read() system call

http://debug-sai.blogbus.com/logs/51711788.html

asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
{
//傳入文件fd
 struct file *file;
 ssize_t ret = -EBADF;
 int fput_needed;
//根據文件fd得到file
 file = fget_light(fd, &fput_needed);
 if (file) {
  //讀出當前偏移
  loff_t pos = file_pos_read(file);
  //從當前偏移讀，pos返回讀取後的偏移
  ret = vfs_read(file, buf, count, &pos);
  //設置新偏移
  file_pos_write(file, pos);
  fput_light(file, fput_needed);//
 }
 return ret;
}
看看fget_light是怎麼根據fd得到file的
struct file *fget_light(unsigned int fd, int *fput_needed)
{
 struct file *file;
 struct files_struct *files = current->files;
 *fput_needed = 0;
 if (likely((atomic_read(&files->count) == 1))) {
//若已經打開了
  file = fcheck_files(files, fd);
 } else {
//若沒有打開，加鎖
  rcu_read_lock();
  file = fcheck_files(files, fd);
  if (file) {
   if (atomic_long_inc_not_zero(&file->f_count))
    *fput_needed = 1;
   else
    /* Didn't get the reference, someone's freed */
    file = NULL;
  }
  rcu_read_unlock();
 }
 return file;

fput_light的流程如下，具體見源代碼：
fput_light->fput(當所有釋放完時還要調用__fput)
接著看 vfs_read：
ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
{
 ssize_t ret;
 //下面都是一系列驗證
 if (!(file->f_mode & FMODE_READ))
  return -EBADF;
 if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
  return -EINVAL;
 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
  return -EFAULT;
 ret = rw_verify_area(READ, file, pos, count);
 if (ret >= 0) {
  count = ret;
 
  if (file->f_op->read)
   //一般都是到這裡調用vfs的read
   ret = file->f_op->read(file, buf, count, pos);
  else
   ret = do_sync_read(file, buf, count, pos);
  if (ret > 0) {
   fsnotify_access(file->f_path.dentry);
   add_rchar(current, ret);
  }
  inc_syscr(current);
 }
 return ret;
}
跟著就是vfs設置的read函數了，對於ext2來說是
const struct file_operations ext2_file_operations = {
 .llseek  = generic_file_llseek,
 .read  = do_sync_read,
 .write  = do_sync_write,
 .aio_read = generic_file_aio_read,
 .aio_write = generic_file_aio_write,
 .unlocked_ioctl = ext2_ioctl,
#ifdef CONFIG_COMPAT
 .compat_ioctl = ext2_compat_ioctl,
#endif
 .mmap  = generic_file_mmap,
 .open  = generic_file_open,
 .release = ext2_release_file,
 .fsync  = ext2_sync_file,
 .splice_read = generic_file_splice_read,
 .splice_write = generic_file_splice_write,
};
所以是do_sync_read：
ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
{
 struct iovec iov = { .iov_base = buf, .iov_len = len };
 struct kiocb kiocb;
 ssize_t ret;
 //初始化kiocb
 init_sync_kiocb(&kiocb, filp);
 kiocb.ki_pos = *ppos;
 kiocb.ki_left = len;
 for (;;) {
  //其實就是變了個參數傳到generic_file_aio_read中。。。具體分析就先到這了
  ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
  if (ret != -EIOCBRETRY)
   break;
  wait_on_retry_sync_kiocb(&kiocb);
 }
 if (-EIOCBQUEUED == ret)
  ret = wait_on_sync_kiocb(&kiocb);
 *ppos = kiocb.ki_pos;
 return ret;
}