As a follow-up to commit 6cab7a37f5c04 ("staging: comedi: (regression)
channel list must be set for COMEDI_CMD ioctl"), Hartley Sweeten pointed
out another couple of bugs stemming from commit 6cab7a37f5c04 ("staging:
comedi: comedi_fops: introduce __comedi_get_user_chanlist()").
Firstly, `do_cmdtest_ioctl()` never frees the kernel copy of the user
chanlist allocated by `__comedi_get_user_chanlist()`, so that memory is
leaked. Fix it by freeing the allocated kernel memory pointed to by
`cmd.chanlist` before that pointer is overwritten with its original
pointer to user memory before `cmd` is copied back to user-space.
Secondly, if `__comedi_get_user_chanlist()` returns an error,
`cmd->chanlist` is left unchanged and in fact will be a pointer to user
memory. This causes `do_cmd_ioctl()` to `goto cleanup` and call
`do_become_nonbusy()` which would attempt to free the memory pointed to
by the user-space pointer. Fix it by setting `cmd->chanlist` to NULL at
the start of `__comedi_get_user_chanlist()`.
Fixes: c6cd0eefb27b ("staging: comedi: comedi_fops: introduce __comedi_get_user_chanlist()")
Reported-by: H Hartley Sweeten <[email protected]>
Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y: 6cab7a37f5c04
Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y
---
Greg, this patch applies to your "staging-linus" branch.
---
drivers/staging/comedi/comedi_fops.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/staging/comedi/comedi_fops.c b/drivers/staging/comedi/comedi_fops.c
index a9b7fe5..a1788e8 100644
--- a/drivers/staging/comedi/comedi_fops.c
+++ b/drivers/staging/comedi/comedi_fops.c
@@ -1462,6 +1462,7 @@ static int __comedi_get_user_chanlist(struct comedi_device *dev,
unsigned int *chanlist;
int ret;
+ cmd->chanlist = NULL;
chanlist = memdup_user(user_chanlist,
cmd->chanlist_len * sizeof(unsigned int));
if (IS_ERR(chanlist))
@@ -1615,6 +1616,8 @@ static int do_cmdtest_ioctl(struct comedi_device *dev,
ret = s->do_cmdtest(dev, s, &cmd);
+ kfree(cmd.chanlist); /* free kernel copy of user chanlist */
+
/* restore chanlist pointer before copying back */
cmd.chanlist = (unsigned int __force *)user_chanlist;
--
2.1.1
On Monday, October 20, 2014 7:11 AM, Ian Abbott wrote:
> As a follow-up to commit 6cab7a37f5c04 ("staging: comedi: (regression)
> channel list must be set for COMEDI_CMD ioctl"), Hartley Sweeten pointed
> out another couple of bugs stemming from commit 6cab7a37f5c04 ("staging:
> comedi: comedi_fops: introduce __comedi_get_user_chanlist()").
>
> Firstly, `do_cmdtest_ioctl()` never frees the kernel copy of the user
> chanlist allocated by `__comedi_get_user_chanlist()`, so that memory is
> leaked. Fix it by freeing the allocated kernel memory pointed to by
> `cmd.chanlist` before that pointer is overwritten with its original
> pointer to user memory before `cmd` is copied back to user-space.
>
> Secondly, if `__comedi_get_user_chanlist()` returns an error,
> `cmd->chanlist` is left unchanged and in fact will be a pointer to user
> memory. This causes `do_cmd_ioctl()` to `goto cleanup` and call
> `do_become_nonbusy()` which would attempt to free the memory pointed to
> by the user-space pointer. Fix it by setting `cmd->chanlist` to NULL at
> the start of `__comedi_get_user_chanlist()`.
>
> Fixes: c6cd0eefb27b ("staging: comedi: comedi_fops: introduce __comedi_get_user_chanlist()")
> Reported-by: H Hartley Sweeten <[email protected]>
> Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y: 6cab7a37f5c04
> Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y
> ---
> Greg, this patch applies to your "staging-linus" branch.
> ---
> drivers/staging/comedi/comedi_fops.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/staging/comedi/comedi_fops.c b/drivers/staging/comedi/comedi_fops.c
> index a9b7fe5..a1788e8 100644
> --- a/drivers/staging/comedi/comedi_fops.c
> +++ b/drivers/staging/comedi/comedi_fops.c
> @@ -1462,6 +1462,7 @@ static int __comedi_get_user_chanlist(struct comedi_device *dev,
> unsigned int *chanlist;
> int ret;
>
> + cmd->chanlist = NULL;
> chanlist = memdup_user(user_chanlist,
> cmd->chanlist_len * sizeof(unsigned int));
> if (IS_ERR(chanlist))
I don't think this covers all the problems in your second issue above.
