vsprintf: protect kernel from panic due to non-canonical pointer dereference

Message ID 20221017191611.2577466-1-jane.chu@oracle.com
State New
Headers
Series vsprintf: protect kernel from panic due to non-canonical pointer dereference |

Commit Message

Jane Chu Oct. 17, 2022, 7:16 p.m. UTC
  While debugging a separate issue, it was found that an invalid string
pointer could very well contain a non-canical address, such as
0x7665645f63616465. In that case, this line of defense isn't enough
to protect the kernel from crashing due to general protection fault

	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
                return "(efault)";

So instead, use kern_addr_valid() to validate the string pointer.

Signed-off-by: Jane Chu <jane.chu@oracle.com>
---
 lib/vsprintf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
  

Comments

Andy Shevchenko Oct. 17, 2022, 7:25 p.m. UTC | #1
On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> While debugging a separate issue, it was found that an invalid string
> pointer could very well contain a non-canical address, such as
> 0x7665645f63616465. In that case, this line of defense isn't enough
> to protect the kernel from crashing due to general protection fault
> 
> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>                 return "(efault)";
> 
> So instead, use kern_addr_valid() to validate the string pointer.

How did you check that value of the (invalid string) pointer?
  
Jane Chu Oct. 17, 2022, 7:31 p.m. UTC | #2
On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>> While debugging a separate issue, it was found that an invalid string
>> pointer could very well contain a non-canical address, such as
>> 0x7665645f63616465. In that case, this line of defense isn't enough
>> to protect the kernel from crashing due to general protection fault
>>
>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>                  return "(efault)";
>>
>> So instead, use kern_addr_valid() to validate the string pointer.
> 
> How did you check that value of the (invalid string) pointer?
> 

In the bug scenario, the invalid string pointer was an out-of-bound 
string pointer. While the OOB referencing is fixed, the lingering issue 
is that the kernel ought to be able to protect itself, as the pointer 
contains a non-canonical address.  That said, I realized that not all 
architecture implement meaningful kern_addr_valid(), so this line
    if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
is still need.  I'll send v2.

thanks,
-jane

>
  
Petr Mladek Oct. 18, 2022, 12:45 p.m. UTC | #3
On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> > On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> >> While debugging a separate issue, it was found that an invalid string
> >> pointer could very well contain a non-canical address, such as
> >> 0x7665645f63616465. In that case, this line of defense isn't enough
> >> to protect the kernel from crashing due to general protection fault
> >>
> >> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> >>                  return "(efault)";
> >>
> >> So instead, use kern_addr_valid() to validate the string pointer.
> > 
> > How did you check that value of the (invalid string) pointer?
> > 
> 
> In the bug scenario, the invalid string pointer was an out-of-bound 
> string pointer. While the OOB referencing is fixed,

Could you please provide more details about the fixed OOB?
What exact vsprintf()/printk() call was broken and eventually
how it was fixed, please?

> the lingering issue
> is that the kernel ought to be able to protect itself, as the pointer 
> contains a non-canonical address.

Was the pointer used only by the vsprintf()?
Or was it accessed also by another code, please?

I wonder if this patch would prevent the crash or if the broken
kernel would crash later anyway.

> That said, I realized that not all
> architecture implement meaningful kern_addr_valid(), so this line
>     if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> is still need.  I'll send v2.

Please, add linux-mm@kvack.org into CC. I wonder if kern_addr_valid()
is safe to use anywhere, especially during early boot. I wonder if
it would make sense to implement it on all architectures.

Best Regards,
Petr
  
Jane Chu Oct. 18, 2022, 6:56 p.m. UTC | #4
On 10/18/2022 5:45 AM, Petr Mladek wrote:
> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>> While debugging a separate issue, it was found that an invalid string
>>>> pointer could very well contain a non-canical address, such as
>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>> to protect the kernel from crashing due to general protection fault
>>>>
>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>                   return "(efault)";
>>>>
>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>
>>> How did you check that value of the (invalid string) pointer?
>>>
>>
>> In the bug scenario, the invalid string pointer was an out-of-bound
>> string pointer. While the OOB referencing is fixed,
> 
> Could you please provide more details about the fixed OOB?
> What exact vsprintf()/printk() call was broken and eventually
> how it was fixed, please?

For sensitive reason, I'd like to avoid mentioning the specific name of 
the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
and describe the precise nature of the issue.

devX_attrY[] is a string array, declared and filled at compile time,
like
   const char const devX_attrY[] = {
	[ATTRY_A] = "Dev X AttributeY A",
	[ATTRY_B] = "Dev X AttributeY B",
	...
	[ATTRY_G] = "Dev X AttributeY G",
   }
such that, when user "cat /sys/devices/systems/.../attry_1",
"Dev X AttributeY B" will show up in the terminal.
That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.

The bug was that the index to the array was wrongfully produced,
leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the 
calculation and that is not an upstream fix.

> 
>> the lingering issue
>> is that the kernel ought to be able to protect itself, as the pointer
>> contains a non-canonical address.
> 
> Was the pointer used only by the vsprintf()?
> Or was it accessed also by another code, please?

The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
No other code uses the OOB pointer, verified both by code examination 
and test.

Here is a snippet of the crash backtrace from an instrumented kernel, 
scratched one line for sensitive reason -

crash> bt
PID: 3250   TASK: ffff9cb50fe23d80  CPU: 18  COMMAND: "cat"
  #0 [ffffc0bacf377998] machine_kexec at ffffffff9b06c7c1
  #1 [ffffc0bacf3779f8] __crash_kexec at ffffffff9b13bb52
  #2 [ffffc0bacf377ac8] crash_kexec at ffffffff9b13cdac
  #3 [ffffc0bacf377ae8] oops_end at ffffffff9b03357a
  #4 [ffffc0bacf377b10] die at ffffffff9b033c32
  #5 [ffffc0bacf377b40] do_general_protection at ffffffff9b030c52
  #6 [ffffc0bacf377b70] general_protection at ffffffff9ba03db4
     [exception RIP: string_nocheck+19]
     RIP: ffffffff9b87cc73  RSP: ffffc0bacf377c20  RFLAGS: 00010286
     RAX: 0000000000000000  RBX: ffff9da13fc17fff  RCX: ffff0a00ffffff04
     RDX: 726f635f63616465  RSI: ffff9da13fc17fff  RDI: ffffffffffffffff
     RBP: ffffc0bacf377c20   R8: ffff9da0bfd2f010   R9: ffff9da0bfc18000
     R10: 0000000000001000  R11: 0000000000000000  R12: 726f635f63616465
     R13: ffff0a00ffffff04  R14: ffffffff9c1a6a4f  R15: ffffffff9c1a6a4f
     ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
  #7 [ffffc0bacf377c28] string at ffffffff9b87ce98
  #8 [ffffc0bacf377c58] vsnprintf at ffffffff9b87efe3
  #9 [ffffc0bacf377cb8] sprintf at ffffffff9b87f506
#10 [ffffc0bacf377d18] <------------------------------>
#11 [ffffc0bacf377d28] dev_attr_show at ffffffff9b56d183
#12 [ffffc0bacf377d48] sysfs_kf_seq_show at ffffffff9b3272dc
#13 [ffffc0bacf377d68] kernfs_seq_show at ffffffff9b32576c
#14 [ffffc0bacf377d78] seq_read at ffffffff9b2be407
#15 [ffffc0bacf377de8] kernfs_fop_read at ffffffff9b325ffe
#16 [ffffc0bacf377e28] __vfs_read at ffffffff9b2940ea
#17 [ffffc0bacf377eb0] vfs_read at ffffffff9b2942ac
#18 [ffffc0bacf377ee0] sys_read at ffffffff9b29485c
#19 [ffffc0bacf377f28] do_syscall_64 at ffffffff9b003ca9
#20 [ffffc0bacf377f50] entry_SYSCALL_64_after_hwframe at ffffffff9ba001b1

crash> dis ffffffff9b87cc73
0xffffffff9b87cc73 <string_nocheck+19>: movzbl (%rdx),%r8d

and RDX: 726f635f63616465 was a non-canonical address.

