See also: last week's Kernel page.

Kernel development

2.3 development thus far has concentrated mostly on internal changes thus far, such as tweaking the way that wait queues are handled. One user-visible change has been the addition of EFS - the Extent File System, used by SGI in older versions of Irix and on their software distribution CD's. More on the Linux EFS implementation can be found at the Linux EFS web site.

The current stable kernel release is 2.2.9, which was put out in a bit of a hurry after some ill-advised buffering changes made 2.2.8 into a rather dangerous thing to run. Quite a few users have complained about extensive file system corruption caused by 2.2.8. Anybody who is still running a 2.2.8 kernel should quickly revert back to an older one or upgrade to 2.2.9, which has the offending changes removed.

There have also been some complaints that recent 2.2 releases have worse memory behavior. One user did some testing and concluded that 2.2.7 introduced the problem, which shows up as excessive swapping. Just how extensive this problem is is still not clear.

BitKeeper is coming out at Linux Expo. There will be a BOF to introduce the system, as well as a booth where interested people can check out a version of the system with a repository containing the entire kernel development history. Larry McVoy has put the slides for his presentation up on the web as well. If all goes according to plan, the core kernel developers will start using BitKeeper before too long.

Linus has finally chimed in on capabilities. The debate, remember, was between those who want capability (privilege) information stored in the file system and those who would rather put it into the header of the ELF executable file. Linus seems to lean toward the file system solution, even though it looks harder in the long run. His reasoning was a bit new: his main concern with the ELF header solution is that it only works with ELF executables. If you want to associate capabilities with scripts (which could serve to reduce privilege as well as increase it), the ELF approach will not work. Here's a brief note giving some examples of what he's thinking about.

Since things generally go the way Linus wants them to in the kernel realm, it is likely that the ELF header approach may not proceed much farther.

A USB success story. Interested folks may want to check out this posting from Ben Pfaff, who got his USB mouse working on a 2.2.8 system. Included is some information on what he did. We're pretty much at the point where Linux can no longer be criticised for lack of USB support.

A NetWare file system for Linux has been made available under the GPL by the Timponogas Research Group. See their press release for details. In a separate posting on the kernel list, the Timpanogas person in charge of the release suggested that a full Novell Directory System (NDS) implementation may be the next thing they release.

Other interesting software and patch releases this week:

• The third release of the input patches from Vojtech Pavlik is out. These patches include USB support and other good stuff.

• H.J. Lu has released knfsd 1.3.1. This updates his kernel NFS patches for recent 2.2 releases.

• PPSkit 0.7-pre3 by Ulrich Windl is out. This patch adds nanosecond time keeping and NTP tweaks; with this release he has also put in an implementation of the PPS API, a draft standard for high resolution event timing.

• DIPC 1.1c is out. The Distributed Inter-Process Communication package aids in the building of clustered applications, with transparent implementations of semaphores, message queues, and shared memory.

• Richard Gooch's eternal devfs patch is up to version 102.

• Andrea Arcangeli posted an extensive set of buffer management patches which fixes and improves a lot of little things.

Section Editor: Jon Corbet

For other kernel news, see: LinuxHQ Kernel traffic

Next: Distributions