xfs: widen ondisk quota expiration timestamps to handle y2038+

Enable the bigtime feature for quota timers.  We decrease the accuracy
of the timers to ~4s in exchange for being able to set timers up to the
bigtime maximum.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Collins <allison.henderson@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>

fs/xfs/libxfs/xfs_dquot_buf.c

@@ -69,6 +69,13 @@ xfs_dquot_verify(
 	    ddq_type != XFS_DQTYPE_GROUP)
 		return __this_address;
 
+	if ((ddq->d_type & XFS_DQTYPE_BIGTIME) &&
+	    !xfs_sb_version_hasbigtime(&mp->m_sb))
+		return __this_address;
+
+	if ((ddq->d_type & XFS_DQTYPE_BIGTIME) && !ddq->d_id)
+		return __this_address;
+
 	if (id != -1 && id != be32_to_cpu(ddq->d_id))
 		return __this_address;
 
@@ -295,7 +302,12 @@ xfs_dquot_from_disk_ts(
 	struct xfs_disk_dquot	*ddq,
 	__be32			dtimer)
 {
-	return be32_to_cpu(dtimer);
+	uint32_t		t = be32_to_cpu(dtimer);
+
+	if (t != 0 && (ddq->d_type & XFS_DQTYPE_BIGTIME))
+		return xfs_dq_bigtime_to_unix(t);
+
+	return t;
 }
 
 /* Convert an incore timer value into an on-disk timer value. */
@@ -304,5 +316,10 @@ xfs_dquot_to_disk_ts(
 	struct xfs_dquot	*dqp,
 	time64_t		timer)
 {
-	return cpu_to_be32(timer);
+	uint32_t		t = timer;
+
+	if (timer != 0 && (dqp->q_type & XFS_DQTYPE_BIGTIME))
+		t = xfs_dq_unix_to_bigtime(timer);
+
+	return cpu_to_be32(t);
 }

fs/xfs/libxfs/xfs_format.h

@@ -1263,13 +1263,15 @@ static inline bool xfs_dinode_has_bigtime(const struct xfs_dinode *dip)
 #define XFS_DQTYPE_USER		0x01		/* user dquot record */
 #define XFS_DQTYPE_PROJ		0x02		/* project dquot record */
 #define XFS_DQTYPE_GROUP	0x04		/* group dquot record */
+#define XFS_DQTYPE_BIGTIME	0x80		/* large expiry timestamps */
 
 /* bitmask to determine if this is a user/group/project dquot */
 #define XFS_DQTYPE_REC_MASK	(XFS_DQTYPE_USER | \
 				 XFS_DQTYPE_PROJ | \
 				 XFS_DQTYPE_GROUP)
 
-#define XFS_DQTYPE_ANY		(XFS_DQTYPE_REC_MASK)
+#define XFS_DQTYPE_ANY		(XFS_DQTYPE_REC_MASK | \
+				 XFS_DQTYPE_BIGTIME)
 
 /*
  * XFS Quota Timers
@@ -1282,6 +1284,10 @@ static inline bool xfs_dinode_has_bigtime(const struct xfs_dinode *dip)
  * ondisk min and max defined here can be used directly to constrain the incore
  * quota expiration timestamps on a Unix system.
  *
+ * When bigtime is enabled, we trade two bits of precision to expand the
+ * expiration timeout range to match that of big inode timestamps.  The min and
+ * max recorded here are the on-disk limits, not a Unix timestamp.
+ *
  * The grace period for each quota type is stored in the root dquot (id = 0)
  * and is applied to a non-root dquot when it exceeds the soft or hard limits.
  * The length of quota grace periods are unsigned 32-bit quantities measured in
@@ -1300,6 +1306,48 @@ static inline bool xfs_dinode_has_bigtime(const struct xfs_dinode *dip)
  */
 #define XFS_DQ_LEGACY_EXPIRY_MAX	((int64_t)U32_MAX)
 
+/*
+ * Smallest possible ondisk quota expiration value with bigtime timestamps.
+ * This corresponds (after conversion to a Unix timestamp) with the incore
+ * expiration of Jan  1 00:00:04 UTC 1970.
+ */
+#define XFS_DQ_BIGTIME_EXPIRY_MIN	(XFS_DQ_LEGACY_EXPIRY_MIN)
+
+/*
+ * Largest supported ondisk quota expiration value with bigtime timestamps.
+ * This corresponds (after conversion to a Unix timestamp) with an incore
+ * expiration of Jul  2 20:20:24 UTC 2486.
+ *
+ * The ondisk field supports values up to -1U, which corresponds to an incore
+ * expiration in 2514.  This is beyond the maximum the bigtime inode timestamp,
+ * so we cap the maximum bigtime quota expiration to the max inode timestamp.
+ */
+#define XFS_DQ_BIGTIME_EXPIRY_MAX	((int64_t)4074815106U)
+
+/*
+ * The following conversion factors assist in converting a quota expiration
+ * timestamp between the incore and ondisk formats.
+ */
+#define XFS_DQ_BIGTIME_SHIFT	(2)
+#define XFS_DQ_BIGTIME_SLACK	((int64_t)(1ULL << XFS_DQ_BIGTIME_SHIFT) - 1)
+
+/* Convert an incore quota expiration timestamp to an ondisk bigtime value. */
+static inline uint32_t xfs_dq_unix_to_bigtime(time64_t unix_seconds)
+{
+	/*
+	 * Round the expiration timestamp up to the nearest bigtime timestamp
+	 * that we can store, to give users the most time to fix problems.
+	 */
+	return ((uint64_t)unix_seconds + XFS_DQ_BIGTIME_SLACK) >>
+			XFS_DQ_BIGTIME_SHIFT;
+}
+
+/* Convert an ondisk bigtime quota expiration value to an incore timestamp. */
+static inline time64_t xfs_dq_bigtime_to_unix(uint32_t ondisk_seconds)
+{
+	return (time64_t)ondisk_seconds << XFS_DQ_BIGTIME_SHIFT;
+}
+
 /*
  * Default quota grace periods, ranging from zero (use the compiled defaults)
  * to ~136 years.  These are applied to a non-root dquot that has exceeded