After applying this patch to the instrumented kernel, instead of panic, 
the "cat" command produced  "(efault)"

> 
> I wonder if this patch would prevent the crash or if the broken
> kernel would crash later anyway.

A broken kernel has a different issue to be fixed, the upstream kernel 
isn't broken, it could just offer better protect in case a bug was 
introduced in future.

> 
>> That said, I realized that not all
>> architecture implement meaningful kern_addr_valid(), so this line
>>      if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>> is still need.  I'll send v2.
> 
> Please, add linux-mm@kvack.org into CC. 

Will do.

> I wonder if kern_addr_valid()
> is safe to use anywhere, especially during early boot. I wonder if
> it would make sense to implement it on all architectures.

On x86 architecture, kern_addr_valid() looks safe to me though, on 
several other architectures, it's defined (1).

> 
> Best Regards,
> Petr

Thanks!
-jane
  
Randy Dunlap Oct. 18, 2022, 7:28 p.m. UTC | #5
Hi--

On 10/18/22 11:56, Jane Chu wrote:
> On 10/18/2022 5:45 AM, Petr Mladek wrote:
>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>>> While debugging a separate issue, it was found that an invalid string
>>>>> pointer could very well contain a non-canical address, such as
>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>>> to protect the kernel from crashing due to general protection fault
>>>>>
>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>>                   return "(efault)";
>>>>>
>>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>>
>>>> How did you check that value of the (invalid string) pointer?
>>>>
>>>
>>> In the bug scenario, the invalid string pointer was an out-of-bound
>>> string pointer. While the OOB referencing is fixed,
>>
>> Could you please provide more details about the fixed OOB?
>> What exact vsprintf()/printk() call was broken and eventually
>> how it was fixed, please?
> 
> For sensitive reason, I'd like to avoid mentioning the specific name of 
> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> and describe the precise nature of the issue.
> 
> devX_attrY[] is a string array, declared and filled at compile time,
> like
>    const char const devX_attrY[] = {
> 	[ATTRY_A] = "Dev X AttributeY A",
> 	[ATTRY_B] = "Dev X AttributeY B",
> 	...
> 	[ATTRY_G] = "Dev X AttributeY G",
>    }
> such that, when user "cat /sys/devices/systems/.../attry_1",
> "Dev X AttributeY B" will show up in the terminal.
> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> 
> The bug was that the index to the array was wrongfully produced,
> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the 
> calculation and that is not an upstream fix.
> 
>>
>>> the lingering issue
>>> is that the kernel ought to be able to protect itself, as the pointer
>>> contains a non-canonical address.
>>
>> Was the pointer used only by the vsprintf()?
>> Or was it accessed also by another code, please?
> 
> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
> No other code uses the OOB pointer, verified both by code examination 
> and test.
> 
> Here is a snippet of the crash backtrace from an instrumented kernel, 
> scratched one line for sensitive reason -
> 
> crash> bt
> PID: 3250   TASK: ffff9cb50fe23d80  CPU: 18  COMMAND: "cat"
>   #0 [ffffc0bacf377998] machine_kexec at ffffffff9b06c7c1
>   #1 [ffffc0bacf3779f8] __crash_kexec at ffffffff9b13bb52
>   #2 [ffffc0bacf377ac8] crash_kexec at ffffffff9b13cdac
>   #3 [ffffc0bacf377ae8] oops_end at ffffffff9b03357a
>   #4 [ffffc0bacf377b10] die at ffffffff9b033c32
>   #5 [ffffc0bacf377b40] do_general_protection at ffffffff9b030c52
>   #6 [ffffc0bacf377b70] general_protection at ffffffff9ba03db4
>      [exception RIP: string_nocheck+19]
>      RIP: ffffffff9b87cc73  RSP: ffffc0bacf377c20  RFLAGS: 00010286
>      RAX: 0000000000000000  RBX: ffff9da13fc17fff  RCX: ffff0a00ffffff04
>      RDX: 726f635f63616465  RSI: ffff9da13fc17fff  RDI: ffffffffffffffff
>      RBP: ffffc0bacf377c20   R8: ffff9da0bfd2f010   R9: ffff9da0bfc18000
>      R10: 0000000000001000  R11: 0000000000000000  R12: 726f635f63616465
>      R13: ffff0a00ffffff04  R14: ffffffff9c1a6a4f  R15: ffffffff9c1a6a4f
>      ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
>   #7 [ffffc0bacf377c28] string at ffffffff9b87ce98
>   #8 [ffffc0bacf377c58] vsnprintf at ffffffff9b87efe3
>   #9 [ffffc0bacf377cb8] sprintf at ffffffff9b87f506
> #10 [ffffc0bacf377d18] <------------------------------>
> #11 [ffffc0bacf377d28] dev_attr_show at ffffffff9b56d183
> #12 [ffffc0bacf377d48] sysfs_kf_seq_show at ffffffff9b3272dc
> #13 [ffffc0bacf377d68] kernfs_seq_show at ffffffff9b32576c
> #14 [ffffc0bacf377d78] seq_read at ffffffff9b2be407
> #15 [ffffc0bacf377de8] kernfs_fop_read at ffffffff9b325ffe
> #16 [ffffc0bacf377e28] __vfs_read at ffffffff9b2940ea
> #17 [ffffc0bacf377eb0] vfs_read at ffffffff9b2942ac
> #18 [ffffc0bacf377ee0] sys_read at ffffffff9b29485c
> #19 [ffffc0bacf377f28] do_syscall_64 at ffffffff9b003ca9
> #20 [ffffc0bacf377f50] entry_SYSCALL_64_after_hwframe at ffffffff9ba001b1
> 
> crash> dis ffffffff9b87cc73
> 0xffffffff9b87cc73 <string_nocheck+19>: movzbl (%rdx),%r8d
> 
> and RDX: 726f635f63616465 was a non-canonical address.
> 
> After applying this patch to the instrumented kernel, instead of panic, 
> the "cat" command produced  "(efault)"
> 
>>
>> I wonder if this patch would prevent the crash or if the broken
>> kernel would crash later anyway.
> 
> A broken kernel has a different issue to be fixed, the upstream kernel 
> isn't broken, it could just offer better protect in case a bug was 
> introduced in future.
> 
>>
>>> That said, I realized that not all
>>> architecture implement meaningful kern_addr_valid(), so this line
>>>      if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>> is still need.  I'll send v2.
>>
>> Please, add linux-mm@kvack.org into CC. 
> 
> Will do.
> 
>> I wonder if kern_addr_valid()
>> is safe to use anywhere, especially during early boot. I wonder if
>> it would make sense to implement it on all architectures.
> 
> On x86 architecture, kern_addr_valid() looks safe to me though, on 
> several other architectures, it's defined (1).

