Re: [PATCH] btrfs-prog: scrub: print message when scrubbing a read-only filesystem without r option

[David Sterba <dsterba@suse.cz>]

On Thu, Jul 04, 2024 at 08:56:47AM +0930, Qu Wenruo wrote:
> > --- a/cmds/scrub.c
> > +++ b/cmds/scrub.c
> > @@ -957,7 +957,10 @@ static void *scrub_one_dev(void *ctx)
> >   		warning("setting ioprio failed: %m (ignored)");
> >
> >   	ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
> > -        pr_verbose(LOG_DEFAULT, "scrub ioctl devid:%llu ret:%d errno:%d\n",sp->scrub_args.devid, ret, errno);
> > +	if (ret){
> > +		pr_verbose(LOG_DEFAULT, "scrub on devid %llu failed ret=%d errno=%d (%m)\n",
> > +			sp->scrub_args.devid, ret, errno);
> > +	}
> 
> Doing direct output inside a thread can lead to race of the output.

We'd have to communicate the ioctl start errors back to the main thread,
or rely on internal locking of printf, that it can race is possible but
as long as the string is printed in one go we'll get only lines
reordered.

> And we do not know if the scrub ioctl would return error early or run
> for a long long time, that's the dilemma.
> 
> I'm wondering if it's possible to do a blkid/mountinfo based probe.
> Then it would avoid the possible output race and error out earilier.

We can detet read/write status of the block devices. I'm not sure about
the mount info, what function can we use for that?