On 2024-11-04 11:19 am, Catalin Marinas wrote:
The introduction of iova_depot_pop() in 911aa1245da8 ("iommu/iova: Make
the rcache depot scale better") confused kmemleak by moving a struct
iova_magazine object from a singly linked list to rcache->depot and
resetting the 'next' pointer referencing it. Unlike doubly linked lists,
the content of the object being referred is never changed on removal
from a singly linked list and the kmemleak checksum heuristics do not
detect such scenario. This leads to false positives like:
unreferenced object 0xffff8881a5301000 (size 1024):
comm "softirq", pid 0, jiffies 4306297099 (age 462.991s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 e7 7d 05 00 00 00 00 00 .........}......
0f b4 05 00 00 00 00 00 b4 96 05 00 00 00 00 00 ................
backtrace:
[<ffffffff819f5f08>] __kmem_cache_alloc_node+0x1e8/0x320
[<ffffffff818a239a>] kmalloc_trace+0x2a/0x60
[<ffffffff8231d31e>] free_iova_fast+0x28e/0x4e0
[<ffffffff82310860>] fq_ring_free_locked+0x1b0/0x310
[<ffffffff8231225d>] fq_flush_timeout+0x19d/0x2e0
[<ffffffff813e95ba>] call_timer_fn+0x19a/0x5c0
[<ffffffff813ea16b>] __run_timers+0x78b/0xb80
[<ffffffff813ea5bd>] run_timer_softirq+0x5d/0xd0
[<ffffffff82f1d915>] __do_softirq+0x205/0x8b5
Introduce kmemleak_transient_leak() which resets the object checksum
requiring another scan pass before it is reported (if still
unreferenced). Call this new API in iova_depot_pop().
Acked-by: Robin Murphy <robin.murphy@xxxxxxx>
Signed-off-by: Catalin Marinas <catalin.marinas@xxxxxxx>
Reported-by: Ido Schimmel <idosch@xxxxxxxxxx>
Tested-by: Ido Schimmel <idosch@xxxxxxxxxx>
Cc: Robin Murphy <robin.murphy@xxxxxxx>
Cc: Joerg Roedel <joro@xxxxxxxxxx>
Cc: Will Deacon <will@xxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Link: https://lore.kernel.org/r/ZY1osaGLyT-sdKE8@shredder/
---
This could be two patches but I thought the rationale for a new kmemleak
API goes better with its use in the iova code. Happy to move the 6 lines
iova change to a separate patch but they should still go in together.
Given that there are more line under mm/, I'd say it better goes in via
the mm tree with the relevant acks from the iommu folk.
Thanks.
Documentation/dev-tools/kmemleak.rst | 1 +
drivers/iommu/iova.c | 6 +++++
include/linux/kmemleak.h | 4 +++
mm/kmemleak.c | 39 ++++++++++++++++++++++++++++
4 files changed, 50 insertions(+)
diff --git a/Documentation/dev-tools/kmemleak.rst b/Documentation/dev-tools/kmemleak.rst
index 2cb00b53339f..7d784e03f3f9 100644
--- a/Documentation/dev-tools/kmemleak.rst
+++ b/Documentation/dev-tools/kmemleak.rst
@@ -161,6 +161,7 @@ See the include/linux/kmemleak.h header for the functions prototype.
- ``kmemleak_free_percpu`` - notify of a percpu memory block freeing
- ``kmemleak_update_trace`` - update object allocation stack trace
- ``kmemleak_not_leak`` - mark an object as not a leak
+- ``kmemleak_transient_leak`` - mark an object as a transient leak
- ``kmemleak_ignore`` - do not scan or report an object as leak
- ``kmemleak_scan_area`` - add scan areas inside a memory block
- ``kmemleak_no_scan`` - do not scan a memory block
diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
index 16c6adff3eb7..5b5400efb657 100644
--- a/drivers/iommu/iova.c
+++ b/drivers/iommu/iova.c
@@ -6,6 +6,7 @@
*/
#include <linux/iova.h>
+#include <linux/kmemleak.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/smp.h>
@@ -673,6 +674,11 @@ static struct iova_magazine *iova_depot_pop(struct iova_rcache *rcache)
{
struct iova_magazine *mag = rcache->depot;
+ /*
+ * As the mag->next pointer is moved to rcache->depot and reset via
+ * the mag->size assignment, mark it as a transient false positive.
+ */
+ kmemleak_transient_leak(mag->next);
rcache->depot = mag->next;
mag->size = IOVA_MAG_SIZE;
rcache->depot_size--;
diff --git a/include/linux/kmemleak.h b/include/linux/kmemleak.h
index 6a3cd1bf4680..93a73c076d16 100644
--- a/include/linux/kmemleak.h
+++ b/include/linux/kmemleak.h
@@ -26,6 +26,7 @@ extern void kmemleak_free_part(const void *ptr, size_t size) __ref;
extern void kmemleak_free_percpu(const void __percpu *ptr) __ref;
extern void kmemleak_update_trace(const void *ptr) __ref;
extern void kmemleak_not_leak(const void *ptr) __ref;
+extern void kmemleak_transient_leak(const void *ptr) __ref;
extern void kmemleak_ignore(const void *ptr) __ref;
extern void kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) __ref;
extern void kmemleak_no_scan(const void *ptr) __ref;
@@ -93,6 +94,9 @@ static inline void kmemleak_update_trace(const void *ptr)
static inline void kmemleak_not_leak(const void *ptr)
{
}
+static inline void kmemleak_transient_leak(const void *ptr)
+{
+}
static inline void kmemleak_ignore(const void *ptr)
{
}
diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index 0400f5e8ac60..72e09ac9140b 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -934,6 +934,28 @@ static void make_black_object(unsigned long ptr, unsigned int objflags)
paint_ptr(ptr, KMEMLEAK_BLACK, objflags);
}
+/*
+ * Reset the checksum of an object. The immediate effect is that it will not
+ * be reported as a leak during the next scan until its checksum is updated.
+ */
+static void reset_checksum(unsigned long ptr)
+{
+ unsigned long flags;
+ struct kmemleak_object *object;
+
+ object = find_and_get_object(ptr, 0);
+ if (!object) {
+ kmemleak_warn("Not resetting the checksum of an unknown object at 0x%08lx\n",
+ ptr);
+ return;
+ }
+
+ raw_spin_lock_irqsave(&object->lock, flags);
+ object->checksum = 0;
+ raw_spin_unlock_irqrestore(&object->lock, flags);
+ put_object(object);
+}
+
/*
* Add a scanning area to the object. If at least one such area is added,
* kmemleak will only scan these ranges rather than the whole memory block.
@@ -1202,6 +1224,23 @@ void __ref kmemleak_not_leak(const void *ptr)
}
EXPORT_SYMBOL(kmemleak_not_leak);
+/**
+ * kmemleak_transient_leak - mark an allocated object as transient false positive
+ * @ptr: pointer to beginning of the object
+ *
+ * Calling this function on an object will cause the memory block to not be
+ * reported as a leak temporarily. This may happen, for example, if the object
+ * is part of a singly linked list and the ->next reference to it is changed.
+ */
+void __ref kmemleak_transient_leak(const void *ptr)
+{
+ pr_debug("%s(0x%px)\n", __func__, ptr);
+
+ if (kmemleak_enabled && ptr && !IS_ERR(ptr))
+ reset_checksum((unsigned long)ptr);
+}
+EXPORT_SYMBOL(kmemleak_transient_leak);
+
/**
* kmemleak_ignore - ignore an allocated object
* @ptr: pointer to beginning of the object