You might want to compare this patch, which seems to have some support:

https://lore.kernel.org/lkml/20221018074014.185687-1-wangkefeng.wang@huawei.com/
  
Jane Chu Oct. 18, 2022, 7:58 p.m. UTC | #6
On 10/18/2022 12:28 PM, Randy Dunlap wrote:
> Hi--
> 
[..]
>>>> That said, I realized that not all
>>>> architecture implement meaningful kern_addr_valid(), so this line
>>>>       if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>> is still need.  I'll send v2.
>>>
>>> Please, add linux-mm@kvack.org into CC.
>>
>> Will do.
>>
>>> I wonder if kern_addr_valid()
>>> is safe to use anywhere, especially during early boot. I wonder if
>>> it would make sense to implement it on all architectures.
>>
>> On x86 architecture, kern_addr_valid() looks safe to me though, on
>> several other architectures, it's defined (1).
> 
> You might want to compare this patch, which seems to have some support:
> 
> https://lore.kernel.org/lkml/20221018074014.185687-1-wangkefeng.wang@huawei.com/
> 

Thank you for alerting me, appreciated!

The patch comment says "copy_from_kernel_nofault() which could check 
whether the address is a valid kernel address, so no need 
kern_addr_valid()", I'm afraid copy_from_kernel_nofault() is more of a 
heavy hammer, and less appropriate for this patch.  I'll take a closer 
look before responding to the submitter.

thanks!
-jane
  
Andy Shevchenko Oct. 18, 2022, 8:07 p.m. UTC | #7
On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
> On 10/18/2022 5:45 AM, Petr Mladek wrote:
> > On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> >> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> >>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> >>>> While debugging a separate issue, it was found that an invalid string
> >>>> pointer could very well contain a non-canical address, such as
> >>>> 0x7665645f63616465. In that case, this line of defense isn't enough
> >>>> to protect the kernel from crashing due to general protection fault
> >>>>
> >>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> >>>>                   return "(efault)";
> >>>>
> >>>> So instead, use kern_addr_valid() to validate the string pointer.
> >>>
> >>> How did you check that value of the (invalid string) pointer?
> >>>
> >>
> >> In the bug scenario, the invalid string pointer was an out-of-bound
> >> string pointer. While the OOB referencing is fixed,
> > 
> > Could you please provide more details about the fixed OOB?
> > What exact vsprintf()/printk() call was broken and eventually
> > how it was fixed, please?
> 
> For sensitive reason, I'd like to avoid mentioning the specific name of 
> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> and describe the precise nature of the issue.
> 
> devX_attrY[] is a string array, declared and filled at compile time,
> like
>    const char const devX_attrY[] = {
> 	[ATTRY_A] = "Dev X AttributeY A",
> 	[ATTRY_B] = "Dev X AttributeY B",
> 	...
> 	[ATTRY_G] = "Dev X AttributeY G",
>    }
> such that, when user "cat /sys/devices/systems/.../attry_1",
> "Dev X AttributeY B" will show up in the terminal.
> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> 
> The bug was that the index to the array was wrongfully produced,
> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the 
> calculation and that is not an upstream fix.
> 
> > 
> >> the lingering issue
> >> is that the kernel ought to be able to protect itself, as the pointer
> >> contains a non-canonical address.
> > 
> > Was the pointer used only by the vsprintf()?
> > Or was it accessed also by another code, please?
> 
> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
> No other code uses the OOB pointer, verified both by code examination 
> and test.

So, then the vsprintf() is _the_ point to crash and why should we hide that?
Because of the crash you found the culprit, right? The efault will hide very
important details.

So to me it sounds like I like this change less and less...

> Here is a snippet of the crash backtrace from an instrumented kernel, 
> scratched one line for sensitive reason -
> 
> crash> bt
> PID: 3250   TASK: ffff9cb50fe23d80  CPU: 18  COMMAND: "cat"
>   #0 [ffffc0bacf377998] machine_kexec at ffffffff9b06c7c1
>   #1 [ffffc0bacf3779f8] __crash_kexec at ffffffff9b13bb52
>   #2 [ffffc0bacf377ac8] crash_kexec at ffffffff9b13cdac
>   #3 [ffffc0bacf377ae8] oops_end at ffffffff9b03357a
>   #4 [ffffc0bacf377b10] die at ffffffff9b033c32
>   #5 [ffffc0bacf377b40] do_general_protection at ffffffff9b030c52
>   #6 [ffffc0bacf377b70] general_protection at ffffffff9ba03db4
>      [exception RIP: string_nocheck+19]
>      RIP: ffffffff9b87cc73  RSP: ffffc0bacf377c20  RFLAGS: 00010286
>      RAX: 0000000000000000  RBX: ffff9da13fc17fff  RCX: ffff0a00ffffff04
>      RDX: 726f635f63616465  RSI: ffff9da13fc17fff  RDI: ffffffffffffffff
>      RBP: ffffc0bacf377c20   R8: ffff9da0bfd2f010   R9: ffff9da0bfc18000
>      R10: 0000000000001000  R11: 0000000000000000  R12: 726f635f63616465
>      R13: ffff0a00ffffff04  R14: ffffffff9c1a6a4f  R15: ffffffff9c1a6a4f
>      ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
>   #7 [ffffc0bacf377c28] string at ffffffff9b87ce98
>   #8 [ffffc0bacf377c58] vsnprintf at ffffffff9b87efe3
>   #9 [ffffc0bacf377cb8] sprintf at ffffffff9b87f506
> #10 [ffffc0bacf377d18] <------------------------------>
> #11 [ffffc0bacf377d28] dev_attr_show at ffffffff9b56d183
> #12 [ffffc0bacf377d48] sysfs_kf_seq_show at ffffffff9b3272dc
> #13 [ffffc0bacf377d68] kernfs_seq_show at ffffffff9b32576c
> #14 [ffffc0bacf377d78] seq_read at ffffffff9b2be407
> #15 [ffffc0bacf377de8] kernfs_fop_read at ffffffff9b325ffe
> #16 [ffffc0bacf377e28] __vfs_read at ffffffff9b2940ea
> #17 [ffffc0bacf377eb0] vfs_read at ffffffff9b2942ac
> #18 [ffffc0bacf377ee0] sys_read at ffffffff9b29485c
> #19 [ffffc0bacf377f28] do_syscall_64 at ffffffff9b003ca9
> #20 [ffffc0bacf377f50] entry_SYSCALL_64_after_hwframe at ffffffff9ba001b1
> 
> crash> dis ffffffff9b87cc73
> 0xffffffff9b87cc73 <string_nocheck+19>: movzbl (%rdx),%r8d
> 
> and RDX: 726f635f63616465 was a non-canonical address.
> 
> After applying this patch to the instrumented kernel, instead of panic, 
> the "cat" command produced  "(efault)"
> 
> > 
> > I wonder if this patch would prevent the crash or if the broken
> > kernel would crash later anyway.
> 
> A broken kernel has a different issue to be fixed, the upstream kernel 
> isn't broken, it could just offer better protect in case a bug was 
> introduced in future.
  
