Repository navigation
dtoverlay security audit - #234
Merged
Merged
Conversation
The state directory, /tmp/.dtoverlays, has a fixed name in a world-writable directory. Any local user can create it first (running "dtoverlay -l" as an unprivileged user was enough), and root then replays the .dtbo files it contains into the kernel and writes files through any symlinks planted there. Move it to /run/dtoverlays, which only root can write to and which is cleared at boot (as the configfs state is). Only create it when running as root, and refuse to use it unless it is a real directory owned and only writable by root. If it doesn't exist, non-root users listing overlays just see that none are loaded. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nothing stops two instances of dtoverlay/dtparam from running at once, and the results can be bad: - Both read the same overlay count and give their overlays the same sequence number, breaking the numbering that removal relies on. - A removal deletes and replays the later overlays, and an add that runs during it corrupts the sequence. - Concurrent dtparam invocations share base.dtb, and can take a snapshot of /proc/device-tree while another overlay is being applied. - Listing can read a .dtbo that is still being written. Take an flock() on /run/dtoverlays/lock before reading the state and hold it until exit, exclusively for adds and removals, shared for listing. The kernel releases it however the process exits, including through fatal_error(). Rather than block forever, give up after 30s. The dtoverlay-pre/post hooks run with the lock held, so a hook that tried to add or remove an overlay would deadlock (or recurse). Mark the hooks' environment and make that an error; listing from a hook is still allowed, and skips the lock because the parent already holds it exclusively. Also check for root before touching the state when adding or removing, rather than after reading it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The .dtbo files in the state directory were written in place, so killing dtoverlay part-way through a write left a truncated file that a later removal would try to replay. Write to a temporary file and rename it into place. The temporary name starts with '.', so a leftover one isn't mistaken for an overlay. Also fail the add if the file can't be written, instead of applying whatever is there. dtoverlay_list passed the result of dtoverlay_load_dtb to dtoverlay_dtb_trailer without checking it for NULL, so an unreadable file crashed it; report the overlay as unreadable instead. In the library, dtoverlay_save_dtb leaked the FILE when a write failed, and ignored the result of fclose, which is where errors in buffered data are reported. dtoverlay_save_dtb is also used to write to configfs, so it still writes directly to the named file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two instances starting at once could both find /config missing, and the loser's mkdir failed with EEXIST, which was fatal; accept EEXIST. They could also both find configfs unmounted and mount it twice. The mount now happens under the exclusive lock; listing only takes a shared lock, so it upgrades the lock before mounting. When applying an overlay failed, dtoverlay_add moved it to error.dtb and freed error_file without clearing it. If a later step then succeeded, as can happen while replaying overlays during a removal, main() passed the freed string to unlink. Clear error_file, which also keeps the error.dtb that has just been written. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
If configfs wasn't at /sys/kernel/config, dtoverlay created /config and then tried to mount configfs on /config/device-tree, which can't exist until configfs is mounted at /config - so the mount always failed. It also ran mount via the shell and then reported errno, which said nothing about why mount had failed. Mount configfs on /run/dtoverlays/configfs instead, which is on a tmpfs and in a directory already checked to belong to root, using mount(2). Then check that it has a device-tree folder, unmounting it and failing if it doesn't. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Removing an overlay deletes the configfs directories of it and every later overlay, then renumbers and reapplies the survivors. Being interrupted part-way (Ctrl-C, kill, or the terminal hanging up) left configfs and the state directory out of step, with no way to recover short of a reboot. Block SIGHUP, SIGINT, SIGQUIT, SIGTERM and SIGPIPE while an overlay is being added or removed, and unblock them afterwards, at which point any that arrived are acted on (so the dtoverlay-post hook doesn't run). A signal handler couldn't clean up safely, and the only consistent state to clean up to is the finished one anyway. SIGKILL still can't be stopped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When removing an overlay, the later overlays are renumbered and reapplied. If one of them failed to apply, its renumbered .dtbo was left in the state directory without the count being advanced, so the next overlay was given the same number, or the next run believed the failed overlay was loaded. The same happened in dtoverlay_add when a failure had already used up error.dtb. Discard the file of an overlay that fails to apply: move the first one to error.dtb as before, and delete any others. It then drops out of the sequence and the numbering stays contiguous. Name the overlay in the error message (it always said "dtparam"), and exit with an error if any overlay was lost. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signals from a terminal or from timeout are sent to the whole process group. While dtoverlay was deferring them during a removal, the dtc that regenerates a dtparam still received them, and testing on a Pi 5 showed that a SIGINT or SIGTERM at the wrong moment killed it. The state stayed consistent, but the dtparam was dropped with "Failed to read active DTB". Run dtc with fork/exec rather than system(), and while the parent is deferring signals make the child ignore them, which survives exec. This also stops the filename being passed through a shell. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fix a number of issues found by a security audit of dtoverlay/dtparam.