Swoole6.x不开启iouring的时候，fread操作疑似存在BUG

[Moxyu]

Swoole 信息

swoole

Swoole => enabled
Author => Swoole Team <team@swoole.com>
Version => 6.0.2
Built => Jun 18 2025 12:07:15
coroutine => enabled with boost asm context
epoll => enabled
eventfd => enabled
signalfd => enabled
cpu_affinity => enabled
spinlock => enabled
rwlock => enabled
sockets => enabled
openssl => OpenSSL 3.0.13 30 Jan 2024
dtls => enabled
http2 => enabled
json => enabled
curl-native => enabled
curl-version => 7.74.0
zlib => 1.2.13
brotli => E16777225/D16777225
mutex_timedlock => enabled
pthread_barrier => enabled
futex => enabled

Directive => Local Value => Master Value
swoole.enable_library => On => On
swoole.enable_fiber_mock => Off => Off
swoole.enable_preemptive_scheduler => Off => Off
swoole.display_errors => On => On
swoole.use_shortname => Off => Off
swoole.unixsock_buffer_size => 8388608 => 8388608

PHP 信息

PHP 8.4.7 (cli) (built: May  9 2025 10:58:14) (NTS)
Copyright (c) The PHP Group
Zend Engine v4.4.7, Copyright (c) Zend Technologies

目前我遇到的一个问题是，我目前的项目是基于Webman框架进行开发，使用了Swoole作为协程驱动

其他场景都正常，但是在一次意外的8M图片静态资源访问场景中，报了一个错：

Workerman[start.php] start in DEBUG mode
-------------------------------------------- WORKERMAN ---------------------------------------------
Workerman/5.1.3         PHP/8.4.7 (JIT off)           Linux/5.15.167.4-microsoft-standard-WSL2
--------------------------------------------- WORKERS ----------------------------------------------
event-loop  proto       user        worker      listen                 count       state            
swoole      tcp         root        webman      http://0.0.0.0:8787    80           [OK]            
----------------------------------------------------------------------------------------------------
Press Ctrl+C to stop. Start success.
ErrorException: fread(): Read of 18446744073709543424 bytes failed with errno=14 Bad address in /www/wwwroot/valorix/server-api/vendor/workerman/workerman/src/Protocols/Http.php:263
Stack trace:
#0 [internal function]: {closure:/www/wwwroot/valorix/server-api/support/bootstrap.php:31}()
#1 /www/wwwroot/valorix/server-api/vendor/workerman/workerman/src/Protocols/Http.php(263): fread()
#2 /www/wwwroot/valorix/server-api/vendor/workerman/workerman/src/Protocols/Http.php(281): Workerman\Protocols\Http::{closure:Workerman\Protocols\Http::sendStream():247}()
#3 /www/wwwroot/valorix/server-api/vendor/workerman/workerman/src/Connection/TcpConnection.php(808): Workerman\Protocols\Http::{closure:Workerman\Protocols\Http::sendStream():279}()
#4 /www/wwwroot/valorix/server-api/vendor/workerman/workerman/src/Events/Swoole.php(288): Workerman\Connection\TcpConnection->baseWrite()
#5 [internal function]: Workerman\Events\Swoole->{closure:Workerman\Events\Swoole::safeCall():286}()
#6 {main}

静态资源是直接通过Webman（核心是Workerman）进程来处理的，没有经过任何Web服务器。经过不断调试，200K的图片没问题，但是当图片大于5M就会报这个错误。

起初调试毫无头绪，后来看了v6发布说明，注意到了iouring这个东西。然后重新编译 Swoole 加上--enable-iouring参数，问题就解决了。

然后为了验证这个结论，我又重新编译 Swoole 故意不加 --enable-iouring ，果然问题又出现了，所以这个问题的原因是 iouring。

本着学习的态度去追溯了下 Workerman 的源码，位置/vendor/workerman/workerman/src/Protocols/Http.php:263。

protected static function sendStream(TcpConnection $connection, $handler, int $offset = 0, int $length = 0): void
    {
        $connection->context->bufferFull = false;
        $connection->context->streamSending = true;
        if ($offset !== 0) {
            fseek($handler, $offset);
        }
        $offsetEnd = $offset + $length;
        // Read file content from disk piece by piece and send to client.
        $doWrite = function () use ($connection, $handler, $length, $offsetEnd) {
            // Send buffer not full.
            while ($connection->context->bufferFull === false) {
                // Read from disk.
                $size = 1024 * 1024;
                if ($length !== 0) {
                    $tell = ftell($handler);
                    $remainSize = $offsetEnd - $tell;
                    if ($remainSize <= 0) {
                        fclose($handler);
                        $connection->onBufferDrain = null;
                        return;
                    }
                    $size = min($remainSize, $size);
                }

                $buffer = fread($handler, $size);
                // Read eof.
                if ($buffer === '' || $buffer === false) {
                    fclose($handler);
                    $connection->onBufferDrain = null;
                    $connection->context->streamSending = false;
                    return;
                }
                $connection->send($buffer, true);
            }
        };
        // Send buffer full.
        $connection->onBufferFull = function ($connection) {
            $connection->context->bufferFull = true;
        };
        // Send buffer drain.
        $connection->onBufferDrain = function ($connection) use ($doWrite) {
            $connection->context->bufferFull = false;
            $doWrite();
        };
        $doWrite();
    }

大概意思就是流式分块发送文件内容，通过分块读取文件内容并动态发送，避免一次性加载大文件导致内存溢出，同时处理TCP连接的缓冲区状态。整体代码看着是没什么问题。

但是Wokerman使用Swoole作为EventLoop的情况下，满足以下3个条件的时候，就会触发这个报错。

• Swoole版本为6.x
• 没有 iouring 的时候
• 静态资源超过5M（大约是这个值）

报错内容就是 ErrorException: fread(): Read of 18446744073709543424 bytes failed with errno=14 Bad address ，像是资源竞争？

然后我本地临时修改了 Workerman 的源码,把 $size 固定了 1024, 然后fread之前加上了一个dump打印操作来用作延迟，问题就没有出现了。

资源竞争？或者是其他？具体我就不太清楚了。

[NathanFreeman]

<?php
use Workerman\Worker;
use Workerman\Connection\TcpConnection;
use Workerman\Protocols\Http\Request;
use Workerman\Events\Swoole;
use Workerman\Protocols\Http\Response;
use Workerman\Coroutine;
require_once __DIR__ . '/vendor/autoload.php';

$http_worker = new Worker("http://0.0.0.0:2345");
$http_worker->eventLoop = Swoole::class;
$http_worker->count = 2;

$http_worker->onMessage = function(TcpConnection $connection, Request $request)
{
    Coroutine::create(function () use ($connection) {
        $response = (new Response())->withFile('/home/8MB_file');
        $connection->send($response);
    });
};


Worker::runAll();

我按照上面的代码没法复现出来问题，看看能否提供一个稳定复现的代码

[matyhtf]

需要检查是否存在对 同一个 fp 存在并发的 fread 或 fseek 、fwrite 行为？