Root/mm/kmemleak.c

1/*
2 * mm/kmemleak.c
3 *
4 * Copyright (C) 2008 ARM Limited
5 * Written by Catalin Marinas <catalin.marinas@arm.com>
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 *
20 *
21 * For more information on the algorithm and kmemleak usage, please see
22 * Documentation/kmemleak.txt.
23 *
24 * Notes on locking
25 * ----------------
26 *
27 * The following locks and mutexes are used by kmemleak:
28 *
29 * - kmemleak_lock (rwlock): protects the object_list modifications and
30 * accesses to the object_tree_root. The object_list is the main list
31 * holding the metadata (struct kmemleak_object) for the allocated memory
32 * blocks. The object_tree_root is a red black tree used to look-up
33 * metadata based on a pointer to the corresponding memory block. The
34 * kmemleak_object structures are added to the object_list and
35 * object_tree_root in the create_object() function called from the
36 * kmemleak_alloc() callback and removed in delete_object() called from the
37 * kmemleak_free() callback
38 * - kmemleak_object.lock (spinlock): protects a kmemleak_object. Accesses to
39 * the metadata (e.g. count) are protected by this lock. Note that some
40 * members of this structure may be protected by other means (atomic or
41 * kmemleak_lock). This lock is also held when scanning the corresponding
42 * memory block to avoid the kernel freeing it via the kmemleak_free()
43 * callback. This is less heavyweight than holding a global lock like
44 * kmemleak_lock during scanning
45 * - scan_mutex (mutex): ensures that only one thread may scan the memory for
46 * unreferenced objects at a time. The gray_list contains the objects which
47 * are already referenced or marked as false positives and need to be
48 * scanned. This list is only modified during a scanning episode when the
49 * scan_mutex is held. At the end of a scan, the gray_list is always empty.
50 * Note that the kmemleak_object.use_count is incremented when an object is
51 * added to the gray_list and therefore cannot be freed. This mutex also
52 * prevents multiple users of the "kmemleak" debugfs file together with
53 * modifications to the memory scanning parameters including the scan_thread
54 * pointer
55 *
56 * The kmemleak_object structures have a use_count incremented or decremented
57 * using the get_object()/put_object() functions. When the use_count becomes
58 * 0, this count can no longer be incremented and put_object() schedules the
59 * kmemleak_object freeing via an RCU callback. All calls to the get_object()
60 * function must be protected by rcu_read_lock() to avoid accessing a freed
61 * structure.
62 */
63
64#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
65
66#include <linux/init.h>
67#include <linux/kernel.h>
68#include <linux/list.h>
69#include <linux/sched.h>
70#include <linux/jiffies.h>
71#include <linux/delay.h>
72#include <linux/export.h>
73#include <linux/kthread.h>
74#include <linux/rbtree.h>
75#include <linux/fs.h>
76#include <linux/debugfs.h>
77#include <linux/seq_file.h>
78#include <linux/cpumask.h>
79#include <linux/spinlock.h>
80#include <linux/mutex.h>
81#include <linux/rcupdate.h>
82#include <linux/stacktrace.h>
83#include <linux/cache.h>
84#include <linux/percpu.h>
85#include <linux/hardirq.h>
86#include <linux/mmzone.h>
87#include <linux/slab.h>
88#include <linux/thread_info.h>
89#include <linux/err.h>
90#include <linux/uaccess.h>
91#include <linux/string.h>
92#include <linux/nodemask.h>
93#include <linux/mm.h>
94#include <linux/workqueue.h>
95#include <linux/crc32.h>
96
97#include <asm/sections.h>
98#include <asm/processor.h>
99#include <linux/atomic.h>
100
101#include <linux/kmemcheck.h>
102#include <linux/kmemleak.h>
103#include <linux/memory_hotplug.h>
104
105/*
106 * Kmemleak configuration and common defines.
