Fallback to dev_urandom doesn't work when py_getrandom returns -1

[jernejs]

BPO	32580
Nosy	@vstinner, @jernejs
Files	random_fallback.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 = None
created_at = <Date 2018-01-17.13:11:08.071>
labels = ['build']
title = "Fallback to dev_urandom doesn't work when py_getrandom returns -1"
updated_at = <Date 2018-01-17.20:52:10.985>
user = 'https://github.com/jernejs'

bugs.python.org fields:

activity = <Date 2018-01-17.20:52:10.985>
actor = 'jernejs'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Build']
creation = <Date 2018-01-17.13:11:08.071>
creator = 'jernejs'
dependencies = []
files = ['47389']
hgrepos = []
issue_num = 32580
keywords = ['patch']
message_count = 5.0
messages = ['310158', '310162', '310196', '310197', '310206']
nosy_count = 2.0
nosy_names = ['vstinner', 'jernejs']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = None
url = 'https://bugs.python.org/issue32580'
versions = ['Python 3.5', 'Python 3.6']

[jernejs]

I'm playing with Gentoo in LXC running on ARM-based QNAP NAS, and when trying to compile Python 3.5.4, the build failed with "Fatal Python error: failed to get random numbers to initialize Python" (full details at https://forums.gentoo.org/viewtopic-p-8172124.html ).

glibc in the environment is 2.25 (supports getrandom), while kernel is 3.10 (doesn't support getrandom).

Looking at random.c, pyurandom returns -1 instead of falling back to dev_urandom when py_getrandom returns -1, so the attached fix is simple.

[vstinner]

I'm playing with Gentoo in LXC running on ARM-based QNAP NAS, and when trying to compile Python 3.5.4, the build failed with "Fatal Python error: failed to get random numbers to initialize Python" (full details at https://forums.gentoo.org/viewtopic-p-8172124.html ).

I'm surprised. Python 3.5.4 does contain my latest fix: commit 035ba5d (bpo-29157).

Looking at random.c, pyurandom returns -1 instead of falling back to dev_urandom when py_getrandom returns -1, so the attached fix is simple.

Ignoring failures is not a good idea.

What is the result of the getrandom() function?

glibc in the environment is 2.25 (supports getrandom), while kernel is 3.10 (doesn't support getrandom).

I expect n < 0 with errno = ENOSYS: py_getrandom() should return 0 in this case, not -1.

[jernejs]

OK, looks like Qnap is doing something weird there - calls to getrandom result in EFAULT, which results in . PHP had the same problem: https://bugs.php.net/bug.php?id=75409

Changing "if (errno == ENOSYS || errno == EPERM) {" to "if (errno == ENOSYS || errno == EPERM || errno == EFAULT) {" also works.

[vstinner]

OK, looks like Qnap is doing something weird there - calls to getrandom result in EFAULT

Aha, first time that I see this error on getrandom().

Is it a kernel change? A libc change? Is Qnap a kind of sandbox?

[jernejs]

Qnap is a NAS, and while I'm running Gentoo inside LXC, EFAULT is returned by getrandom syscall even in the host OS. The model I'm doing this on is TS-228, uname -a of the host OS reports:

Linux qnap 3.10.20-al-2.5.3 #67 SMP PREEMPT Sat Dec 30 04:29:34 CST 2017 armv7l unknown

This is with the latest firmware installed as of this writing - 4.3.4.0435 (30.12.2017).