diff mbox

[RFC] binfmt_elf: Use ELF_ET_DYN_BASE only for PIE

Message ID 20170621055835.GA27467@beast (mailing list archive)
State New, archived
Headers show

Commit Message

Kees Cook June 21, 2017, 5:58 a.m. UTC
The ELF_ET_DYN_BASE position was originally intended to keep loaders
away from ET_EXEC binaries. (For example, running "/lib/ld-linux.so.2
/bin/cat" might cause the subsequent load of /bin/cat into where the
loader had been loaded.) With the advent of PIE (ET_DYN binaries with
an INTERP Program Header), ELF_ET_DYN_BASE continued to be used since
the kernel was only looking at ET_DYN. However, since ELF_ET_DYN_BASE
is traditionally set at the top 1/3rd of the TASK_SIZE, a substantial
portion of the address space is unused.

For 32-bit tasks when RLIMIT_STACK is set to RLIM_INFINITY, programs
are loaded below the mmap region. This means they can be made to collide
(CVE-2017-1000370) or nearly collide (CVE-2017-1000371) with pathological
stack regions. Lowering ELF_ET_DYN_BASE solves both by moving programs
above the mmap region in all cases, and will now additionally avoid
programs falling back to the mmap region by enforcing MAP_FIXED for
program loads (i.e. if it would have collided with the stack, now it
will fail to load instead of falling back to the mmap region).

To allow for a lower ELF_ET_DYN_BASE, loaders (ET_DYN without INTERP)
are loaded into the mmap region, leaving space available for either an
ET_EXEC binary with a fixed location or PIE being loaded into mmap by the
loader. Only PIE programs are loaded offset from ELF_ET_DYN_BASE, which
means architectures can now safely lower their values without risk of
loaders colliding with their subsequently loaded programs.

Thanks go to PaX for inspiration on how to approach this solution.

Fixes: d1fd836dcf00 ("mm: split ET_DYN ASLR from mmap ASLR")
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 arch/x86/include/asm/elf.h |  8 ++------
 fs/binfmt_elf.c            | 38 +++++++++++++++++++++++++++++++-------
 2 files changed, 33 insertions(+), 13 deletions(-)

Comments

Rik van Riel June 21, 2017, 12:07 p.m. UTC | #1
On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
> The ELF_ET_DYN_BASE position was originally intended to keep loaders
> away from ET_EXEC binaries. (For example, running "/lib/ld-linux.so.2
> /bin/cat" might cause the subsequent load of /bin/cat into where the
> loader had been loaded.) With the advent of PIE (ET_DYN binaries with
> an INTERP Program Header), ELF_ET_DYN_BASE continued to be used since
> the kernel was only looking at ET_DYN. However, since ELF_ET_DYN_BASE
> is traditionally set at the top 1/3rd of the TASK_SIZE, a substantial
> portion of the address space is unused.
> 
> For 32-bit tasks when RLIMIT_STACK is set to RLIM_INFINITY, programs
> are loaded below the mmap region. This means they can be made to
> collide
> (CVE-2017-1000370) or nearly collide (CVE-2017-1000371) with
> pathological
> stack regions. Lowering ELF_ET_DYN_BASE solves both by moving
> programs
> above the mmap region in all cases, and will now additionally avoid
> programs falling back to the mmap region by enforcing MAP_FIXED for
> program loads (i.e. if it would have collided with the stack, now it
> will fail to load instead of falling back to the mmap region).
> 
> To allow for a lower ELF_ET_DYN_BASE, loaders (ET_DYN without INTERP)
> are loaded into the mmap region, leaving space available for either
> an
> ET_EXEC binary with a fixed location or PIE being loaded into mmap by
> the
> loader. Only PIE programs are loaded offset from ELF_ET_DYN_BASE,
> which
> means architectures can now safely lower their values without risk of
> loaders colliding with their subsequently loaded programs.
> 
> Thanks go to PaX for inspiration on how to approach this solution.
> 
> Fixes: d1fd836dcf00 ("mm: split ET_DYN ASLR from mmap ASLR")
> Signed-off-by: Kees Cook <keescook@chromium.org>
> ---
>  arch/x86/include/asm/elf.h |  8 ++------
>  fs/binfmt_elf.c            | 38 +++++++++++++++++++++++++++++++-----
> --
>  2 files changed, 33 insertions(+), 13 deletions(-)
> 
> diff --git a/arch/x86/include/asm/elf.h b/arch/x86/include/asm/elf.h
> index e8ab9a46bc68..46549973ea98 100644
> --- a/arch/x86/include/asm/elf.h
> +++ b/arch/x86/include/asm/elf.h
> @@ -245,12 +245,8 @@ extern int force_personality32;
>  #define CORE_DUMP_USE_REGSET
>  #define ELF_EXEC_PAGESIZE	4096
>  
> -/* This is the location that an ET_DYN program is loaded if
> exec'ed.  Typical
> -   use of this is to invoke "./ld.so someprog" to test out a new
> version of
> -   the loader.  We need to make sure that it is out of the way of
> the program
> -   that it will "exec", and that there is sufficient room for the
> brk.  */
> -
> -#define ELF_ET_DYN_BASE		(TASK_SIZE / 3 * 2)
> +/* This is the base location for PIE (ET_DYN with INTERP) loads. */
> +#define ELF_ET_DYN_BASE		0x400000UL

