Re: [PATCH v2 3/4] mm: kmemleak: handle address stored in object based on its type

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 2022/6/6 23:01, Catalin Marinas wrote:
On Fri, Jun 03, 2022 at 11:54:14AM +0800, Patrick Wang wrote:
Treat the address stored in object in different way according
to its type:

- Only use kasan_reset_tag for virtual address
- Only update min_addr and max_addr for virtual address
- Convert physical address to virtual address in scan_object

Suggested-by: Catalin Marinas <catalin.marinas@xxxxxxx>
Signed-off-by: Patrick Wang <patrick.wang.shcn@xxxxxxxxx>
---
  mm/kmemleak.c | 34 ++++++++++++++++++++++++----------
  1 file changed, 24 insertions(+), 10 deletions(-)

diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index 218144392446..246a70b7218f 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -297,7 +297,9 @@ static void hex_dump_object(struct seq_file *seq,
  	warn_or_seq_printf(seq, "  hex dump (first %zu bytes):\n", len);
  	kasan_disable_current();
  	warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
-			     HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
+			     HEX_GROUP_SIZE, object->flags & OBJECT_PHYS ? ptr :
+			     kasan_reset_tag((void *)ptr),
+			     len, HEX_ASCII);
  	kasan_enable_current();
  }

This will go wrong since ptr is the actual physical address, it cannot
be dereferenced. This should only be used on virtual pointers and this
is the case already as we never print unreferenced objects from the phys
tree. What we could do though is something like an early exit from this
function (together with a comment that it doesn't support dumping such
objects):

	if (WARN_ON_ONCE(object->flags & OBJECT_PHYS))
		return;


I also found this. Will do.

@@ -389,14 +391,15 @@ static struct kmemleak_object *lookup_object(unsigned long ptr, int alias,
  {
  	struct rb_node *rb = is_phys ? object_phys_tree_root.rb_node :
  			     object_tree_root.rb_node;
-	unsigned long untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
+	unsigned long untagged_ptr = is_phys ? ptr : (unsigned long)kasan_reset_tag((void *)ptr);
while (rb) {
  		struct kmemleak_object *object;
  		unsigned long untagged_objp;
object = rb_entry(rb, struct kmemleak_object, rb_node);
-		untagged_objp = (unsigned long)kasan_reset_tag((void *)object->pointer);
+		untagged_objp = is_phys ? object->pointer :
+				(unsigned long)kasan_reset_tag((void *)object->pointer);
if (untagged_ptr < untagged_objp)
  			rb = object->rb_node.rb_left;

You could leave this unchanged. A phys pointer is already untagged, so
it wouldn't make any difference.

Right, will do.


@@ -643,16 +646,19 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
raw_spin_lock_irqsave(&kmemleak_lock, flags); - untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
-	min_addr = min(min_addr, untagged_ptr);
-	max_addr = max(max_addr, untagged_ptr + size);
+	untagged_ptr = is_phys ? ptr : (unsigned long)kasan_reset_tag((void *)ptr);

Same here.

Will do.


+	if (!is_phys) {
+		min_addr = min(min_addr, untagged_ptr);
+		max_addr = max(max_addr, untagged_ptr + size);
+	}
  	link = is_phys ? &object_phys_tree_root.rb_node :
  		&object_tree_root.rb_node;
  	rb_parent = NULL;
  	while (*link) {
  		rb_parent = *link;
  		parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
-		untagged_objp = (unsigned long)kasan_reset_tag((void *)parent->pointer);
+		untagged_objp = is_phys ? parent->pointer :
+				(unsigned long)kasan_reset_tag((void *)parent->pointer);

And here.

Will do.


  		if (untagged_ptr + size <= untagged_objp)
  			link = &parent->rb_node.rb_left;
  		else if (untagged_objp + parent->size <= untagged_ptr)
@@ -1202,7 +1208,9 @@ static bool update_checksum(struct kmemleak_object *object)
kasan_disable_current();
  	kcsan_disable_current();
-	object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
+	object->checksum = crc32(0, object->flags & OBJECT_PHYS ? (void *)object->pointer :
+				    kasan_reset_tag((void *)object->pointer),
+				    object->size);

Luckily that's never called on a phys object, otherwise *object->pointer
would segfault. As for hex_dump, just return early with a warning if
that's the case.

Right, will do.

Thanks,
Patrick





[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux