netns xfrm: fix "ip xfrm state|policy count" misreport
[safe/jmp/linux-2.6] / mm / kmemleak.c
index c977f7a..5b069e4 100644 (file)
 #include <linux/string.h>
 #include <linux/nodemask.h>
 #include <linux/mm.h>
+#include <linux/workqueue.h>
+#include <linux/crc32.h>
 
 #include <asm/sections.h>
 #include <asm/processor.h>
 #include <asm/atomic.h>
 
+#include <linux/kmemcheck.h>
 #include <linux/kmemleak.h>
 
 /*
 #define MSECS_MIN_AGE          5000    /* minimum object age for reporting */
 #define SECS_FIRST_SCAN                60      /* delay before the first scan */
 #define SECS_SCAN_WAIT         600     /* subsequent auto scanning delay */
-#define GRAY_LIST_PASSES       25      /* maximum number of gray list scans */
 #define MAX_SCAN_SIZE          4096    /* maximum size of a scanned block */
 
 #define BYTES_PER_POINTER      sizeof(void *)
 /* scanning area inside a memory block */
 struct kmemleak_scan_area {
        struct hlist_node node;
-       unsigned long offset;
-       size_t length;
+       unsigned long start;
+       size_t size;
 };
 
+#define KMEMLEAK_GREY  0
+#define KMEMLEAK_BLACK -1
+
 /*
  * Structure holding the metadata for each allocated memory block.
  * Modifications to such objects should be made while holding the
@@ -144,6 +149,8 @@ struct kmemleak_object {
        int min_count;
        /* the total number of pointers found pointing to this object */
        int count;
+       /* checksum for detecting modified objects */
+       u32 checksum;
        /* memory ranges to be scanned inside an object (empty for all) */
        struct hlist_head area_list;
        unsigned long trace[MAX_TRACE];
@@ -159,8 +166,15 @@ struct kmemleak_object {
 #define OBJECT_REPORTED                (1 << 1)
 /* flag set to not scan the object */
 #define OBJECT_NO_SCAN         (1 << 2)
-/* flag set on newly allocated objects */
-#define OBJECT_NEW             (1 << 3)
+
+/* number of bytes to print per line; must be 16 or 32 */
+#define HEX_ROW_SIZE           16
+/* number of bytes to print at a time (1, 2, 4, 8) */
+#define HEX_GROUP_SIZE         1
+/* include ASCII after the hex output */
+#define HEX_ASCII              1
+/* max number of lines to be printed */
+#define HEX_MAX_LINES          2
 
 /* the list of all allocated objects */
 static LIST_HEAD(object_list);
@@ -227,13 +241,14 @@ struct early_log {
        const void *ptr;                /* allocated/freed memory block */
        size_t size;                    /* memory block size */
        int min_count;                  /* minimum reference count */
-       unsigned long offset;           /* scan area offset */
-       size_t length;                  /* scan area length */
+       unsigned long trace[MAX_TRACE]; /* stack trace */
+       unsigned int trace_len;         /* stack trace length */
 };
 
 /* early logging buffer and current position */
-static struct early_log early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE];
-static int crt_early_log;
+static struct early_log
+       early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
+static int crt_early_log __initdata;
 
 static void kmemleak_disable(void);
 
@@ -256,6 +271,35 @@ static void kmemleak_disable(void);
 } while (0)
 
