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_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_inode.h"
  29 #include "xfs_ialloc.h"
  30 #include "xfs_itable.h"
  31 #include "xfs_quota.h"
  32 #include "xfs_error.h"
  33 #include "xfs_bmap.h"
  34 #include "xfs_bmap_btree.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_trans_space.h"
  37 #include "xfs_qm.h"
  38 #include "xfs_trace.h"
  39 #include "xfs_icache.h"
  40 #include "xfs_cksum.h"
  41 #include "xfs_dinode.h"
  42
  43 /*
  44  * The global quota manager. There is only one of these for the entire
  45  * system, _not_ one per file system. XQM keeps track of the overall
  46  * quota functionality, including maintaining the freelist and hash
  47  * tables of dquots.
  48  */
  49 STATIC int      xfs_qm_init_quotainos(xfs_mount_t *);
  50 STATIC int      xfs_qm_init_quotainfo(xfs_mount_t *);
  51
  52
  53 STATIC void     xfs_qm_dqfree_one(struct xfs_dquot *dqp);
  54 /*
  55  * We use the batch lookup interface to iterate over the dquots as it
  56  * currently is the only interface into the radix tree code that allows
  57  * fuzzy lookups instead of exact matches.  Holding the lock over multiple
  58  * operations is fine as all callers are used either during mount/umount
  59  * or quotaoff.
  60  */
  61 #define XFS_DQ_LOOKUP_BATCH     32
  62
  63 STATIC int
  64 xfs_qm_dquot_walk(
  65         struct xfs_mount        *mp,
  66         int                     type,
  67         int                     (*execute)(struct xfs_dquot *dqp, void *data),
  68         void                    *data)
  69 {
  70         struct xfs_quotainfo    *qi = mp->m_quotainfo;
  71         struct radix_tree_root  *tree = xfs_dquot_tree(qi, type);
  72         uint32_t                next_index;
  73         int                     last_error = 0;
  74         int                     skipped;
  75         int                     nr_found;
  76
  77 restart:
  78         skipped = 0;
  79         next_index = 0;
  80         nr_found = 0;
  81
  82         while (1) {
  83                 struct xfs_dquot *batch[XFS_DQ_LOOKUP_BATCH];
  84                 int             error = 0;
  85                 int             i;
  86
  87                 mutex_lock(&qi->qi_tree_lock);
  88                 nr_found = radix_tree_gang_lookup(tree, (void **)batch,
  89                                         next_index, XFS_DQ_LOOKUP_BATCH);
  90                 if (!nr_found) {
  91                         mutex_unlock(&qi->qi_tree_lock);
  92                         break;
  93                 }
  94
  95                 for (i = 0; i < nr_found; i++) {
  96                         struct xfs_dquot *dqp = batch[i];
  97
  98                         next_index = be32_to_cpu(dqp->q_core.d_id) + 1;
  99
 100                         error = execute(batch[i], data);
 101                         if (error == EAGAIN) {
 102                                 skipped++;
 103                                 continue;
 104                         }
 105                         if (error && last_error != EFSCORRUPTED)
 106                                 last_error = error;
 107                 }
 108
 109                 mutex_unlock(&qi->qi_tree_lock);
 110
 111                 /* bail out if the filesystem is corrupted.  */
 112                 if (last_error == EFSCORRUPTED) {
 113                         skipped = 0;
 114                         break;
 115                 }
 116         }
 117
 118         if (skipped) {
 119                 delay(1);
 120                 goto restart;
 121         }
 122
 123         return last_error;
 124 }
 125
 126
 127 /*
 128  * Purge a dquot from all tracking data structures and free it.
 129  */
 130 STATIC int
 131 xfs_qm_dqpurge(
 132         struct xfs_dquot        *dqp,
 133         void                    *data)
 134 {
 135         struct xfs_mount        *mp = dqp->q_mount;
 136         struct xfs_quotainfo    *qi = mp->m_quotainfo;
 137
 138         xfs_dqlock(dqp);
 139         if ((dqp->dq_flags & XFS_DQ_FREEING) || dqp->q_nrefs != 0) {
 140                 xfs_dqunlock(dqp);
 141                 return EAGAIN;
 142         }
 143
 144         dqp->dq_flags |= XFS_DQ_FREEING;
 145
 146         xfs_dqflock(dqp);
 147
 148         /*
 149          * If we are turning this type of quotas off, we don't care
 150          * about the dirty metadata sitting in this dquot. OTOH, if
 151          * we're unmounting, we do care, so we flush it and wait.
 152          */
 153         if (XFS_DQ_IS_DIRTY(dqp)) {
 154                 struct xfs_buf  *bp = NULL;
 155                 int             error;
 156
 157                 /*
 158                  * We don't care about getting disk errors here. We need
 159                  * to purge this dquot anyway, so we go ahead regardless.
 160                  */
 161                 error = xfs_qm_dqflush(dqp, &bp);
 162                 if (error) {
 163                         xfs_warn(mp, "%s: dquot %p flush failed",
 164                                 __func__, dqp);
 165                 } else {
 166                         error = xfs_bwrite(bp);
 167                         xfs_buf_relse(bp);
 168                 }
 169                 xfs_dqflock(dqp);
 170         }
 171
 172         ASSERT(atomic_read(&dqp->q_pincount) == 0);
 173         ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
 174                !(dqp->q_logitem.qli_item.li_flags & XFS_LI_IN_AIL));
 175
 176         xfs_dqfunlock(dqp);
 177         xfs_dqunlock(dqp);
 178
 179         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
 180                           be32_to_cpu(dqp->q_core.d_id));
 181         qi->qi_dquots--;
 182
 183         /*
 184          * We move dquots to the freelist as soon as their reference count
 185          * hits zero, so it really should be on the freelist here.
 186          */
 187         ASSERT(!list_empty(&dqp->q_lru));
 188         list_lru_del(&qi->qi_lru, &dqp->q_lru);
 189         XFS_STATS_DEC(xs_qm_dquot_unused);
 190
 191         xfs_qm_dqdestroy(dqp);
 192         return 0;
 193 }
 194
 195 /*
 196  * Release the group or project dquot pointers the user dquots maybe carrying
 197  * around as a hint, and proceed to purge the user dquot cache if requested.
 198 */
 199 STATIC int
 200 xfs_qm_dqpurge_hints(
 201         struct xfs_dquot        *dqp,
 202         void                    *data)
 203 {
 204         struct xfs_dquot        *gdqp = NULL;
 205         struct xfs_dquot        *pdqp = NULL;
 206         uint                    flags = *((uint *)data);
 207
 208         xfs_dqlock(dqp);
 209         if (dqp->dq_flags & XFS_DQ_FREEING) {
 210                 xfs_dqunlock(dqp);
 211                 return EAGAIN;
 212         }
 213
 214         /* If this quota has a hint attached, prepare for releasing it now */
 215         gdqp = dqp->q_gdquot;
 216         if (gdqp)
 217                 dqp->q_gdquot = NULL;
 218
 219         pdqp = dqp->q_pdquot;
 220         if (pdqp)
 221                 dqp->q_pdquot = NULL;
 222
 223         xfs_dqunlock(dqp);
 224
 225         if (gdqp)
 226                 xfs_qm_dqrele(gdqp);
 227         if (pdqp)
 228                 xfs_qm_dqrele(pdqp);
 229
 230         if (flags & XFS_QMOPT_UQUOTA)
 231                 return xfs_qm_dqpurge(dqp, NULL);
 232
 233         return 0;
 234 }
 235
 236 /*
 237  * Purge the dquot cache.
 238  */
 239 void
 240 xfs_qm_dqpurge_all(
 241         struct xfs_mount        *mp,
 242         uint                    flags)
 243 {
 244         /*
 245          * We have to release group/project dquot hint(s) from the user dquot
 246          * at first if they are there, otherwise we would run into an infinite
 247          * loop while walking through radix tree to purge other type of dquots
 248          * since their refcount is not zero if the user dquot refers to them
 249          * as hint.
 250          *
 251          * Call the special xfs_qm_dqpurge_hints() will end up go through the
 252          * general xfs_qm_dqpurge() against user dquot cache if requested.
 253          */
 254         xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_dqpurge_hints, &flags);
 255
 256         if (flags & XFS_QMOPT_GQUOTA)
 257                 xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_dqpurge, NULL);
 258         if (flags & XFS_QMOPT_PQUOTA)
 259                 xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_dqpurge, NULL);
 260 }
 261
 262 /*
 263  * Just destroy the quotainfo structure.
 264  */
 265 void
 266 xfs_qm_unmount(
 267         struct xfs_mount        *mp)
 268 {
 269         if (mp->m_quotainfo) {
 270                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
 271                 xfs_qm_destroy_quotainfo(mp);
 272         }
 273 }
 274
 275
 276 /*
 277  * This is called from xfs_mountfs to start quotas and initialize all
 278  * necessary data structures like quotainfo.  This is also responsible for
 279  * running a quotacheck as necessary.  We are guaranteed that the superblock
 280  * is consistently read in at this point.
 281  *
 282  * If we fail here, the mount will continue with quota turned off. We don't
 283  * need to inidicate success or failure at all.
 284  */
 285 void
 286 xfs_qm_mount_quotas(
 287         xfs_mount_t     *mp)
 288 {
 289         int             error = 0;
 290         uint            sbf;
 291
 292         /*
 293          * If quotas on realtime volumes is not supported, we disable
 294          * quotas immediately.
 295          */
 296         if (mp->m_sb.sb_rextents) {
 297                 xfs_notice(mp, "Cannot turn on quotas for realtime filesystem");
 298                 mp->m_qflags = 0;
 299                 goto write_changes;
 300         }
 301
 302         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 303
 304         /*
 305          * Allocate the quotainfo structure inside the mount struct, and
 306          * create quotainode(s), and change/rev superblock if necessary.
 307          */
 308         error = xfs_qm_init_quotainfo(mp);
 309         if (error) {
 310                 /*
 311                  * We must turn off quotas.
 312                  */
 313                 ASSERT(mp->m_quotainfo == NULL);
 314                 mp->m_qflags = 0;
 315                 goto write_changes;
 316         }
 317         /*
 318          * If any of the quotas are not consistent, do a quotacheck.
