Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 8 Nov 2011 19:50:50 GMT
From:      Garrett Cooper <yanegomi@gmail.com>
To:        freebsd-gnats-submit@FreeBSD.org
Subject:   kern/162382: Orphaned swap references not garbage collected; renders impossible to remove
Message-ID:  <201111081950.pA8JooTh051419@red.freebsd.org>
Resent-Message-ID: <201111082000.pA8K0O3R035772@freefall.freebsd.org>

next in thread | raw e-mail | index | archive | help

>Number:         162382
>Category:       kern
>Synopsis:       Orphaned swap references not garbage collected; renders impossible to remove
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Tue Nov 08 20:00:24 UTC 2011
>Closed-Date:
>Last-Modified:
>Originator:     Garrett Cooper
>Release:        9.0-RC1
>Organization:
iXsystems, Inc.
>Environment:
FreeBSD bayonetta.local 9.0-RC1 FreeBSD 9.0-RC1 #0: Sat Nov  5 17:19:05 PDT 2011     root@bayonetta.local:/usr/obj/store/freebsd/stable/9/sys/BAYONETTA  amd64
>Description:
If one sets up multiple swaps that are striped across disks and one goes south, the kernel retains a reference to the lost swap, but unfortunately doesn't remove the phantom reference; furthermore, the user can't remove the phantom reference, because the bogus device doesn't exist. This can cause issues if the swap was being used, either at a kernel or userland level.

Eventually [with mps at least] the system gets cranky and kicks the device out of some subsystems, but doesn't get rid of the bogus reference -- thus it could cause serious issues if a process tries to swap back from the phantom swap device.

What I'm recommending is that the phantom reference be reaped immediately. This should trickle up through all affected subsystems (GEOM -> vm -> etc). Any processes trying to access memory that was swapped out should be SIGKILLed immediately in a worst case scenario, but other potentially more graceful (or quicker) methods of notifying affected processes could be employed so processes that carve out a large chunk of memory that's gone MIA (SIGABRT? SIGSEGV?).

delphij@ might have more details or context into the underlying issue.
>How-To-Repeat:
1. Install FreeBSD on a system with an HBA or gmirror that's capable of hotswap and setup the RAID in a redundant manner.
2. Setup swaps on multiple devices in a striped manner.
3. Yank a disk with a swap on it.
4. Look at swapctl -l; you'll see a gobbledygook reference to a dead device (something like "/dev/C@#$A6^7").
>Fix:


>Release-Note:
>Audit-Trail:
>Unformatted:



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201111081950.pA8JooTh051419>