Red Hat Bugzilla – Bug 223393
SCSI disk I/O with Emulex adapters on Linux RHEL4U4 may experience Linux �mkfs� process hangs.
Last modified: 2010-06-07 01:36:33 EDT
Description of problem: I/O intensive user applications running on RedHat Enterprise Linux 4 update 4 and using Emulex host bus adapters and the Emulex driver version 8.0.16.27, may experience process hangs of the Linux filesystem creation process called mkfs. An examination of both the Emulex code tree and behaviors, as well as RHEL 4 U4 mkfs behaviors, leads HP and Emulex to believe this is a filesystem or other RHEL 4 related problem. When a mkfs process hang occurs on a SCSI disk device, concurrent raw I/O to the disk device in question is unaffected. All filesystem I/O to the affected device will wait for successful completion of the mkfs process before they can execute. Filesystem and raw I/O processes to other disk devices are unaffected. Version-Release number of selected component (if applicable): How reproducible: 100% Steps to Reproduce: 1. Use test tool that will mount, mkfs and run intensive FS I/O to several devices attached in a SAN (fibre-channel) 2. Monitor various system logs and ps functions to watch for hung mkfs process, as well as cessation of FS I/O to the affected device (can take up to 9 hours to see). 3. Monitor any test-tool alerts, if provided, for hung or time-out processes. Actual results: mkfs process hang occurs after several hours of I/O, can take up to nine hours to clear. FS I/O ceases to the affected device. Expected results: mkfs should complete gracefully, terminate, and I/O should be successfully executed on the device(s) under test. Additional info: Please see attached strace file
Created attachment 145954 [details] strace on hung mkfs process
Created attachment 145955 [details] Clipping of test-tool log showing hung mkfs process
Kernel request, reassigning. Read ya, Phil
Hi, Paul. Are you trying to do raw i/o to the same device that mkfs is using? Are you getting physical i/o errors, and if so, what are they?