Re: [Libcdio-devel] udf_read_sectors cannot handle offsets past 2 GB

[Pete Batard]

On 2012.01.24 21:15, Pete Batard wrote:
> I just ran a test on an ARM based Linux platform (a Sheevaplug running
> ArmedSlack), that has LFS support, but a lot of the extracted files are
> getting corrupted, such as /sources/readme.rtf (should be mostly ASCII
> text). Many others, such as the > 4GB install.wim, seem to be fine
> though, and I also haven't observed any corruption on x86 so far.

Looks like this occurs when a platform has LFS (=>64 bit off_t) but 
fseeko was not detected.

For some reason on the ARM machine, even as it is available and 
_FILE_OFFSET_BITS is set to 64, fseeko was not detected.
I am able to reproduce the issue on cygwin as well, where off_t is 64 
bit by default (at least on x64 Windows) whereas autoconf doesn't detect 
any LFS settings

I guess that explains why any file past the 2 GB mark in the stream 
would get corrupted, but I'm not sure I understand why the 4.25GB 
install.wim still seemed OK...

If one forces HAVE_FSEEKO, then the files are back to normal on all 
platforms.

I'll try to produce a patch to avoid the issue.

Regards,

/Pete