GIF>I8DD@E> Kernel News Q8:BËJB

KiXZb`e^B\ie\c ;lZb8e[:fm\i ?Xi[nXi\:fdgXk`Y`c`kp :feki`Ylk`fej Matt Mackall was a bit shaken to discover Adam Osuchowski was poking around in Wang Chen created a web page for kernel a kernel error message that began “Trea- the deep dark places of the kernel and contribution stats. The page breaks down son uncloaked!” The message went on to came upon some hard-coded assembly information by person and employer, as say that the TCP code had detected a bro- that used the xadd instruction. Because well as by change sets and number of ken peer. Matt thought the message was the 386 CPU didn’t implement an xadd lines of code affected. Also, it's possible alarming, so he posted a patch to change instruction, Adam asked whether Linux to search for a given person or employer it. Herbert Xu’s reply was, “What’s next, still supported the 386. The xadd in- and see a graph of only their contribu- you’re going to remove ‘printer on fire’ as struction turned out to be just a bug, but tions. According to Wang’s stats, for ex- well? This message has been there for the incident sparked a discussion about ample, Yahoo employees have contri- eons and is part of Linux lore.” But Alan which older systems were and were not buted much more to version 2.6.25 than Cox said, “It was changed. The printer on supported under Linux. to other recent kernels, whereas Micro- fire bit was adjusted to make it clear be- In terms of systems supporting Sym- soft employees are not listed as contrib- cause the original message did confuse metric Multi-Processing (SMP), Alan Cox uting anything at all. and worry a few people.” He felt Matt’s remarked that the first system to support There are still some problems with the patch was a good idea. Intel’s MP standard was the 486 with ex- engine. For example, if a person has con- However, it turned out that Alan was ternal APIC; he reckoned those would be tributed to the kernel using multiple not quite right. As Herbert pointed out the oldest systems capable of running email addresses, it still doesn’t seem pos- via grep, the printer on fire message was SMP Linux, although he thought the sible to see a graph representing the total still in the code. Alan remarked, “Inter- Earth might have been denuded of such contributions from all those addresses. esting. It got changed and then the systems long ago. Maciej W. Rozycki During the course of discussion, a num- change got lost between 2.2 and 2.4 commented, “I failed to track down a ber of folks pointed out this and other re- somewhere. It was changed to report single 486 SMP system that would ad- lated issues to Wang, so presumably the ‘lp0 reported invalid error status (on fire, here to the MP spec. There were and problem will be addressed. eh?)‘ and it looks like that needs re-fix- possibly still are APIC-based 486 SMP ing.” But David S. Miller said this just systems out there, but most likely they showed how much it didn’t matter, and are not Intel MPS-compliant, by not pro- The Linux kernel that having fun with kernel development viding the MP header at the very least. mailing list com- was one of the best parts. And Matt said Thus Linux would have to be ported and prises the core of that because no one had printer ports I gather the interest in doing so is epsi- Linux development anymore, this particular message proba- lon. Myself, I could not resist trying an activities. Traffic vol- bly wasn’t so dangerous. But he added, APIC-based 486 SMP box and possibly umes are immense, the “Treason uncloaked” message was fixing issues if I found one and it was often reaching ten filling the bulk of his system messages. MPS-compliant, but nothing beyond that thousand messages I would say. Life’s too short.” in a given week, and

70 ISSUE 101 APRIL 2009

070-071_kernel.indd 70 11.02.2009 15:59:11 Uhr Kernel News GIF>I8DD@E>

=lik_\iE\knfib`e^ Klo*:fdgXk`Y`c`kpJkXklj JkXkljF]JhlXj_=J J\Zli`kp Daniel Phillips announced that Tux3 was Phillip Lougher is submitting SquashFS Michael Stone proposed implementing a abandoning compatibility with older ker- for inclusion in the main kernel tree kernel feature to allow a process to irrevo- nels, and going forward it will compile again. The last SquashFS submission cably give up the ability to perform unre- only on kernels newer than 2.6.28. Tux3 was in 2005, and he’s made significant stricted network I/ O, not just for itself but would, however, continue to track the progress since then, including two major for all its future child-processes and their current version of the Linux kernel git revisions and a nearly complete rewrite descendants. He said this would let users repository. Tux3’s own development of all kernel-based code. Among other protect themselves from a whole big continues to take place using Mercurial changes, the filesystem is now 64-bit, batch of evil software. Andi Kleen replied (http:// www. selenic. com/ mercurial/ supporting fantasmagorically large files, that this could be partially done, at least wiki/), a version control system that and it presents the traditional ‘.’ and ‘..’ with outgoing packets, using netfilter. Mi- came out at about the same time as git, directories in directory listings. chael didn’t like this idea, though, and and seems to be the only free version With alternatives like CramFS already clarified a number of issues surrounding control system able to compete with git in existence, Phillip’s earlier attempt to what he wanted this thing to do – the in terms of speed and distributed devel- submit his code met with resistance, main points being that it should be pro- opment features. partly on the grounds that we don’t re- cess-centric and easy for distributions Getting back to Tux3, Daniel has an- ally need more than one compressed and vendors to adopt. He and Andi went nounced that he is about to embark on filesystem. This time, Phillip had pre- back and forth on the technical merits of implementing atomic commits, and so pared a number of counters to that argu- netfilter’s ability to support these require- the code will be in a period of instability ment. In filesystem size, file size, block ments, but at one point Alan Cox pointed while that happens. size, and various other measures, out that there were ways for a process to In any event, Tux3 is very much a SquashFS dramatically exceeds CramFS’s break out of any such restrictions and development project, and should only be capabilities. And as Phillip pointed out, that, really, more fine-grained security so- used by people who really enjoy testing CramFS is currently listed as orphaned, lutions like SELinux were the way to go. filesystems. while SquashFS is actively maintained.

C`eloDX^Xq`e\

APRIL 2009 ISSUE 101 71

070-071_kernel.indd 71 11.02.2009 15:59:12 Uhr