Force a core to dump from an active, normally running program on FreeBSD

Question

I'm writing error handling code for a server on FreeBSD. For extremely serious errors I want to avoid data corruption by immediately terminating. That's easy, exit(3). Before I exit, I output my relevant variables that led me there. However, ideally, this termination would be accompanied by a .core so that I could fully investigate what got me to this catastrophic (and likely hard to reproduce) state.

How can I force this to happen?

msw · Accepted Answer · 2016-11-20T14:49:40.720

49

kill -QUIT process_id will cause a core dump from a running process (assuming that resource limits allow it).

Or see man 3 abort for causing a program to dump itself.

Added: From an interactive shell, a running program can be made to abort with the quit key, usually Ctrl+\, which sends a SIGQUIT just as the more common Ctrl+C sends a SIGINT. This is identical to the kill -QUIT… it's just easier to type if you are on the controlling terminal. See man 1 stty if your default quit key is different.

edited Nov 20 '16 at 14:49

answered Jul 03 '11 at 05:10

msw

42,753
9
87
112

1

You can also use kill -3 [pid] - same deal, less typing. You can see the other kill flags on the manpage: man kill. – smcphill Jul 03 '11 at 05:13
Well, I want to do it from the process itself; so `abort(3)` seems ok but from some testing here I don't seem to be able to access any symbols when examining the resulting `.core`. I can get a backtrace fine, but if I try to print any of my variables they aren't found. Is this expected with `abort(3)` or am I doing something stupid? `gcc -g -O0 test.c -o test` then `./test` then `gdb ./test test.core` – Nektarios Jul 03 '11 at 05:26
2

I think your problem is with using gdb: you need to pick a frame from the `where` backtrace to give scope to the variable names. See http://sourceware.org/gdb/current/onlinedocs/gdb/Selection.html#Selection – msw Jul 04 '11 at 01:36

score 8 · Answer 2 · answered Jun 13 '17 at 11:18

8

This helped me! kill -11 always works for me. 11 is SIGSEGV (invalid memory reference)

answered Jun 13 '17 at 11:18

Aadishri

1,341
2
18
26

score 3 · Answer 3 · answered Jun 14 '17 at 15:51

3

You might also want to take a look at gcore(1) (http://man.freebsd.org/gcore).

answered Jun 14 '17 at 15:51

Edward Tomasz Napierala

1,766
13
12

score 3 · Answer 4 · answered Aug 10 '22 at 21:43

3

I had no luck with kill -QUIT in a particular case. What worked for me is:

gdb --pid <process_id> -ex "set confirm off" -ex generate-core-file -ex q

answered Aug 10 '22 at 21:43

Pedro Gimeno

2,837
1
25
33

1

Use `--batch` and you can skip the `-ex "set confirm off"` and `-ex q`. – Aaron Wright Dec 23 '22 at 17:45

score 1 · Answer 5 · edited Jul 17 '18 at 07:57

1

On sles12.. Below code worked for me :

kill -11

The previous suggestions did not do anything.

edited Jul 17 '18 at 07:57

LuFFy

8,799
10
41
59

answered Jul 17 '18 at 07:25

Marvin

11
1

Maik Ro · Answer 6 · 2021-11-01T12:32:24.010

0

You could also try kill -BUS <pid> which worked for me (though on ubuntu 20.04)

edited Nov 01 '21 at 12:32

answered Nov 01 '21 at 12:26

Maik Ro

320
1
11

Force a core to dump from an active, normally running program on FreeBSD

6 Answers6

Linked