Jacopo / legdbs

Ptrace-less breakpoints. Useful if ptrace() is prohibited but signals work.

Ptrace-less breakpoints, a.k.a. poor-mans detours.

Mainly to serve LegitBS' idea that ptrace is Bad(TM). (But works for any case where ptrace is disabled / undesirable.)

All it needs is an opcode that causes a fault signal to be sent, and freedom to catch it. Put them in legdbs.c.

Limitations:

• Cannot handle jumping in the middle of a multi-byte breakpoint
• No multithreading (may want to use SIGSTOP to emulate mutexes)
• Since it is mainly useful for static programs, you cannot use the regular libc. Weird stuff is necessary to use C. See my inline-notlibc repo.

TODO:

• Needs manual info on the next instruction (use print_set_breakpoint_line.py)
• More testing :)

Usage (see demo_handler.c)

void foo(int sig, siginfo_t *info, struct ucontext *) { ... }
install_signal_handler();
set_breakpoint(where_to_break, postbreak_instruction, foo);
set_breakpoint(other, other_postbreak_instruction, foo);

Note that set_breakpoint specifies TWO separate breakpoints

• The "wanted" one, disabled after hitting it.
• The "service" one used to set the "wanted" breakpoint again. It must be set at the first instruction after addr that does NOT overlap with the breakpoint opcode bytes.

If the original code has instructions A, B, C, D:

orig.  want.  serv.
 A_1    BRK    A_1   <- addr
 A_2    BRK    A_2
 B_1    BRK    B_1
 B_2    B_2    B_2   <- addr + sizeof(brk_opcode)
 B_3    B_3    B_3
 C_1    C_1    BRK   <- postbreak_instruction
 C_2    C_2    BRK
 D_1    D_1    BRK
 D_2    D_2    D_2   <- postbreak_instruction + sizeof(brk_opcode)
 ...    ...    ...

After hitting the breakpoint at addr, execution will resume as in serv., so that the code behaves as it was supposed to. However, we need to stop as soon as possible to restore the breakpoint, hence the service breakpoint.

print_set_breakpoint_line.py program addr sizeof(brk_opcode)

Helps you find postbreak_instruction.

Note: will exit(109) on failure.