There was a recent change in blktrace.c that added a RCU protection to
`q->blk_trace` in order to fix a use-after-free issue during access.
However the change missed an edge case that can lead to dereferencing of
`bt` pointer even when it's NULL:
```
bt->act_mask = value; // bt can still be NULL here
```
Added a reassignment into the NULL check block to fix the issue.
Fixes: c780e86dd48 ("blktrace: Protect q->blk_trace with RCU")
Signed-off-by: Cengiz Can <[email protected]>
---
Huge thanks goes to Steven Rostedt for his assistance.
kernel/trace/blktrace.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 4560878f0bac..29ea88f10b87 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -1896,8 +1896,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
}
ret = 0;
- if (bt == NULL)
+ if (bt == NULL) {
ret = blk_trace_setup_queue(q, bdev);
+ bt = q->blk_trace;
+ }
if (ret == 0) {
if (attr == &dev_attr_act_mask)
--
2.25.1
On 3/3/20 3:33 PM, Cengiz Can wrote:
> There was a recent change in blktrace.c that added a RCU protection to
> `q->blk_trace` in order to fix a use-after-free issue during access.
>
> However the change missed an edge case that can lead to dereferencing of
> `bt` pointer even when it's NULL:
>
> ```
> bt->act_mask = value; // bt can still be NULL here
> ```
>
> Added a reassignment into the NULL check block to fix the issue.
>
> Fixes: c780e86dd48 ("blktrace: Protect q->blk_trace with RCU")
>
> Signed-off-by: Cengiz Can <[email protected]>
> ---
> Huge thanks goes to Steven Rostedt for his assistance.
>
> kernel/trace/blktrace.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> index 4560878f0bac..29ea88f10b87 100644
> --- a/kernel/trace/blktrace.c
> +++ b/kernel/trace/blktrace.c
> @@ -1896,8 +1896,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
> }
>
> ret = 0;
> - if (bt == NULL)
> + if (bt == NULL) {
> ret = blk_trace_setup_queue(q, bdev);
> + bt = q->blk_trace;
The return value 'ret' should be judged, it's wrong to set 'bt' if blk_trace_setup_queue()
return failure.
> + }
>
> if (ret == 0) {
> if (attr == &dev_attr_act_mask)
> --
> 2.25.1
>
On Tue, 3 Mar 2020 21:29:08 +0800
Bob Liu <[email protected]> wrote:
> On 3/3/20 3:33 PM, Cengiz Can wrote:
> > There was a recent change in blktrace.c that added a RCU protection to
> > `q->blk_trace` in order to fix a use-after-free issue during access.
> >
> > However the change missed an edge case that can lead to dereferencing of
> > `bt` pointer even when it's NULL:
> >
> > ```
> > bt->act_mask = value; // bt can still be NULL here
> > ```
> >
> > Added a reassignment into the NULL check block to fix the issue.
> >
> > Fixes: c780e86dd48 ("blktrace: Protect q->blk_trace with RCU")
> >
> > Signed-off-by: Cengiz Can <[email protected]>
> > ---
> > Huge thanks goes to Steven Rostedt for his assistance.
> >
> > kernel/trace/blktrace.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> > index 4560878f0bac..29ea88f10b87 100644
> > --- a/kernel/trace/blktrace.c
> > +++ b/kernel/trace/blktrace.c
> > @@ -1896,8 +1896,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
> > }
> >
> > ret = 0;
> > - if (bt == NULL)
> > + if (bt == NULL) {
> > ret = blk_trace_setup_queue(q, bdev);
> > + bt = q->blk_trace;
>
> The return value 'ret' should be judged, it's wrong to set 'bt' if blk_trace_setup_queue()
> return failure.
Why? If ret is an error, q is still valid, and bt would just be garbage. bt
is ignored below if ret is anything but zero. Why add an unnecessary if
condition here?
That said, the bt assignment still needs rcu annotation:
bt = rcu_dereference_protected(q->blk_trace,
lockdep_is_held(&q->blk_trace_mutex));
-- Steve
>
> > + }
> >
> > if (ret == 0) {
> > if (attr == &dev_attr_act_mask)
> > --
> > 2.25.1
> >
On 3/3/20 9:35 PM, Steven Rostedt wrote:
> On Tue, 3 Mar 2020 21:29:08 +0800
> Bob Liu <[email protected]> wrote:
>
>> On 3/3/20 3:33 PM, Cengiz Can wrote:
>>> There was a recent change in blktrace.c that added a RCU protection to
>>> `q->blk_trace` in order to fix a use-after-free issue during access.