 319          */
 320         if (XFS_QM_NEED_QUOTACHECK(mp)) {
 321                 error = xfs_qm_quotacheck(mp);
 322                 if (error) {
 323                         /* Quotacheck failed and disabled quotas. */
 324                         return;
 325                 }
 326         }
 327         /*
 328          * If one type of quotas is off, then it will lose its
 329          * quotachecked status, since we won't be doing accounting for
 330          * that type anymore.
 331          */
 332         if (!XFS_IS_UQUOTA_ON(mp))
 333                 mp->m_qflags &= ~XFS_UQUOTA_CHKD;
 334         if (!XFS_IS_GQUOTA_ON(mp))
 335                 mp->m_qflags &= ~XFS_GQUOTA_CHKD;
 336         if (!XFS_IS_PQUOTA_ON(mp))
 337                 mp->m_qflags &= ~XFS_PQUOTA_CHKD;
 338
 339  write_changes:
 340         /*
 341          * We actually don't have to acquire the m_sb_lock at all.
 342          * This can only be called from mount, and that's single threaded. XXX
 343          */
 344         spin_lock(&mp->m_sb_lock);
 345         sbf = mp->m_sb.sb_qflags;
 346         mp->m_sb.sb_qflags = mp->m_qflags & XFS_MOUNT_QUOTA_ALL;
 347         spin_unlock(&mp->m_sb_lock);
 348
 349         if (sbf != (mp->m_qflags & XFS_MOUNT_QUOTA_ALL)) {
 350                 if (xfs_qm_write_sb_changes(mp, XFS_SB_QFLAGS)) {
 351                         /*
 352                          * We could only have been turning quotas off.
 353                          * We aren't in very good shape actually because
 354                          * the incore structures are convinced that quotas are
 355                          * off, but the on disk superblock doesn't know that !
 356                          */
 357                         ASSERT(!(XFS_IS_QUOTA_RUNNING(mp)));
 358                         xfs_alert(mp, "%s: Superblock update failed!",
 359                                 __func__);
 360                 }
 361         }
 362
 363         if (error) {
 364                 xfs_warn(mp, "Failed to initialize disk quotas.");
 365                 return;
 366         }
 367 }
 368
 369 /*
 370  * Called from the vfsops layer.
 371  */
 372 void
 373 xfs_qm_unmount_quotas(
 374         xfs_mount_t     *mp)
 375 {
 376         /*
 377          * Release the dquots that root inode, et al might be holding,
 378          * before we flush quotas and blow away the quotainfo structure.
 379          */
 380         ASSERT(mp->m_rootip);
 381         xfs_qm_dqdetach(mp->m_rootip);
 382         if (mp->m_rbmip)
 383                 xfs_qm_dqdetach(mp->m_rbmip);
 384         if (mp->m_rsumip)
 385                 xfs_qm_dqdetach(mp->m_rsumip);
 386
 387         /*
 388          * Release the quota inodes.
 389          */
 390         if (mp->m_quotainfo) {
 391                 if (mp->m_quotainfo->qi_uquotaip) {
 392                         IRELE(mp->m_quotainfo->qi_uquotaip);
 393                         mp->m_quotainfo->qi_uquotaip = NULL;
 394                 }
 395                 if (mp->m_quotainfo->qi_gquotaip) {
 396                         IRELE(mp->m_quotainfo->qi_gquotaip);
 397                         mp->m_quotainfo->qi_gquotaip = NULL;
 398                 }
 399                 if (mp->m_quotainfo->qi_pquotaip) {
 400                         IRELE(mp->m_quotainfo->qi_pquotaip);
 401                         mp->m_quotainfo->qi_pquotaip = NULL;
 402                 }
 403         }
 404 }
 405
 406 STATIC int
 407 xfs_qm_dqattach_one(
 408         xfs_inode_t     *ip,
 409         xfs_dqid_t      id,
 410         uint            type,
 411         uint            doalloc,
 412         xfs_dquot_t     *udqhint, /* hint */
 413         xfs_dquot_t     **IO_idqpp)
 414 {
 415         xfs_dquot_t     *dqp;
 416         int             error;
 417
 418         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 419         error = 0;
 420
 421         /*
 422          * See if we already have it in the inode itself. IO_idqpp is
 423          * &i_udquot or &i_gdquot. This made the code look weird, but
 424          * made the logic a lot simpler.
 425          */
 426         dqp = *IO_idqpp;
 427         if (dqp) {
 428                 trace_xfs_dqattach_found(dqp);
 429                 return 0;
 430         }
 431
 432         /*
 433          * udqhint is the i_udquot field in inode, and is non-NULL only
 434          * when the type arg is group/project. Its purpose is to save a
 435          * lookup by dqid (xfs_qm_dqget) by caching a group dquot inside
 436          * the user dquot.
 437          */
 438         if (udqhint) {
 439                 ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 440                 xfs_dqlock(udqhint);
 441
 442                 /*
 443                  * No need to take dqlock to look at the id.
 444                  *
 445                  * The ID can't change until it gets reclaimed, and it won't
 446                  * be reclaimed as long as we have a ref from inode and we
 447                  * hold the ilock.
 448                  */
 449                 if (type == XFS_DQ_GROUP)
 450                         dqp = udqhint->q_gdquot;
 451                 else
 452                         dqp = udqhint->q_pdquot;
 453                 if (dqp && be32_to_cpu(dqp->q_core.d_id) == id) {
 454                         ASSERT(*IO_idqpp == NULL);
 455
 456                         *IO_idqpp = xfs_qm_dqhold(dqp);
 457                         xfs_dqunlock(udqhint);
 458                         return 0;
 459                 }
 460
 461                 /*
 462                  * We can't hold a dquot lock when we call the dqget code.
 463                  * We'll deadlock in no time, because of (not conforming to)
 464                  * lock ordering - the inodelock comes before any dquot lock,
 465                  * and we may drop and reacquire the ilock in xfs_qm_dqget().
 466                  */
 467                 xfs_dqunlock(udqhint);
 468         }
 469
 470         /*
 471          * Find the dquot from somewhere. This bumps the
 472          * reference count of dquot and returns it locked.
 473          * This can return ENOENT if dquot didn't exist on
 474          * disk and we didn't ask it to allocate;
 475          * ESRCH if quotas got turned off suddenly.
 476          */
 477         error = xfs_qm_dqget(ip->i_mount, ip, id, type,
 478                              doalloc | XFS_QMOPT_DOWARN, &dqp);
 479         if (error)
 480                 return error;
 481
 482         trace_xfs_dqattach_get(dqp);
 483
 484         /*
 485          * dqget may have dropped and re-acquired the ilock, but it guarantees
 486          * that the dquot returned is the one that should go in the inode.
 487          */
 488         *IO_idqpp = dqp;
 489         xfs_dqunlock(dqp);
 490         return 0;
 491 }
 492
 493
 494 /*
 495  * Given a udquot and group/project type, attach the group/project
 496  * dquot pointer to the udquot as a hint for future lookups.
 497  */
 498 STATIC void
 499 xfs_qm_dqattach_hint(
 500         struct xfs_inode        *ip,
 501         int                     type)
 502 {
 503         struct xfs_dquot **dqhintp;
 504         struct xfs_dquot *dqp;
 505         struct xfs_dquot *udq = ip->i_udquot;
 506
 507         ASSERT(type == XFS_DQ_GROUP || type == XFS_DQ_PROJ);
 508
 509         xfs_dqlock(udq);
 510
 511         if (type == XFS_DQ_GROUP) {
 512                 dqp = ip->i_gdquot;
 513                 dqhintp = &udq->q_gdquot;
 514         } else {
 515                 dqp = ip->i_pdquot;
 516                 dqhintp = &udq->q_pdquot;
 517         }
 518
 519         if (*dqhintp) {
 520                 struct xfs_dquot *tmp;
 521
 522                 if (*dqhintp == dqp)
 523                         goto done;
 524
 525                 tmp = *dqhintp;
 526                 *dqhintp = NULL;
 527                 xfs_qm_dqrele(tmp);
 528         }
 529
 530         *dqhintp = xfs_qm_dqhold(dqp);
 531 done:
 532         xfs_dqunlock(udq);
 533 }
 534
 535 static bool
 536 xfs_qm_need_dqattach(
 537         struct xfs_inode        *ip)
 538 {
 539         struct xfs_mount        *mp = ip->i_mount;
 540
 541         if (!XFS_IS_QUOTA_RUNNING(mp))
 542                 return false;
 543         if (!XFS_IS_QUOTA_ON(mp))
 544                 return false;
 545         if (!XFS_NOT_DQATTACHED(mp, ip))
 546                 return false;
 547         if (xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
 548                 return false;
 549         return true;
 550 }
 551
 552 /*
 553  * Given a locked inode, attach dquot(s) to it, taking U/G/P-QUOTAON
 554  * into account.
 555  * If XFS_QMOPT_DQALLOC, the dquot(s) will be allocated if needed.
 556  * Inode may get unlocked and relocked in here, and the caller must deal with
 557  * the consequences.
 558  */
 559 int
 560 xfs_qm_dqattach_locked(
 561         xfs_inode_t     *ip,
 562         uint            flags)
 563 {
 564         xfs_mount_t     *mp = ip->i_mount;
 565         uint            nquotas = 0;
 566         int             error = 0;
 567
 568         if (!xfs_qm_need_dqattach(ip))
 569                 return 0;
 570
 571         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 572
 573         if (XFS_IS_UQUOTA_ON(mp)) {
 574                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_uid, XFS_DQ_USER,
 575                                                 flags & XFS_QMOPT_DQALLOC,
 576                                                 NULL, &ip->i_udquot);
 577                 if (error)
 578                         goto done;
 579                 nquotas++;
 580         }
 581
 582         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 583         if (XFS_IS_GQUOTA_ON(mp)) {
 584                 error = xfs_qm_dqattach_one(ip, ip->i_d.di_gid, XFS_DQ_GROUP,
 585                                                 flags & XFS_QMOPT_DQALLOC,
 586                                                 ip->i_udquot, &ip->i_gdquot);
 587                 /*
 588                  * Don't worry about the udquot that we may have
 589                  * attached above. It'll get detached, if not already.
