[lkml]   [2018]   [Jan]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
SubjectUDF driver is not able to read UDF filesystem without Terminating Descriptor
Hi! In attachment you can find two minimalistic UDF filesystem images.
Strictly speaking they are not compliant to Ecma-167 which requires to
have at least 258 blocks for filesystem, but Linux kernel has no problem
to read also such small UDF filesystems. First one is "udf_with_td", has
89 blocks and udf.ko reads it without problem. Second one is
"udf_without_td", has 88 blocks and udf.ko reject it with error message:

UDF-fs: linux/fs/udf/misc.c:223:udf_read_tagged: location mismatch block 1, tag 0 != 1
UDF-fs: warning (device loop1): udf_fill_super: No fileset found

The only difference between these two images is presence/absence of
Terminating Descriptor (TD).

Looking at the code in function udf_process_sequence(), I found out that
udf.ko does not read Partition Descriptor when filesystem does not
contain Terminating Descriptor. This is of course wrong as without
reading Partition Descriptor it is not possible to locale root directory
of the UDF filesystem.

In that function is following code:

if (vds[VDS_POS_PARTITION_DESC].block) {
* We rescan the whole descriptor sequence to find
* partition descriptor blocks and process them.
for (block = vds[VDS_POS_PARTITION_DESC].block;
block < vds[VDS_POS_TERMINATING_DESC].block;
block++) {
ret = udf_load_partdesc(sb, block);
if (ret < 0)
return ret;

Array vds is initialized to zeros, so when Partition Descriptor is not
present then vds[VDS_POS_TERMINATING_DESC].block is zero and so above
for loop is immediately stopped.

As a quick fix I applied following patch:

--- a/fs/udf/super.c
+++ b/fs/udf/super.c
@@ -1620,6 +1620,7 @@ static noinline int udf_process_sequence(
unsigned int indirections = 0;

memset(vds, 0, sizeof(struct udf_vds_record) * VDS_POS_LENGTH);
+ vds[VDS_POS_TERMINATING_DESC].block = lastblock;

* Read the main descriptor sequence and find which descriptors
which forces above loop to call udf_load_partdesc() also when TD is not

Note that this problem happen also with large enough UDF filesystems
without Terminating Descriptor, not only for those small.

And I have not found any restrictions in UDF or ECMA-167 specification
that Terminating Descriptor is required.

But I'm not sure if above for loop is correct at all.

In function udf_process_sequence() is code for handling TAG_IDENT_VDP,
which allow to scanning nested Volume Descriptor Sequences. And each
identifier descriptor handles (increasing) volDescSeqNum except
TAG_IDENT_PD which is for Partition Descriptor. Partition Descriptor is
always used one which was processed first:

case TAG_IDENT_PD: /* ISO 13346 3/10.5 */
if (!curr->block)
curr->block = block;
But only the last processed Terminating Descriptor is used:

case TAG_IDENT_TD: /* ISO 13346 3/10.9 */
vds[VDS_POS_TERMINATING_DESC].block = block;
And if nested Volume Descriptor Sequences "jumps" to block with smaller
number and in that nested sequence is Terminating Descriptor, it would
have smaller block number as Partition Descriptor block number (read
before nested jump) and so loop for calling udf_load_partdesc() would be
stopped at beginning too.

Also question is, which Partition Descriptor should be loaded? First
processed? Last processed? One with smallest or biggest Sequence Number?
Or all which are found, including all in nested sequence?

Pali Rohár
[unhandled content-type:application/octet-stream][unhandled content-type:application/octet-stream]
 \ /
  Last update: 2018-01-21 13:09    [W:0.042 / U:0.036 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site