107 */
108#define MAX_TRACE 16 /* stack trace length */
109#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
110#define SECS_FIRST_SCAN 60 /* delay before the first scan */
111#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
112#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
113
114#define BYTES_PER_POINTER sizeof(void *)
115
116/* GFP bitmask for kmemleak internal allocations */
117#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
118                 __GFP_NORETRY | __GFP_NOMEMALLOC | \
119                 __GFP_NOWARN)
120
121/* scanning area inside a memory block */
122struct kmemleak_scan_area {
123    struct hlist_node node;
124    unsigned long start;
125    size_t size;
126};
127
128#define KMEMLEAK_GREY 0
129#define KMEMLEAK_BLACK -1
130
131/*
132 * Structure holding the metadata for each allocated memory block.
133 * Modifications to such objects should be made while holding the
134 * object->lock. Insertions or deletions from object_list, gray_list or
135 * rb_node are already protected by the corresponding locks or mutex (see
136 * the notes on locking above). These objects are reference-counted
137 * (use_count) and freed using the RCU mechanism.
138 */
139struct kmemleak_object {
140    spinlock_t lock;
141    unsigned long flags; /* object status flags */
142    struct list_head object_list;
143    struct list_head gray_list;
144    struct rb_node rb_node;
145    struct rcu_head rcu; /* object_list lockless traversal */
146    /* object usage count; object freed when use_count == 0 */
147    atomic_t use_count;
148    unsigned long pointer;
149    size_t size;
150    /* minimum number of a pointers found before it is considered leak */
151    int min_count;
152    /* the total number of pointers found pointing to this object */
153    int count;
154    /* checksum for detecting modified objects */
155    u32 checksum;
156    /* memory ranges to be scanned inside an object (empty for all) */
157    struct hlist_head area_list;
158    unsigned long trace[MAX_TRACE];
159    unsigned int trace_len;
160    unsigned long jiffies; /* creation timestamp */
161    pid_t pid; /* pid of the current task */
162    char comm[TASK_COMM_LEN]; /* executable name */
163};
164
165/* flag representing the memory block allocation status */
166#define OBJECT_ALLOCATED (1 << 0)
167/* flag set after the first reporting of an unreference object */
168#define OBJECT_REPORTED (1 << 1)
169/* flag set to not scan the object */
170#define OBJECT_NO_SCAN (1 << 2)
171
172/* number of bytes to print per line; must be 16 or 32 */
173#define HEX_ROW_SIZE 16
174/* number of bytes to print at a time (1, 2, 4, 8) */
175#define HEX_GROUP_SIZE 1
176/* include ASCII after the hex output */
177#define HEX_ASCII 1
178/* max number of lines to be printed */
179#define HEX_MAX_LINES 2
180
181/* the list of all allocated objects */
182static LIST_HEAD(object_list);
183/* the list of gray-colored objects (see color_gray comment below) */
184static LIST_HEAD(gray_list);
185/* search tree for object boundaries */
186static struct rb_root object_tree_root = RB_ROOT;
187/* rw_lock protecting the access to object_list and object_tree_root */
188static DEFINE_RWLOCK(kmemleak_lock);
189
190/* allocation caches for kmemleak internal data */
191static struct kmem_cache *object_cache;
192static struct kmem_cache *scan_area_cache;
193
194/* set if tracing memory operations is enabled */
195static int kmemleak_enabled;
196/* set in the late_initcall if there were no errors */
197static int kmemleak_initialized;
198/* enables or disables early logging of the memory operations */
199static int kmemleak_early_log = 1;
200/* set if a kmemleak warning was issued */
201static int kmemleak_warning;
202/* set if a fatal kmemleak error has occurred */
203static int kmemleak_error;
204
205/* minimum and maximum address that may be valid pointers */
206static unsigned long min_addr = ULONG_MAX;
207static unsigned long max_addr;
208
209static struct task_struct *scan_thread;
210/* used to avoid reporting of recently allocated objects */
211static unsigned long jiffies_min_age;
212static unsigned long jiffies_last_scan;
213/* delay between automatic memory scannings */
214static signed long jiffies_scan_wait;
215/* enables or disables the task stacks scanning */
216static int kmemleak_stack_scan = 1;
217/* protects the memory scanning, parameters and debug/kmemleak file access */
218static DEFINE_MUTEX(scan_mutex);
219/* setting kmemleak=on, will set this var, skipping the disable */
220static int kmemleak_skip_disable;
221/* If there are leaks that can be reported */
222static bool kmemleak_found_leaks;
223
224/*
225 * Early object allocation/freeing logging. Kmemleak is initialized after the
226 * kernel allocator. However, both the kernel allocator and kmemleak may
227 * allocate memory blocks which need to be tracked. Kmemleak defines an
228 * arbitrary buffer to hold the allocation/freeing information before it is
229 * fully initialized.
