x86: Include asm/ptrace.h in syscall_wrapper header

Message ID 20221018122708.823792-1-jolsa@kernel.org
State New
Headers
Series x86: Include asm/ptrace.h in syscall_wrapper header |

Commit Message

Jiri Olsa Oct. 18, 2022, 12:27 p.m. UTC
  From: Jiri Olsa <olsajiri@gmail.com>

With just the forward declaration of the 'struct pt_regs' in
syscall_wrapper.h, the syscall stub functions:

  __[x64|ia32]_sys_*(struct pt_regs *regs)

will have different definition of 'regs' argument in BTF data
based on which object file they are defined in.

If the syscall's object includes 'struct pt_regs' definition,
the BTF argument data will point to 'struct pt_regs' record,
like:

  [226] STRUCT 'pt_regs' size=168 vlen=21
         'r15' type_id=1 bits_offset=0
         'r14' type_id=1 bits_offset=64
         'r13' type_id=1 bits_offset=128
  ...

If not, it will point to fwd declaration record:

  [15439] FWD 'pt_regs' fwd_kind=struct

and make bpf tracing program hooking on those functions unable
to access fields from 'struct pt_regs'.

Including asm/ptrace.h directly in syscall_wrapper.h to make
sure all syscalls see 'struct pt_regs' definition and resulted
BTF for '__*_sys_*(struct pt_regs *regs)' functions point to
actual struct, not just forward declaration.

Reported-by: Akihiro HARAI <jharai0815@gmail.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 arch/x86/include/asm/syscall_wrapper.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
  

Comments

Stanislav Fomichev Oct. 18, 2022, 6:23 p.m. UTC | #1
On 10/18, Jiri Olsa wrote:
> From: Jiri Olsa <olsajiri@gmail.com>

> With just the forward declaration of the 'struct pt_regs' in
> syscall_wrapper.h, the syscall stub functions:

>    __[x64|ia32]_sys_*(struct pt_regs *regs)

> will have different definition of 'regs' argument in BTF data
> based on which object file they are defined in.

> If the syscall's object includes 'struct pt_regs' definition,
> the BTF argument data will point to 'struct pt_regs' record,
> like:

>    [226] STRUCT 'pt_regs' size=168 vlen=21
>           'r15' type_id=1 bits_offset=0
>           'r14' type_id=1 bits_offset=64
>           'r13' type_id=1 bits_offset=128
>    ...

> If not, it will point to fwd declaration record:

>    [15439] FWD 'pt_regs' fwd_kind=struct

> and make bpf tracing program hooking on those functions unable
> to access fields from 'struct pt_regs'.

Is the core issue here is that we can't / don't resolve FWD declarations
at load time (or dedup time)? I'm assuming 'struct pt_regs' is still
exposed somewhere in BTF via some other obj file, it's just those
syscall definitions that have FWD decl?

Applying this patch seems fine for now, but also seems fragile. Should
we instead/also teach verifier/dedup/whatever to resolve those FWD
declarations?

> Including asm/ptrace.h directly in syscall_wrapper.h to make
> sure all syscalls see 'struct pt_regs' definition and resulted
> BTF for '__*_sys_*(struct pt_regs *regs)' functions point to
> actual struct, not just forward declaration.

> Reported-by: Akihiro HARAI <jharai0815@gmail.com>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>   arch/x86/include/asm/syscall_wrapper.h | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)

> diff --git a/arch/x86/include/asm/syscall_wrapper.h  
> b/arch/x86/include/asm/syscall_wrapper.h
> index 59358d1bf880..fd2669b1cb2d 100644
> --- a/arch/x86/include/asm/syscall_wrapper.h
> +++ b/arch/x86/include/asm/syscall_wrapper.h
> @@ -6,7 +6,7 @@
>   #ifndef _ASM_X86_SYSCALL_WRAPPER_H
>   #define _ASM_X86_SYSCALL_WRAPPER_H

> -struct pt_regs;
> +#include <asm/ptrace.h>

>   extern long __x64_sys_ni_syscall(const struct pt_regs *regs);
>   extern long __ia32_sys_ni_syscall(const struct pt_regs *regs);
> --
> 2.37.3
  
