深入理解PIPE

转载：http://blog.ddup.us/?p=285

 

在linux中要进行进程间通信有多种方法：pipe、fifo、共享内存，信号量，消息队列，共享文件等等。其中pipe和fifo 使用最广泛，二者的区别为pipe为匿名管道，只能用在有父子关系的进程间通信，而fifo可以通过文件系统中的一个文件取得，所以不受上述限制。作为父子进程间通信的通道，pipe同样可以看作是一个先进先出的队列。

PIPE基本用法

pipe的使用很简单，原型如下：

<span style="color:#339933">#include <unistd.h></span>

<span style="color:#993333">int</span> pipe<span style="color:#0990">(</span><span style="color:#993333">int</span> pipefd<span style="color:#0990">[</span><span style="color:#00dd">2</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#339933">;</span>

典型的使用方法如下：

<span style="color:#339933">#include <stdlib.h></span>
<span style="color:#339933">#include <stdio.h></span>
<span style="color:#339933">#include <unistd.h></span>

<span style="color:#993333">int</span> main<span style="color:#0990">(</span><span style="color:#0990">)</span>
<span style="color:#0990">{</span>
    <span style="color:#993333">int</span> pfd<span style="color:#0990">[</span><span style="color:#00dd">2</span><span style="color:#0990">]</span><span style="color:#339933">;</span>
    <span style="color:#993333">int</span> ret <span style="color:#339933">=</span> pipe<span style="color:#0990">(</span>pfd<span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#b1b10">if</span> <span style="color:#0990">(</span>ret <span style="color:#339933">!=</span> <span style="color:#00dd">0</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
        perror<span style="color:#0990">(</span><span style="color:#ff00">"pipe"</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        exit<span style="color:#0990">(</span><span style="color:#339933">-</span><span style="color:#00dd">1</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span>

    pid_t pid <span style="color:#339933">=</span> fork<span style="color:#0990">(</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#b1b10">if</span> <span style="color:#0990">(</span>pid <span style="color:#339933">==</span> <span style="color:#00dd">0</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
        close<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">1</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#993333">char</span> rdbuf<span style="color:#0990">[</span><span style="color:#00dd">1024</span><span style="color:#0990">]</span><span style="color:#339933">;</span>
        read<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">0</span><span style="color:#0990">]</span><span style="color:#339933">,</span> rdbuf<span style="color:#339933">,</span> <span style="color:#993333">sizeof</span><span style="color:#0990">(</span>rdbuf<span style="color:#0990">)</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#b1b10">return</span> <span style="color:#00dd">0</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span> <span style="color:#b1b10">else</span> <span style="color:#b1b10">if</span> <span style="color:#0990">(</span>pid <span style="color:#339933">></span> <span style="color:#00dd">0</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
        close<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">0</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#993333">char</span> wrbuf<span style="color:#0990">[</span><span style="color:#0990">]</span> <span style="color:#339933">=</span> <span style="color:#ff00">"abcd"</span><span style="color:#339933">;</span>
        write<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">1</span><span style="color:#0990">]</span><span style="color:#339933">,</span> wrbuf<span style="color:#339933">,</span> <span style="color:#993333">sizeof</span><span style="color:#0990">(</span>wrbuf<span style="color:#0990">)</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        pid_t cpid <span style="color:#339933">=</span> waitpid<span style="color:#0990">(</span><span style="color:#339933">-</span><span style="color:#00dd">1</span><span style="color:#339933">,</span> NULL<span style="color:#339933">,</span> <span style="color:#00dd">0</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span> <span style="color:#b1b10">else</span> <span style="color:#0990">{</span>
        perror<span style="color:#0990">(</span><span style="color:#ff00">"fork"</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span>
    <span style="color:#b1b10">return</span> <span style="color:#00dd">0</span><span style="color:#339933">;</span>
<span style="color:#0990">}</span>

从代码中可以看出，调用pipe之后，pfd即为一对相关联的管道，其中pfd[0]对应的是数据输入端，pfd[1]对应数据输出端。当父进程想要给子进程发送数据时，直接将数据write到pfd[1]中即可，对应的子进程从pfd[0]中read。

PIPE缓冲区限制

从上边的父子进程间关系可以看到，这是操作系统中典型的1:1生产者消费者模型。管道作为联系生产者与消费者的缓冲区同样会涉及到两个问题：1）缓冲区的大小问题。 2）缓冲区读写的互斥问题。

将上述代码中父进程的write一行去掉，即可测试出在管道为空的情况下，子进程读管道会阻塞（默认是阻塞的，亦可以将其设置为非阻塞，这样返回的将是-1，同时errno设为EAGAIN）。