230 */
231
232/* kmemleak operation type for early logging */
233enum {
234    KMEMLEAK_ALLOC,
235    KMEMLEAK_ALLOC_PERCPU,
236    KMEMLEAK_FREE,
237    KMEMLEAK_FREE_PART,
238    KMEMLEAK_FREE_PERCPU,
239    KMEMLEAK_NOT_LEAK,
240    KMEMLEAK_IGNORE,
241    KMEMLEAK_SCAN_AREA,
242    KMEMLEAK_NO_SCAN
243};
244
245/*
246 * Structure holding the information passed to kmemleak callbacks during the
247 * early logging.
248 */
249struct early_log {
250    int op_type; /* kmemleak operation type */
251    const void *ptr; /* allocated/freed memory block */
252    size_t size; /* memory block size */
253    int min_count; /* minimum reference count */
254    unsigned long trace[MAX_TRACE]; /* stack trace */
255    unsigned int trace_len; /* stack trace length */
256};
257
258/* early logging buffer and current position */
259static struct early_log
260    early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
261static int crt_early_log __initdata;
262
263static void kmemleak_disable(void);
264
265/*
266 * Print a warning and dump the stack trace.
267 */
268#define kmemleak_warn(x...) do { \
269    pr_warning(x); \
270    dump_stack(); \
271    kmemleak_warning = 1; \
272} while (0)
273
274/*
275 * Macro invoked when a serious kmemleak condition occurred and cannot be
276 * recovered from. Kmemleak will be disabled and further allocation/freeing
277 * tracing no longer available.
278 */
279#define kmemleak_stop(x...) do { \
280    kmemleak_warn(x); \
281    kmemleak_disable(); \
282} while (0)
283
284/*
285 * Printing of the objects hex dump to the seq file. The number of lines to be
286 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
287 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
288 * with the object->lock held.
289 */
290static void hex_dump_object(struct seq_file *seq,
291                struct kmemleak_object *object)
292{
293    const u8 *ptr = (const u8 *)object->pointer;
294    int i, len, remaining;
295    unsigned char linebuf[HEX_ROW_SIZE * 5];
296
297    /* limit the number of lines to HEX_MAX_LINES */
298    remaining = len =
299        min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
300
301    seq_printf(seq, " hex dump (first %d bytes):\n", len);
302    for (i = 0; i < len; i += HEX_ROW_SIZE) {
303        int linelen = min(remaining, HEX_ROW_SIZE);
304
305        remaining -= HEX_ROW_SIZE;
306        hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
307                   HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
308                   HEX_ASCII);
309        seq_printf(seq, " %s\n", linebuf);
310    }
311}
312
313/*
314 * Object colors, encoded with count and min_count:
315 * - white - orphan object, not enough references to it (count < min_count)
316 * - gray - not orphan, not marked as false positive (min_count == 0) or
317 * sufficient references to it (count >= min_count)
318 * - black - ignore, it doesn't contain references (e.g. text section)
319 * (min_count == -1). No function defined for this color.
320 * Newly created objects don't have any color assigned (object->count == -1)
321 * before the next memory scan when they become white.
322 */
323static bool color_white(const struct kmemleak_object *object)
324{
325    return object->count != KMEMLEAK_BLACK &&
326        object->count < object->min_count;
327}
328
329static bool color_gray(const struct kmemleak_object *object)
330{
331    return object->min_count != KMEMLEAK_BLACK &&
332        object->count >= object->min_count;
333}
334
335/*
336 * Objects are considered unreferenced only if their color is white, they have
337 * not be deleted and have a minimum age to avoid false positives caused by
338 * pointers temporarily stored in CPU registers.
