Chinaunix首页 | 论坛 | 博客
  • 博客访问: 31204
  • 博文数量: 5
  • 博客积分: 0
  • 博客等级: 民兵
  • 技术积分: 111
  • 用 户 组: 普通用户
  • 注册时间: 2013-06-10 18:38
文章分类
文章存档

2014年(1)

2013年(4)

我的朋友

分类: LINUX

2013-06-14 10:48:17

原文地址:linux 管道的实现分析 作者:zghover

         linux 管道的实现分析
1, 数据结构

//inode结点信息结构
struct inode {
...
    struct pipe_inode_info  *i_pipe;
... 
};

//管道缓冲区个数
#define PIPE_BUFFERS (16)
//管道缓存区对象结构
struct pipe_buffer {
    struct page *page; //管道缓冲区页框的描述符地址
    unsigned int offset, len; //页框内有效数据的当前位置,和有效数据的长度
    struct pipe_buf_operations *ops; //管道缓存区方法表的地址
};

//管道信息结构
struct pipe_inode_info {
    wait_queue_head_t wait; //管道等待队列
    unsigned int nrbufs, curbuf; //包含待读数据的缓冲区数和包含待读数据的第一个缓冲区的索引
    struct pipe_buffer bufs[PIPE_BUFFERS]; //管道缓冲区描述符数组
    struct page *tmp_page; //高速缓存区页框指针
    unsigned int start;  //当前管道缓存区读的位置
    unsigned int readers; //读进程的标志,或编号
    unsigned int writers; //写进程的标志,或编号
    unsigned int waiting_writers; //在等待队列中睡眠的写进程的个数
    unsigned int r_counter; //与readers类似,但当等待写入FIFO的进程是使用
    unsigned int w_counter; //与writers类似,但当等待写入FIFO的进程时使用
    struct fasync_struct *fasync_readers; //用于通过信号进行的异步I/O通知
    struct fasync_struct *fasync_writers; //用于通过信号的异步I/O通知
};
 
 
2, 管道的实现
管道可以看着是打开的文件,但在已安装的文件系统中没有相应的影像。
管道是作为一组VFS对象来实现的,因此没有对应的磁盘映像。
而在2.6中把这些VFS对象组织成pipfs特殊文件系统以加速他们的处理。
但这种文件系统在系统目录树中没有安装点,用户根本看不到它。
2.1 pipefs 文件系统的安装
//文件系统的安装可以参考文件系统的实现一章
static struct super_block *pipefs_get_sb(struct file_system_type *fs_type,
    int flags, const char *dev_name, void *data)
{
    return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC);
}
static struct file_system_type pipe_fs_type = {
    .name       = "pipefs",
    .get_sb     = pipefs_get_sb,
    .kill_sb    = kill_anon_super,
};
static int __init init_pipe_fs(void)
{
    int err = register_filesystem(&pipe_fs_type);
    if (!err) {
        pipe_mnt = kern_mount(&pipe_fs_type);
        if (IS_ERR(pipe_mnt)) {
            err = PTR_ERR(pipe_mnt);
            unregister_filesystem(&pipe_fs_type);
        }  
    }  
    return err;
}
static void __exit exit_pipe_fs(void)
{
    unregister_filesystem(&pipe_fs_type);
    mntput(pipe_mnt);
}

