tarfile-write issue with file/folder permissions
ubidefeo opened this issue · comments
Context
I've been playing with tarfile
for a few days and yesterday started trying to create my own archives using tarfile-write
installed via mip
.
Adding a folder to it works recursively, but transferring the file to the PC via mpremote
and expanding it yields all files and folders with --- --- ---
permissions (0x000).
Discovery
I investigated and found the source of the issue is related to how the TarInfo
object is composed in these lines
in particular on line 102 we find mode
to be set to the os.stat(FILE)[0]
tarinfo.mode = stat[0]
stat[0]
is either 0x8000
(file) or 0x4000
(folder).
And the subsequently applied masking yields 0x000
as permissions set while adding it to the archive.
I think that mode should be safely set to 0x1a4 > 644 > rw- r-- r--
I have tested by manually composing a TarInfo object to be handed to addfile()
rather than patching add()
How to reproduce:
import tarfile
import os
os.mkdir('test_folder')
f = open('test_folder/test_file_01.txt', 'w')
f.write('file 01')
f.close()
os.mkdir('test_folder/sub_folder')
f = open('test_folder/sub_folder/test_file_02.txt', 'w')
f.write('file 02')
f.close()
archive = tarfile.TarFile('test_archive.tar', 'w')
archive.add('test_folder')
archive.close()
Copy the archive from the board to the PC via mpremote cp :test_archive.tar test_archive.tar
Expand the archive and verify the permissions for test_folder.
From Mac os I was not able to chmod 644 test_folder
, but using the UI I could unrestrict access.
Looking at the folder content you can verify the same permission issue happens with sub_folder
and contained file(s)
Environment:
I run a custom build for ESP32-S3, but this issue is not related to changes I have applied
(name='micropython', version=(1, 23, 0, 'preview'), _machine='LilyGo T-QT Pro with ESP32S3', _mpy=10758)
MicroPython v1.23.0-preview.48.g076516d88.dirty on 2024-01-21; LilyGo T-QT Pro with ESP32S3
Board: ESP32-S3
TarFile version: '0.4.1'
Hope someone can take a look :)
u.
upon further digging, the .mode
attribute seems to be doing its thing when it comes to TarInfo
, and the issue lies into how the Tar header is composed.
Dissecting this line
hdr.mode[:] = b"%07o\0" % (tarinfo.mode & 0o7777)
the permissions always come out as 000
The fix I have tested is to replace that line with
hdr.mode[:] = b"%07o\0" % (0x1ed & 0o7777)
This yields tar files which can be expanded on the host computer, but of course all the files will be 755
, which might have some implications if one wants to be picky
Moreover, the way the initial mode
is set in the TarInfo
init is wrong, since it characterises the mode
based on the presence of a /
at the end of the resource name.
The correct way to do this is to check for os.stat(name)[0]
- self.mode = _S_IFDIR if self.name[-1] == "/" else _S_IFREG
+ self.mode = _S_IFDIR if os.stat(name)[0] == 0x4000 else _S_IFREG
I left a review on your PR, but I wanted to say thanks for a very comprehensive and clear bug report and analysis!