Linux Kernel Development——虛擬檔案系統

文章目錄

• 通用檔案系統介面
• 超級塊對象

虛擬檔案系統（VFS）為使用者空間提供了檔案系統相關的介面，使用者程式可以通過標準的Unix檔案系統調用對不同介質上的不同檔案系統進行讀寫操作。

通用檔案系統介面

VFS使得使用者可以直接使用open()、read()和write()而無需考慮具體的檔案系統和實際物理介質。標準系統調用也可以在不同的介質和檔案系統之間執行，VFS負責這種不同介質和不同檔案系統之間的協調，並對上提供一種通用的存取方法。

之所以這種通用介面對所有類型的檔案系統都可以操作，是因為核心在它的底層檔案系統之上建立了一個抽象層。這個抽象層提供了一個通用檔案系統模型，支援各種檔案系統。VFS定義了所有檔案系統都支援的基本資料結構和介面，而實際檔案系統都實現了這些基本介面。由於實際檔案系統的代碼在統一的介面和資料結構下隱藏了實現的細節，所以在VFS層和核心的其他部分來看，所有的檔案系統都是相同的。

VFS中有四個主要的物件模型，分別是：

• 超級塊對象：一個已安裝的檔案系統；
• 索引節點對象：代表一個檔案；
• 目錄項對象：代表路徑的一個組成部分；
• 檔案對象：代表檔案，注意目錄也是檔案。

每種物件模型核心都定義了對應的操作對象，描述了核心針對該對象可以使用的方法。

超級塊對象

每種檔案系統都必須實現超級塊，用於儲存特定檔案系統的資訊，通常對應於存放在磁碟特定扇區中的檔案系統超級塊或檔案系統控制塊。對於非基於磁碟的檔案系統，會在使用現場建立超級塊並儲存在記憶體中。

超級塊用struct super_block結構體表示：

1400struct super_block {1401        struct list_head        s_list;         /* Keep this first 指向超級塊鏈表的指標 */1402        dev_t                   s_dev;          /* search index; _not_ kdev_t 裝置標誌符 */1403        unsigned char           s_dirt;　　　　　/* 修改（髒）標誌 */1404        unsigned char           s_blocksize_bits; /* 塊大小 單位bits */1405        unsigned long           s_blocksize;　　/* 塊大小 單位Bytes*/1406        loff_t                  s_maxbytes;     /* Max file size */1407        struct file_system_type *s_type;1408        const struct super_operations   *s_op;　　/× 超級塊方法 ×/1409        const struct dquot_operations   *dq_op;  /× 磁碟限額方法 ×/1410        const struct quotactl_ops       *s_qcop; /× 限額控制方法 ×/1411        const struct export_operations *s_export_op; /× 匯出方法 ×/1412        unsigned long           s_flags;1413        unsigned long           s_magic;1414        struct dentry           *s_root;1415        struct rw_semaphore     s_umount;1416        struct mutex            s_lock;1417        int                     s_count;1418        atomic_t                s_active;1419#ifdef CONFIG_SECURITY1420        void                    *s_security;1421#endif1422        const struct xattr_handler **s_xattr;14231424        struct list_head        s_inodes;       /* all inodes */1425        struct hlist_bl_head    s_anon;         /* anonymous dentries for (nfs) exporting */1426#ifdef CONFIG_SMP1427        struct list_head __percpu *s_files;1428#else1429        struct list_head        s_files;1430#endif1431        /* s_dentry_lru, s_nr_dentry_unused protected by dcache.c lru locks */1432        struct list_head        s_dentry_lru;   /* unused dentry lru */1433        int                     s_nr_dentry_unused;     /* # of dentry on lru */14341435        /* s_inode_lru_lock protects s_inode_lru and s_nr_inodes_unused */1436        spinlock_t              s_inode_lru_lock ____cacheline_aligned_in_smp;1437        struct list_head        s_inode_lru;            /* unused inode lru */1438        int                     s_nr_inodes_unused;     /* # of inodes on lru */14391440        struct block_device     *s_bdev;1441        struct backing_dev_info *s_bdi;1442        struct mtd_info         *s_mtd;1443        struct list_head        s_instances;1444        struct quota_info       s_dquot;        /* Diskquota specific options */14451446        int                     s_frozen;1447        wait_queue_head_t       s_wait_unfrozen;14481449        char s_id[32];                          /* Informational name */1450        u8 s_uuid[16];                          /* UUID */14511452        void                    *s_fs_info;     /* Filesystem private info */1453        fmode_t                 s_mode;14541455        /* Granularity of c/m/atime in ns.1456           Cannot be worse than a second */1457        u32                s_time_gran;14581459        /*1460         * The next field is for VFS *only*. No filesystems have any business1461         * even looking at it. You had been warned.1462         */1463        struct mutex s_vfs_rename_mutex;        /* Kludge */14641465        /*1466         * Filesystem subtype.  If non-empty the filesystem type field1467         * in /proc/mounts will be "type.subtype"1468         */1469        char *s_subtype;14701471        /*1472         * Saved mount options for lazy filesystems using1473         * generic_show_options()1474         */1475        char __rcu *s_options;1476        const struct dentry_operations *s_d_op; /* default d_op for dentries */14771478        /*1479         * Saved pool identifier for cleancache (-1 means none)1480         */1481        int cleancache_poolid;14821483        struct shrinker s_shrink;       /* per-sb shrinker handle */1484};

