Uprobe: Bug(?) when probing small binaries
Hi Oleg,
I'm observing a bug in the uprobe infrastructure. When target binary is quite small, uprobe replaces 'trap' instruction at two different places. Ex,
Simple test.c that loops for 100 seconds:
void main()
{
int i = 0;
while (i++ != 100) {
printf("hi: %d", i);
sleep(1);
}
}
Add a probe on function main() in test.c
$ perf probe -x ./a.out main Added new event: probe_a:main (on main in /home/ravi/a.out)
Start the target program and pause it.
$ gdb --args ./a.out (gdb) r main: 1 main: 2 ^C
(gdb) disassemble main
0x000000001000069c <+8>: mflr r0
(gdb) x/w 0x1001069c
0x1001069c: 2080899750
Now enable the probe:
# echo 1 > events/probe_a/main/enable
Check probed instruction:
(gdb) disassemble main
0x000000001000069c <+8>: trap
*Bug*:
(gdb) x/w 0x1001069c
0x1001069c: 2145386504
In short, when it replaces the probe instruction, it does some corruption in the readonly vma. This seems to be a bug.
How did I get the other address 0x1001069c?I found build_map_info() returns these two vmas for the single probe:
10000000-10010000 r-xp 00000000 08:05 67325595 /home/ravi/a.out 10010000-10020000 r--p 00000000 08:05 67325595 /home/ravi/a.out
and thusregister_for_each_vmas() calls install_breakpoint() on both of thesevmas with different vaddr.
The example is on powerpc but same issue is observed on x86 as well. As, the code is common, it should be reproducible on every architecture.
Also, I don't observe this issue for bigger binaries (maybe for those whose vma spans across multiple pages).
Thanks,
Ravi
Hi Ravi,
Quoted 2 lines · Ravi Bangoria
I'm observing a bug in the uprobe infrastructure.
I am not sure,
Quoted 3 lines
When target binary
is quite small, uprobe replaces 'trap' instruction at two different places. Ex,
but this is actually "the same place" (in the file), just 2 virtual addrs differ,
Quoted 17 lines
(gdb) x/w
0x1001069c
0x1001069c: 2080899750Now enable the probe:
# echo 1 > events/probe_a/main/enableCheck probed instruction:
(gdb) disassemble main
0x000000001000069c<+8>: trap*Bug*:
(gdb) x/w
0x1001069c
0x1001069c: 2145386504
Quoted 8 lines
In short, when it replaces the probe instruction, it does some corruption in the readonly vma. This seems to be a bug.
How did I get the other address 0x1001069c?I found
build_map_info()returns these two vmas for the single probe:10000000-10010000 r-xp 00000000 08:05 67325595
/home/ravi/a.out10010000-10020000 r--p 00000000 08:05 67325595/home/ravi/a.out
please note that these 2 vma's mmap the same region in a.out, so *0x1001069c and *0x1000069c point to the same insn.
Oleg.