Jane Chu Oct. 18, 2022, 8:30 p.m. UTC | #8
On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
> On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
>> On 10/18/2022 5:45 AM, Petr Mladek wrote:
>>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>>>> While debugging a separate issue, it was found that an invalid string
>>>>>> pointer could very well contain a non-canical address, such as
>>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>>>> to protect the kernel from crashing due to general protection fault
>>>>>>
>>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>>>                    return "(efault)";
>>>>>>
>>>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>>>
>>>>> How did you check that value of the (invalid string) pointer?
>>>>>
>>>>
>>>> In the bug scenario, the invalid string pointer was an out-of-bound
>>>> string pointer. While the OOB referencing is fixed,
>>>
>>> Could you please provide more details about the fixed OOB?
>>> What exact vsprintf()/printk() call was broken and eventually
>>> how it was fixed, please?
>>
>> For sensitive reason, I'd like to avoid mentioning the specific name of
>> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
>> and describe the precise nature of the issue.
>>
>> devX_attrY[] is a string array, declared and filled at compile time,
>> like
>>     const char const devX_attrY[] = {
>> 	[ATTRY_A] = "Dev X AttributeY A",
>> 	[ATTRY_B] = "Dev X AttributeY B",
>> 	...
>> 	[ATTRY_G] = "Dev X AttributeY G",
>>     }
>> such that, when user "cat /sys/devices/systems/.../attry_1",
>> "Dev X AttributeY B" will show up in the terminal.
>> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
>>
>> The bug was that the index to the array was wrongfully produced,
>> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
>> calculation and that is not an upstream fix.
>>
>>>
>>>> the lingering issue
>>>> is that the kernel ought to be able to protect itself, as the pointer
>>>> contains a non-canonical address.
>>>
>>> Was the pointer used only by the vsprintf()?
>>> Or was it accessed also by another code, please?
>>
>> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
>> No other code uses the OOB pointer, verified both by code examination
>> and test.
> 
> So, then the vsprintf() is _the_ point to crash and why should we hide that?
> Because of the crash you found the culprit, right? The efault will hide very
> important details.
> 
> So to me it sounds like I like this change less and less...

What about the existing check
  	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
                     return "(efault)";
?

In an experiment just to print the raw OOB pointer values, I saw below
(the devX attrY stuff are substitutes of the real attributes, other 
values and strings are verbatim copy from "dmesg"):

[ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
[ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
[ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
[ 3002.772349] devX_attrY[29]: (0) (null)
[ 3002.772350] devX_attrY[30]: (0) (null)
[ 3002.772351] devX_attrY[31]: (0) (null)
[ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
[ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
[ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
[ 3002.772355] devX_attrY[35]: (746e75) (einval)

where starting from index 29 are all OOB pointers.

As you can see, if the OOBs are NULL, "(null)" was printed due to the 
existing checking, but when the OOBs are turned to non-canonical which
is detectable, the fact the pointer value deviates from
   (ffffffff84d60aee + 4 * sizeof(void *))
evidently shown that the OOBs are detectable.

The question then is why should the non-canonical OOBs be treated 
differently from NULL and ERR_VALUE?

Thanks,
-jane
  
Andy Shevchenko Oct. 18, 2022, 8:49 p.m. UTC | #9
On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
> On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
> > On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
> >> On 10/18/2022 5:45 AM, Petr Mladek wrote:
> >>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> >>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> >>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> >>>>>> While debugging a separate issue, it was found that an invalid string
> >>>>>> pointer could very well contain a non-canical address, such as
> >>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
> >>>>>> to protect the kernel from crashing due to general protection fault
> >>>>>>
> >>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> >>>>>>                    return "(efault)";
> >>>>>>
> >>>>>> So instead, use kern_addr_valid() to validate the string pointer.
> >>>>>
> >>>>> How did you check that value of the (invalid string) pointer?
> >>>>>
> >>>>
> >>>> In the bug scenario, the invalid string pointer was an out-of-bound
> >>>> string pointer. While the OOB referencing is fixed,
> >>>
> >>> Could you please provide more details about the fixed OOB?
> >>> What exact vsprintf()/printk() call was broken and eventually
> >>> how it was fixed, please?
> >>
> >> For sensitive reason, I'd like to avoid mentioning the specific name of
> >> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> >> and describe the precise nature of the issue.
> >>
> >> devX_attrY[] is a string array, declared and filled at compile time,
> >> like
> >>     const char const devX_attrY[] = {
> >> 	[ATTRY_A] = "Dev X AttributeY A",
> >> 	[ATTRY_B] = "Dev X AttributeY B",
> >> 	...
> >> 	[ATTRY_G] = "Dev X AttributeY G",
> >>     }
> >> such that, when user "cat /sys/devices/systems/.../attry_1",
> >> "Dev X AttributeY B" will show up in the terminal.
> >> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> >>
> >> The bug was that the index to the array was wrongfully produced,
> >> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
> >> calculation and that is not an upstream fix.
> >>
> >>>
> >>>> the lingering issue
> >>>> is that the kernel ought to be able to protect itself, as the pointer
> >>>> contains a non-canonical address.
> >>>
> >>> Was the pointer used only by the vsprintf()?
> >>> Or was it accessed also by another code, please?
> >>
> >> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
> >> No other code uses the OOB pointer, verified both by code examination
> >> and test.
> > 
> > So, then the vsprintf() is _the_ point to crash and why should we hide that?
> > Because of the crash you found the culprit, right? The efault will hide very
> > important details.
> > 
> > So to me it sounds like I like this change less and less...
> 
> What about the existing check
>   	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>                      return "(efault)";
> ?

Because it's _special_. We know that First page is equivalent to a NULL pointer
and the last one is dedicated for so called error pointers. There are no more
special exceptions to the addresses in the Linux kernel (I don't talk about
alignment requirements by the certain architectures).

> In an experiment just to print the raw OOB pointer values, I saw below
> (the devX attrY stuff are substitutes of the real attributes, other 
> values and strings are verbatim copy from "dmesg"):
> 
> [ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
> [ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
> [ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
> [ 3002.772349] devX_attrY[29]: (0) (null)
> [ 3002.772350] devX_attrY[30]: (0) (null)
> [ 3002.772351] devX_attrY[31]: (0) (null)
> [ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
> [ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
> [ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
> [ 3002.772355] devX_attrY[35]: (746e75) (einval)
> 
> where starting from index 29 are all OOB pointers.
> 
> As you can see, if the OOBs are NULL, "(null)" was printed due to the 
> existing checking, but when the OOBs are turned to non-canonical which
> is detectable, the fact the pointer value deviates from
>    (ffffffff84d60aee + 4 * sizeof(void *))
> evidently shown that the OOBs are detectable.
> 
> The question then is why should the non-canonical OOBs be treated 
> differently from NULL and ERR_VALUE?

Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
to see a bug as early as possible?
  