This value is good for 32 bit binaries, but for 64
bit binaries you probably want to put ELF_ET_DYN_BASE
at 4GB or higher.

The latter is necessary because Android uses the
lower 4GB of address space for its JVM runtime,
with 32 bit pointers inside that part of the otherwise
64 bit address space.

In other words:

#define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL : 0x100000000UL)

> +++ b/fs/binfmt_elf.c
> 
> +			 * Therefore, programs are loaded offset
> from
> +			 * ELF_ET_DYN_BASE and loaders are loaded
> into the
> +			 * independently randomized mmap region (0
> load_bias
> +			 * without MAP_FIXED).
> +			 */
> +			if (elf_interpreter) {
> +				load_bias = ELF_ET_DYN_BASE;
> +				if (current->flags & PF_RANDOMIZE)
> +					load_bias +=
> arch_mmap_rnd();
> +				elf_flags |= MAP_FIXED;
> +			} else
> +				load_bias = 0;
> +
> +			load_bias -= vaddr;

I like this, and the big comment telling people how it
works :)
Kees Cook June 21, 2017, 5:23 p.m. UTC | #2
On Wed, Jun 21, 2017 at 5:07 AM, Rik van Riel <riel@redhat.com> wrote:
> On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
>> +/* This is the base location for PIE (ET_DYN with INTERP) loads. */
>> +#define ELF_ET_DYN_BASE              0x400000UL
>
> This value is good for 32 bit binaries, but for 64
> bit binaries you probably want to put ELF_ET_DYN_BASE
> at 4GB or higher.
>
> The latter is necessary because Android uses the
> lower 4GB of address space for its JVM runtime,
> with 32 bit pointers inside that part of the otherwise
> 64 bit address space.
>
> In other words:
>
> #define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL : 0x100000000UL)

Ah, interesting. Okay, that should be fine. I'll adjust it.

>> +++ b/fs/binfmt_elf.c
>>
>> +                      * Therefore, programs are loaded offset
>> from
>> +                      * ELF_ET_DYN_BASE and loaders are loaded
>> into the
>> +                      * independently randomized mmap region (0
>> load_bias
>> +                      * without MAP_FIXED).
>> +                      */
>> +                     if (elf_interpreter) {
>> +                             load_bias = ELF_ET_DYN_BASE;
>> +                             if (current->flags & PF_RANDOMIZE)
>> +                                     load_bias +=
>> arch_mmap_rnd();
>> +                             elf_flags |= MAP_FIXED;
>> +                     } else
>> +                             load_bias = 0;
>> +
>> +                     load_bias -= vaddr;
>
> I like this, and the big comment telling people how it
> works :)

Thanks! It looks like your patch for commenting load_bias never got
picked up, so I've added some more comments for that and some other
things too. (Mostly for all the stuff I have to read a few times when
I look at this code.)

