11:36 p.m.
We'll take a look.
Thanks,
Bob
-----Original Message-----
From: Dan Carpenter [mailto:dan.carpenter@oracle.com]
Sent: Thursday, October 17, 2013 5:29 AM
To: Moore, Robert
Cc: linux-acpi(a)vger.kernel.org; devel(a)acpica.org
Subject: re: ACPICA: Resource Mgr: Prevent infinite loops in resource
walks
Hello Bob Moore,
The patch c13085e519e8: "ACPICA: Resource Mgr: Prevent infinite loops in
resource walks" from Mar 8, 2013 is not beautiful. My static checker
complains about the loop because:
"drivers/acpi/acpica/rscalc.c:197 acpi_rs_get_aml_length()
warn: 'resource' can't be NULL."
drivers/acpi/acpica/rscalc.c
195 /* Traverse entire list of internal resource descriptors
*/
196
197 while (resource) {
^^^^^^^^
My static checker is wrong because we use the -fno-strict-overflow to
prevent GCC from optimizing this check away. But we are looping over a
list of pointers until our pointer wraps to NULL. In other words we loop
over all the 2**64 - 1 addresses until we wrap to NULL or we find
something with an invalid type or something with ->length zero.
I assume the last element in the list always has length zero? If so then
we could replace "while (resource)" with "while
(resource->length)"
198
199 /* Validate the descriptor type */
200
201 if (resource->type > ACPI_RESOURCE_TYPE_MAX) {
202
return_ACPI_STATUS(AE_AML_INVALID_RESOURCE_TYPE);
203 }
204
205 /* Sanity check the length. It must not be zero,
or we loop forever */
206
207 if (!resource->length) {
208
return_ACPI_STATUS(AE_AML_BAD_RESOURCE_LENGTH);
209 }
210
211 /* Get the base size of the (external stream)
resource descriptor */
212
regards,
dan carpenter
8:06 a.m.
New subject: ACPICA: Resource Mgr: Prevent infinite loops in resource walks
The last resource should always be ACPI_RESOURCE_TYPE_END_TAG and we have nsrepair
mechanism to ensure that this entry is always there for resource based properties (there
are still 2 kinds of properties haven't been fixed yet).
If we do not want to rely on this logic, then we may need to use buffer->length and
pass this one to the acpi_rs_get_aml_length to avoid potential infinite loop.
I'll draft an ACPICA patch to achieve the latter.
Thanks and best regards
-Lv
-----Original Message-----
From: devel-bounces(a)acpica.org [mailto:devel-bounces@acpica.org] On Behalf Of Moore,
Robert
Sent: Wednesday, October 23, 2013 5:36 AM
To: Dan Carpenter
Cc: linux-acpi(a)vger.kernel.org; devel(a)acpica.org
Subject: Re: [Devel] ACPICA: Resource Mgr: Prevent infinite loops in resource walks
We'll take a look.
Thanks,
Bob
> -----Original Message-----
> From: Dan Carpenter [mailto:dan.carpenter@oracle.com]
> Sent: Thursday, October 17, 2013 5:29 AM
> To: Moore, Robert
> Cc: linux-acpi(a)vger.kernel.org; devel(a)acpica.org
> Subject: re: ACPICA: Resource Mgr: Prevent infinite loops in resource
> walks
>
> Hello Bob Moore,
>
> The patch c13085e519e8: "ACPICA: Resource Mgr: Prevent infinite loops in
> resource walks" from Mar 8, 2013 is not beautiful. My static checker
> complains about the loop because:
> "drivers/acpi/acpica/rscalc.c:197 acpi_rs_get_aml_length()
> warn: 'resource' can't be NULL."
>
> drivers/acpi/acpica/rscalc.c
> 195 /* Traverse entire list of internal resource descriptors
> */
> 196
> 197 while (resource) {
> ^^^^^^^^
>
> My static checker is wrong because we use the -fno-strict-overflow to
> prevent GCC from optimizing this check away. But we are looping over a
> list of pointers until our pointer wraps to NULL. In other words we loop
> over all the 2**64 - 1 addresses until we wrap to NULL or we find
> something with an invalid type or something with ->length zero.
>
> I assume the last element in the list always has length zero? If so then
> we could replace "while (resource)" with "while
(resource->length)"
>
> 198
> 199 /* Validate the descriptor type */
> 200
> 201 if (resource->type > ACPI_RESOURCE_TYPE_MAX) {
> 202
> return_ACPI_STATUS(AE_AML_INVALID_RESOURCE_TYPE);
> 203 }
> 204
> 205 /* Sanity check the length. It must not be zero,
> or we loop forever */
> 206
> 207 if (!resource->length) {
> 208
> return_ACPI_STATUS(AE_AML_BAD_RESOURCE_LENGTH);
> 209 }
> 210
> 211 /* Get the base size of the (external stream)
> resource descriptor */
> 212
>
> regards,
> dan carpenter
_______________________________________________
Devel mailing list
Devel(a)acpica.org
https://lists.acpica.org/mailman/listinfo/devel
3133
days inactive
3134
days old
1 comments
2 participants
participants (2)
-
Moore, Robert -
Zheng, Lv