Workaround lingering dirty bit on Mk.7 USBMS partition

[NiLuJe]

NOTE: This was officially implemented in FW 4.22.15190 . Nothing to see here anymore if you're running an up-to-date firmware .

-----

As people with a Mk.7 device are probably aware, there's been a recurrent glitch which sometimes leaves the FAT filesystem of the partition exported over USBMS with the dirty bit set.

This makes most OS complain loudly about it when trying to mount it (and might cause the most stringent of 'em to refuse to mount it with write access).

I'm still not quite sure what's triggering the issue in the first place (but I figure Nickel is sometimes a bit hasty with how it unmounts the partition), but there are mechanisms in place to attempt to fix minor issues with the FAT partition automatically. These rely on the fsck tool from dosfstools. Unfortunately, like most of the sysroot, the version Kobo ships is *hilariously* old. So much so, that it doesn't actually know how to deal with the dirty bit.
Which means the existing fail-safe mechanisms are powerless to "fix" this issue.

I stumbled onto this entirely by chance (c.f., this post for more details), but the good news is that replacing that old binary with a recent version should take care of the problem (well, not the root cause, of course, but at least the built-in "fix" will now actually work).

So, here it is, in the form of a custom update package (i.e., a KoboRoot tarball), which you'll find inside the attached ZIP. As usual, put it (the tgz, not the zip) in the .kobo folder to install it.

----

fsck.fat -> dosfsck: DOSFSTools, c.f., individual source files for detailed Copyright attribution,
Released under the GNU General Public License version 3+ (https://www.gnu.org/licenses/gpl.html),
(https://github.com/dosfstools/dosfstools)

[NiLuJe]

Pinging @davidfor or anyone on the beta team to elevate this to someone who can actually do something about it for real .

(Might take a while, because GPLv3, but, there you have it).

[Semwize]

I checked (Clara) and yes, the popup with complain disappeared.
Very well.

Edit:

After rebooting the device, the complaint appears once. I don’t remember if it was on Mark 5 and 6.

Checked - does not appear there. But still good, for they popped up too often.

[npr]

Thanks for the workaround, is it safe to use on mk6 and other devices?

[NiLuJe]

It (should) be completely harmless even on earlier devices, not that you'd necessarily have any need of it there.

Won't magically fix truly broke filesystems, though .

[npr]

It's magically fixed the problem on my H2OE2.

I'll try it on my glow and aura later.
Thanks for a great fix.

[frostschutz]

Originally Posted by NiLuJe

I'm still not quite sure what's triggering the issue in the first place (but I figure Nickel is sometimes a bit hasty with how it unmounts the partition)

well it still uses the umount -l (lazy umount) in the /etc/local/Kobo/udev scripts. lazy umount only makes the filesystem invisible. it does not umount anything at all, as long as the filesystem is still in use. thus you export mounted filesystems and the result is corruption.

and the kobo device never reports any error so no one knows that there is a problem or what triggers it or how to fix it

remove the -l option and the script would run into "filesystem is busy" errors instead. e.g. it used to happen when connecting to usb from inside the kepub reader or at least that was the case years ago when I wrote UsbDebug to try and track down the culprits (outdated by now)

remove the -l and add a few loops of retry umount and it'll still work as long as the filesystem would unbusy itself eventually

but if something blocks the filesystem permanent, with this fix you'd lose usb connectivity altogether, and end up locked out if you don't have another channel like telnet/ftp to access the device

maybe if it fails to umount five times, it should umount -l then display a warning message on the screen just to notify that there was a problem ...

[NiLuJe]

@frostschutz: it's also fairly easy to identify & murder anything that might be holding a fd to the fs on a non-lazy umount failure (and/or fail gracefully (i.e., reboot) if it's unrecoverable, e.g., from nickel itself)).

But, yeah, USBMS is tricky.

Amazon went with a giant Jenga tower of indirection to tackle the problem, and it's been the root of a number of sneaky issues over the years, too.

[geek1011]

I wonder if a remount ro before umount would prevent the worst of the corruption without hanging.

[davidfor]

Originally Posted by NiLuJe

Pinging @davidfor or anyone on the beta team to elevate this to someone who can actually do something about it for real .

(Might take a while, because GPLv3, but, there you have it).

Already done when I saw your original post yesterday. I'm under an NDA for that, but, I can probably say I am happy with the response

[NiLuJe]

Originally Posted by davidfor

Already done when I saw your original post yesterday. I'm under an NDA for that, but, I can probably say I am happy with the response

Yay, cheers!

[npr]

Fixed the issue in my kobo glow.
windows nolonger complains when USB is connected.

[JSWolf]

I've never had Windows complain about my original H2O. That's about 5.5 years of use.

[NiLuJe]

@JSWolf: Same, which is why I initially thought it was limited to either Mk. 7 and/or devices with an eMMc instead of an SD card.

[DNSB]

Originally Posted by NiLuJe

@JSWolf: Same, which is why I initially thought it was limited to either Mk. 7 and/or devices with an eMMc instead of an SD card.

Of my Kobo device collection, the Clara HD probably complained the most.