Finally got around to checking in some mechanisms to protect against meta-data race conditions. it seems to pass normal regressions and dbench.
While in the process, found a slew of errors using Inferno as the server. Will be pushing those later this week as they seem somewhat less critical.
The version of 9p in the 2.6.19 (and possibly 2.6.18) kernel was broken. Fixes have been posted and will hopefully get into 2.6.20. Updates are available in my repository on kernel.org. Sorry for the gap in coverage, but personal life events as well as my day job really got in the way of keeping on top of 9p maintainership in 2006—things have stabalized somewhat, so things should be better this year.
It is currently recommended that you use spfs (available from the NPFS sourceforge project) for serving 9p2000.u and Inferno for standard 9p. Developing a stable, multi-threaded server solution will be a primary focus on 2007.
I’ve got the new v9fs development branch somewhat stable on kernel.org—included in it our branches representing the current mm snapshot, my current release (patches submitted) and test (patches being tested) branches as well as some helper branches (systemsim-test includes patches to run the kernel on IBM’s Full System Simulator for ppc64 regression and the standalone branches which are meant to be the source for extracting stand-alone module tarballs which can be compiled on earlier versions of the kernel (ie. they have the ifdefs).
Right now the test branch is approximately the same at the -mm branch (it has all our changes, but is missing some of the impacts of more global changes that have been applied to -mm by others). Those changes are going through regression scripts now and will be pulled into the release branch sometime today.
I’ve setup a ppc64 auto-regression mechanism using IBM’s systemsim. I need to try and incorporate Lucho’s scripts into my suite, but right now am doing the basic sanity tests: fsx/Bonnie/postmark. I’m also in the process of setting up a rootdisk for qemu to do similar testing on unmodified kernels (systemsim requires some patches so its a little bit of pain to use for just regression). If there is interest, I’ll post the rootdisk somewhere where folks can download it.
My current work queue and patch queue is available off of SWiK – I’ll try and keep it up to date. If I’m missing something, please let me know and I’ll add it to the queue.
I’ve posted the first set of changes trying to fix our fid management issues. At Russ Cox’s suggestion, I’ve dramatically simplified the fid management and matching code. The previous design was too zealous, trying to account for scenerios that just don’t exist yet—so we’ll keep the complexity out till we need it. The code has passed basic regressions and is going through the longer regression runs now. This fix definitely fixes the ‘pwd’ problem, and may also fix the problems Ron had with some of his synthetics.
Not sure if these changes will make it into 2.6.16, but looks like they’ll be ready when the 2.6.17 window opens up.
Next on deck is addressing security issues, which will bring back some of the complexity of the fid matching (we need to at least match against uids) and will also bring up the new attaches for new users issues, how to deal with auth, and how to map uids/gids.
We are starting the transition to Lucho’s new npfs libraries and servers from u9fs. This should provide a better base for our future synthetic file server applications. Anyone who hasn’t tried it should go to the npfs sourceforge page and download a tar-ball to try it out. We’ll be making a formal release soon including rpms, debs, and ebuilds to send to the various distros.
I’m also going through and trying to update the bugzilla list. Most of the problems should have been fixed by some of Lucho’s rework in the kernel code.
Oh yeah, and as you must of noticed by now—we’ve moved the web page to SWiK hoping that it’ll help me keep things up to date easier.
- There are no tarballs on the npfs sourceforge page! -
The v9fs kernel code was merged into Linux’s 2.6.14 development tree last week – marking a fairly major transition for the v9fs project. Over the next few weeks I’m turning my focus onto some stability issues and the user-space support applications in order to help encourage its adoption and use. Thanks to everyone who has helped in the past with bug-reports, code, and suggestions.
Post a new entry