339 */
340static bool unreferenced_object(struct kmemleak_object *object)
341{
342    return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
343        time_before_eq(object->jiffies + jiffies_min_age,
344                   jiffies_last_scan);
345}
346
347/*
348 * Printing of the unreferenced objects information to the seq file. The
349 * print_unreferenced function must be called with the object->lock held.
350 */
351static void print_unreferenced(struct seq_file *seq,
352                   struct kmemleak_object *object)
353{
354    int i;
355    unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
356
357    seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
358           object->pointer, object->size);
359    seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
360           object->comm, object->pid, object->jiffies,
361           msecs_age / 1000, msecs_age % 1000);
362    hex_dump_object(seq, object);
363    seq_printf(seq, " backtrace:\n");
364
365    for (i = 0; i < object->trace_len; i++) {
366        void *ptr = (void *)object->trace[i];
367        seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
368    }
369}
370
371/*
372 * Print the kmemleak_object information. This function is used mainly for
373 * debugging special cases when kmemleak operations. It must be called with
374 * the object->lock held.
375 */
376static void dump_object_info(struct kmemleak_object *object)
377{
378    struct stack_trace trace;
379
380    trace.nr_entries = object->trace_len;
381    trace.entries = object->trace;
382
383    pr_notice("Object 0x%08lx (size %zu):\n",
384          object->pointer, object->size);
385    pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
386          object->comm, object->pid, object->jiffies);
387    pr_notice(" min_count = %d\n", object->min_count);
388    pr_notice(" count = %d\n", object->count);
389    pr_notice(" flags = 0x%lx\n", object->flags);
390    pr_notice(" checksum = %u\n", object->checksum);
391    pr_notice(" backtrace:\n");
392    print_stack_trace(&trace, 4);
393}
394
395/*
396 * Look-up a memory block metadata (kmemleak_object) in the object search
397 * tree based on a pointer value. If alias is 0, only values pointing to the
398 * beginning of the memory block are allowed. The kmemleak_lock must be held
399 * when calling this function.
400 */
401static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
402{
403    struct rb_node *rb = object_tree_root.rb_node;
404
405    while (rb) {
406        struct kmemleak_object *object =
407            rb_entry(rb, struct kmemleak_object, rb_node);
408        if (ptr < object->pointer)
409            rb = object->rb_node.rb_left;
410        else if (object->pointer + object->size <= ptr)
411            rb = object->rb_node.rb_right;
412        else if (object->pointer == ptr || alias)
413            return object;
414        else {
415            kmemleak_warn("Found object by alias at 0x%08lx\n",
416                      ptr);
417            dump_object_info(object);
418            break;
419        }
420    }
421    return NULL;
422}
423
424/*
425 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
426 * that once an object's use_count reached 0, the RCU freeing was already
427 * registered and the object should no longer be used. This function must be
428 * called under the protection of rcu_read_lock().
429 */
430static int get_object(struct kmemleak_object *object)
431{
432    return atomic_inc_not_zero(&object->use_count);
433}
434
435/*
436 * RCU callback to free a kmemleak_object.
437 */
438static void free_object_rcu(struct rcu_head *rcu)
439{
440    struct hlist_node *tmp;
441    struct kmemleak_scan_area *area;
442    struct kmemleak_object *object =
443        container_of(rcu, struct kmemleak_object, rcu);
444
445    /*
446     * Once use_count is 0 (guaranteed by put_object), there is no other
447     * code accessing this object, hence no need for locking.
448     */
449    hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
450        hlist_del(&area->node);
451        kmem_cache_free(scan_area_cache, area);
452    }
453    kmem_cache_free(object_cache, object);
454}
455
456/*
457 * Decrement the object use_count. Once the count is 0, free the object using
458 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
459 * delete_object() path, the delayed RCU freeing ensures that there is no
460 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
461 * is also possible.
