Fix tst_find_backing_dev when no initramfs
After having discussed this patch content with Jan Stancek, I realized that my patch log is too minimal.
By this message, I want to substantiate the case.
I will start by saying the problem I want to fix happens in minimal systems:
the kernel and minimal onit scripts.
This is the case of many embedded system powered by linux.
Here is the link of a minimal system I used to test the patch.
https://gitlab.com/acarmina/test-files/-/blob/main/testenv.tar.gz
In such a system, you have:
# cat /proc/cmdline console=ttyS0 root=/dev/sda rw rootwait # cat /proc/self/mountinfo 14 1 8:0 / / rw,relatime - ext2 /dev/root rw 15 14 0:5 / /dev rw,relatime - devtmpfs devtmpfs rw,size=505412k,nr_inodes=126353,mode=755 16 14 0:13 / /proc rw,relatime - proc proc rw 17 15 0:14 / /dev/pts rw,relatime - devpts devpts rw,gid=5,mode=620,ptmxmode=666 18 15 0:15 / /dev/shm rw,relatime - tmpfs tmpfs rw,mode=777 19 14 0:16 / /tmp rw,relatime - tmpfs tmpfs rw 20 14 0:17 / /run rw,nosuid,nodev,relatime - tmpfs tmpfs rw,mode=755 21 14 0:18 / /sys rw,relatime - sysfs sysfs rw # /usr/lib/ltp-testsuite/testcases/bin/ioctl_loop05 tst_test.c:1363: TINFO: Timeout per run is 0h 05m 00s tst_test.c:1115: TINFO: ext2/ext3/ext4 is supported by the test tst_device.c:88: TINFO: Found free device 0 '/dev/loop0' loop0: detected capacity change from 0 to 2048 tst_device.c:566: TWARN: stat(/dev/root) failed: ENOENT (2)
Summary:
passed 0 failed 0 broken 0 skipped 0
warnings 1
#
/dev/root is simply not there, and the test fails with a warning.
In the systems where the /dev/root is there, it is typically a symbolic link to the actual device.
Digging on why this device is not there, despite the mount info reports the / to be mounted over it, I discovered something weird in the early kernel initialization:
https://elixir.bootlin.com/linux/v6.0.3/source/init/do_mounts.c#L566
In this recent kernel release it is evident how the kernel creates the /dev/root device node and after it mounts the new device on the / .
The problem is not widely known since in systems where the /dev/root is used, typically, the init scripts handle it.
Although it may look strange, this behaviour is considered the intended behaviour for the kernel.
An indirect proof is that as early as 2013 a patch aimed to fix this behaviour circulated in the kernel mailing lists
https://www.mail-archive.com/[email protected]/msg378443.html
But it never made to the upstream
Assuming the current the intended kernel behaviour, the thesis I'm trying to push is that the test, meant to test the kernel, should not rely on work done in userspace init scripts.
In the patch I am pushing, my intent is to deal with this /dev/root peculiarity at the LTP level, by basically add a logic that replaces the actual device in case of stat failing on the /dev/root case.
I hope this make my patch context clear.
Cheers
Alessandro
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp