Re: [Linux-kernel-mentees] [PATCH v3] ptrace: Prevent kernel-infoleak in ptrace_get_syscall_info()

From: Christian Brauner
Date: Sat Aug 01 2020 - 16:11:05 EST


On Sat, Aug 01, 2020 at 07:08:19PM +0300, Dmitry V. Levin wrote:
> On Sat, Aug 01, 2020 at 11:20:44AM -0400, Peilin Ye wrote:
> > ptrace_get_syscall_info() is potentially copying uninitialized stack
> > memory to userspace, since the compiler may leave a 3-byte hole near the
> > beginning of `info`. Fix it by adding a padding field to `struct
> > ptrace_syscall_info`.
> >
> > Cc: stable@xxxxxxxxxxxxxxx
> > Fixes: 201766a20e30 ("ptrace: add PTRACE_GET_SYSCALL_INFO request")
> > Suggested-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> > Signed-off-by: Peilin Ye <yepeilin.cs@xxxxxxxxx>
> > ---
> > Change in v3:
> > - Remove unnecessary `__aligned__` attribute. (Suggested by
> > Dmitry V. Levin <ldv@xxxxxxxxxxxx>)
> >
> > Change in v2:
> > - Add a padding field to `struct ptrace_syscall_info`, instead of
> > doing memset() on `info`. (Suggested by Dmitry V. Levin
> > <ldv@xxxxxxxxxxxx>)
> >
> > Reference: https://lwn.net/Articles/417989/
> >
> > $ # before:
> > $ pahole -C "ptrace_syscall_info" kernel/ptrace.o
> > struct ptrace_syscall_info {
> > __u8 op; /* 0 1 */
> >
> > /* XXX 3 bytes hole, try to pack */
> >
> > __u32 arch __attribute__((__aligned__(4))); /* 4 4 */
> > __u64 instruction_pointer; /* 8 8 */
> > __u64 stack_pointer; /* 16 8 */
> > union {
> > struct {
> > __u64 nr; /* 24 8 */
> > __u64 args[6]; /* 32 48 */
> > } entry; /* 24 56 */
> > struct {
> > __s64 rval; /* 24 8 */
> > __u8 is_error; /* 32 1 */
> > } exit; /* 24 16 */
> > struct {
> > __u64 nr; /* 24 8 */
> > __u64 args[6]; /* 32 48 */
> > /* --- cacheline 1 boundary (64 bytes) was 16 bytes ago --- */
> > __u32 ret_data; /* 80 4 */
> > } seccomp; /* 24 64 */
> > }; /* 24 64 */
> >
> > /* size: 88, cachelines: 2, members: 5 */
> > /* sum members: 85, holes: 1, sum holes: 3 */
> > /* forced alignments: 1, forced holes: 1, sum forced holes: 3 */
> > /* last cacheline: 24 bytes */
> > } __attribute__((__aligned__(8)));
> > $
> > $ # after:
> > $ pahole -C "ptrace_syscall_info" kernel/ptrace.o
> > struct ptrace_syscall_info {
> > __u8 op; /* 0 1 */
> > __u8 pad[3]; /* 1 3 */
> > __u32 arch; /* 4 4 */
> > __u64 instruction_pointer; /* 8 8 */
> > __u64 stack_pointer; /* 16 8 */
> > union {
> > struct {
> > __u64 nr; /* 24 8 */
> > __u64 args[6]; /* 32 48 */
> > } entry; /* 24 56 */
> > struct {
> > __s64 rval; /* 24 8 */
> > __u8 is_error; /* 32 1 */
> > } exit; /* 24 16 */
> > struct {
> > __u64 nr; /* 24 8 */
> > __u64 args[6]; /* 32 48 */
> > /* --- cacheline 1 boundary (64 bytes) was 16 bytes ago --- */
> > __u32 ret_data; /* 80 4 */
> > } seccomp; /* 24 64 */
> > }; /* 24 64 */
> >
> > /* size: 88, cachelines: 2, members: 6 */
> > /* last cacheline: 24 bytes */
> > };
> > $ _
> >
> > include/uapi/linux/ptrace.h | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/include/uapi/linux/ptrace.h b/include/uapi/linux/ptrace.h
> > index a71b6e3b03eb..83ee45fa634b 100644
> > --- a/include/uapi/linux/ptrace.h
> > +++ b/include/uapi/linux/ptrace.h
> > @@ -81,7 +81,8 @@ struct seccomp_metadata {
> >
> > struct ptrace_syscall_info {
> > __u8 op; /* PTRACE_SYSCALL_INFO_* */
> > - __u32 arch __attribute__((__aligned__(sizeof(__u32))));
> > + __u8 pad[3];
> > + __u32 arch;
> > __u64 instruction_pointer;
> > __u64 stack_pointer;
> > union {
>
> Reviewed-by: Dmitry V. Levin <ldv@xxxxxxxxxxxx>

Acked-by: Christian Brauner <christian.brauner@xxxxxxxxxx>

Oh fun.
I'd pick this up and run the ptrace tests that we have for this. If they
pass I'd apply to my fixes branch and send after the merge window unless
I hear objections.

Fwiw, what was the original reason for using
__attribute__((__aligned__(sizeof(__u32))))?
b4 mbox is failing to download the relevant thread(s) for me.

Thanks!
Christian