2.2 pipe的建立的实现
对于每个管道来说,内核都创建一个inode结点对象,两个file对象,一个用于读,一个用于写。

int do_pipe(int *fd)
{
 struct qstr this;
 char name[32];
 struct dentry *dentry;
 struct inode * inode;
 struct file *f1, *f2;
 int error;
 int i,j;
 error = -ENFILE;
 f1 = get_empty_filp();  //获取文件对象1
 if (!f1)
  goto no_files;
 f2 = get_empty_filp();  //获取文件对象2
 if (!f2)
  goto close_f1;
 inode = get_pipe_inode(); //获取pipe的inode结点
 if (!inode)
  goto close_f12;
 error = get_unused_fd(); //获取没有使用的fd1
 if (error < 0)
  goto close_f12_inode;
 i = error;
 error = get_unused_fd(); //获取没有使用的fd2
 if (error < 0)
  goto close_f12_inode_i;
 j = error;
 error = -ENOMEM;
 sprintf(name, "[%lu]", inode->i_ino); //设置索引节点号
 this.name = name;
 this.len = strlen(name);
 this.hash = inode->i_ino; /* will go */
 dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &this); //获取一个目录对象
 if (!dentry)
  goto close_f12_inode_i_j;
 dentry->d_op = &pipefs_dentry_operations;
 //把目录对象和inode结点联系在一起
 d_add(dentry, inode);
 f1->f_vfsmnt = f2->f_vfsmnt = mntget(mntget(pipe_mnt));
 f1->f_dentry = f2->f_dentry = dget(dentry);
 f1->f_mapping = f2->f_mapping = inode->i_mapping;
 /* read file */ //给读描述符的文件对象赋值
 f1->f_pos = f2->f_pos = 0; //读的位置从0偏移量开始
 f1->f_flags = O_RDONLY; //只读
 f1->f_op = &read_pipe_fops; //读操作时执行的函数
 f1->f_mode = FMODE_READ; //读模式
 f1->f_version = 0;
 /* write file */
 f2->f_flags = O_WRONLY; //只写
 f2->f_op = &write_pipe_fops; //写操作执行函数
 f2->f_mode = FMODE_WRITE; //写模式
 f2->f_version = 0;
 fd_install(i, f1); //给文件对象f1中的fd赋值
 fd_install(j, f2); //给文件对象f2中的fd赋值
 fd[0] = i; //把值赋给用户空间
 fd[1] = j; //把值赋给用户空间
 return 0;
close_f12_inode_i_j:
 put_unused_fd(j);
close_f12_inode_i:
 put_unused_fd(i);
close_f12_inode:
 free_pipe_info(inode);
 iput(inode);
close_f12:
 put_filp(f2);
close_f1:
 put_filp(f1);
no_files:
 return error; 
}

//获取管道的inode结构
static struct inode * get_pipe_inode(void)
{
 struct inode *inode = new_inode(pipe_mnt->mnt_sb);
 if (!inode)
  goto fail_inode;
 if(!pipe_new(inode))
  goto fail_iput;
 PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1;
 inode->i_fop = &rdwr_pipe_fops;
 /*
  * Mark the inode dirty from the very beginning,
  * that way it will never be moved to the dirty
  * list because "mark_inode_dirty()" will think
  * that it already _is_ on the dirty list.
  */
 inode->i_state = I_DIRTY;
 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 inode->i_uid = current->fsuid;
 inode->i_gid = current->fsgid;
 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 inode->i_blksize = PAGE_SIZE;
 return inode;
fail_iput:
 iput(inode);
fail_inode:
 return NULL;
}
 
*管道的读操作的实现
 
管道读操作的规则如下:
 
 
//管道读操作函数
static ssize_t
pipe_readv(struct file *filp, const struct iovec *_iov,
    unsigned long nr_segs, loff_t *ppos)
{
 struct inode *inode = filp->f_dentry->d_inode; //获取inode结点指针
 struct pipe_inode_info *info;
 int do_wakeup;
 ssize_t ret;
 struct iovec *iov = (struct iovec *)_iov; //获取读缓冲区的结构
 size_t total_len;
 total_len = iov_length(iov, nr_segs);
 /* Null read succeeds. */
 if (unlikely(total_len == 0))
  return 0;
 do_wakeup = 0;
 ret = 0;
 down(PIPE_SEM(*inode)); //获取inode中的i_sem信号量
 info = inode->i_pipe; //获取inode 结构的pipe_inode_info结构指针
 for (;;) {
  int bufs = info->nrbufs; //检查有几个管道缓冲区有被读取的数据
  if (bufs) { //说明有其中有缓冲区包含了读数据
   int curbuf = info->curbuf; //获取当前读数据的管道缓存区的索引
   struct pipe_buffer *buf = info->bufs + curbuf; //共有16个缓冲区,curbuf是当前的
   struct pipe_buf_operations *ops = buf->ops; //获取操作函数列表
   void *addr;
   size_t chars = buf->len;
   int error;
   //若缓冲区长度大于要求读取的数据长度,chars设置成要求读的长度
   if (chars > total_len)
    chars = total_len;
   //执行Map方法
   addr = ops->map(filp, info, buf);
   //从缓存区中复制数据
   error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars);
   //执行umap方法
   ops->unmap(info, buf);
   if (unlikely(error)) {
    if (!ret) ret = -EFAULT; //第一次读失败
    break;
   }
   //更新管道的offset和len字段
   ret += chars;
   buf->offset += chars;
   buf->len -= chars;
   
