https://bugzilla.novell.com/show_bug.cgi?id=409011 Cyril Hrubis changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |chrubis@novell.com AssignedTo|bnc-team-screening@forge.provo.novell.com |ro@novell.com -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User vegard.myrvang@gmail.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c2 vegard myrvang changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |vegard.myrvang@gmail.com Status|NEEDINFO |NEW Info Provider|vegard.myrvang@gmail.com | --- Comment #2 from vegard myrvang 2008-07-25 07:17:02 MDT --- Hi I checked once more to be sure. And my halt.local does not have any content except comment lines starting with #. vegard -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User xenoterracide@gmail.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c4 --- Comment #4 from Caleb Cushing 2008-08-08 11:07:19 MDT --- I suppose I should have checked that file out... not sure what it has to do with a kernel upgrade... downgrading the kernel to the previous patch and re-upgrading worked for me. before you do it however, might want to double check exactly what's in those patches and see what files change. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c6 --- Comment #6 from Jean Delvare 2008-09-21 12:26:09 MDT --- Further investigations reveal that, at least in my case, what hangs is not halt.local but another command later in the reboot script: test -z "$hupon" || fuser -sk -HUP $hupon In my case, $hupon has value "/dev/tty1 /dev/tty2 /dev/tty3 /dev/tty4 /dev/tty5 /dev/tty6". It is also worth noting that, again at least in my case, the hang isn't actually a hang. After several minutes, the machine finally reboots as instructed. And a last note, for what it's worth... Although I am totally unable to explain why, it seems that the phenomenon is amplified by network mounts. If I manually unmount all my network mounts before rebooting the machine, the hang doesn't happen. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User ro@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c8 Ruediger Oertel changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |NEEDINFO Info Provider| |werner@novell.com --- Comment #8 from Ruediger Oertel 2008-09-22 07:25:54 MDT --- werner: any idea ? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c10 Dr. Werner Fink changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |werner@novell.com Info Provider|werner@novell.com |jdelvare@novell.com --- Comment #10 from Dr. Werner Fink 2008-09-22 09:03:20 MDT --- Could be due the changes for bug #346776 ... the unpatched fuser does always a stat(2) on every mount point. This should not happen with the psmisc-22.6-nfs4fuser.patch or better only if the option `-m' is choosen. AFAICS see from my strace here, this is true: strace -e stat64 src/fuser /dev 2>&1 | \ grep -E "(`mount | grep 'type nfs ' | awk '{print $3}' | xargs | tr ' ' '|'`)" returns nothing but strace -e stat64 src/fuser -m /dev 2>&1 | \ grep -E "(`mount | grep 'type nfs ' | awk '{print $3}' | xargs | tr ' ' '|'`)" shows stat64("/suse/werner", {st_mode=S_IFDIR|0755, st_size=28672, ...}) = 0 therefore I'd like to know *which` mount points are *not* ignored by the stat(2) scheme of fuser? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c12 Jean Delvare changed: What |Removed |Added ---------------------------------------------------------------------------- Info Provider|jdelvare@novell.com |werner@novell.com --- Comment #12 from Jean Delvare 2008-09-22 09:34:06 MDT --- Werner, your patch only prevents stat() from being called on NFS mounts. And it works for NFS, I'm not arguing that. For all other mount types, stat() is still being called. But this is hardly the problem anyway. The question is, why does your patch move the calls to scan_mount_devices() and fill_unix_cache() inside the command-line parsing loop? This is where the big performance regression stands. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c14 --- Comment #14 from Dr. Werner Fink 2008-09-22 09:41:36 MDT --- Beside this I'd like to have a way to use stat(2) or readlink(2) even on active NFS mount points but be able to be interrupted after a few ms in case the network is down or the NFS server concerned is not responding. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c16 --- Comment #16 from Dr. Werner Fink 2008-09-23 03:12:53 MDT --- The parsing of the mount points was never NFS specific, this *is* the problem as comparision based on inode and device numbers for each file or directory will cause a stat(2). It is not that easy to avoid stat(2) for each entry and make fuser to work as without the patch. Comparison based on strings as the oldfuser does is not that secure and was missing very often files and/or directory ... or even more reports wrong files and/or directory. And yes the move of the two functions was done in a block, means both lines together, and the problem with the performance impact (and my fault) I detected during debugging for bug #346776. