2017 年 2 月 1 日 GitLab 数据库被误删引起了广大争议.
linux不像windows有个回收站，使用rm -rf *基本上文件是找不回来的。
那么问题来了：
对于linux下误删的文件，我们是否真的无法通过软件进行恢复呢?
答案当然是否定的，对于误删的文件，我们还是能通过软件恢复过来的。对于误删文件还原可以分为两种情况:
一种是删除以后在进程存在删除信息
一种是删除以后进程都找不到，只有借助于工具还原。
接下来以例子分别解说下两种不同的误删还原方式：
误删除文件进程还在的情况:
这种一般是有活动的进程存在持续标准输入或输出，到时文件被删除后，进程PID依旧存在。这也是有些服务器删除一些文件但是磁盘不释放的原因。
打开一个终端对一个测试文件做cat追加操作：
[root@docking ~]# echo "This is DeleteFile test." > deletefile.txt
[root@docking ~]# ls
deletefile.txt
[root@docking ~]# cat >> deletefile.txt
Add SomeLine into deletefile for fun.
打开另外一个终端查看这个文件可以清楚看到内容：
[root@docking ~]# ls
deletefile.txt
[root@docking ~]# cat deletefile.txt
This is DeleteFile test.
Add SomeLine into deletefile for fun.
此时，删除文件rm -f deletefile.txt
[root@docking ~]# rm -f deletefile.txt
[root@docking ~]# ls
#命令查看这个目录，文件已经不存在了，那么现在我们将其恢复出来。
lsof查看删除的文件进程是否还存在。
如没有安装请自行yum install lsof或者apt-get install lsof
1.类似这种情况，我们可以先lsof查看删除的文件 是否还在
[root@docking ~]# lsof | grep deletefile
cat 21796 root 1w REG 253,1 63 138860 /root/deletefile.txt (deleted)
2.恢复cp /proc/pid/fd/1 /指定目录/文件名
进入 进程目录，一般是进入/proc/pid/fd/,针对当前情况：
[root@docking ~]# cd /proc/21796/fd
[root@docking fd]# ll
总用量 0
lrwx------ 1 root root 64 1月 18 22:21 0 -> /dev/pts/0
l-wx------ 1 root root 64 1月 18 22:21 1 -> /root/deletefile.txt (deleted)
lrwx------ 1 root root 64 1月 18 22:21 2 -> /dev/pts/0
恢复操作：
[root@docking fd]# cp 1 ~/deletefile.txt.backup
[root@docking fd]# cat ~/deletefile.txt.backup
This is DeleteFile test.
Add SomeLine into deletefile for fun.
3.恢复完成。
误删除的文件进程已经不存在，借助于工具还原
准备一些文件目录
#准备一份挂载的盘
mkdir backuptest
cd backuptest
mkdir deletetest
mkdir deletetest/innerfolder
echo "Delete a folder test." > deletetest/innerfolder/deletefile.txt
echo "tcpdump:x:172:72::/:/sbin/nologin" > tmppasswd
最后准备的目录结构如下：
taroballs@taroballs-PC:/media/taroballs/taroballs/backuptest$ cd ..
taroballs@taroballs-PC:/media/taroballs/taroballs$ tree backuptest/
backuptest/
├── deletetest
│ └── innerfolder
│ └── deletefile.txt
└── tmppasswd
2 directories, 2 files
现在开始删除该目录rm -rf backuptest/
taroballs@taroballs-PC:/media/taroballs/taroballs$ rm -rf backuptest/
taroballs@taroballs-PC:/media/taroballs/taroballs$ ls -l
总用量 0
这种情况一般是没有守护进行或者后台进程对其持续输入，所以删除就真的删除了。lsof也看不到，故需要采用工具进行恢复。
现在开始进行误删除文件的恢复。
我们采用的工具是extundelete第三方工具。恢复步骤以及注意事项如下：
停止对当前分区做任何操作，防止inode被覆盖。inode被覆盖基本就告别恢复了。
夸张一点讲，比如停止所在分区的服务，卸载目录所在的设备，有必要的情况下都可以断网。
通过dd命令对 当前分区进行备份，防止第三方软件恢复失败导致数据丢失。
适合数据非常重要的情况，这里是例子，所以就没有备份，如备份可以考虑如下方式：dd if=/path/filename of=/dev/vdc1
通过umount命令，对当前设备分区卸载。或者fuser 命令umount /dev/vdb1
如果提示设备busy，可以用fuser命令强制卸载：fuser -m -v -i -k ./
下载第三方工具extundelete安装，搜索误删除的文件进行还原
extundelete工具安装
wget https://nchc.dl.sourceforge.net/project/extundelete/extundelete/0.2.4/extundelete-0.2.4.tar.bz2
解压该文件tar jxvf extundelete-0.2.4.tar.bz2
若报这种错误
[root@docking ~]# tar jxvf extundelete-0.2.4.tar.bz2
tar (child): bzip2：无法 exec: 没有那个文件或目录
tar (child): Error is not recoverable: exiting now
tar: Child returned status 2
tar: Error is not recoverable: exiting now
则使用yum -y install bzip2进行解决
[root@docking ~]# tar jxvf extundelete-0.2.4.tar.bz2
extundelete-0.2.4/
extundelete-0.2.4/acinclude.m4
extundelete-0.2.4/missing
extundelete-0.2.4/autogen.sh
extundelete-0.2.4/aclocal.m4
extundelete-0.2.4/configure
extundelete-0.2.4/LICENSE
extundelete-0.2.4/README
...................................................
