netdev - Re: [RFC PATCH] tracing: change syscall number type in struct syscall_trace

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]

Message-ID: <20231005123413.GA488417@alecto.usersys.redhat.com>
Date: Thu, 5 Oct 2023 14:34:13 +0200
From: Artem Savkov <asavkov@...hat.com>
To: Alexei Starovoitov <ast@...nel.org>,
	Andrii Nakryiko <andrii@...nel.org>
Cc: Masami Hiramatsu <mhiramat@...nel.org>, linux-kernel@...r.kernel.org,
	linux-trace-kernel@...r.kernel.org,
	Steven Rostedt <rostedt@...dmis.org>,
	Daniel Borkmann <daniel@...earbox.net>, bpf@...r.kernel.org,
	netdev@...r.kernel.org, Thomas Gleixner <tglx@...utronix.de>,
	linux-rt-users@...r.kernel.org, Jiri Olsa <jolsa@...nel.org>
Subject: Re: [RFC PATCH] tracing: change syscall number type in struct
 syscall_trace_*

On Wed, Oct 04, 2023 at 02:55:47PM +0200, Artem Savkov wrote:
> On Tue, Oct 03, 2023 at 09:38:44PM -0400, Steven Rostedt wrote:
> > On Mon,  2 Oct 2023 15:52:42 +0200
> > Artem Savkov <asavkov@...hat.com> wrote:
> > 
> > > linux-rt-devel tree contains a patch that adds an extra member to struct
> > > trace_entry. This causes the offset of args field in struct
> > > trace_event_raw_sys_enter be different from the one in struct
> > > syscall_trace_enter:
> > 
> > This patch looks like it's fixing the symptom and not the issue. No code
> > should rely on the two event structures to be related. That's an unwanted
> > coupling, that will likely cause issues down the road (like the RT patch
> > you mentioned).
> 
> I agree, but I didn't see a better solution and that was my way of
> starting conversation, thus the RFC.
> 
> > > 
> > > struct trace_event_raw_sys_enter {
> > >         struct trace_entry         ent;                  /*     0    12 */
> > > 
> > >         /* XXX last struct has 3 bytes of padding */
> > >         /* XXX 4 bytes hole, try to pack */
> > > 
> > >         long int                   id;                   /*    16     8 */
> > >         long unsigned int          args[6];              /*    24    48 */
> > >         /* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
> > >         char                       __data[];             /*    72     0 */
> > > 
> > >         /* size: 72, cachelines: 2, members: 4 */
> > >         /* sum members: 68, holes: 1, sum holes: 4 */
> > >         /* paddings: 1, sum paddings: 3 */
> > >         /* last cacheline: 8 bytes */
> > > };
> > > 
> > > struct syscall_trace_enter {
> > >         struct trace_entry         ent;                  /*     0    12 */
> > > 
> > >         /* XXX last struct has 3 bytes of padding */
> > > 
> > >         int                        nr;                   /*    12     4 */
> > >         long unsigned int          args[];               /*    16     0 */
> > > 
> > >         /* size: 16, cachelines: 1, members: 3 */
> > >         /* paddings: 1, sum paddings: 3 */
> > >         /* last cacheline: 16 bytes */
> > > };
> > > 
> > > This, in turn, causes perf_event_set_bpf_prog() fail while running bpf
> > > test_profiler testcase because max_ctx_offset is calculated based on the
> > > former struct, while off on the latter:
> > 
> > The above appears to be pointing to the real bug. The "is calculated based
> > on the former struct while off on the latter" Why are the two being used
> > together? They are supposed to be *unrelated*!
> > 
> > 
> > > 
> > >   10488         if (is_tracepoint || is_syscall_tp) {
> > >   10489                 int off = trace_event_get_offsets(event->tp_event);
> > 
> > So basically this is clumping together the raw_syscalls with the syscalls
> > events as if they are the same. But the are not. They are created
> > differently. It's basically like using one structure to get the offsets of
> > another structure. That would be a bug anyplace else in the kernel. Sounds
> > like it's a bug here too.
> > 
> > I think the issue is with this code, not the tracing code.
> > 
> > We could expose the struct syscall_trace_enter and syscall_trace_exit if
> > the offsets to those are needed.
> 
> I don't think we need syscall_trace_* offsets, looks like
> trace_event_get_offsets() should return offset trace_event_raw_sys_enter
> instead. I am still trying to figure out how all of this works together.
> Maybe Alexei or Andrii have more context here.

Turns out it is even more confusing. The tests dereference the context as
struct trace_event_raw_sys_enter so bpf verifier sets max_ctx_offset
based on that, then perf_event_set_bpf_prog() checks this offset against
the one in struct syscall_trace_enter, but what bpf prog really gets is
a pointer to struct syscall_tp_t from kernel/trace/trace_syscalls.c.

I don't know the history behind these decisions, but should the tests
dereference context as struct syscall_trace_enter instead and struct
syscall_tp_t be changed to have syscall_nr as int?

-- 
 Artem