tests: detect loop devices with deleted files as an invariant #14568
Conversation
When a loop device is created with a given backing file, the file itself may be deleted. The kernel keeps the reference to the file alive but it also adjusts the name of the file to have the suffix " (deleted)". We should not have loop devices with this property across tests. They might indicate that a mount sticks around and leaks through a reference somehow, or that the low-level loop-back utilities are buggy. Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>
zyga
force-pushed
the
tweak/deleted-loop-invariant
branch
from
4daf969
to
346c55e
Compare
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #14568 +/- ##
==========================================
+ Coverage 78.85% 78.87% +0.01%
==========================================
Files 1079 1083 +4
Lines 145615 146105 +490
==========================================
+ Hits 114828 115239 +411
- Misses 23601 23671 +70
- Partials 7186 7195 +9
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
| check_deleted_loop_devices() { | ||
| n="$1" # invariant name | ||
|
|
||
| losetup -l | grep -F ' (deleted)' > "$TESTSTMP/tests.invariant.$n" |
There was a problem hiding this comment.
(deleted) is unlikely to shwo up in device name or file name, so we can be less picky about the format I think
| losetup -l | grep -F ' (deleted)' > "$TESTSTMP/tests.invariant.$n" | |
| losetup -l | grep -F '(deleted)' > "$TESTSTMP/tests.invariant.$n" |
There was a problem hiding this comment.
I think this part is relevant. The space is not quoted otherwise. I'd keep this.
There was a problem hiding this comment.
I did an experiment and I ran into another fragile part of the userspace API where escaping is just not done. The files are literally called like that, neither is deleted:
zyga@falka:~/snapd$ losetup -l
NAME SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE DIO LOG-SEC
/dev/loop1 0 0 0 0 /home/zyga/snapd/potato (deleted) 0 512
/dev/loop0 0 0 0 0 /home/zyga/snapd/potato 0 512
zyga@falka:~/snapd$ cat /sys/class/block/loop[01]/loop/backing_file
/home/zyga/snapd/potato
/home/zyga/snapd/potato (deleted)
There was a problem hiding this comment.
zyga@falka:~/snapd$ losetup -l
NAME SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE DIO LOG-SEC
/dev/loop1 0 0 0 0 /home/zyga/snapd/potato (deleted) (deleted) 0 512
/dev/loop0 0 0 0 0 /home/zyga/snapd/potato (deleted) 0 512
The joke is that text formats are easy for humans and machines to parse.
When a loop device is created with a given backing file, the file itself may be deleted. The kernel keeps the reference to the file alive but it also adjusts the name of the file to have the suffix " (deleted)".
We should not have loop devices with this property across tests. They might indicate that a mount sticks around and leaks through a reference somehow, or that the low-level loop-back utilities are buggy.