ful fsck whenever journal replay is needed?

eccerr0r

Is there a way to make bootup fsck to do a full check whenever a filesystem journal replay is needed?

There's someone I know that thinks that crashes are normal (because that's how windows works) and it's ok to hard power off machines because there seems to be no consequence. Yes, the journal does help, but I want to make it more painful (i.e. fsck) to not actually think about fixing the problem...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Hu · Administrator Joined: 06 Mar 2007 Posts: 23066

If you really want to make it painful, just disable the journal.

Mount everything unjournaled. It will be slightly faster in the good case, and have a much higher risk of serious problems in the bad case.

eccerr0r · Posted: Fri Sep 01, 2023 11:09 pm Post subject:

That unfortunately would cause more pain if the machine wouldn't start up again the next boot. I'm remote administering but sent components to them to self repair but doesn't know their way around root, etc. - and also the main reason why there's no effort to repair the cause of the crashes - rather it's better that something runs and crashes than doesn't run at all...

sigh.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

NeddySeagoon · Posted: Sat Sep 02, 2023 2:22 pm Post subject:

eccerr0r,

Some checkers accept the -f option.
Some run that way if you

eccerr0r · Posted: Sun Sep 03, 2023 3:30 am Post subject:

somehow need to touch the file if and only if the journal was replayed?
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

NeddySeagoon · Posted: Sun Sep 03, 2023 9:27 am Post subject:

eccerr0r,

I don't see that happening.
mount will do the journal replay if the filesystem is flagged 'dirty' no fsck required.

You would need to check the dirty bit before mount got to it, then run a full filesystem check before any mount was attempted.
You can do that in the initrd if you really wan to. It has to be before root is mounted.

That's the same as just turning off the journal though - which on extX is far easier.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

Goverp · Advocate Joined: 07 Mar 2007 Posts: 2202

szatox · Advocate Joined: 27 Aug 2013 Posts: 3491

eccerr0r · Posted: Sun Sep 03, 2023 12:11 pm Post subject:

How about this as a reason for not wanting hard power offs: I'm logged in remotely (to do emerge -u kind of maintenance that is never done) and whether or not the console crashed or not, my remote session is fine. I don't mind the shutdowns but it's annoying to try to tell whether it was a crash or network failure which is possible. I'd rather be alerted of a clean shutdown which is perfectly fine.
I believe the crashes are caused by failing hardware which I have sent, but so far refuses to install that would make the crashes go away.
Not sure why I care so much about keeping the machine up to date and healthy more than the user but that's the way it goes...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

szatox · Advocate Joined: 27 Aug 2013 Posts: 3491

I see. Well, a multiuser machine is a different case.
It's a bit weird though, I think my PC does run fsck after every power outage and after 3 months since the last check (adjustable with tune2fs).

eccerr0r · Posted: Sun Sep 03, 2023 8:48 pm Post subject:

Interesting, I usually do not need to fsck unless the max mount count is exceeded whether it's resetting from a hard crash or power outage. Are you using a journalled file system? It should (usually) just replay and you're on your way for ext3, ext4fs...

But thanks that seems like something to think about, perhaps keep an actual count in-band or somewhere so that if it really has been 50 mounts it will fsck anyway.

Anyway I suppose I like having root access on remote machines that I don't need to pay for internet, electricity, or "rack space" ... it's too bad it's not up 24/7 but I tend to still get honored uptime on occasion, just need to make sure it's not -too- often. Alas much of that uptime is dedicated to emerge -uD firefox ...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

szatox · Advocate Joined: 27 Aug 2013 Posts: 3491

Yeah, ext3 has always had journal enabled by default. You got me thinking though, something ain't right :lol:

You can tune2fs routine scans based on time since the last fsck. E.g. debian used to set it to 1 month. Back in the days of lenny, at least. I think they changed it to 3 months afterwards.
It's independent of mount count, tripping either threshold is enough to trigger a check.

man tune2fs does not recommend those scheduled checks though. I guess it depends on how paranoid you are.

eccerr0r · Posted: Mon Sep 04, 2023 7:03 pm Post subject:

Yeah I don't think it's of much value to fsck a disk, put it away for a year, and next time you use it, it needs to fsck again -- especially since fsck doesn't read every single block anyway. Only if it reads every block or one needed to write to the disk should it have some value for a check, and mounting a disk r/w counts as a write.

Then there's the argument that replaying a journal is a write too...it sure is...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Hu · Administrator Joined: 06 Mar 2007 Posts: 23066

A filesystem check can be useful for finding filesystems which have started to go corrupt due to software bugs, as well as damage caused by hardware returning wrong data. I suspect the idea behind time-based checks was that a server which is powered up constantly would only ever mount the filesystem when forced to reboot by kernel update or power outage. Such a system might accumulate many hours of usage and writes to the filesystem (and thus many opportunities for rare bugs to manifest at least once), but take a year to accumulate even 12 mounts (supposing that it only takes major kernel updates, rather than every time "all users must upgrade"). Time-based checks ensure that it is still checked periodically, possibly even every reboot if uptime is normally measured in months.