mknod devices can be >32 bits

[jcea]

BPO	23098
Nosy	@jcea, @serhiy-storchaka
Files	posix_dev_t_converter-3.5.patch posix_dev_t_converter-3.5_2.patch posix_dev_t_converter-2.7.patch posix_dev_t_converter-3.4.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/jcea'
closed_at = <Date 2015-01-18.09:47:58.037>
created_at = <Date 2014-12-21.22:36:24.874>
labels = ['extension-modules', 'easy', 'type-bug']
title = 'mknod devices can be >32 bits'
updated_at = <Date 2015-01-18.09:47:58.036>
user = 'https://github.com/jcea'

bugs.python.org fields:

activity = <Date 2015-01-18.09:47:58.036>
actor = 'serhiy.storchaka'
assignee = 'jcea'
closed = True
closed_date = <Date 2015-01-18.09:47:58.037>
closer = 'serhiy.storchaka'
components = ['Extension Modules']
creation = <Date 2014-12-21.22:36:24.874>
creator = 'jcea'
dependencies = []
files = ['37526', '37527', '37532', '37533']
hgrepos = []
issue_num = 23098
keywords = ['patch', 'easy']
message_count = 8.0
messages = ['233004', '233015', '233019', '233022', '233026', '233037', '234229', '234232']
nosy_count = 3.0
nosy_names = ['jcea', 'python-dev', 'serhiy.storchaka']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue23098'
versions = ['Python 2.7', 'Python 3.4', 'Python 3.5']

[jcea]

Dan MacDonald told me that "os.mknod()" should accept devices >32 bits.

I wrote this code in linux 64 bits:

"""
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <stdio.h>

int main(void) {
    printf("%d", sizeof(dev_t));
    return 0;
}
"""

The result of running this is "8". We need a long long.

I did this change in Python 2.7 branch:

"""
diff -r f00412d32b41 Modules/posixmodule.c
--- a/Modules/posixmodule.c	Sat Dec 20 13:41:14 2014 -0600
+++ b/Modules/posixmodule.c	Sun Dec 21 23:30:00 2014 +0100
@@ -7009,9 +7009,9 @@
 {
     char *filename;
     int mode = 0600;
-    int device = 0;
+    long long device = 0;
     int res;
-    if (!PyArg_ParseTuple(args, "s|ii:mknod", &filename, &mode, &device))
+    if (!PyArg_ParseTuple(args, "s|iL:mknod", &filename, &mode, &device))
         return NULL;
     Py_BEGIN_ALLOW_THREADS
     res = mknod(filename, mode, device);
"""

Looks like this patch is trivial. Please, comment.

[serhiy-storchaka]

This is not so easy as look at first glance. PY_LONG_LONG should be used instead of long long. And as far as this type is optional, its use should be conditional if HAVE_LONG_LONG is defined.

May be needed complex converter like _Py_Uid_Converter because dev_t on Linux is 64-bit unsigned int and can not fit in the range of 64-bit signed int.

The code for 3.x should be even more complex due to using Argument Clinic.

[jcea]

Serhiy, could you consider to create a patch for 2.7 and 3.4?. I am not familiar with the details of Argument Clinic.

[serhiy-storchaka]

Here is a patch against 3.5. Unsigned bitwise int converter is used for dev_t because dev_t can be unsigned (and it is on Linux). This is not ideal solution, there is no overflow check, but at least it should work for correct code.

[serhiy-storchaka]

As pointed by Antoine there are other affected functions. Updated patch fixes converting of arguments or results in makedev(), major() and minor().

[serhiy-storchaka]

And here are patches for 2.7 and 3.4 (unfortunately none of patches is applied
clear to other version).

[python-dev]

New changeset 7ee09e2fec13 by Serhiy Storchaka in branch '2.7':
Issue bpo-23098: 64-bit dev_t is now supported in the os module.
https://hg.python.org/cpython/rev/7ee09e2fec13

New changeset 18703ffea2b3 by Serhiy Storchaka in branch '3.4':
Issue bpo-23098: 64-bit dev_t is now supported in the os module.
https://hg.python.org/cpython/rev/18703ffea2b3

New changeset fe0fddd6fd21 by Serhiy Storchaka in branch 'default':
Issue bpo-23098: 64-bit dev_t is now supported in the os module.
https://hg.python.org/cpython/rev/fe0fddd6fd21

[serhiy-storchaka]

Committed to 2.7 with small change: stat() and makedev() should return int instead of long if possible.