 /*
+ * Printing of the objects hex dump to the seq file. The number of lines to be
+ * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
+ * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
+ * with the object->lock held.
+ */
+static void hex_dump_object(struct seq_file *seq,
+                           struct kmemleak_object *object)
+{
+       const u8 *ptr = (const u8 *)object->pointer;
+       int i, len, remaining;
+       unsigned char linebuf[HEX_ROW_SIZE * 5];
+
+       /* limit the number of lines to HEX_MAX_LINES */
+       remaining = len =
+               min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
+
+       seq_printf(seq, "  hex dump (first %d bytes):\n", len);
+       for (i = 0; i < len; i += HEX_ROW_SIZE) {
+               int linelen = min(remaining, HEX_ROW_SIZE);
+
+               remaining -= HEX_ROW_SIZE;
+               hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
+                                  HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
+                                  HEX_ASCII);
+               seq_printf(seq, "    %s\n", linebuf);
+       }
+}
+
+/*
  * Object colors, encoded with count and min_count:
  * - white - orphan object, not enough references to it (count < min_count)
  * - gray  - not orphan, not marked as false positive (min_count == 0) or
@@ -265,19 +309,16 @@ static void kmemleak_disable(void);
  * Newly created objects don't have any color assigned (object->count == -1)
  * before the next memory scan when they become white.
  */
-static int color_white(const struct kmemleak_object *object)
-{
-       return object->count != -1 && object->count < object->min_count;
-}
-
-static int color_gray(const struct kmemleak_object *object)
+static bool color_white(const struct kmemleak_object *object)
 {
-       return object->min_count != -1 && object->count >= object->min_count;
+       return object->count != KMEMLEAK_BLACK &&
+               object->count < object->min_count;
 }
 
-static int color_black(const struct kmemleak_object *object)
+static bool color_gray(const struct kmemleak_object *object)
 {
-       return object->min_count == -1;
+       return object->min_count != KMEMLEAK_BLACK &&
+               object->count >= object->min_count;
 }
 
 /*
@@ -285,9 +326,9 @@ static int color_black(const struct kmemleak_object *object)
  * not be deleted and have a minimum age to avoid false positives caused by
  * pointers temporarily stored in CPU registers.
  */
-static int unreferenced_object(struct kmemleak_object *object)
+static bool unreferenced_object(struct kmemleak_object *object)
 {
-       return (object->flags & OBJECT_ALLOCATED) && color_white(object) &&
+       return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
                time_before_eq(object->jiffies + jiffies_min_age,
                               jiffies_last_scan);
 }
@@ -300,11 +341,14 @@ static void print_unreferenced(struct seq_file *seq,
                               struct kmemleak_object *object)
 {
        int i;
+       unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
 
        seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
                   object->pointer, object->size);
-       seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu\n",
-                  object->comm, object->pid, object->jiffies);
+       seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
+                  object->comm, object->pid, object->jiffies,
+                  msecs_age / 1000, msecs_age % 1000);
+       hex_dump_object(seq, object);
        seq_printf(seq, "  backtrace:\n");
 
        for (i = 0; i < object->trace_len; i++) {
@@ -332,6 +376,7 @@ static void dump_object_info(struct kmemleak_object *object)
        pr_notice("  min_count = %d\n", object->min_count);
        pr_notice("  count = %d\n", object->count);
        pr_notice("  flags = 0x%lx\n", object->flags);
+       pr_notice("  checksum = %d\n", object->checksum);
        pr_notice("  backtrace:\n");
        print_stack_trace(&trace, 4);
 }
@@ -436,21 +481,36 @@ static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
 }
 
 /*
+ * Save stack trace to the given array of MAX_TRACE size.
+ */
+static int __save_stack_trace(unsigned long *trace)
+{
+       struct stack_trace stack_trace;
+
+       stack_trace.max_entries = MAX_TRACE;
+       stack_trace.nr_entries = 0;
+       stack_trace.entries = trace;
+       stack_trace.skip = 2;
+       save_stack_trace(&stack_trace);
+
+       return stack_trace.nr_entries;
+}
+
+/*
  * Create the metadata (struct kmemleak_object) corresponding to an allocated
  * memory block and add it to the object_list and object_tree_root.
  */
-static void create_object(unsigned long ptr, size_t size, int min_count,
-                         gfp_t gfp)
+static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
+                                            int min_count, gfp_t gfp)
 {
        unsigned long flags;
        struct kmemleak_object *object;
        struct prio_tree_node *node;
-       struct stack_trace trace;
 
        object = kmem_cache_alloc(object_cache, gfp & GFP_KMEMLEAK_MASK);
        if (!object) {
                kmemleak_stop("Cannot allocate a kmemleak_object structure\n");
-               return;
+               return NULL;
        }
 
