[RFC 5/5] uprobes: add global breakpoints

[Peter Zijlstra]

On Tue, 2012-08-07 at 18:12 +0200, Sebastian Andrzej Siewior wrote:
> By setting an uprobe tracepoint, one learns whenever a certain point
> within a program is reached / passed. This is recorded and the
> application continues.
> This patch adds the ability to hold the program once this point has been
> passed and the user may attach to the program via ptrace.
> First, setup a global breakpoint which is very similar to a uprobe trace
> point:
> 
> |echo 'g /home/bigeasy/sample:0x0000044d %ip %ax' > uprobe_events
> 
> This is exactly what uprobe does except that it starts with the letter
> 'g' instead of 'p'.
> 
> Step two is to enable it:
> |echo 1 > events/uprobes/enable
> 
> Lets assume you execute ./sample and the breakpoint is hit. In ps you will
> see:
> |1938 pts/1    t+     0:00 ./sample

This seems particularly dangerous.. suppose you tag a frequent function
(say malloc) and the entire userspace freezes, including your shell.

> Now you can attach gdb via 'gdb -p 1938'. The gdb can now interact with
> the tracee and inspect its registers, its stack, single step, let it
> run…
> In case the process is not of great interest, the user may continue
> without gdb by writting its pid into the uprobe_gp_wakeup file
> 
> |echo 1938 > uprobe_gp_wakeup
> 
> What I miss right now is an interface to tell the user/gdb that there is a
> program that hit a global breakpoint and is waiting for further instructions.
> A "tail -f trace" does not work and may contain also a lot of other
> informations. I've been thinking about a poll()able file which returns pids of
> tasks which are put on hold. Other suggestions? 

I'm not really happy with any of this. I would suggest limiting this
stuff much further, like say only have it affect ptraced
processes/tasks. That way you cannot accidentally freeze the entire
system into oblivion.

GDB (and assorted stuff) can already track an entire process hierarchy
with fork follow stuffs.