On Fri, Jan 28, 2005 at 03:47:58PM +0100, Slawek Zak wrote: +> Happened on yesterday's 5.3-STABLE. +> +> It is a mount from SPARC/Solaris serving a ~600GB raw disk to FreeBSD. +> On the FreeBSD side were 10 iozone processes looping in automated mode +> (-a). Iozone runs fine up to 512K block until the server dies terrible +> death. After reboot the filesystem (UFS2 with softupdates) cannot be +> mounted or checked: +> +> ravell# fsck -t ufs /dev/ggate0 +> ** /dev/ggate0 +> Cannot find file system superblock +> ioctl (GCINFO): Inappropriate ioctl for device +> fsck_ufs: /dev/ggate0: can't read disk label +> ravell# dumpfs /dev/ggate0 +> dumpfs: /dev/ggate0: could not read superblock to fill out disk It is possible that ggate has returned an error to the file system, because of full queue or something simlar. If so, it is possible that file system can crash (UFS do really poor job in I/O errors handling). You can catch errors from ggate by setting kern.geom.gate.debug to 1. +> As a side note, would it be hard to eliminate userland pages from the +> crashdump to both increase secrecy and shorten crashfile/dump duration? +> Can be done in both Solaris and Tru64, would be good to have in FreeBSD too. Yeah... -- Pawel Jakub Dawidek http://www.wheel.pl pjd_at_FreeBSD.org http://www.FreeBSD.org FreeBSD committer Am I Evil? Yes, I Am!
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:27 UTC