patches/kernel/0138-x86-unwinder-orc-Dont-bail-on-stack-overflow.patch

   1 From bb0be747b5ee45f07f5514a214231c9061261b50 Mon Sep 17 00:00:00 2001
   2 From: Andy Lutomirski <luto@kernel.org>
   3 Date: Mon, 4 Dec 2017 15:07:08 +0100
   4 Subject: [PATCH 138/233] x86/unwinder/orc: Dont bail on stack overflow
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 CVE-2017-5754
  10
  11 If the stack overflows into a guard page and the ORC unwinder should work
  12 well: by construction, there can't be any meaningful data in the guard page
  13 because no writes to the guard page will have succeeded.
  14
  15 But there is a bug that prevents unwinding from working correctly: if the
  16 starting register state has RSP pointing into a stack guard page, the ORC
  17 unwinder bails out immediately.
  18
  19 Instead of bailing out immediately check whether the next page up is a
  20 valid check page and if so analyze that. As a result the ORC unwinder will
  21 start the unwind.
  22
  23 Tested by intentionally overflowing the task stack.  The result is an
  24 accurate call trace instead of a trace consisting purely of '?' entries.
  25
  26 There are a few other bugs that are triggered if the unwinder encounters a
  27 stack overflow after the first step, but they are outside the scope of this
  28 fix.
  29
  30 Signed-off-by: Andy Lutomirski <luto@kernel.org>
  31 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
  32 Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
  33 Cc: Borislav Petkov <bp@alien8.de>
  34 Cc: Borislav Petkov <bpetkov@suse.de>
  35 Cc: Brian Gerst <brgerst@gmail.com>
  36 Cc: Dave Hansen <dave.hansen@intel.com>
  37 Cc: Dave Hansen <dave.hansen@linux.intel.com>
  38 Cc: David Laight <David.Laight@aculab.com>
  39 Cc: Denys Vlasenko <dvlasenk@redhat.com>
  40 Cc: Eduardo Valentin <eduval@amazon.com>
  41 Cc: Greg KH <gregkh@linuxfoundation.org>
  42 Cc: H. Peter Anvin <hpa@zytor.com>
  43 Cc: Josh Poimboeuf <jpoimboe@redhat.com>
  44 Cc: Juergen Gross <jgross@suse.com>
  45 Cc: Linus Torvalds <torvalds@linux-foundation.org>
  46 Cc: Peter Zijlstra <peterz@infradead.org>
  47 Cc: Rik van Riel <riel@redhat.com>
  48 Cc: Will Deacon <will.deacon@arm.com>
  49 Cc: aliguori@amazon.com
  50 Cc: daniel.gruss@iaik.tugraz.at
  51 Cc: hughd@google.com
  52 Cc: keescook@google.com
  53 Link: https://lkml.kernel.org/r/20171204150604.991389777@linutronix.de
  54 Signed-off-by: Ingo Molnar <mingo@kernel.org>
  55 (cherry picked from commit d3a09104018cf2ad5973dfa8a9c138ef9f5015a3)
  56 Signed-off-by: Andy Whitcroft <apw@canonical.com>
  57 Signed-off-by: Kleber Sacilotto de Souza <kleber.souza@canonical.com>
  58 (cherry picked from commit e5c3115ac69cddd384d6f7abc4a0ef030b247498)
  59 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  60 ---
  61  arch/x86/kernel/unwind_orc.c | 14 ++++++++++++--
  62  1 file changed, 12 insertions(+), 2 deletions(-)
  63
  64 diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
  65 index 570b70d3f604..cea85bfe93f7 100644
  66 --- a/arch/x86/kernel/unwind_orc.c
  67 +++ b/arch/x86/kernel/unwind_orc.c
  68 @@ -552,8 +552,18 @@ void __unwind_start(struct unwind_state *state, struct task_struct *task,
  69         }
  70
  71         if (get_stack_info((unsigned long *)state->sp, state->task,
  72 -                          &state->stack_info, &state->stack_mask))
  73 -               return;
  74 +                          &state->stack_info, &state->stack_mask)) {
  75 +               /*
  76 +                * We weren't on a valid stack.  It's possible that
  77 +                * we overflowed a valid stack into a guard page.
  78 +                * See if the next page up is valid so that we can
  79 +                * generate some kind of backtrace if this happens.
  80 +                */
  81 +               void *next_page = (void *)PAGE_ALIGN((unsigned long)state->sp);
  82 +               if (get_stack_info(next_page, state->task, &state->stack_info,
  83 +                                  &state->stack_mask))
  84 +                       return;
  85 +       }
  86
  87         /*
  88          * The caller can provide the address of the first frame directly
  89 --
  90 2.14.2
  91