超級塊對象中的s_op定義了超級塊的操作函數表，用super_operations結構體表示，其中的每一項都定義了一種操作的函數指標：

1658struct super_operations {1659        struct inode *(*alloc_inode)(struct super_block *sb);1660        void (*destroy_inode)(struct inode *);16611662        void (*dirty_inode) (struct inode *, int flags);1663        int (*write_inode) (struct inode *, struct writeback_control *wbc);1664        int (*drop_inode) (struct inode *);1665        void (*evict_inode) (struct inode *);1666        void (*put_super) (struct super_block *);1667        void (*write_super) (struct super_block *);1668        int (*sync_fs)(struct super_block *sb, int wait);1669        int (*freeze_fs) (struct super_block *);1670        int (*unfreeze_fs) (struct super_block *);1671        int (*statfs) (struct dentry *, struct kstatfs *);1672        int (*remount_fs) (struct super_block *, int *, char *);1673        void (*umount_begin) (struct super_block *);16741675        int (*show_options)(struct seq_file *, struct vfsmount *);1676        int (*show_devname)(struct seq_file *, struct vfsmount *);1677        int (*show_path)(struct seq_file *, struct vfsmount *);1678        int (*show_stats)(struct seq_file *, struct vfsmount *);1679#ifdef CONFIG_QUOTA1680        ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);1681        ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);1682#endif1683        int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);1684        int (*nr_cached_objects)(struct super_block *);1685        void (*free_cached_objects)(struct super_block *, int);1686};

在具體的檔案系統中，存在一個指向超級塊結構的指標，其內會向該結構體傳遞對應該檔案系統的操作函數實現。上述的動作表也不必全部實現，檔案系統可以將不需要的函數指標設定為NULL。另外，上述函數均在進程上下文中調用，必要時均可阻塞。

索引節點對象
索引節點對象包含了核心在操作檔案和目錄時需要的全部資訊，這些資訊可以從磁碟索引節點直接讀入。
索引節點使用inode結構體表示，一個索引節點代表了檔案系統中的一個檔案，它也可以是裝置或者管道這樣的特殊檔案。