为了测试缓冲区的大小，再次修改上述代码，使得子进程不再读取数据，同时父进程一直向其中写入数据，代码如下：

pid_t pid <span style="color:#339933">=</span> fork<span style="color:#0990">(</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#b1b10">if</span> <span style="color:#0990">(</span>pid <span style="color:#339933">==</span> <span style="color:#00dd">0</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
        close<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">1</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        pause<span style="color:#0990">(</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#b1b10">return</span> <span style="color:#00dd">0</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span> <span style="color:#b1b10">else</span> <span style="color:#b1b10">if</span> <span style="color:#0990">(</span>pid <span style="color:#339933">></span> <span style="color:#00dd">0</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
        close<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">0</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        pause<span style="color:#0990">(</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#993333">char</span> buf<span style="color:#0990">[</span><span style="color:#0990">]</span> <span style="color:#339933">=</span> <span style="color:#ff00">"a"</span><span style="color:#339933">;</span>
        <span style="color:#b1b10">for</span> <span style="color:#0990">(</span><span style="color:#993333">int</span> i <span style="color:#339933">=</span> <span style="color:#00dd">0</span><span style="color:#339933">;</span> <span style="color:#339933">;</span> i<span style="color:#339933">++</span><span style="color:#0990">)</span> <span style="color:#0990">{</span>
            write<span style="color:#0990">(</span>pfd<span style="color:#0990">[</span><span style="color:#00dd">1</span><span style="color:#0990">]</span><span style="color:#339933">,</span> buf<span style="color:#339933">,</span> <span style="color:#993333">sizeof</span><span style="color:#0990">(</span>buf<span style="color:#0990">[</span><span style="color:#00dd">0</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
            <span style="color:#0066">printf</span><span style="color:#0990">(</span><span style="color:#ff00">"write pipe %d, total %d<span style="color:#0099; font-weight:bold">\n</span>"</span><span style="color:#339933">,</span> i<span style="color:#339933">+</span><span style="color:#00dd">1</span><span style="color:#339933">,</span> <span style="color:#0990">(</span>i<span style="color:#339933">+</span><span style="color:#00dd">1</span><span style="color:#0990">)</span><span style="color:#339933">*</span><span style="color:#993333">sizeof</span><span style="color:#0990">(</span>buf<span style="color:#0990">[</span><span style="color:#00dd">0</span><span style="color:#0990">]</span><span style="color:#0990">)</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
        <span style="color:#0990">}</span>
        pid_t cpid <span style="color:#339933">=</span> waitpid<span style="color:#0990">(</span><span style="color:#339933">-</span><span style="color:#00dd">1</span><span style="color:#339933">,</span> NULL<span style="color:#339933">,</span> <span style="color:#00dd">0</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span> <span style="color:#b1b10">else</span> <span style="color:#0990">{</span>
        perror<span style="color:#0990">(</span><span style="color:#ff00">"fork"</span><span style="color:#0990">)</span><span style="color:#339933">;</span>
    <span style="color:#0990">}</span>

修改之后的父进程每次向管道中写入一个字符，并将已经写入的字节数打印出来。当程序停住时，在Linux2.6.27上显示的结果为：

write pipe <span style="color:#00dd">65536</span><span style="color:#339933">,</span> total <span style="color:#00dd">65536</span>

由此可以看出，此时pipe的缓冲区大小为64KB。相同的程序在MacOSX中显示为：

write pipe <span style="color:#00dd">16384</span><span style="color:#339933">,</span> total <span style="color:#00dd">16384</span>

可以看出MacOSX的pipe默认缓冲区大小16KB。

已知了pipe默认缓冲区的大小了，那么自然就会想这个缓冲区大小是不是可以人工设定呢？搜索一番之后，发现好多人都说这个值是限定死了的，在内核代码中固定就是64K。后来下了一份linux 3.0代码发现这个默认值已经是可以改的了。相关代码在include/linux/pipe_fs_i.h中：

<span style="color:#339933">#define PIPE_DEF_BUFFERS    16</span>
…
…
<span style="color:#808080; font-style:italic">/* for F_SETPIPE_SZ and F_GETPIPE_SZ */</span>
<span style="color:#993333">long</span> pipe_fcntl<span style="color:#0990">(</span><span style="color:#993333">struct</span> file <span style="color:#339933">*,</span> <span style="color:#993333">unsigned</span> <span style="color:#993333">int</span><span style="color:#339933">,</span> <span style="color:#993333">unsigned</span> <span style="color:#993333">long</span> arg<span style="color:#0990">)</span><span style="color:#339933">;</span>

看到PIPE_DEF_BUFFERS就是默认的pipe缓冲区大小，不过是以页为单位的，默认的页大小为4K，所以默认的pipe缓冲区大小即为64KB。但是在最下边又有个函数pipe_fcntl，同时有两个常量F_SETPIPE_SZ,F_GETPIPE_SZ。看来应该是用来修改默认缓冲区大小的。

