> From: Richard Daemon [mailto:[EMAIL PROTECTED] > > On Jan 19, 2008 8:31 PM, Schvberle Daniel > <[EMAIL PROTECTED]> wrote: > > > Hi all! > > I've just upgraded my firewall from OpenBSD 4.0 to > 4.2-stable and ran > into a small problem regarding mount_mfs. I solved it, > but in case > anybody else runs into it, here's something for the archives. > > I run the box from a 512MB CF and, originally, with very limited > memory. The /var, tmp and /dev are mount_mfs and during > the upgrade I > had trobule with mounting /dev. > > I used to mount /dev with the following line: > > swap /dev mfs rw,-P=/proto/dev,-s=700,-i=256 0 0 > > It seems that sometime after 4.1 was released (probably > during ffs2 > development) mount_mfs was changed in such a way that > it doesn't allow > very high density for inodes. This resulted in > mount_mfs failing on > replicating the /dev and me getting a readonly /dev, > which resulted > in a box that I couldn't login into remotly (with ssh). > Luckily you I > could still issue commands with winscp or login > locally. After couple > of tests I concluded that mount_mfs simply ignores > density settings > lower than 1024, so I changed the /dev to settings to > the following > line: > > swap /dev mfs rw,-P=/proto/dev,-s=4000,-i=1024 0 0 > > Now everything is ok, I'm happy and sice CF is in a new > box with lots > of memory I'm not trying to squeeze every byte out of it. > > Maybe this maximal density could be documented somehow? > I glanced at > the mkfs.c and saw that, in theory, it should warn the user when > reducing the density but I never got a warning during my tests. > > dmesg in case anybody needs it:
<snip dmesg> > > Wow, very weird that you post this. I just noticed the exact > same thing yesterday too. Upgraded from 4.0-stable to > 4.2-stable on a WRAP (pcengines.ch) box with my 512M CF and > /dev entries failing as well. My previous inode settings used to be: > > swap /dev mfs rw,-P=/.devtmp,-s=1200,-i=128 0 0 but that > crapped out in 4.2. > > I changed it to -s=3072, -i=128 just to get it fully working > properly and I haven't looked into it further yet, but > wondering if I'm better off maybe trying higher inode (like > yours) but lower MFS size such as -s=1024 because I'm limited > in memory (128M total). Other than that, is an MFS /dev size > bigger than 1M even needed? I'd really like to reduce as much > as possible. > > Thanks for the post! > > I'm new to this mailing list and so far, it's great! No, I don't think you'd ever need a /dev this big, but in order to get the needed number of inodes you have to push the size up. Your line is ok, but maybe you should put i=1024 instead of i=128, so you know what the real values are - that's what it's using anyway With 128MB you really shouldn'y worry. I was concerned because I had only 32MB or 48MB. mount_mfs doesn't really use the memory untill it's needed, so you could make, say 100GB mfs on a box with 128MB of RAM and it would work as long as you've got memory to hold the files. Regardnig /dev, you really don't need much as it's a small filesystem, but sometimes you can get real files in there. This is what happend once to my lil' box (I had a _real_ /dev/null) and it crapped out because it ran out of memory. After that I reduced the /dev as much as I could, I didn't want another local DoS to happen. I have 512MB now and couldn't care less if /dev is 0.1 or 1 MB, and with 128MB you shouldn't either, especially since it gets allocated only if really needed by the files.
