Re: [PATCH v2 1/1] arm64: Do not forget syscall when starting a new thread.

From: Francis Laniel <flaniel@linux.microsoft.com>
To: Will Deacon <will@kernel.org>
Cc: linux-trace-devel@vger.kernel.org,
	James Morse <james.morse@arm.com>,
	Daniel Kiss <daniel.kiss@arm.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Mark Brown <broonie@kernel.org>,
	Peter Collingbourne <pcc@google.com>,
	Kees Cook <keescook@chromium.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Christophe Leroy <christophe.leroy@csgroup.eu>,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 1/1] arm64: Do not forget syscall when starting a new thread.
Date: Fri, 01 Jul 2022 14:10:44 +0200	[thread overview]
Message-ID: <12007198.O9o76ZdvQC@pwmachine> (raw)
In-Reply-To: <20220701113631.GC28070@willie-the-truck>

Hi.

Le vendredi 1 juillet 2022, 13:36:32 CEST Will Deacon a écrit :
> On Thu, Jun 30, 2022 at 07:16:44PM +0200, Francis Laniel wrote:
> > Hi.
> > 
> > Le mardi 28 juin 2022, 21:26:32 CEST Francis Laniel a écrit :
> > > Hi.
> > > 
> > > Le mardi 28 juin 2022, 15:58:35 CEST Will Deacon a écrit :
> > > > On Wed, Jun 08, 2022 at 05:24:46PM +0100, Francis Laniel wrote:
> > > > > This patch enables exeve*() to be traced with
> > > > > syscalls:sys_exit_execve
> > > > > tracepoint.
> > > > > Previous to it, by calling forget_syscall(), this tracepoint would
> > > > > not
> > > > > print its information as syscall is -1.
> > > > > So, this patch removes call to forget_syscall() and set
> > > > > regs->syscallno
> > > > > to its previous value.
> > > > > 
> > > > > Signed-off-by: Francis Laniel <flaniel@linux.microsoft.com>
> > > > > ---
> > > > > 
> > > > >  arch/arm64/include/asm/processor.h | 3 ++-
> > > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/arch/arm64/include/asm/processor.h
> > > > > b/arch/arm64/include/asm/processor.h index
> > > > > 9e58749db21d..86eb0bfe3b38
> > > > > 100644
> > > > > --- a/arch/arm64/include/asm/processor.h
> > > > > +++ b/arch/arm64/include/asm/processor.h
> > > > > @@ -272,8 +272,9 @@ void tls_preserve_current_state(void);
> > > > > 
> > > > >  static inline void start_thread_common(struct pt_regs *regs,
> > > > >  unsigned
> > > > >  long pc) {
> > > > > 
> > > > > +	s32 previous_syscall = regs->syscallno;
> > > > > 
> > > > >  	memset(regs, 0, sizeof(*regs));
> > > > > 
> > > > > -	forget_syscall(regs);
> > > > > +	regs->syscallno = previous_syscall;
> > > > 
> > > > I'm still unsure about this. Even if we preserve the syscall number
> > > > here,
> > > > won't all the arguments be reported as 0?
> > > 
> > > I am not really sure what you meant about arguments, can you please
> > > precise
> > > between command line arguments (ls -al) and syscall arguments (argp,
> > > envp,
> > > etc.)?
> > > Indeed, if my understanding is correct syscall arguments are showed by
> > > sys_enter_* while sys_exit_* only reports the syscall return code.
> > > 
> > > Regarding the return code I think the value is correct as it is used in
> > > syscall_trace_exit() but set in invoke_syscall() after the syscall
> > > finishes
> > > [1, 2].
> > > The comparison of arm64 and amd64 output also shows no difference:
> > > # amd64
> > > ls 435739 [002] 24689.292479:  syscalls:sys_exit_execve: 0x0
> > > 
> > >             7fc43732e100 _start+0x0
> > >             (/usr/lib/x86_64-linux-gnu/ld-2.31.so)
> > > 
> > > # arm64
> > > ls   266 [000]    34.708444:  syscalls:sys_exit_execve: 0x0
> > > 
> > >                     1140 [unknown]
> > >                     (/usr/lib/aarch64-linux-gnu/ld-2.31.so)
> > > > 
> > > > I also looked quickly at the 32-bit arch/arm/ code and it looks like
> > > > the
> > > > same behaviour exists there (module CONFIG_BINFMT_ELF_FDPIC).
> > 
> > I tested arm32 and it is not affected (even though I did not have
> > CONFIG_BINFMT_ELF_FDPIC set).
> > 
> > Here is ftrace output for arm64 without this patch:
> >             bash-316     [000] .....    72.167342: sys_execve(filename:
> > aaaaf9bbcd30, argv: aaaaf9bb54f0, envp: aaaaf9a7d9b0)
> > 
> > Here is the output for arm64 with this patch:
> >              cat-313     [000] .....   417.926073: sys_execve(filename:
> > aaaaee7ce9f0, argv: aaaaee7833a0, envp: aaaaee6a69b0)
> > 
> >              cat-313     [000] .....   417.939619: sys_execve -> 0x0
> > 
> > And here is output for arm32:
> >              cat-254     [000] .....   127.804128: sys_execve(filename:
> >              5bff18,
> > 
> > argv: 53bb00, envp: 5543a8)
> > 
> >              cat-254     [000] .....   127.809142: sys_execve -> 0x0
> > 
> > From the above, the arm32 output seems correct even though:
> > # CONFIG_BINFMT_ELF_FDPIC is not set
> > 
> > After some debugging, I realized that arm32 syscall_get_nr() uses
> > abi_syscall to get the syscall number and not a register (I guess
> > abi_syscall was set to value of R7 before) [1].
> > So the fact that regs->uregs are memset'ed to 0 is not a problem.
> 
> Thanks for confirming this, I'll go ahead and queue your patch and let's
> hope nothing breaks :)

You are welcome!
If there any problem, feel free to ping and I will try to handle them.

> Will

Best regards.