 590                  */
 591                 if (error)
 592                         goto done;
 593                 nquotas++;
 594         }
 595
 596         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 597         if (XFS_IS_PQUOTA_ON(mp)) {
 598                 error = xfs_qm_dqattach_one(ip, xfs_get_projid(ip), XFS_DQ_PROJ,
 599                                                 flags & XFS_QMOPT_DQALLOC,
 600                                                 ip->i_udquot, &ip->i_pdquot);
 601                 /*
 602                  * Don't worry about the udquot that we may have
 603                  * attached above. It'll get detached, if not already.
 604                  */
 605                 if (error)
 606                         goto done;
 607                 nquotas++;
 608         }
 609
 610         /*
 611          * Attach this group/project quota to the user quota as a hint.
 612          * This WON'T, in general, result in a thrash.
 613          */
 614         if (nquotas > 1 && ip->i_udquot) {
 615                 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 616                 ASSERT(ip->i_gdquot || !XFS_IS_GQUOTA_ON(mp));
 617                 ASSERT(ip->i_pdquot || !XFS_IS_PQUOTA_ON(mp));
 618
 619                 /*
 620                  * We do not have i_udquot locked at this point, but this check
 621                  * is OK since we don't depend on the i_gdquot to be accurate
 622                  * 100% all the time. It is just a hint, and this will
 623                  * succeed in general.
 624                  */
 625                 if (ip->i_udquot->q_gdquot != ip->i_gdquot)
 626                         xfs_qm_dqattach_hint(ip, XFS_DQ_GROUP);
 627
 628                 if (ip->i_udquot->q_pdquot != ip->i_pdquot)
 629                         xfs_qm_dqattach_hint(ip, XFS_DQ_PROJ);
 630         }
 631
 632  done:
 633 #ifdef DEBUG
 634         if (!error) {
 635                 if (XFS_IS_UQUOTA_ON(mp))
 636                         ASSERT(ip->i_udquot);
 637                 if (XFS_IS_GQUOTA_ON(mp))
 638                         ASSERT(ip->i_gdquot);
 639                 if (XFS_IS_PQUOTA_ON(mp))
 640                         ASSERT(ip->i_pdquot);
 641         }
 642         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 643 #endif
 644         return error;
 645 }
 646
 647 int
 648 xfs_qm_dqattach(
 649         struct xfs_inode        *ip,
 650         uint                    flags)
 651 {
 652         int                     error;
 653
 654         if (!xfs_qm_need_dqattach(ip))
 655                 return 0;
 656
 657         xfs_ilock(ip, XFS_ILOCK_EXCL);
 658         error = xfs_qm_dqattach_locked(ip, flags);
 659         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 660
 661         return error;
 662 }
 663
 664 /*
 665  * Release dquots (and their references) if any.
 666  * The inode should be locked EXCL except when this's called by
 667  * xfs_ireclaim.
 668  */
 669 void
 670 xfs_qm_dqdetach(
 671         xfs_inode_t     *ip)
 672 {
 673         if (!(ip->i_udquot || ip->i_gdquot || ip->i_pdquot))
 674                 return;
 675
 676         trace_xfs_dquot_dqdetach(ip);
 677
 678         ASSERT(!xfs_is_quota_inode(&ip->i_mount->m_sb, ip->i_ino));
 679         if (ip->i_udquot) {
 680                 xfs_qm_dqrele(ip->i_udquot);
 681                 ip->i_udquot = NULL;
 682         }
 683         if (ip->i_gdquot) {
 684                 xfs_qm_dqrele(ip->i_gdquot);
 685                 ip->i_gdquot = NULL;
 686         }
 687         if (ip->i_pdquot) {
 688                 xfs_qm_dqrele(ip->i_pdquot);
 689                 ip->i_pdquot = NULL;
 690         }
 691 }
 692
 693 struct xfs_qm_isolate {
 694         struct list_head        buffers;
 695         struct list_head        dispose;
 696 };
 697
 698 static enum lru_status
 699 xfs_qm_dquot_isolate(
 700         struct list_head        *item,
 701         spinlock_t              *lru_lock,
 702         void                    *arg)
 703 {
 704         struct xfs_dquot        *dqp = container_of(item,
 705                                                 struct xfs_dquot, q_lru);
 706         struct xfs_qm_isolate   *isol = arg;
 707
 708         if (!xfs_dqlock_nowait(dqp))
 709                 goto out_miss_busy;
 710
 711         /*
 712          * This dquot has acquired a reference in the meantime remove it from
 713          * the freelist and try again.
 714          */
 715         if (dqp->q_nrefs) {
 716                 xfs_dqunlock(dqp);
 717                 XFS_STATS_INC(xs_qm_dqwants);
 718
 719                 trace_xfs_dqreclaim_want(dqp);
 720                 list_del_init(&dqp->q_lru);
 721                 XFS_STATS_DEC(xs_qm_dquot_unused);
 722                 return LRU_REMOVED;
 723         }
 724
 725         /*
 726          * If the dquot is dirty, flush it. If it's already being flushed, just
 727          * skip it so there is time for the IO to complete before we try to
 728          * reclaim it again on the next LRU pass.
 729          */
 730         if (!xfs_dqflock_nowait(dqp)) {
 731                 xfs_dqunlock(dqp);
 732                 goto out_miss_busy;
 733         }
 734
 735         if (XFS_DQ_IS_DIRTY(dqp)) {
 736                 struct xfs_buf  *bp = NULL;
 737                 int             error;
 738
 739                 trace_xfs_dqreclaim_dirty(dqp);
 740
 741                 /* we have to drop the LRU lock to flush the dquot */
 742                 spin_unlock(lru_lock);
 743
 744                 error = xfs_qm_dqflush(dqp, &bp);
 745                 if (error) {
 746                         xfs_warn(dqp->q_mount, "%s: dquot %p flush failed",
 747                                  __func__, dqp);
 748                         goto out_unlock_dirty;
 749                 }
 750
 751                 xfs_buf_delwri_queue(bp, &isol->buffers);
 752                 xfs_buf_relse(bp);
 753                 goto out_unlock_dirty;
 754         }
 755         xfs_dqfunlock(dqp);
 756
 757         /*
 758          * Prevent lookups now that we are past the point of no return.
 759          */
 760         dqp->dq_flags |= XFS_DQ_FREEING;
 761         xfs_dqunlock(dqp);
 762
 763         ASSERT(dqp->q_nrefs == 0);
 764         list_move_tail(&dqp->q_lru, &isol->dispose);
 765         XFS_STATS_DEC(xs_qm_dquot_unused);
 766         trace_xfs_dqreclaim_done(dqp);
 767         XFS_STATS_INC(xs_qm_dqreclaims);
 768         return LRU_REMOVED;
 769
 770 out_miss_busy:
 771         trace_xfs_dqreclaim_busy(dqp);
 772         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 773         return LRU_SKIP;
 774
 775 out_unlock_dirty:
 776         trace_xfs_dqreclaim_busy(dqp);
 777         XFS_STATS_INC(xs_qm_dqreclaim_misses);
 778         xfs_dqunlock(dqp);
 779         spin_lock(lru_lock);
 780         return LRU_RETRY;
 781 }
 782
 783 static unsigned long
 784 xfs_qm_shrink_scan(
 785         struct shrinker         *shrink,
 786         struct shrink_control   *sc)
 787 {
 788         struct xfs_quotainfo    *qi = container_of(shrink,
 789                                         struct xfs_quotainfo, qi_shrinker);
 790         struct xfs_qm_isolate   isol;
 791         unsigned long           freed;
 792         int                     error;
 793         unsigned long           nr_to_scan = sc->nr_to_scan;
 794
 795         if ((sc->gfp_mask & (__GFP_FS|__GFP_WAIT)) != (__GFP_FS|__GFP_WAIT))
 796                 return 0;
 797
 798         INIT_LIST_HEAD(&isol.buffers);
 799         INIT_LIST_HEAD(&isol.dispose);
 800
 801         freed = list_lru_walk_node(&qi->qi_lru, sc->nid, xfs_qm_dquot_isolate, &isol,
 802                                         &nr_to_scan);
 803
 804         error = xfs_buf_delwri_submit(&isol.buffers);
 805         if (error)
 806                 xfs_warn(NULL, "%s: dquot reclaim failed", __func__);
 807
 808         while (!list_empty(&isol.dispose)) {
 809                 struct xfs_dquot        *dqp;
 810
 811                 dqp = list_first_entry(&isol.dispose, struct xfs_dquot, q_lru);
 812                 list_del_init(&dqp->q_lru);
 813                 xfs_qm_dqfree_one(dqp);
 814         }
 815
 816         return freed;
 817 }
 818
 819 static unsigned long
 820 xfs_qm_shrink_count(
 821         struct shrinker         *shrink,
 822         struct shrink_control   *sc)
 823 {
 824         struct xfs_quotainfo    *qi = container_of(shrink,
 825                                         struct xfs_quotainfo, qi_shrinker);
 826
 827         return list_lru_count_node(&qi->qi_lru, sc->nid);
 828 }
 829
 830 /*
 831  * This initializes all the quota information that's kept in the
 832  * mount structure
 833  */
 834 STATIC int
 835 xfs_qm_init_quotainfo(
 836         xfs_mount_t     *mp)
 837 {
 838         xfs_quotainfo_t *qinf;
 839         int             error;
 840         xfs_dquot_t     *dqp;
 841
 842         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
 843
 844         qinf = mp->m_quotainfo = kmem_zalloc(sizeof(xfs_quotainfo_t), KM_SLEEP);
 845
 846         if ((error = list_lru_init(&qinf->qi_lru))) {
 847                 kmem_free(qinf);
 848                 mp->m_quotainfo = NULL;
 849                 return error;
 850         }
 851
 852         /*
 853          * See if quotainodes are setup, and if not, allocate them,
 854          * and change the superblock accordingly.
 855          */
 856         if ((error = xfs_qm_init_quotainos(mp))) {
 857                 list_lru_destroy(&qinf->qi_lru);
 858                 kmem_free(qinf);
 859                 mp->m_quotainfo = NULL;
 860                 return error;
 861         }
 862
 863         INIT_RADIX_TREE(&qinf->qi_uquota_tree, GFP_NOFS);
 864         INIT_RADIX_TREE(&qinf->qi_gquota_tree, GFP_NOFS);
 865         INIT_RADIX_TREE(&qinf->qi_pquota_tree, GFP_NOFS);
 866         mutex_init(&qinf->qi_tree_lock);
 867
 868         /* mutex used to serialize quotaoffs */
 869         mutex_init(&qinf->qi_quotaofflock);
 870
 871         /* Precalc some constants */
 872         qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 873         qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(mp,
 874                                                         qinf->qi_dqchunklen);
 875
 876         mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 877
 878         /*
 879          * We try to get the limits from the superuser's limits fields.
