kmemleak: enable tracking for percpu pointers

Patch series "kmemleak: support for percpu memory leak detect'.

This is a rework of this series:
https://lore.kernel.org/lkml/20200921020007.35803-1-chenjun102@huawei.com/

Originally I was investigating a percpu leak on our customer nodes and
having this functionality was a huge help, which lead to this fix [1].

So probably it's a good idea to have it in mainstream too, especially as
after [2] it became much easier to implement (we already have a separate
tree for percpu pointers).

[1] commit 0af8c09c ("netfilter: x_tables: fix percpu counter block leak on error path when creating new netns")
[2] commit 39042079 ("kmemleak: avoid RCU stalls when freeing metadata for per-CPU pointers")


This patch (of 2):

This basically does:

- Add min_percpu_addr and max_percpu_addr to filter out unrelated data
  similar to min_addr and max_addr;

- Set min_count for percpu pointers to 1 to start tracking them;

- Calculate checksum of percpu area as xor of crc32 for each cpu;

- Split pointer lookup and update refs code into separate helper and use
  it twice: once as if the pointer is a virtual pointer and once as if
  it's percpu.

[ptikhomirov@virtuozzo.com: v2]
  Link: https://lkml.kernel.org/r/20240731025526.157529-2-ptikhomirov@virtuozzo.com
Link: https://lkml.kernel.org/r/20240725041223.872472-1-ptikhomirov@virtuozzo.com
Link: https://lkml.kernel.org/r/20240725041223.872472-2-ptikhomirov@virtuozzo.comSigned-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Wei Yongjun <weiyongjun1@huawei.com>
Cc: Chen Jun <chenjun102@huawei.com>
Cc: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm/kmemleak.c

@@ -224,6 +224,10 @@ static int kmemleak_error;
 static unsigned long min_addr = ULONG_MAX;
 static unsigned long max_addr;
 
+/* minimum and maximum address that may be valid per-CPU pointers */
+static unsigned long min_percpu_addr = ULONG_MAX;
+static unsigned long max_percpu_addr;
+
 static struct task_struct *scan_thread;
 /* used to avoid reporting of recently allocated objects */
 static unsigned long jiffies_min_age;
@@ -294,12 +298,19 @@ static void hex_dump_object(struct seq_file *seq,
 	const u8 *ptr = (const u8 *)object->pointer;
 	size_t len;
 
-	if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
+	if (WARN_ON_ONCE(object->flags & OBJECT_PHYS))
 		return;
 
+	if (object->flags & OBJECT_PERCPU)
+		ptr = (const u8 *)this_cpu_ptr((void __percpu *)object->pointer);
+
 	/* limit the number of lines to HEX_MAX_LINES */
 	len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
 
+	if (object->flags & OBJECT_PERCPU)
+		warn_or_seq_printf(seq, "  hex dump (first %zu bytes on cpu %d):\n",
+				   len, raw_smp_processor_id());
+	else
 		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,
@@ -695,10 +706,14 @@ static int __link_object(struct kmemleak_object *object, unsigned long ptr,
 
 	untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
 	/*
-	 * Only update min_addr and max_addr with object
-	 * storing virtual address.
-	 */
-	if (!(objflags & (OBJECT_PHYS | OBJECT_PERCPU))) {
+	 * Only update min_addr and max_addr with object storing virtual
+	 * address. And update min_percpu_addr max_percpu_addr for per-CPU
+	 * objects.
+	 */
+	if (objflags & OBJECT_PERCPU) {
+		min_percpu_addr = min(min_percpu_addr, untagged_ptr);
+		max_percpu_addr = max(max_percpu_addr, untagged_ptr + size);
+	} else if (!(objflags & OBJECT_PHYS)) {
 		min_addr = min(min_addr, untagged_ptr);
 		max_addr = max(max_addr, untagged_ptr + size);
 	}
@@ -1055,12 +1070,8 @@ void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
 {
 	pr_debug("%s(0x%px, %zu)\n", __func__, ptr, size);
 
-	/*
-	 * Percpu allocations are only scanned and not reported as leaks
-	 * (min_count is set to 0).
-	 */
 	if (kmemleak_enabled && ptr && !IS_ERR(ptr))
-		create_object_percpu((unsigned long)ptr, size, 0, gfp);
+		create_object_percpu((unsigned long)ptr, size, 1, gfp);
 }
 EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
 
@@ -1304,12 +1315,23 @@ static bool update_checksum(struct kmemleak_object *object)
 {
 	u32 old_csum = object->checksum;
 
