On Mon, 2011-08-08 at 12:59 +1000, Dave Chinner wrote:
> From: Dave Chinner <dchinner@xxxxxxxxxx>
>
> Because repair uses direct IO, it cannot do IO smaller than a sector
> on the underlying device. When repairing a filesystem image, the
> filesystem hosting the image may have a sector size larger than the
> sector size of the image, and so single image sector reads and
> writes will fail.
>
> To avoid this, when checking a file and there is a sector size
> mismatch like this, turn off direct IO. While there, fix a compile
> bug in the IO_DEBUG option for libxfs which was found during triage.
>
> Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>
Looks good, but I have a few pretty minor things I think
you should consider fixing before you commit.
Reviewed-by: Alex Elder <aelder@xxxxxxx>
. . .
> diff --git a/repair/sb.c b/repair/sb.c
> index 0ee2345..c8df076 100644
> --- a/repair/sb.c
> +++ b/repair/sb.c
> @@ -689,7 +689,14 @@ verify_set_primary_sb(xfs_sb_t *rsb,
> */
> num_sbs = MIN(NUM_SBS, rsb->sb_agcount);
> skip = howmany(num_sbs, rsb->sb_agcount);
> - size = NUM_AGH_SECTS * rsb->sb_sectsize;
> +
> + /*
> + * We haven't been able to validate the sector size yet properly
> + * (e.g. in the case of repairing an image in a file), so we need to
> + * take into account sector mismatches and so use the maximum possible
> + * sector size rather than that in @rsb.
How about "rather than the sector size in @rsb".
What you have is OK, but my parser had to backtrack
to be sure I knew what "that" was referring to.
> + */
> + size = NUM_AGH_SECTS * (1 << (XFS_MAX_SECTORSIZE_LOG));
> retval = 0;
> list = NULL;
> num_ok = 0;
> diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> index 4707b83..dda9daa 100644
> --- a/repair/xfs_repair.c
> +++ b/repair/xfs_repair.c
> @@ -569,6 +569,36 @@ main(int argc, char **argv)
> memset(&xfs_m, 0, sizeof(xfs_mount_t));
> libxfs_sb_from_disk(&xfs_m.m_sb, XFS_BUF_TO_SBP(sbp));
>
> + /*
> + * if the sector size of the filesystem we are trying to repair is
> + * smaller than that of the underlying filesystem (i.e. we are repairing
> + * an image), the we have to turn off direct IO because we cannot do IO
> + * smaller than the host filesystem's sector size.
> + */
> + if (isa_file) {
> + int fd = libxfs_device_to_fd(x.ddev);
> + struct xfs_fsop_geom_v1 geom = {
> + .sectsize = BBSIZE,
Why do you fill in sectsize here? I don't know that
you're guaranteed it will still be intact if the ioctl()
below fails. If that's your reason I'd prefer to see it
filled in after the do_warn() call.
> + };
> +
> + if (ioctl(fd, XFS_IOC_FSGEOMETRY_V1, &geom) < 0) {
> + do_warn(_("Cannot get host fiesystem geometry.\n"
filesystem
> + "Repair may fail if there is a sector size mismatch between "
Put a newline at the end of this line too...
> + "the image and the host filesystem.\n"));
...otherwise the combination is a pretty long line.
> + }
> +
> + if (xfs_m.m_sb.sb_sectsize < geom.sectsize) {
> + long old_flags;
> +
> + old_flags = fcntl(fd, F_GETFL, 0);
Maybe don't bother turning off direct I/O if it's not
already set in old_flags.
> + if (fcntl(fd, F_SETFL, old_flags & ~O_DIRECT) < 0) {
> + do_warn(_(
> + "Sector size on host filesystem larger than image sector
> size.\n"
> + "Cannot turn off direct IO, so exiting.\n"));
> + exit(1);
> + }
> + }
> + }
> mp = libxfs_mount(&xfs_m, &xfs_m.m_sb, x.ddev, x.logdev, x.rtdev, 0);
>
> if (!mp) {
|