cd extundelete-0.2.4
./configure
若这步骤报错
[root@docking extundelete-0.2.4]# ./configure
Configuring extundelete 0.2.4
configure: error: in `/root/extundelete-0.2.4':
configure: error: C++ compiler cannot create executables
See `config.log' for more details
则使用yum -y install gcc-c++解决.
若执行上一步仍然报错，
[root@docking extundelete-0.2.4]# ./configure
Configuring extundelete 0.2.4
configure: error: Can't find ext2fs library
则使用yum -y install e2fsprogs e2fsprogs-devel来解决。
#Ubuntu的解决办法为sudo apt-get install e2fslibs-dev e2fslibs-dev
不出意外的话到这里应该configure能够顺利完成.
[root@docking extundelete-0.2.4]# ./configure
Configuring extundelete 0.2.4
Writing generated files to disk
[root@docking extundelete-0.2.4]#
最后make然后 make install
[root@docking extundelete-0.2.4]# make
make -s all-recursive
Making all in src
extundelete.cc: 在函数‘ext2_ino_t find_inode(ext2_filsys, ext2_filsys, ext2_inode*, std::string, int)’中:
extundelete.cc:1272:29: 警告：在 {} 内将‘search_flags’从‘int’转换为较窄的类型‘ext2_ino_t {aka unsigned int}’ [-Wnarrowing]
buf, match_name2, priv, 0};
^
[root@docking extundelete-0.2.4]# make install
Making install in src
/usr/bin/install -c extundelete '/usr/local/bin'
extundelete安装完成.
扫描误删除的文件：
使用df -lh查看挂载：
taroballs@taroballs-PC:~$ df -lh
文件系统 容量 已用 可用 已用% 挂载点
udev 1.9G 0 1.9G 0% /dev
tmpfs 387M 1.8M 385M 1% /run
/dev/sda2 92G 61G 26G 71% /
tmpfs 1.9G 49M 1.9G 3% /dev/shm
tmpfs 5.0M 4.0K 5.0M 1% /run/lock
tmpfs 1.9G 0 1.9G 0% /sys/fs/cgroup
/dev/sda3 104G 56G 44G 57% /home
tmpfs 387M 40K 387M 1% /run/user/1000
/dev/sda4 70G 20G 47G 30% /media/taroballs/d8423f8c-d687-4c03-a7c8-06a7fb57f96d
/dev/sdb1 6.8G 4.1G 2.8G 60% /media/taroballs/taroballs
/dev/sr0 4.0G 4.0G 0 100% /media/taroballs/2018-01-16-12-36-00-00
taroballs@taroballs-PC:~$ cd /media/taroballs/taroballs/
taroballs@taroballs-PC:/media/taroballs/taroballs$
可以看到，我们的目录/media/taroballs/taroballs
挂载到/dev/sdb1 这个文件系统中.
umount我们的挂载盘
比如：
taroballs@taroballs-PC:~$ df -lh | grep /dev/sdb1
/dev/sdb1 6.8G 4.1G 2.8G 60% /media/taroballs/taroballs
umount这个目录
taroballs@taroballs-PC:~$ umount /media/taroballs/taroballs
taroballs@taroballs-PC:~$ df -lh | grep /dev/sdb1
taroballs@taroballs-PC:~$
#记得删除一定要后umount哦，不然二次写入谁也帮不了你呢。
通过inode节点恢复
taroballs@taroballs-PC:~$ mkdir recovertest
taroballs@taroballs-PC:~$ cd recovertest/
taroballs@taroballs-PC:~/recovertest$
执行恢复extundelete /dev/sdb1 --inode 2
taroballs@taroballs-PC:/media/taroballs/taroballs$ sudo extundelete /dev/sdb1 --inode 2
NOTICE: Extended attributes are not restored.
