[REVIEW] Fix end of device zeroing with > 1K sector sizes

[REVIEW] Fix end of device zeroing with > 1K sector sizes

[Barry Naujok]

[-- Attachment #1: Type: text/plain, Size: 481 bytes --]

It has been found that mkfs with big sector sizes and devices/partitions
that aren't aligned to the sector size can cause short writes when zeroing
the end of the device. As mkfs.xfs didn't complete, the filesystem is
unmountable.

Eg:

# mkfs.xfs -f -L fs_1 -s size=4096 -l size=64m,sunit=512 /dev/hdd1
mkfs.xfs: error - wrote only 130048 of 131072 bytes


The attached patch rounds down the device size to either 1K or
the sector size, whichever is greater.

Barry.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: fix_eodev_zeroing.patch --]
[-- Type: text/x-patch; name=fix_eodev_zeroing.patch, Size: 1248 bytes --]

--- a/xfsprogs/mkfs/xfs_mkfs.c	2008-02-22 14:48:27.000000000 +1100
+++ b/xfsprogs/mkfs/xfs_mkfs.c	2008-02-22 14:41:38.093984235 +1100
@@ -703,6 +703,7 @@ main(
 	int			sectoralign;
 	int			sectorlog;
 	unsigned int		sectorsize;
+	__uint64_t		sector_mask;
 	int			slflag;
 	int			ssflag;
 	__uint64_t		tmp_agsize;
@@ -1637,12 +1638,18 @@ main(
 	 * Ok, Linux only has a 1024-byte resolution on device _size_,
 	 * and the sizes below are in basic 512-byte blocks,
 	 * so if we have (size % 2), on any partition, we can't get
-	 * to the last 512 bytes.  Just chop it down by a block.
+	 * to the last 512 bytes.  The same issue exists for larger
+	 * sector sizes - we cannot write past the last sector.
+	 *
+	 * So, we reduce the size (in basic blocks) to a perfect
+	 * multiple of the sector size, or 1024, whichever is larger.
 	 */
 
-	xi.dsize -= (xi.dsize % 2);
-	xi.rtsize -= (xi.rtsize % 2);
-	xi.logBBsize -= (xi.logBBsize % 2);
+	sector_mask = xfs_mask64hi((sectorlog < 10 ? 10 : sectorlog) - BBSHIFT);
+	xi.dsize &= sector_mask;
+	xi.rtsize &= sector_mask;
+	xi.logBBsize &= xfs_mask64hi((lsectorlog < 10 ? 10 : lsectorlog) -
+				BBSHIFT);
 
 	if (!force_overwrite) {
 		if (check_overwrite(dfile) ||

Re: [REVIEW] Fix end of device zeroing with > 1K sector sizes

[Eric Sandeen]

Barry Naujok wrote:
> It has been found that mkfs with big sector sizes and devices/partitions
> that aren't aligned to the sector size can cause short writes when zeroing
> the end of the device. As mkfs.xfs didn't complete, the filesystem is
> unmountable.
> 
> Eg:
> 
> # mkfs.xfs -f -L fs_1 -s size=4096 -l size=64m,sunit=512 /dev/hdd1
> mkfs.xfs: error - wrote only 130048 of 131072 bytes
> 
> 
> The attached patch rounds down the device size to either 1K or
> the sector size, whichever is greater.

Hopefully to a multiple thereof ;)

Looks good to me... and strangely familiar!

-Eric

> Barry.
>