        INIT_LIST_HEAD(&object->object_list);
@@ -458,12 +518,13 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
        INIT_HLIST_HEAD(&object->area_list);
        spin_lock_init(&object->lock);
        atomic_set(&object->use_count, 1);
-       object->flags = OBJECT_ALLOCATED | OBJECT_NEW;
+       object->flags = OBJECT_ALLOCATED;
        object->pointer = ptr;
        object->size = size;
        object->min_count = min_count;
-       object->count = -1;                     /* no color initially */
+       object->count = 0;                      /* white color initially */
        object->jiffies = jiffies;
+       object->checksum = 0;
 
        /* task information */
        if (in_irq()) {
@@ -484,18 +545,14 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
        }
 
        /* kernel backtrace */
-       trace.max_entries = MAX_TRACE;
-       trace.nr_entries = 0;
-       trace.entries = object->trace;
-       trace.skip = 1;
-       save_stack_trace(&trace);
-       object->trace_len = trace.nr_entries;
+       object->trace_len = __save_stack_trace(object->trace);
 
        INIT_PRIO_TREE_NODE(&object->tree_node);
        object->tree_node.start = ptr;
        object->tree_node.last = ptr + size - 1;
 
        write_lock_irqsave(&kmemleak_lock, flags);
+
        min_addr = min(min_addr, ptr);
        max_addr = max(max_addr, ptr + size);
        node = prio_tree_insert(&object_tree_root, &object->tree_node);
@@ -506,20 +563,19 @@ static void create_object(unsigned long ptr, size_t size, int min_count,
         * random memory blocks.
         */
        if (node != &object->tree_node) {
-               unsigned long flags;
-
                kmemleak_stop("Cannot insert 0x%lx into the object search tree "
                              "(already existing)\n", ptr);
                object = lookup_object(ptr, 1);
-               spin_lock_irqsave(&object->lock, flags);
+               spin_lock(&object->lock);
                dump_object_info(object);
-               spin_unlock_irqrestore(&object->lock, flags);
+               spin_unlock(&object->lock);
 
                goto out;
        }
        list_add_tail_rcu(&object->object_list, &object_list);
 out:
        write_unlock_irqrestore(&kmemleak_lock, flags);
+       return object;
 }
 
 /*
@@ -606,61 +662,68 @@ static void delete_object_part(unsigned long ptr, size_t size)
 
        put_object(object);
 }
-/*
- * Make a object permanently as gray-colored so that it can no longer be
- * reported as a leak. This is used in general to mark a false positive.
- */
-static void make_gray_object(unsigned long ptr)
+
+static void __paint_it(struct kmemleak_object *object, int color)
+{
+       object->min_count = color;
+       if (color == KMEMLEAK_BLACK)
+               object->flags |= OBJECT_NO_SCAN;
+}
+
+static void paint_it(struct kmemleak_object *object, int color)
 {
        unsigned long flags;
+
+       spin_lock_irqsave(&object->lock, flags);
+       __paint_it(object, color);
+       spin_unlock_irqrestore(&object->lock, flags);
+}
+
+static void paint_ptr(unsigned long ptr, int color)
+{
        struct kmemleak_object *object;
 
        object = find_and_get_object(ptr, 0);
        if (!object) {
-               kmemleak_warn("Graying unknown object at 0x%08lx\n", ptr);
+               kmemleak_warn("Trying to color unknown object "
+                             "at 0x%08lx as %s\n", ptr,
+                             (color == KMEMLEAK_GREY) ? "Grey" :
+                             (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
                return;
        }
-
-       spin_lock_irqsave(&object->lock, flags);
-       object->min_count = 0;
-       spin_unlock_irqrestore(&object->lock, flags);
+       paint_it(object, color);
        put_object(object);
 }
 
