lkml.org 
[lkml]   [2018]   [Jan]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: Possible Memory Leak in KCOV Linux 4.15-rc1
Below is a reproducer.

#define _GNU_SOURCE
#include <fcntl.h>
#include <sys/ioctl.h>
#include <sys/mman.h>
#include <stdlib.h>
#include <unistd.h>

#define KCOV_INIT_TRACE _IOR('c', 1, unsigned long)
#define KCOV_ENABLE _IO('c', 100)
#define KCOV_DISABLE _IO('c', 101)
#define COVER_SIZE (16 << 20)


void kcov_setup() {
unsigned long *cover;
int fd;

fd = open("/sys/kernel/debug/kcov", O_RDWR);
if (fd == -1)
perror("open"), exit(1);
if (ioctl(fd, KCOV_INIT_TRACE, COVER_SIZE))
perror("ioctl"), exit(1);
cover = (unsigned long*)mmap(NULL,
COVER_SIZE * sizeof(unsigned long),
PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
if ((void*)cover == MAP_FAILED)
perror("mmap"), exit(1);
if (ioctl(fd, KCOV_ENABLE, 0))
perror("ioctl"), exit(1);

}

void main() {
int i;
for (i = 0; i < 4; i++)
kcov_setup();
sleep(10);
}

On Sun, Jan 21, 2018 at 1:11 AM, Shankara Pailoor <sp3485@columbia.edu> wrote:
> Hi Dmitry,
>
> The leaks went away when I disabled and closed the old file
> descriptors before opening new ones.
>
> The patch you sent wouldn't work because t is not initialized at the
> line. This seems to work for me
>
> diff --git a/kernel/kcov.c b/kernel/kcov.c
> index 7594c03..1397006 100644
> --- a/kernel/kcov.c
> +++ b/kernel/kcov.c
> @@ -371,6 +371,8 @@ static int kcov_ioctl_locked(struct kcov *kcov,
> unsigned int cmd,
> else
> return -EINVAL;
> t = current;
> + if (!t->kcov)
> + return -EBUSY;
> /* Cache in task struct for performance. */
> t->kcov_size = kcov->size;
> t->kcov_area = kcov->area;
>
> On Sat, Jan 20, 2018 at 7:06 AM, Dmitry Vyukov <dvyukov@google.com> wrote:
>> On Sat, Jan 20, 2018 at 4:01 PM, Shankara Pailoor <sp3485@columbia.edu> wrote:
>>> Hi Dmitry,
>>>
>>> I will try and get something to you tomorrow. Just wondering, but what
>>> happens to the old struct kcov if a task opens /sys/kernel/debug/kcov
>>> twice? I am looking here
>>> https://elixir.free-electrons.com/linux/v4.15-rc8/source/kernel/kcov.c#L381
>>> and I don't see where the previous struct would get freed.
>>
>> Good question. Perhaps we need something like:
>>
>> diff --git a/kernel/kcov.c b/kernel/kcov.c
>> index 7594c033d98a..c76498018500 100644
>> --- a/kernel/kcov.c
>> +++ b/kernel/kcov.c
>> @@ -358,7 +358,7 @@ static int kcov_ioctl_locked(struct kcov *kcov,
>> unsigned int cmd,
>> */
>> if (kcov->mode != KCOV_MODE_INIT || !kcov->area)
>> return -EINVAL;
>> - if (kcov->t != NULL)
>> + if (kcov->t != NULL || t->kcov != NULL)
>> return -EBUSY;
>> if (arg == KCOV_TRACE_PC)
>> kcov->mode = KCOV_MODE_TRACE_PC;
>>
>>
>>
>>> On Sat, Jan 20, 2018 at 4:38 AM, Dmitry Vyukov <dvyukov@google.com> wrote:
>>>> On Fri, Jan 19, 2018 at 8:29 PM, Shankara Pailoor <sp3485@columbia.edu> wrote:
>>>>> Hi Dmitry,
>>>>>
>>>>> I added support for kcov in strace and I have been tracing a fairly
>>>>> large program but after a little while, I notice that when I mmap a
>>>>> new cover buffer, the call fails with ENOMEM. After killing the
>>>>> program, I try and rerun and I notice that there is nearly no memory
>>>>> on the system. When I do a kmemleak scan I get the following reports:
>>>>>
>>>>> I believe the problem occurs when I try and setup the kcov buffer
>>>>> again after an exec. Instead of reusing the old file descriptor I open
>>>>> kcov again within that process. In that case, I don't know what
>>>>> happens to the old kcov struct.
>>>>>
>>>>> I don't see a maintainers list for kcov so I decided to email you
>>>>> directly. Let me know what more information I can provide.
>>>>
>>>>
>>>> Hi Shankara,
>>>>
>>>> Looks bad. Can you provide a reproducer?
>>>> We extensively use kcov with syzkaller, but have not observed such
>>>> leaks. Also I don't see anything obvious in the code.
>>>>
>>>> Thanks
>>>

\
 
 \ /
  Last update: 2018-01-21 18:55    [W:0.041 / U:2.568 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site