Submitter : dxm Status : open
Assigned Engineer : lord Priority : 3
*Modified Date : 06/22/00 *Modified User : jtk
*Modified User Domain : sgi.com *Description :
syscalls into XFS can get hung up sleeping on an event
that doesn't ever happen (or just delayed for many seconds).
XFS qa 013 will often trip this bug: the fsstress process sits
in 'R' state, waiting for a syscall to return.
kdb> btp 16558
EBP EIP Function(args)
0xc311fbd8 0xc01134b2 schedule+0x2b6 (0xc311fbec)
kernel .text 0xc0100000 0xc01131fc 0xc0113660
.....
==========================
ADDITIONAL INFORMATION (ADD)
From: ted kline <jtk@xxxxxxx>
Date: Jun 22 2000 06:35:04PM
[pvnews version: 1.71]
==========================
>
> p.p.ps. Ted can answer questions about why the delay is there.
>
There was a create/unlink race, where one thread, in an unlink,
has ifree'd the inode #, another thread, in a create, has grabbed
the inode # before the unlink is done, xfs_iget recognizes this,
drops locks, and gives up control for 1 tick (1/100 second),
thus allowing the iextract to complete, then tries again.
This happens fairly rarely..
-Ted
|