 /*
+ * Make a object permanently as gray-colored so that it can no longer be
+ * reported as a leak. This is used in general to mark a false positive.
+ */
+static void make_gray_object(unsigned long ptr)
+{
+       paint_ptr(ptr, KMEMLEAK_GREY);
+}
+
+/*
  * Mark the object as black-colored so that it is ignored from scans and
  * reporting.
  */
 static void make_black_object(unsigned long ptr)
 {
-       unsigned long flags;
-       struct kmemleak_object *object;
-
-       object = find_and_get_object(ptr, 0);
-       if (!object) {
-               kmemleak_warn("Blacking unknown object at 0x%08lx\n", ptr);
-               return;
-       }
-
-       spin_lock_irqsave(&object->lock, flags);
-       object->min_count = -1;
-       object->flags |= OBJECT_NO_SCAN;
-       spin_unlock_irqrestore(&object->lock, flags);
-       put_object(object);
+       paint_ptr(ptr, KMEMLEAK_BLACK);
 }
 
 /*
  * 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.
  */
-static void add_scan_area(unsigned long ptr, unsigned long offset,
-                         size_t length, gfp_t gfp)
+static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
 {
        unsigned long flags;
        struct kmemleak_object *object;
        struct kmemleak_scan_area *area;
 
-       object = find_and_get_object(ptr, 0);
+       object = find_and_get_object(ptr, 1);
        if (!object) {
                kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
                              ptr);
@@ -674,7 +737,7 @@ static void add_scan_area(unsigned long ptr, unsigned long offset,
        }
 
        spin_lock_irqsave(&object->lock, flags);
-       if (offset + length > object->size) {
+       if (ptr + size > object->pointer + object->size) {
                kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
                dump_object_info(object);
                kmem_cache_free(scan_area_cache, area);
@@ -682,8 +745,8 @@ static void add_scan_area(unsigned long ptr, unsigned long offset,
        }
 
        INIT_HLIST_NODE(&area->node);
-       area->offset = offset;
-       area->length = length;
+       area->start = ptr;
+       area->size = size;
 
        hlist_add_head(&area->node, &object->area_list);
 out_unlock:
@@ -718,14 +781,15 @@ static void object_no_scan(unsigned long ptr)
  * Log an early kmemleak_* call to the early_log buffer. These calls will be
  * processed later once kmemleak is fully initialized.
  */
-static void log_early(int op_type, const void *ptr, size_t size,
-                     int min_count, unsigned long offset, size_t length)
+static void __init log_early(int op_type, const void *ptr, size_t size,
+                            int min_count)
 {
        unsigned long flags;
        struct early_log *log;
 
        if (crt_early_log >= ARRAY_SIZE(early_log)) {
-               pr_warning("Early log buffer exceeded\n");
+               pr_warning("Early log buffer exceeded, "
+                          "please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n");
                kmemleak_disable();
                return;
        }
@@ -740,25 +804,55 @@ static void log_early(int op_type, const void *ptr, size_t size,
        log->ptr = ptr;
        log->size = size;
        log->min_count = min_count;
-       log->offset = offset;
-       log->length = length;
+       if (op_type == KMEMLEAK_ALLOC)
+               log->trace_len = __save_stack_trace(log->trace);
        crt_early_log++;
        local_irq_restore(flags);
 }
 
 /*
+ * Log an early allocated block and populate the stack trace.
+ */
+static void early_alloc(struct early_log *log)
+{
+       struct kmemleak_object *object;
+       unsigned long flags;
+       int i;
+
+       if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
+               return;
+
+       /*
+        * RCU locking needed to ensure object is not freed via put_object().
+        */
+       rcu_read_lock();
+       object = create_object((unsigned long)log->ptr, log->size,
+                              log->min_count, GFP_ATOMIC);
+       if (!object)
+               goto out;
+       spin_lock_irqsave(&object->lock, flags);
+       for (i = 0; i < log->trace_len; i++)
+               object->trace[i] = log->trace[i];
+       object->trace_len = log->trace_len;
+       spin_unlock_irqrestore(&object->lock, flags);
+out:
+       rcu_read_unlock();
+}
+
+/*
  * Memory allocation function callback. This function is called from the
  * kernel allocators when a new block is allocated (kmem_cache_alloc, kmalloc,
  * vmalloc etc.).
  */
-void kmemleak_alloc(const void *ptr, size_t size, int min_count, gfp_t gfp)
+void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
+                         gfp_t gfp)
 {
        pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                create_object((unsigned long)ptr, size, min_count, gfp);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_ALLOC, ptr, size, min_count, 0, 0);
+               log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
 }
 EXPORT_SYMBOL_GPL(kmemleak_alloc);
 