-	if (WARN_ON_ONCE(object->flags & (OBJECT_PHYS | OBJECT_PERCPU)))
+	if (WARN_ON_ONCE(object->flags & OBJECT_PHYS))
 		return false;
 
 	kasan_disable_current();
 	kcsan_disable_current();
+	if (object->flags & OBJECT_PERCPU) {
+		unsigned int cpu;
+
+		object->checksum = 0;
+		for_each_possible_cpu(cpu) {
+			void *ptr = per_cpu_ptr((void __percpu *)object->pointer, cpu);
+
+			object->checksum ^= crc32(0, kasan_reset_tag((void *)ptr), object->size);
+		}
+	} else {
 		object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
+	}
 	kasan_enable_current();
 	kcsan_enable_current();
 
@@ -1340,6 +1362,64 @@ static void update_refs(struct kmemleak_object *object)
 	}
 }
 
+static void pointer_update_refs(struct kmemleak_object *scanned,
+			 unsigned long pointer, unsigned int objflags)
+{
+	struct kmemleak_object *object;
+	unsigned long untagged_ptr;
+	unsigned long excess_ref;
+
+	untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
+	if (objflags & OBJECT_PERCPU) {
+		if (untagged_ptr < min_percpu_addr || untagged_ptr >= max_percpu_addr)
+			return;
+	} else {
+		if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
+			return;
+	}
+
+	/*
+	 * No need for get_object() here since we hold kmemleak_lock.
+	 * object->use_count cannot be dropped to 0 while the object
+	 * is still present in object_tree_root and object_list
+	 * (with updates protected by kmemleak_lock).
+	 */
+	object = __lookup_object(pointer, 1, objflags);
+	if (!object)
+		return;
+	if (object == scanned)
+		/* self referenced, ignore */
+		return;
+
+	/*
+	 * Avoid the lockdep recursive warning on object->lock being
+	 * previously acquired in scan_object(). These locks are
+	 * enclosed by scan_mutex.
+	 */
+	raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
+	/* only pass surplus references (object already gray) */
+	if (color_gray(object)) {
+		excess_ref = object->excess_ref;
+		/* no need for update_refs() if object already gray */
+	} else {
+		excess_ref = 0;
+		update_refs(object);
+	}
+	raw_spin_unlock(&object->lock);
+
+	if (excess_ref) {
+		object = lookup_object(excess_ref, 0);
+		if (!object)
+			return;
+		if (object == scanned)
+			/* circular reference, ignore */
+			return;
+		raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
+		update_refs(object);
+		raw_spin_unlock(&object->lock);
+	}
+}
+
 /*
  * Memory scanning is a long process and it needs to be interruptible. This
  * function checks whether such interrupt condition occurred.
@@ -1372,13 +1452,10 @@ static void scan_block(void *_start, void *_end,
 	unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
 	unsigned long *end = _end - (BYTES_PER_POINTER - 1);
 	unsigned long flags;
-	unsigned long untagged_ptr;
 
 	raw_spin_lock_irqsave(&kmemleak_lock, flags);
 	for (ptr = start; ptr < end; ptr++) {
-		struct kmemleak_object *object;
 		unsigned long pointer;
-		unsigned long excess_ref;
 
 		if (scan_should_stop())
 			break;
@@ -1387,50 +1464,8 @@ static void scan_block(void *_start, void *_end,
 		pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
 		kasan_enable_current();
 
-		untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
-		if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
-			continue;
-
-		/*
-		 * No need for get_object() here since we hold kmemleak_lock.
-		 * object->use_count cannot be dropped to 0 while the object
-		 * is still present in object_tree_root and object_list
-		 * (with updates protected by kmemleak_lock).
-		 */
-		object = lookup_object(pointer, 1);
-		if (!object)
-			continue;
-		if (object == scanned)
-			/* self referenced, ignore */
-			continue;
-
-		/*
-		 * Avoid the lockdep recursive warning on object->lock being
-		 * previously acquired in scan_object(). These locks are
-		 * enclosed by scan_mutex.
-		 */
-		raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
-		/* only pass surplus references (object already gray) */
-		if (color_gray(object)) {
-			excess_ref = object->excess_ref;
-			/* no need for update_refs() if object already gray */
-		} else {
-			excess_ref = 0;
-			update_refs(object);
-		}
-		raw_spin_unlock(&object->lock);
-
-		if (excess_ref) {
-			object = lookup_object(excess_ref, 0);
-			if (!object)
-				continue;
-			if (object == scanned)
-				/* circular reference, ignore */
-				continue;
-			raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
-			update_refs(object);
-			raw_spin_unlock(&object->lock);
-		}
+		pointer_update_refs(scanned, pointer, 0);
+		pointer_update_refs(scanned, pointer, OBJECT_PERCPU);
 	}
 	raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
 }