2014-01-17 09:40:28

by Jianguo Wu

[permalink] [raw]
Subject: [PATCH] mm/kmemleak: add support for re-enable kmemleak at runtime

Now disabling kmemleak is an irreversible operation, but sometimes
we may need to re-enable kmemleak at runtime. So add a knob to enable
kmemleak at runtime:
echo on > /sys/kernel/debug/kmemleak

Signed-off-by: Jianguo Wu <[email protected]>
---
Documentation/kmemleak.txt | 3 ++-
mm/kmemleak.c | 37 +++++++++++++++++++++++++++++++++----
2 files changed, 35 insertions(+), 5 deletions(-)

diff --git a/Documentation/kmemleak.txt b/Documentation/kmemleak.txt
index b6e3973..8ec56ad 100644
--- a/Documentation/kmemleak.txt
+++ b/Documentation/kmemleak.txt
@@ -44,7 +44,8 @@ objects to be reported as orphan.
Memory scanning parameters can be modified at run-time by writing to the
/sys/kernel/debug/kmemleak file. The following parameters are supported:

- off - disable kmemleak (irreversible)
+ off - disable kmemleak
+ on - enable kmemleak
stack=on - enable the task stacks scanning (default)
stack=off - disable the tasks stacks scanning
scan=on - start the automatic memory scanning thread (default)
diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index 31f01c5..02f292c 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -260,6 +260,7 @@ static struct early_log
static int crt_early_log __initdata;

static void kmemleak_disable(void);
+static void kmemleak_enable(void);

/*
* Print a warning and dump the stack trace.
@@ -1616,9 +1617,6 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
int buf_size;
int ret;

- if (!atomic_read(&kmemleak_enabled))
- return -EBUSY;
-
buf_size = min(size, (sizeof(buf) - 1));
if (strncpy_from_user(buf, user_buf, buf_size) < 0)
return -EFAULT;
@@ -1628,6 +1626,19 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
if (ret < 0)
return ret;

+ if (strncmp(buf, "on", 2) == 0) {
+ if (atomic_read(&kmemleak_enabled))
+ ret = -EBUSY;
+ else
+ kmemleak_enable();
+ goto out;
+ }
+
+ if (!atomic_read(&kmemleak_enabled)) {
+ ret = -EBUSY;
+ goto out;
+ }
+
if (strncmp(buf, "off", 3) == 0)
kmemleak_disable();
else if (strncmp(buf, "stack=on", 8) == 0)
@@ -1703,7 +1714,7 @@ static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);

/*
* Disable kmemleak. No memory allocation/freeing will be traced once this
- * function is called. Disabling kmemleak is an irreversible operation.
+ * function is called.
*/
static void kmemleak_disable(void)
{
@@ -1721,6 +1732,24 @@ static void kmemleak_disable(void)
pr_info("Kernel memory leak detector disabled\n");
}

+static void kmemleak_enable(void)
+{
+ struct kmemleak_object *object;
+
+ /* free the kmemleak internal objects the previous thread scanned */
+ rcu_read_lock();
+ list_for_each_entry_rcu(object, &object_list, object_list)
+ delete_object_full(object->pointer);
+ rcu_read_unlock();
+
+ atomic_set(&kmemleak_enabled, 1);
+ atomic_set(&kmemleak_error, 0);
+
+ start_scan_thread();
+
+ pr_info("Kernel memory leak detector enabled\n");
+}
+
/*
* Allow boot-time kmemleak disabling (enabled by default).
*/
--
1.7.7


2014-01-17 12:05:12

by Catalin Marinas

[permalink] [raw]
Subject: Re: [PATCH] mm/kmemleak: add support for re-enable kmemleak at runtime

On Fri, Jan 17, 2014 at 09:40:02AM +0000, Jianguo Wu wrote:
> Now disabling kmemleak is an irreversible operation, but sometimes
> we may need to re-enable kmemleak at runtime. So add a knob to enable
> kmemleak at runtime:
> echo on > /sys/kernel/debug/kmemleak

It is irreversible for very good reason: once it missed the initial
memory allocations, there is no way for kmemleak to build the object
reference graph and you'll get lots of false positives, pretty much
making it unusable.

--
Catalin

2014-01-18 08:42:05

by Jianguo Wu

[permalink] [raw]
Subject: Re: [PATCH] mm/kmemleak: add support for re-enable kmemleak at runtime

On 2014/1/17 20:04, Catalin Marinas wrote:

> On Fri, Jan 17, 2014 at 09:40:02AM +0000, Jianguo Wu wrote:
>> Now disabling kmemleak is an irreversible operation, but sometimes
>> we may need to re-enable kmemleak at runtime. So add a knob to enable
>> kmemleak at runtime:
>> echo on > /sys/kernel/debug/kmemleak
>
> It is irreversible for very good reason: once it missed the initial
> memory allocations, there is no way for kmemleak to build the object
> reference graph and you'll get lots of false positives, pretty much
> making it unusable.
>

Do you mean we didn't trace memory allocations during kmemleak disable period,
and these memory may reference to new allocated objects after re-enable?

2014-01-18 10:45:43

by Catalin Marinas

[permalink] [raw]
Subject: Re: [PATCH] mm/kmemleak: add support for re-enable kmemleak at runtime

On 18 Jan 2014, at 08:41, Jianguo Wu <[email protected]> wrote:
> On 2014/1/17 20:04, Catalin Marinas wrote:
>> On Fri, Jan 17, 2014 at 09:40:02AM +0000, Jianguo Wu wrote:
>>> Now disabling kmemleak is an irreversible operation, but sometimes
>>> we may need to re-enable kmemleak at runtime. So add a knob to enable
>>> kmemleak at runtime:
>>> echo on > /sys/kernel/debug/kmemleak
>>
>> It is irreversible for very good reason: once it missed the initial
>> memory allocations, there is no way for kmemleak to build the object
>> reference graph and you'll get lots of false positives, pretty much
>> making it unusable.
>
> Do you mean we didn't trace memory allocations during kmemleak disable period,
> and these memory may reference to new allocated objects after re-enable?

Yes. Those newly allocated objects would be reported as leaks.

Catalin-