@@ -766,14 +860,14 @@ EXPORT_SYMBOL_GPL(kmemleak_alloc);
  * Memory freeing function callback. This function is called from the kernel
  * allocators when a block is freed (kmem_cache_free, kfree, vfree etc.).
  */
-void kmemleak_free(const void *ptr)
+void __ref kmemleak_free(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                delete_object_full((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_FREE, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_FREE, ptr, 0, 0);
 }
 EXPORT_SYMBOL_GPL(kmemleak_free);
 
@@ -781,14 +875,14 @@ EXPORT_SYMBOL_GPL(kmemleak_free);
  * Partial memory freeing function callback. This function is usually called
  * from bootmem allocator when (part of) a memory block is freed.
  */
-void kmemleak_free_part(const void *ptr, size_t size)
+void __ref kmemleak_free_part(const void *ptr, size_t size)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                delete_object_part((unsigned long)ptr, size);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_FREE_PART, ptr, size, 0, 0, 0);
+               log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
 }
 EXPORT_SYMBOL_GPL(kmemleak_free_part);
 
@@ -796,14 +890,14 @@ EXPORT_SYMBOL_GPL(kmemleak_free_part);
  * Mark an already allocated memory block as a false positive. This will cause
  * the block to no longer be reported as leak and always be scanned.
  */
-void kmemleak_not_leak(const void *ptr)
+void __ref kmemleak_not_leak(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                make_gray_object((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_not_leak);
 
@@ -812,47 +906,60 @@ EXPORT_SYMBOL(kmemleak_not_leak);
  * corresponding block is not a leak and does not contain any references to
  * other allocated memory blocks.
  */
-void kmemleak_ignore(const void *ptr)
+void __ref kmemleak_ignore(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                make_black_object((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_IGNORE, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_ignore);
 
 /*
  * Limit the range to be scanned in an allocated memory block.
  */
-void kmemleak_scan_area(const void *ptr, unsigned long offset, size_t length,
-                       gfp_t gfp)
+void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
-               add_scan_area((unsigned long)ptr, offset, length, gfp);
+               add_scan_area((unsigned long)ptr, size, gfp);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_SCAN_AREA, ptr, 0, 0, offset, length);
+               log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
 }
 EXPORT_SYMBOL(kmemleak_scan_area);
 
 /*
  * Inform kmemleak not to scan the given memory block.
  */
-void kmemleak_no_scan(const void *ptr)
+void __ref kmemleak_no_scan(const void *ptr)
 {
        pr_debug("%s(0x%p)\n", __func__, ptr);
 
        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
                object_no_scan((unsigned long)ptr);
        else if (atomic_read(&kmemleak_early_log))
-               log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0, 0, 0);
+               log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
 }
 EXPORT_SYMBOL(kmemleak_no_scan);
 
 /*
+ * Update an object's checksum and return true if it was modified.
+ */
+static bool update_checksum(struct kmemleak_object *object)
+{
+       u32 old_csum = object->checksum;
+
+       if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
+               return false;
+
+       object->checksum = crc32(0, (void *)object->pointer, object->size);
+       return object->checksum != old_csum;
+}
+
+/*
  * Memory scanning is a long process and it needs to be interruptable. This
  * function checks whether such interrupt condition occured.
  */
@@ -885,15 +992,22 @@ static void scan_block(void *_start, void *_end,
        unsigned long *end = _end - (BYTES_PER_POINTER - 1);
 
        for (ptr = start; ptr < end; ptr++) {
-               unsigned long flags;
-               unsigned long pointer = *ptr;
                struct kmemleak_object *object;
+               unsigned long flags;
+               unsigned long pointer;
 
                if (allow_resched)
                        cond_resched();
                if (scan_should_stop())
                        break;
 
