Date: Wed, 18 Jun 2003 02:30:09 +0000 (UTC) From: Ryan Lackey <ryan@metacolo.com> To: FreeBSD-gnats-submit@FreeBSD.org Subject: kern/53433: heavy i/o on GBDE partition on SMP locks up box on 5.0/5.1 Message-ID: <20030618023009.B84C1284F4@mx1.metacolo.net> Resent-Message-ID: <200306180240.h5I2e5xh061667@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 53433 >Category: kern >Synopsis: heavy i/o on GBDE partition on SMP locks up box on 5.0/5.1 >Confidential: no >Severity: critical >Priority: high >Responsible: freebsd-bugs >State: open >Quarter: >Keywords: >Date-Required: >Class: sw-bug >Submitter-Id: current-users >Arrival-Date: Tue Jun 17 19:40:05 PDT 2003 >Closed-Date: >Last-Modified: >Originator: Ryan Lackey >Release: FreeBSD 5.0-RELEASE-p7 i386 >Organization: metacolo >Environment: System: FreeBSD bb1-sfo.metacolo.net 5.0-RELEASE-p7 FreeBSD 5.0-RELEASE-p7 #9: Wed Jun 18 01:25:58 UTC 2003 root@bb1-sfo.metacolo.net:/usr/obj/usr/src/sys/RYAN i386 >Description: My system reliably locks up when operating in SMP mode and doing large disk I/O to a gbde-encrypted partition (postfix mailspool) on 5_1_0-RELEASE and 5_0. Specifically a directory of >200M with lots of postfix mail spool files (>2000?) >How-To-Repeat: On my system, a 5.1-RELEASE VA Linux 2230 Dual p3-800Mhz L440gx based SMP system, I have a 1.4G GBDE partition. In this partition, I had a postfix mail spool which due to a mail flood had expanded to 400M. Upon noticing this full partition, I ssh'd in and began to delete files. Part of this involved deleting the 400M mailspool and manually removing files...during this process, the machine locked up (remotely). No syslog messages were present from between normal operation and reboot, so I assume the machine paniced. I rebooted and had the same behavior on trying to delete the spool. I downgraded to 5_0 and had exactly the same problem when attempting rm -rf on the postfix spool. I recompiled the kernel without SMP support and on 5_0 have not yet had another lockup. I successfully deleted the mailspool. syslog messages included repeats of the following pre-crash: Jun 15 22:20:08 bb1-sfo kernel: Jun 15 22:20:08 bb1-sfo remailer/qmgr[20985]: fatal: myrealloc: insufficient memory: Cannot allocate memory Jun 15 22:29:52 bb1-sfo kernel: Jun 15 22:29:52 bb1-sfo kernel: /home: optimiza\tion changed from TIME to SPACE Jun 15 22:30:25 bb1-sfo kernel: Jun 15 22:30:25 bb1-sfo remailer/qmgr[35323]: f\atal: scan_dir_push: open directory active: No such file or directory >Fix: Disabling SMP support seems to make 5_0 reliable, at least. I will investigate with 5_1, and assemble a spare machine for more extensive testing of -CURRENT SMP+GBDE interaction under load, since I plan to use this configuration in heavy production as soon as possible. >Release-Note: >Audit-Trail: >Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030618023009.B84C1284F4>