Thanks to visit codestin.com
Credit goes to github.com

Skip to content

gh-137335: Fix unlikely name conflicts for named pipes in multiprocessing and asyncio on Windows #137389

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
serhiy-storchaka merged 9 commits into from
Feb 24, 2026
Merged

gh-137335: Fix unlikely name conflicts for named pipes in multiprocessing and asyncio on Windows #137389

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
2a0a5e9
gh-137335: Fix unlikely nmae conflicts for named pipes in multiproces…
serhiy-storchaka Aug 4, 2025
5b91a56
Fix typo
serhiy-storchaka Aug 5, 2025
ed050a0
Do not use os.getpid() and tempfile._get_candidate_names().
serhiy-storchaka Aug 5, 2025
7363c99
Merge branch 'main' into temp-named-pipes
serhiy-storchaka Feb 5, 2026
5243b9e
Use os.urandom().
serhiy-storchaka Feb 5, 2026
81cb66e
Merge branch 'main' into temp-named-pipes
serhiy-storchaka Feb 10, 2026
5207bf6
Restore the PID and sequential number for debugging and reducing chan…
serhiy-storchaka Feb 10, 2026
cae80b9
Limit the number of attempts.
serhiy-storchaka Feb 10, 2026
de04089
Apply suggestion from @serhiy-storchaka
serhiy-storchaka Feb 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 12 additions & 10 deletions Lib/asyncio/windows_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,11 +6,10 @@
raise ImportError('win32 only')

import _winapi
import itertools
import msvcrt
import os
import random
import subprocess
import tempfile
import warnings


Expand All @@ -23,18 +22,13 @@
BUFSIZE = 8192
PIPE = subprocess.PIPE
STDOUT = subprocess.STDOUT
_mmap_counter = itertools.count()


# Replacement for os.pipe() using handles instead of fds


def pipe(*, duplex=False, overlapped=(True, True), bufsize=BUFSIZE):
"""Like os.pipe() but with overlapped support and using handles not fds."""
address = tempfile.mktemp(
Comment thread
kumaraditya303 marked this conversation as resolved.
prefix=r'\\.\pipe\python-pipe-{:d}-{:d}-'.format(
os.getpid(), next(_mmap_counter)))

if duplex:
openmode = _winapi.PIPE_ACCESS_DUPLEX
access = _winapi.GENERIC_READ | _winapi.GENERIC_WRITE
Expand All @@ -56,9 +50,17 @@ def pipe(*, duplex=False, overlapped=(True, True), bufsize=BUFSIZE):

h1 = h2 = None
try:
h1 = _winapi.CreateNamedPipe(
address, openmode, _winapi.PIPE_WAIT,
1, obsize, ibsize, _winapi.NMPWAIT_WAIT_FOREVER, _winapi.NULL)
while True:
address = r'\\.\pipe\python-pipe-' + random.randbytes(8).hex()
Copy link
Copy Markdown
Member

@picnixz picnixz Aug 5, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A comment:

we could use os.urandom(8).hex() but this would only reduce the number of loop iterations we would do until we find a suitable name. OTOH, this makes normal cases much slower (usually, we are not in the presence of an adversary that is trying to create pipes...), so we would be only doing the loop once.

Now, I'm actually worried that it if we're able to interact with the process that is creating the pipes, then we could actually recover enough samples from the underlying PRNG instance and get the original seed. But this is only if 1) there are no random calls in between and 2) we can get 624 consecutive 32-bit samples from the random source.

Condition 2) already holds because random.randbytes(8) is actually equivalent to two calls of random.randbytes(4) that are concatenated. Since reverting MT-19937 requires only consecutive 624 32-bit words, this is the same as doing 312 pipe creations where we don't have calls to random.* in between (and then inspect the named file). This could be possible in practice, especially if we're talking about the multiprocessing and the asyncio components which could have some interactivness (e.g., a server).

Therefore, I would still suggest using os.urandom(8).hex() even if it slows down the calls.

Copy link
Copy Markdown
Member Author

@serhiy-storchaka serhiy-storchaka Feb 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good catch, I did not thought about this.

