Re: [PATCH v3 3/4] arm64: implement live patching

[Miroslav Benes <mbenes@suse.cz>]

On Mon, 1 Oct 2018, Torsten Duwe wrote:

> Based on ftrace with regs, do the usual thing. Also allocate a
> task flag for whatever consistency handling will be used.
> Watch out for interactions with the graph tracer.

Similar to what Mark wrote about 2/4, I'd appreciate a better commit log. 
Could you explain traditional "what/why/how", please?
 
> Signed-off-by: Torsten Duwe <duwe@suse.de>
> 
> --- a/arch/arm64/Kconfig
> +++ b/arch/arm64/Kconfig
> @@ -119,6 +119,7 @@ config ARM64
>  	select HAVE_GENERIC_DMA_COHERENT
>  	select HAVE_HW_BREAKPOINT if PERF_EVENTS
>  	select HAVE_IRQ_TIME_ACCOUNTING
> +	select HAVE_LIVEPATCH
>  	select HAVE_MEMBLOCK
>  	select HAVE_MEMBLOCK_NODE_MAP if NUMA
>  	select HAVE_NMI
> @@ -1349,4 +1350,6 @@ if CRYPTO
>  source "arch/arm64/crypto/Kconfig"
>  endif
>  
> +source "kernel/livepatch/Kconfig"
> +
>  source "lib/Kconfig"
> --- a/arch/arm64/include/asm/thread_info.h
> +++ b/arch/arm64/include/asm/thread_info.h
> @@ -76,6 +76,7 @@ void arch_release_task_struct(struct tas
>  #define TIF_FOREIGN_FPSTATE	3	/* CPU's FP state is not current's */
>  #define TIF_UPROBE		4	/* uprobe breakpoint or singlestep */
>  #define TIF_FSCHECK		5	/* Check FS is USER_DS on return */
> +#define TIF_PATCH_PENDING	6
>  #define TIF_NOHZ		7
>  #define TIF_SYSCALL_TRACE	8
>  #define TIF_SYSCALL_AUDIT	9
> @@ -94,6 +95,7 @@ void arch_release_task_struct(struct tas
>  #define _TIF_NEED_RESCHED	(1 << TIF_NEED_RESCHED)
>  #define _TIF_NOTIFY_RESUME	(1 << TIF_NOTIFY_RESUME)
>  #define _TIF_FOREIGN_FPSTATE	(1 << TIF_FOREIGN_FPSTATE)
> +#define _TIF_PATCH_PENDING	(1 << TIF_PATCH_PENDING)
>  #define _TIF_NOHZ		(1 << TIF_NOHZ)
>  #define _TIF_SYSCALL_TRACE	(1 << TIF_SYSCALL_TRACE)
>  #define _TIF_SYSCALL_AUDIT	(1 << TIF_SYSCALL_AUDIT)
> @@ -106,7 +108,8 @@ void arch_release_task_struct(struct tas
>  
>  #define _TIF_WORK_MASK		(_TIF_NEED_RESCHED | _TIF_SIGPENDING | \
>  				 _TIF_NOTIFY_RESUME | _TIF_FOREIGN_FPSTATE | \
> -				 _TIF_UPROBE | _TIF_FSCHECK)
> +				 _TIF_UPROBE | _TIF_FSCHECK | \
> +				 _TIF_PATCH_PENDING)

Could you add a note to the changelog what this means? My ability to read 
arm64 entry.S is very limited, but I can see that _TIF_WORK_MASK is 
process in a syscall exit and irq return paths. That's good. It is also 
called (via "b ret_to_user") in a couple of different places (el0_* 
labels). I guess those are returns from exception handling. A comment 
about it in the changelog would be appreciated.
  
>  #define _TIF_SYSCALL_WORK	(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
>  				 _TIF_SYSCALL_TRACEPOINT | _TIF_SECCOMP | \
> --- /dev/null
> +++ b/arch/arm64/include/asm/livepatch.h
> @@ -0,0 +1,36 @@
> +/* SPDX-License-Identifier: GPL-2.0
> + *
> + * livepatch.h - arm64-specific Kernel Live Patching Core
> + *
> + * Copyright (C) 2016,2018 SUSE
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version 2
> + * of the License, or (at your option) any later version.
> + *
> + * This program is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program; if not, see <http://www.gnu.org/licenses/>.
> + */
> +#ifndef _ASM_ARM64_LIVEPATCH_H
> +#define _ASM_ARM64_LIVEPATCH_H
> +
> +#include <linux/module.h>
> +#include <linux/ftrace.h>

I think that only

#include <asm/ptrace.h>

is in fact needed, because of "struct pt_regs".


Ad relocations. I checked that everything in struct mod_arch_specific 
stays after the module is load. Both core and init get SHF_ALLOC set 
(mod->arch.core.plt->sh_flags in module_frob_arch_sections(). It is 
important because apply_relocate_add() may use those sections 
through module_emit_plt_entry() call.

ftrace_trampoline section gets SHF_ALLOC as well. Btw, don't you want to 
do something similar for new ftrace_regs_trampoline in 
module_frob_arch_sections()? Perhaps even edit 
arch/arm64/kernel/module.lds? I'm completely clueless here, so it may be 
ok. And it applies to 2/4 patch.

The last thing is count_plts() function called from 
module_frob_arch_sections(). It needed to be changed in 2016 as well. See 
Jessica's patch 
(20160713001113.GA30925@packer-debian-8-amd64.digitalocean.com). The logic 
in the function has changed since then. If I am not mistaken, 
count_plts() is fine as it is right now. It does not consider SHN_UNDEF 
anymore, it looks at the destination section (where a symbol should 
resolved to) only.

Jessica, could you doublecheck, please?

Torsten, please doublecheck all this as well and add a comment about all 
this to the changelog, so we don't have to analyze it from the beginning 
again in the future.

Thanks,
Miroslav