+               /* don't scan uninitialized memory */
+               if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
+                                                 BYTES_PER_POINTER))
+                       continue;
+
+               pointer = *ptr;
+
                object = find_and_get_object(pointer, 1);
                if (!object)
                        continue;
@@ -924,11 +1038,14 @@ static void scan_block(void *_start, void *_end,
                 * added to the gray_list.
                 */
                object->count++;
-               if (color_gray(object))
+               if (color_gray(object)) {
                        list_add_tail(&object->gray_list, &gray_list);
-               else
-                       put_object(object);
+                       spin_unlock_irqrestore(&object->lock, flags);
+                       continue;
+               }
+
                spin_unlock_irqrestore(&object->lock, flags);
+               put_object(object);
        }
 }
 
@@ -943,8 +1060,8 @@ static void scan_object(struct kmemleak_object *object)
        unsigned long flags;
 
        /*
-        * Once the object->lock is aquired, the corresponding memory block
-        * cannot be freed (the same lock is aquired in delete_object).
+        * Once the object->lock is acquired, the corresponding memory block
+        * cannot be freed (the same lock is acquired in delete_object).
         */
        spin_lock_irqsave(&object->lock, flags);
        if (object->flags & OBJECT_NO_SCAN)
@@ -968,14 +1085,47 @@ static void scan_object(struct kmemleak_object *object)
                }
        } else
                hlist_for_each_entry(area, elem, &object->area_list, node)
-                       scan_block((void *)(object->pointer + area->offset),
-                                  (void *)(object->pointer + area->offset
-                                           + area->length), object, 0);
+                       scan_block((void *)area->start,
+                                  (void *)(area->start + area->size),
+                                  object, 0);
 out:
        spin_unlock_irqrestore(&object->lock, flags);
 }
 
 /*
+ * Scan the objects already referenced (gray objects). More objects will be
+ * referenced and, if there are no memory leaks, all the objects are scanned.
+ */
+static void scan_gray_list(void)
+{
+       struct kmemleak_object *object, *tmp;
+
+       /*
+        * The list traversal is safe for both tail additions and removals
+        * from inside the loop. The kmemleak objects cannot be freed from
+        * outside the loop because their use_count was incremented.
+        */
+       object = list_entry(gray_list.next, typeof(*object), gray_list);
+       while (&object->gray_list != &gray_list) {
+               cond_resched();
+
+               /* may add new objects to the list */
+               if (!scan_should_stop())
+                       scan_object(object);
+
+               tmp = list_entry(object->gray_list.next, typeof(*object),
+                                gray_list);
+
+               /* remove the object from the list and release it */
+               list_del(&object->gray_list);
+               put_object(object);
+
+               object = tmp;
+       }
+       WARN_ON(!list_empty(&gray_list));
+}
+
+/*
  * Scan data sections and all the referenced memory blocks allocated via the
  * kernel's standard allocators. This function must be called with the
  * scan_mutex held.
@@ -983,11 +1133,9 @@ out:
 static void kmemleak_scan(void)
 {
        unsigned long flags;
-       struct kmemleak_object *object, *tmp;
-       struct task_struct *task;
+       struct kmemleak_object *object;
        int i;
        int new_leaks = 0;
-       int gray_list_pass = 0;
 
        jiffies_last_scan = jiffies;
 
@@ -1008,7 +1156,6 @@ static void kmemleak_scan(void)
 #endif
                /* reset the reference count (whiten the object) */
                object->count = 0;
-               object->flags &= ~OBJECT_NEW;
                if (color_gray(object) && get_object(object))
                        list_add_tail(&object->gray_list, &gray_list);
 
