linux-kernel - Re: [RFC PATCH -tip 3/3] x86/kprobes,orc: Unwind optprobe trampoline correctly

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 31 Mar 2021 10:57:36 -0500
From:   Josh Poimboeuf <jpoimboe@...hat.com>
To:     Masami Hiramatsu <mhiramat@...nel.org>
Cc:     Ingo Molnar <mingo@...nel.org>, X86 ML <x86@...nel.org>,
        Daniel Xu <dxu@...uu.xyz>, linux-kernel@...r.kernel.org,
        bpf@...r.kernel.org, kuba@...nel.org, mingo@...hat.com,
        ast@...nel.org, tglx@...utronix.de, kernel-team@...com, yhs@...com,
        Steven Rostedt <rostedt@...dmis.org>
Subject: Re: [RFC PATCH -tip 3/3] x86/kprobes,orc: Unwind optprobe trampoline
 correctly

On Wed, Mar 31, 2021 at 02:44:56PM +0900, Masami Hiramatsu wrote:
> +#ifdef CONFIG_UNWINDER_ORC
> +unsigned long recover_optprobe_trampoline(unsigned long addr, unsigned long *sp)
> +{
> +	unsigned long offset, entry, probe_addr;
> +	struct optimized_kprobe *op;
> +	struct orc_entry *orc;
> +
> +	entry = find_kprobe_optinsn_slot_entry(addr);
> +	if (!entry)
> +		return addr;
> +
> +	offset = addr - entry;
> +
> +	/* Decode arg1 and get the optprobe */
> +	op = (void *)extract_set_arg1((void *)(entry + TMPL_MOVE_IDX));
> +	if (!op)
> +		return addr;
> +
> +	probe_addr = (unsigned long)op->kp.addr;
> +
> +	if (offset < TMPL_END_IDX) {
> +		orc = orc_find((unsigned long)optprobe_template_func + offset);
> +		if (!orc || orc->sp_reg != ORC_REG_SP)
> +			return addr;
> +		/*
> +		 * Since optprobe trampoline doesn't push caller on the stack,
> +		 * need to decrement 1 stack entry size
> +		 */
> +		*sp += orc->sp_offset - sizeof(long);
> +		return probe_addr;
> +	} else {
> +		return probe_addr + offset - TMPL_END_IDX;
> +	}
> +}
> +#endif

Hm, I'd like to avoid intertwining kprobes and ORC like this.

ORC unwinds other generated code by assuming the generated code uses a
frame pointer.  Could we do that here?

With CONFIG_FRAME_POINTER, unwinding works because SAVE_REGS_STRING has
ENCODE_FRAME_POINTER, but that's not going to work for ORC.

Instead of these patches, can we 'push %rbp; mov %rsp, %rbp' at the
beginning of the template and 'pop %rbp' at the end?

I guess SAVE_REGS_STRING would need to be smart enough to push the
original saved version of %rbp.  Of course then that breaks the
kretprobe_trampoline() usage, so it may need to be a separate macro.

[ Or make the same change to kretprobe_trampoline().  Then the other
  patch set wouldn't be needed either ;-) ]

Of course the downside is, when you get an interrupt during the frame
pointer setup, unwinding is broken.  But I think that's acceptable for
generated code.  We've lived with that limitation for all code, with
CONFIG_FRAME_POINTER, for many years.

Eventually we may want to have a way to register generated code (and the
ORC for it).

-- 
Josh