Packages: python3: UnicodeDecodeError on calling socket.gethostbyaddr("hostname")

[FlorianFieber]

I've stumbled over an UnicodeDecodeError which backtracks to socket.getfqdn() calling socket.gethostbyaddr("hostname").

Python3:

spam@egg:~$ python3
Python 3.4.3 (default, May 26 2015, 12:25:24) 
[GCC 4.8.3] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import socket
>>> socket.getfqdn()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.4/socket.py", line 463, in getfqdn
    hostname, aliases, ipaddrs = gethostbyaddr(name)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc0 in position 0: invalid start byte
>>> socket.gethostbyaddr("egg")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc0 in position 0: invalid start byte

Whereas it's working in Python2:

spam@egg:~$ python2
Python 2.7.9 (default, May  1 2015, 23:21:05) 
[GCC 4.8.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import socket
>>> socket.getfqdn()
'spam.egg'
>>> socket.gethostbyaddr("egg")
('spam.egg', ['\xc0\xa8\x01\x01'], ['192.168.1.1'])

[commodo]

will look into it;
by default python3 uses unicode, whereas python2 uses ASCII as default encoding;
it could be a bug in python3

[commodo]

@FlorianFieber
Just re-tried again with latest trunk.
Seems to work now.
I wanted to update my trunk build before getting to investigate deeper.
Most likely something was fixed/changed underneath the hood (DNS stuff in OpenWRT), because Python3 was unchanged.

Mind to retry ?
I know this may be a cheap fix, in the sense that this issue might still reside somewhere in Python3.
In my case, I tested with x86 (both 32 & 64 bit versions, to make sure it's not a 32-64-issue).

If this re-appears I'll dig deeper into it, but with this one, it seems like it was either a pythnon3 bug, or something was not agree-ing with python3.

[FlorianFieber]

I happened to build and flash the latest trunk with NLS (full language support) for mips yesterday. The problem persists:

spam@egg:~# python3 -c "import socket;print(socket.getfqdn())"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/lib/python3.4/socket.py", line 463, in getfqdn
    hostname, aliases, ipaddrs = gethostbyaddr(name)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc0 in position 0: invalid start byte

[commodo]

Ok.
Will check further.

Thanks

[commodo]

Right; so this is a heart-beat message (i.e. to let people know I haven't forgot about this).
Just time constraints are (well...) constraining me to allocate not so much time on this.

Last time I checked this (and did not reproduce it) was on my home network.
Managed to re-reproduce this on another network; seems it's important that gethostbyaddr() returns some value, otherwise the system's hostname will be printed.
Not sure when a fix for this will be found.

[ziadsawalha]

I was able to work around this by making sure my /etc/resolve.conf had a _valid_ nameserver in it (see bug).

I changed my default /etc/resolve.conf from:

search lan
nameserver 127.0.0.1

to

search lan
nameserver 192.168.1.55  #< this is my ISP connection's DNS address (modified for this post)
nameserver 127.0.0.1

I'm not sure what was _invalid_ about 127.0.0.1 as it responds if queried.

[commodo]

@FlorianFieber
can you retry again ? :p
[ i know it's been 2 years ]

this looks a lot like an issue that's more related to Python3 or libc
i tried real quick now ; and it looks to be running ok [or at least consistently ]
Python & Python3 have run through a few upgrades, musl as well [ I don't remember if it was uClibc that was being used in the initial report ]

root@LEDE:/#  python3 -c "import socket;print(socket.getfqdn('google.com'))"
bud02s23-in-f206.1e100.net
root@LEDE:/#  python -c "import socket;print(socket.getfqdn('google.com'))"
bud02s23-in-f206.1e100.net

root@LEDE:/#  python -c "import socket;print(socket.getfqdn())"
LEDE.lan
root@LEDE:/#  python3 -c "import socket;print(socket.getfqdn())"
LEDE.lan

root@LEDE:/#  python3 -c "import socket;print(socket.getfqdn('8.8.8.8'))"
google-public-dns-a.google.com
root@LEDE:/# ^C
root@LEDE:/#  python -c "import socket;print(socket.getfqdn('8.8.8.8'))"
google-public-dns-a.google.com

[FlorianFieber]

I'm sorry for the late reply. Sadly, I don't have access to a flashable OpenWRT router right now to test it.

