URL:
<>
Summary: /hurd/crash doesn't work
Project: The GNU Hurd
Submitted by: tschwinge
Submitted on: Friday 08/04/06 at 16:14
Category: Hurd Servers
Severity: 3 - Normal
Priority: 5 - Normal
Item Group: None
Status: None
Privacy: Public
Assigned to: None
Name:
Email:
/Closed:
Reproducibility: Every Time
Size (loc): None
Planned Release: None
Effort: 0.00
Details:
`/hurd/crash ' is usually running as a server on
`/servers/crash-suspend' and may be linked to from `/servers/crash' to
actually use it.
It's meant to suspend a crashed process, i.e. keep it in a frozen state in
memory (as opposed to dumping a core file) to make it attachable and
debuggable by gdb.
Unfortunately this doesn't work:
#v+
thomas@leibniz:/var/tmp $ cat crash.c
int main (void) { return *(int *) 0; }
thomas@leibniz:/var/tmp $ gcc crash.c
thomas@leibniz:/var/tmp $ ./a.out
Segmentation fault (core dumped)
thomas@leibniz:/var/tmp $ ls -l core
-rw 1 thomas root 16912384 Jun 15 20:43 core
thomas@leibniz:/var/tmp $ ls -l /servers/crash
lrwxr-xr-x 1 root root 13 Jun 15 20:42 /servers/crash -crash-suspend
thomas@leibniz:/var/tmp $ showtrans /servers/crash-suspend
/hurd/crash
#v-
According to Roland it has been working ``in the past'' (``Long before core
dumping ever worked at all.'') and he also added that ``the crash server is
one of the easier ones to debug with gdb.''
Reply to this item at:
<>
Message sent via/by Savannah
http://savannah.gnu.org/
Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org

Follow-up Comment #1, bug #17319 (project hurd):

Roland commented on bug-hurd:

Make sure it's not just a bug in the option parsing or something. Check
with fsysopts /servers/crash. I note that its trivfs_append_args doesn't
show the setting. You might fix that.

Try fsysopts and see if that changes the behavior.
If it does, then the problem is in deciding whether it's an orphan.

Then debug the server and see if crash_how is set correctly. Because
nothing else (that's not dying) depends on it, it is easy to use gdb on
crash.
Break in S_crash_dump_task and see what's going on.

Reply to this item at:

<>

Message sent via/by Savannah
http://savannah.gnu.org/

Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org

Follow-up Comment #2, bug #17319 (project hurd):

Here is a patch that makes the crash server support in
fsysopts.

Incidentally I couldn't set the fsysopts as a regular user. Perhaps this is
a bug also.

Additional Item Attachment:

File name:
Size:0 KB

<>

Reply to this item at:

<>

Message sent via/by Savannah
http://savannah.gnu.org/

Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org

Follow-up Comment #3, bug #17319 (project hurd):

I tried the patched crash server, and it didn't quite work as expected.

I retried Thomas' example:

#v+
thomas@leibniz:/var/tmp $ cat crash.c
int main (void) { return *(int *) 0; }
thomas@leibniz:/var/tmp $ gcc crash.c
thomas@leibniz:/var/tmp $ ./a.out
Segmentation fault (core dumped)
thomas@leibniz:/var/tmp $ ls -l core
-rw 1 thomas root 16912384 Jun 15 20:43 core
thomas@leibniz:/var/tmp $ ls -l /servers/crash
lrwxr-xr-x 1 root root 13 Jun 15 20:42 /servers/crash -crash-suspend
thomas@leibniz:/var/tmp $ showtrans /servers/crash-suspend
/hurd/crash
#v-
But I added:

$ fsysopts /servers/crash-suspend

It didn't change the behaviour -- it still dumped core. but,

$ fsysopts /servers/crash-suspend

did. It suspended the task.

I then tried to attach to the crash server using gdb, but it said that the
pid didn't have any children.

Reply to this item at:

<>

Message sent via/by Savannah
http://savannah.gnu.org/

Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org

Follow-up Comment #4, bug #17319 (project hurd):

I had some success with the patched crash-server and I thought I should share
it here.

After doing this:

# ln -sf crash-suspend /servers/crash
# fsysopts /servers/crash-suspend

I could successfully attach to crashed suspended processes using gdb.

I also tried , and , but no
luck.

Fun stuff!

Reply to this item at:

<>

Message sent via/by Savannah
http://savannah.gnu.org/

Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org

Follow-up Comment #5, bug #17319 (project hurd):

I installed file #10477
(`') after a
tiny tweak.

#v+
2006-11-27 Ben Asselstine <benasselstine (AT) canada (DOT) com>
Thomas Schwinge <tschwinge (AT) gnu (DOT) org>

* crash.c (trivfs_append_args): Handle CRASHRPHANS_HW.
#v-

This does -- of course -- not yet fix this bug report's main issue.

Reply to this item at:

<>

Message sent via/by Savannah
http://savannah.gnu.org/

Bug-hurd mailing list
Bug-hurd (AT) gnu (DOT) org