If /proc/<PID>/environ gets read before the envp[] array is fully set
up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
to read more bytes than are actually written, as env_start will already
be set but env_end will still be zero, making the range calculation
underflow, allowing to read beyond the end of what has been written.
Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
zero. It is, apparently, intentionally set last in create_*_tables().
This bug was found by the PaX size_overflow plugin that detected the
arithmetic underflow of 'this_len = env_end - (env_start + src)' when
env_end is still zero.
Reported-at: https://forums.grsecurity.net/viewtopic.php?f=3&t=4363
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=116461
Signed-off-by: Mathias Krause <[email protected]>
Cc: Emese Revfy <[email protected]>
Cc: Pax Team <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Mateusz Guzik <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Cc: Cyrill Gorcunov <[email protected]>
Cc: Jarod Wilson <[email protected]>
---
fs/proc/base.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 4f764c2ac1a5..45f2162e55b2 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -955,7 +955,8 @@ static ssize_t environ_read(struct file *file, char __user *buf,
struct mm_struct *mm = file->private_data;
unsigned long env_start, env_end;
- if (!mm)
+ /* Ensure the process spawned far enough to have an environment. */
+ if (!mm || !mm->env_end)
return 0;
page = (char *)__get_free_page(GFP_TEMPORARY);
--
1.7.10.4
On Thu, Apr 28, 2016 at 09:04:18PM +0200, Mathias Krause wrote:
> If /proc/<PID>/environ gets read before the envp[] array is fully set
> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
> to read more bytes than are actually written, as env_start will already
> be set but env_end will still be zero, making the range calculation
> underflow, allowing to read beyond the end of what has been written.
>
> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
> zero. It is, apparently, intentionally set last in create_*_tables().
>
> This bug was found by the PaX size_overflow plugin that detected the
> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
> env_end is still zero.
>
> Reported-at: https://forums.grsecurity.net/viewtopic.php?f=3&t=4363
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=116461
> ---
> fs/proc/base.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 4f764c2ac1a5..45f2162e55b2 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -955,7 +955,8 @@ static ssize_t environ_read(struct file *file, char __user *buf,
> struct mm_struct *mm = file->private_data;
> unsigned long env_start, env_end;
>
> - if (!mm)
> + /* Ensure the process spawned far enough to have an environment. */
> + if (!mm || !mm->env_end)
> return 0;
>
> page = (char *)__get_free_page(GFP_TEMPORARY);
In this case get_cmdline in mm/util.c should also be patched for
completness. It tests for arg_end, but later accesses env_end.
--
Mateusz Guzik
On Thu, Apr 28, 2016 at 09:04:18PM +0200, Mathias Krause wrote:
> If /proc/<PID>/environ gets read before the envp[] array is fully set
> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
> to read more bytes than are actually written, as env_start will already
> be set but env_end will still be zero, making the range calculation
> underflow, allowing to read beyond the end of what has been written.
>
> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
> zero. It is, apparently, intentionally set last in create_*_tables().
>
> This bug was found by the PaX size_overflow plugin that detected the
> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
> env_end is still zero.
>
> Reported-at: https://forums.grsecurity.net/viewtopic.php?f=3&t=4363
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=116461
> Signed-off-by: Mathias Krause <[email protected]>
> Cc: Emese Revfy <[email protected]>
> Cc: Pax Team <[email protected]>
> Cc: Al Viro <[email protected]>
> Cc: Mateusz Guzik <[email protected]>
> Cc: Alexey Dobriyan <[email protected]>
> Cc: Cyrill Gorcunov <[email protected]>
> Cc: Jarod Wilson <[email protected]>
> ---
> fs/proc/base.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 4f764c2ac1a5..45f2162e55b2 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -955,7 +955,8 @@ static ssize_t environ_read(struct file *file, char __user *buf,
> struct mm_struct *mm = file->private_data;
> unsigned long env_start, env_end;
>
> - if (!mm)
> + /* Ensure the process spawned far enough to have an environment. */
> + if (!mm || !mm->env_end)
> return 0;
At least in proc_pid_cmdline_read such test is done.
Acked-by: Cyrill Gorcunov <[email protected]>
On 28 April 2016 at 21:20, Mateusz Guzik <[email protected]> wrote:
> In this case get_cmdline in mm/util.c should also be patched for
> completness. It tests for arg_end, but later accesses env_end.
But it'll do this only when argv[] was modified from what the kernel
initially wrote, which, in turn, either requires the process to have
started executing and messing with it's own argv[] or another process
poking at it via ptrace(). In the former case env_end will be non-zero
already and I don't know if the latter case is actually possible, i.e.
if one can already attach to a process this early. If one can, then
yes, that place needs to be modified, too.
Thanks,
Mathias
On Thu, 28 Apr 2016 21:04:18 +0200 Mathias Krause <[email protected]> wrote:
> If /proc/<PID>/environ gets read before the envp[] array is fully set
> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
> to read more bytes than are actually written, as env_start will already
> be set but env_end will still be zero, making the range calculation
> underflow, allowing to read beyond the end of what has been written.
>
> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
> zero. It is, apparently, intentionally set last in create_*_tables().
>
> This bug was found by the PaX size_overflow plugin that detected the
> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
> env_end is still zero.
So what are the implications of this? From my reading, a craftily
constructed application could occasionally read arbitrarily large
amounts of kernel memory?
On Thu, 28 Apr 2016 21:04:18 +0200 Mathias Krause <[email protected]> wrote:
> If /proc/<PID>/environ gets read before the envp[] array is fully set
> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
> to read more bytes than are actually written, as env_start will already
> be set but env_end will still be zero, making the range calculation
> underflow, allowing to read beyond the end of what has been written.
>
> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
> zero. It is, apparently, intentionally set last in create_*_tables().
Also, if this is indeed our design then
a) the various create_*_tables() should have comments in there which
explain this subtlety to the reader. Or, better, they use a common
helper function for this readiness-signaling operation because..
b) we'll need some barriers there to ensure that the environ_read()
caller sees the create_*_tables() writes in the correct order.
> This bug was found by the PaX size_overflow plugin that detected the
> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
> env_end is still zero.
On 28 April 2016 at 23:26, Andrew Morton <[email protected]> wrote:
> On Thu, 28 Apr 2016 21:04:18 +0200 Mathias Krause <[email protected]> wrote:
>
>> If /proc/<PID>/environ gets read before the envp[] array is fully set
>> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
>> to read more bytes than are actually written, as env_start will already
>> be set but env_end will still be zero, making the range calculation
>> underflow, allowing to read beyond the end of what has been written.
>>
>> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
>> zero. It is, apparently, intentionally set last in create_*_tables().
>>
>> This bug was found by the PaX size_overflow plugin that detected the
>> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
>> env_end is still zero.
>
> So what are the implications of this? From my reading, a craftily
> constructed application could occasionally read arbitrarily large
> amounts of kernel memory?
I don't think access_remote_vm() is capable of that. So, the only
consequence is, userland trying to access /proc/<PID>/environ of a not
yet fully set up process may get inconsistent data as we're in the
middle of copying in the environment variables.
Regards,
Mathias
On 28 April 2016 at 23:30, Andrew Morton <[email protected]> wrote:
> On Thu, 28 Apr 2016 21:04:18 +0200 Mathias Krause <[email protected]> wrote:
>
>> If /proc/<PID>/environ gets read before the envp[] array is fully set
>> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
>> to read more bytes than are actually written, as env_start will already
>> be set but env_end will still be zero, making the range calculation
>> underflow, allowing to read beyond the end of what has been written.
>>
>> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
>> zero. It is, apparently, intentionally set last in create_*_tables().
