Sascha Wildner’s preformatted DragonFly documentation has been cleaned up, and had title pages and content lists added.
Month: May 2004
Matt Dillon mentioned that he’s considering using apt-get combined with VFS to create a sort of ‘per-user’ port visibility. Quoting his examples:
“So the ‘apache user’ might only have 10 ports installed (and only access to those 10) while a GUI user might have 50 ports installed (and only access to those 50). Overlaps would be allowed… the environments would be considered independant.
Such environments would also enforce basic isolation/separation/security features such as “/usr is read-only except for /usr/local”, and would be stackable, but instead of trying to use them on a per-package basis we would use them on a per-user basis (or something like that).
This gives us the best of both worlds in my view.
This would mean a new ‘port’ system would have to wait until VFS is finished, though using apt-get without the per-user separation would be achieveable almost immediately.
This log page passed 20,000 views for May. It’ll be interesting to see the bump after the 1.0 release happens…
David Rhodus mentioned in his GoBSD journal that DragonFly is a better choice right now than FreeBSD 4.10, highlighting a FreeBSD-4 unfixed bug in his journal. He also let slip that he’s working on a commercial operating system based on DragonFly, with the first release candidate coming very soon.
Rahul Siddharthan pointed at this post as evidence of softupdates not working as quickly as it used to, for FreeBSD and by extension DragonFly. Matt Dillon replied that:
“FreeBSD-5 has a lot overhead, especially when it comes to the buffer manipulation that softupdates does. DFly should be nearly the same as 4.x in regards to FS performance.
There isn’t much I can do about softupdates but the (slowly progressing) namecache work will eventually allow us to release the exclusive lock on the directory vnode during directory searches and this will bring up our lots-of-little-file benchmark numbers considerably.
Another issue that slows down filesystem operations is the busy-page lockout that occurs when the system is writing data to disk and some other operation wants to modify the page undergoing I/O. That is ”on the table’ as well.”
There’s a whole slew of screenshots of the curses interface to the Nympha installer at http://www.livebsd.com/dfly/screenshots/.
The kernel mailing list is getting hit with a Windows mail virus in a big way; be careful when viewing it through your mail client/news interface if you use Windows.
Update: fixed now.
Matt Dillon has added bind-9.2.4rc4 to
contrib. This replaces bind8 that was being used for DragonFly (and I assume FreeBSD-4) by default. Things may be somwhat unstable for a few days; if that is trouble for you, don’t update your system this week.
Sascha Wildner has placed converted DragonFly man pages at http://www.yoyodyne.de/dfly/doc/
‘GeekGod’ has an alpha version of the DragonFly Live CD with the Nympha installer. Try it out, if you feel lucky – it is alpha, so read the caveats.
Update: now works on VMWare.
Emiel Kollof’s changes to the NVIDIA binary video driver have been committed; it will no longer break with the recent /dev changes.
Hiten Pandya noted that, if you are creating a port override in dfports, setting CCVER=gcc3 and testing your override using the different compiler is a good idea. He’s found a few ports where poeple didn’t do that and the build was broken.
Sascha Wildner described in a post to dragonfly.bugs,
/etc/rc.conf will enable your system to save crash dumps in
/var/crash, saving the effort of retyping them in a post. (“makeoptions DEBUG=-g” in your kernel config is also needed.)
Emiel Kollof has fixed up the NVIDIA driver – This diff makes it possible to update your kernel and still use the driver; the change to the dfports override is forthcoming.
Chris Pressey committed the last stage of Andre Nathan’s gigantic K&R -> ANSI cleanup of source. 21 stages, in all.
The latest dev numbering changes from Matt Dillon will break the NVIDIA binary driver, if you are using it. Emiel Kollof is working on a new version, and until that’s set, avoid updating if you want to keep your NVIDIA driver working.
cvs is now at version 1.12.8, updated by Matt Dillon.
The network connection to dragonflybsd.org appears to be down – nntp is not accessible, and the website isn’t either.
Update: It’s back online.
I’ve committed a partial section on installation in docs, which – if we’re lucky – will be completely superseded by a DragonFly installer program.
Joerg Sonnenberger has updated the
bktr(4) driver to match what’s current in FreeBSD, including a new
msp driver (from Linux), and support for the Terratec TValue, which was submitted by Patrick Mauritz.
cc3 has been switched back to stabs output, temporarily until DWARF2 support comes into
gdb5 or DragonFly gets
Brock Johnson submitted and Eirik Nygaard committed version 3.8.3 of
tcpdumpand 0.8.3 of
Chris Pressey wrote down how to fake a hard drive as a ‘live CD’.
