corrupted octal value in tar header

[daleiliu]

I was trying to make tarballs to the subdirectories of a rust vendor directory with this command:
tar jcf ../tarballs/$CRATE.tar.bz2 $CRATE

The command runs OK for all the crates (>200). But later when I tried decompress them, two of them failed:
tar xf tarballs/windows_x86_64_gnu-0.48.5.tar.bz2
tar xf tarballs/windows_x86_64_gnu.tar.bz2
with this error:
tar: corrupted octal value in tar header

The original crate file can be downloaded from:
curl -L https://crates.io/api/v1/crates/windows_x86_64_gnu/0.52.6/download | tar -zxf -

And I found the issue is caused by just one file, lib/libwindows.0.52.0.a

[rmyorston]

I'm unable to reproduce the problem.

This is with the latest busybox-w32 release on 64-bit Windows 10:

~ $ busybox | head -2
BusyBox v1.37.0-FRP-5467-g9376eebd8 (2024-09-15 08:56:36 UTC)
(mingw64-gcc 14.1.1-3.fc40; mingw64-crt 11.0.1-3.fc40; glob; Unicode)
~ $ wget -O - https://crates.io/api/v1/crates/windows_x86_64_gnu/0.52.6/download | tar xfz -
Connecting to crates.io (52.84.90.74:443)
Connecting to static.crates.io (151.101.190.137:443)
writing to stdout
-                    100% |**********************************************************|  816k  0:00:00 ETA
written to stdout
~ $ tar cfj windows_x86_64_gnu-0.52.6.tar.bz2 windows_x86_64_gnu-0.52.6/
~ $ tar tvfj windows_x86_64_gnu-0.52.6.tar.bz2
drwxrwxr-x 4095/4095         0 2024-11-08 07:28:41 windows_x86_64_gnu-0.52.6/
-rw-rw-r-- 4095/4095       119 1970-01-01 00:00:01 windows_x86_64_gnu-0.52.6/.cargo_vcs_info.json
-rw-rw-r-- 4095/4095       198 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/build.rs
-rw-rw-r-- 4095/4095       959 1970-01-01 00:00:01 windows_x86_64_gnu-0.52.6/Cargo.toml
-rw-rw-r-- 4095/4095       327 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/Cargo.toml.orig
drwxrwxr-x 4095/4095         0 2024-11-08 07:28:40 windows_x86_64_gnu-0.52.6/lib/
-rw-rw-r-- 4095/4095  12695688 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/lib/libwindows.0.52.0.a
-rw-rw-r-- 4095/4095     11351 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/license-apache-2.0
-rw-rw-r-- 4095/4095      1141 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/license-mit
drwxrwxr-x 4095/4095         0 2024-11-08 07:28:41 windows_x86_64_gnu-0.52.6/src/
-rwxrwxr-x 4095/4095        11 2006-07-24 02:21:28 windows_x86_64_gnu-0.52.6/src/lib.rs
~ $

[daleiliu]

Thanks for the double check. I did some more tests and found the issue seems about the "tar" auto detection of compression type function. The first command is OK but the second one reports the error:
$ tar jxf windows_x86_64_gnu-0.52.6.tar.bz2
$ tar xf windows_x86_64_gnu-0.52.6.tar.bz2

Also if I compress it from a Ubuntu PC, busybox-w32's "tar xf" generate tar.bz2 file reports the same error. So it should be something in the decompression path.

[rmyorston]

I'm still unable to reproduce the problem on my Windows system. However, I have been able to create tar files on Linux which display the issue. Both GNU tar and BusyBox tar generate files which fail to extract with BusyBox tar on both Windows and Linux.

Now that I have a file that reliably reproduces the problem I'll be able to investigate.

[rmyorston]

The code to autodetect compression was looking for the magic string 'ustar' or a run of five NULs to determine if an archive is uncompressed. In my sample file the bzip2 header had seven NULs, it was therefore considered to be uncompressed.

Looking for a longer run of NULs should avoid the problem. Really old tar files can have 127 NULs, but checking for 16 should be plenty.

New prerelease binaries have been issued with this fix (PRE-5531 or above).