fs/xfs/libxfs/xfs_quota_defs.h

@@ -23,7 +23,8 @@ typedef uint8_t		xfs_dqtype_t;
 #define XFS_DQTYPE_STRINGS \
 	{ XFS_DQTYPE_USER,	"USER" }, \
 	{ XFS_DQTYPE_PROJ,	"PROJ" }, \
-	{ XFS_DQTYPE_GROUP,	"GROUP" }
+	{ XFS_DQTYPE_GROUP,	"GROUP" }, \
+	{ XFS_DQTYPE_BIGTIME,	"BIGTIME" }
 
 /*
  * flags for q_flags field in the dquot.

fs/xfs/xfs_dquot.c

@@ -223,6 +223,8 @@ xfs_qm_init_dquot_blk(
 		d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
 		d->dd_diskdq.d_id = cpu_to_be32(curid);
 		d->dd_diskdq.d_type = type;
+		if (curid > 0 && xfs_sb_version_hasbigtime(&mp->m_sb))
+			d->dd_diskdq.d_type |= XFS_DQTYPE_BIGTIME;
 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
 			uuid_copy(&d->dd_uuid, &mp->m_sb.sb_meta_uuid);
 			xfs_update_cksum((char *)d, sizeof(struct xfs_dqblk),
@@ -1179,6 +1181,14 @@ xfs_qm_dqflush_check(
 	    !dqp->q_rtb.timer)
 		return __this_address;
 
+	/* bigtime flag should never be set on root dquots */
+	if (dqp->q_type & XFS_DQTYPE_BIGTIME) {
+		if (!xfs_sb_version_hasbigtime(&dqp->q_mount->m_sb))
+			return __this_address;
+		if (dqp->q_id == 0)
+			return __this_address;
+	}
+
 	return NULL;
 }
 

fs/xfs/xfs_ondisk.h

@@ -167,6 +167,11 @@ xfs_check_ondisk_structs(void)
 			XFS_LEGACY_TIME_MIN);
 	XFS_CHECK_VALUE(XFS_BIGTIME_TIME_MAX - XFS_BIGTIME_EPOCH_OFFSET,
 			16299260424LL);
+
+	/* Do the same with the incore quota expiration range. */
+	XFS_CHECK_VALUE(XFS_DQ_BIGTIME_EXPIRY_MIN << XFS_DQ_BIGTIME_SHIFT, 4);
+	XFS_CHECK_VALUE(XFS_DQ_BIGTIME_EXPIRY_MAX << XFS_DQ_BIGTIME_SHIFT,
+			16299260424LL);
 }
 
 #endif /* __XFS_ONDISK_H */

fs/xfs/xfs_qm.c

@@ -661,8 +661,15 @@ xfs_qm_init_quotainfo(
 	/* Precalc some constants */
 	qinf->qi_dqchunklen = XFS_FSB_TO_BB(mp, XFS_DQUOT_CLUSTER_SIZE_FSB);
 	qinf->qi_dqperchunk = xfs_calc_dquots_per_chunk(qinf->qi_dqchunklen);
+	if (xfs_sb_version_hasbigtime(&mp->m_sb)) {
+		qinf->qi_expiry_min =
+			xfs_dq_bigtime_to_unix(XFS_DQ_BIGTIME_EXPIRY_MIN);
+		qinf->qi_expiry_max =
+			xfs_dq_bigtime_to_unix(XFS_DQ_BIGTIME_EXPIRY_MAX);
+	} else {
 		qinf->qi_expiry_min = XFS_DQ_LEGACY_EXPIRY_MIN;
 		qinf->qi_expiry_max = XFS_DQ_LEGACY_EXPIRY_MAX;
+	}
 
 	mp->m_qflags |= (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_CHKD);
 
@@ -881,6 +888,8 @@ xfs_qm_reset_dqcounts(
 			ddq->d_bwarns = 0;
 			ddq->d_iwarns = 0;
 			ddq->d_rtbwarns = 0;
+			if (xfs_sb_version_hasbigtime(&mp->m_sb))
+				ddq->d_type |= XFS_DQTYPE_BIGTIME;
 		}
 
 		if (xfs_sb_version_hascrc(&mp->m_sb)) {

fs/xfs/xfs_trans_dquot.c

@@ -55,6 +55,12 @@ xfs_trans_log_dquot(
 {
 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
 
+	/* Upgrade the dquot to bigtime format if possible. */
+	if (dqp->q_id != 0 &&
+	    xfs_sb_version_hasbigtime(&tp->t_mountp->m_sb) &&
+	    !(dqp->q_type & XFS_DQTYPE_BIGTIME))
+		dqp->q_type |= XFS_DQTYPE_BIGTIME;
+
 	tp->t_flags |= XFS_TRANS_DIRTY;
 	set_bit(XFS_LI_DIRTY, &dqp->q_logitem.qli_item.li_flags);
 }