462 */
463static void put_object(struct kmemleak_object *object)
464{
465    if (!atomic_dec_and_test(&object->use_count))
466        return;
467
468    /* should only get here after delete_object was called */
469    WARN_ON(object->flags & OBJECT_ALLOCATED);
470
471    call_rcu(&object->rcu, free_object_rcu);
472}
473
474/*
475 * Look up an object in the object search tree and increase its use_count.
476 */
477static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
478{
479    unsigned long flags;
480    struct kmemleak_object *object = NULL;
481
482    rcu_read_lock();
483    read_lock_irqsave(&kmemleak_lock, flags);
484    if (ptr >= min_addr && ptr < max_addr)
485        object = lookup_object(ptr, alias);
486    read_unlock_irqrestore(&kmemleak_lock, flags);
487
488    /* check whether the object is still available */
489    if (object && !get_object(object))
490        object = NULL;
491    rcu_read_unlock();
492
493    return object;
494}
495
496/*
497 * Save stack trace to the given array of MAX_TRACE size.
498 */
499static int __save_stack_trace(unsigned long *trace)
500{
501    struct stack_trace stack_trace;
502
503    stack_trace.max_entries = MAX_TRACE;
504    stack_trace.nr_entries = 0;
505    stack_trace.entries = trace;
506    stack_trace.skip = 2;
507    save_stack_trace(&stack_trace);
508
509    return stack_trace.nr_entries;
510}
511
512/*
513 * Create the metadata (struct kmemleak_object) corresponding to an allocated
514 * memory block and add it to the object_list and object_tree_root.
515 */
516static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
517                         int min_count, gfp_t gfp)
518{
519    unsigned long flags;
520    struct kmemleak_object *object, *parent;
521    struct rb_node **link, *rb_parent;
522
523    object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
524    if (!object) {
525        pr_warning("Cannot allocate a kmemleak_object structure\n");
526        kmemleak_disable();
527        return NULL;
528    }
529
530    INIT_LIST_HEAD(&object->object_list);
531    INIT_LIST_HEAD(&object->gray_list);
532    INIT_HLIST_HEAD(&object->area_list);
533    spin_lock_init(&object->lock);
534    atomic_set(&object->use_count, 1);
535    object->flags = OBJECT_ALLOCATED;
536    object->pointer = ptr;
537    object->size = size;
538    object->min_count = min_count;
539    object->count = 0; /* white color initially */
540    object->jiffies = jiffies;
541    object->checksum = 0;
542
543    /* task information */
544    if (in_irq()) {
545        object->pid = 0;
546        strncpy(object->comm, "hardirq", sizeof(object->comm));
547    } else if (in_softirq()) {
548        object->pid = 0;
549        strncpy(object->comm, "softirq", sizeof(object->comm));
550    } else {
551        object->pid = current->pid;
552        /*
553         * There is a small chance of a race with set_task_comm(),
554         * however using get_task_comm() here may cause locking
555         * dependency issues with current->alloc_lock. In the worst
556         * case, the command line is not correct.
557         */
558        strncpy(object->comm, current->comm, sizeof(object->comm));
559    }
560
561    /* kernel backtrace */
562    object->trace_len = __save_stack_trace(object->trace);
563
564    write_lock_irqsave(&kmemleak_lock, flags);
565
566    min_addr = min(min_addr, ptr);
567    max_addr = max(max_addr, ptr + size);
568    link = &object_tree_root.rb_node;
569    rb_parent = NULL;
570    while (*link) {
571        rb_parent = *link;
572        parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
573        if (ptr + size <= parent->pointer)
574            link = &parent->rb_node.rb_left;
575        else if (parent->pointer + parent->size <= ptr)
576            link = &parent->rb_node.rb_right;
577        else {
578            kmemleak_stop("Cannot insert 0x%lx into the object "
579                      "search tree (overlaps existing)\n",
580                      ptr);
581            kmem_cache_free(object_cache, object);
582            object = parent;
583            spin_lock(&object->lock);
584            dump_object_info(object);
585            spin_unlock(&object->lock);
586            goto out;
587        }
588    }
589    rb_link_node(&object->rb_node, rb_parent, link);
590    rb_insert_color(&object->rb_node, &object_tree_root);
591
592    list_add_tail_rcu(&object->object_list, &object_list);
593out:
594    write_unlock_irqrestore(&kmemleak_lock, flags);
595    return object;
596}
597
598/*
599 * Remove the metadata (struct kmemleak_object) for a memory block from the
600 * object_list and object_tree_root and decrement its use_count.
