使用Fuse编写文件系统

FUSE的全称是”Filesystem in Userspace”,即“用户空间的文件系统”,这是一个内核模块,能够让用户在用户空间实现文件系统并且挂载到某个目录,就像在内核实现的文件系统一样。使用FUSE有几个好处:一是因为在用户空间实现,开发和调试都比较方便;二是可以把一些常用的服务以文件系统的形式展现,方便操作,如ftpfs,sshfs,mailfs等;另外可以避免一些版权问题,如linux上对ntfs,zfs的操作都是通过FUSE实现的。当然用户空间的实现也有缺点,最明显的就是由多次在用户态/内核态切换带来的性能下降。

 

根据参考资料[1]的介绍,用户通过FUSE和内核的通信过程如下:

 

          +----------------+
          | myfs /tmp/fuse |
          +----------------+
             |  ^
+--------------+     v  |
| ls /tmp/fuse |  +--------------+
+--------------+  |  libfuse  |
   ^ |     +--------------+
   | v        |  |
+--------------+  +--------------+
|   glibc  |  |   glibc  |
+--------------+  +--------------+
   ^ |        |  ^
~.~.~.|.~|~.~.~.~.~.~.~.~.|.~.|.~.~.~.~.~.~.~.~.
   | v        v  |
+--------------+  +--------------+
|       |----|   FUSE   |
|       |  +--------------+
|   VFS   |      ...
|       |  +--------------+
|       |----|   Ext3   |
+--------------+  +--------------+

从图中可以看到,FUSE和ext3一样,是内核里的一个文件系统模块。例如我们用FUSE实现了一个文件系统并挂载在/tmp/fuse,当我们对该目录执行ls时,内核里的FUSE从VFS获得参数,然后调用我们自己实现的myfs中相应的函数,得到结果后再通过VFS返回给ls。

以下实验的环境是debian 6,需要安装libfuse-dev,fuse-utils及其它相关依赖。

hello, world

下面是一个简单的文件系统oufs,只支持ls操作:

 

#define FUSE_USE_VERSION 26
 
#include <stdio.h>
#include <string.h>
#include <fuse.h>
 
static int ou_readdir(const char* path, void* buf, fuse_fill_dir_t filler,
           off_t offset, struct fuse_file_info* fi)
{
  return filler(buf, "hello-world", NULL, 0);
}
 
static int ou_getattr(const char* path, struct stat* st)
{
  memset(st, 0, sizeof(struct stat));
 
  if (strcmp(path, "/") == 0)
    st->st_mode = 0755 | S_IFDIR;
  else
    st->st_mode = 0644 | S_IFREG;
 
  return 0;
}
 
static struct fuse_operations oufs_ops = {
  .readdir  =  ou_readdir,
  .getattr  =  ou_getattr,
};
 
int main(int argc, char* argv[])
{
  return fuse_main(argc, argv, &oufs_ops, NULL);
}

 

和编译程序的Makefile:

 


CC := gcc
CFLAGS := -g -Wall -D_FILE_OFFSET_BITS=64
OBJS := $(patsubst %.c, %.o, $(wildcard *.c))
 
LIBS := -lfuse
 
TARGET := oufs
 
.PHONY: all clean
 
all: $(TARGET)
 
$(TARGET): $(OBJS)
$(CC) $(CFLAGS) -o $@ $^ $(LIBS)
 
.c.o:
$(CC) $(CFLAGS) -c $<
 
clean:
rm -f $(TARGET) $(OBJS)

编译成功后会看到生成的可执行文件oufs。建立一个挂载点/tmp/mnt,然后运行

./oufs /tmp/mnt

成功后试试”ls /tmp/mnt”,就能看到一个文件hello-world。要调试的时候可以加上-d选项,这样就能看到FUSE和自己printf的调试输出。

代码第一行指定了要使用的FUSE API版本。这里使用的是2.6版本。

要实现ls的功能,FUSE需要提供两个函数:readdir()和getattr(),这两个接口是struct fuse_operations里的两个函数指针:

