[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ifupdown writes to /etc... a bug?

On Wed, Mar 12, 2003 at 01:26:58PM +0000, Miquel van Smoorenburg wrote:
> In article <20030312124745.GB27390@azure.humbug.org.au>,
> Anthony Towns  <aj@azure.humbug.org.au> wrote:
> >On Wed, Mar 12, 2003 at 11:45:48AM +0000, Miquel van Smoorenburg wrote:
> >> So, it would be better to mount /run automatically without an
> >> /etc/fstab entry, since it's hard to say what that entry is.
> >> Besides, for a ramdisk, you still need to mkfs a filesystem on
> >> it before mounting it, so it's all special case code anyway.
> >
> >We are in a position where we can cheat, though. Since we don't
> >automatically support read-only root partitions, we can just make /run
> >be on the root fs, and assume that admins who've already demonstrated
> >enough cleverness to cope with /etc/motd, /etc/network/ifstate,
> >/lib/modules/*/modules.* and so forth can cope with adding an fstab
> >entry for /run.

> I agree. In that case, we don't need to worry about 2.2 kernels
> and ramdisks either; just state 'use a 2.4 kernel with tmpfs
> (Virtual memory file system support) enabled'. And the debian kernels
> should enable it by default (it's very small as it is just an
> extension on the always-builtin shm support).

> Only things left to be done:

> - Enable CONFIG_TMPFS in default debian kernels
> - Include /run in sysvinit
> - Mount /run ASAP if it is a seperate filesystem
> - Clean /run as soon as it is read/write.
> - On installation of sysvinit offer to put /run as tmpfs in fstab
>   automatically if the kernel supports it

- amend policy so that people don't file serious bugs against packages
  for using a directory not defined in the FHS :)

Steve Langasek
postmodern programmer

Attachment: pgp3Z6J5lDAea.pgp
Description: PGP signature

Reply to: