Partially revert commit 5f501d555653 ("binfmt_elf: reintroduce using
MAP_FIXED_NOREPLACE").
At least ia64 has ET_EXEC PT_LOAD segments that are not virtual-address
contiguous (but _are_ file-offset contiguous). This would result in
giant mapping attempts to cover the entire span, including the virtual
address range hole. Disable total_mapping_size for ET_EXEC, which
reduces the MAP_FIXED_NOREPLACE coverage to only the first PT_LOAD:
$ readelf -lW /usr/bin/gcc
...
Program Headers:
Type Offset VirtAddr PhysAddr FileSiz MemSiz ...
...
LOAD 0x000000 0x4000000000000000 0x4000000000000000 0x00b5a0 0x00b5a0 ...
LOAD 0x00b5a0 0x600000000000b5a0 0x600000000000b5a0 0x0005ac 0x000710 ...
...
^^^^^^^^ ^^^^^^^^^^^^^^^^^^ ^^^^^^^^ ^^^^^^^^
File offset range : 0x000000-0x00bb4c
0x00bb4c bytes
Virtual address range : 0x4000000000000000-0x600000000000bcb0
0x200000000000bcb0 bytes
Ironically, this is the reverse of the problem that originally caused
problems with ET_EXEC and MAP_FIXED_NOREPLACE: overlaps. This problem is
with holes. Future work could restore full coverage if load_elf_binary()
were to perform mappings in a separate phase from the loading (where
it could resolve both overlaps and holes).
Cc: Alexander Viro <[email protected]>
Cc: Eric Biederman <[email protected]>
Cc: [email protected]
Cc: [email protected]
Reported-by: matoro <[email protected]>
Reported-by: John Paul Adrian Glaubitz <[email protected]>
Fixes: 5f501d555653 ("binfmt_elf: reintroduce using MAP_FIXED_NOREPLACE")
Link: https://lore.kernel.org/r/[email protected]
Cc: [email protected]
Signed-off-by: Kees Cook <[email protected]>
---
Here's the v5.16 backport.
---
fs/binfmt_elf.c | 25 ++++++++++++++++++-------
1 file changed, 18 insertions(+), 7 deletions(-)
diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index f8c7f26f1fbb..911a9e7044f4 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -1135,14 +1135,25 @@ static int load_elf_binary(struct linux_binprm *bprm)
* is then page aligned.
*/
load_bias = ELF_PAGESTART(load_bias - vaddr);
- }
- /*
- * Calculate the entire size of the ELF mapping (total_size).
- * (Note that load_addr_set is set to true later once the
- * initial mapping is performed.)
- */
- if (!load_addr_set) {
+ /*
+ * Calculate the entire size of the ELF mapping
+ * (total_size), used for the initial mapping,
+ * due to first_pt_load which is set to false later
+ * once the initial mapping is performed.
+ *
+ * Note that this is only sensible when the LOAD
+ * segments are contiguous (or overlapping). If
+ * used for LOADs that are far apart, this would
+ * cause the holes between LOADs to be mapped,
+ * running the risk of having the mapping fail,
+ * as it would be larger than the ELF file itself.
+ *
+ * As a result, only ET_DYN does this, since
+ * some ET_EXEC (e.g. ia64) may have virtual
+ * memory holes between LOADs.
+ *
+ */
total_size = total_mapping_size(elf_phdata,
elf_ex->e_phnum);
if (!total_size) {
--
2.32.0
On 2022-02-28 15:55, Kees Cook wrote:
> Partially revert commit 5f501d555653 ("binfmt_elf: reintroduce using
> MAP_FIXED_NOREPLACE").
>
> At least ia64 has ET_EXEC PT_LOAD segments that are not virtual-address
> contiguous (but _are_ file-offset contiguous). This would result in
> giant mapping attempts to cover the entire span, including the virtual
> address range hole. Disable total_mapping_size for ET_EXEC, which
> reduces the MAP_FIXED_NOREPLACE coverage to only the first PT_LOAD:
>
> $ readelf -lW /usr/bin/gcc
> ...
> Program Headers:
> Type Offset VirtAddr PhysAddr FileSiz MemSiz
> ...
> ...
> LOAD 0x000000 0x4000000000000000 0x4000000000000000 0x00b5a0 0x00b5a0
> ...
> LOAD 0x00b5a0 0x600000000000b5a0 0x600000000000b5a0 0x0005ac 0x000710
> ...
> ...
> ^^^^^^^^ ^^^^^^^^^^^^^^^^^^ ^^^^^^^^ ^^^^^^^^
>
> File offset range : 0x000000-0x00bb4c
> 0x00bb4c bytes
>
> Virtual address range : 0x4000000000000000-0x600000000000bcb0
> 0x200000000000bcb0 bytes
>
> Ironically, this is the reverse of the problem that originally caused
> problems with ET_EXEC and MAP_FIXED_NOREPLACE: overlaps. This problem
> is
> with holes. Future work could restore full coverage if
> load_elf_binary()
> were to perform mappings in a separate phase from the loading (where
> it could resolve both overlaps and holes).
>
> Cc: Alexander Viro <[email protected]>
> Cc: Eric Biederman <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Reported-by: matoro <[email protected]>
> Reported-by: John Paul Adrian Glaubitz <[email protected]>
> Fixes: 5f501d555653 ("binfmt_elf: reintroduce using
> MAP_FIXED_NOREPLACE")
> Link:
> https://lore.kernel.org/r/[email protected]
> Cc: [email protected]
> Signed-off-by: Kees Cook <[email protected]>
> ---
> Here's the v5.16 backport.
> ---
> fs/binfmt_elf.c | 25 ++++++++++++++++++-------
> 1 file changed, 18 insertions(+), 7 deletions(-)
>
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index f8c7f26f1fbb..911a9e7044f4 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -1135,14 +1135,25 @@ static int load_elf_binary(struct linux_binprm
> *bprm)
> * is then page aligned.
> */
> load_bias = ELF_PAGESTART(load_bias - vaddr);
> - }
>
> - /*
> - * Calculate the entire size of the ELF mapping (total_size).
> - * (Note that load_addr_set is set to true later once the
> - * initial mapping is performed.)
> - */
> - if (!load_addr_set) {
> + /*
> + * Calculate the entire size of the ELF mapping
> + * (total_size), used for the initial mapping,
> + * due to first_pt_load which is set to false later
> + * once the initial mapping is performed.
> + *
> + * Note that this is only sensible when the LOAD
> + * segments are contiguous (or overlapping). If
> + * used for LOADs that are far apart, this would
> + * cause the holes between LOADs to be mapped,
> + * running the risk of having the mapping fail,
> + * as it would be larger than the ELF file itself.
> + *
> + * As a result, only ET_DYN does this, since
> + * some ET_EXEC (e.g. ia64) may have virtual
> + * memory holes between LOADs.
> + *
> + */
> total_size = total_mapping_size(elf_phdata,
> elf_ex->e_phnum);
> if (!total_size) {
This does the trick! Thank you so much!!
On Mon, Feb 28, 2022 at 05:14:26PM -0500, matoro wrote:
> On 2022-02-28 15:55, Kees Cook wrote:
> > Partially revert commit 5f501d555653 ("binfmt_elf: reintroduce using
> > MAP_FIXED_NOREPLACE").
> >
> > At least ia64 has ET_EXEC PT_LOAD segments that are not virtual-address
> > contiguous (but _are_ file-offset contiguous). This would result in
> > giant mapping attempts to cover the entire span, including the virtual
> > address range hole. Disable total_mapping_size for ET_EXEC, which
> > reduces the MAP_FIXED_NOREPLACE coverage to only the first PT_LOAD:
> >
> > $ readelf -lW /usr/bin/gcc
> > ...
> > Program Headers:
> > Type Offset VirtAddr PhysAddr FileSiz MemSiz
> > ...
> > ...
> > LOAD 0x000000 0x4000000000000000 0x4000000000000000 0x00b5a0 0x00b5a0
> > ...
> > LOAD 0x00b5a0 0x600000000000b5a0 0x600000000000b5a0 0x0005ac 0x000710
> > ...
> > ...
> > ^^^^^^^^ ^^^^^^^^^^^^^^^^^^ ^^^^^^^^ ^^^^^^^^
> >
> > File offset range : 0x000000-0x00bb4c
> > 0x00bb4c bytes
> >
> > Virtual address range : 0x4000000000000000-0x600000000000bcb0
> > 0x200000000000bcb0 bytes
> >
> > Ironically, this is the reverse of the problem that originally caused
> > problems with ET_EXEC and MAP_FIXED_NOREPLACE: overlaps. This problem is
> > with holes. Future work could restore full coverage if load_elf_binary()
> > were to perform mappings in a separate phase from the loading (where
> > it could resolve both overlaps and holes).
> >
> > Cc: Alexander Viro <[email protected]>
> > Cc: Eric Biederman <[email protected]>
> > Cc: [email protected]
> > Cc: [email protected]
> > Reported-by: matoro <[email protected]>
> > Reported-by: John Paul Adrian Glaubitz <[email protected]>
> > Fixes: 5f501d555653 ("binfmt_elf: reintroduce using
> > MAP_FIXED_NOREPLACE")
> > Link:
> > https://lore.kernel.org/r/[email protected]
> > Cc: [email protected]
> > Signed-off-by: Kees Cook <[email protected]>
> > ---
> > Here's the v5.16 backport.
> > ---
> > fs/binfmt_elf.c | 25 ++++++++++++++++++-------
> > 1 file changed, 18 insertions(+), 7 deletions(-)
> >
> > diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> > index f8c7f26f1fbb..911a9e7044f4 100644
> > --- a/fs/binfmt_elf.c
> > +++ b/fs/binfmt_elf.c
> > @@ -1135,14 +1135,25 @@ static int load_elf_binary(struct linux_binprm
> > *bprm)
> > * is then page aligned.
> > */
> > load_bias = ELF_PAGESTART(load_bias - vaddr);
> > - }
> >
> > - /*
> > - * Calculate the entire size of the ELF mapping (total_size).
> > - * (Note that load_addr_set is set to true later once the
> > - * initial mapping is performed.)
> > - */
> > - if (!load_addr_set) {
> > + /*
> > + * Calculate the entire size of the ELF mapping
> > + * (total_size), used for the initial mapping,
> > + * due to first_pt_load which is set to false later
> > + * once the initial mapping is performed.
> > + *
> > + * Note that this is only sensible when the LOAD
> > + * segments are contiguous (or overlapping). If
> > + * used for LOADs that are far apart, this would
> > + * cause the holes between LOADs to be mapped,
> > + * running the risk of having the mapping fail,
> > + * as it would be larger than the ELF file itself.
> > + *
> > + * As a result, only ET_DYN does this, since
> > + * some ET_EXEC (e.g. ia64) may have virtual
> > + * memory holes between LOADs.
> > + *
> > + */
> > total_size = total_mapping_size(elf_phdata,
> > elf_ex->e_phnum);
> > if (!total_size) {
>
> This does the trick! Thank you so much!!
Excellent; thank you for testing! I'll send this to Linus shortly.
--
Kees Cook
Hello!
On 2/28/22 21:55, Kees Cook wrote:
> Partially revert commit 5f501d555653 ("binfmt_elf: reintroduce using
> MAP_FIXED_NOREPLACE").
>
> At least ia64 has ET_EXEC PT_LOAD segments that are not virtual-address
> contiguous (but _are_ file-offset contiguous). This would result in
> giant mapping attempts to cover the entire span, including the virtual
> address range hole. Disable total_mapping_size for ET_EXEC, which
> reduces the MAP_FIXED_NOREPLACE coverage to only the first PT_LOAD:
>
> $ readelf -lW /usr/bin/gcc
> ...
> Program Headers:
> Type Offset VirtAddr PhysAddr FileSiz MemSiz ...
> ...
> LOAD 0x000000 0x4000000000000000 0x4000000000000000 0x00b5a0 0x00b5a0 ...
> LOAD 0x00b5a0 0x600000000000b5a0 0x600000000000b5a0 0x0005ac 0x000710 ...
> ...
> ^^^^^^^^ ^^^^^^^^^^^^^^^^^^ ^^^^^^^^ ^^^^^^^^
>
> File offset range : 0x000000-0x00bb4c
> 0x00bb4c bytes
>
> Virtual address range : 0x4000000000000000-0x600000000000bcb0
> 0x200000000000bcb0 bytes
>
> Ironically, this is the reverse of the problem that originally caused
> problems with ET_EXEC and MAP_FIXED_NOREPLACE: overlaps. This problem is
> with holes. Future work could restore full coverage if load_elf_binary()
> were to perform mappings in a separate phase from the loading (where
> it could resolve both overlaps and holes).
>
> Cc: Alexander Viro <[email protected]>
> Cc: Eric Biederman <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Reported-by: matoro <[email protected]>
> Reported-by: John Paul Adrian Glaubitz <[email protected]>
> Fixes: 5f501d555653 ("binfmt_elf: reintroduce using MAP_FIXED_NOREPLACE")
> Link: https://lore.kernel.org/r/[email protected]
> Cc: [email protected]
> Signed-off-by: Kees Cook <[email protected]>
> ---
> Here's the v5.16 backport.
> ---
> fs/binfmt_elf.c | 25 ++++++++++++++++++-------
> 1 file changed, 18 insertions(+), 7 deletions(-)
>
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index f8c7f26f1fbb..911a9e7044f4 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -1135,14 +1135,25 @@ static int load_elf_binary(struct linux_binprm *bprm)
> * is then page aligned.
> */
> load_bias = ELF_PAGESTART(load_bias - vaddr);
> - }
>
> - /*
> - * Calculate the entire size of the ELF mapping (total_size).
> - * (Note that load_addr_set is set to true later once the
> - * initial mapping is performed.)
> - */
> - if (!load_addr_set) {
> + /*
> + * Calculate the entire size of the ELF mapping
> + * (total_size), used for the initial mapping,
> + * due to first_pt_load which is set to false later
> + * once the initial mapping is performed.
> + *
> + * Note that this is only sensible when the LOAD
> + * segments are contiguous (or overlapping). If
> + * used for LOADs that are far apart, this would
> + * cause the holes between LOADs to be mapped,
> + * running the risk of having the mapping fail,
> + * as it would be larger than the ELF file itself.
> + *
> + * As a result, only ET_DYN does this, since
> + * some ET_EXEC (e.g. ia64) may have virtual
> + * memory holes between LOADs.
> + *
> + */
> total_size = total_mapping_size(elf_phdata,
> elf_ex->e_phnum);
> if (!total_size) {
I can confirm that this patch fixes the issue for me.
Tested-By: John Paul Adrian Glaubitz <[email protected]>
Thanks,
Adrian
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer - [email protected]
`. `' Freie Universitaet Berlin - [email protected]
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913