A new ‘known stable’ ISO is on the Download page. This one should hopefully be free of the wierd filesystem corruption bug that’s been hitting people.
Matt Dillon’s tracked down a nasty filesystem corruption bug in the
lockf code; everyone should update, rebuild, reboot, and do a
fsck to make sure your disk is intact.
‘Gabor MICSKO’ has written up the current DragonFly installation process in Hungarian, with screenshots, even.
Jeroen Ruigrok tried using ‘CPUTYPE=p4’ in
make.conf as an option to gcc. The resulting kernel crashed; and several people pointed out that the CPUTYPE optimization does not work – especially with gcc 2.x. Matt Dillon went on further to explain how little it helps:
“Trying to use cpu-specific optimizations in 2.95.x is roughly equivalent to a blind man driving an 18 wheeler down the highway. In other words: “don’t do it, it doesn’t work”. GCC2 is known to produce bad code not only with CPUTYPE, but also with higher optimization levels like -O2. Blame the GCC folks for this.
GCC3 does a better job but even so I would not use CPUTYPE without being prepared to turn it off when something breaks, and at best I might use -O2 or -Os (under GCC3 only, and even then we don’t really officially support it, since differentiating between compiler code generation bugs and DFly bugs is extremely difficult).
Besides, you won’t notice any improvement in kernel performance.”
‘walt’ found and Chris Pressey fixed a bug in the port system that could cause some ports to fail building with a error similar to ‘
cd: can't cd to &&‘. Extraneous text was being inserted by the ‘build-depends-list’ and ‘run-depends-list’ targets.
A number of people have noticed problems with sendfile(2); especially with Apache. Jeffrey Hsu’s fixed it already.
Matt Dillon posted more ideas on how the DragonFly releases were to go. Specifically, different versions would be coded by slip tags instead of the branches that go on with FreeBSD:
“releaseX_Y: Specific official release, build and regression tested, including security fixes. A sub-sub version (Z) would be embedded in the uname to differentiate between updates to a release, but not be incorporated into the tag name.
release: Current official release.
developer: Developers who want to be cutting edge but still be ‘reasonably’ stable, with occassional glitches.
HEAD: Absolute latest work, might have build and other issues.
Not being branches, slip tags are limited in scope. It might not be possible to incorporate a security fix into a particular release tag, for example, all we would be able to do is tell people to upgrade to a later release. When we are able to stabilize the system interfaces (about a year and a half from now) this won’t be a big deal.”
Matt Dillon’s added fsstress in
src/test/stress/fsstress. This tool has been found useful before.
Commit message quoted here:
“Add the filesystem/NFS stress tester program, adapted for BSD by Jordan H ubbard of Apple, originally written by Silicon Graphics, Inc, and modified by many people, including the SUSE Linux project.”
Matt Dillon has updated the ‘known stable’ ISO on the download page. He mentioned he plans to create a ‘stable’ tag in cvs (not a branch, but a slip tag) so that people who don’t care for experimentation can track that target.
Matt Dillon has found further problems with the mmx/xmm/FP code, and committed fixes for them, describer further in the changelog for various file. (The one linked is one of several changed.)
If you built a kernel between now and April 29th, it’d be a good idea to do a full update and buildworld/buildkernel.
I’ve added my documentation work to the framwork that Hiten Pandya put together. I’ve committed sections on backups, X Windows, ports, and upgrading. You can see the current HTML output at http://www.forknibbler.com/guide/.
R. E. Ceiver posted his DragonFly install experience. It follows the same form as the regular from-CD install, but a little more verbosely.
Matt Dillon has found the solution to the wierd MMX problem; it’s now safe to try
kern.mmxopt=1 again. The problem turned out to be (quoting Matt) “related to the PIPE code, but only when the sysctl’s were set up for legacy algorithms. By default the PIPE code uses SFBUFs which are not effected[sic] by the bug.”
Matt Dillon’s placed a new “known stable” ISO at ftp://ftp.dragonflybsd.org/iso-images/dfly-20040502.iso.gz.
Devon H. O’Dell has mirrored this image in San Jose, California, US, and Amsterdam, Holland. Only one month until June, and 1.0! (Though it probably won’t released this early in June.)