On Wed, Mar 12, 2003 at 07:27:30PM -0500, Joey Hess wrote: > Anthony Towns wrote: > > On Wed, Mar 12, 2003 at 12:25:37PM -0500, Joey Hess wrote: > > > [1] probably with a rcS.d/S46mountrun.sh that does a "mount /var/run". > > > > Uh, there's no point having it later than S35mountall.sh; and it certainly > > can't be later than S40networking. > No, that's the second bind mount. The first (to get a /var/run up before > ifupdown runs) is accomplished by putting the bind mount in /etc/fstab, > where S35mountall.sh will mount it. So the sequence is: > 1. boot, root mounted > 2. S35mountall binds /var/run to /whatever > 3. S39ifupdown writes its state to /whatever via /var/run > 4. S45mountnfs.sh mounts over /var from network, which shadows the bind mount > 5. S46mountrun.sh puts the bind mount back over top of the network > mounted /var Compare with: 1. S10checkroot.sh -- grep -q '[[:space:]]/run[[:space:]]' /etc/fstab && mount -n /run mount -n -o remount,$rootmode / rm -f /run/mtab~ : > /run/mtab mount -f -o remount / mount -f /proc grep -q '[[:space:]]/run[[:space:]]' /etc/fstab && mount -f -o remount /run 2. Everything else just works and it raises the question of why implementing elaborate, non-portable bind mount tricks is preferable over getting the FHS amended. -- Steve Langasek postmodern programmer
Attachment:
pgpROvOgmXsRw.pgp
Description: PGP signature