Linux虚拟文件系统之文件系统卸载（sys_umount())

Linux中卸载文件系统由umount系统调用实现，入口函数为sys_umount()。较于文件系统的安装较为简单，下面是具体的实现。

   1. /*sys_umont系统调用*/  
   2. SYSCALL_DEFINE2(umount, char __user *, name, int, flags)  
   3. {  
   4.     struct path path;  
   5.     int retval;  
   6.     /*找到装载点的vfsmount实例和dentry实例，二者包装 
   7.     在一个nameidata结构中*/  
   8.     retval = user_path(name, &path);  
   9.     if (retval)  
  10.         goto out;  
  11.     retval = -EINVAL;  
  12.     /*如果查找的最终目录不是文件系统的挂载点*/  
  13.     if (path.dentry != path.mnt->mnt_root)  
  14.         goto dput_and_out;  
  15.     /*如果要卸载的文件系统还没有安装在命名空间中*/  
  16.     if (!check_mnt(path.mnt))  
  17.         goto dput_and_out;  
  18.   
  19.     retval = -EPERM;  
  20.     /*如果用户不具有卸载文件系统的特权*/  
  21.     if (!capable(CAP_SYS_ADMIN))  
  22.         goto dput_and_out;  
  23.     /*实际umount工作*/  
  24.     retval = do_umount(path.mnt, flags);  
  25. dput_and_out:  
  26.     /* we mustn't call path_put() as that would clear mnt_expiry_mark */  
  27.     dput(path.dentry);  
  28.     mntput_no_expire(path.mnt);  
  29. out:  
  30.     return retval;  
  31. }  

卸载实际工作

   1. static int do_umount(struct vfsmount *mnt, int flags)  
   2. {  
   3.     /*从vfsmount对象的mnt_sb字段检索超级块对象sb的地址*/  
   4.     struct super_block *sb = mnt->mnt_sb;  
   5.     int retval;  
   6.     /*初始化umount_list，该链表在后面的释放中会做临时链表 
   7.     用*/  
   8.     LIST_HEAD(umount_list);  
   9.   
  10.     retval = security_sb_umount(mnt, flags);  
  11.     if (retval)  
  12.         return retval;  
  13.   
  14.     /* 
  15.      * Allow userspace to request a mountpoint be expired rather than 
  16.      * unmounting unconditionally. Unmount only happens if: 
  17.      *  (1) the mark is already set (the mark is cleared by mntput()) 
  18.      *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 
  19.      */  
  20.      /*如果设置了MNT_EXPIRE标志，即要标记挂载点“到期”*/  
  21.     if (flags & MNT_EXPIRE) {  
  22.         /*若要卸载的文件系统是根文件系统或者同时设置了 
  23.         MNT_FORCE或MNT_DETACH，则返回-EINVAL*/  
  24.         if (mnt == current->fs->root.mnt ||  
  25.             flags & (MNT_FORCE | MNT_DETACH))  
  26.             return -EINVAL;  
  27.         /*检查vfsmount的引用计数，若不为2，则返回-EBUSY， 
  28.         要卸载的文件系统在卸载的时候不能有引用者， 
  29.         这个2代表vfsmount的父vfsmount和sys_umount()对本对象的引用*/  
  30.         if (atomic_read(&mnt->mnt_count) != 2)  
  31.             return -EBUSY;  
  32.         /*设置vfsmount对象的mnt_expiry_mark字段为1。*/  
  33.         if (!xchg(&mnt->mnt_expiry_mark, 1))  
  34.             return -EAGAIN;  
  35.     }  
  36.   
  37.     /* 
  38.      * If we may have to abort operations to get out of this 
  39.      * mount, and they will themselves hold resources we must 
  40.      * allow the fs to do things. In the Unix tradition of 
  41.      * 'Gee thats tricky lets do it in userspace' the umount_begin 
  42.      * might fail to complete on the first run through as other tasks 
  43.      * must return, and the like. Thats for the mount program to worry 
  44.      * about for the moment. 
  45.      */  
  46.      /*如果用户要求强制卸载操作，则调用umount_begin 
  47.      超级块操作中断任何正在进行的安装操作*/  
  48.     /*当然如果特定的文件系统定义了下面函数则调用它*/  
  49.     if (flags & MNT_FORCE && sb->s_op->umount_begin) {  
  50.         sb->s_op->umount_begin(sb);  
  51.     }  
  52.   
  53.     /* 
  54.      * No sense to grab the lock for this test, but test itself looks 
  55.      * somewhat bogus. Suggestions for better replacement? 
  56.      * Ho-hum... In principle, we might treat that as umount + switch 
  57.      * to rootfs. GC would eventually take care of the old vfsmount. 
  58.      * Actually it makes sense, especially if rootfs would contain a 
  59.      * /reboot - static binary that would close all descriptors and 
  60.      * call reboot(9). Then init(8) could umount root and exec /reboot. 
  61.      */  
  62.      /*如果要卸载的文件系统是根文件系统，且用户 
  63.      并不要求真正地把它卸载下来（即设置了MNT_DETACH标志， 
  64.      这个标志仅仅标记挂载点为不能再访问，知道挂载不busy 
  65.      时才卸载），则调用do_remount_sb()重新安装根文件系统为只 
  66.      读并终止，并返回do_remount_sb()的返回值。*/  
  67.     if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {  
  68.         /* 
  69.          * Special case for "unmounting" root ... 
  70.          * we just try to remount it readonly. 
  71.          */  
  72.         down_write(&sb->s_umount);  
  73.         if (!(sb->s_flags & MS_RDONLY))  
  74.             retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);  
  75.         up_write(&sb->s_umount);  
  76.         return retval;  
  77.     }  
  78.       
  79.     down_write(&namespace_sem);  
  80.     /*为进行写操作而获取当前进程的namespace_sem读/写信号量和vfsmount_lock自旋锁*/  
  81.     spin_unlock(&vfsmount_lock);  
  82.     spin_lock(&vfsmount_lock);  
  83.     event++;  
  84.   
  85.     if (!(flags & MNT_DETACH))  
  86.         shrink_submounts(mnt, &umount_list);  
  87.   
  88.     retval = -EBUSY;  
  89.     /*如果已安装文件系统不包含任何子安装文件系统的安装点，或者用户要求强制 
  90.     卸载文件系统，则调用umount_tree()卸载文件系统（及其所有子文件系统）。*/  
  91.     if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {  
  92.         if (!list_empty(&mnt->mnt_list))  
  93.             /*完成实际的底层的卸载文件系统的任务。首先他将mnt的所有孩子移动至kill链表中， 
  94.             也就是传递进去的umount_list，然后将kill链表中的所有的vfsmount对象的一些字段设为无效状态。 
  95.             */  
  96.             umount_tree(mnt, 1, &umount_list);  
  97.         retval = 0;  
  98.     }  
  99.       
 100.     if (retval)  
 101.         security_sb_umount_busy(mnt);  
 102.     /*释放vfsmount_lock自旋锁和当前进程的namespace_sem读/写信号量*/  
 103.     up_write(&namespace_sem);  
 104.     /*减小相应文件系统根目录的目录项对象和已经安装文件系统 
 105.     描述符的引用计数器值，这些计数器值由path_lookup()增加*/  
 106.     release_mounts(&umount_list);  
 107.     return retval;  
 108. }  

从内核链表中脱离

   1. /*完成实际的底层的卸载文件系统的任务。首先他将mnt的所有子移动至kill链表中， 
   2. 也就是传递进去的umount_list，然后将kill链表中的所有的vfsmount对象的一些字段设为无效状态。 
   3. */  
   4. void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)  
   5. {  
   6.     struct vfsmount *p;  
   7.   
   8.     for (p = mnt; p; p = next_mnt(p, mnt))  
   9.         list_move(&p->mnt_hash, kill);  
  10.   
  11.     if (propagate)  
  12.         propagate_umount(kill);  
  13.   
  14.     list_for_each_entry(p, kill, mnt_hash) {  
  15.         list_del_init(&p->mnt_expire);  
  16.         list_del_init(&p->mnt_list);  
  17.         __touch_mnt_namespace(p->mnt_ns);  
  18.         p->mnt_ns = NULL;  
  19.         list_del_init(&p->mnt_child);  
  20.         if (p->mnt_parent != p) {  
  21.             p->mnt_parent->mnt_ghosts++;  
  22.             p->mnt_mountpoint->d_mounted--;  
  23.         }  
  24.         change_mnt_propagation(p, MS_PRIVATE);  
  25.     }  
  26. }  

释放引用计数

   1. void release_mounts(struct list_head *head)  
   2. {  
   3.     struct vfsmount *mnt;  
   4.     while (!list_empty(head)) {  
   5.         mnt = list_first_entry(head, struct vfsmount, mnt_hash);  
   6.         list_del_init(&mnt->mnt_hash);  
   7.         if (mnt->mnt_parent != mnt) {  
   8.             struct dentry *dentry;  
   9.             struct vfsmount *m;  
  10.             spin_lock(&vfsmount_lock);  
  11.             dentry = mnt->mnt_mountpoint;  
  12.             m = mnt->mnt_parent;  
  13.             mnt->mnt_mountpoint = mnt->mnt_root;  
  14.             mnt->mnt_parent = mnt;  
  15.             m->mnt_ghosts--;  
  16.             spin_unlock(&vfsmount_lock);  
  17.             /*下面两个函数为减小引用计数，减到0时释放*/  
  18.             dput(dentry);             
  19.             mntput(m);  
  20.         }  
  21.         /*vfsmount对象所占的内存空间最终在mntput()函数中释放*/  
  22.         mntput(mnt);  
  23.     }  
  24. }