 880          * This is quite hacky, but it is standard quota practice.
 881          *
 882          * We look at the USR dquot with id == 0 first, but if user quotas
 883          * are not enabled we goto the GRP dquot with id == 0.
 884          * We don't really care to keep separate default limits for user
 885          * and group quotas, at least not at this point.
 886          *
 887          * Since we may not have done a quotacheck by this point, just read
 888          * the dquot without attaching it to any hashtables or lists.
 889          */
 890         error = xfs_qm_dqread(mp, 0,
 891                         XFS_IS_UQUOTA_RUNNING(mp) ? XFS_DQ_USER :
 892                          (XFS_IS_GQUOTA_RUNNING(mp) ? XFS_DQ_GROUP :
 893                           XFS_DQ_PROJ),
 894                         XFS_QMOPT_DOWARN, &dqp);
 895         if (!error) {
 896                 xfs_disk_dquot_t        *ddqp = &dqp->q_core;
 897
 898                 /*
 899                  * The warnings and timers set the grace period given to
 900                  * a user or group before he or she can not perform any
 901                  * more writing. If it is zero, a default is used.
 902                  */
 903                 qinf->qi_btimelimit = ddqp->d_btimer ?
 904                         be32_to_cpu(ddqp->d_btimer) : XFS_QM_BTIMELIMIT;
 905                 qinf->qi_itimelimit = ddqp->d_itimer ?
 906                         be32_to_cpu(ddqp->d_itimer) : XFS_QM_ITIMELIMIT;
 907                 qinf->qi_rtbtimelimit = ddqp->d_rtbtimer ?
 908                         be32_to_cpu(ddqp->d_rtbtimer) : XFS_QM_RTBTIMELIMIT;
 909                 qinf->qi_bwarnlimit = ddqp->d_bwarns ?
 910                         be16_to_cpu(ddqp->d_bwarns) : XFS_QM_BWARNLIMIT;
 911                 qinf->qi_iwarnlimit = ddqp->d_iwarns ?
 912                         be16_to_cpu(ddqp->d_iwarns) : XFS_QM_IWARNLIMIT;
 913                 qinf->qi_rtbwarnlimit = ddqp->d_rtbwarns ?
 914                         be16_to_cpu(ddqp->d_rtbwarns) : XFS_QM_RTBWARNLIMIT;
 915                 qinf->qi_bhardlimit = be64_to_cpu(ddqp->d_blk_hardlimit);
 916                 qinf->qi_bsoftlimit = be64_to_cpu(ddqp->d_blk_softlimit);
 917                 qinf->qi_ihardlimit = be64_to_cpu(ddqp->d_ino_hardlimit);
 918                 qinf->qi_isoftlimit = be64_to_cpu(ddqp->d_ino_softlimit);
 919                 qinf->qi_rtbhardlimit = be64_to_cpu(ddqp->d_rtb_hardlimit);
 920                 qinf->qi_rtbsoftlimit = be64_to_cpu(ddqp->d_rtb_softlimit);
 921
 922                 xfs_qm_dqdestroy(dqp);
 923         } else {
 924                 qinf->qi_btimelimit = XFS_QM_BTIMELIMIT;
 925                 qinf->qi_itimelimit = XFS_QM_ITIMELIMIT;
 926                 qinf->qi_rtbtimelimit = XFS_QM_RTBTIMELIMIT;
 927                 qinf->qi_bwarnlimit = XFS_QM_BWARNLIMIT;
 928                 qinf->qi_iwarnlimit = XFS_QM_IWARNLIMIT;
 929                 qinf->qi_rtbwarnlimit = XFS_QM_RTBWARNLIMIT;
 930         }
 931
 932         qinf->qi_shrinker.count_objects = xfs_qm_shrink_count;
 933         qinf->qi_shrinker.scan_objects = xfs_qm_shrink_scan;
 934         qinf->qi_shrinker.seeks = DEFAULT_SEEKS;
 935         qinf->qi_shrinker.flags = SHRINKER_NUMA_AWARE;
 936         register_shrinker(&qinf->qi_shrinker);
 937         return 0;
 938 }
 939
 940
 941 /*
 942  * Gets called when unmounting a filesystem or when all quotas get
 943  * turned off.
 944  * This purges the quota inodes, destroys locks and frees itself.
 945  */
 946 void
 947 xfs_qm_destroy_quotainfo(
 948         xfs_mount_t     *mp)
 949 {
 950         xfs_quotainfo_t *qi;
 951
 952         qi = mp->m_quotainfo;
 953         ASSERT(qi != NULL);
 954
 955         unregister_shrinker(&qi->qi_shrinker);
 956         list_lru_destroy(&qi->qi_lru);
 957
 958         if (qi->qi_uquotaip) {
 959                 IRELE(qi->qi_uquotaip);
 960                 qi->qi_uquotaip = NULL; /* paranoia */
 961         }
 962         if (qi->qi_gquotaip) {
 963                 IRELE(qi->qi_gquotaip);
 964                 qi->qi_gquotaip = NULL;
 965         }
 966         if (qi->qi_pquotaip) {
 967                 IRELE(qi->qi_pquotaip);
 968                 qi->qi_pquotaip = NULL;
 969         }
 970         mutex_destroy(&qi->qi_quotaofflock);
 971         kmem_free(qi);
 972         mp->m_quotainfo = NULL;
 973 }
 974
 975 /*
 976  * Create an inode and return with a reference already taken, but unlocked
 977  * This is how we create quota inodes
 978  */
 979 STATIC int
 980 xfs_qm_qino_alloc(
 981         xfs_mount_t     *mp,
 982         xfs_inode_t     **ip,
 983         __int64_t       sbfields,
 984         uint            flags)
 985 {
 986         xfs_trans_t     *tp;
 987         int             error;
 988         int             committed;
 989
 990         *ip = NULL;
 991         /*
 992          * With superblock that doesn't have separate pquotino, we
 993          * share an inode between gquota and pquota. If the on-disk
 994          * superblock has GQUOTA and the filesystem is now mounted
 995          * with PQUOTA, just use sb_gquotino for sb_pquotino and
 996          * vice-versa.
 997          */
 998         if (!xfs_sb_version_has_pquotino(&mp->m_sb) &&
 999                         (flags & (XFS_QMOPT_PQUOTA|XFS_QMOPT_GQUOTA))) {
1000                 xfs_ino_t ino = NULLFSINO;
1001
1002                 if ((flags & XFS_QMOPT_PQUOTA) &&
1003                              (mp->m_sb.sb_gquotino != NULLFSINO)) {
1004                         ino = mp->m_sb.sb_gquotino;
1005                         ASSERT(mp->m_sb.sb_pquotino == NULLFSINO);
1006                 } else if ((flags & XFS_QMOPT_GQUOTA) &&
1007                              (mp->m_sb.sb_pquotino != NULLFSINO)) {
1008                         ino = mp->m_sb.sb_pquotino;
1009                         ASSERT(mp->m_sb.sb_gquotino == NULLFSINO);
1010                 }
1011                 if (ino != NULLFSINO) {
1012                         error = xfs_iget(mp, NULL, ino, 0, 0, ip);
1013                         if (error)
1014                                 return error;
1015                         mp->m_sb.sb_gquotino = NULLFSINO;
1016                         mp->m_sb.sb_pquotino = NULLFSINO;
1017                 }
1018         }
1019
1020         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_QINOCREATE);
1021         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_create,
1022                                   XFS_QM_QINOCREATE_SPACE_RES(mp), 0);
1023         if (error) {
1024                 xfs_trans_cancel(tp, 0);
1025                 return error;
1026         }
1027
1028         if (!*ip) {
1029                 error = xfs_dir_ialloc(&tp, NULL, S_IFREG, 1, 0, 0, 1, ip,
1030                                                                 &committed);
1031                 if (error) {
1032                         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES |
1033                                          XFS_TRANS_ABORT);
1034                         return error;
1035                 }
1036         }
1037
1038         /*
1039          * Make the changes in the superblock, and log those too.
1040          * sbfields arg may contain fields other than *QUOTINO;
1041          * VERSIONNUM for example.
1042          */
1043         spin_lock(&mp->m_sb_lock);
1044         if (flags & XFS_QMOPT_SBVERSION) {
1045                 ASSERT(!xfs_sb_version_hasquota(&mp->m_sb));
1046                 ASSERT((sbfields & (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1047                         XFS_SB_GQUOTINO | XFS_SB_PQUOTINO | XFS_SB_QFLAGS)) ==
1048                                 (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1049                                  XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1050                                  XFS_SB_QFLAGS));
1051
1052                 xfs_sb_version_addquota(&mp->m_sb);
1053                 mp->m_sb.sb_uquotino = NULLFSINO;
1054                 mp->m_sb.sb_gquotino = NULLFSINO;
1055                 mp->m_sb.sb_pquotino = NULLFSINO;
1056
1057                 /* qflags will get updated fully _after_ quotacheck */
1058                 mp->m_sb.sb_qflags = mp->m_qflags & XFS_ALL_QUOTA_ACCT;
1059         }
1060         if (flags & XFS_QMOPT_UQUOTA)
1061                 mp->m_sb.sb_uquotino = (*ip)->i_ino;
1062         else if (flags & XFS_QMOPT_GQUOTA)
1063                 mp->m_sb.sb_gquotino = (*ip)->i_ino;
1064         else
1065                 mp->m_sb.sb_pquotino = (*ip)->i_ino;
1066         spin_unlock(&mp->m_sb_lock);
1067         xfs_mod_sb(tp, sbfields);
1068
1069         if ((error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES))) {
1070                 xfs_alert(mp, "%s failed (error %d)!", __func__, error);
1071                 return error;
1072         }
1073         return 0;
1074 }
1075
1076
1077 STATIC void
1078 xfs_qm_reset_dqcounts(
1079         xfs_mount_t     *mp,
1080         xfs_buf_t       *bp,
1081         xfs_dqid_t      id,
1082         uint            type)
1083 {
1084         struct xfs_dqblk        *dqb;
1085         int                     j;
1086
1087         trace_xfs_reset_dqcounts(bp, _RET_IP_);
1088
1089         /*
1090          * Reset all counters and timers. They'll be
1091          * started afresh by xfs_qm_quotacheck.
