PEP 524: Add os.getrandom()

[vstinner]

BPO	27778
Nosy	@ncoghlan, @vstinner, @tiran, @vadmium
PRs	[Do Not Merge] Convert Misc/NEWS so that it is managed by towncrier #552
Files	getrandom.patch getrandom_errno.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 = None
closed_at = <Date 2016-10-17.16:21:08.288>
created_at = <Date 2016-08-16.17:25:41.671>
labels = ['3.7', 'type-feature', 'library', 'docs']
title = 'PEP 524: Add os.getrandom()'
updated_at = <Date 2017-03-31.16:36:09.313>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2017-03-31.16:36:09.313>
actor = 'dstufft'
assignee = 'docs@python'
closed = True
closed_date = <Date 2016-10-17.16:21:08.288>
closer = 'vstinner'
components = ['Documentation', 'Library (Lib)']
creation = <Date 2016-08-16.17:25:41.671>
creator = 'vstinner'
dependencies = []
files = ['44127', '44760']
hgrepos = []
issue_num = 27778
keywords = ['patch']
message_count = 17.0
messages = ['272867', '273024', '274662', '274698', '274717', '274722', '274724', '274730', '274733', '276323', '276356', '276386', '276389', '277068', '277070', '277335', '278815']
nosy_count = 6.0
nosy_names = ['ncoghlan', 'vstinner', 'christian.heimes', 'docs@python', 'python-dev', 'martin.panter']
pr_nums = ['552']
priority = 'low'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue27778'
versions = ['Python 3.6', 'Python 3.7']

[vstinner]

Attached patch adds os.getrandom(): thin wrapper on the Linux getrandom() syscall.

os.getrandom() can return less bytes than requested.

The patch is incomplete: it doesn't include documentation.

I chose to not implement a loop to not loose entropy if a following call fails (ex: fail with EINTR). Rationale:
https://mail.python.org/pipermail/security-sig/2016-July/000072.html

We should also add Solaris support later.

See also bpo-27776: "PEP-524: Make os.urandom() blocking on Linux".

[ncoghlan]

Given docs (with the Linux-only platform support disclaimer), +1 for this as an initial implementation.

Providing it on Solaris as well can be a separate patch, but it's less important there (since /dev/urandom and os.urandom() are already blocking APIs)

[python-dev]

New changeset 27267d2fb091 by Victor Stinner in branch 'default':
Add os.getrandom()
https://hg.python.org/cpython/rev/27267d2fb091

[vadmium]

HAVE_GETRANDOM_SYSCALL seems to be a compile-time library check, not a runtime check. I compiled and run on Linux 3.15.5, and os.getrandom() exists but raises ENOSYS:

======================================================================
ERROR: test_getrandom0 (test.test_os.GetRandomTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/media/disk/home/proj/python/cpython/Lib/test/test_os.py", line 1280, in test_getrandom0
    empty = os.getrandom(0)
OSError: [Errno 38] Function not implemented

[ncoghlan]

Huh, I thought I'd already filed an issue for that, but it looks like it was only a security-sig thread: https://mail.python.org/pipermail/security-sig/2016-June/000060.html

I've now remedied that omission and filed http://bugs.python.org/issue27990 to cover it explicitly.

Since that was a pre-existing problem that also happens to affect this API, rather than something new introduced by Victor's patch, closing this again.

[python-dev]

New changeset 7a243a40b421 by Victor Stinner in branch 'default':
Fix test_os.GetRandomTests()
https://hg.python.org/cpython/rev/7a243a40b421

[vstinner]

HAVE_GETRANDOM_SYSCALL seems to be a compile-time library check, not a runtime check. I compiled and run on Linux 3.15.5, and os.getrandom() exists but raises ENOSYS:

Oh, I'm surprised the configure sees getrandom() as available. But well ok, the error can occur if you compile Python on a more recent kernel than the running kernel.

I fixed the unit test: skip getrandom() tests if getrandom() fails with ENOSYS.

Do you think that it's worth to document this case?

[ncoghlan]

Ah, I'd missed that Martin was talking about the other way around from bpo-27990.

Yes, I think it's worth documenting that os.getrandom() may raise OSError if the running kernel doesn't provide the syscall - that's going to be pretty easy to trigger by running a container with Python 3.6 on a container host running an older Linux kernel.

[vadmium]

I run Arch Linux, but only update packages when I have to. As a result, I am running Linux 3.15 installed and running, but the linux-api-headers has more recently been updated to 4.7 (i.e. matching Linux 4.7).