601 */
602static void __delete_object(struct kmemleak_object *object)
603{
604    unsigned long flags;
605
606    write_lock_irqsave(&kmemleak_lock, flags);
607    rb_erase(&object->rb_node, &object_tree_root);
608    list_del_rcu(&object->object_list);
609    write_unlock_irqrestore(&kmemleak_lock, flags);
610
611    WARN_ON(!(object->flags & OBJECT_ALLOCATED));
612    WARN_ON(atomic_read(&object->use_count) < 2);
613
614    /*
615     * Locking here also ensures that the corresponding memory block
616     * cannot be freed when it is being scanned.
617     */
618    spin_lock_irqsave(&object->lock, flags);
619    object->flags &= ~OBJECT_ALLOCATED;
620    spin_unlock_irqrestore(&object->lock, flags);
621    put_object(object);
622}
623
624/*
625 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
626 * delete it.
627 */
628static void delete_object_full(unsigned long ptr)
629{
630    struct kmemleak_object *object;
631
632    object = find_and_get_object(ptr, 0);
633    if (!object) {
634#ifdef DEBUG
635        kmemleak_warn("Freeing unknown object at 0x%08lx\n",
636                  ptr);
637#endif
638        return;
639    }
640    __delete_object(object);
641    put_object(object);
642}
643
644/*
645 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
646 * delete it. If the memory block is partially freed, the function may create
647 * additional metadata for the remaining parts of the block.
648 */
649static void delete_object_part(unsigned long ptr, size_t size)
650{
651    struct kmemleak_object *object;
652    unsigned long start, end;
653
654    object = find_and_get_object(ptr, 1);
655    if (!object) {
656#ifdef DEBUG
657        kmemleak_warn("Partially freeing unknown object at 0x%08lx "
658                  "(size %zu)\n", ptr, size);
659#endif
660        return;
661    }
662    __delete_object(object);
663
664    /*
665     * Create one or two objects that may result from the memory block
666     * split. Note that partial freeing is only done by free_bootmem() and
667     * this happens before kmemleak_init() is called. The path below is
668     * only executed during early log recording in kmemleak_init(), so
669     * GFP_KERNEL is enough.
670     */
671    start = object->pointer;
672    end = object->pointer + object->size;
673    if (ptr > start)
674        create_object(start, ptr - start, object->min_count,
675                  GFP_KERNEL);
676    if (ptr + size < end)
677        create_object(ptr + size, end - ptr - size, object->min_count,
678                  GFP_KERNEL);
679
680    put_object(object);
681}
682
683static void __paint_it(struct kmemleak_object *object, int color)
684{
685    object->min_count = color;
686    if (color == KMEMLEAK_BLACK)
687        object->flags |= OBJECT_NO_SCAN;
688}
689
690static void paint_it(struct kmemleak_object *object, int color)
691{
692    unsigned long flags;
693
694    spin_lock_irqsave(&object->lock, flags);
695    __paint_it(object, color);
696    spin_unlock_irqrestore(&object->lock, flags);
697}
698
699static void paint_ptr(unsigned long ptr, int color)
700{
701    struct kmemleak_object *object;
702
703    object = find_and_get_object(ptr, 0);
704    if (!object) {
705        kmemleak_warn("Trying to color unknown object "
706                  "at 0x%08lx as %s\n", ptr,
707                  (color == KMEMLEAK_GREY) ? "Grey" :
708                  (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
709        return;
710    }
711    paint_it(object, color);
712    put_object(object);
713}
714
715/*
716 * Mark an object permanently as gray-colored so that it can no longer be
717 * reported as a leak. This is used in general to mark a false positive.