   //若现在的缓存区的数据长度为0
   if (!buf->len) {
    buf->ops = NULL;
    ops->release(info, buf);
    curbuf = (curbuf + 1) & (PIPE_BUFFERS-1);
    info->curbuf = curbuf;
    info->nrbufs = --bufs;
    do_wakeup = 1;
   }
   total_len -= chars;  //更新读的总长度
   if (!total_len)  //该读的已读完成
    break; /* common path: read succeeded */
  }
  if (bufs) /* More to do? */
   continue;
  //若bufs为0,说明所有管道为NULL,此时进行一下操作
  if (!PIPE_WRITERS(*inode)) //是否有写操作正在进行
   break;
  if (!PIPE_WAITING_WRITERS(*inode)) { //是否需要等待
   /* syscall merging: Usually we must not sleep
    * if O_NONBLOCK is set, or if we got some data.
    * But if a writer sleeps in kernel space, then
    * we can wait for that data without violating POSIX.
    */
   if (ret)
    break;
   if (filp->f_flags & O_NONBLOCK) { //要等待但又设置了NONBLOCK标记,矛盾了
    ret = -EAGAIN;
    break;
   }
  }
  if (signal_pending(current)) { //设置进程阻塞标志
   if (!ret) ret = -ERESTARTSYS;
   break;
  }
  if (do_wakeup) {
   wake_up_interruptible_sync(PIPE_WAIT(*inode));
    kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
  }
  pipe_wait(inode);
 }
 up(PIPE_SEM(*inode));
 /* Signal writers asynchronously that there is more room.  */
 if (do_wakeup) {
  wake_up_interruptible(PIPE_WAIT(*inode));
  kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
 }
 if (ret > 0)
  file_accessed(filp);  //更新文件结构的atime对象
 return ret;
}
static ssize_t
pipe_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
{
 struct iovec iov = { .iov_base = buf, .iov_len = count };
 return pipe_readv(filp, &iov, 1, ppos);
}

/* Drop the inode semaphore and wait for a pipe event, atomically */
void pipe_wait(struct inode * inode)
{
    DEFINE_WAIT(wait);
 //把current添加到管道的等待队列中
    prepare_to_wait(PIPE_WAIT(*inode), &wait, TASK_INTERRUPTIBLE);
 //释放i_sem
    up(PIPE_SEM(*inode));
    schedule();
 //被呼醒,把它从等待队列中删除
    finish_wait(PIPE_WAIT(*inode), &wait);
 //再次获取i_sem索引节点信号量
    down(PIPE_SEM(*inode));
}
 
 
*管道的写操作
在管道创建函数do_pipe中可以看到,管道的写操作结构是write_pipe_fops,
该操作列表中的写操作是调用pipe_write实现的。
POSIX标准定义了写操作的一些规则:
 
 
 
 
 

static ssize_t
pipe_writev(struct file *filp, const struct iovec *_iov,
     unsigned long nr_segs, loff_t *ppos)
{
    struct inode *inode = filp->f_dentry->d_inode;
    struct pipe_inode_info *info;
    ssize_t ret;
    int do_wakeup;
    struct iovec *iov = (struct iovec *)_iov;
    size_t total_len;

    total_len = iov_length(iov, nr_segs);
    /* Null write succeeds. */
    if (unlikely(total_len == 0))
        return 0;

    do_wakeup = 0;
    ret = 0;
    down(PIPE_SEM(*inode));
    info = inode->i_pipe;

    //是否有读者进程存在,若没有写管道操作就没有任何意义

    //此时产生SIGPIPE信号

    if (!PIPE_READERS(*inode)) {
        send_sig(SIGPIPE, current, 0);
        ret = -EPIPE;
        goto out;
    }