果然这个特性是在2.6.35的内核中加入的。发行说明可以见这里，相应的commit log中看到有相应的Commit。这样就可以通过使用fcntl配合上边两个常量指令更改pipe缓冲区大小。同时在/proc/sys/fs/pipe-max-size中可以方便查看pipe缓冲区最大值。

至此已经知道了linux默认的pipe缓冲区为64KB，同时在2.6.35之后的内核可以使用fcntl更改缓冲区大小。还剩另一个问题，就是原子读写问题。

PIPE的原子访问

同样在include/linux/pipe_fs_i.h中，有如下代码：

<span style="color:#808080; font-style:italic">/* Differs from PIPE_BUF in that PIPE_SIZE is the length of the actual
   memory allocation, whereas PIPE_BUF makes atomicity guarantees.  */</span>
<span style="color:#339933">#define PIPE_SIZE       PAGE_SIZE</span>

看到定义了PIPE_SIZE这个常量，大小为一个页面大小（默认是4K）。从注释中可以看到这个PIPE_SIZE值即为PIPE最大可保证的原子操作字节数。同样在Write Man Page中写到：

Write requests of <span style="color:#0990">{</span>PIPE_BUF<span style="color:#0990">}</span> bytes or less shall not be interleaved with data from other processes doing writes on the same pipe.
<span style="color:#202020">Writes</span> of greater than <span style="color:#0990">{</span>PIPE_BUF<span style="color:#0990">}</span> bytes may have data interleaved<span style="color:#339933">,</span> on arbitrary boundaries<span style="color:#339933">,</span> with writes by other processes<span style="color:#339933">,</span> whether or not the O_NONBLOCK flag of the file status flags is set.

这个PIPE_BUF常量在/usr/include/linux/limits.h中定义：

<span style="color:#339933">#define PIPE_BUF        4096    /* # bytes in atomic write to a pipe */</span>

另外也可以通过ulimit -p查看这个限制，不过这是个只读值，不能修改的。ulimit -a输出如下，其中的pipe size 即是：

core file size          <span style="color:#0990">(</span>blocks<span style="color:#339933">,</span> <span style="color:#339933">-</span>c<span style="color:#0990">)</span> <span style="color:#00dd">0</span>
data seg size           <span style="color:#0990">(</span>kbytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>d<span style="color:#0990">)</span> unlimited
scheduling priority             <span style="color:#0990">(</span><span style="color:#339933">-</span>e<span style="color:#0990">)</span> <span style="color:#00dd">0</span>
file size               <span style="color:#0990">(</span>blocks<span style="color:#339933">,</span> <span style="color:#339933">-</span>f<span style="color:#0990">)</span> unlimited
pending signals                 <span style="color:#0990">(</span><span style="color:#339933">-</span>i<span style="color:#0990">)</span> <span style="color:#00dd">81920</span>
max locked memory       <span style="color:#0990">(</span>kbytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>l<span style="color:#0990">)</span> <span style="color:#00dd">32</span>
max memory size         <span style="color:#0990">(</span>kbytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>m<span style="color:#0990">)</span> unlimited
open files                      <span style="color:#0990">(</span><span style="color:#339933">-</span>n<span style="color:#0990">)</span> <span style="color:#00dd">1024</span>
pipe size            <span style="color:#0990">(</span><span style="color:#00dd">512</span> bytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>p<span style="color:#0990">)</span> <span style="color:#00dd">8</span>
POSIX message queues     <span style="color:#0990">(</span>bytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>q<span style="color:#0990">)</span> <span style="color:#00dd">819200</span>
real<span style="color:#339933">-</span>time priority              <span style="color:#0990">(</span><span style="color:#339933">-</span>r<span style="color:#0990">)</span> <span style="color:#00dd">0</span>
stack size              <span style="color:#0990">(</span>kbytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>s<span style="color:#0990">)</span> <span style="color:#00dd">10240</span>
cpu time               <span style="color:#0990">(</span>seconds<span style="color:#339933">,</span> <span style="color:#339933">-</span>t<span style="color:#0990">)</span> unlimited
max user processes              <span style="color:#0990">(</span><span style="color:#339933">-</span>u<span style="color:#0990">)</span> <span style="color:#00dd">1024</span>
virtual memory          <span style="color:#0990">(</span>kbytes<span style="color:#339933">,</span> <span style="color:#339933">-</span>v<span style="color:#0990">)</span> unlimited
file locks                      <span style="color:#0990">(</span><span style="color:#339933">-</span>x<span style="color:#0990">)</span> unlimited

参考

http://pubs.opengroup.org/onlinepubs/9699919799/functions/write.html
http://stackoverflow.com/questions/4624071/pipe-buffer-size-is-4k-or-64k
http://home.gna.org/pysfst/tests/pipe-limit.html
http://unix.stackexchange.com/questions/11946/how-big-is-the-pipe-buffer

http://stackoverflow.com/questions/4739348/is-it-possible-to-change-the-size-of-a-named-pipe-on-linux