Date: Wed, 12 Jun 2013 00:21:22 +0400 From: Vagner <vagner@bsdway.ru> To: Konstantin Belousov <kostikbel@gmail.com>, FreeBSD hackers Mail List <freebsd-hackers@freebsd.org> Subject: Re: Where is stack of program? Message-ID: <20130611202122.GA3348@vagner-wrk.bsdway.ru> In-Reply-To: <20130611080538.GA2645@vagner-wrk.bsdway.ru> References: <20130610164425.GA2966@vagner-wrk.bsdway.ru> <20130611054544.GI3047@kib.kiev.ua> <20130611080538.GA2645@vagner-wrk.bsdway.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
--SUOF0GtieIMvvwua
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
On 12:05 Tue 11 Jun , Vagner wrote:
> On 08:45 Tue 11 Jun , Konstantin Belousov wrote:
> > On Mon, Jun 10, 2013 at 08:44:25PM +0400, Vagner wrote:
> > > Hi!
> > > I need method of finding in struct vm_map or vm_object segments of
> > > memory which is situated in the stack.
> > > Can you help me please?
> >
> > Note that the stack is per-thread. The concept is somewhat machine-specific,
> > some architectures utilize two stacks, on some the stack is purely software
> > convention.
> >
> > You did not specified what context your code is to be run, e.g. is
> > it kernel or user space ? Assuming it is kernel since you mentioned
> > vm_something. The least error prone route is the thread context (frame)
> > -> tf_esp on i386 (or tf_rsp on amd64) -> lookup of the map entry in the
> > process p_vmspace.
> >
> > In reality, the stack is often fragmented, since the stack grow code
> > does not coalesce the adjacent grow-down map entries.
>
> I asket the question because in my servers run very large daemons (writed their own). Then daemons get
> signal for create coredump, downtime approximately 1h. This is very long for daemons in
> production. Often in coredump for debug need only stack and current
> frame of code, but in function, which initialise for create dump of
> memory haven't this possibility. Linux have /proc/self/coredump_filter
> for settings what segments included in core file, but in our FreeBSD I
> don't find this:( Help me for find solution please
>
My solution.
I added new option MALLOC_OPTIONS ("S") in environment variable for malloc(3).
Also I added new `if' in function pages_map at libc/stdlib. In this `if' I added option MAP_NOCORE
(mmap(2)) for segments allocated with MAP_PRIVATE | MAP_ANON. Patch file
attached.
As a result: I have a corefile with stack and text of program ~ 2Mb (above
- 2Gb).
Thanks to all
--
Respectfully,
Stanislav Putrya
System administrator
FotoStrana.Ru Ltd.
ICQ IM: 328585847
Jabber-GoogleTalk: root.vagner
mob.phone SPB: +79215788755
mob.phone RND: +79525600664
email: vagner[at]bsdway.ru
email: putrya[at]playform.ru
email: root.vagner[at]gmail.com
site: bsdway.ru
site: fotostrana.ru
----------------------------------------
( ) ASCII ribbon campaign
X - against HTML, vCards and
/ \ - proprietary attachments in e-mail
--SUOF0GtieIMvvwua
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment; filename="core_only_stack.patch"
Index: lib/libc/stdlib/malloc.c
===================================================================
--- lib/libc/stdlib/malloc.c (revision 249257)
+++ lib/libc/stdlib/malloc.c (working copy)
@@ -1155,9 +1155,11 @@
#ifndef MALLOC_PRODUCTION
static bool opt_abort = true;
static bool opt_junk = true;
+static bool opt_stackcore = false;
#else
static bool opt_abort = false;
static bool opt_junk = false;
+static bool opt_stackcore = false;
#endif
#ifdef MALLOC_TCACHE
static size_t opt_lg_tcache_nslots = LG_TCACHE_NSLOTS_DEFAULT;
@@ -1797,8 +1799,12 @@
* We don't use MAP_FIXED here, because it can cause the *replacement*
* of existing mappings, and we only want to create new mappings.
*/
- ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON,
+ if (opt_stackcore)
+ ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON | MAP_NOCORE,
-1, 0);
+ else
+ ret = mmap(addr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON,
+ -1, 0);
assert(ret != NULL);
if (ret == MAP_FAILED)
@@ -5623,6 +5629,9 @@
case 'P':
opt_stats_print = true;
break;
+ case 'S':
+ opt_stackcore = true;
+ break;
case 'q':
if (opt_lg_qspace_max > LG_QUANTUM)
opt_lg_qspace_max--;
--SUOF0GtieIMvvwua--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130611202122.GA3348>