SUMMARY: Processes hang opening /dev/ksyms

From: Brian Tao (taob@risc.org)
Date: Tue May 11 1999 - 23:55:13 CDT


    I don't know if Sun officially has a patch for this yet, but
Casper Dik suggested a successful workaround. The kernel module
loader and /dev/ksyms can deadlock under certain conditions (it may
have something to do with leftover NFS client state and lock files
after a reboot). Forceloading the loader modules at boot allows
things to work again. In /etc/system:

        forceload: misc/klmmod
        forceload: misc/klmops

On Fri, 19 Mar 1999, Brian Tao wrote:
>
> I have an Ultra-1 running Solaris 2.5.1 (SunOS tor-dev1 5.5.1
> Generic_103640-17 sun4u sparc) that had been up for 336 days with no
> apparent problems. Last weekend, it had to be physically moved, so it
> was shutdown and rebooted in its new home.
>
> When it came back up, everything seemed fine at first, but I
> slowly discovered that anything that reads /dev/ksyms would hang.
> This includes arp, netstat, prtconf, lsof, dmesg or even a dd of the
> device file. Truss shows that execution of a program will freeze on
> open("/dev/ksyms", O_RDONLY) .
>
> A modinfo will also halt, but not on the /dev/ksyms module:
>
> # modinfo &
> [...]
> 70 600df008 181a 13 1 telmod (telnet module)
> 71 6077a000 1b98 4 1 logindmux ( LOGIND MUX Driver)
> 72 6055e800 778 22 1 sy (Indirect driver for tty 'sy')
> 73 6055e000 6b4 90 1 kstat (kernel statistics driver)
> [hangs]
>
> # modinfo -i 76
> can't get module information: Invalid argument
>
> # modinfo -i 75
> Id Loadaddr Size Info Rev Module Name
> 75 608fb000 ea0 72 1 ksyms (kernel symbols driver)
>
> # modinfo -i 74
> [hangs]
>
> # ls -lL /dev/ksyms
> crw-rw-rw- 1 root sys 72, 0 May 3 1996 /dev/ksyms
>
>
> A boot -v of the system looks perfectly normal, and the usual
> services on that box (NFS client daemons, sshd, Apache httpd, named,
> etc.) all start up correctly and problem-free. There have not been
> any patches applied to the system during its 336-day stint. Anything
> else I should be looking at here?
> --
> Brian Tao (BT300, taob@risc.org)
> "Though this be madness, yet there is method in't"
>

-- 
Brian Tao (BT300, taob@risc.org)
"Though this be madness, yet there is method in't"



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:19 CDT