From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7) Gecko/20040706 Firefox/0.9.1 Description of problem: When doing an install onto a new reiserfs / partition (and not installing onto any other partitions), the installer freezes up after successfully formatting the new reiserfs filesystem. "ps" on the command line (on virtual terminal 2) shows that one "anaconda" process is stuck in D state, and it's possible to accidentally freeze up the bash shell on VT 2 as well. (Now that I'm writing this up, I think I should see what mount is freezing things up, but I'll do that and add it as a comment after the initial bug report.) Due to the way it's freezing up, I suspect the kernel is involved somehow. Version-Release number of selected component (if applicable): kernel-2.6.7-1.478, anaconda-10.0.1-2 How reproducible: Always Steps to Reproduce: (some steps have been omitted for brevity, but the important stuff is here) 1. Boot the installer with "linux reiserfs" or some variant thereof. 2. Choose any kind of install (personal workstation, custom, etc.) 3. Format a / partition as reiserfs, and do not format or mount any other partitions. 4. Go with whatever type of package selection you want. 5. Click "Next" on the last screen and let the install begin. Actual Results: The root filesystem is formatted and (I think) mounted, then the installer just freezes up. Expected Results: Installation should proceed without problems. Additional info: On my test system I am now reinstalling FC2 on a reiserfs partition. I will then attempt an upgrade to FC-devel (from 20040712); that will let me see whether this bug only occurs on fresh installs. I can reproduce this bug 100% of the time on both of the test systems on which I have tried fc-devel reiserfs installs. I will be attaching a few different log files which may provide some hints.
Created attachment 101852 [details] syslog at the final screen before installation starts
why not use ext3 ?
Created attachment 101853 [details] syslog after the installer freezes This is the last line from the log before the freeze, and it's one I've never seen before from the kernel: <4>inode_doinit_with_dentry: getxattr returned 1 for dev=hdc10 ino=3 BTW, this is with SELinux "Disabled" in the SELinux menu in the installer GUI. The same freeze also happens with SELinux set to "Enabled" or whatever the 3rd choice in the menu is (the choice that activates enforcing mode).
Created attachment 101854 [details] anaconda.log at the final screen before installation starts
Created attachment 101855 [details] anaconda.log after the installer freezes
At least for some of my workloads, reiserfs feels considerably faster. (I haven't timed it however.)
At the same level of dataprotection, reiserfs doesn't seem to be faster than the 2.6 ext3. It does seem to however create longer latencies ;(
Hmmm, I'll retest the two filesystems with newer 2.6 kernels, paying more attention to the "same level of dataprotection" aspect. (With 2.4, I did time some of my workloads, and reiserfs was typically twice as fast as ext3. Of course, 2.6 changes things enough that I can't rely on those numbers anymore and I need to repeat the tests.) Anyway, reiserfs has other advantages -- for instance, when tails are enabled, it uses much less space if you're storing tons of tiny files.
Let me clarify my last comment: My reiserfs vs. ext3 tests under 2.4 did not involve writing any data and were conducted with atimes disabled, so I doubt the journalling modes of the two filesystems would have made a significant difference. (BTW, I also tested JFS at the time, and ext3 was 4.5X as fast as JFS.) Also, I'm *not* trying to get Red Hat to change the default filesystem. ext3 is a good default, and I do use it on most of my systems. Nonetheless, it would be nice if reiserfs worked as an option.
I've finished testing an anaconda upgrade from FC2-on-reiserfs to FC-devel: anaconda gets stuck in D state at "Finding packages to upgrade..." -- there's some disk activity for maybe a few minutes, and then it freezes up (although the hourglass cursor keeps tumbling). AFAICT, once the freeze happens, anything else that tries to access /mnt/sysimage (e.g. "ls /mnt/sysimage" on VT 2) gets stuck in D state.
Comment #3 is not a surprise, because reiserfs from upstream 2.6 does not support xattrs. I do not know however if that is related to the behavior you are seeing, or if this is fixable with only Anaconda changes. Realistically Red Hat probably will not expend any resources into looking into this issue for FC3 because reiserfs is not a supported feature. However if the problem is diagnosed and a good fix is submitted by the community, and it is limited to Anaconda, we will try to review and include it. If it requires kernel changes, then pretty much the only way it will get into FC3 is if you get it into the upstream kernel.
Somehow I thought xattr support had been added to reiserfs upstream at some point recently. Perhaps I'm just confusing it with something else, though... Er, wait, what's this? CONFIG_REISERFS_FS_XATTR=y (this is from the .config for kernel-2.6.7-1.478 i686)
Wow, that's surprising. I recall somebody saying 'maybe 2.7' for that feature. Maybe they changed their mind later. In any case I think FC2's original kernel lacked that capability.
Yes, it was definitely missing from the original FC2 kernel. IIRC there was some kind of argument on lkml about this; Hans Reiser wanted to wait (for reiserfs v4, if I'm not mistaken) but he was overriden by other people. BTW, JFS is also failing, but in a different manner. (Maybe this belongs in a bug of its own, but I want to look at this a bit more first.) Instead of freezing up, it gets far enough to fail with this: ---begin quote--- Error Installing Package There was an error installing fedora-logos-1.1.25-1.2. This can indicate media failure, lack of disk space, and/or hardware problems. This is a fatal error and your install will be aborted. Please verify your media and try your install again. Press the OK button to reboot your system. ---end quote--- (I also tried XFS. To make a long story short, my testing was inconclusive because of mistakes I made, and I'll repeat it correctly later, but XFS is completely unaffected by anything remotely resembling this bug, and it's probably in the same shape as it was in FC2.)
I'm lowering the severity now that I have a workaround: Add "selinux=0" to the installer command line as well as "reiserfs" and/or "jfs". "selinux=0" prevents the problems with both jfs and reiserfs. I can think of possible changes to anaconda to keep jfs and reiserfs installs from failing without "selinux=0", but I think it would also be an additional, unnecessary maintenance burden on Red Hat down the road. There's a patch that makes the JFS xattrs SELinux-compatible, but it hasn't been committed upstream for some reason. Anyway, I'm not going to say anything else about JFS in this bug, to prevent this bug from becoming hydra-headed. My best guess about the reiserfs freeze issue is that it's something I'll have to handle w/the upstream maintainers. So, I'll close this bug as UPSTREAM for now, and if it turns out to be specific to the FC kernel, I'll consider reopening the bug.
I've now reported this to upstream as bugme.osdl.org #3080: http://bugme.osdl.org/show_bug.cgi?id=3080