Bug 667226

Summary:	linux_ro_checker produces false positives in the presence of read-only bind mounts
Product:	Portage Development	Reporter:	Mira Ressel <aranea>
Component:	Core	Assignee:	Portage team <dev-portage>
Status:	CONFIRMED ---
Severity:	normal
Priority:	Normal
Version:	unspecified
Hardware:	All
OS:	Linux
Whiteboard:
Package list:		Runtime testing required:	---
Bug Depends on:
Bug Blocks:	23851

Description Mira Ressel 2018-09-28 13:27:13 UTC

Portage's logic to detect read-only filesystems (lib/portage/util/writeable_check.py) is broken because it incorrectly assumes that the st_dev value returned by stat() matches only one of the filesystems listed in /proc/self/mountinfo.

This assumption is incorrect because bind mounts share the st_dev value of their parent, but have an independent read-only/read-write status. (There may be other scenarios where the assumption fails, for example with btrfs snapshots.)

This isn't a huge issue because this whole check will only be triggered when Portage has already detected that a directory isn't writable -- so in general, when this bug is triggered, there's indeed a reason to abort the merge, Portage just does so with a nonsensical error message.

For example, I ran into this when I tried to emerge openssh -- /var/empty wasn't writable due to a problem on my system, but instead of exiting with the error message "Hey, I can't write /var/empty", Portage instead complained that /home/www/packages (a read-only bind mount from /var/portage/distfiles) was mounted read-only.