-Kees
Daniel Micay June 21, 2017, 5:27 p.m. UTC | #3
On Wed, 2017-06-21 at 10:23 -0700, Kees Cook wrote:
> On Wed, Jun 21, 2017 at 5:07 AM, Rik van Riel <riel@redhat.com> wrote:
> > On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
> > > +/* This is the base location for PIE (ET_DYN with INTERP) loads.
> > > */
> > > +#define ELF_ET_DYN_BASE              0x400000UL
> > 
> > This value is good for 32 bit binaries, but for 64
> > bit binaries you probably want to put ELF_ET_DYN_BASE
> > at 4GB or higher.
> > 
> > The latter is necessary because Android uses the
> > lower 4GB of address space for its JVM runtime,
> > with 32 bit pointers inside that part of the otherwise
> > 64 bit address space.
> > 
> > In other words:
> > 
> > #define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL :
> > 0x100000000UL)
> 
> Ah, interesting. Okay, that should be fine. I'll adjust it.
> 
> > > +++ b/fs/binfmt_elf.c
> > > 
> > > +                      * Therefore, programs are loaded offset
> > > from
> > > +                      * ELF_ET_DYN_BASE and loaders are loaded
> > > into the
> > > +                      * independently randomized mmap region (0
> > > load_bias
> > > +                      * without MAP_FIXED).
> > > +                      */
> > > +                     if (elf_interpreter) {
> > > +                             load_bias = ELF_ET_DYN_BASE;
> > > +                             if (current->flags & PF_RANDOMIZE)
> > > +                                     load_bias +=
> > > arch_mmap_rnd();
> > > +                             elf_flags |= MAP_FIXED;
> > > +                     } else
> > > +                             load_bias = 0;
> > > +
> > > +                     load_bias -= vaddr;
> > 
> > I like this, and the big comment telling people how it
> > works :)
> 
> Thanks! It looks like your patch for commenting load_bias never got
> picked up, so I've added some more comments for that and some other
> things too. (Mostly for all the stuff I have to read a few times when
> I look at this code.)
> 
> -Kees
> 

The stack rlimit calculation fix for space potentially lost to ASLR is
probably still needed too, right?
Kees Cook June 21, 2017, 5:28 p.m. UTC | #4
On Wed, Jun 21, 2017 at 10:27 AM, Daniel Micay <danielmicay@gmail.com> wrote:
> On Wed, 2017-06-21 at 10:23 -0700, Kees Cook wrote:
>> On Wed, Jun 21, 2017 at 5:07 AM, Rik van Riel <riel@redhat.com> wrote:
>> > On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
>> > > +/* This is the base location for PIE (ET_DYN with INTERP) loads.
>> > > */
>> > > +#define ELF_ET_DYN_BASE              0x400000UL
>> >
>> > This value is good for 32 bit binaries, but for 64
>> > bit binaries you probably want to put ELF_ET_DYN_BASE
>> > at 4GB or higher.
>> >
>> > The latter is necessary because Android uses the
>> > lower 4GB of address space for its JVM runtime,
>> > with 32 bit pointers inside that part of the otherwise
>> > 64 bit address space.
>> >
>> > In other words:
>> >
>> > #define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL :
>> > 0x100000000UL)
>>
>> Ah, interesting. Okay, that should be fine. I'll adjust it.
>>
>> > > +++ b/fs/binfmt_elf.c
>> > >
>> > > +                      * Therefore, programs are loaded offset
>> > > from
>> > > +                      * ELF_ET_DYN_BASE and loaders are loaded
>> > > into the
>> > > +                      * independently randomized mmap region (0
>> > > load_bias
>> > > +                      * without MAP_FIXED).
>> > > +                      */
>> > > +                     if (elf_interpreter) {
>> > > +                             load_bias = ELF_ET_DYN_BASE;
>> > > +                             if (current->flags & PF_RANDOMIZE)
>> > > +                                     load_bias +=
>> > > arch_mmap_rnd();
>> > > +                             elf_flags |= MAP_FIXED;
>> > > +                     } else
>> > > +                             load_bias = 0;
>> > > +
>> > > +                     load_bias -= vaddr;
>> >
>> > I like this, and the big comment telling people how it
>> > works :)
>>
>> Thanks! It looks like your patch for commenting load_bias never got
>> picked up, so I've added some more comments for that and some other
>> things too. (Mostly for all the stuff I have to read a few times when
>> I look at this code.)
>>
>> -Kees
>>
>
> The stack rlimit calculation fix for space potentially lost to ASLR is
> probably still needed too, right?

Yes. Was that picked up by akpm already?