Jiri Olsa Oct. 19, 2022, 9:30 a.m. UTC | #2
On Tue, Oct 18, 2022 at 11:23:21AM -0700, sdf@google.com wrote:
> On 10/18, Jiri Olsa wrote:
> > From: Jiri Olsa <olsajiri@gmail.com>
> 
> > With just the forward declaration of the 'struct pt_regs' in
> > syscall_wrapper.h, the syscall stub functions:
> 
> >    __[x64|ia32]_sys_*(struct pt_regs *regs)
> 
> > will have different definition of 'regs' argument in BTF data
> > based on which object file they are defined in.
> 
> > If the syscall's object includes 'struct pt_regs' definition,
> > the BTF argument data will point to 'struct pt_regs' record,
> > like:
> 
> >    [226] STRUCT 'pt_regs' size=168 vlen=21
> >           'r15' type_id=1 bits_offset=0
> >           'r14' type_id=1 bits_offset=64
> >           'r13' type_id=1 bits_offset=128
> >    ...
> 
> > If not, it will point to fwd declaration record:
> 
> >    [15439] FWD 'pt_regs' fwd_kind=struct
> 
> > and make bpf tracing program hooking on those functions unable
> > to access fields from 'struct pt_regs'.
> 
> Is the core issue here is that we can't / don't resolve FWD declarations
> at load time (or dedup time)? I'm assuming 'struct pt_regs' is still
> exposed somewhere in BTF via some other obj file, it's just those
> syscall definitions that have FWD decl?

yes, BTF is generated from dwarf debug info, so it's object based,
and in some objects the regs argument will point to full struct
definition and in some just to forward declaration

> 
> Applying this patch seems fine for now, but also seems fragile. Should
> we instead/also teach verifier/dedup/whatever to resolve those FWD
> declarations?

I'm not sure how hard it'd be to connect forward declarations
to definitions.. it'd need to be probably in dedup or verifier
as you suggest

and I think it'd need to be done just based on struct name search,
so I expect all sort of unexpected problems ;-) other than to be
sure you connect to proper struct

Andrii will have probably better idea if and where this is possible

jirka

> 
> > Including asm/ptrace.h directly in syscall_wrapper.h to make
> > sure all syscalls see 'struct pt_regs' definition and resulted
> > BTF for '__*_sys_*(struct pt_regs *regs)' functions point to
> > actual struct, not just forward declaration.
> 
> > Reported-by: Akihiro HARAI <jharai0815@gmail.com>
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >   arch/x86/include/asm/syscall_wrapper.h | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> > diff --git a/arch/x86/include/asm/syscall_wrapper.h
> > b/arch/x86/include/asm/syscall_wrapper.h
> > index 59358d1bf880..fd2669b1cb2d 100644
> > --- a/arch/x86/include/asm/syscall_wrapper.h
> > +++ b/arch/x86/include/asm/syscall_wrapper.h
> > @@ -6,7 +6,7 @@
> >   #ifndef _ASM_X86_SYSCALL_WRAPPER_H
> >   #define _ASM_X86_SYSCALL_WRAPPER_H
> 
> > -struct pt_regs;
> > +#include <asm/ptrace.h>
> 
> >   extern long __x64_sys_ni_syscall(const struct pt_regs *regs);
> >   extern long __ia32_sys_ni_syscall(const struct pt_regs *regs);
> > --
> > 2.37.3
>
  
Andrii Nakryiko Oct. 21, 2022, 9:49 p.m. UTC | #3
On Wed, Oct 19, 2022 at 2:30 AM Jiri Olsa <olsajiri@gmail.com> wrote:
>
> On Tue, Oct 18, 2022 at 11:23:21AM -0700, sdf@google.com wrote:
> > On 10/18, Jiri Olsa wrote:
> > > From: Jiri Olsa <olsajiri@gmail.com>
> >
> > > With just the forward declaration of the 'struct pt_regs' in
> > > syscall_wrapper.h, the syscall stub functions:
> >
> > >    __[x64|ia32]_sys_*(struct pt_regs *regs)
> >
> > > will have different definition of 'regs' argument in BTF data
> > > based on which object file they are defined in.
> >
> > > If the syscall's object includes 'struct pt_regs' definition,
> > > the BTF argument data will point to 'struct pt_regs' record,
> > > like:
> >
> > >    [226] STRUCT 'pt_regs' size=168 vlen=21
> > >           'r15' type_id=1 bits_offset=0
> > >           'r14' type_id=1 bits_offset=64
> > >           'r13' type_id=1 bits_offset=128
> > >    ...
> >
> > > If not, it will point to fwd declaration record:
> >
> > >    [15439] FWD 'pt_regs' fwd_kind=struct
> >
> > > and make bpf tracing program hooking on those functions unable
> > > to access fields from 'struct pt_regs'.
> >
> > Is the core issue here is that we can't / don't resolve FWD declarations
> > at load time (or dedup time)? I'm assuming 'struct pt_regs' is still
> > exposed somewhere in BTF via some other obj file, it's just those
> > syscall definitions that have FWD decl?
>
> yes, BTF is generated from dwarf debug info, so it's object based,
> and in some objects the regs argument will point to full struct
> definition and in some just to forward declaration
>
> >
> > Applying this patch seems fine for now, but also seems fragile. Should
> > we instead/also teach verifier/dedup/whatever to resolve those FWD
> > declarations?
>
> I'm not sure how hard it'd be to connect forward declarations
> to definitions.. it'd need to be probably in dedup or verifier
> as you suggest
>
> and I think it'd need to be done just based on struct name search,
> so I expect all sort of unexpected problems ;-) other than to be
> sure you connect to proper struct