Loading filesystem metadata ... 8 groups loaded.
Group: 0
Contents of inode 2:
.
.省略N行
File name | Inode number | Deleted status
. 2
.. 2
deletetest 12 Deleted
tmppasswd 14 Deleted
通过扫描发现了我们删除的文件夹，现在执行恢复操作。
(1)恢复单一文件tmppasswd
taroballs@taroballs-PC:~/recovertest$ extundelete /dev/sdb1 --restore-file passwd
NOTICE: Extended attributes are not restored.
Loading filesystem metadata ... 8 groups loaded.
Loading journal descriptors ... 46 descriptors loaded.
Successfully restored file tmppasswd
恢复文件是放到了当前目录RECOVERED_FILES。
查看恢复的文件：
taroballs@taroballs-PC:~/recovertest$ cat tmppasswd
tcpdump:x:172:72::/:/sbin/nologin
(2)恢复目录deletetest
extundelete /dev/sdb1 --restore-directory deletetest
NOTICE: Extended attributes are not restored.
Loading filesystem metadata ... 8 groups loaded.
Loading journal descriptors ... 46 descriptors loaded.
Searching for recoverable inodes in directory deletetest ...
5 recoverable inodes found.
Looking through the directory structure for deleted files ...
(3)恢复所有
taroballs@taroballs-PC:~/recovertest$ extundelete /dev/sdb1 --restore-all
NOTICE: Extended attributes are not restored.
Loading filesystem metadata ... 8 groups loaded.
Loading journal descriptors ... 46 descriptors loaded.
Searching for recoverable inodes in directory / ...
5 recoverable inodes found.
Looking through the directory structure for deleted files ...
0 recoverable inodes still lost.
taroballs@taroballs-PC:~/recovertest$ tree
backuptest/
├── deletetest
│ └── innerfolder
│ └── deletefile.txt
└── tmppasswd
2 directories, 2 files
(4)恢复指定inode
taroballs@taroballs-PC:~/recovertest$ extundelete /dev/sdb1 --restore-inode 14
NOTICE: Extended attributes are not restored.
Loading filesystem metadata ... 8 groups loaded.
Loading journal descriptors ... 46 descriptors loaded.
taroballs@taroballs-PC:~/recovertest$ cat file.14
tcpdump:x:172:72::/:/sbin/nologin
#注意恢复inode的时候，恢复 出来的文件名和之前不一样，需要单独进行改名。
最后附上extundelete的用法:
$ extundelete --help
Usage: extundelete [options] [--] device-file
Options:
--version, -[vV] Print version and exit successfully.
--help, Print this help and exit successfully.
--superblock Print contents of superblock in addition to the rest.
If no action is specified then this option is implied.
--journal Show content of journal.
--after dtime Only process entries deleted on or after 'dtime'.
--before dtime Only process entries deleted before 'dtime'.
Actions:
--inode ino Show info on inode 'ino'.
--block blk Show info on block 'blk'.
--restore-inode ino[,ino,...]
Restore the file(s) with known inode number 'ino'.
The restored files are created in ./RECOVERED_FILES
with their inode number as extension (ie, file.12345).
--restore-file 'path' Will restore file 'path'. 'path' is relative to root
of the partition and does not start with a '/'
The restored file is created in the current
directory as 'RECOVERED_FILES/path'.
--restore-files 'path' Will restore files which are listed in the file 'path'.
Each filename should be in the same format as an option
to --restore-file, and there should be one per line.
--restore-directory 'path'
Will restore directory 'path'. 'path' is relative to the
root directory of the file system. The restored
directory is created in the output directory as 'path'.
--restore-all Attempts to restore everything.
-j journal Reads an external journal from the named file.
-b blocknumber Uses the backup superblock at blocknumber when opening
the file system.
-B blocksize Uses blocksize as the block size when opening the file
system. The number should be the number of bytes.
--log 0 Make the program silent.
--log filename Logs all messages to filename.
--log D1=0,D2=filename Custom control of log messages with comma-separated
Examples below: list of options. Dn must be one of info, warn, or
--log info,error error. Omission of the '=name' results in messages
--log warn=0 with the specified level to be logged to the console.
--log error=filename If the parameter is '=0', logging for the specified
level will be turned off. If the parameter is
'=filename', messages with that level will be written
to filename.
-o directory Save the recovered files to the named directory.
The restored files are created in a directory
named 'RECOVERED_FILES/' by default.