@@ -1051,76 +1198,51 @@ static void kmemleak_scan(void)
        }
 
        /*
-        * Scanning the task stacks may introduce false negatives and it is
-        * not enabled by default.
+        * Scanning the task stacks (may introduce false negatives).
         */
        if (kmemleak_stack_scan) {
+               struct task_struct *p, *g;
+
                read_lock(&tasklist_lock);
-               for_each_process(task)
-                       scan_block(task_stack_page(task),
-                                  task_stack_page(task) + THREAD_SIZE,
-                                  NULL, 0);
+               do_each_thread(g, p) {
+                       scan_block(task_stack_page(p), task_stack_page(p) +
+                                  THREAD_SIZE, NULL, 0);
+               } while_each_thread(g, p);
                read_unlock(&tasklist_lock);
        }
 
        /*
         * Scan the objects already referenced from the sections scanned
-        * above. More objects will be referenced and, if there are no memory
-        * leaks, all the objects will be scanned. The list traversal is safe
-        * for both tail additions and removals from inside the loop. The
-        * kmemleak objects cannot be freed from outside the loop because their
-        * use_count was increased.
+        * above.
         */
-repeat:
-       object = list_entry(gray_list.next, typeof(*object), gray_list);
-       while (&object->gray_list != &gray_list) {
-               cond_resched();
-
-               /* may add new objects to the list */
-               if (!scan_should_stop())
-                       scan_object(object);
-
-               tmp = list_entry(object->gray_list.next, typeof(*object),
-                                gray_list);
-
-               /* remove the object from the list and release it */
-               list_del(&object->gray_list);
-               put_object(object);
-
-               object = tmp;
-       }
-
-       if (scan_should_stop() || ++gray_list_pass >= GRAY_LIST_PASSES)
-               goto scan_end;
+       scan_gray_list();
 
        /*
-        * Check for new objects allocated during this scanning and add them
-        * to the gray list.
+        * Check for new or unreferenced objects modified since the previous
+        * scan and color them gray until the next scan.
         */
        rcu_read_lock();
        list_for_each_entry_rcu(object, &object_list, object_list) {
                spin_lock_irqsave(&object->lock, flags);
-               if ((object->flags & OBJECT_NEW) && !color_black(object) &&
-                   get_object(object)) {
-                       object->flags &= ~OBJECT_NEW;
+               if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
+                   && update_checksum(object) && get_object(object)) {
+                       /* color it gray temporarily */
+                       object->count = object->min_count;
                        list_add_tail(&object->gray_list, &gray_list);
                }
                spin_unlock_irqrestore(&object->lock, flags);
        }
        rcu_read_unlock();
 
-       if (!list_empty(&gray_list))
-               goto repeat;
-
-scan_end:
-       WARN_ON(!list_empty(&gray_list));
+       /*
+        * Re-scan the gray list for modified unreferenced objects.
+        */
+       scan_gray_list();
 
        /*
-        * If scanning was stopped or new objects were being allocated at a
-        * higher rate than gray list scanning, do not report any new
-        * unreferenced objects.
+        * If scanning was stopped do not report any new unreferenced objects.
         */
-       if (scan_should_stop() || gray_list_pass >= GRAY_LIST_PASSES)
+       if (scan_should_stop())
                return;
 
        /*
@@ -1184,7 +1306,7 @@ static int kmemleak_scan_thread(void *arg)
  * Start the automatic memory scanning thread. This function must be called
  * with the scan_mutex held.
  */
-void start_scan_thread(void)
+static void start_scan_thread(void)
 {
        if (scan_thread)
                return;
@@ -1199,7 +1321,7 @@ void start_scan_thread(void)
  * Stop the automatic memory scanning thread. This function must be called
  * with the scan_mutex held.
  */
-void stop_scan_thread(void)
+static void stop_scan_thread(void)
 {
        if (scan_thread) {
                kthread_stop(scan_thread);
@@ -1330,6 +1452,28 @@ static int dump_str_object_info(const char *str)
 }
 
