[netatalk-admins] pre-a19 available and database thoughts


Subject: [netatalk-admins] pre-a19 available and database thoughts
From: a sun (asun@saul5.u.washington.edu)
Date: Mon May 11 1998 - 12:35:48 EDT


i've made available a pre-a19 for people to try out. i still want to
stick in appledouble v2 for a19, but i wanted to see how what i have
works for people. once that is done, here's what's on my current top
list for future patches (as time permits):
        1) per-volume permanent did allocation
        2) work on a per-directory hashed file database. this should
           get us case insensitivity, filename length tags, more parts
           of AFP 2.1, and faster enumerates.
        3) look into better interopability with samba. do the people
           @whistle.com already have some patches for this? in any
           case, i'll probably try to make the locking code co-exist
           better with samba.

so, how do we handle dids using appledouble v2?
        1) did allocation handled by a bitmap in the top level being
           served (with nextdid at root level and actual did storage in
           .AppleDouble headers).
        2) the lower dids are handled separately and not actually handed
           out permanently. root vols always get 2, similar to how
           it's handled now, even if they might have another number in
           another circumstance.
        3) otherwise, everybody keeps their did's (stored in
           .AppleDouble header).
        4) if a new root directory is used, the afpd will step through
           and fill in its bitmap for subdirectories. in case of
           conflicts, it will just go through and force one of the
           dids to get re-allocated. as a means of reducing the
           chances of future conflicts in the case of a parent
           directory/child subdirectory relationship, the parent afpd
           can reset the .nextdid of a subdirectory mount point to
           something currently out of range.

-----
anyway, here's what's different in the patch:
        a) there's a workaround for people nfs-mounting directories
           who don't have working lockds. add -DHAVE_BROKEN_LOCKS to
           your DEFS line if you're one of those people. it resorts to
           flock-style locks in that case. as far as i can tell, lots
           of oses don't have proper lockds.

           NOTE: this means that flocks will now be used to implement
                 transient locks on a file-wide basis. this is
                 different from the 1.4b2 use of flocks. as a
                 consequence, having serving the same file from
                 different machines may result in lossage.

        b) -nouservol option added to afpd.conf so that you can
           prevent ~/.AppleVolumes from ever being read.
           
        c) amount transferred in kilobytes recorded in syslog on
           exit.

        d) i stuck in the beta solaris streams module. if anybody
           knows how to stick in socket options into it, i would
           appreciate it. thanks to wes and folks for allowing me to
           do this!

        e) i fiddled with some of the AFPLIBS a little in the hopes
           that i can just get rid of the inet_aton glue in
           libatalk/compat. let me know if i did enought.

        f) -DNEED_QUOTACTL_WRAPPER is again the default for linux
           machines.

it's retrievable at
<ftp://ftp.u.washington.edu:/public/asun/pre-a19.tar.gz>.

-a



This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:32:40 EST