718 */
719static void make_gray_object(unsigned long ptr)
720{
721    paint_ptr(ptr, KMEMLEAK_GREY);
722}
723
724/*
725 * Mark the object as black-colored so that it is ignored from scans and
726 * reporting.
727 */
728static void make_black_object(unsigned long ptr)
729{
730    paint_ptr(ptr, KMEMLEAK_BLACK);
731}
732
733/*
734 * Add a scanning area to the object. If at least one such area is added,
735 * kmemleak will only scan these ranges rather than the whole memory block.
736 */
737static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
738{
739    unsigned long flags;
740    struct kmemleak_object *object;
741    struct kmemleak_scan_area *area;
742
743    object = find_and_get_object(ptr, 1);
744    if (!object) {
745        kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
746                  ptr);
747        return;
748    }
749
750    area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
751    if (!area) {
752        pr_warning("Cannot allocate a scan area\n");
753        goto out;
754    }
755
756    spin_lock_irqsave(&object->lock, flags);
757    if (size == SIZE_MAX) {
758        size = object->pointer + object->size - ptr;
759    } else if (ptr + size > object->pointer + object->size) {
760        kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
761        dump_object_info(object);
762        kmem_cache_free(scan_area_cache, area);
763        goto out_unlock;
764    }
765
766    INIT_HLIST_NODE(&area->node);
767    area->start = ptr;
768    area->size = size;
769
770    hlist_add_head(&area->node, &object->area_list);
771out_unlock:
772    spin_unlock_irqrestore(&object->lock, flags);
773out:
774    put_object(object);
775}
776
777/*
778 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
779 * pointer. Such object will not be scanned by kmemleak but references to it
780 * are searched.
781 */
782static void object_no_scan(unsigned long ptr)
783{
784    unsigned long flags;
785    struct kmemleak_object *object;
786
787    object = find_and_get_object(ptr, 0);
788    if (!object) {
789        kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
790        return;
791    }
792
793    spin_lock_irqsave(&object->lock, flags);
794    object->flags |= OBJECT_NO_SCAN;
795    spin_unlock_irqrestore(&object->lock, flags);
796    put_object(object);
797}
798
799/*
800 * Log an early kmemleak_* call to the early_log buffer. These calls will be
801 * processed later once kmemleak is fully initialized.
802 */
803static void __init log_early(int op_type, const void *ptr, size_t size,
804                 int min_count)
805{
806    unsigned long flags;
807    struct early_log *log;
808
809    if (kmemleak_error) {
810        /* kmemleak stopped recording, just count the requests */
811        crt_early_log++;
812        return;
813    }
814
815    if (crt_early_log >= ARRAY_SIZE(early_log)) {
816        kmemleak_disable();
817        return;
818    }
819
820    /*
821     * There is no need for locking since the kernel is still in UP mode
822     * at this stage. Disabling the IRQs is enough.
823     */
824    local_irq_save(flags);
825    log = &early_log[crt_early_log];
826    log->op_type = op_type;
827    log->ptr = ptr;
828    log->size = size;
829    log->min_count = min_count;
830    log->trace_len = __save_stack_trace(log->trace);
831    crt_early_log++;
832    local_irq_restore(flags);
833}
834
835/*
836 * Log an early allocated block and populate the stack trace.
837 */
838static void early_alloc(struct early_log *log)
839{
840    struct kmemleak_object *object;
841    unsigned long flags;
842    int i;
843
844    if (!kmemleak_enabled || !log->ptr || IS_ERR(log->ptr))
845        return;
846
847    /*
848     * RCU locking needed to ensure object is not freed via put_object().