>
> Also, if this is indeed our design then
>
> a) the various create_*_tables() should have comments in there which
> explain this subtlety to the reader. Or, better, they use a common
> helper function for this readiness-signaling operation because..
>
> b) we'll need some barriers there to ensure that the environ_read()
> caller sees the create_*_tables() writes in the correct order.
I totally agree that this kind of "synchronization" is rather fragile.
Adding comments won't help much, I fear. Rather a dedicated flag,
signaling "process ready for inspection" may be needed. So far, that's
what env_end is (ab-)used for.
Regards,
Mathias
On Thu, Apr 28, 2016 at 10:20 PM, Mateusz Guzik <[email protected]> wrote:
> On Thu, Apr 28, 2016 at 09:04:18PM +0200, Mathias Krause wrote:
>> If /proc/<PID>/environ gets read before the envp[] array is fully set
>> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
>> to read more bytes than are actually written, as env_start will already
>> be set but env_end will still be zero, making the range calculation
>> underflow, allowing to read beyond the end of what has been written.
>>
>> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
>> zero. It is, apparently, intentionally set last in create_*_tables().
>>
>> This bug was found by the PaX size_overflow plugin that detected the
>> arithmetic underflow of 'this_len = env_end - (env_start + src)' when
>> env_end is still zero.
>>
>> Reported-at: https://forums.grsecurity.net/viewtopic.php?f=3&t=4363
>> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=116461
>> ---
>> fs/proc/base.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/proc/base.c b/fs/proc/base.c
>> index 4f764c2ac1a5..45f2162e55b2 100644
>> --- a/fs/proc/base.c
>> +++ b/fs/proc/base.c
>> @@ -955,7 +955,8 @@ static ssize_t environ_read(struct file *file, char __user *buf,
>> struct mm_struct *mm = file->private_data;
>> unsigned long env_start, env_end;
>>
>> - if (!mm)
>> + /* Ensure the process spawned far enough to have an environment. */
>> + if (!mm || !mm->env_end)
>> return 0;
>>
>> page = (char *)__get_free_page(GFP_TEMPORARY);
>
> In this case get_cmdline in mm/util.c should also be patched for
> completness. It tests for arg_end, but later accesses env_end.
Sort of. get_cmdline() is only really used in audit code applied
to an exiting process which has cmdline setup long ago.
Should have rewrote /proc/*/environ as well... :-(
Alexey
On Fri, Apr 29, 2016 at 9:02 AM, Mathias Krause <[email protected]> wrote:
> On 28 April 2016 at 23:30, Andrew Morton <[email protected]> wrote:
>> On Thu, 28 Apr 2016 21:04:18 +0200 Mathias Krause <[email protected]> wrote:
>>
>>> If /proc/<PID>/environ gets read before the envp[] array is fully set
>>> up in create_{aout,elf,elf_fdpic,flat}_tables(), we might end up trying
>>> to read more bytes than are actually written, as env_start will already
>>> be set but env_end will still be zero, making the range calculation
>>> underflow, allowing to read beyond the end of what has been written.
>>>
>>> Fix this as it is done for /proc/<PID>/cmdline by testing env_end for
>>> zero. It is, apparently, intentionally set last in create_*_tables().
>>
>> Also, if this is indeed our design then
>>
>> a) the various create_*_tables() should have comments in there which
>> explain this subtlety to the reader. Or, better, they use a common
>> helper function for this readiness-signaling operation because..
>>
>> b) we'll need some barriers there to ensure that the environ_read()
>> caller sees the create_*_tables() writes in the correct order.
>
> I totally agree that this kind of "synchronization" is rather fragile.
> Adding comments won't help much, I fear. Rather a dedicated flag,
> signaling "process ready for inspection" may be needed. So far, that's
> what env_end is (ab-)used for.
If MM Cabal is OK with MMF_LOAD_BINARY_OK flag
applied at search_binary_handler(), it should work for /proc .
Alexey