/usr/include/fuse/fuse.h
/** Function to add an entry in a readdir() operation
 *
 * @param buf the buffer passed to the readdir() operation
 * @param name the file name of the directory entry
 * @param stat file attributes, can be NULL
 * @param off offset of the next entry or zero
 * @return 1 if buffer is full, zero otherwise
 */
typedef int (*fuse_fill_dir_t) (void *buf, const char *name,
                const struct stat *stbuf, off_t off);
 
struct fuse_operations {
    /** Get file attributes.
     *
     * Similar to stat(). The 'st_dev' and 'st_blksize' fields are
     * ignored.   The 'st_ino' field is ignored except if the 'use_ino'
     * mount option is given.
     */
    int (*getattr) (const char *, struct stat *);
 
    /** Read directory
     *
     * This supersedes the old getdir() interface. New applications
     * should use this.
     *
     * The filesystem may choose between two modes of operation:
     *
     * 1) The readdir implementation ignores the offset parameter, and
     * passes zero to the filler function's offset. The filler
     * function will not return '1' (unless an error happens), so the
     * whole directory is read in a single readdir operation. This
     * works just like the old getdir() method.
     *
     * 2) The readdir implementation keeps track of the offsets of the
     * directory entries. It uses the offset parameter and always
     * passes non-zero offset to the filler function. When the buffer
     * is full (or an error happens) the filler function will return
     * '1'.
     *
     * Introduced in version 2.3
     */
    int (*readdir) (const char *, void *, fuse_fill_dir_t, off_t,
            struct fuse_file_info *); 
 
    ......
};

这里要实现getattr()是因为我们要遍历根目录的内容,要通过getattr()获取根目录权限等信息。getattr()实现类似stat()的功能,返回相关的信息如文件权限,类型,uid等。参数里的path有可能是文件或目录或软链接或其它,因此除了权限外还要填充类型信息。程序里除了根目录就只有一个文件hello-world,因此只有目录(S_IFDIR)和普通文件(S_IFREG)两种情况。使用”ls -l”查看/tmp/mnt下的内容可以发现,hello-world的link数,修改时间等都是错误的,那是因为ou_getattr()实现中没有填充这些信息。查看manpage可以知道stat都有哪些字段。

readdir()的参数fuse_fill_dir_t是一个函数指针,每次往buf中填充一个entry的信息。程序中的ou_readdir()采用了注释中的第一种模式,即把所有的entry一次性放到buffer中。如果entry较多,把readdir()中的offset传给filler即可,buffer满了filler返回1。

最后在main函数中调用的是fuse_main(),把文件系统的操作函数和参数(如这里的挂载点/tmp/mnt)传给FUSE。

创建/删除文件

接下来让我们扩充上面的程序,增加创建/删除普通文件的功能:

#define FUSE_USE_VERSION 26
 
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#include <fuse.h>
 
#include "list.h"
 
#define MAX_NAMELEN 255
 
struct ou_entry {
  mode_t mode;
  struct list_node node;
  char name[MAX_NAMELEN + 1];
};
 
static struct list_node entries;
 
static int ou_readdir(const char* path, void* buf, fuse_fill_dir_t filler,
           off_t offset, struct fuse_file_info* fi)
{
  struct list_node* n;
 
  filler(buf, ".", NULL, 0);
  filler(buf, "..", NULL, 0);
 
  list_for_each (n, &entries) {
    struct ou_entry* o = list_entry(n, struct ou_entry, node);
    filler(buf, o->name, NULL, 0);
  }
 
  return 0;
}
 
static int ou_getattr(const char* path, struct stat* st)
{
  struct list_node* n;
 
  memset(st, 0, sizeof(struct stat));
 
  if (strcmp(path, "/") == 0) {
    st->st_mode = 0755 | S_IFDIR;
    st->st_nlink = 2;
    st->st_size = 0;
 
    list_for_each (n, &entries) {
      struct ou_entry* o = list_entry(n, struct ou_entry, node);
      ++st->st_nlink;
      st->st_size += strlen(o->name);
    }
 
