Date: Tue, 10 Nov 1998 03:36:28 +0200 (EET) From: Alexander Litvin <archer@lucky.net> To: Luoqi Chen <luoqi@watermarkgroup.com> Cc: current@FreeBSD.ORG Subject: Re: The infamous dying daemons bug Message-ID: <199811100136.DAA00563@grape.carrier.kiev.ua> In-Reply-To: <199811092056.PAA22043@lor.watermarkgroup.com>
next in thread | previous in thread | raw e-mail | index | archive | help
In article <199811092056.PAA22043@lor.watermarkgroup.com> you wrote:
>> Totally unrelated to the problem. It seems, so far as I was able to
>> characterize, to happen to daemons which are *swapped out* at the time
>> of the memory shortage. If it's active enough to still be in core, it
>> doesn't get spammed.
>>
LC> I went through swap_pager.c today and found a problem that could potentially
LC> have bad consequences. It's a comparison between page index in the swap pager
LC> and the size of the vm object, since a shadowed object may have a non-zero
LC> paging offset with respect to the swap pager, the comparison should have taken
LC> the offset into account. This piece of code has been there since '95, so
LC> I can't say if this was responsible for the daemon dying problem.
It is definitely not responsible. At least for this particular problem,
though it may fix something else ;)
After applying the patch, and artificially exhausting memory, I promptly
got:
Nov 10 03:15:34 grape /kernel: swap_pager: suggest more swap space: 61 MB
Nov 10 03:16:26 grape /kernel: pid 310 (sendmail), uid 0: exited on signal 11
Nov 10 03:17:26 grape /kernel: pid 311 (sendmail), uid 0: exited on signal 11
Nov 10 03:18:25 grape /kernel: pid 313 (sendmail), uid 0: exited on signal 11
Nov 10 03:19:25 grape /kernel: pid 353 (sendmail), uid 0: exited on signal 11
Nov 10 03:20:26 grape /kernel: pid 394 (sendmail), uid 0: exited on signal 11
LC> -lq
LC> Index: swap_pager.c
LC> ===================================================================
LC> RCS file: /home/ncvs/src/sys/vm/swap_pager.c,v
LC> retrieving revision 1.103
LC> diff -u -r1.103 swap_pager.c
LC> --- swap_pager.c 1998/10/31 15:31:28 1.103
LC> +++ swap_pager.c 1998/11/09 11:02:54
LC> @@ -349,7 +349,7 @@
LC> *valid = 0;
LC> ix = pindex / SWB_NPAGES;
LC> if ((ix >= object->un_pager.swp.swp_nblocks) ||
LC> - (pindex >= object->size)) {
LC> + (pindex >= object->size + OFF_TO_IDX(object->paging_offset))) {
LC> return (FALSE);
LC> }
LC> swb = &object->un_pager.swp.swp_blocks[ix];
LC> @@ -1227,8 +1227,8 @@
LC> * intent of this code is to allocate small chunks for
LC> * small objects)
LC> */
LC> - if ((off == 0) && ((fidx + ntoget) > object->size)) {
LC> - ntoget = object->size - fidx;
LC> + if ((off == 0) && ((fidx + ntoget) > object->size + paging_pindex)) {
LC> + ntoget = object->size + paging_pindex - fidx;
LC> }
LC> retrygetspace:
LC> if (!swap_pager_full && ntoget > 1 &&
---
We are the people our parents warned us about.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199811100136.DAA00563>