-Kees
Daniel Micay June 21, 2017, 5:32 p.m. UTC | #5
On Wed, 2017-06-21 at 10:28 -0700, Kees Cook wrote:
> On Wed, Jun 21, 2017 at 10:27 AM, Daniel Micay <danielmicay@gmail.com>
> wrote:
> > On Wed, 2017-06-21 at 10:23 -0700, Kees Cook wrote:
> > > On Wed, Jun 21, 2017 at 5:07 AM, Rik van Riel <riel@redhat.com>
> > > wrote:
> > > > On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
> > > > > +/* This is the base location for PIE (ET_DYN with INTERP)
> > > > > loads.
> > > > > */
> > > > > +#define ELF_ET_DYN_BASE              0x400000UL
> > > > 
> > > > This value is good for 32 bit binaries, but for 64
> > > > bit binaries you probably want to put ELF_ET_DYN_BASE
> > > > at 4GB or higher.
> > > > 
> > > > The latter is necessary because Android uses the
> > > > lower 4GB of address space for its JVM runtime,
> > > > with 32 bit pointers inside that part of the otherwise
> > > > 64 bit address space.
> > > > 
> > > > In other words:
> > > > 
> > > > #define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL :
> > > > 0x100000000UL)
> > > 
> > > Ah, interesting. Okay, that should be fine. I'll adjust it.
> > > 
> > > > > +++ b/fs/binfmt_elf.c
> > > > > 
> > > > > +                      * Therefore, programs are loaded offset
> > > > > from
> > > > > +                      * ELF_ET_DYN_BASE and loaders are
> > > > > loaded
> > > > > into the
> > > > > +                      * independently randomized mmap region
> > > > > (0
> > > > > load_bias
> > > > > +                      * without MAP_FIXED).
> > > > > +                      */
> > > > > +                     if (elf_interpreter) {
> > > > > +                             load_bias = ELF_ET_DYN_BASE;
> > > > > +                             if (current->flags &
> > > > > PF_RANDOMIZE)
> > > > > +                                     load_bias +=
> > > > > arch_mmap_rnd();
> > > > > +                             elf_flags |= MAP_FIXED;
> > > > > +                     } else
> > > > > +                             load_bias = 0;
> > > > > +
> > > > > +                     load_bias -= vaddr;
> > > > 
> > > > I like this, and the big comment telling people how it
> > > > works :)
> > > 
> > > Thanks! It looks like your patch for commenting load_bias never
> > > got
> > > picked up, so I've added some more comments for that and some
> > > other
> > > things too. (Mostly for all the stuff I have to read a few times
> > > when
> > > I look at this code.)
> > > 
> > > -Kees
> > > 
> > 
> > The stack rlimit calculation fix for space potentially lost to ASLR
> > is
> > probably still needed too, right?
> 
> Yes. Was that picked up by akpm already?
> 
> -Kees

I think it was dropped when the ET_DYN changes were dropped.
Kees Cook June 21, 2017, 5:33 p.m. UTC | #6
On Wed, Jun 21, 2017 at 10:32 AM, Daniel Micay <danielmicay@gmail.com> wrote:
> On Wed, 2017-06-21 at 10:28 -0700, Kees Cook wrote:
>> On Wed, Jun 21, 2017 at 10:27 AM, Daniel Micay <danielmicay@gmail.com>
>> wrote:
>> > On Wed, 2017-06-21 at 10:23 -0700, Kees Cook wrote:
>> > > On Wed, Jun 21, 2017 at 5:07 AM, Rik van Riel <riel@redhat.com>
>> > > wrote:
>> > > > On Tue, 2017-06-20 at 22:58 -0700, Kees Cook wrote:
>> > > > > +/* This is the base location for PIE (ET_DYN with INTERP)
>> > > > > loads.
>> > > > > */
>> > > > > +#define ELF_ET_DYN_BASE              0x400000UL
>> > > >
>> > > > This value is good for 32 bit binaries, but for 64
>> > > > bit binaries you probably want to put ELF_ET_DYN_BASE
>> > > > at 4GB or higher.
>> > > >
>> > > > The latter is necessary because Android uses the
>> > > > lower 4GB of address space for its JVM runtime,
>> > > > with 32 bit pointers inside that part of the otherwise
>> > > > 64 bit address space.
>> > > >
>> > > > In other words:
>> > > >
>> > > > #define ELF_ET_DYN_BASE (mmap_is_ia32() ? 0x400000UL :
>> > > > 0x100000000UL)
>> > >
>> > > Ah, interesting. Okay, that should be fine. I'll adjust it.
>> > >
>> > > > > +++ b/fs/binfmt_elf.c
>> > > > >
>> > > > > +                      * Therefore, programs are loaded offset
>> > > > > from
>> > > > > +                      * ELF_ET_DYN_BASE and loaders are
>> > > > > loaded
>> > > > > into the
>> > > > > +                      * independently randomized mmap region
>> > > > > (0
>> > > > > load_bias
>> > > > > +                      * without MAP_FIXED).
>> > > > > +                      */
>> > > > > +                     if (elf_interpreter) {
>> > > > > +                             load_bias = ELF_ET_DYN_BASE;
>> > > > > +                             if (current->flags &
>> > > > > PF_RANDOMIZE)
>> > > > > +                                     load_bias +=
>> > > > > arch_mmap_rnd();
>> > > > > +                             elf_flags |= MAP_FIXED;
>> > > > > +                     } else
>> > > > > +                             load_bias = 0;
>> > > > > +
>> > > > > +                     load_bias -= vaddr;
>> > > >
>> > > > I like this, and the big comment telling people how it
>> > > > works :)
>> > >
>> > > Thanks! It looks like your patch for commenting load_bias never
>> > > got
>> > > picked up, so I've added some more comments for that and some
>> > > other
>> > > things too. (Mostly for all the stuff I have to read a few times
>> > > when
>> > > I look at this code.)
>> > >
>> > > -Kees
>> > >
>> >
>> > The stack rlimit calculation fix for space potentially lost to ASLR
>> > is
>> > probably still needed too, right?
>>
>> Yes. Was that picked up by akpm already?
>>
>> -Kees
>
> I think it was dropped when the ET_DYN changes were dropped.

