管道读写规则和Pipe Capacity、PIPE_BUF

时间 2019-11-18

标签管道读写规则 pipe capacity buf 繁體版

原文原文链接

1、当没有数据可读时
O_NONBLOCK disable：read调用阻塞，即进程暂停执行，一直等到有数据来到为止。
linux

O_NONBLOCK enable：read调用返回-1，errno值为EAGAIN。ubuntu

示例程序以下：小程序

    C++ Code  
 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

特地在子进程中sleep了3s，让父进程先被调度运行，并且读端文件状态标志设置为非阻塞，即马上出错返回，以下。api

simba@ubuntu:~/Documents/code/linux_programming/APUE/pipe$ ./pipe_block
read error: Resource temporarily unavailable
socket

2、当管道满的时候
O_NONBLOCK disable： write调用阻塞，直到有进程读走数据
O_NONBLOCK enable：调用返回-1，errno值为EAGAIN函数

管道是一块内存缓冲区，能够写个小程序测试一下管道的容量Pipe Capacity：测试

    C++ Code  
 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

程序中将写端文件状态标志设置为非阻塞，当管道被写满时不会等待其余进程读取数据，而是直接返回-1并置errno，输出以下：
simba@ubuntu:~/Documents/code/linux_programming/APUE/pipe$ ./pipe_capacity
err=Resource temporarily unavailable
count=65536
atom

打印了错误码，能够看到管道的容量是64kB，man 7 pipe中也有提到在2.6.11内核之前是4096，如今是65536。spa

3、若是全部管道读端对应的文件描述符被关闭（管道读端的引用计数等于0），则write操做会产生SIGPIPE信号，默认终止当前进程code

示例代码以下：

    C++ Code  
 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

输出测试：

simba@ubuntu:~/Documents/code/linux_programming/APUE/pipe$ ./close_fd_read
recv sig=13
err=Broken pipe

父进程睡眠1s确保全部读端文件描述符都已经关闭，若是没有安装SIGPIPE信号的处理函数，则默认终止当前进程，即write函数不会返回，如今write错误返回-1，并置errno=EPIPE，对应的出错信息是Broken pipe。

4、若是全部管道写端对应的文件描述符被关闭（管道写端的引用计数等于0），那么管道中剩余的数据都被读取后，再次read会返回0

示例程序以下：

    C++ Code  
 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55

输出测试以下：

simba@ubuntu:~/Documents/code/linux_programming/APUE/pipe$ ./close_fd_write
ret = 0

一样地父进程睡眠1s确保全部的写端文件描述符都已经关闭，read返回0。

5、当要写入的数据量不大于PIPE_BUF时，linux将保证写入的原子性；当要写入的数据量大于PIPE_BUF时，linux将再也不保证写入的原子性。

On Linux, PIPE_BUF is 4096 bytes。

The precise semantics depend on whether the file descriptor is nonblocking (O_NONBLOCK), whether there are multiple writers to the pipe, and on n, the number of bytes to be written。即由文件描述符是不是非阻塞的，是否有多个进程向管道写入以及写入的字节数所决定准确的语义，总共分4种状况，具体可man一下。

下面的程序演示 O_NONBLOCK disabled ，size > PIPE_BUF(4K)的状况：

    C++ Code  
 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76

/*************************************************************************
    > File Name: process_.c
    > Author: Simba
    > Mail: dameng34@163.com
    > Created Time: Sat 23 Feb 2013 02:34:02 PM CST
************************************************************************/
#include<sys/types.h>
#include<sys/stat.h>
#include<unistd.h>
#include<fcntl.h>
#include<stdio.h>
#include<stdlib.h>
#include<errno.h>
#include<string.h>
#include<signal.h>
#define ERR_EXIT(m) \
    do { \
        perror(m); \
        exit(EXIT_FAILURE); \
    } while(0)

#define TEST_SIZE 68*1024 // 68KB
/* 默认O_NONBLOCK disabled ，这里验证 size > PIPE_BUF(4K)的状况 */
int main(int argc, char *argv[])
{
    char a[TEST_SIZE];
    char b[TEST_SIZE];

    memset(a, 'A', sizeof(a));
    memset(b, 'B', sizeof(b));

    int pipefd[2];
    int ret = pipe(pipefd);
    if (ret == -1)
        ERR_EXIT("pipe error");

    int pid = fork();
    if (pid == 0)
    {

        close(pipefd[0]);
        ret = write(pipefd[1], a, sizeof(a)); // 所有写完才返回
        printf("apid=%d write %d bytes to pipe\n", getpid(), ret);
        exit(0);
    }

    pid = fork();