 744/* 745 * Keep mostly read-only and often accessed (especially for 746 * the RCU path lookup and 'stat' data) fields at the beginning 747 * of the 'struct inode' 748 */ 749struct inode { 750        umode_t                 i_mode; 751        unsigned short          i_opflags; 752        uid_t                   i_uid; 753        gid_t                   i_gid; 754        unsigned int            i_flags; 755 756#ifdef CONFIG_FS_POSIX_ACL 757        struct posix_acl        *i_acl; 758        struct posix_acl        *i_default_acl; 759#endif 760 761        const struct inode_operations   *i_op; 762        struct super_block      *i_sb; 763        struct address_space    *i_mapping; 764 765#ifdef CONFIG_SECURITY 766        void                    *i_security; 767#endif 768 769        /* Stat data, not accessed from path walking */ 770        unsigned long           i_ino; 771        /* 772         * Filesystems may only read i_nlink directly.  They shall use the 773         * following functions for modification: 774         * 775         *    (set|clear|inc|drop)_nlink 776         *    inode_(inc|dec)_link_count 777         */ 778        union { 779                const unsigned int i_nlink; 780                unsigned int __i_nlink; 781        }; 782        dev_t                   i_rdev; 783        struct timespec         i_atime; 784        struct timespec         i_mtime; 785        struct timespec         i_ctime; 786        spinlock_t              i_lock; /* i_blocks, i_bytes, maybe i_size */ 787        unsigned short          i_bytes; 788        blkcnt_t                i_blocks; 789        loff_t                  i_size; 790 791#ifdef __NEED_I_SIZE_ORDERED 792        seqcount_t              i_size_seqcount; 793#endif 794 795        /* Misc */ 796        unsigned long           i_state; 797        struct mutex            i_mutex; 798 799        unsigned long           dirtied_when;   /* jiffies of first dirtying */ 800 801        struct hlist_node       i_hash; 802        struct list_head        i_wb_list;      /* backing dev IO list */ 803        struct list_head        i_lru;          /* inode LRU list */ 804        struct list_head        i_sb_list; 805        union { 806                struct list_head        i_dentry; 807                struct rcu_head         i_rcu; 808        }; 809        atomic_t                i_count; 810        unsigned int            i_blkbits; 811        u64                     i_version; 812        atomic_t                i_dio_count; 813        atomic_t                i_writecount; 814        const struct file_operations    *i_fop; /* former ->i_op->default_file_ops */ 815        struct file_lock        *i_flock; 816        struct address_space    i_data; 817#ifdef CONFIG_QUOTA 818        struct dquot            *i_dquot[MAXQUOTAS]; 819#endif 820        struct list_head        i_devices; 821        union { 822                struct pipe_inode_info  *i_pipe; 823                struct block_device     *i_bdev; 824                struct cdev             *i_cdev; 825        }; 826 827        __u32                   i_generation; 828 829#ifdef CONFIG_FSNOTIFY 830        __u32                   i_fsnotify_mask; /* all events this inode cares about */ 831        struct hlist_head       i_fsnotify_marks; 832#endif 833 834#ifdef CONFIG_IMA 835        atomic_t                i_readcount; /* struct files open RO */ 836#endif 837        void                    *i_private; /* fs or device private pointer */ 838}; 839

其中，i_op定義了索引節點對象的所有操作方法：

1613struct inode_operations {
1614        struct dentry * (*lookup) (struct inode *,struct dentry *, struct nameidata *);
1615        void * (*follow_link) (struct dentry *, struct nameidata *);
1616        int (*permission) (struct inode *, int);
1617        struct posix_acl * (*get_acl)(struct inode *, int);
1618
1619        int (*readlink) (struct dentry *, char __user *,int);
1620        void (*put_link) (struct dentry *, struct nameidata *, void *);
1621
1622        int (*create) (struct inode *,struct dentry *,int, struct nameidata *);
1623        int (*link) (struct dentry *,struct inode *,struct dentry *);
1624        int (*unlink) (struct inode *,struct dentry *);
1625        int (*symlink) (struct inode *,struct dentry *,const char *);
1626        int (*mkdir) (struct inode *,struct dentry *,int);
1627        int (*rmdir) (struct inode *,struct dentry *);
1628        int (*mknod) (struct inode *,struct dentry *,int,dev_t);
1629        int (*rename) (struct inode *, struct dentry *,
1630                        struct inode *, struct dentry *);
1631        void (*truncate) (struct inode *);
1632        int (*setattr) (struct dentry *, struct iattr *);
1633        int (*getattr) (struct vfsmount *mnt, struct dentry *, struct kstat *);
1634        int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
1635        ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
1636        ssize_t (*listxattr) (struct dentry *, char *, size_t);
1637        int (*removexattr) (struct dentry *, const char *);
1638        void (*truncate_range)(struct inode *, loff_t, loff_t);
1639        int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start,
1640                      u64 len);