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c18 --- Comment #18 from Dr. Werner Fink 2008-09-23 03:24:11 MDT --- The lsof utility seems to use a workaround, as described in the manual page lsof(3) in section `BLOCKS AND TIMEOUTS': Lsof can be blocked by some kernel functions that it uses - lstat(2), readlink(2), and stat(2). These functions are stalled in the kernel, for example, when the hosts where mounted NFS file systems reside become inaccessible. Lsof attempts to break these blocks with timers and child pro- cesses, but the techniques are not wholly reliable. When lsof does manage to break a block, it will report the break with an error message. The messages may be suppressed with the -t and -w options. maybe this is the way to go. But AFAIK the /etc/mtab has no dev=... within the options filed, therefore the technique described below will not work on Linux. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c20 --- Comment #20 from Jean Delvare 2008-09-23 06:18:33 MDT --- Created an attachment (id=241124) --> (https://bugzilla.novell.com/attachment.cgi?id=241124) Make fuser parse mtab only when needed Werner, that's what I had in mind. This patch goes on top of the current psmisc package in head. It modifies the code to fill out the mount_devices list only when needed. This prevents any network timeout/blocking issue as long as -m/-c isn't used. I've tested it and it works for me. Do you think this could be applied for openSuse 11.1? And possibly backported to 11.0 together with psmisc-22.6-fdleak.patch? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c21 Dr. Werner Fink changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEEDINFO |NEW Info Provider|werner@novell.com | --- Comment #21 from Dr. Werner Fink 2008-09-23 06:46:37 MDT --- Created an attachment (id=241132) --> (https://bugzilla.novell.com/attachment.cgi?id=241132) statsafely.c Looks fine ... just had worked out a nfssafe() to reduce the timeout. Only question is: does the fork cause a performance impact? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c23 --- Comment #23 from Jean Delvare 2008-09-23 07:33:18 MDT --- Ah, sorry, I hadn't noticed that you had modified scan_mount_devices() to have undocumented side effects. So indeed my patch can't be applied as is. Most probably, with your changes, it can't be applied at all. Oh well, I probably better stop looking at your code. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User cthiel@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c25 Christoph Thiel changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mhopf@novell.com --- Comment #25 from Christoph Thiel 2008-09-25 04:15:00 MDT --- *** Bug 429204 has been marked as a duplicate of this bug. *** https://bugzilla.novell.com/show_bug.cgi?id=429204 -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User werner@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c27 Dr. Werner Fink changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |samba-maintainers@SuSE.de Info Provider|jdelvare@novell.com |samba-maintainers@SuSE.de --- Comment #27 from Dr. Werner Fink 2008-09-25 07:41:08 MDT --- The patch is for NFS only ... maybe the samba people are knowing how to detect the CIFS mount in /etc/mtab or /proc/mounts -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c29 --- Comment #29 from Jean Delvare 2008-09-26 03:38:46 MDT --- I have the feeling that we're getting off-track here. The simple truth is that "fuser -sk -HUP /dev/tty1" doesn't need to know anything about mounted disks, so it shouldn't access them. That's what my patch in comment #20 was doing, and I am confident that it's the way to go. In its current form, it breaks Werner's NFS patch, but I see no reason why both patches couldn't be made to work together. Trying to handle nfs and cifs separately sounds plain wrong. There are several other network disk types and we certainly don't want to add specific code for all of them, especially when there is no need for that in the first place. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User jdelvare@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c31 --- Comment #31 from Jean Delvare 2008-09-26 06:30:30 MDT --- Scanning the mount table is one thing, and it's not too expensive so we don't really care. Running stat on every mount point is another which _is_ expensive and may lead to long timeouts, so it should only be done when we really need it. That is, only when -m is used. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=409011 User ro@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=409011#c33 Ruediger Oertel changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEEDINFO |RESOLVED Info Provider|sjayaraman@novell.com | Resolution| |FIXED --- Comment #33 from Ruediger Oertel 2008-11-05 08:51:22 MST --- closing due to last comment, please reopen if the same issue re-appears. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.