    if (pid == 0)
    {

        close(pipefd[0]);
        ret = write(pipefd[1], b, sizeof(b));
        printf("bpid=%d write %d bytes to pipe\n", getpid(), ret);
        exit(0);
    }

    close(pipefd[1]);

    sleep(1);

    int fd = open("test.txt", O_WRONLY | O_CREAT | O_TRUNC, 0664);
    char buf[1024 * 4] = {0};
    int n = 1;
    while (1)
    {
        ret = read(pipefd[0], buf, sizeof(buf)); //当管道被写入数据，就已经能够开始读了,每次读取4k
        if (ret == 0) // 管道写端所有关闭,即读到告终尾
            break;
        printf("n=%02d pid=%d read %d bytes from pipe buf[4095]=%c\n",
               n++, getpid(), ret, buf[4095]);
        write(fd, buf, ret);
    }

    return 0;
}

输出测试以下：

simba@ubuntu:~/Documents/code/linux_programming/APUE/pipe$ ./pipe_buf
n=01 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=02 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=03 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=04 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=05 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=06 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=07 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=08 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=09 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=10 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=11 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=12 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=13 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=14 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=15 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=16 pid=7137 read 4096 bytes from pipe buf[4095]=B
n=17 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=18 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=19 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=20 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=21 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=22 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=23 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=24 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=25 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=26 pid=7137 read 4096 bytes from pipe buf[4095]=A
apid=7138 write 69632 bytes to pipe
n=27 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=28 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=29 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=30 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=31 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=32 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=33 pid=7137 read 4096 bytes from pipe buf[4095]=A
n=34 pid=7137 read 4096 bytes from pipe buf[4095]=B
bpid=7139 write 69632 bytes to pipe

分析一下：如今的状况是有两个子进程在对管道进行阻塞写入各68k，即每一个子进程彻底写入68k才返回，而父进程对管道进行阻塞读取，每次读取4k，打印每4k中的最后一个字符，若是没有数据到达就阻塞等待，若是管道剩余数据不足4k，read 极可能返回 < 4k，但由于咱们写入68k是4k整数倍，故不存在这种状况。须要注意的是是边写边读，由于前面说过管道的容量只有64k，当管道被写满时子进程就阻塞等待父进程读取后再写入。由上面输出能够看出B进程先写入64k的B，而后A进程写入68k的A以后B进程接着写完最后4K的B，而后write返回。由A进程write完毕输出的提示可知此时A进程已经写完成了，但父进程还没读取A完毕，当两个子进程所有写完退出时关闭写端文件描述符，则父进程read就会返回0，退出while循环。能够得出结论：当多个进程对管道进行写入，且一次性写入数据量大于PIPE_BUF时，则不能保证写入的原子性，便可能数据是穿插着的。man 手册的解释以下：

O_NONBLOCK disabled, n > PIPE_BUF
The write is nonatomic: the data given to write(2) may be interleaved with write(2)s by other process; the write(2) blocks until n bytes have been written.

注意咱们这里设定了size=68k，则写端不能设置成非阻塞，由于Pipe Capacity 只有64k，不能一次性写入68k，若是此时管道是满的（64k)，则只能返回-1并置错误码为EAGAIN，且一个字符也不写入，若不是满的，则写入的字节数是不肯定的，须要检查write的返回值，并且这些字节极可能也与其余进程写入的数据穿插着。读端也不能设置为非阻塞，若是此时还没有有数据写入（管道为空）则返回-1并置错误码为EAGAIN，若是有部分数据已经写入，则读取的数据字节数也是不肯定的，须要检查read的返回值。总之测试4种不一样情形下的状况也应设置不一样的条件。

O_NONBLOCK disabled, n <= PIPE_BUF
All n bytes are written atomically; write(2) may block if there is not room for n bytes to be written imme‐
diately

O_NONBLOCK enabled, n <= PIPE_BUF
If there is room to write n bytes to the pipe, then write(2) succeeds immediately, writing all n bytes;
otherwise write(2) fails, with errno set to EAGAIN.

O_NONBLOCK disabled, n > PIPE_BUF
The write is nonatomic: the data given to write(2) may be interleaved with write(2)s by other process; the
write(2) blocks until n bytes have been written.

O_NONBLOCK enabled, n > PIPE_BUF
If the pipe is full, then write(2) fails, with errno set to EAGAIN. Otherwise, from 1 to n bytes may be
written (i.e., a "partial write" may occur; the caller should check the return value from write(2) to see
how many bytes were actually written), and these bytes may be interleaved with writes by other processes.

管道的前4种读写规则具备广泛意义，Tcp socket 也具备管道的这些特性。

参考：《APUE》