1092          */
1093 #ifdef DEBUG
1094         j = XFS_FSB_TO_B(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
1095         do_div(j, sizeof(xfs_dqblk_t));
1096         ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
1097 #endif
1098         dqb = bp->b_addr;
1099         for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
1100                 struct xfs_disk_dquot   *ddq;
1101
1102                 ddq = (struct xfs_disk_dquot *)&dqb[j];
1103
1104                 /*
1105                  * Do a sanity check, and if needed, repair the dqblk. Don't
1106                  * output any warnings because it's perfectly possible to
1107                  * find uninitialised dquot blks. See comment in xfs_dqcheck.
1108                  */
1109                 xfs_dqcheck(mp, ddq, id+j, type, XFS_QMOPT_DQREPAIR,
1110                             "xfs_quotacheck");
1111                 ddq->d_bcount = 0;
1112                 ddq->d_icount = 0;
1113                 ddq->d_rtbcount = 0;
1114                 ddq->d_btimer = 0;
1115                 ddq->d_itimer = 0;
1116                 ddq->d_rtbtimer = 0;
1117                 ddq->d_bwarns = 0;
1118                 ddq->d_iwarns = 0;
1119                 ddq->d_rtbwarns = 0;
1120
1121                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1122                         xfs_update_cksum((char *)&dqb[j],
1123                                          sizeof(struct xfs_dqblk),
1124                                          XFS_DQUOT_CRC_OFF);
1125                 }
1126         }
1127 }
1128
1129 STATIC int
1130 xfs_qm_dqiter_bufs(
1131         struct xfs_mount        *mp,
1132         xfs_dqid_t              firstid,
1133         xfs_fsblock_t           bno,
1134         xfs_filblks_t           blkcnt,
1135         uint                    flags,
1136         struct list_head        *buffer_list)
1137 {
1138         struct xfs_buf          *bp;
1139         int                     error;
1140         int                     type;
1141
1142         ASSERT(blkcnt > 0);
1143         type = flags & XFS_QMOPT_UQUOTA ? XFS_DQ_USER :
1144                 (flags & XFS_QMOPT_PQUOTA ? XFS_DQ_PROJ : XFS_DQ_GROUP);
1145         error = 0;
1146
1147         /*
1148          * Blkcnt arg can be a very big number, and might even be
1149          * larger than the log itself. So, we have to break it up into
1150          * manageable-sized transactions.
1151          * Note that we don't start a permanent transaction here; we might
1152          * not be able to get a log reservation for the whole thing up front,
1153          * and we don't really care to either, because we just discard
1154          * everything if we were to crash in the middle of this loop.
1155          */
1156         while (blkcnt--) {
1157                 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1158                               XFS_FSB_TO_DADDR(mp, bno),
1159                               mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1160                               &xfs_dquot_buf_ops);
1161
1162                 /*
1163                  * CRC and validation errors will return a EFSCORRUPTED here. If
1164                  * this occurs, re-read without CRC validation so that we can
1165                  * repair the damage via xfs_qm_reset_dqcounts(). This process
1166                  * will leave a trace in the log indicating corruption has
1167                  * been detected.
1168                  */
1169                 if (error == EFSCORRUPTED) {
1170                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
1171                                       XFS_FSB_TO_DADDR(mp, bno),
1172                                       mp->m_quotainfo->qi_dqchunklen, 0, &bp,
1173                                       NULL);
1174                 }
1175
1176                 if (error)
1177                         break;
1178
1179                 /*
1180                  * A corrupt buffer might not have a verifier attached, so
1181                  * make sure we have the correct one attached before writeback
1182                  * occurs.
1183                  */
1184                 bp->b_ops = &xfs_dquot_buf_ops;
1185                 xfs_qm_reset_dqcounts(mp, bp, firstid, type);
1186                 xfs_buf_delwri_queue(bp, buffer_list);
1187                 xfs_buf_relse(bp);
1188
1189                 /* goto the next block. */
1190                 bno++;
1191                 firstid += mp->m_quotainfo->qi_dqperchunk;
1192         }
1193
1194         return error;
1195 }
1196
1197 /*
1198  * Iterate over all allocated USR/GRP/PRJ dquots in the system, calling a
1199  * caller supplied function for every chunk of dquots that we find.
1200  */
1201 STATIC int
1202 xfs_qm_dqiterate(
1203         struct xfs_mount        *mp,
1204         struct xfs_inode        *qip,
1205         uint                    flags,
1206         struct list_head        *buffer_list)
1207 {
1208         struct xfs_bmbt_irec    *map;
1209         int                     i, nmaps;       /* number of map entries */
1210         int                     error;          /* return value */
1211         xfs_fileoff_t           lblkno;
1212         xfs_filblks_t           maxlblkcnt;
1213         xfs_dqid_t              firstid;
1214         xfs_fsblock_t           rablkno;
1215         xfs_filblks_t           rablkcnt;
1216
1217         error = 0;
1218         /*
1219          * This looks racy, but we can't keep an inode lock across a
1220          * trans_reserve. But, this gets called during quotacheck, and that
1221          * happens only at mount time which is single threaded.
1222          */
1223         if (qip->i_d.di_nblocks == 0)
1224                 return 0;
1225
1226         map = kmem_alloc(XFS_DQITER_MAP_SIZE * sizeof(*map), KM_SLEEP);
1227
1228         lblkno = 0;
1229         maxlblkcnt = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1230         do {
1231                 uint            lock_mode;
1232
1233                 nmaps = XFS_DQITER_MAP_SIZE;
1234                 /*
1235                  * We aren't changing the inode itself. Just changing
1236                  * some of its data. No new blocks are added here, and
1237                  * the inode is never added to the transaction.
1238                  */
1239                 lock_mode = xfs_ilock_data_map_shared(qip);
1240                 error = xfs_bmapi_read(qip, lblkno, maxlblkcnt - lblkno,
1241                                        map, &nmaps, 0);
1242                 xfs_iunlock(qip, lock_mode);
1243                 if (error)
1244                         break;
1245
1246                 ASSERT(nmaps <= XFS_DQITER_MAP_SIZE);
1247                 for (i = 0; i < nmaps; i++) {
1248                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
1249                         ASSERT(map[i].br_blockcount);
1250
1251
1252                         lblkno += map[i].br_blockcount;
1253
1254                         if (map[i].br_startblock == HOLESTARTBLOCK)
1255                                 continue;
1256
1257                         firstid = (xfs_dqid_t) map[i].br_startoff *
1258                                 mp->m_quotainfo->qi_dqperchunk;
1259                         /*
1260                          * Do a read-ahead on the next extent.
1261                          */
1262                         if ((i+1 < nmaps) &&
1263                             (map[i+1].br_startblock != HOLESTARTBLOCK)) {
1264                                 rablkcnt =  map[i+1].br_blockcount;
1265                                 rablkno = map[i+1].br_startblock;
1266                                 while (rablkcnt--) {
1267                                         xfs_buf_readahead(mp->m_ddev_targp,
1268                                                XFS_FSB_TO_DADDR(mp, rablkno),
1269                                                mp->m_quotainfo->qi_dqchunklen,
1270                                                &xfs_dquot_buf_ops);
1271                                         rablkno++;
1272                                 }
1273                         }
1274                         /*
1275                          * Iterate thru all the blks in the extent and
1276                          * reset the counters of all the dquots inside them.
1277                          */
1278                         error = xfs_qm_dqiter_bufs(mp, firstid,
1279                                                    map[i].br_startblock,
1280                                                    map[i].br_blockcount,
1281                                                    flags, buffer_list);
1282                         if (error)
1283                                 goto out;
1284                 }
1285         } while (nmaps > 0);
1286
1287 out:
1288         kmem_free(map);
1289         return error;
1290 }
1291
1292 /*
1293  * Called by dqusage_adjust in doing a quotacheck.
1294  *
1295  * Given the inode, and a dquot id this updates both the incore dqout as well
1296  * as the buffer copy. This is so that once the quotacheck is done, we can
1297  * just log all the buffers, as opposed to logging numerous updates to
1298  * individual dquots.
1299  */
1300 STATIC int
1301 xfs_qm_quotacheck_dqadjust(
1302         struct xfs_inode        *ip,
1303         xfs_dqid_t              id,
1304         uint                    type,
1305         xfs_qcnt_t              nblks,
1306         xfs_qcnt_t              rtblks)
1307 {
1308         struct xfs_mount        *mp = ip->i_mount;
1309         struct xfs_dquot        *dqp;
1310         int                     error;
1311
1312         error = xfs_qm_dqget(mp, ip, id, type,
1313                              XFS_QMOPT_DQALLOC | XFS_QMOPT_DOWARN, &dqp);
1314         if (error) {
1315                 /*
1316                  * Shouldn't be able to turn off quotas here.
1317                  */
1318                 ASSERT(error != ESRCH);
1319                 ASSERT(error != ENOENT);
1320                 return error;
1321         }
1322
1323         trace_xfs_dqadjust(dqp);
1324
1325         /*
1326          * Adjust the inode count and the block count to reflect this inode's
1327          * resource usage.
1328          */
1329         be64_add_cpu(&dqp->q_core.d_icount, 1);
1330         dqp->q_res_icount++;
1331         if (nblks) {
1332                 be64_add_cpu(&dqp->q_core.d_bcount, nblks);
1333                 dqp->q_res_bcount += nblks;
1334         }
1335         if (rtblks) {
1336                 be64_add_cpu(&dqp->q_core.d_rtbcount, rtblks);
1337                 dqp->q_res_rtbcount += rtblks;
1338         }
1339
1340         /*
1341          * Set default limits, adjust timers (since we changed usages)
1342          *
1343          * There are no timers for the default values set in the root dquot.