[commodo]

no worries :)

[CarlEdman]

Hi guys, just posting to confirm that Florian is not the only one experiencing this problem. Under OpenWrt Chaos Calmer 15.05.1, the gethostbyaddr() (and as a consequence getfqdn() and much more) in standard Python 3.4 from opkg will crash with a UnicodeDecodeError for almost address:

% python3.4 -c "import _socket; print(_socket.gethostbyaddr('google.com'))"
Traceback (most recent call last):
File "", line 1, in
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xf8 in position 2: invalid start byte

The sole exception I found was for localhost:
% python3.4 -c "import _socket; print(_socket.gethostbyaddr('localhost'))"
('localhost', [], ['127.0.0.1'])

If I was to guess, the reason is that the series of bytes 127, 0, 0, 1 is valid ASCII (and hence Unicode) while any IP address which contains any byte with the high bit set may not be.

[poranje]

Could have to do with the locale setting. See nice article in LWN https://lwn.net/Articles/741176/
Paul

Op 23 dec. 2017, om 13:53 heeft Carl Edman [e-mail] het volgende geschreven:

Hi guys, just posting to confirm that Florian is not the only one experiencing this problem. Under OpenWrt Chaos Calmer 15.05.1, the gethostbyaddr() (and as a consequence getfqdn() and much more) in standard Python 3.4 from opkg will crash with a UnicodeDecodeError for almost address:

python3.4 -c "import _socket; print(_socket.gethostbyaddr('google.com'))"

Traceback (most recent call last):
File "", line 1, in
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xf8 in position 2: invalid start byte

The sole exception I found was for localhost:

python3.4 -c "import _socket; print(_socket.gethostbyaddr('localhost'))"

('localhost', [], ['127.0.0.1'])

If I was to guess, the reason is that the series of bytes 127, 0, 0, 1 is valid ASCII (and hence Unicode) while any IP address which contains any byte with the high bit set may not be.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

[commodo]

@CarlEdman
I think the issue may have been fixed in Python/Python3 here:
https://bugs.python.org/issue26227
I'm not completely sure that, it's the same issue as here ; I could try.

I don't know if ChaosCalmer is supported anymore, and I can't find a build infrastructure that would build another package.
Technically, CC is using Python3 ver 3.4.3 ; ver 3.4.7 is the more recent one, and I could probably backport the patch from the Python bugs-site.

If this issue is still of interest for CC, I could try to see about fixing it.
My preference would be to upgrade to 17.01 if possible.

[commodo]

Technically, CC is using Python3 ver 3.4.3 ; ver 3.4.7 is the more recent one, and I could probably backport the patch from the Python bugs-site.

3.4.7 is the more recent from the 3.4.x series

3.6.4 is the more recent one

[CarlEdman]

@commodo Thanks for the response! I'm planning to upgrade to a new version one of these days anyways. If it requires more than minimal work on your part, please don't bother to fix my issue until I can confirm it happens on a current version of OpenWRT (hopefully after the eagerly awaited LEDE merger; that is still happening, right?).

[commodo]

First of all: Happy New Year :)

I tried upgrading 3.4.3 to 3.4.7 and applying the patch from : https://bugs.python.org/issue26227
It did not help ; it still fails like

One next thing I could try, is to upgrade to 3.6.4 and try it, but I'm not sure about upgrading it officially into the 15.05 version.
On LEDE trunk with 3.6.x [and 3.5.x] it seems to work [the issue does not reproduce], but it could also work due to a newer kernel. 15.05 is on kernel 3.18 [i think].

[commodo]

@CarlEdman
Python 3.6.4 behaves the same.

root@OpenWrt:/# python3 -V
Python 3.6.4
root@OpenWrt:/# python3 -c "import socket;print(socket.getfqdn())"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/lib/python3.6/socket.py", line 673, in getfqdn
    hostname, aliases, ipaddrs = gethostbyaddr(name)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xfd in position 0: invalid start byte
root@OpenWrt:/# uname -a
Linux OpenWrt 3.18.84 #6 SMP Tue Jan 2 16:59:13 EET 2018 x86_64 GNU/Linux

I just noticed that uClibc is used vs musl in OpenWrt 15.05.
At this point, I am not sure if it's libc or kernel version.

I guess at this point, a full system upgrade is required to make this work.

[CarlEdman]

Just a quick update in case anybody is interested. I just bought a new router (Linksys WRT3200ACM, $119 refurb from Amazon) which actually has functional WiFi under OpenWRT and installed the latest stable firmware (LEDE Reboot 17.01.4 r3560-79f57e422d / LuCI lede-17.01 branch (git-18.061.17832-d092772) ) and python 3.6.0.

And now the same code works just fine:

# python3 -c "import _socket;print(_socket.gethostbyaddr('google.com'))"
('iad23s58-in-f14.1e100.net', ['iad23s58-in-f14.1e100.net'], ['172.217.7.238'])

[commodo]

ah ; 3 years later here :)
shall we close this and re-open if needed ? :)
@FlorianFieber @hnyman ?