[RFC] tracing: change syscall number type in struct syscall_trace_*

  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:

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:

  10488         if (is_tracepoint || is_syscall_tp) {
  10489                 int off = trace_event_get_offsets(event->tp_event);
  10490
  10491                 if (prog->aux->max_ctx_offset > off)
  10492                         return -EACCES;
  10493         }

This patch changes the type of nr member in syscall_trace_* structs to
be long so that "args" offset is equal to that in struct
trace_event_raw_sys_enter.

Signed-off-by: Artem Savkov <asavkov@redhat.com>
---
 kernel/trace/trace.h          | 4 ++--
 kernel/trace/trace_syscalls.c | 7 ++++---
 2 files changed, 6 insertions(+), 5 deletions(-)
  

On Mon, Oct 2, 2023 at 6:53 AM Artem Savkov <asavkov@redhat.com> wrote:
>
> linux-rt-devel tree contains a patch that adds an extra member to struct

can you please point to the patch itself that makes that change?

> 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:
>
> 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:
>
>   10488         if (is_tracepoint || is_syscall_tp) {
>   10489                 int off = trace_event_get_offsets(event->tp_event);
>   10490
>   10491                 if (prog->aux->max_ctx_offset > off)
>   10492                         return -EACCES;
>   10493         }
>
> This patch changes the type of nr member in syscall_trace_* structs to
> be long so that "args" offset is equal to that in struct
> trace_event_raw_sys_enter.
>
> Signed-off-by: Artem Savkov <asavkov@redhat.com>
> ---
>  kernel/trace/trace.h          | 4 ++--
>  kernel/trace/trace_syscalls.c | 7 ++++---
>  2 files changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> index 77debe53f07cf..cd1d24df85364 100644
> --- a/kernel/trace/trace.h
> +++ b/kernel/trace/trace.h
> @@ -135,13 +135,13 @@ enum trace_type {
>   */
>  struct syscall_trace_enter {
>         struct trace_entry      ent;
> -       int                     nr;
> +       long                    nr;
>         unsigned long           args[];
>  };
>
>  struct syscall_trace_exit {
>         struct trace_entry      ent;
> -       int                     nr;
> +       long                    nr;
>         long                    ret;
>  };
>
> diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
> index de753403cdafb..c26939119f2e4 100644
> --- a/kernel/trace/trace_syscalls.c
> +++ b/kernel/trace/trace_syscalls.c
> @@ -101,7 +101,7 @@ find_syscall_meta(unsigned long syscall)
>         return NULL;
>  }
>
> -static struct syscall_metadata *syscall_nr_to_meta(int nr)
> +static struct syscall_metadata *syscall_nr_to_meta(long nr)
>  {
>         if (IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR))
>                 return xa_load(&syscalls_metadata_sparse, (unsigned long)nr);
> @@ -132,7 +132,8 @@ print_syscall_enter(struct trace_iterator *iter, int flags,
>         struct trace_entry *ent = iter->ent;
>         struct syscall_trace_enter *trace;
>         struct syscall_metadata *entry;
> -       int i, syscall;
> +       int i;
> +       long syscall;
>
>         trace = (typeof(trace))ent;
>         syscall = trace->nr;
> @@ -177,7 +178,7 @@ print_syscall_exit(struct trace_iterator *iter, int flags,
>         struct trace_seq *s = &iter->seq;
>         struct trace_entry *ent = iter->ent;
>         struct syscall_trace_exit *trace;
> -       int syscall;
> +       long syscall;
>         struct syscall_metadata *entry;
>
>         trace = (typeof(trace))ent;
> --
> 2.41.0
>
>
  

On Mon,  2 Oct 2023 15:52:42 +0200
Artem Savkov <asavkov@redhat.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).


> 
> 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.