Ah! Rik, can you resend just the stack calculation fixes? I can Ack those.

-Kees
Rik van Riel June 21, 2017, 6:47 p.m. UTC | #7
On Wed, 2017-06-21 at 13:32 -0400, Daniel Micay wrote:
> On Wed, 2017-06-21 at 10:28 -0700, Kees Cook wrote:
> > > The stack rlimit calculation fix for space potentially lost to
> > > ASLR
> > > is
> > > probably still needed too, right?
> > 
> > Yes. Was that picked up by akpm already?
> > 
> > -Kees
> 
> I think it was dropped when the ET_DYN changes were dropped.

I can resend those.
diff mbox

Patch

diff --git a/arch/x86/include/asm/elf.h b/arch/x86/include/asm/elf.h
index e8ab9a46bc68..46549973ea98 100644
--- a/arch/x86/include/asm/elf.h
+++ b/arch/x86/include/asm/elf.h
@@ -245,12 +245,8 @@  extern int force_personality32;
 #define CORE_DUMP_USE_REGSET
 #define ELF_EXEC_PAGESIZE	4096
 
-/* This is the location that an ET_DYN program is loaded if exec'ed.  Typical
-   use of this is to invoke "./ld.so someprog" to test out a new version of
-   the loader.  We need to make sure that it is out of the way of the program
-   that it will "exec", and that there is sufficient room for the brk.  */
-
-#define ELF_ET_DYN_BASE		(TASK_SIZE / 3 * 2)
+/* This is the base location for PIE (ET_DYN with INTERP) loads. */
+#define ELF_ET_DYN_BASE		0x400000UL
 
 /* This yields a mask that user programs can use to figure out what
    instruction set this CPU supports.  This could be done in user space,
diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 5075fd5c62c8..a998c7251d1c 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -930,13 +930,37 @@  static int load_elf_binary(struct linux_binprm *bprm)
 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 			elf_flags |= MAP_FIXED;
 		} else if (loc->elf_ex.e_type == ET_DYN) {
-			/* Try and get dynamic programs out of the way of the
-			 * default mmap base, as well as whatever program they
-			 * might try to exec.  This is because the brk will
-			 * follow the loader, and is not movable.  */
-			load_bias = ELF_ET_DYN_BASE - vaddr;
-			if (current->flags & PF_RANDOMIZE)
-				load_bias += arch_mmap_rnd();
+			/*
+			 * There are effectively two types of ET_DYN
+			 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
+			 * and loaders (ET_DYN without INTERP, since they
+			 * _are_ the ELF interpreter). The loaders must
+			 * be loaded away from programs since the program
+			 * may otherwise collide with the loader (especially
+			 * for ET_EXEC which does not have a randomized
+			 * position). For example to handle invocations of
+			 * "./ld.so someprog" to test out a new version of
+			 * the loader, the subsequent program that the
+			 * loader loads must avoid the loader itself, so
+			 * they cannot share the same load range. Sufficient
+			 * room for the brk must be allocated with the
+			 * loader as well, since brk must be available with
+			 * the loader.
+			 *
+			 * Therefore, programs are loaded offset from
+			 * ELF_ET_DYN_BASE and loaders are loaded into the
+			 * independently randomized mmap region (0 load_bias
+			 * without MAP_FIXED).
+			 */
+			if (elf_interpreter) {
+				load_bias = ELF_ET_DYN_BASE;
+				if (current->flags & PF_RANDOMIZE)
+					load_bias += arch_mmap_rnd();
+				elf_flags |= MAP_FIXED;
+			} else
+				load_bias = 0;
+
+			load_bias -= vaddr;
 			load_bias = ELF_PAGESTART(load_bias);
 			total_size = total_mapping_size(elf_phdata,
 							loc->elf_ex.e_phnum);