Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

socket.recv() and recvfrom() with MSG_TRUNC raise ValueError and lose the datagram since 3.15.1

未关闭
  • 哪里不对？
#160,622 0 条评论 1 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
43/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
backend, networking

调研方向

Start by tracing socket.recv() and recvfrom() through sock_recv() and sock_recvfrom() to PyBytesWriter_FinishWithSize(), then inspect the change from gh-157226 and the existing behavior discussed in gh-69121. Reproduce the reported case on Linux and check how recvmsg() handles truncated datagrams. Done means defining and verifying behavior that avoids raising after consuming the datagram and avoids returning invalid padding bytes.

由索引模型根据 Issue 内容生成。

描述

extension-modules topic-socket type-bug
Bug report
Bug description:

Since 3.15.1, socket.recv(bufsize, MSG_TRUNC) and socket.recvfrom(bufsize, MSG_TRUNC) raise ValueError: size larger than allocated size when the datagram is longer than bufsize. Without MSG_PEEK, the datagram is already consumed by the time the exception is raised, so the data is lost:

import socket
a, b = socket.socketpair(socket.AF_UNIX, socket.SOCK_DGRAM)
a.send(b"abcdefgh")
a.send(b"second")
try:
    print(b.recv(2, socket.MSG_TRUNC))
except ValueError as e:
    print(e)
try:
    print(b.recvfrom(2, socket.MSG_TRUNC))
except ValueError as e:
    print(e)
b.setblocking(False)
b.recv(100)  # BlockingIOError: both datagrams are gone

3.15.1:

size larger than allocated size
size larger than allocated size

3.15.0 and 3.14 (the long-standing behavior, see gh-69121):

b'ab\x00\x00\x00\x00\x00\x00'
(b'se\x00\x00\x00\x00', None)

With MSG_TRUNC, Linux recv(2) returns the real length of the datagram even if it was longer than the buffer. sock_recv() and sock_recvfrom() pass that length to PyBytesWriter_FinishWithSize(), which, after c38a347e2c3199d7d3f6a9eefbf46df52caf725d (backport of gh-157226 to 3.15), rejects a size larger than the allocated one. The same code is in main.

The old behavior was a bug too (gh-69121): the returned object was grown to the real datagram length, with only the first bufsize bytes being valid. Still, it is a behavior change in a bugfix release that breaks working code. For example, CRIU's pycriu used len(s.recv(1, socket.MSG_TRUNC | socket.MSG_PEEK)) to find out the message size (fixed in checkpoint-restore/criu#3268 by using recv_into(), which returns the value from recv(2) as is).

Raising after the data is consumed seems to be the worst option. Returning the min(outlen, bufsize) bytes actually received (as suggested in gh-69121, and similar to what recvmsg() does) would avoid both the data loss and the garbage bytes.

CPython versions tested on:

3.14, 3.15

Operating systems tested on:

Linux

Linked PRs
  • gh-160677
主要语言
Python
星标
77.8k
派生
37.7k
平均合并
11 小时 25 分钟
30 天内合并 PR
723

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

python/cpython 的其他 Issue

查看 python/cpython 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。