Background fsck is broken
Poul-Henning Kamp
phk at phk.freebsd.dk
Wed Dec 15 02:46:22 PST 2004
In message <20041215095901.GK25967 at ip.net.ua>, Ruslan Ermilov writes:
>
>--KjSGHOmKKB2VUiQn
>Content-Type: text/plain; charset=us-ascii
>Content-Disposition: inline
>Content-Transfer-Encoding: quoted-printable
>
>Hi,
>
>Here's another fsck(8) buglet. While booting single-user, / is
>mounted read-only, and "fsck -p /" succeeds as expected. While
>remounting / read-only (e.g., after shutting down from multi-user
>to single-user), it doesn't:
This is working as designed.
The way it works is that when you boot, the root filesystem opens
the device (r=1, w=0, e=0) thereby permitting fsck to open the
device for write.
When the root filesystem is upgraded to RW, the open is opgraded
to (r=1,w=1,e=1) and writing via /dev/mumble is no longer permitted.
Architecturally the way we fsck the root filesystem is highly bogus
and it would be much cleaner if mounted filesystems _always_ were
fsck'ed through a snapshot, but there are a unknown code to be written
to allow that to happen for the case where "unexpected softupdates
inconsistencies" are found.
--
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
More information about the freebsd-current
mailing list