linux: fix reading zram values

[eworm-de]

This is a follow-up on an earlier commit 'linux: avoid repeated zram ENOENT probes'.
Actually /sys/block/zramN are symlinks, so setting O_NOFOLLOW breaks this. Let's drop the option.

Fixes: 9fe974b

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 1f40aa4f-22e3-4a17-ad10-5d2ed28c093e

📥 Commits

Reviewing files that changed from the base of the PR and between 1ef4226 and bdf4ffc.

📒 Files selected for processing (1)

• linux/LinuxMachine.c

---

📝 Walkthrough

Walkthrough

The zram device scanner’s HAVE_OPENAT path now opens each per-device directory with O_DIRECTORY | O_PATH instead of also specifying O_NOFOLLOW. The fallback path remains unchanged.

Suggested reviewers: intelfx

Poem

A zram door swings wide,
With directory and path allied,
One flag steps away,
The fallback holds its sway,
And scanning proceeds inside.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[eworm-de]

Well, alternatively we could look into /sys/devices/virtual/block/...

Wondering if the paths and/or symlinks changed with recent linux versions?

[Komzpa]

Thanks for catching this and for opening #2045.

/sys/block/zramN has been zram's documented sysfs interface since the
original 2010 sysfs conversion,
and the current kernel ABI still documents
disksize and mm_stat there.

On current kernels, entries under /sys/block are symlinks, so
openat(..., O_DIRECTORY | O_PATH | O_NOFOLLOW) fails with ENOTDIR before
the relative disksize and mm_stat reads can run. Dropping O_NOFOLLOW
fixes that while retaining the dirfd-relative reads.

I would keep /sys/block: switching to /sys/devices/virtual/block would bind
htop to the backing sysfs topology instead of the documented userspace ABI.
The one-line fix looks correct to me. Thank you!

[eworm-de]

Thanks for revealing the details. Sounds like keeping the documented path is the correct approach.