 /*
+ * We use grey instead of black to ensure we can do future scans on the same
+ * objects. If we did not do future scans these black objects could
+ * potentially contain references to newly allocated objects in the future and
+ * we'd end up with false positives.
+ */
+static void kmemleak_clear(void)
+{
+       struct kmemleak_object *object;
+       unsigned long flags;
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(object, &object_list, object_list) {
+               spin_lock_irqsave(&object->lock, flags);
+               if ((object->flags & OBJECT_REPORTED) &&
+                   unreferenced_object(object))
+                       __paint_it(object, KMEMLEAK_GREY);
+               spin_unlock_irqrestore(&object->lock, flags);
+       }
+       rcu_read_unlock();
+}
+
+/*
  * File write operation to configure kmemleak at run-time. The following
  * commands can be written to the /sys/kernel/debug/kmemleak file:
  *   off       - disable kmemleak (irreversible)
@@ -1340,6 +1484,8 @@ static int dump_str_object_info(const char *str)
  *   scan=...  - set the automatic memory scanning period in seconds (0 to
  *               disable it)
  *   scan      - trigger a memory scan
+ *   clear     - mark all current reported unreferenced kmemleak objects as
+ *               grey to ignore printing them
  *   dump=...  - dump information about the object found at the given address
  */
 static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
@@ -1381,6 +1527,8 @@ static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
                }
        } else if (strncmp(buf, "scan", 4) == 0)
                kmemleak_scan();
+       else if (strncmp(buf, "clear", 5) == 0)
+               kmemleak_clear();
        else if (strncmp(buf, "dump=", 5) == 0)
                ret = dump_str_object_info(buf + 5);
        else
@@ -1409,7 +1557,7 @@ static const struct file_operations kmemleak_fops = {
  * Perform the freeing of the kmemleak internal objects after waiting for any
  * current memory scan to complete.
  */
-static int kmemleak_cleanup_thread(void *arg)
+static void kmemleak_do_cleanup(struct work_struct *work)
 {
        struct kmemleak_object *object;
 
@@ -1421,22 +1569,9 @@ static int kmemleak_cleanup_thread(void *arg)
                delete_object_full(object->pointer);
        rcu_read_unlock();
        mutex_unlock(&scan_mutex);
-
-       return 0;
 }
 
-/*
- * Start the clean-up thread.
- */
-static void kmemleak_cleanup(void)
-{
-       struct task_struct *cleanup_thread;
-
-       cleanup_thread = kthread_run(kmemleak_cleanup_thread, NULL,
-                                    "kmemleak-clean");
-       if (IS_ERR(cleanup_thread))
-               pr_warning("Failed to create the clean-up thread\n");
-}
+static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
 
 /*
  * Disable kmemleak. No memory allocation/freeing will be traced once this
@@ -1454,7 +1589,7 @@ static void kmemleak_disable(void)
 
        /* check whether it is too early for a kernel thread */
        if (atomic_read(&kmemleak_initialized))
-               kmemleak_cleanup();
+               schedule_work(&cleanup_work);
 
        pr_info("Kernel memory leak detector disabled\n");
 }
@@ -1507,8 +1642,7 @@ void __init kmemleak_init(void)
 
                switch (log->op_type) {
                case KMEMLEAK_ALLOC:
-                       kmemleak_alloc(log->ptr, log->size, log->min_count,
-                                      GFP_KERNEL);
+                       early_alloc(log);
                        break;
                case KMEMLEAK_FREE:
                        kmemleak_free(log->ptr);
@@ -1523,8 +1657,7 @@ void __init kmemleak_init(void)
                        kmemleak_ignore(log->ptr);
                        break;
                case KMEMLEAK_SCAN_AREA:
-                       kmemleak_scan_area(log->ptr, log->offset, log->length,
-                                          GFP_KERNEL);
+                       kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
                        break;
                case KMEMLEAK_NO_SCAN:
                        kmemleak_no_scan(log->ptr);
@@ -1551,7 +1684,7 @@ static int __init kmemleak_late_init(void)
                 * after setting kmemleak_initialized and we may end up with
                 * two clean-up threads but serialized by scan_mutex.
                 */
-               kmemleak_cleanup();
+               schedule_work(&cleanup_work);
                return -ENOMEM;
        }