d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
                d->dd_diskdq.d_id = cpu_to_be32(curid);
                d->dd_diskdq.d_flags = type;
-               if (xfs_sb_version_hascrc(&mp->m_sb))
+               if (xfs_sb_version_hascrc(&mp->m_sb)) {
                        uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
+                       xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
+                                        XFS_DQUOT_CRC_OFF);
+               }
        }
 
        xfs_trans_dquot_buf(tp, bp,
        dqp->q_low_space[XFS_QLOWSP_5_PCNT] = space * 5;
 }
 
-STATIC void
-xfs_dquot_buf_calc_crc(
-       struct xfs_mount        *mp,
-       struct xfs_buf          *bp)
-{
-       struct xfs_dqblk        *d = (struct xfs_dqblk *)bp->b_addr;
-       int                     i;
-
-       if (!xfs_sb_version_hascrc(&mp->m_sb))
-               return;
-
-       for (i = 0; i < mp->m_quotainfo->qi_dqperchunk; i++, d++) {
-               xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
-                                offsetof(struct xfs_dqblk, dd_crc));
-       }
-}
-
 STATIC bool
 xfs_dquot_buf_verify_crc(
        struct xfs_mount        *mp,
 
        for (i = 0; i < ndquots; i++, d++) {
                if (!xfs_verify_cksum((char *)d, sizeof(struct xfs_dqblk),
-                                offsetof(struct xfs_dqblk, dd_crc)))
+                                XFS_DQUOT_CRC_OFF))
                        return false;
                if (!uuid_equal(&d->dd_uuid, &mp->m_sb.sb_uuid))
                        return false;
        }
-
        return true;
 }
 
        }
 }
 
+/*
+ * we don't calculate the CRC here as that is done when the dquot is flushed to
+ * the buffer after the update is done. This ensures that the dquot in the
+ * buffer always has an up-to-date CRC value.
+ */
 void
 xfs_dquot_buf_write_verify(
        struct xfs_buf  *bp)
                xfs_buf_ioerror(bp, EFSCORRUPTED);
                return;
        }
-       xfs_dquot_buf_calc_crc(mp, bp);
 }
 
 const struct xfs_buf_ops xfs_dquot_buf_ops = {
         * copy the lsn into the on-disk dquot now while we have the in memory
         * dquot here. This can't be done later in the write verifier as we
         * can't get access to the log item at that point in time.
+        *
+        * We also calculate the CRC here so that the on-disk dquot in the
+        * buffer always has a valid CRC. This ensures there is no possibility
+        * of a dquot without an up-to-date CRC getting to disk.
         */
        if (xfs_sb_version_hascrc(&mp->m_sb)) {
                struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddqp;
 
                dqb->dd_lsn = cpu_to_be64(dqp->q_logitem.qli_item.li_lsn);
+               xfs_update_cksum((char *)dqb, sizeof(struct xfs_dqblk),
+                                XFS_DQUOT_CRC_OFF);
        }
 
        /*
 
        d->dd_diskdq.d_flags = type;
        d->dd_diskdq.d_id = cpu_to_be32(id);
 
+       if (xfs_sb_version_hascrc(&mp->m_sb)) {
+               uuid_copy(&d->dd_uuid, &mp->m_sb.sb_uuid);
+               xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
+                                XFS_DQUOT_CRC_OFF);
+       }
+
        return errs;
 }
 
        }
 
        memcpy(ddq, recddq, item->ri_buf[1].i_len);
+       if (xfs_sb_version_hascrc(&mp->m_sb)) {
+               xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
+                                XFS_DQUOT_CRC_OFF);
+       }
 
        ASSERT(dq_f->qlf_size == 2);
        ASSERT(bp->b_target->bt_mount == mp);
 
 #include "xfs_qm.h"
 #include "xfs_trace.h"
 #include "xfs_icache.h"
+#include "xfs_cksum.h"
 
 /*
  * The global quota manager. There is only one of these for the entire
        xfs_dqid_t      id,
        uint            type)
 {
-       xfs_disk_dquot_t        *ddq;
+       struct xfs_dqblk        *dqb;
        int                     j;
 
        trace_xfs_reset_dqcounts(bp, _RET_IP_);
        do_div(j, sizeof(xfs_dqblk_t));
        ASSERT(mp->m_quotainfo->qi_dqperchunk == j);
 #endif
-       ddq = bp->b_addr;
+       dqb = bp->b_addr;
        for (j = 0; j < mp->m_quotainfo->qi_dqperchunk; j++) {
+               struct xfs_disk_dquot   *ddq;
+
+               ddq = (struct xfs_disk_dquot *)&dqb[j];
+
                /*
                 * Do a sanity check, and if needed, repair the dqblk. Don't
                 * output any warnings because it's perfectly possible to
                ddq->d_bwarns = 0;
                ddq->d_iwarns = 0;
                ddq->d_rtbwarns = 0;
-               ddq = (xfs_disk_dquot_t *) ((xfs_dqblk_t *)ddq + 1);
+
+               if (xfs_sb_version_hascrc(&mp->m_sb)) {
+                       xfs_update_cksum((char *)&dqb[j],
+                                        sizeof(struct xfs_dqblk),
+                                        XFS_DQUOT_CRC_OFF);
+               }
        }
 }
 
                              XFS_FSB_TO_DADDR(mp, bno),
                              mp->m_quotainfo->qi_dqchunklen, 0, &bp,
                              &xfs_dquot_buf_ops);
-               if (error)
-                       break;
 
                /*
-                * XXX(hch): need to figure out if it makes sense to validate
-                *           the CRC here.
+                * CRC and validation errors will return a EFSCORRUPTED here. If
+                * this occurs, re-read without CRC validation so that we can
+                * repair the damage via xfs_qm_reset_dqcounts(). This process
+                * will leave a trace in the log indicating corruption has
+                * been detected.
                 */
+               if (error == EFSCORRUPTED) {
+                       error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
+                                     XFS_FSB_TO_DADDR(mp, bno),
+                                     mp->m_quotainfo->qi_dqchunklen, 0, &bp,
+                                     NULL);
+               }
+
+               if (error)
+                       break;
+
                xfs_qm_reset_dqcounts(mp, bp, firstid, type);
                xfs_buf_delwri_queue(bp, buffer_list);
                xfs_buf_relse(bp);
-               /*
-                * goto the next block.
-                */
+
+               /* goto the next block. */
                bno++;
                firstid += mp->m_quotainfo->qi_dqperchunk;
        }
 
        uuid_t            dd_uuid;      /* location information */
 } xfs_dqblk_t;
 
+#define XFS_DQUOT_CRC_OFF      offsetof(struct xfs_dqblk, dd_crc)
+
 /*
  * flags for q_flags field in the dquot.
  */