Linux XFS filesystem development
 help / color / mirror / Atom feed
From: "Darrick J. Wong" <djwong@kernel.org>
To: Eric Sandeen <sandeen@sandeen.net>
Cc: Eric Sandeen <sandeen@redhat.com>, linux-xfs@vger.kernel.org
Subject: Re: [PATCH V2] mkfs: Emit a clearer message if too-small log is due to AG geometry
Date: Thu, 27 Aug 2026 07:38:44 -0700	[thread overview]
Message-ID: <20260827143844.GC2114998@frogsfrogsfrogs> (raw)
In-Reply-To: <8d2e6c40-0d22-4eda-8f7f-be2678b53d2d@sandeen.net>

On Thu, Aug 27, 2026 at 08:19:52AM -0500, Eric Sandeen wrote:
> On 8/27/26 12:01 AM, Darrick J. Wong wrote:
> > On Wed, Aug 26, 2026 at 10:49:38PM -0500, Eric Sandeen wrote:
> >> There is a constraint in mkfs.xfs that logs must not be smaller than
> >> 64MiB. We can arrive at a too-small log without specifying its size,
> >> though, if we specify so many AGs (or such small AGs) that a single
> >> AG cannot hold the minimum-sized log. This leads to confusing output:
> >>
> >> $ truncate --size=500m testfile
> >> $ mkfs.xfs -dagcount=10 testfile
> >> Log size must be at least 64MiB.
> >> Usage: mkfs.xfs
> >> ...
> >>
> >> because the user did not specify the log size at all! If this error
> >> occurs due to the AG geometry specification, make that clear with a
> >> better error message.
> >>
> >> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> >> ---
> >>  mkfs/xfs_mkfs.c | 19 ++++++++++++++++++-
> >>  1 file changed, 18 insertions(+), 1 deletion(-)
> >>
> >> V2: Ugh as soon as I sent this I realized I forgot to restrict this to internal
> >> logs. An external log may be too small all on its own, unrelated to AG size.
> >>
> >> diff --git a/mkfs/xfs_mkfs.c b/mkfs/xfs_mkfs.c
> >> index a4864c37..4f11c9de 100644
> >> --- a/mkfs/xfs_mkfs.c
> >> +++ b/mkfs/xfs_mkfs.c
> >> @@ -3456,8 +3456,25 @@ validate_supported(
> >>  	 */
> >>  	if (mp->m_sb.sb_logblocks <
> >>  			XFS_MIN_REALISTIC_LOG_BLOCKS(mp->m_sb.sb_blocklog)) {
> >> -		fprintf(stderr,
> >> +		/*
> >> +		 * An internal log must fit within a single allocation group.
> >> +		 * If the user specified the AG geometry but didn't ask for a
> >> +		 * specific log size, the undersized log is due to allocation
> >> +		 * groups that are too small, not a log size the user chose.
> >> +		 */
> >> +		if (cli->loginternal &&
> >> +		    !cli_opt_set(&lopts, L_SIZE) &&
> >> +		    (cli_opt_set(&dopts, D_AGCOUNT) ||
> >> +		     cli_opt_set(&dopts, D_AGSIZE))) {
> >> +			fprintf(stderr,
> >> + _("Allocation group size (%lld MiB) is too small to hold the minimum 64MiB log.\n"
> >> +   "Specify fewer or larger allocation groups, or use a larger data device.\n"),
> > 
> > I like this!  With a couple of comments:
> > 
> > Can the warning message be more targeted to the option?  e.g.
> > 
> > 	if (loginternal && !set(L_SIZE)) {
> > 		if (set(D_AGCOUNT))
> > 			fprintf("Specify fewer AGs or a larger data device...");
> > 		else
> > 			fprintf("Specify larger allocation groups...");
> > 	}
> 
> And if the user specified both size and count? I'm not really convinced
> that we need to go that far, I guess. All the logic in that block is
> kind of eye-bleeding already IMHO.

Eh, ok.  I forgot that maybe the user did both, so they get both
messages. :)

Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>

--D


> > Also, what do you think about not hardcoding 64MiB?
> > 
> > printf("Log size must be at least %uMiB\n",
> > 		XFS_MIN_REALISTIC_LOG_BLOCKS(20));
> 
> That give blocks, not MiB I think, right? 
> > (separate change obviously)
> 
> Yeah I noticed that but was just following djwong's lead from the existing
> printf. ;) 

Yeah but djwong is laaaaazy.

> #define XFS_MIN_REALISTIC_LOG_BLOCKS(blog)      (MEGABLOCKS(64, (blog)))
> 
> Maybe we'd need:
> 
> #define XFS_MIN_REALISTIC_LOG_MB	64
> 
> and then
> 
> #define XFS_MIN_REALISTIC_LOG_BLOCKS(blog)      (MEGABLOCKS(XFS_MIN_REALISTIC_LOG_MB, (blog)))
> 
> and then
> 
> printf("Log size must be at least %uMiB\n", XFS_MIN_REALISTIC_LOG_MB);
> 
> or something?

MEGABLOCKS(xx, 20) gets you megabytes:

#define MEGABLOCKS(count, blog)      ((uint64_t)(count) << (20 - (blog)))

so XFS_MIN_REALISTIC_LOG_BLOCKS(20) gets you the min realistic log size
in megabytes.

#define XFS_MIN_REALISTIC_LOG_MB \
	(XFS_MIN_REALISTIC_LOG_BLOCKS(20))

Your way also works. :)

--D

> -Eric
> 
> > --D
> > 
> >> +				(long long)XFS_FSB_TO_B(mp,
> >> +					mp->m_sb.sb_agblocks) >> 20);
> >> +		} else {
> >> +			fprintf(stderr,
> >>   _("Log size must be at least 64MiB.\n"));
> >> +		}
> >>  		usage();
> >>  	}
> >>  
> >> -- 
> >> 2.55.0
> >>
> >>
> > 
> 

      reply	other threads:[~2026-08-27 14:38 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-08-27  3:42 [PATCH] mkfs: Emit a clearer message if too-small log is due to AG geometry Eric Sandeen
2026-08-27  3:49 ` [PATCH V2] " Eric Sandeen
2026-08-27  5:01   ` Darrick J. Wong
2026-08-27 13:19     ` Eric Sandeen
2026-08-27 14:38       ` Darrick J. Wong [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260827143844.GC2114998@frogsfrogsfrogs \
    --to=djwong@kernel.org \
    --cc=linux-xfs@vger.kernel.org \
    --cc=sandeen@redhat.com \
    --cc=sandeen@sandeen.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox