Date: Fri, 26 Jan 2007 04:20:23 GMT From: Bruce Evans <bde@zeta.org.au> To: freebsd-bugs@FreeBSD.org Subject: Re: bin/108320: dumpfs(8) lists wrong file system size [PATCH?] Message-ID: <200701260420.l0Q4KN7p005187@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR bin/108320; it has been noted by GNATS.
From: Bruce Evans <bde@zeta.org.au>
To: Toby Burress <kurin@delete.org>
Cc: freebsd-gnats-submit@freebsd.org, jmallett@freebsd.org
Subject: Re: bin/108320: dumpfs(8) lists wrong file system size [PATCH?]
Date: Fri, 26 Jan 2007 15:15:32 +1100 (EST)
On Thu, 25 Jan 2007, Toby Burress wrote:
> On Thu, Jan 25, 2007 at 06:56:23PM +1100, Bruce Evans wrote:
>> On Wed, 24 Jan 2007, Toby Burress wrote:
>>
>>>> Description:
>>> When using dumpfs(8) to check the file system size, and in particular
>>> dumpfs -m to find a command that creates the file system, the size
>>> listed is smaller than the actual size of the file system.
>>
>> No, dumpfs is correct. It just reports the size being used ffs. The
>> size is in blocks (frags). This size is reported by dumpfs as `fsize`.
>
> If dumpfs is correct, then why are the file systems created with
> the command `dumpfs -m` smaller than the original file systems?
Sorry, I didn't notice that you wanted -m to work (I didn't know that
-m existed). -m just never actually worked, except accidentally when
the frag size equals the sector size. It prints "-s <size in frags>"
but newfs's -s arg takes a size in sectors. These sizes normally
differ by a factor of 4, since the frag size defaults to 2048 and most
disks have a sector size of 512. Another bug in -m is that it conflicts
with NetBSD's -m.
It's bogus for newfs to take a size in sectors. The sector size is
not recorded in struct fs or used by ffs. It just needs to be a divisor
of the frag size (and maybe a power of 2). newfs normally determines
it using a disk ioctl and only uses it to scale the sizes. This default
can can be overridden by specifying the size using newfs -S or -T.
Another bug: -T apparently clobbers any setting by -S, even if -S is
after -T on the command line.
Possible fixes:
- use a disk ioctl in dumpfs, so as to reverse the conversion of units
that newfs will do. I don't like this. It doesn't work if the file
system is a copy of a disk in a regular file.
- use a fake sector size: print "-S <frag size>" in dumpfs. This should
give identical results since ffs doesn't actually use the sector size,
except newfs randomizes some things in the new file system. The
undocumented -R flag can be used to avoid the randomness for testing.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200701260420.l0Q4KN7p005187>