Skip to contentSkip to conversation

xfs_mdrestore: don't allow restoring onto zoned block devices

PATCH In review Christoph Hellwig 3 participants 4 messages May 12, 2025 – May 13, 2025

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs_mdrestore: don't allow restoring onto zoned block devices

The way mdrestore works is not very amendable to zone devices. The code that checks the device size tries to write to the highest offset, which doesn't match the write pointer of a clean zone device. And while that is relatively easily fixable, the metadata for each RTG records the highest written offset, and the mount code compares that to the hardware write pointer, which will mismatch. This could be fixed by using write zeroes to pad the RTG until the expected write pointer, but this turns the quick metadata operation that mdrestore is supposed to be into something that could take hours on HDD.

So instead error out when someone tries to mdrestore onto a zoned device to clearly document that this won't work. Doing a mdrestore into a file still works perfectly fine, and we might look into a new mdrestore option to restore into a set of files suitable for the zoned loop device driver to make mdrestore fully usable for debugging.

---
 mdrestore/xfs_mdrestore.c | 8 ++++++++
 1 file changed, 8 insertions(+)
mdrestore/xfs_mdrestore.c +8 -0
index 95b01a99a154..f10c4befb2fc 100644
@@ -8,6 +8,7 @@
 #include "xfs_metadump.h"
 #include <libfrog/platform.h>
 #include "libfrog/div64.h"
+#include <linux/blkzoned.h>
 
 union mdrestore_headers {
 	__be32				magic;
@@ -148,6 +149,13 @@ open_device(
 	dev->fd = open(path, open_flags, 0644);
 	if (dev->fd < 0)
 		fatal("couldn't open \"%s\"\n", path);
+
+	if (!dev->is_file) {
+		uint32_t zone_size;
+
+		if (ioctl(dev->fd, BLKGETZONESZ, &zone_size) == 0 && zone_size)
+			fatal("can't restore to zoned device \"%s\"\n", path);
+	}
 }
 
 static void
Signature · 1 line
2.47.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 26 lines · Christoph Hellwig

The way mdrestore works is not very amendable to zone devices. The code that checks the device size tries to write to the highest offset, which doesn't match the write pointer of a clean zone device. And while that is relatively easily fixable, the metadata for each RTG records the highest written offset, and the mount code compares that to the hardware write pointer, which will mismatch. This could be fixed by using write zeroes to pad the RTG until the expected write pointer, but this turns the quick metadata operation that mdrestore is supposed to be into something that could take hours on HDD.

So instead error out when someone tries to mdrestore onto a zoned device to clearly document that this won't work. Doing a mdrestore into a file still works perfectly fine, and we might look into a new mdrestore option to restore into a set of files suitable for the zoned loop device driver to make mdrestore fully usable for debugging.

---
 mdrestore/xfs_mdrestore.c | 8 ++++++++
 1 file changed, 8 insertions(+)
mdrestore/xfs_mdrestore.c +0 -0
index 95b01a99a154..f10c4befb2fc 100644

mdrestore/xfs_mdrestore.c line 11

Quoted 4 lines
mdrestore/xfs_mdrestore.c +1 -0
@@ -8,6 +8,7 @@
 #include "xfs_metadump.h"
 #include <libfrog/platform.h>
 #include "libfrog/div64.h"
+#include <linux/blkzoned.h>

I wonder if there ought to be guards around blkzoned.h, but OTOH that seems to have been introduced in 4.9 around 8 years ago so maybe it's fine?

/me is willing to go along with that if the maintainer is. Meanwhile the code changes make sense so as long as there isn't some "set the write pointer to an arbitrary LBA" command that I missed,

since we wouldn't be mdrestoring user file data to the zoned section, right?

--D

Quoted 21 lines
mdrestore/xfs_mdrestore.c +7 -0
 union mdrestore_headers {
 	__be32				magic;
@@ -148,6 +149,13 @@ open_device(
 	dev->fd = open(path, open_flags, 0644);
 	if (dev->fd < 0)
 		fatal("couldn't open \"%s\"\n", path);
+
+	if (!dev->is_file) {
+		uint32_t zone_size;
+
+		if (ioctl(dev->fd, BLKGETZONESZ, &zone_size) == 0 && zone_size)
+			fatal("can't restore to zoned device \"%s\"\n", path);
+	}
 }
 
 static void
Signature · 1 line
2.47.2
#2 lore raw
Quoted 3 lines · Darrick J. Wong

I wonder if there ought to be guards around blkzoned.h, but OTOH that seems to have been introduced in 4.9 around 8 years ago so maybe it's fine?

I think we're fine. If not we'll also need to do this for mkfs and repair anyway.

Quoted 3 lines

/me is willing to go along with that if the maintainer is. Meanwhile the code changes make sense so as long as there isn't some "set the write pointer to an arbitrary LBA" command that I missed,

There isn't. It would have been really useful for testing, though..

Quoted 4 lines

since we wouldn't be mdrestoring user file data to the zoned section, right?

Well, mdrestore can't even restore user data :)

#3 lore raw
Quoted 9 lines · Christoph Hellwig
Quoted 3 lines · Darrick J. Wong

I wonder if there ought to be guards around blkzoned.h, but OTOH that seems to have been introduced in 4.9 around 8 years ago so maybe it's fine?

I think we're fine. If not we'll also need to do this for mkfs and repair anyway.

Quoted 1 line

/me is willing to go along with that if the maintainer is. Meanwhile

Fine by me, looks reasonable

Quoted 13 lines
Quoted 2 lines

the code changes make sense so as long as there isn't some "set the write pointer to an arbitrary LBA" command that I missed,

There isn't. It would have been really useful for testing, though..

Quoted 4 lines

since we wouldn't be mdrestoring user file data to the zoned section, right?

Well, mdrestore can't even restore user data :)

Signature · 1 line
- Andrey