>>>
>>> However the change missed an edge case that can lead to dereferencing of
>>> `bt` pointer even when it's NULL:
>>>
>>> ```
>>> bt->act_mask = value; // bt can still be NULL here
>>> ```
>>>
>>> Added a reassignment into the NULL check block to fix the issue.
>>>
>>> Fixes: c780e86dd48 ("blktrace: Protect q->blk_trace with RCU")
>>>
>>> Signed-off-by: Cengiz Can <[email protected]>
>>> ---
>>> Huge thanks goes to Steven Rostedt for his assistance.
>>>
>>> kernel/trace/blktrace.c | 4 +++-
>>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
>>> index 4560878f0bac..29ea88f10b87 100644
>>> --- a/kernel/trace/blktrace.c
>>> +++ b/kernel/trace/blktrace.c
>>> @@ -1896,8 +1896,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
>>> }
>>>
>>> ret = 0;
>>> - if (bt == NULL)
>>> + if (bt == NULL) {
>>> ret = blk_trace_setup_queue(q, bdev);
>>> + bt = q->blk_trace;
>>
>> The return value 'ret' should be judged, it's wrong to set 'bt' if blk_trace_setup_queue()
>> return failure.
>
> Why? If ret is an error, q is still valid, and bt would just be garbage. bt
> is ignored below if ret is anything but zero. Why add an unnecessary if
> condition here?
>
Oh..yes. You are right, sorry for missed the if (ret == 0) below.
Reviewed-by: Bob Liu <[email protected]>
> That said, the bt assignment still needs rcu annotation:
> > bt = rcu_dereference_protected(q->blk_trace,
> lockdep_is_held(&q->blk_trace_mutex));
>
> -- Steve
>
>
>>
>>> + }
>>>
>>> if (ret == 0) {
>>> if (attr == &dev_attr_act_mask)
>>> --
>>> 2.25.1
>>>
>
On Tue, 3 Mar 2020 10:33:59 +0300
Cengiz Can <[email protected]> wrote:
> There was a recent change in blktrace.c that added a RCU protection to
> `q->blk_trace` in order to fix a use-after-free issue during access.
>
> However the change missed an edge case that can lead to dereferencing of
> `bt` pointer even when it's NULL:
>
> ```
> bt->act_mask = value; // bt can still be NULL here
> ```
>
> Added a reassignment into the NULL check block to fix the issue.
Note, you should probably add a note in your change log that this issue was
found by Coverity. Sometimes static analyzers have a tag of some kind that
they would like patches that fix the issues they discover to be in the
change log. This way they can track the fixes that are found by the tool.
>
> Fixes: c780e86dd48 ("blktrace: Protect q->blk_trace with RCU")
>
> Signed-off-by: Cengiz Can <[email protected]>
> ---
> Huge thanks goes to Steven Rostedt for his assistance.
>
> kernel/trace/blktrace.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> index 4560878f0bac..29ea88f10b87 100644
> --- a/kernel/trace/blktrace.c
> +++ b/kernel/trace/blktrace.c
> @@ -1896,8 +1896,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
> }
>
> ret = 0;
> - if (bt == NULL)
> + if (bt == NULL) {
> ret = blk_trace_setup_queue(q, bdev);
> + bt = q->blk_trace;
As I said in the other email, the above assignment still needs RCU
annotation:
bt = rcu_dereference_protected(q->blk_trace,
lockdep_is_held(&q->blk_trace_mutex));
Otherwise, other static analyzers will flag this as a problem.
-- Steve
> + }
>
> if (ret == 0) {
> if (attr == &dev_attr_act_mask)
> --
> 2.25.1
On 2020-03-03 17:14, Steven Rostedt wrote:
>
> Note, you should probably add a note in your change log that this issue
> was
> found by Coverity. Sometimes static analyzers have a tag of some kind
> that
> they would like patches that fix the issues they discover to be in the
> change log. This way they can track the fixes that are found by the
> tool.
>
I have added the Coverity issue ID and relevant notes to mark it.
Thanks.
>
> As I said in the other email, the above assignment still needs RCU
> annotation:
>
> bt = rcu_dereference_protected(q->blk_trace,
> lockdep_is_held(&q->blk_trace_mutex));
I missed that. Thanks for reminding. TIL.
Please check PATCH-v2.
--
Cengiz Can
@cengiz_io