debugging C program with gdb

Question

I'm trying to test a scheduler that I wrote. I schedule two processes - both are infinite while loops (just while(1) statements). When I run the program sometimes it segfaults after like ten seconds (sometimes 5 sec, sometimes 15 or more). Sometimes it doesn't segfault at all and runs as expected. I have a log file which shows me that both processes are scheduled as expected before the segfault occurs. I'm trying to debug the errors using gdb but it's not being very helpful. Here's what I got with backtrace:

#0  0x00007ffff7ff1000 in ?? ()
#1  0x000000000000002b in ?? ()
#2  0x00007ffff78b984a in new_do_write () from /lib64/libc.so.6
#3  0x000000000061e3d0 in ?? ()
#4  0x0000000000000000 in ?? ()

I don't really understand #2.

I think this may be a stack overflow related error. However, I only malloc twice in the whole process - both times when I'm setting up the two processes, I malloc a pcb block in the pcb table I wrote. Has anyone run into similar issues before? Could this be something with how I'm setting/swapping the contexts in the scheduler? Why does it segfault sometimes, and sometimes not?

You should compile your code with debug information (-g in gcc), then your stack trace will make more sense — onon15
– onon15, Commented Nov 22, 2012 at 21:20
Seeing this obviously smashed stack I suspect -Wall would help also. — alk
– alk, Commented Nov 22, 2012 at 21:26
Instead of describing your code, you should post it here. It's easier to comment on concrete code than on prose. — Olaf Dietsche
– Olaf Dietsche, Commented Nov 22, 2012 at 21:28
Also, please don't cast malloc return value in C, it can hide at least one insidious error that will bite you at some point. And it clutters your code. C is perfectly capable of implicitly converting void* to any other pointer type. — paxdiablo
– paxdiablo, Commented Nov 22, 2012 at 22:06

Employed Russian · Accepted Answer · 2012-11-23 02:03:02Z

1

You didn't tell how you obtained the stack trace that you show in the question.

It is very likely that the stack trace is bogus not because the stack is corrupt, but because you've invoked GDB incorrectly, e.g. specified wrong executable when attaching the process or examining core dump.

One common mistake is to build the executable with -O2 (let's call this executable E1), then rebuild it with -g (let's call this E2) and try to analyze core of live process that is running E1 giving GDB E2 as the symbol file.

Don't do that, it doesn't work and isn't expected to work.

answered Nov 23, 2012 at 2:03

Employed Russian

217k36 gold badges324 silver badges396 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

paxdiablo · Accepted Answer · 2012-11-22 21:42:07Z

0

Since your stack seems corrupted, you're probably correct that you have a stack buffer overflow somewhere. Without the code, it's a little difficult to tell.

But this has nothing to do with your malloc calls. Overflowing dynamically allocated buffers would corrupt the heap, not the stack.

Whay you'll probably need to be looking at is local variables that aren't big enough for the data you're trying to copy in to them, like:

char xyzzy[5];
strcpy (xyzzy, "this is a bad idea";

Or passing a buffer (again, most likely on the stack) to a system call that writes more data to it than you provide for.

They're the most likely causes though theoretically, of course, any undefined behaviour on your part could cause this. If the solution is not evident based on this answer, you'll probably need to post the code that caused it. Try to ensure you trim it down as much as possible when you do that so that it's the shortest complete program that exhibits the bug.

Often you'll find by doing that, the problem becomes evident :-)

edited Nov 22, 2012 at 21:42

answered Nov 22, 2012 at 21:34

paxdiablo

888k243 gold badges1.6k silver badges2k bronze badges

2 Comments

user1710707 Over a year ago

Hey I commented out some code (which sets up the second process). It's above. Can you help me figure out whats wrong there? thanks

paxdiablo Over a year ago

@user1710707, not a lot of code to work with :-) However, it's likely to be the latter issue in my answer here, passing invalid params to a call. Exec-like calls generally expect an extra element at the end of the argv array, set to null. Not sure if that's the case with p_spawn_internal, but you should look into that.

Collectives™ on Stack Overflow

debugging C program with gdb

2 Answers 2

Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related