xfs: fix log CRC mismatches between i386 and other architectures

When mounting file systems with a log that was dirtied on i386 on
other architectures or vice versa, log recovery is unhappy:

[   11.068052] XFS (vdb): Torn write (CRC failure) detected at log block 0x2. Truncating head block from 0xc.

This is because the CRCs generated by i386 and other architectures
always diff.  The reason for that is that sizeof(struct xlog_rec_header)
returns different values for i386 vs the rest (324 vs 328), because the
struct is not sizeof(uint64_t) aligned, and i386 has odd struct size
alignment rules.

This issue goes back to commit 13cdc853c519 ("Add log versioning, and new
super block field for the log stripe") in the xfs-import tree, which
adds log v2 support and the h_size field that causes the unaligned size.
At that time it only mattered for the crude debug only log header
checksum, but with commit 0e446be ("xfs: add CRC checks to the log")
it became a real issue for v5 file system, because now there is a proper
CRC, and regular builds actually expect it match.

Fix this by allowing checksums with and without the padding.

Fixes: 0e446be ("xfs: add CRC checks to the log")
Cc: <[email protected]> # v3.8
Signed-off-by: Christoph Hellwig <[email protected]>
Signed-off-by: Carlos Maiolino <[email protected]>

Author: Christoph Hellwig

fs/xfs/libxfs/xfs_log_format.h

@@ -173,12 +173,40 @@ typedef struct xlog_rec_header {
 	__be32	  h_prev_block; /* block number to previous LR		:  4 */
 	__be32	  h_num_logops;	/* number of log operations in this LR	:  4 */
 	__be32	  h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
-	/* new fields */
+
+	/* fields added by the Linux port: */
 	__be32    h_fmt;        /* format of log record                 :  4 */
 	uuid_t	  h_fs_uuid;    /* uuid of FS                           : 16 */
+
+	/* fields added for log v2: */
 	__be32	  h_size;	/* iclog size				:  4 */
+
+	/*
+	 * When h_size added for log v2 support, it caused structure to have
+	 * a different size on i386 vs all other architectures because the
+	 * sum of the size ofthe  member is not aligned by that of the largest
+	 * __be64-sized member, and i386 has really odd struct alignment rules.
+	 *
+	 * Due to the way the log headers are placed out on-disk that alone is
+	 * not a problem becaue the xlog_rec_header always sits alone in a
+	 * BBSIZEs area, and the rest of that area is padded with zeroes.
+	 * But xlog_cksum used to calculate the checksum based on the structure
+	 * size, and thus gives different checksums for i386 vs the rest.
+	 * We now do two checksum validation passes for both sizes to allow
+	 * moving v5 file systems with unclean logs between i386 and other
+	 * (little-endian) architectures.
+	 */
+	__u32	  h_pad0;
 } xlog_rec_header_t;
 
+#ifdef __i386__
+#define XLOG_REC_SIZE		offsetofend(struct xlog_rec_header, h_size)
+#define XLOG_REC_SIZE_OTHER	sizeof(struct xlog_rec_header)
+#else
+#define XLOG_REC_SIZE		sizeof(struct xlog_rec_header)
+#define XLOG_REC_SIZE_OTHER	offsetofend(struct xlog_rec_header, h_size)
+#endif /* __i386__ */
+
 typedef struct xlog_rec_ext_header {
 	__be32	  xh_cycle;	/* write cycle of log			: 4 */
 	__be32	  xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /*	: 256 */

fs/xfs/libxfs/xfs_ondisk.h

@@ -174,6 +174,8 @@ xfs_check_ondisk_structs(void)
 	XFS_CHECK_STRUCT_SIZE(struct xfs_rud_log_format,	16);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_map_extent,		32);
 	XFS_CHECK_STRUCT_SIZE(struct xfs_phys_extent,		16);
+	XFS_CHECK_STRUCT_SIZE(struct xlog_rec_header,		328);
+	XFS_CHECK_STRUCT_SIZE(struct xlog_rec_ext_header,	260);
 
 	XFS_CHECK_OFFSET(struct xfs_bui_log_format, bui_extents,	16);
 	XFS_CHECK_OFFSET(struct xfs_cui_log_format, cui_extents,	16);

fs/xfs/xfs_log.c

@@ -1568,13 +1568,13 @@ xlog_cksum(
 	struct xlog		*log,
 	struct xlog_rec_header	*rhead,
 	char			*dp,
-	int			size)
+	unsigned int		hdrsize,
+	unsigned int		size)
 {
 	uint32_t		crc;
 
 	/* first generate the crc for the record header ... */
-	crc = xfs_start_cksum_update((char *)rhead,
-			      sizeof(struct xlog_rec_header),
+	crc = xfs_start_cksum_update((char *)rhead, hdrsize,
 			      offsetof(struct xlog_rec_header, h_crc));
 
 	/* ... then for additional cycle data for v2 logs ... */
@@ -1818,7 +1818,7 @@ xlog_sync(
 
 	/* calculcate the checksum */
 	iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
-					    iclog->ic_datap, size);
+			iclog->ic_datap, XLOG_REC_SIZE, size);
 	/*
 	 * Intentionally corrupt the log record CRC based on the error injection
 	 * frequency, if defined. This facilitates testing log recovery in the

fs/xfs/xfs_log_priv.h

@@ -499,8 +499,8 @@ xlog_recover_finish(
 extern void
 xlog_recover_cancel(struct xlog *);
 
-extern __le32	 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
-			    char *dp, int size);
+__le32	 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
+		char *dp, unsigned int hdrsize, unsigned int size);
 
 extern struct kmem_cache *xfs_log_ticket_cache;
 struct xlog_ticket *xlog_ticket_alloc(struct xlog *log, int unit_bytes,

fs/xfs/xfs_log_recover.c

@@ -2894,9 +2894,24 @@ xlog_recover_process(
 	int			pass,
 	struct list_head	*buffer_list)
 {
-	__le32			expected_crc = rhead->h_crc, crc;
+	__le32			expected_crc = rhead->h_crc, crc, other_crc;
 
-	crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
+	crc = xlog_cksum(log, rhead, dp, XLOG_REC_SIZE,
+			be32_to_cpu(rhead->h_len));
+
+	/*
+	 * Look at the end of the struct xlog_rec_header definition in
+	 * xfs_log_format.h for the glory details.
+	 */
+	if (expected_crc && crc != expected_crc) {
+		other_crc = xlog_cksum(log, rhead, dp, XLOG_REC_SIZE_OTHER,
+				be32_to_cpu(rhead->h_len));
+		if (other_crc == expected_crc) {
+			xfs_notice_once(log->l_mp,
+	"Fixing up incorrect CRC due to padding.");
+			crc = other_crc;
+		}
+	}
 
 	/*
 	 * Nothing else to do if this is a CRC verification pass. Just return