    /* We try to merge small writes */
    //若有待读数据的缓冲区,而且写入的数据长度小于PAGE_SIZE

    if (info->nrbufs && total_len < PAGE_SIZE) {
        //第一个待读缓冲区+可读缓冲区数-1得到第一个可写缓冲区的地址

        int lastbuf = (info->curbuf + info->nrbufs - 1) & (PIPE_BUFFERS-1);
        struct pipe_buffer *buf = info->bufs + lastbuf;
        struct pipe_buf_operations *ops = buf->ops;
        int offset = buf->offset + buf->len;
        //若可写缓冲区的剩余的空间大于写入的数据总量total_len

        if (ops->can_merge && offset + total_len <= PAGE_SIZE) {
            void *addr = ops->map(filp, info, buf);
            //把数据复制到管道缓冲区

            int error = pipe_iov_copy_from_user(offset + addr, iov, total_len);
            ops->unmap(info, buf);
            ret = error;
            do_wakeup = 1;
            if (error)
                goto out;
            //更新有效数据长度字段

            buf->len += total_len;
            ret = total_len;
            goto out;
        }
            
    }

    // 若全部可写(可读缓冲区数为0),

    // 或写入数据长度大于管道缓冲区的长度单位(PAGE_SIZE)

    for (;;) {
        int bufs;
        //是否有读者进程存在

        if (!PIPE_READERS(*inode)) {
            send_sig(SIGPIPE, current, 0);
            if (!ret) ret = -EPIPE;
            break;
        }
        //获取读缓冲区数

        bufs = info->nrbufs;
        if (bufs < PIPE_BUFFERS) {
            ssize_t chars;
            //用第一个可读缓冲区+可读缓冲区数得到可写(空)缓冲区的地址

            int newbuf = (info->curbuf + bufs) & (PIPE_BUFFERS-1);
            struct pipe_buffer *buf = info->bufs + newbuf;
            struct page *page = info->tmp_page;
            int error;

            //若page的值为空,从伙伴系统中获取一页

            if (!page) {
                page = alloc_page(GFP_HIGHUSER);
                if (unlikely(!page)) {
                    ret = ret ? : -ENOMEM;
                    break;
                }
                info->tmp_page = page;
            }
            /* Always wakeup, even if the copy fails. Otherwise
             * we lock up (O_NONBLOCK-)readers that sleep due to
             * syscall merging.
             * FIXME! Is this really true?
             */

            do_wakeup = 1;
            chars = PAGE_SIZE;
            if (chars > total_len)
                chars = total_len;

            //写chars字节到缓冲区中

            error = pipe_iov_copy_from_user(kmap(page), iov, chars);
            kunmap(page);
            if (unlikely(error)) {
                if (!ret) ret = -EFAULT;
                break;
            }
            ret += chars;

            /* Insert it into the buffer array */
            /更新nrbufs,和len字段。
            buf->page = page;
            buf->ops = &anon_pipe_buf_ops;
            buf->offset = 0;
            buf->len = chars;
            info->nrbufs = ++bufs;
            info->tmp_page = NULL;

            //若没有写完继续写入剩下的数据

            total_len -= chars;
            if (!total_len)
                break;
        }
        //还有可写缓冲区,继续写

        if (bufs < PIPE_BUFFERS)
            continue;
        //若设置非阻塞,

        //若没有写入任何的数据ret=0,此时返回错误

        //若已经写完了数据,结束写操作。

        if (filp->f_flags & O_NONBLOCK) {
            if (!ret) ret = -EAGAIN;
            break;
        }
        if (signal_pending(current)) {
            if (!ret) ret = -ERESTARTSYS;
            break;
        }
        if (do_wakeup) {
            wake_up_interruptible_sync(PIPE_WAIT(*inode));
            kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
            do_wakeup = 0;
        }
        PIPE_WAITING_WRITERS(*inode)++;
        pipe_wait(inode);
        PIPE_WAITING_WRITERS(*inode)--;
    }
out:
    up(PIPE_SEM(*inode));
    if (do_wakeup) {
        wake_up_interruptible(PIPE_WAIT(*inode));
        kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
    }
    if (ret > 0)
        inode_update_time(inode, 1);    /* mtime and ctime */
    return ret;
}


阅读(2305) | 评论(0) | 转发(0) |
给主人留下些什么吧!~~