It does fix the attempt to free the memory pointed to by the user-space pointer
if the memdub_user() fails.
But in do_cmd_ioctl() we still have an issue if the s->do_cmdtest() fails or
the CMDF_BOGUS flag is set. There we restore the users cmd.chanlist
pointer to allow copy_to_user() to return the cmd/ But the kernel allocated
chanlist is not freed. The goto cleanup will again try to free the user-space
pointer.
The remaining goto cleanup jumps will correctly free the kernel allocated
pointer stored in cmd.chanlist.
> @@ -1615,6 +1616,8 @@ static int do_cmdtest_ioctl(struct comedi_device *dev,
>
> ret = s->do_cmdtest(dev, s, &cmd);
>
> + kfree(cmd.chanlist); /* free kernel copy of user chanlist */
> +
> /* restore chanlist pointer before copying back */
> cmd.chanlist = (unsigned int __force *)user_chanlist;
>
This one does fix the issue in do_cmdtest_ioctl().
Regards,
Hartley
On 20/10/14 22:38, Hartley Sweeten wrote:
> On Monday, October 20, 2014 7:11 AM, Ian Abbott wrote:
>> As a follow-up to commit 6cab7a37f5c04 ("staging: comedi: (regression)
>> channel list must be set for COMEDI_CMD ioctl"), Hartley Sweeten pointed
>> out another couple of bugs stemming from commit 6cab7a37f5c04 ("staging:
>> comedi: comedi_fops: introduce __comedi_get_user_chanlist()").
>>
>> Firstly, `do_cmdtest_ioctl()` never frees the kernel copy of the user
>> chanlist allocated by `__comedi_get_user_chanlist()`, so that memory is
>> leaked. Fix it by freeing the allocated kernel memory pointed to by
>> `cmd.chanlist` before that pointer is overwritten with its original
>> pointer to user memory before `cmd` is copied back to user-space.
>>
>> Secondly, if `__comedi_get_user_chanlist()` returns an error,
>> `cmd->chanlist` is left unchanged and in fact will be a pointer to user
>> memory. This causes `do_cmd_ioctl()` to `goto cleanup` and call
>> `do_become_nonbusy()` which would attempt to free the memory pointed to
>> by the user-space pointer. Fix it by setting `cmd->chanlist` to NULL at
>> the start of `__comedi_get_user_chanlist()`.
>>
>> Fixes: c6cd0eefb27b ("staging: comedi: comedi_fops: introduce __comedi_get_user_chanlist()")
>> Reported-by: H Hartley Sweeten <[email protected]>
>> Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y: 6cab7a37f5c04
>> Cc: <[email protected]> # 3.15.y 3.16.y 3.17.y
>> ---
>> Greg, this patch applies to your "staging-linus" branch.
>> ---
>> drivers/staging/comedi/comedi_fops.c | 3 +++
>> 1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/staging/comedi/comedi_fops.c b/drivers/staging/comedi/comedi_fops.c
>> index a9b7fe5..a1788e8 100644
>> --- a/drivers/staging/comedi/comedi_fops.c
>> +++ b/drivers/staging/comedi/comedi_fops.c
>> @@ -1462,6 +1462,7 @@ static int __comedi_get_user_chanlist(struct comedi_device *dev,
>> unsigned int *chanlist;
>> int ret;
>>
>> + cmd->chanlist = NULL;
>> chanlist = memdup_user(user_chanlist,
>> cmd->chanlist_len * sizeof(unsigned int));
>> if (IS_ERR(chanlist))
>
> I don't think this covers all the problems in your second issue above.
>
> It does fix the attempt to free the memory pointed to by the user-space pointer
> if the memdub_user() fails.
>
> But in do_cmd_ioctl() we still have an issue if the s->do_cmdtest() fails or
> the CMDF_BOGUS flag is set. There we restore the users cmd.chanlist
> pointer to allow copy_to_user() to return the cmd/ But the kernel allocated
> chanlist is not freed. The goto cleanup will again try to free the user-space
> pointer.
No, 'async->cmd' is copied (via struct assignment) to local variable
'cmd' and it is the local 'cmd.chanlist' pointer that is set to the
original user-space pointer and the local 'cmd' that is copied back to
user-space. 'async->cmd.chanlist' still points to the kernel copy of
the chanlist that gets cleaned up by 'goto cleanup' / 'do_become_nonbusy().
--
-=( Ian Abbott @ MEV Ltd. E-mail: <[email protected]> )=-
-=( Web: http://www.mev.co.uk/ )=-