-- Steve


>   10490
>   10491                 if (prog->aux->max_ctx_offset > off)
>   10492                         return -EACCES;
>   10493         }
> 
> This patch changes the type of nr member in syscall_trace_* structs to
> be long so that "args" offset is equal to that in struct
> trace_event_raw_sys_enter.
>
  

On Tue, Oct 03, 2023 at 03:11:15PM -0700, Andrii Nakryiko wrote:
> On Mon, Oct 2, 2023 at 6:53 AM Artem Savkov <asavkov@redhat.com> wrote:
> >
> > linux-rt-devel tree contains a patch that adds an extra member to struct
> 
> can you please point to the patch itself that makes that change?

Of course, some context would be useful. The patch in question is b1773eac3f29c
("sched: Add support for lazy preemption") from rt-devel tree [0]. It came up
a couple of times before: [1] [2] [3] [4].

[0] https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git/commit/?id=b1773eac3f29cbdcdfd16e0339f1a164066e9f71
[1] https://lore.kernel.org/linux-rt-users/20200221153541.681468-1-jolsa@kernel.org/t/#u
[2] https://github.com/iovisor/bpftrace/commit/a2e3d5dbc03ceb49b776cf5602d31896158844a7
[3] https://lore.kernel.org/bpf/xunyjzy64q9b.fsf@redhat.com/t/#u
[4] https://lore.kernel.org/bpf/20230727150647.397626-1-ykaliuta@redhat.com/t/#u

> > 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:
> >
> > 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:
> >
> >   10488         if (is_tracepoint || is_syscall_tp) {
> >   10489                 int off = trace_event_get_offsets(event->tp_event);
> >   10490
> >   10491                 if (prog->aux->max_ctx_offset > off)
> >   10492                         return -EACCES;
> >   10493         }
> >
> > This patch changes the type of nr member in syscall_trace_* structs to
> > be long so that "args" offset is equal to that in struct
> > trace_event_raw_sys_enter.
> >
> > Signed-off-by: Artem Savkov <asavkov@redhat.com>
> > ---
> >  kernel/trace/trace.h          | 4 ++--
> >  kernel/trace/trace_syscalls.c | 7 ++++---
> >  2 files changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
> > index 77debe53f07cf..cd1d24df85364 100644
> > --- a/kernel/trace/trace.h
> > +++ b/kernel/trace/trace.h
> > @@ -135,13 +135,13 @@ enum trace_type {
> >   */
> >  struct syscall_trace_enter {
> >         struct trace_entry      ent;
> > -       int                     nr;
> > +       long                    nr;
> >         unsigned long           args[];
> >  };
> >
> >  struct syscall_trace_exit {
> >         struct trace_entry      ent;
> > -       int                     nr;
> > +       long                    nr;
> >         long                    ret;
> >  };
> >
> > diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
> > index de753403cdafb..c26939119f2e4 100644
> > --- a/kernel/trace/trace_syscalls.c
> > +++ b/kernel/trace/trace_syscalls.c
> > @@ -101,7 +101,7 @@ find_syscall_meta(unsigned long syscall)
> >         return NULL;
> >  }
> >
> > -static struct syscall_metadata *syscall_nr_to_meta(int nr)
> > +static struct syscall_metadata *syscall_nr_to_meta(long nr)
> >  {
> >         if (IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR))
> >                 return xa_load(&syscalls_metadata_sparse, (unsigned long)nr);
> > @@ -132,7 +132,8 @@ print_syscall_enter(struct trace_iterator *iter, int flags,
> >         struct trace_entry *ent = iter->ent;
> >         struct syscall_trace_enter *trace;
> >         struct syscall_metadata *entry;
> > -       int i, syscall;
> > +       int i;
> > +       long syscall;
> >
> >         trace = (typeof(trace))ent;
> >         syscall = trace->nr;
> > @@ -177,7 +178,7 @@ print_syscall_exit(struct trace_iterator *iter, int flags,
> >         struct trace_seq *s = &iter->seq;
> >         struct trace_entry *ent = iter->ent;
> >         struct syscall_trace_exit *trace;
> > -       int syscall;
> > +       long syscall;
> >         struct syscall_metadata *entry;
> >
> >         trace = (typeof(trace))ent;
> > --
> > 2.41.0
> >
> >
>
  

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@redhat.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.
  

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@redhat.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?