xfs
[Top] [All Lists]

Re: [PATCH 00/19 v2] mkfs cleaning

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 00/19 v2] mkfs cleaning
From: Jan Tulak <jtulak@xxxxxxxxxx>
Date: Wed, 1 Jun 2016 15:19:34 +0200
Cc: xfs-oss <xfs@xxxxxxxxxxx>, Eric Sandeen <sandeen@xxxxxxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20160510061054.GI26977@dastard>
References: <1461231593-31294-1-git-send-email-jtulak@xxxxxxxxxx> <20160510061054.GI26977@dastard>
On Tue, May 10, 2016 at 8:10 AM, Dave Chinner <david@xxxxxxxxxxxxx> wrote:

Looking at xfstests runs, new failures are:

generic/054 -
generic/055 - both fail with:

+*** mkfs failed: -l version=2,su=4096 ***

and the .full file has this specific error:

Illegal value 4096 for -l su option. value is too small

indicating that we should be allowing (2^N * block size) log
stripe units to be set. This will be a limit configuration issue,
most likely needing fixing in mkfs.

âI'm looking on it, but this issue was introduced by Eric's fix for patchÂ"mkfs: table based parsing for converted parameters":

â> â
The kernel enforces a max of XLOG_MAX_RECORD_BSIZE,
â> â
and it should match the limits in L_SUNIT after all ...

And looking onÂfs/xfs/xfs_super.c, the MIN value is enforced too. So maybe it is the test what needs fixing? Otherwise, I would put something likeÂXFS_MIN_SECTORSIZE there.


Â

xfs/096 6s ... - output mismatch (see /home/dave/src/xfstests-dev/results//xfs/xfs/096.out.bad)
  --- tests/xfs/096.out   Â2016-04-05 11:11:36.814382107 +1000
  +++ /home/dave/src/xfstests-dev/results//xfs/xfs/096.out.bad    2016-05-10 12:54:25.956199714 +1000
  @@ -2,18 +2,62 @@

  Â# su too big but must be a multiple of fs block size too
  Â--- mkfs=-l version=2,su=262656 ---
  -log stripe unit (262656) must be a multiple of the block size (4096)
  +Illegal value 262656 for -l su option. value is too large
  +Usage: mkfs.xfs
  +/* blocksize */      [-b log=n|size=num]
  ...
  (Run 'diff -u tests/xfs/096.out /home/dave/src/xfstests-dev/results//xfs/xfs/096.out.bad' to see the entire diff)

This test is explicitly testing an invalid value, checking that it
falls back to a sane default. This patchset has changed the
behaviour, so the xfstest will need some rework to handle both old
and new mkfs behaviour.

Can you send patches to fix all these up?

âSure.

Jan
â
Â


--
<Prev in Thread] Current Thread [Next in Thread>