Skip to contentSkip to conversation

UDF 1GB file size limit after CVE-2006-4145 fix

DISCUSSION Sergey Vlasov 2 participants 2 messages Sep 3, 2006 – Sep 4, 2006 VFS

lore
Share & tools
lore
#0 lore raw

Hello!

Quoted 18 lines · Jan Kara

UDF code is not really ready to handle extents larger that 1GB. This is the easy way to forbid creating those.

Also truncation code did not count with the case when there are no extents in the file and we are extending the file.

---
 fs/udf/super.c    |    2 +-
 fs/udf/truncate.c |   64 ++++++++++++++++++++++++++++++++---------------------
 2 files changed, 40 insertions(+), 26 deletions(-)
fs/udf/super.c +0 -0
index 7de172e..fcce1a2 100644

fs/udf/super.c #endif · line 1662

Quoted 5 lines
fs/udf/super.c +1 -1
@@ -1659,7 +1659,7 @@ #endif
 		iput(inode);
 		goto error_out;
 	}
-	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_maxbytes = 1<<30;

[... rest of patch skipped ...]

After this change the size of files which can be created on an UDF filesystem becomes limited to 1GB. This is very unfortunate - in particular, it means that there will be no way to write a file larger than 4GB to a DVD under Linux (mkisofs -udf does not support files larger than 4GB, so the typical workaround was to use mkudffs and mount -o loop). In fact, this change may be considered as a regression - large files on UDF seemed to work before (at least in simple cases), and now they are forbidden.

Files larger than 1GB can be read even after this patch (because s_maxbytes is not checked in read paths, and udf does not use generic_file_lseek()), so old disks at least can be read.

What issues with files larger than 1GB have been found in the code? Is someone working to fix these problems?

Signature · 1 line
Sergey Vlasov
reviewer
#1 lore raw

Hello,

Quoted 26 lines · Jan Kara
Quoted 18 lines

UDF code is not really ready to handle extents larger that 1GB. This is the easy way to forbid creating those.

Also truncation code did not count with the case when there are no extents in the file and we are extending the file.

---
 fs/udf/super.c    |    2 +-
 fs/udf/truncate.c |   64 ++++++++++++++++++++++++++++++++---------------------
 2 files changed, 40 insertions(+), 26 deletions(-)
fs/udf/super.c +0 -0
index 7de172e..fcce1a2 100644

fs/udf/super.c #endif · line 1662

Quoted 5 lines
fs/udf/super.c +1 -1
@@ -1659,7 +1659,7 @@ #endif
 		iput(inode);
 		goto error_out;
 	}
-	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_maxbytes = 1<<30;

[... rest of patch skipped ...]

Quoted 8 lines

After this change the size of files which can be created on an UDF filesystem becomes limited to 1GB. This is very unfortunate - in particular, it means that there will be no way to write a file larger than 4GB to a DVD under Linux (mkisofs -udf does not support files larger than 4GB, so the typical workaround was to use mkudffs and mount -o loop). In fact, this change may be considered as a regression - large files on UDF seemed to work before (at least in simple cases), and now they are forbidden.

Actually I've been trying this and I have not been able to create file
larger than 1GB on my computer without UDF corrupting slab or doing some other nasty thing. OK, maybe if you created it in 1GB pieces it could work but anyway the problem is that currently if you have UDF rw-mounted, ordinary user could make UDF corrupt kernel memory... So consider this limitation more as a hotfix to the security problem - real fix is to rewrite UDF write path to not create extents larger than 1 GB but that is quite some work and will definitely need more testing.

Quoted 5 lines

Files larger than 1GB can be read even after this patch (because s_maxbytes is not checked in read paths, and udf does not use generic_file_lseek()), so old disks at least can be read.

What issues with files larger than 1GB have been found in the code?

See above.

Quoted 1 line

Is someone working to fix these problems?

Yes, I plan to have a look into a proper fix of this problem (i.e. fix
UDF write path).

Honza

Signature · 5 lines
Jan Kara <[email protected]>
SuSE CR Labs

-- 
VGER BF report: H 0.000799476