tempfile._get_default_tempdir() leaves files behind when HD is full

[kichik]

BPO	16800
Nosy	@birkenfeld, @ncoghlan, @pitrou, @serhiy-storchaka
Files	fix_tempfile_leaving_files_behind.patch: Suggested patch against today's tip 81121:6e2e5adc0400 fix_tempfile_leaving_files_behind_2.patch: Patch for 3.3+ fix_tempfile_leaving_files_behind_2-2.7.patch fix_tempfile_leaving_files_behind_2-3.2.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/serhiy-storchaka'
closed_at = <Date 2013-02-12.23:06:03.867>
created_at = <Date 2012-12-28.05:37:37.621>
labels = ['library', 'performance']
title = 'tempfile._get_default_tempdir() leaves files behind when HD is full'
updated_at = <Date 2013-02-12.23:06:03.865>
user = 'https://bugs.python.org/kichik'

bugs.python.org fields:

activity = <Date 2013-02-12.23:06:03.865>
actor = 'serhiy.storchaka'
assignee = 'serhiy.storchaka'
closed = True
closed_date = <Date 2013-02-12.23:06:03.867>
closer = 'serhiy.storchaka'
components = ['Library (Lib)']
creation = <Date 2012-12-28.05:37:37.621>
creator = 'kichik'
dependencies = []
files = ['28464', '28966', '28968', '28969']
hgrepos = []
issue_num = 16800
keywords = ['patch']
message_count = 9.0
messages = ['178362', '178365', '181469', '181471', '181498', '181587', '181981', '181982', '181983']
nosy_count = 6.0
nosy_names = ['georg.brandl', 'ncoghlan', 'pitrou', 'python-dev', 'serhiy.storchaka', 'kichik']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'resource usage'
url = 'https://bugs.python.org/issue16800'
versions = ['Python 2.7', 'Python 3.2', 'Python 3.3', 'Python 3.4']

[kichik]

tempfile._get_default_tempdir() tries to find a good temporary directory by attempting to create a file and write a string into it for all candidate directories. It deletes those files right after closing them. But if writing rather than creating them fails, the files get left behind. This can happen, for example, when the disk is full on Linux. It will let you create the file with open() and even write() to it. But as soon as close() is called, OSError is raised with errno 28 (No space left on device) and unlink() is never called.

On our system (Linux 2.6.32, ext3), after filling the disk by mistake, we suddenly found dozens of random empty files scattered across the file system. The candidate list, where they were all found, even includes the root directory so the issue is quite visible.

We noticed this problem on Python 2.4.4, but I have been able to reproduce it with today's tip.

A possible solution would be wrapping everything between open() and unlink() in a try-finally block. This way, no matter what happens, the file will get deleted. I have attached a patch implementing this solution and a new test case that reproduces the issue. Verified with make test && make patchcheck.

[kichik]

Reproduction details:

[root@meh /]# uname -a
Linux meh 2.6.32-131.17.1.el6-0.0.3.smp.pae.gcc4.1.x86.i686 #1 SMP Tue Dec 20 02:40:22 PST 2011 i686 GNU/Linux
[root@meh /]# pwd
/
[root@meh /]# dd if=/dev/zero of=/tmp/BIG bs=4M
dd: writing `/tmp/BIG': No space left on device
31+0 records in
30+0 records out
127254528 bytes (127 MB) copied, 3.7028 s, 34.4 MB/s
[root@meh /]# ls
bin  boot  dev  etc  home  initrd  lib  lost+found  media  mnt  opt  proc  root  sbin  srv  sys  tmp  usr  var
[root@meh /]# python -c "import tempfile; tempfile.mktemp()"
Traceback (most recent call last):
  File "<string>", line 1, in ?
  File "/usr/lib/python2.4/tempfile.py", line 362, in mktemp
    dir = gettempdir()
  File "/usr/lib/python2.4/tempfile.py", line 268, in gettempdir
    tempdir = _get_default_tempdir()
  File "/usr/lib/python2.4/tempfile.py", line 214, in _get_default_tempdir
    raise IOError, (_errno.ENOENT,
IOError: [Errno 2] No usable temporary directory found in ['/tmp', '/var/tmp', '/usr/tmp', '/']
[root@meh /]# ls
bin  boot  dev  etc  home  initrd  lib  lost+found  media  mnt  opt  proc  -qE-nI  root  sbin  srv  sys  tmp  usr  var
[root@meh /]#

[serhiy-storchaka]

In general it looks good to me. But we can get rid of Python file and use only os.open/os.write/os.close. Here is an updated patch which carefully close a file descriptor and remove a file even if write or close failed.

Amir Szekely, can you please submit a contributor form?

http://python.org/psf/contrib/contrib-form/
http://python.org/psf/contrib/

[serhiy-storchaka]

And here are patches for 2.7 and 3.2 (there are some peculiarities, i.e. in
exception handling).

[kichik]

I've submitted a contribution form by e-mail.

[pitrou]

There are reasons to use buffered I/O rather than os.write: os.write can fail with EINTR, for example.

[python-dev]

New changeset b368fc93dca8 by Serhiy Storchaka in branch '2.7':
Issue bpo-16800: tempfile.gettempdir() no longer left temporary files when
http://hg.python.org/cpython/rev/b368fc93dca8

New changeset 377123f10820 by Serhiy Storchaka in branch '3.2':
Issue bpo-16800: tempfile.gettempdir() no longer left temporary files when
http://hg.python.org/cpython/rev/377123f10820

New changeset 6f432bb11b28 by Serhiy Storchaka in branch '3.3':
Issue bpo-16800: tempfile.gettempdir() no longer left temporary files when
http://hg.python.org/cpython/rev/6f432bb11b28

New changeset b66a5b41d82f by Serhiy Storchaka in branch 'default':
Issue bpo-16800: tempfile.gettempdir() no longer left temporary files when
http://hg.python.org/cpython/rev/b66a5b41d82f

[python-dev]

New changeset a43f67e95ef0 by Serhiy Storchaka in branch '2.7':
Fix for issue bpo-16800: Use buffered write to handle EINTR.
http://hg.python.org/cpython/rev/a43f67e95ef0

New changeset 4622206db91b by Serhiy Storchaka in branch '3.2':
Fix for issue bpo-16800: Use buffered write to handle EINTR.
http://hg.python.org/cpython/rev/4622206db91b

New changeset 2fb03fe354e3 by Serhiy Storchaka in branch '3.3':
Fix for issue bpo-16800: Use buffered write to handle EINTR.
http://hg.python.org/cpython/rev/2fb03fe354e3

New changeset fec33725f319 by Serhiy Storchaka in branch 'default':
Fix for issue bpo-16800: Use buffered write to handle EINTR.
http://hg.python.org/cpython/rev/fec33725f319

[serhiy-storchaka]

Thank you for explanation, Antoine.

Thank you for your contribution, Amir.