    return 0;
  }
 
  list_for_each (n, &entries) {
    struct ou_entry* o = list_entry(n, struct ou_entry, node);
    if (strcmp(path + 1, o->name) == 0) {
      st->st_mode = o->mode;
      st->st_nlink = 1;
      return 0;
    }
  }
 
  return -ENOENT;
}
 
static int ou_create(const char* path, mode_t mode, struct fuse_file_info* fi)
{
  struct ou_entry* o;
  struct list_node* n;
 
  if (strlen(path + 1) > MAX_NAMELEN)
    return -ENAMETOOLONG;
 
  list_for_each (n, &entries) {
    o = list_entry(n, struct ou_entry, node);
    if (strcmp(path + 1, o->name) == 0)
      return -EEXIST;
  }
 
  o = malloc(sizeof(struct ou_entry));
  strcpy(o->name, path + 1); /* skip leading '/' */
  o->mode = mode | S_IFREG;
  list_add_prev(&o->node, &entries);
 
  return 0;
}
 
static int ou_unlink(const char* path)
{
  struct list_node *n, *p;
 
  list_for_each_safe (n, p, &entries) {
    struct ou_entry* o = list_entry(n, struct ou_entry, node);
    if (strcmp(path + 1, o->name) == 0) {
      __list_del(n);
      free(o);
      return 0;
    }
  }
 
  return -ENOENT;
}
 
static struct fuse_operations oufs_ops = {
  .getattr  =  ou_getattr,
  .readdir  =  ou_readdir,
  .create   =  ou_create,
  .unlink   =  ou_unlink,
};
 
int main(int argc, char* argv[])
{
  list_init(&entries);
 
  return fuse_main(argc, argv, &oufs_ops, NULL);
}

代码中用到的链表操作list.h见附录。本来想用c++的list,但是用g++编译时出错了,搜错误信息的时候发现了一个gcc的bug。多嘴插一句,clang3非常好用,错误提示非常人性化,谁用谁知道。

新增了两个函数ou_create()和ou_unlink(),分别用于创建和删除文件;增加了一个全局变量entries用于保存所有的entry;ou_readdir()和ou_getattr()也做了相应的修改。由于这些函数相当于实现VFS中的接口,因此不能在错误时返回-1了事,而是要根据不同的错误类型返回不同的值。关于错误代码头文件为errno.h,值和含义在/usr/include/asm-generic/errno-base.h和/usr/include/asm-generic/errno.h有定义。

编译挂载后试试执行”touch /tmp/mnt/abc”,然后ls看看是否多了一个文件abc?

程序中用到的list.h(相关原理可以参考这里):

#ifndef __LIST_H__
#define __LIST_H__
 
/* doubly linked list implementation from linux kernel */
 
#define offset_of(type, member) \
  ((unsigned long)(&(((type*)0)->member)))
 
#define container_of(ptr, type, member) \
  ((type*)((unsigned long)(ptr) - offset_of(type, member)))
 
#define list_entry(ptr, type, member) container_of(ptr, type, member)
 
struct list_node {
  struct list_node *prev, *next;
};
 
static inline void list_init(struct list_node* list)
{
  list->prev = list;
  list->next = list;
}
 
static inline void list_add_prev(struct list_node* node,
                 struct list_node* next)
{
  node->next = next;
  node->prev = next->prev;
  next->prev = node;
  node->prev->next = node;
}
 
static inline void __list_del(struct list_node* node)
{
  node->prev->next = node->next;
  node->next->prev = node->prev;
}
 
static inline void list_del(struct list_node* node)
{
  __list_del(node);
  list_init(node);
}
 
#define list_for_each(p, head) \
  for ((p) = (head)->next; (p) != (head); (p) = (p)->next)
 
#define list_for_each_safe(p, n, head) \
  for ((p) = (head)->next, (n) = (p)->next; \
     (p) != (head); \
     (p) = (n), (n) = (p)->next)
 
#endif

原文地址:http://ouonline.net/building-your-own-fs-with-fuse-1