Linux Kernel系列 - VFS核心数据结构

Hanks.Wang - 专注于操作系统与移动安全研究，Linux-Kernel/SELinux/SEAndroid/TrustZone/Encription/MDM    Mail - byhankswang@gmail.com

VFS核心数据结构

本文中参照的Linux Kernel版本为3.4.5

URL: https://www.kernel.org/pub/linux/kernel/v3.x/linux-3.4.5.tar.gz

阅读Linux文件系统源代码的重要性

Linux VFS通过抽象的手段对多种文件系统类型均有良好的支持，通过阅读VFS源码可以很好的学习使用C语言开发面向对象的大型项目；Linux系统中本身对物理设备抽象成文件的设计，通过阅读VFS源代码有助于深度理解设备管理和文件系统关系；Linux程序开发中大量的glibc API的实现都和VFS有一定的关系，通过阅读VFS源代码有助于理解和帮助用户态和内核态程序的开发；由于VFS在内核中的重要地位，通过阅读VFS源代码有助于理解内核中其他模块的实现。

VFS中比较重要的数据结构包括struct super_block, struct file_system_type, struct super_operations, struct dentry,struct inode, struct inode_operations, struct file, struct file_operations ;其中相关的基本数据结构请认真阅读linux源码文件include/linux/fs.h

                                                  

图1 - linux文件系统体系结构

1.超级块结构体 super_block

struct super_block定义在include/linux/fs.h:1418

struct super_block {
    struct list_heads_list;/* Keep this first */
    dev_t s_dev;/* search index; _not_ kdev_t */
    unsigned char s_dirt;
    unsigned char s_blocksize_bits;
    unsigned long s_blocksize;
    loff_ts_maxbytes;/* Max file size */
    struct file_system_type* s_type;
    const struct super_operations* s_op;
    const struct dquot_operations* dq_op;
    const struct quotactl_ops* s_qcop;
    const struct export_operations * s_export_op;
    unsigned longs_flags;
    unsigned longs_magic;
    struct dentry*s_root;
    struct rw_semaphore s_umount;
    struct mutex s_lock;
    ints_count;
    atomic_t s_active;
    #ifdef CONFIG_SECURITY
    void                    *s_security;
    #endif
    const struct xattr_handler ** s_xattr;


    struct list_heads_inodes;/* all inodes */
    struct hlist_bl_head s_anon;/* anonymous dentries for (nfs) exporting */
    #ifdef CONFIG_SMP
    struct list_head __percpu *s_files;
    #else
    struct list_heads_files;
    #endif

    struct list_heads_mounts; /* list of mounts; _not_ for fs use */
    /* s_dentry_lru, s_nr_dentry_unused protected by dcache.c lru locks */
    struct list_heads_dentry_lru;/* unused dentry lru */
    ints_nr_dentry_unused;/* # of dentry on lru */


    /* s_inode_lru_lock protects s_inode_lru and s_nr_inodes_unused */
    spinlock_t s_inode_lru_lock ____cacheline_aligned_in_smp;
    struct list_head s_inode_lru;/* unused inode lru */
    ints_nr_inodes_unused;/* # of inodes on lru */


    struct block_device* s_bdev;
    struct backing_dev_info * s_bdi;
    struct mtd_info* s_mtd;
    struct hlist_node s_instances;
    struct quota_info s_dquot;/* Diskquota specific options */


    int s_frozen;
    wait_queue_head_ts_wait_unfrozen;


    char s_id[32];/* Informational name */
    u8 s_uuid[16];/* UUID */


    void * s_fs_info;/* Filesystem private info */
    unsigned int s_max_links;
    fmode_t s_mode;


    /* Granularity of c/m/atime in ns.
       Cannot be worse than a second */
    u32   s_time_gran;


    /*
    * The next field is for VFS *only*. No filesystems have any business
    * even looking at it. You had been warned.
    */
    struct mutex s_vfs_rename_mutex;/* Kludge */


    /*
    * Filesystem subtype.  If non-empty the filesystem type field
    * in /proc/mounts will be "type.subtype"
    */
    char *s_subtype;


    /*
    * Saved mount options for lazy filesystems using
    * generic_show_options()
    */
    char __rcu *s_options;
    const struct dentry_operations *s_d_op; /* default d_op for dentries */


    /*
    * Saved pool identifier for cleancache (-1 means none)
    */
    int cleancache_poolid;


    struct shrinker s_shrink;/* per-sb shrinker handle */


    /* Number of inodes with nlink == 0 but still referenced */
    atomic_long_t s_remove_count;


     /* Being remounted read-only */
    int s_readonly_remount;
};

2. 文件系统类型 file_system_type

struct file_system_type定义在include/linux/fs.h:1835

struct file_system_type{
    const char *name;
    int fs_flags;
    struct dentry *(*mount) (struct file_system_type *, int,
              const char *, void *);
    void (*kill_sb) (struct super_block *);
    struct module *owner;
    struct file_system_type * next;
    struct hlist_head fs_supers;


    struct lock_class_key s_lock_key;
    struct lock_class_key s_umount_key;
    struct lock_class_key s_vfs_rename_key;


    struct lock_class_key i_lock_key;
    struct lock_class_key i_mutex_key;
    struct lock_class_key i_mutex_dir_key;
};

3. 文件系统超级操作 super_operations

super_operations定义在include/linux/fs.h1864行

struct super_operations {

    struct inode *(*alloc_inode)(struct super_block *sb);
    void (*destroy_inode)(struct inode *);


    void (*dirty_inode) (struct inode *, int flags);
    int (*write_inode) (struct inode *, struct writeback_control *wbc);
    int (*drop_inode) (struct inode *);
    void (*evict_inode) (struct inode *);
    void (*put_super) (struct super_block *);
    void (*write_super) (struct super_block *);
    int (*sync_fs)(struct super_block *sb, int wait);
    int (*freeze_fs) (struct super_block *);
    int (*unfreeze_fs) (struct super_block *);
    int (*statfs) (struct dentry *, struct kstatfs *);
    int (*remount_fs) (struct super_block *, int *, char *);
    void (*umount_begin) (struct super_block *);


    int (*show_options)(struct seq_file *, struct dentry *);
    int (*show_devname)(struct seq_file *, struct dentry *);
    int (*show_path)(struct seq_file *, struct dentry *);
    int (*show_stats)(struct seq_file *, struct dentry *);

    #ifdef CONFIG_QUOTA
    ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
    ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
    #endif

    int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
    int (*nr_cached_objects)(struct super_block *);
    void (*free_cached_objects)(struct super_block *, int);
};

相关的参考文献：

https://www.ibm.com/developerworks/cn/aix/library/au-speakingunix14/