Sorry for not answering immediately, I missed your comment.

try:
h1 = _winapi.CreateNamedPipe(
address, openmode, _winapi.PIPE_WAIT,
1, obsize, ibsize, _winapi.NMPWAIT_WAIT_FOREVER, _winapi.NULL)
break
except OSError as e:
if e.winerror not in (_winapi.ERROR_PIPE_BUSY,
_winapi.ERROR_ACCESS_DENIED):
Copy link
Copy Markdown
Member

@gpshead gpshead Feb 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

an infinite loop retrying on this error could potentially hide an error if the system simply doesn't allow this process to create a pipe with the given address? (here and below)

Copy link
Copy Markdown
Member Author

@serhiy-storchaka serhiy-storchaka Feb 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do not know if this is even possible (\\.\pipe\ is not a real directory which you can forbid access to), but it will not harm if limit the number of attempts.

raise

h2 = _winapi.CreateFile(
address, access, 0, _winapi.NULL, _winapi.OPEN_EXISTING,
Expand Down
56 changes: 35 additions & 21 deletions Lib/multiprocessing/connection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,8 +11,8 @@

import errno
import io
import itertools
import os
import random
import sys
import socket
import struct
Expand Down Expand Up @@ -45,8 +45,6 @@
# A very generous timeout when it comes to local connections...
CONNECTION_TIMEOUT = 20.

_mmap_counter = itertools.count()

default_family = 'AF_INET'
families = ['AF_INET']

Expand Down Expand Up @@ -78,8 +76,7 @@ def arbitrary_address(family):
elif family == 'AF_UNIX':
return tempfile.mktemp(prefix='sock-', dir=util.get_temp_dir())
elif family == 'AF_PIPE':
return tempfile.mktemp(prefix=r'\\.\pipe\pyc-%d-%d-' %
(os.getpid(), next(_mmap_counter)), dir="")
return r'\\.\pipe\pyc-' + random.randbytes(8).hex()
else:
raise ValueError('unrecognized family')

Expand Down Expand Up @@ -472,17 +469,27 @@ class Listener(object):
def __init__(self, address=None, family=None, backlog=1, authkey=None):
family = family or (address and address_type(address)) \
or default_family
address = address or arbitrary_address(family)

_validate_family(family)
if authkey is not None and not isinstance(authkey, bytes):
raise TypeError('authkey should be a byte string')

if family == 'AF_PIPE':
self._listener = PipeListener(address, backlog)
if address:
self._listener = PipeListener(address, backlog)
else:
while True:
address = arbitrary_address(family)
try:
self._listener = PipeListener(address, backlog)
break
except OSError as e:
if e.winerror not in (_winapi.ERROR_PIPE_BUSY,
_winapi.ERROR_ACCESS_DENIED):
raise
else:
address = address or arbitrary_address(family)
self._listener = SocketListener(address, family, backlog)

if authkey is not None and not isinstance(authkey, bytes):
raise TypeError('authkey should be a byte string')

self._authkey = authkey

def accept(self):
Expand Down Expand Up @@ -570,7 +577,6 @@ def Pipe(duplex=True):
'''
Returns pair of connection objects at either end of a pipe
'''
address = arbitrary_address('AF_PIPE')
if duplex:
openmode = _winapi.PIPE_ACCESS_DUPLEX
access = _winapi.GENERIC_READ | _winapi.GENERIC_WRITE
Expand All @@ -580,15 +586,23 @@ def Pipe(duplex=True):
access = _winapi.GENERIC_WRITE
obsize, ibsize = 0, BUFSIZE

h1 = _winapi.CreateNamedPipe(
address, openmode | _winapi.FILE_FLAG_OVERLAPPED |
_winapi.FILE_FLAG_FIRST_PIPE_INSTANCE,
_winapi.PIPE_TYPE_MESSAGE | _winapi.PIPE_READMODE_MESSAGE |
_winapi.PIPE_WAIT,
1, obsize, ibsize, _winapi.NMPWAIT_WAIT_FOREVER,
# default security descriptor: the handle cannot be inherited
_winapi.NULL
)
while True:
address = arbitrary_address('AF_PIPE')
try:
h1 = _winapi.CreateNamedPipe(
address, openmode | _winapi.FILE_FLAG_OVERLAPPED |
_winapi.FILE_FLAG_FIRST_PIPE_INSTANCE,
_winapi.PIPE_TYPE_MESSAGE | _winapi.PIPE_READMODE_MESSAGE |
_winapi.PIPE_WAIT,
1, obsize, ibsize, _winapi.NMPWAIT_WAIT_FOREVER,
# default security descriptor: the handle cannot be inherited
_winapi.NULL
)
break
except OSError as e:
if e.winerror not in (_winapi.ERROR_PIPE_BUSY,
_winapi.ERROR_ACCESS_DENIED):
raise
h2 = _winapi.CreateFile(
address, access, 0, _winapi.NULL, _winapi.OPEN_EXISTING,
_winapi.FILE_FLAG_OVERLAPPED, _winapi.NULL
Expand Down
2 changes: 2 additions & 0 deletions Misc/NEWS.d/next/Library/2025-08-04-23-20-43.gh-issue-137335.IIjDJN.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
Get rid of any possibility of a name conflict for named pipes in
:mod:`multiprocessing` and :mod:`asyncio` on Windows, no matter how small.
Loading