On Mar 2, 2018, at 12:35 PM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote:
>
> On Thu, Mar 01, 2018 at 02:17:17PM -0700, Andreas Dilger wrote:
>> On Mar 1, 2018, at 11:23 AM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote:
>>>
>>> From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
>>>
>>> Implement online fsck for ext* filesystems which live on LVM-managed
>>> logical volumes. The basic strategy mirrors that of e2croncheck --
>>> create a snapshot, fsck the snapshot, report whatever errors appear,
>>> remove snapshot. Unlike e2croncheck, this utility accepts any LVM
>>> device path, knows about snapshots running out of space, and can call
>>> fstrim having validated that the fs metadata is ok.
>>
>> One high-level note - in my lvcheck script, there was an option to disable
>> background checking if the system was running on a battery:
>
> I don't think e2scrub or e2scrub_all should refuse to run if the system
> is on AC power -- if the user runs them from the cli then they shouldn't
> have to override that kind of decision to get what they asked for.
>
> However, checking for AC power certainly makes sense for the background
> scrubber. The systemd service description contains a predicate to
> disable the service if the system is running on a battery. However, the
> cron job does not, so I will add that.
Good point, I didn't really notice that in the systemd service, since that
isn't my cup of tea.
>>> +mark_corrupt() {
>>> + ${DBG} "@root_sbindir@/tune2fs" -C 16000 -T "19000101" "${dev}"
>>
>> This won't actually do anything if the time/mount-based checks are disabled
>> (which has been the default for a long time already, not that I agree with it).
>> You need to add something like "-i 720" to force a check on the filesystem on
>> the next mount.
>
> The trouble with this (all of it, really) is that this clobbers whatever
> setting the administrator might have written into the superblock. Given
> that corruption failures will be logged and produce emailed reports, I
> wonder if it would be easier to dispense with the force-fsck part
> entirely?
How about adding an option to tune2fs to mark the filesystem in error
from userspace, so there is no need to mess with the timestamps or clobber
the time-based fsck settings in the superblock? "-E" would have been good
to match "-e errors_behaviour", but it is already used. Other options
include "-E force_fsck" or "-F", or something else.
Cheers, Andreas
Attachment:
signature.asc
Description: Message signed with OpenPGP