Haakon Bugge Oct. 19, 2022, 10:43 a.m. UTC | #10
> On 18 Oct 2022, at 22:49, Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> 
> On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
>> On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
>>> On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
>>>> On 10/18/2022 5:45 AM, Petr Mladek wrote:
>>>>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>>>>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>>>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>>>>>> While debugging a separate issue, it was found that an invalid string
>>>>>>>> pointer could very well contain a non-canical address, such as
>>>>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>>>>>> to protect the kernel from crashing due to general protection fault
>>>>>>>> 
>>>>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>>>>>                   return "(efault)";
>>>>>>>> 
>>>>>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>>>>> 
>>>>>>> How did you check that value of the (invalid string) pointer?
>>>>>>> 
>>>>>> 
>>>>>> In the bug scenario, the invalid string pointer was an out-of-bound
>>>>>> string pointer. While the OOB referencing is fixed,
>>>>> 
>>>>> Could you please provide more details about the fixed OOB?
>>>>> What exact vsprintf()/printk() call was broken and eventually
>>>>> how it was fixed, please?
>>>> 
>>>> For sensitive reason, I'd like to avoid mentioning the specific name of
>>>> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
>>>> and describe the precise nature of the issue.
>>>> 
>>>> devX_attrY[] is a string array, declared and filled at compile time,
>>>> like
>>>>    const char const devX_attrY[] = {
>>>> 	[ATTRY_A] = "Dev X AttributeY A",
>>>> 	[ATTRY_B] = "Dev X AttributeY B",
>>>> 	...
>>>> 	[ATTRY_G] = "Dev X AttributeY G",
>>>>    }
>>>> such that, when user "cat /sys/devices/systems/.../attry_1",
>>>> "Dev X AttributeY B" will show up in the terminal.
>>>> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
>>>> 
>>>> The bug was that the index to the array was wrongfully produced,
>>>> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
>>>> calculation and that is not an upstream fix.
>>>> 
>>>>> 
>>>>>> the lingering issue
>>>>>> is that the kernel ought to be able to protect itself, as the pointer
>>>>>> contains a non-canonical address.
>>>>> 
>>>>> Was the pointer used only by the vsprintf()?
>>>>> Or was it accessed also by another code, please?
>>>> 
>>>> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
>>>> No other code uses the OOB pointer, verified both by code examination
>>>> and test.
>>> 
>>> So, then the vsprintf() is _the_ point to crash and why should we hide that?
>>> Because of the crash you found the culprit, right? The efault will hide very
>>> important details.
>>> 
>>> So to me it sounds like I like this change less and less...
>> 
>> What about the existing check
>>  	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>                     return "(efault)";
>> ?
> 
> Because it's _special_. We know that First page is equivalent to a NULL pointer
> and the last one is dedicated for so called error pointers. There are no more
> special exceptions to the addresses in the Linux kernel (I don't talk about
> alignment requirements by the certain architectures).
> 
>> In an experiment just to print the raw OOB pointer values, I saw below
>> (the devX attrY stuff are substitutes of the real attributes, other 
>> values and strings are verbatim copy from "dmesg"):
>> 
>> [ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
>> [ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
>> [ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
>> [ 3002.772349] devX_attrY[29]: (0) (null)
>> [ 3002.772350] devX_attrY[30]: (0) (null)
>> [ 3002.772351] devX_attrY[31]: (0) (null)
>> [ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
>> [ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
>> [ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
>> [ 3002.772355] devX_attrY[35]: (746e75) (einval)
>> 
>> where starting from index 29 are all OOB pointers.
>> 
>> As you can see, if the OOBs are NULL, "(null)" was printed due to the 
>> existing checking, but when the OOBs are turned to non-canonical which
>> is detectable, the fact the pointer value deviates from
>>   (ffffffff84d60aee + 4 * sizeof(void *))
>> evidently shown that the OOBs are detectable.
>> 
>> The question then is why should the non-canonical OOBs be treated 
>> differently from NULL and ERR_VALUE?
> 
> Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> to see a bug as early as possible?

If you follow that argument, why doesn't the kernel crash when the pointer is, e.g., a NULL pointer? According to you, shouldn't it crash a early as possible in that case also?


Thxs, HÃ¥kon
  
Andy Shevchenko Oct. 19, 2022, 11:25 a.m. UTC | #11
On Wed, Oct 19, 2022 at 10:43:07AM +0000, Haakon Bugge wrote:
> > On 18 Oct 2022, at 22:49, Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> > On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:

...

> > Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> > to see a bug as early as possible?
> 
> If you follow that argument, why doesn't the kernel crash when the pointer
> is, e.g., a NULL pointer? According to you, shouldn't it crash a early as
> possible in that case also?

Because it is _special_. It's not just an invalid pointer. There may be
very well good cases where we supply (valid!) NULL pointers to the printf().
  
Jane Chu Oct. 19, 2022, 6:36 p.m. UTC | #12
On 10/18/2022 1:49 PM, Andy Shevchenko wrote:
> On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
>> On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
>>> On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
>>>> On 10/18/2022 5:45 AM, Petr Mladek wrote:
>>>>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>>>>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>>>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>>>>>> While debugging a separate issue, it was found that an invalid string
>>>>>>>> pointer could very well contain a non-canical address, such as
>>>>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>>>>>> to protect the kernel from crashing due to general protection fault
>>>>>>>>
>>>>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>>>>>                     return "(efault)";
>>>>>>>>
>>>>>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>>>>>
>>>>>>> How did you check that value of the (invalid string) pointer?
>>>>>>>
>>>>>>
>>>>>> In the bug scenario, the invalid string pointer was an out-of-bound
>>>>>> string pointer. While the OOB referencing is fixed,
>>>>>
>>>>> Could you please provide more details about the fixed OOB?
>>>>> What exact vsprintf()/printk() call was broken and eventually
>>>>> how it was fixed, please?
>>>>
>>>> For sensitive reason, I'd like to avoid mentioning the specific name of
>>>> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
>>>> and describe the precise nature of the issue.
>>>>
>>>> devX_attrY[] is a string array, declared and filled at compile time,
>>>> like
>>>>      const char const devX_attrY[] = {
>>>> 	[ATTRY_A] = "Dev X AttributeY A",
>>>> 	[ATTRY_B] = "Dev X AttributeY B",
>>>> 	...
>>>> 	[ATTRY_G] = "Dev X AttributeY G",
>>>>      }
>>>> such that, when user "cat /sys/devices/systems/.../attry_1",
>>>> "Dev X AttributeY B" will show up in the terminal.
>>>> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
>>>>
>>>> The bug was that the index to the array was wrongfully produced,
>>>> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
>>>> calculation and that is not an upstream fix.
>>>>
>>>>>
>>>>>> the lingering issue
>>>>>> is that the kernel ought to be able to protect itself, as the pointer
>>>>>> contains a non-canonical address.
>>>>>
>>>>> Was the pointer used only by the vsprintf()?
>>>>> Or was it accessed also by another code, please?
>>>>
>>>> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
>>>> No other code uses the OOB pointer, verified both by code examination
>>>> and test.
>>>
>>> So, then the vsprintf() is _the_ point to crash and why should we hide that?
>>> Because of the crash you found the culprit, right? The efault will hide very
>>> important details.
>>>
>>> So to me it sounds like I like this change less and less...
>>
>> What about the existing check
>>    	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>                       return "(efault)";
>> ?
> 
> Because it's _special_. We know that First page is equivalent to a NULL pointer
> and the last one is dedicated for so called error pointers. There are no more
> special exceptions to the addresses in the Linux kernel (I don't talk about
> alignment requirements by the certain architectures).
> 
>> In an experiment just to print the raw OOB pointer values, I saw below
>> (the devX attrY stuff are substitutes of the real attributes, other
>> values and strings are verbatim copy from "dmesg"):
>>
>> [ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
>> [ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
>> [ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
>> [ 3002.772349] devX_attrY[29]: (0) (null)
>> [ 3002.772350] devX_attrY[30]: (0) (null)
>> [ 3002.772351] devX_attrY[31]: (0) (null)
>> [ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
>> [ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
>> [ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
>> [ 3002.772355] devX_attrY[35]: (746e75) (einval)
>>
>> where starting from index 29 are all OOB pointers.
>>
>> As you can see, if the OOBs are NULL, "(null)" was printed due to the
>> existing checking, but when the OOBs are turned to non-canonical which
>> is detectable, the fact the pointer value deviates from
>>     (ffffffff84d60aee + 4 * sizeof(void *))
>> evidently shown that the OOBs are detectable.
>>
>> The question then is why should the non-canonical OOBs be treated
>> differently from NULL and ERR_VALUE?
> 
> Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> to see a bug as early as possible?
> 

If the purpose is to see the bug as early as possible, then getting
"(efault)" from reading sysfs attribute would serve the purpose, right?