exactly, which is why BTF dedup algorithm resolves FWD to STRUCT/UNION
only when we have to outer structs/unions that we are deduplicating,
and one of their field points to FWD on one side and STRUCT on another
side. Looking up by name is ambiguous and could be incorrect, so BTF
dedup needs sort of a "structural proof".

As far as the patch goes, it's a good thing to have more complete
types wherever possible, so:

Acked-by: Andrii Nakryiko <andrii@kernel.org>

>
> Andrii will have probably better idea if and where this is possible
>
> jirka
>
> >
> > > Including asm/ptrace.h directly in syscall_wrapper.h to make
> > > sure all syscalls see 'struct pt_regs' definition and resulted
> > > BTF for '__*_sys_*(struct pt_regs *regs)' functions point to
> > > actual struct, not just forward declaration.
> >
> > > Reported-by: Akihiro HARAI <jharai0815@gmail.com>
> > > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > > ---
> > >   arch/x86/include/asm/syscall_wrapper.h | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > > diff --git a/arch/x86/include/asm/syscall_wrapper.h
> > > b/arch/x86/include/asm/syscall_wrapper.h
> > > index 59358d1bf880..fd2669b1cb2d 100644
> > > --- a/arch/x86/include/asm/syscall_wrapper.h
> > > +++ b/arch/x86/include/asm/syscall_wrapper.h
> > > @@ -6,7 +6,7 @@
> > >   #ifndef _ASM_X86_SYSCALL_WRAPPER_H
> > >   #define _ASM_X86_SYSCALL_WRAPPER_H
> >
> > > -struct pt_regs;
> > > +#include <asm/ptrace.h>
> >
> > >   extern long __x64_sys_ni_syscall(const struct pt_regs *regs);
> > >   extern long __ia32_sys_ni_syscall(const struct pt_regs *regs);
> > > --
> > > 2.37.3
> >
  
Lorenz Bauer Oct. 24, 2022, 3:26 p.m. UTC | #4
On Tue, 18 Oct 2022, at 13:27, Jiri Olsa wrote:
> From: Jiri Olsa <olsajiri@gmail.com>
>
> With just the forward declaration of the 'struct pt_regs' in
> syscall_wrapper.h, the syscall stub functions:
>
>   __[x64|ia32]_sys_*(struct pt_regs *regs)

I think arm64 has a similar problem: https://elixir.bootlin.com/linux/v6.1-rc2/source/arch/arm64/include/asm/syscall_wrapper.h#L11

Best
Lorenz
  

Patch

diff --git a/arch/x86/include/asm/syscall_wrapper.h b/arch/x86/include/asm/syscall_wrapper.h
index 59358d1bf880..fd2669b1cb2d 100644
--- a/arch/x86/include/asm/syscall_wrapper.h
+++ b/arch/x86/include/asm/syscall_wrapper.h
@@ -6,7 +6,7 @@ 
 #ifndef _ASM_X86_SYSCALL_WRAPPER_H
 #define _ASM_X86_SYSCALL_WRAPPER_H
 
-struct pt_regs;
+#include <asm/ptrace.h>
 
 extern long __x64_sys_ni_syscall(const struct pt_regs *regs);
 extern long __ia32_sys_ni_syscall(const struct pt_regs *regs);