DragonFly BSD Digest

César Duque posted his own take on how the DragonFly page should look:
http://www.notfound.org/df (it’s a large jpeg.)

Hiten Pandya has apparently been working on a new version of the main site. Those changes are not (yet?) in CVS.

The conversation about how to handle third-party utilities in the base system is continuing. Most everyone is coming around to the idea of variant symlinks – file links that use a variable to determine destination, with that variable value determined by the rc system.

Matt Dillon added to his earlier plan with a description of the framework for this variable data, using a system MTA as an example:
Continue reading “Even more replacement”

David P. Reese, Jr. has completed the 4th stage of his work separating syscalls. This patch splits getsockopt() and setsockopt(). It’s not yet committed as of this writing, but there are no objections yet. A side benefit of this work is that calls like stackgap_init() and stackgap_alloc() can be removed, as they copy data into an area of kernel memory that’s easily overwritten.

The aforementioned K&R -> ANSI cleanups have been committed – that’s about 160 files in bin and sbin that have been cleaned up.

David Rhodus kindly pointed out to me that I missed his commit of ACPI code from Intel, back on the 23rd.

It’s not installed by default. To use it, you’ll need to add ‘device acpica’ to your kernel config file.

Matt Dillon described a mail ‘API’, where a particular MTA port would be installed into a /usr/local/ directory, and then an /etc/rc.d/mail scipt would correlate the right variables to control behavior for that MTA.

Full post added below, in case someone decides to use it as a template for work… (hint)

Also: the slab allocator is now in by default.
Continue reading “More on mail replacement”

Timothy Cava brought up the topic of removing sendmail from the base system, and what would be needed to do so without breaking many, many things. Mike Porter pointed out that an rc.mail file should be able to control it. Jeremy Messenger added that this was talked about in an earlier thread revolving around Perl and other base components from FreeBSD. David Rhodus closed this discussion, somewhat, by saying a final DragonFly BSD packaging mechanism is necessary first.

Matt Dillon posted a patch for the next stage of his namecache work. I’m pasting the intro here, where he talks about what this patch does, and some of what the last stage will do.
Continue reading “namecache, stage 2 of 3”

David Leimbach asked for links that describe what a slab allocator is, and Hiten Pandya helpfully noted Bonwick94, Bonwick2001, and linux/mm/slab.c in the Linux kernel.

Joe “Floid” Kanowitz brought up the idea of RAM being used as disk cache, to which a number of replies were made. I mentioned softupdate snapshots as a possible mechanism, which Sander Vesik corrected to say that it would be more like a “write through” union mount of a MFS and real filesystem.

Kip Macy pointed out that ‘tmpfs‘, on Solaris and Linux does something similar. BSD also has mfs, which requires that you allocate memory ahead of time. (Most commonly used to speed up buildwords in /usr/obj/ .)

The concept of journaling file systems that save the state of a file at given chronological points was brought up, and several people noted ‘Elephant‘. (Link from Hiten Pandya.)

Matt Dillon also wrote up several paragraphs on using RAM for storage, which are sufficiently technical that I’ll paste them in the extended link for this entry, rather then sum up.
Continue reading “Ways to use up RAM”

David P. Reese, Jr. noted that he will be at the Neal Stephenson talk in the Menlo Park Kepler’s Books on the 26th, and if there’s any other developers around, he’d like to meet up afterwards. Also, Matt Dillon noted he should throw a party when the namecache work is done.

Matt Dillon noted that if/when replacing K&R declarations with ANSI-compatible ones, watch for argument order:

(old style)
int
fubar(a, b, c)
int c;
int b;
int a;
{
}


(new style)
int
fubar(int a, int b, int c)


He also added: indent when the argument list wraps, and don’t try it in /usr/src/sys/ for the next few weeks, as he’s working there!

Also: ibotty submitted a patch that would let users mount devices to which they do not have write access, though this may be delayed until/overridden by VFS work.

Pedro Giffuni noted that there is an effort in NetBSD to protect the stack that may be useful for DragonFly BSD; several others noted that the x86 platform, where DragonFly is currently focused, does not explicitly support this behavior, so it is more hassle than use at this time.

You may say to yourself, “Gosh, I hope the recent SSH and Sendmail holes are fixed in DragonFly BSD, too!” And you’d be right.

Miguel Mendez brought up the idea of using NetBSD’s Veriexec for more secure prebinding (which will also be committed, but not used by default); David Rhodus reports he has Veriexec working and will commit it soon.

David P. Reese, Jr. has separated out the sendto/sendmesg/recvfrom/recvmsg calls and created kernel versions. 4.3BSD syscalls have been separated out, too, to src/sys/emulation/43bsd. (Patch pending)

Matt Dillon pointed out that GEOM in FreeBSD may be something DragonFly can present a compatible API for, but it requires the messaging work for VFS and DEV first. Incidentally, Matt will be working on namecache, VOP_LOOKUP, Simon ‘corecode’ Schubert’s prebinding, and Jeffrey Hsu’s changes to ifq.

If, like me, you are clueless on what prebinding and ifq are, here an explanation of prebinding for OS X (should apply here) and ifq… appears to be the queue in which chunks of network information are held/passed? I haven’t found a good explanation. (Comments welcome, if you know)

asmodai pointed out xenofarm during a discussion of the tinderbox builds of DragonFly.