The fact an OOB pointer has already being turned into either NULL or
non-canonical value implies that *if* kernel code other than
vsprintf() references the pointer, it'll crash else where; but *if* no
other code referencing the pointer, why crash?

thanks,
-jane
  
Andy Shevchenko Oct. 19, 2022, 7:26 p.m. UTC | #13
On Wed, Oct 19, 2022 at 06:36:07PM +0000, Jane Chu wrote:
> On 10/18/2022 1:49 PM, Andy Shevchenko wrote:
> > On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
> >> On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
> >>> On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
> >>>> On 10/18/2022 5:45 AM, Petr Mladek wrote:
> >>>>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> >>>>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> >>>>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> >>>>>>>> While debugging a separate issue, it was found that an invalid string
> >>>>>>>> pointer could very well contain a non-canical address, such as
> >>>>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
> >>>>>>>> to protect the kernel from crashing due to general protection fault
> >>>>>>>>
> >>>>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> >>>>>>>>                     return "(efault)";
> >>>>>>>>
> >>>>>>>> So instead, use kern_addr_valid() to validate the string pointer.
> >>>>>>>
> >>>>>>> How did you check that value of the (invalid string) pointer?
> >>>>>>>
> >>>>>>
> >>>>>> In the bug scenario, the invalid string pointer was an out-of-bound
> >>>>>> string pointer. While the OOB referencing is fixed,
> >>>>>
> >>>>> Could you please provide more details about the fixed OOB?
> >>>>> What exact vsprintf()/printk() call was broken and eventually
> >>>>> how it was fixed, please?
> >>>>
> >>>> For sensitive reason, I'd like to avoid mentioning the specific name of
> >>>> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> >>>> and describe the precise nature of the issue.
> >>>>
> >>>> devX_attrY[] is a string array, declared and filled at compile time,
> >>>> like
> >>>>      const char const devX_attrY[] = {
> >>>> 	[ATTRY_A] = "Dev X AttributeY A",
> >>>> 	[ATTRY_B] = "Dev X AttributeY B",
> >>>> 	...
> >>>> 	[ATTRY_G] = "Dev X AttributeY G",
> >>>>      }
> >>>> such that, when user "cat /sys/devices/systems/.../attry_1",
> >>>> "Dev X AttributeY B" will show up in the terminal.
> >>>> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> >>>>
> >>>> The bug was that the index to the array was wrongfully produced,
> >>>> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
> >>>> calculation and that is not an upstream fix.
> >>>>
> >>>>>
> >>>>>> the lingering issue
> >>>>>> is that the kernel ought to be able to protect itself, as the pointer
> >>>>>> contains a non-canonical address.
> >>>>>
> >>>>> Was the pointer used only by the vsprintf()?
> >>>>> Or was it accessed also by another code, please?
> >>>>
> >>>> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
> >>>> No other code uses the OOB pointer, verified both by code examination
> >>>> and test.
> >>>
> >>> So, then the vsprintf() is _the_ point to crash and why should we hide that?
> >>> Because of the crash you found the culprit, right? The efault will hide very
> >>> important details.
> >>>
> >>> So to me it sounds like I like this change less and less...
> >>
> >> What about the existing check
> >>    	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> >>                       return "(efault)";
> >> ?
> > 
> > Because it's _special_. We know that First page is equivalent to a NULL pointer
> > and the last one is dedicated for so called error pointers. There are no more
> > special exceptions to the addresses in the Linux kernel (I don't talk about
> > alignment requirements by the certain architectures).
> > 
> >> In an experiment just to print the raw OOB pointer values, I saw below
> >> (the devX attrY stuff are substitutes of the real attributes, other
> >> values and strings are verbatim copy from "dmesg"):
> >>
> >> [ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
> >> [ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
> >> [ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
> >> [ 3002.772349] devX_attrY[29]: (0) (null)
> >> [ 3002.772350] devX_attrY[30]: (0) (null)
> >> [ 3002.772351] devX_attrY[31]: (0) (null)
> >> [ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
> >> [ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
> >> [ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
> >> [ 3002.772355] devX_attrY[35]: (746e75) (einval)
> >>
> >> where starting from index 29 are all OOB pointers.
> >>
> >> As you can see, if the OOBs are NULL, "(null)" was printed due to the
> >> existing checking, but when the OOBs are turned to non-canonical which
> >> is detectable, the fact the pointer value deviates from
> >>     (ffffffff84d60aee + 4 * sizeof(void *))
> >> evidently shown that the OOBs are detectable.
> >>
> >> The question then is why should the non-canonical OOBs be treated
> >> differently from NULL and ERR_VALUE?
> > 
> > Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> > to see a bug as early as possible?
> > 
> 
> If the purpose is to see the bug as early as possible, then getting
> "(efault)" from reading sysfs attribute would serve the purpose, right?
> 
> The fact an OOB pointer has already being turned into either NULL or
> non-canonical value implies that *if* kernel code other than
> vsprintf() references the pointer, it'll crash else where;

No, not the case for error pointers and NULL.

> but *if* no
> other code referencing the pointer, why crash?

Because how else you can see the bug?! The trace will give you essential
information about registers, etc that gives you a hint what the _cause_ of the
crash. And we need that cause. The "(efault)" has not even a bit close to what
crash gives us.

So, this is my last message in the discussion.

Here is a formal NAK. Up to maintainers to decide what to do with this.
  
