LEAmDNS: _parseResponse keeps every answer record of a packet on the heap, so one large answer can abort with out of memory

オープン
  • 何か問題がありますか？
#9,343 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
52/100
issue の種類
バグ
明瞭さ
おおむね明確
活発さ
活発
技術スタック
arduino, cpp
領域
embedded-iot

調査の方向性

Start in libraries/ESP8266mDNS/src/LEAmDNS_Control.cpp: trace _parseResponse through _readRRAnswer and _processAnswers, then compare with _parseQuery's record handling. The report proposes processing and freeing each response record before reading the next; done means response handling no longer retains all packet records at once while preserving probe and service query behavior. The reporter says they have a patched copy, but provides no test or captured packet.

索引モデルが issue の本文から書いたものです。

説明

What happens

An ESP8266 (core 3.1.2, a GeekMagic SmallTV-Ultra with about 6 KB of free heap while running) restarted after
11.4 hours with an out-of-memory abort inside LEAmDNS. Our crash record, decoded with addr2line:

operator new (288 bytes) failed
  <- MDNSResponder::_readRRAnswer
  <- MDNSResponder::_parseResponse
  <- MDNSResponder::_parseMessage
  <- MDNSResponder::_process
free heap 952 bytes, largest free block 248 bytes

The lowest values recorded before, during those hours, were about 4.4 KB free and a 1.5 KB largest block. So
about 3.5 KB went within a moment. The sketch only advertises (MDNS.begin, addService, addServiceTxt, and
MDNS.update() in loop()); it sends no queries.

Why, from reading the code (master, libraries/ESP8266mDNS/src/LEAmDNS_Control.cpp)
  • _parseResponse reads answers only while a service query waits or a probe runs (line 742-743). Probing runs
    after MDNS.begin and again after every interface status change (LEAmDNS.cpp, the statusChangeCB that
    calls _restart()), for example a Wi-Fi reconnect.
  • In that state, every record of the incoming packet (ANCount + NSCount + ARCount, taken from the packet
    header) is read with _readRRAnswer into its own heap object and kept in a list (lines 764-775).
  • The list goes to _processAnswers as a whole (line 794) and is freed only after that (lines 804-810).

So the heap needed grows with the number of records in someone else's answer. A phone, TV or speaker that
answers with many records during the probe window can take several KB at once. new aborts on failure in this
core, so the library's own checks for a failed allocation never run, and the result is a restart.

What we did not measure: we have not captured the packet, and we don't know for sure that a reconnect came just
before. That part is our reading of the code.

_parseQuery handles known answers differently: it frees each record right after reading it (so one record at a
time), which is why we think the same can work for responses.

Possible fix

Handle each answer record in _parseResponse as soon as it is read, and free it before reading the next, so the
peak is one record instead of the whole packet. That means splitting _processAnswers so it can take one record
at a time. A simpler, partial alternative: skip (read past without allocating) records whose names cannot
concern this responder.

We are trying the first approach in a patched copy for our own firmware and can open a PR if this direction is
welcome.

主要言語
C++
スター
16.7k
フォーク
13k
平均マージ
1日 2時間
マージ済み PR（30日）
7

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

esp8266/Arduino のほかの issue

esp8266/Arduino の issue をすべて見る

似ている issue

C++ の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。