1344          */
1345         if (dqp->q_core.d_id) {
1346                 xfs_qm_adjust_dqlimits(mp, dqp);
1347                 xfs_qm_adjust_dqtimers(mp, &dqp->q_core);
1348         }
1349
1350         dqp->dq_flags |= XFS_DQ_DIRTY;
1351         xfs_qm_dqput(dqp);
1352         return 0;
1353 }
1354
1355 STATIC int
1356 xfs_qm_get_rtblks(
1357         xfs_inode_t     *ip,
1358         xfs_qcnt_t      *O_rtblks)
1359 {
1360         xfs_filblks_t   rtblks;                 /* total rt blks */
1361         xfs_extnum_t    idx;                    /* extent record index */
1362         xfs_ifork_t     *ifp;                   /* inode fork pointer */
1363         xfs_extnum_t    nextents;               /* number of extent entries */
1364         int             error;
1365
1366         ASSERT(XFS_IS_REALTIME_INODE(ip));
1367         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1368         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1369                 if ((error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK)))
1370                         return error;
1371         }
1372         rtblks = 0;
1373         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1374         for (idx = 0; idx < nextents; idx++)
1375                 rtblks += xfs_bmbt_get_blockcount(xfs_iext_get_ext(ifp, idx));
1376         *O_rtblks = (xfs_qcnt_t)rtblks;
1377         return 0;
1378 }
1379
1380 /*
1381  * callback routine supplied to bulkstat(). Given an inumber, find its
1382  * dquots and update them to account for resources taken by that inode.
1383  */
1384 /* ARGSUSED */
1385 STATIC int
1386 xfs_qm_dqusage_adjust(
1387         xfs_mount_t     *mp,            /* mount point for filesystem */
1388         xfs_ino_t       ino,            /* inode number to get data for */
1389         void            __user *buffer, /* not used */
1390         int             ubsize,         /* not used */
1391         int             *ubused,        /* not used */
1392         int             *res)           /* result code value */
1393 {
1394         xfs_inode_t     *ip;
1395         xfs_qcnt_t      nblks, rtblks = 0;
1396         int             error;
1397
1398         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1399
1400         /*
1401          * rootino must have its resources accounted for, not so with the quota
1402          * inodes.
1403          */
1404         if (xfs_is_quota_inode(&mp->m_sb, ino)) {
1405                 *res = BULKSTAT_RV_NOTHING;
1406                 return XFS_ERROR(EINVAL);
1407         }
1408
1409         /*
1410          * We don't _need_ to take the ilock EXCL. However, the xfs_qm_dqget
1411          * interface expects the inode to be exclusively locked because that's
1412          * the case in all other instances. It's OK that we do this because
1413          * quotacheck is done only at mount time.
1414          */
1415         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip);
1416         if (error) {
1417                 *res = BULKSTAT_RV_NOTHING;
1418                 return error;
1419         }
1420
1421         ASSERT(ip->i_delayed_blks == 0);
1422
1423         if (XFS_IS_REALTIME_INODE(ip)) {
1424                 /*
1425                  * Walk thru the extent list and count the realtime blocks.
1426                  */
1427                 error = xfs_qm_get_rtblks(ip, &rtblks);
1428                 if (error)
1429                         goto error0;
1430         }
1431
1432         nblks = (xfs_qcnt_t)ip->i_d.di_nblocks - rtblks;
1433
1434         /*
1435          * Add the (disk blocks and inode) resources occupied by this
1436          * inode to its dquots. We do this adjustment in the incore dquot,
1437          * and also copy the changes to its buffer.
1438          * We don't care about putting these changes in a transaction
1439          * envelope because if we crash in the middle of a 'quotacheck'
1440          * we have to start from the beginning anyway.
1441          * Once we're done, we'll log all the dquot bufs.
1442          *
1443          * The *QUOTA_ON checks below may look pretty racy, but quotachecks
1444          * and quotaoffs don't race. (Quotachecks happen at mount time only).
1445          */
1446         if (XFS_IS_UQUOTA_ON(mp)) {
1447                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_uid,
1448                                                    XFS_DQ_USER, nblks, rtblks);
1449                 if (error)
1450                         goto error0;
1451         }
1452
1453         if (XFS_IS_GQUOTA_ON(mp)) {
1454                 error = xfs_qm_quotacheck_dqadjust(ip, ip->i_d.di_gid,
1455                                                    XFS_DQ_GROUP, nblks, rtblks);
1456                 if (error)
1457                         goto error0;
1458         }
1459
1460         if (XFS_IS_PQUOTA_ON(mp)) {
1461                 error = xfs_qm_quotacheck_dqadjust(ip, xfs_get_projid(ip),
1462                                                    XFS_DQ_PROJ, nblks, rtblks);
1463                 if (error)
1464                         goto error0;
1465         }
1466
1467         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1468         IRELE(ip);
1469         *res = BULKSTAT_RV_DIDONE;
1470         return 0;
1471
1472 error0:
1473         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1474         IRELE(ip);
1475         *res = BULKSTAT_RV_GIVEUP;
1476         return error;
1477 }
1478
1479 STATIC int
1480 xfs_qm_flush_one(
1481         struct xfs_dquot        *dqp,
1482         void                    *data)
1483 {
1484         struct list_head        *buffer_list = data;
1485         struct xfs_buf          *bp = NULL;
1486         int                     error = 0;
1487
1488         xfs_dqlock(dqp);
1489         if (dqp->dq_flags & XFS_DQ_FREEING)
1490                 goto out_unlock;
1491         if (!XFS_DQ_IS_DIRTY(dqp))
1492                 goto out_unlock;
1493
1494         xfs_dqflock(dqp);
1495         error = xfs_qm_dqflush(dqp, &bp);
1496         if (error)
1497                 goto out_unlock;
1498
1499         xfs_buf_delwri_queue(bp, buffer_list);
1500         xfs_buf_relse(bp);
1501 out_unlock:
1502         xfs_dqunlock(dqp);
1503         return error;
1504 }
1505
1506 /*
1507  * Walk thru all the filesystem inodes and construct a consistent view
1508  * of the disk quota world. If the quotacheck fails, disable quotas.
1509  */
1510 int
1511 xfs_qm_quotacheck(
1512         xfs_mount_t     *mp)
1513 {
1514         int                     done, count, error, error2;
1515         xfs_ino_t               lastino;
1516         size_t                  structsz;
1517         uint                    flags;
1518         LIST_HEAD               (buffer_list);
1519         struct xfs_inode        *uip = mp->m_quotainfo->qi_uquotaip;
1520         struct xfs_inode        *gip = mp->m_quotainfo->qi_gquotaip;
1521         struct xfs_inode        *pip = mp->m_quotainfo->qi_pquotaip;
1522
1523         count = INT_MAX;
1524         structsz = 1;
1525         lastino = 0;
1526         flags = 0;
1527
1528         ASSERT(uip || gip || pip);
1529         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
1530
1531         xfs_notice(mp, "Quotacheck needed: Please wait.");
1532
1533         /*
1534          * First we go thru all the dquots on disk, USR and GRP/PRJ, and reset
1535          * their counters to zero. We need a clean slate.
1536          * We don't log our changes till later.
1537          */
1538         if (uip) {
1539                 error = xfs_qm_dqiterate(mp, uip, XFS_QMOPT_UQUOTA,
1540                                          &buffer_list);
1541                 if (error)
1542                         goto error_return;
1543                 flags |= XFS_UQUOTA_CHKD;
1544         }
1545
1546         if (gip) {
1547                 error = xfs_qm_dqiterate(mp, gip, XFS_QMOPT_GQUOTA,
1548                                          &buffer_list);
1549                 if (error)
1550                         goto error_return;
1551                 flags |= XFS_GQUOTA_CHKD;
1552         }
1553
1554         if (pip) {
1555                 error = xfs_qm_dqiterate(mp, pip, XFS_QMOPT_PQUOTA,
1556                                          &buffer_list);
1557                 if (error)
1558                         goto error_return;
1559                 flags |= XFS_PQUOTA_CHKD;
1560         }
1561
1562         do {
1563                 /*
1564                  * Iterate thru all the inodes in the file system,
1565                  * adjusting the corresponding dquot counters in core.
1566                  */
1567                 error = xfs_bulkstat(mp, &lastino, &count,
1568                                      xfs_qm_dqusage_adjust,
1569                                      structsz, NULL, &done);
1570                 if (error)
1571                         break;
1572
1573         } while (!done);
1574
1575         /*
1576          * We've made all the changes that we need to make incore.  Flush them
1577          * down to disk buffers if everything was updated successfully.
1578          */
1579         if (XFS_IS_UQUOTA_ON(mp)) {
1580                 error = xfs_qm_dquot_walk(mp, XFS_DQ_USER, xfs_qm_flush_one,
1581                                           &buffer_list);
1582         }
1583         if (XFS_IS_GQUOTA_ON(mp)) {
1584                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_GROUP, xfs_qm_flush_one,
1585                                            &buffer_list);
1586                 if (!error)
1587                         error = error2;
1588         }
1589         if (XFS_IS_PQUOTA_ON(mp)) {
1590                 error2 = xfs_qm_dquot_walk(mp, XFS_DQ_PROJ, xfs_qm_flush_one,
1591                                            &buffer_list);
1592                 if (!error)
1593                         error = error2;
1594         }
1595
1596         error2 = xfs_buf_delwri_submit(&buffer_list);
1597         if (!error)
1598                 error = error2;
1599
1600         /*
1601          * We can get this error if we couldn't do a dquot allocation inside
1602          * xfs_qm_dqusage_adjust (via bulkstat). We don't care about the
1603          * dirty dquots that might be cached, we just want to get rid of them
1604          * and turn quotaoff. The dquots won't be attached to any of the inodes
1605          * at this point (because we intentionally didn't in dqget_noattach).
1606          */
1607         if (error) {
1608                 xfs_qm_dqpurge_all(mp, XFS_QMOPT_QUOTALL);
1609                 goto error_return;
1610         }
1611
1612         /*
1613          * If one type of quotas is off, then it will lose its
1614          * quotachecked status, since we won't be doing accounting for
1615          * that type anymore.
1616          */
1617         mp->m_qflags &= ~XFS_ALL_QUOTA_CHKD;
1618         mp->m_qflags |= flags;
1619
1620  error_return:
1621         while (!list_empty(&buffer_list)) {
1622                 struct xfs_buf *bp =
1623                         list_first_entry(&buffer_list, struct xfs_buf, b_list);
1624                 list_del_init(&bp->b_list);
1625                 xfs_buf_relse(bp);
1626         }
1627
1628         if (error) {
1629                 xfs_warn(mp,
1630         "Quotacheck: Unsuccessful (Error %d): Disabling quotas.",
1631                         error);
1632                 /*
1633                  * We must turn off quotas.