Jane Chu Oct. 19, 2022, 8:16 p.m. UTC | #14
On 10/19/2022 12:26 PM, Andy Shevchenko wrote:
> On Wed, Oct 19, 2022 at 06:36:07PM +0000, Jane Chu wrote:
>> On 10/18/2022 1:49 PM, Andy Shevchenko wrote:
>>> On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
>>>> On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
>>>>> On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
>>>>>> On 10/18/2022 5:45 AM, Petr Mladek wrote:
>>>>>>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
>>>>>>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
>>>>>>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
>>>>>>>>>> While debugging a separate issue, it was found that an invalid string
>>>>>>>>>> pointer could very well contain a non-canical address, such as
>>>>>>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>>>>>>>> to protect the kernel from crashing due to general protection fault
>>>>>>>>>>
>>>>>>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>>>>>>>                      return "(efault)";
>>>>>>>>>>
>>>>>>>>>> So instead, use kern_addr_valid() to validate the string pointer.
>>>>>>>>>
>>>>>>>>> How did you check that value of the (invalid string) pointer?
>>>>>>>>>
>>>>>>>>
>>>>>>>> In the bug scenario, the invalid string pointer was an out-of-bound
>>>>>>>> string pointer. While the OOB referencing is fixed,
>>>>>>>
>>>>>>> Could you please provide more details about the fixed OOB?
>>>>>>> What exact vsprintf()/printk() call was broken and eventually
>>>>>>> how it was fixed, please?
>>>>>>
>>>>>> For sensitive reason, I'd like to avoid mentioning the specific name of
>>>>>> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
>>>>>> and describe the precise nature of the issue.
>>>>>>
>>>>>> devX_attrY[] is a string array, declared and filled at compile time,
>>>>>> like
>>>>>>       const char const devX_attrY[] = {
>>>>>> 	[ATTRY_A] = "Dev X AttributeY A",
>>>>>> 	[ATTRY_B] = "Dev X AttributeY B",
>>>>>> 	...
>>>>>> 	[ATTRY_G] = "Dev X AttributeY G",
>>>>>>       }
>>>>>> such that, when user "cat /sys/devices/systems/.../attry_1",
>>>>>> "Dev X AttributeY B" will show up in the terminal.
>>>>>> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
>>>>>>
>>>>>> The bug was that the index to the array was wrongfully produced,
>>>>>> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
>>>>>> calculation and that is not an upstream fix.
>>>>>>
>>>>>>>
>>>>>>>> the lingering issue
>>>>>>>> is that the kernel ought to be able to protect itself, as the pointer
>>>>>>>> contains a non-canonical address.
>>>>>>>
>>>>>>> Was the pointer used only by the vsprintf()?
>>>>>>> Or was it accessed also by another code, please?
>>>>>>
>>>>>> The OOB pointer was used only by vsprintf() for the "cat" sysfs case.
>>>>>> No other code uses the OOB pointer, verified both by code examination
>>>>>> and test.
>>>>>
>>>>> So, then the vsprintf() is _the_ point to crash and why should we hide that?
>>>>> Because of the crash you found the culprit, right? The efault will hide very
>>>>> important details.
>>>>>
>>>>> So to me it sounds like I like this change less and less...
>>>>
>>>> What about the existing check
>>>>     	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>                        return "(efault)";
>>>> ?
>>>
>>> Because it's _special_. We know that First page is equivalent to a NULL pointer
>>> and the last one is dedicated for so called error pointers. There are no more
>>> special exceptions to the addresses in the Linux kernel (I don't talk about
>>> alignment requirements by the certain architectures).
>>>
>>>> In an experiment just to print the raw OOB pointer values, I saw below
>>>> (the devX attrY stuff are substitutes of the real attributes, other
>>>> values and strings are verbatim copy from "dmesg"):
>>>>
>>>> [ 3002.772329] devX_attrY[26]: (ffffffff84d60ad3) Dev X AttributeY E
>>>> [ 3002.772346] devX_attrY[27]: (ffffffff84d60ae4) Dev X AttributeY F
>>>> [ 3002.772347] devX_attrY[28]: (ffffffff84d60aee) Dev X AttributeY G
>>>> [ 3002.772349] devX_attrY[29]: (0) (null)
>>>> [ 3002.772350] devX_attrY[30]: (0) (null)
>>>> [ 3002.772351] devX_attrY[31]: (0) (null)
>>>> [ 3002.772352] devX_attrY[32]: (7665645f63616465) (einval)
>>>> [ 3002.772354] devX_attrY[33]: (646e61685f656369) (einval)
>>>> [ 3002.772355] devX_attrY[34]: (6f635f65755f656c) (einval)
>>>> [ 3002.772355] devX_attrY[35]: (746e75) (einval)
>>>>
>>>> where starting from index 29 are all OOB pointers.
>>>>
>>>> As you can see, if the OOBs are NULL, "(null)" was printed due to the
>>>> existing checking, but when the OOBs are turned to non-canonical which
>>>> is detectable, the fact the pointer value deviates from
>>>>      (ffffffff84d60aee + 4 * sizeof(void *))
>>>> evidently shown that the OOBs are detectable.
>>>>
>>>> The question then is why should the non-canonical OOBs be treated
>>>> differently from NULL and ERR_VALUE?
>>>
>>> Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
>>> to see a bug as early as possible?
>>>
>>
>> If the purpose is to see the bug as early as possible, then getting
>> "(efault)" from reading sysfs attribute would serve the purpose, right?
>>
>> The fact an OOB pointer has already being turned into either NULL or
>> non-canonical value implies that *if* kernel code other than
>> vsprintf() references the pointer, it'll crash else where;
> 
> No, not the case for error pointers and NULL.

Sorry, I don't understand, what about Oops from NUll pointer dereference?

> 
>> but *if* no
>> other code referencing the pointer, why crash?
> 
> Because how else you can see the bug?! The trace will give you essential
> information about registers, etc that gives you a hint what the _cause_ of the
> crash. And we need that cause. The "(efault)" has not even a bit close to what
> crash gives us.
> 
> So, this is my last message in the discussion.
> 
> Here is a formal NAK. Up to maintainers to decide what to do with this.
> 

Sigh, but thanks for taking the time articulating your point of view.

-jane
  
Petr Mladek Oct. 20, 2022, 7:44 a.m. UTC | #15
On Tue 2022-10-18 23:49:27, Andy Shevchenko wrote:
> On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
> > On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
> > > On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
> > >> On 10/18/2022 5:45 AM, Petr Mladek wrote:
> > >>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> > >>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> > >>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> > >>>>>> While debugging a separate issue, it was found that an invalid string
> > >>>>>> pointer could very well contain a non-canical address, such as
> > >>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
> > >>>>>> to protect the kernel from crashing due to general protection fault
> > >>>>>>
> > >>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> > >>>>>>                    return "(efault)";
> > >>>>>>
> > >>>>>> So instead, use kern_addr_valid() to validate the string pointer.
> > >>>>>
> > >>>>> How did you check that value of the (invalid string) pointer?
> > >>>>>
> > >>>>
> > >>>> In the bug scenario, the invalid string pointer was an out-of-bound
> > >>>> string pointer. While the OOB referencing is fixed,
> > >>>
> > >>> Could you please provide more details about the fixed OOB?
> > >>> What exact vsprintf()/printk() call was broken and eventually
> > >>> how it was fixed, please?
> > >>
> > >> For sensitive reason, I'd like to avoid mentioning the specific name of
> > >> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> > >> and describe the precise nature of the issue.
> > >>
> > >> devX_attrY[] is a string array, declared and filled at compile time,
> > >> like
> > >>     const char const devX_attrY[] = {
> > >> 	[ATTRY_A] = "Dev X AttributeY A",
> > >> 	[ATTRY_B] = "Dev X AttributeY B",
> > >> 	...
> > >> 	[ATTRY_G] = "Dev X AttributeY G",
> > >>     }
> > >> such that, when user "cat /sys/devices/systems/.../attry_1",
> > >> "Dev X AttributeY B" will show up in the terminal.
> > >> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> > >>
> > >> The bug was that the index to the array was wrongfully produced,
> > >> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
> > >> calculation and that is not an upstream fix.

I see. printk()/vsprintf() is the only code that accesses this pointer.
If vsprintf() survives than the system survives.

> > As you can see, if the OOBs are NULL, "(null)" was printed due to the 
> > existing checking, but when the OOBs are turned to non-canonical which
> > is detectable, the fact the pointer value deviates from
> >    (ffffffff84d60aee + 4 * sizeof(void *))
> > evidently shown that the OOBs are detectable.
> > 
> > The question then is why should the non-canonical OOBs be treated 
> > differently from NULL and ERR_VALUE?
> 
> Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> to see a bug as early as possible?

