2082
2083         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2084                 return;
2085
2086         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2087         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2088
2089         if (udqp) {
2090                 ASSERT(ip->i_udquot == NULL);
2091                 ASSERT(XFS_IS_UQUOTA_ON(mp));
2092                 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2093
2094                 ip->i_udquot = xfs_qm_dqhold(udqp);
2095                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2096         }
2097         if (gdqp) {
2098                 ASSERT(ip->i_gdquot == NULL);
2099                 ASSERT(XFS_IS_GQUOTA_ON(mp));
2100                 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
2101                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
2102                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2103         }
2104         if (pdqp) {
2105                 ASSERT(ip->i_pdquot == NULL);
2106                 ASSERT(XFS_IS_PQUOTA_ON(mp));
2107                 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2108
2109                 ip->i_pdquot = xfs_qm_dqhold(pdqp);
2110                 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2111         }
2112 }
2113