1634                  */
1635                 ASSERT(mp->m_quotainfo != NULL);
1636                 xfs_qm_destroy_quotainfo(mp);
1637                 if (xfs_mount_reset_sbqflags(mp)) {
1638                         xfs_warn(mp,
1639                                 "Quotacheck: Failed to reset quota flags.");
1640                 }
1641         } else
1642                 xfs_notice(mp, "Quotacheck: Done.");
1643         return (error);
1644 }
1645
1646 /*
1647  * This is called after the superblock has been read in and we're ready to
1648  * iget the quota inodes.
1649  */
1650 STATIC int
1651 xfs_qm_init_quotainos(
1652         xfs_mount_t     *mp)
1653 {
1654         struct xfs_inode        *uip = NULL;
1655         struct xfs_inode        *gip = NULL;
1656         struct xfs_inode        *pip = NULL;
1657         int                     error;
1658         __int64_t               sbflags = 0;
1659         uint                    flags = 0;
1660
1661         ASSERT(mp->m_quotainfo);
1662
1663         /*
1664          * Get the uquota and gquota inodes
1665          */
1666         if (xfs_sb_version_hasquota(&mp->m_sb)) {
1667                 if (XFS_IS_UQUOTA_ON(mp) &&
1668                     mp->m_sb.sb_uquotino != NULLFSINO) {
1669                         ASSERT(mp->m_sb.sb_uquotino > 0);
1670                         error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
1671                                              0, 0, &uip);
1672                         if (error)
1673                                 return XFS_ERROR(error);
1674                 }
1675                 if (XFS_IS_GQUOTA_ON(mp) &&
1676                     mp->m_sb.sb_gquotino != NULLFSINO) {
1677                         ASSERT(mp->m_sb.sb_gquotino > 0);
1678                         error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
1679                                              0, 0, &gip);
1680                         if (error)
1681                                 goto error_rele;
1682                 }
1683                 if (XFS_IS_PQUOTA_ON(mp) &&
1684                     mp->m_sb.sb_pquotino != NULLFSINO) {
1685                         ASSERT(mp->m_sb.sb_pquotino > 0);
1686                         error = xfs_iget(mp, NULL, mp->m_sb.sb_pquotino,
1687                                              0, 0, &pip);
1688                         if (error)
1689                                 goto error_rele;
1690                 }
1691         } else {
1692                 flags |= XFS_QMOPT_SBVERSION;
1693                 sbflags |= (XFS_SB_VERSIONNUM | XFS_SB_UQUOTINO |
1694                             XFS_SB_GQUOTINO | XFS_SB_PQUOTINO |
1695                             XFS_SB_QFLAGS);
1696         }
1697
1698         /*
1699          * Create the three inodes, if they don't exist already. The changes
1700          * made above will get added to a transaction and logged in one of
1701          * the qino_alloc calls below.  If the device is readonly,
1702          * temporarily switch to read-write to do this.
1703          */
1704         if (XFS_IS_UQUOTA_ON(mp) && uip == NULL) {
1705                 error = xfs_qm_qino_alloc(mp, &uip,
1706                                               sbflags | XFS_SB_UQUOTINO,
1707                                               flags | XFS_QMOPT_UQUOTA);
1708                 if (error)
1709                         goto error_rele;
1710
1711                 flags &= ~XFS_QMOPT_SBVERSION;
1712         }
1713         if (XFS_IS_GQUOTA_ON(mp) && gip == NULL) {
1714                 error = xfs_qm_qino_alloc(mp, &gip,
1715                                           sbflags | XFS_SB_GQUOTINO,
1716                                           flags | XFS_QMOPT_GQUOTA);
1717                 if (error)
1718                         goto error_rele;
1719
1720                 flags &= ~XFS_QMOPT_SBVERSION;
1721         }
1722         if (XFS_IS_PQUOTA_ON(mp) && pip == NULL) {
1723                 error = xfs_qm_qino_alloc(mp, &pip,
1724                                           sbflags | XFS_SB_PQUOTINO,
1725                                           flags | XFS_QMOPT_PQUOTA);
1726                 if (error)
1727                         goto error_rele;
1728         }
1729
1730         mp->m_quotainfo->qi_uquotaip = uip;
1731         mp->m_quotainfo->qi_gquotaip = gip;
1732         mp->m_quotainfo->qi_pquotaip = pip;
1733
1734         return 0;
1735
1736 error_rele:
1737         if (uip)
1738                 IRELE(uip);
1739         if (gip)
1740                 IRELE(gip);
1741         if (pip)
1742                 IRELE(pip);
1743         return XFS_ERROR(error);
1744 }
1745
1746 STATIC void
1747 xfs_qm_dqfree_one(
1748         struct xfs_dquot        *dqp)
1749 {
1750         struct xfs_mount        *mp = dqp->q_mount;
1751         struct xfs_quotainfo    *qi = mp->m_quotainfo;
1752
1753         mutex_lock(&qi->qi_tree_lock);
1754         radix_tree_delete(xfs_dquot_tree(qi, dqp->q_core.d_flags),
1755                           be32_to_cpu(dqp->q_core.d_id));
1756
1757         qi->qi_dquots--;
1758         mutex_unlock(&qi->qi_tree_lock);
1759
1760         xfs_qm_dqdestroy(dqp);
1761 }
1762
1763 /*
1764  * Start a transaction and write the incore superblock changes to
1765  * disk. flags parameter indicates which fields have changed.
1766  */
1767 int
1768 xfs_qm_write_sb_changes(
1769         xfs_mount_t     *mp,
1770         __int64_t       flags)
1771 {
1772         xfs_trans_t     *tp;
1773         int             error;
1774
1775         tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1776         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0);
1777         if (error) {
1778                 xfs_trans_cancel(tp, 0);
1779                 return error;
1780         }
1781
1782         xfs_mod_sb(tp, flags);
1783         error = xfs_trans_commit(tp, 0);
1784
1785         return error;
1786 }
1787
1788
1789 /* --------------- utility functions for vnodeops ---------------- */
1790
1791
1792 /*
1793  * Given an inode, a uid, gid and prid make sure that we have
1794  * allocated relevant dquot(s) on disk, and that we won't exceed inode
1795  * quotas by creating this file.
1796  * This also attaches dquot(s) to the given inode after locking it,
1797  * and returns the dquots corresponding to the uid and/or gid.
1798  *
1799  * in   : inode (unlocked)
1800  * out  : udquot, gdquot with references taken and unlocked
1801  */
1802 int
1803 xfs_qm_vop_dqalloc(
1804         struct xfs_inode        *ip,
1805         xfs_dqid_t              uid,
1806         xfs_dqid_t              gid,
1807         prid_t                  prid,
1808         uint                    flags,
1809         struct xfs_dquot        **O_udqpp,
1810         struct xfs_dquot        **O_gdqpp,
1811         struct xfs_dquot        **O_pdqpp)
1812 {
1813         struct xfs_mount        *mp = ip->i_mount;
1814         struct xfs_dquot        *uq = NULL;
1815         struct xfs_dquot        *gq = NULL;
1816         struct xfs_dquot        *pq = NULL;
1817         int                     error;
1818         uint                    lockflags;
1819
1820         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
1821                 return 0;
1822
1823         lockflags = XFS_ILOCK_EXCL;
1824         xfs_ilock(ip, lockflags);
1825
1826         if ((flags & XFS_QMOPT_INHERIT) && XFS_INHERIT_GID(ip))
1827                 gid = ip->i_d.di_gid;
1828
1829         /*
1830          * Attach the dquot(s) to this inode, doing a dquot allocation
1831          * if necessary. The dquot(s) will not be locked.
1832          */
1833         if (XFS_NOT_DQATTACHED(mp, ip)) {
1834                 error = xfs_qm_dqattach_locked(ip, XFS_QMOPT_DQALLOC);
1835                 if (error) {
1836                         xfs_iunlock(ip, lockflags);
1837                         return error;
1838                 }
1839         }
1840
1841         if ((flags & XFS_QMOPT_UQUOTA) && XFS_IS_UQUOTA_ON(mp)) {
1842                 if (ip->i_d.di_uid != uid) {
1843                         /*
1844                          * What we need is the dquot that has this uid, and
1845                          * if we send the inode to dqget, the uid of the inode
1846                          * takes priority over what's sent in the uid argument.
1847                          * We must unlock inode here before calling dqget if
1848                          * we're not sending the inode, because otherwise
1849                          * we'll deadlock by doing trans_reserve while
1850                          * holding ilock.
1851                          */
1852                         xfs_iunlock(ip, lockflags);
1853                         error = xfs_qm_dqget(mp, NULL, uid,
1854                                                  XFS_DQ_USER,
1855                                                  XFS_QMOPT_DQALLOC |
1856                                                  XFS_QMOPT_DOWARN,
1857                                                  &uq);
1858                         if (error) {
1859                                 ASSERT(error != ENOENT);
1860                                 return error;
1861                         }
1862                         /*
1863                          * Get the ilock in the right order.