I do not agree here. Kernel tries to survive many situations when
thighs does not work as expected. It prints a warning so that
users/developers are aware of the problem and could fix it.

In our case, the crash happened when reading a sysfs file.
IMHO, it is much better to show (-EINVAL) than crash. The bug
when accessing devX_attrY[] does not affect the stability of
the system at all.

And the broken string might be passed in a very rare case,
e.g. in an error path. So that it might be hard to catch
when testing.

Best Regards,
Petr
  
Petr Mladek Oct. 20, 2022, 9:18 a.m. UTC | #16
On Thu 2022-10-20 09:44:06, Petr Mladek wrote:
> On Tue 2022-10-18 23:49:27, Andy Shevchenko wrote:
> > On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:
> > > On 10/18/2022 1:07 PM, Andy Shevchenko wrote:
> > > > On Tue, Oct 18, 2022 at 06:56:31PM +0000, Jane Chu wrote:
> > > >> On 10/18/2022 5:45 AM, Petr Mladek wrote:
> > > >>> On Mon 2022-10-17 19:31:53, Jane Chu wrote:
> > > >>>> On 10/17/2022 12:25 PM, Andy Shevchenko wrote:
> > > >>>>> On Mon, Oct 17, 2022 at 01:16:11PM -0600, Jane Chu wrote:
> > > >>>>>> While debugging a separate issue, it was found that an invalid string
> > > >>>>>> pointer could very well contain a non-canical address, such as
> > > >>>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
> > > >>>>>> to protect the kernel from crashing due to general protection fault
> > > >>>>>>
> > > >>>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
> > > >>>>>>                    return "(efault)";
> > > >>>>>>
> > > >>>>>> So instead, use kern_addr_valid() to validate the string pointer.
> > > >>>>>
> > > >>>>> How did you check that value of the (invalid string) pointer?
> > > >>>>>
> > > >>>>
> > > >>>> In the bug scenario, the invalid string pointer was an out-of-bound
> > > >>>> string pointer. While the OOB referencing is fixed,
> > > >>>
> > > >>> Could you please provide more details about the fixed OOB?
> > > >>> What exact vsprintf()/printk() call was broken and eventually
> > > >>> how it was fixed, please?
> > > >>
> > > >> For sensitive reason, I'd like to avoid mentioning the specific name of
> > > >> the sysfs attribute in the bug, instead, just call it "devX_attrY[]",
> > > >> and describe the precise nature of the issue.
> > > >>
> > > >> devX_attrY[] is a string array, declared and filled at compile time,
> > > >> like
> > > >>     const char const devX_attrY[] = {
> > > >> 	[ATTRY_A] = "Dev X AttributeY A",
> > > >> 	[ATTRY_B] = "Dev X AttributeY B",
> > > >> 	...
> > > >> 	[ATTRY_G] = "Dev X AttributeY G",
> > > >>     }
> > > >> such that, when user "cat /sys/devices/systems/.../attry_1",
> > > >> "Dev X AttributeY B" will show up in the terminal.
> > > >> That's it, no more reference to the pointer devX_attrY[ATTRY_B] after that.
> > > >>
> > > >> The bug was that the index to the array was wrongfully produced,
> > > >> leading up to OOB, e.g. devX_attrY[11].  The fix was to fix the
> > > >> calculation and that is not an upstream fix.
> 
> I see. printk()/vsprintf() is the only code that accesses this pointer.
> If vsprintf() survives than the system survives.
> 
> > > As you can see, if the OOBs are NULL, "(null)" was printed due to the 
> > > existing checking, but when the OOBs are turned to non-canonical which
> > > is detectable, the fact the pointer value deviates from
> > >    (ffffffff84d60aee + 4 * sizeof(void *))
> > > evidently shown that the OOBs are detectable.
> > > 
> > > The question then is why should the non-canonical OOBs be treated 
> > > differently from NULL and ERR_VALUE?
> > 
> > Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> > to see a bug as early as possible?
> 
> I do not agree here. Kernel tries to survive many situations when
> thighs does not work as expected. It prints a warning so that
> users/developers are aware of the problem and could fix it.
> 
> In our case, the crash happened when reading a sysfs file.
> IMHO, it is much better to show (-EINVAL) than crash. The bug
> when accessing devX_attrY[] does not affect the stability of
> the system at all.
> 
> And the broken string might be passed in a very rare case,
> e.g. in an error path. So that it might be hard to catch
> when testing.

That said, there is definitely a difference between NULL or error code
code and a random pointer address.

The pointers in ERR_RANGE are likely to stay in this range.
It means that this pointer is hardly usable as a security
attack.

On the other hand, "random" pointer has a bigger chance to be
used for a security attack. From this POV, it is more important
to catch and fix random pointer issues. And shoving just -EINVAL
might not be enough to catch attention.

I guess that this was what Andy wanted to explain. And kernel
crash would definitely catch attention. Showing some warning
with KERN_WARNING or even WARN() might be an alternative.


Anyway, I think that this patch is not worth it:

   + kern_addr_valid() always succeeds on all architectures
     except on x86_64. It means that the check would help
     only on x86_64.

   + kern_addr_valid() always fails on x86 when build with SPARSEMEM.
     This is not acceptable for vsprintf().

   + the situation when only vsprintf() would access the wrong pointer
     are rare. In most cases, the pointer is later used and the kernel
     crashes anyway.

Best Regards,
Petr
  
Andy Shevchenko Oct. 20, 2022, 1:57 p.m. UTC | #17
On Thu, Oct 20, 2022 at 09:44:05AM +0200, Petr Mladek wrote:
> On Tue 2022-10-18 23:49:27, Andy Shevchenko wrote:
> > On Tue, Oct 18, 2022 at 08:30:01PM +0000, Jane Chu wrote:

...

> > Obviously, to see the crash. And let kernel _to crash_. Isn't it what we need
> > to see a bug as early as possible?
> 
> I do not agree here. Kernel tries to survive many situations when
> thighs does not work as expected. It prints a warning so that
> users/developers are aware of the problem and could fix it.

How the user will know what the root cause and how to fix it? The crash
report will give all needed information, the "(eXXXXXX)" will hide it all,
which I consider inappropriate approach.

I.o.w. consider "(eXXXXXX)" vs. something like "your stuff crashed kernel
because of misaligned / etc pointer which has value of 0xXXXXXXXX and other
registers have these values" and so on, so on...

> In our case, the crash happened when reading a sysfs file.
> IMHO, it is much better to show (-EINVAL) than crash. The bug
> when accessing devX_attrY[] does not affect the stability of
> the system at all.

When I got "eXXXXX" from cat /sys/... I think "OK, something went wrong,
I shouldn't really take it seriously". And completely different feelings
when you got a crash, right?

> And the broken string might be passed in a very rare case,
> e.g. in an error path. So that it might be hard to catch
> when testing.
  

Patch

diff --git a/lib/vsprintf.c b/lib/vsprintf.c
index c414a8d9f1ea..2e8a9efc7c12 100644
--- a/lib/vsprintf.c
+++ b/lib/vsprintf.c
@@ -695,7 +695,7 @@  static const char *check_pointer_msg(const void *ptr)
 	if (!ptr)
 		return "(null)";
 
-	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
+	if (!kern_addr_valid((unsigned long)ptr))
 		return "(efault)";
 
 	return NULL;