1864                          */
1865                         xfs_dqunlock(uq);
1866                         lockflags = XFS_ILOCK_SHARED;
1867                         xfs_ilock(ip, lockflags);
1868                 } else {
1869                         /*
1870                          * Take an extra reference, because we'll return
1871                          * this to caller
1872                          */
1873                         ASSERT(ip->i_udquot);
1874                         uq = xfs_qm_dqhold(ip->i_udquot);
1875                 }
1876         }
1877         if ((flags & XFS_QMOPT_GQUOTA) && XFS_IS_GQUOTA_ON(mp)) {
1878                 if (ip->i_d.di_gid != gid) {
1879                         xfs_iunlock(ip, lockflags);
1880                         error = xfs_qm_dqget(mp, NULL, gid,
1881                                                  XFS_DQ_GROUP,
1882                                                  XFS_QMOPT_DQALLOC |
1883                                                  XFS_QMOPT_DOWARN,
1884                                                  &gq);
1885                         if (error) {
1886                                 ASSERT(error != ENOENT);
1887                                 goto error_rele;
1888                         }
1889                         xfs_dqunlock(gq);
1890                         lockflags = XFS_ILOCK_SHARED;
1891                         xfs_ilock(ip, lockflags);
1892                 } else {
1893                         ASSERT(ip->i_gdquot);
1894                         gq = xfs_qm_dqhold(ip->i_gdquot);
1895                 }
1896         }
1897         if ((flags & XFS_QMOPT_PQUOTA) && XFS_IS_PQUOTA_ON(mp)) {
1898                 if (xfs_get_projid(ip) != prid) {
1899                         xfs_iunlock(ip, lockflags);
1900                         error = xfs_qm_dqget(mp, NULL, (xfs_dqid_t)prid,
1901                                                  XFS_DQ_PROJ,
1902                                                  XFS_QMOPT_DQALLOC |
1903                                                  XFS_QMOPT_DOWARN,
1904                                                  &pq);
1905                         if (error) {
1906                                 ASSERT(error != ENOENT);
1907                                 goto error_rele;
1908                         }
1909                         xfs_dqunlock(pq);
1910                         lockflags = XFS_ILOCK_SHARED;
1911                         xfs_ilock(ip, lockflags);
1912                 } else {
1913                         ASSERT(ip->i_pdquot);
1914                         pq = xfs_qm_dqhold(ip->i_pdquot);
1915                 }
1916         }
1917         if (uq)
1918                 trace_xfs_dquot_dqalloc(ip);
1919
1920         xfs_iunlock(ip, lockflags);
1921         if (O_udqpp)
1922                 *O_udqpp = uq;
1923         else if (uq)
1924                 xfs_qm_dqrele(uq);
1925         if (O_gdqpp)
1926                 *O_gdqpp = gq;
1927         else if (gq)
1928                 xfs_qm_dqrele(gq);
1929         if (O_pdqpp)
1930                 *O_pdqpp = pq;
1931         else if (pq)
1932                 xfs_qm_dqrele(pq);
1933         return 0;
1934
1935 error_rele:
1936         if (gq)
1937                 xfs_qm_dqrele(gq);
1938         if (uq)
1939                 xfs_qm_dqrele(uq);
1940         return error;
1941 }
1942
1943 /*
1944  * Actually transfer ownership, and do dquot modifications.
1945  * These were already reserved.
1946  */
1947 xfs_dquot_t *
1948 xfs_qm_vop_chown(
1949         xfs_trans_t     *tp,
1950         xfs_inode_t     *ip,
1951         xfs_dquot_t     **IO_olddq,
1952         xfs_dquot_t     *newdq)
1953 {
1954         xfs_dquot_t     *prevdq;
1955         uint            bfield = XFS_IS_REALTIME_INODE(ip) ?
1956                                  XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
1957
1958
1959         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1960         ASSERT(XFS_IS_QUOTA_RUNNING(ip->i_mount));
1961
1962         /* old dquot */
1963         prevdq = *IO_olddq;
1964         ASSERT(prevdq);
1965         ASSERT(prevdq != newdq);
1966
1967         xfs_trans_mod_dquot(tp, prevdq, bfield, -(ip->i_d.di_nblocks));
1968         xfs_trans_mod_dquot(tp, prevdq, XFS_TRANS_DQ_ICOUNT, -1);
1969
1970         /* the sparkling new dquot */
1971         xfs_trans_mod_dquot(tp, newdq, bfield, ip->i_d.di_nblocks);
1972         xfs_trans_mod_dquot(tp, newdq, XFS_TRANS_DQ_ICOUNT, 1);
1973
1974         /*
1975          * Take an extra reference, because the inode is going to keep
1976          * this dquot pointer even after the trans_commit.
1977          */
1978         *IO_olddq = xfs_qm_dqhold(newdq);
1979
1980         return prevdq;
1981 }
1982
1983 /*
1984  * Quota reservations for setattr(AT_UID|AT_GID|AT_PROJID).
1985  */
1986 int
1987 xfs_qm_vop_chown_reserve(
1988         struct xfs_trans        *tp,
1989         struct xfs_inode        *ip,
1990         struct xfs_dquot        *udqp,
1991         struct xfs_dquot        *gdqp,
1992         struct xfs_dquot        *pdqp,
1993         uint                    flags)
1994 {
1995         struct xfs_mount        *mp = ip->i_mount;
1996         uint                    delblks, blkflags, prjflags = 0;
1997         struct xfs_dquot        *udq_unres = NULL;
1998         struct xfs_dquot        *gdq_unres = NULL;
1999         struct xfs_dquot        *pdq_unres = NULL;
2000         struct xfs_dquot        *udq_delblks = NULL;
2001         struct xfs_dquot        *gdq_delblks = NULL;
2002         struct xfs_dquot        *pdq_delblks = NULL;
2003         int                     error;
2004
2005
2006         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2007         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2008
2009         delblks = ip->i_delayed_blks;
2010         blkflags = XFS_IS_REALTIME_INODE(ip) ?
2011                         XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS;
2012
2013         if (XFS_IS_UQUOTA_ON(mp) && udqp &&
2014             ip->i_d.di_uid != be32_to_cpu(udqp->q_core.d_id)) {
2015                 udq_delblks = udqp;
2016                 /*
2017                  * If there are delayed allocation blocks, then we have to
2018                  * unreserve those from the old dquot, and add them to the
2019                  * new dquot.
2020                  */
2021                 if (delblks) {
2022                         ASSERT(ip->i_udquot);
2023                         udq_unres = ip->i_udquot;
2024                 }
2025         }
2026         if (XFS_IS_GQUOTA_ON(ip->i_mount) && gdqp &&
2027             ip->i_d.di_gid != be32_to_cpu(gdqp->q_core.d_id)) {
2028                 gdq_delblks = gdqp;
2029                 if (delblks) {
2030                         ASSERT(ip->i_gdquot);
2031                         gdq_unres = ip->i_gdquot;
2032                 }
2033         }
2034
2035         if (XFS_IS_PQUOTA_ON(ip->i_mount) && pdqp &&
2036             xfs_get_projid(ip) != be32_to_cpu(pdqp->q_core.d_id)) {
2037                 prjflags = XFS_QMOPT_ENOSPC;
2038                 pdq_delblks = pdqp;
2039                 if (delblks) {
2040                         ASSERT(ip->i_pdquot);
2041                         pdq_unres = ip->i_pdquot;
2042                 }
2043         }
2044
2045         error = xfs_trans_reserve_quota_bydquots(tp, ip->i_mount,
2046                                 udq_delblks, gdq_delblks, pdq_delblks,
2047                                 ip->i_d.di_nblocks, 1,
2048                                 flags | blkflags | prjflags);
2049         if (error)
2050                 return error;
2051
2052         /*
2053          * Do the delayed blks reservations/unreservations now. Since, these
2054          * are done without the help of a transaction, if a reservation fails
2055          * its previous reservations won't be automatically undone by trans
2056          * code. So, we have to do it manually here.
2057          */
2058         if (delblks) {
2059                 /*
2060                  * Do the reservations first. Unreservation can't fail.
2061                  */
2062                 ASSERT(udq_delblks || gdq_delblks || pdq_delblks);
2063                 ASSERT(udq_unres || gdq_unres || pdq_unres);
2064                 error = xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2065                             udq_delblks, gdq_delblks, pdq_delblks,
2066                             (xfs_qcnt_t)delblks, 0,
2067                             flags | blkflags | prjflags);
2068                 if (error)
2069                         return error;
2070                 xfs_trans_reserve_quota_bydquots(NULL, ip->i_mount,
2071                                 udq_unres, gdq_unres, pdq_unres,
2072                                 -((xfs_qcnt_t)delblks), 0, blkflags);
2073         }
2074
2075         return (0);
2076 }
2077
2078 int
2079 xfs_qm_vop_rename_dqattach(
2080         struct xfs_inode        **i_tab)
2081 {
2082         struct xfs_mount        *mp = i_tab[0]->i_mount;
2083         int                     i;
2084
2085         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2086                 return 0;
2087
2088         for (i = 0; (i < 4 && i_tab[i]); i++) {
2089                 struct xfs_inode        *ip = i_tab[i];
2090                 int                     error;
2091
2092                 /*
2093                  * Watch out for duplicate entries in the table.
2094                  */
2095                 if (i == 0 || ip != i_tab[i-1]) {
2096                         if (XFS_NOT_DQATTACHED(mp, ip)) {
2097                                 error = xfs_qm_dqattach(ip, 0);
2098                                 if (error)
2099                                         return error;
2100                         }
2101                 }
2102         }
2103         return 0;
2104 }
2105
2106 void
2107 xfs_qm_vop_create_dqattach(
2108         struct xfs_trans        *tp,
2109         struct xfs_inode        *ip,
2110         struct xfs_dquot        *udqp,
2111         struct xfs_dquot        *gdqp,
2112         struct xfs_dquot        *pdqp)
2113 {
2114         struct xfs_mount        *mp = tp->t_mountp;
2115
2116         if (!XFS_IS_QUOTA_RUNNING(mp) || !XFS_IS_QUOTA_ON(mp))
2117                 return;
2118
2119         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2120         ASSERT(XFS_IS_QUOTA_RUNNING(mp));
2121
2122         if (udqp && XFS_IS_UQUOTA_ON(mp)) {
2123                 ASSERT(ip->i_udquot == NULL);
2124                 ASSERT(ip->i_d.di_uid == be32_to_cpu(udqp->q_core.d_id));
2125
2126                 ip->i_udquot = xfs_qm_dqhold(udqp);
2127                 xfs_trans_mod_dquot(tp, udqp, XFS_TRANS_DQ_ICOUNT, 1);
2128         }
2129         if (gdqp && XFS_IS_GQUOTA_ON(mp)) {
2130                 ASSERT(ip->i_gdquot == NULL);
2131                 ASSERT(ip->i_d.di_gid == be32_to_cpu(gdqp->q_core.d_id));
2132                 ip->i_gdquot = xfs_qm_dqhold(gdqp);
2133                 xfs_trans_mod_dquot(tp, gdqp, XFS_TRANS_DQ_ICOUNT, 1);
2134         }
2135         if (pdqp && XFS_IS_PQUOTA_ON(mp)) {
2136                 ASSERT(ip->i_pdquot == NULL);
2137                 ASSERT(xfs_get_projid(ip) == be32_to_cpu(pdqp->q_core.d_id));
2138
2139                 ip->i_pdquot = xfs_qm_dqhold(pdqp);
2140                 xfs_trans_mod_dquot(tp, pdqp, XFS_TRANS_DQ_ICOUNT, 1);
2141         }
2142 }
2143