849     */
850    rcu_read_lock();
851    object = create_object((unsigned long)log->ptr, log->size,
852                   log->min_count, GFP_ATOMIC);
853    if (!object)
854        goto out;
855    spin_lock_irqsave(&object->lock, flags);
856    for (i = 0; i < log->trace_len; i++)
857        object->trace[i] = log->trace[i];
858    object->trace_len = log->trace_len;
859    spin_unlock_irqrestore(&object->lock, flags);
860out:
861    rcu_read_unlock();
862}
863
864/*
865 * Log an early allocated block and populate the stack trace.
866 */
867static void early_alloc_percpu(struct early_log *log)
868{
869    unsigned int cpu;
870    const void __percpu *ptr = log->ptr;
871
872    for_each_possible_cpu(cpu) {
873        log->ptr = per_cpu_ptr(ptr, cpu);
874        early_alloc(log);
875    }
876}
877
878/**
879 * kmemleak_alloc - register a newly allocated object
880 * @ptr: pointer to beginning of the object
881 * @size: size of the object
882 * @min_count: minimum number of references to this object. If during memory
883 * scanning a number of references less than @min_count is found,
884 * the object is reported as a memory leak. If @min_count is 0,
885 * the object is never reported as a leak. If @min_count is -1,
886 * the object is ignored (not scanned and not reported as a leak)
887 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
888 *
889 * This function is called from the kernel allocators when a new object
890 * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.).
891 */
892void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
893              gfp_t gfp)
894{
895    pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
896
897    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
898        create_object((unsigned long)ptr, size, min_count, gfp);
899    else if (kmemleak_early_log)
900        log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
901}
902EXPORT_SYMBOL_GPL(kmemleak_alloc);
903
904/**
905 * kmemleak_alloc_percpu - register a newly allocated __percpu object
906 * @ptr: __percpu pointer to beginning of the object
907 * @size: size of the object
908 *
909 * This function is called from the kernel percpu allocator when a new object
910 * (memory block) is allocated (alloc_percpu). It assumes GFP_KERNEL
911 * allocation.
912 */
913void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size)
914{
915    unsigned int cpu;
916
917    pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
918
919    /*
920     * Percpu allocations are only scanned and not reported as leaks
921     * (min_count is set to 0).
922     */
923    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
924        for_each_possible_cpu(cpu)
925            create_object((unsigned long)per_cpu_ptr(ptr, cpu),
926                      size, 0, GFP_KERNEL);
927    else if (kmemleak_early_log)
928        log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
929}
930EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
931
932/**
933 * kmemleak_free - unregister a previously registered object
934 * @ptr: pointer to beginning of the object
935 *
936 * This function is called from the kernel allocators when an object (memory
937 * block) is freed (kmem_cache_free, kfree, vfree etc.).
938 */
939void __ref kmemleak_free(const void *ptr)
940{
941    pr_debug("%s(0x%p)\n", __func__, ptr);
942
943    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
944        delete_object_full((unsigned long)ptr);
945    else if (kmemleak_early_log)
946        log_early(KMEMLEAK_FREE, ptr, 0, 0);
947}
948EXPORT_SYMBOL_GPL(kmemleak_free);
949
950/**
951 * kmemleak_free_part - partially unregister a previously registered object
952 * @ptr: pointer to the beginning or inside the object. This also
953 * represents the start of the range to be freed
954 * @size: size to be unregistered
955 *
956 * This function is called when only a part of a memory block is freed
957 * (usually from the bootmem allocator).
958 */
959void __ref kmemleak_free_part(const void *ptr, size_t size)
960{
961    pr_debug("%s(0x%p)\n", __func__, ptr);
962
963    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
964        delete_object_part((unsigned long)ptr, size);
965    else if (kmemleak_early_log)
966        log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
967}
968EXPORT_SYMBOL_GPL(kmemleak_free_part);
969
970/**
971 * kmemleak_free_percpu - unregister a previously registered __percpu object
972 * @ptr: __percpu pointer to beginning of the object
973 *
974 * This function is called from the kernel percpu allocator when an object
975 * (memory block) is freed (free_percpu).
976 */
977void __ref kmemleak_free_percpu(const void __percpu *ptr)
978{
979    unsigned int cpu;
980
981    pr_debug("%s(0x%p)\n", __func__, ptr);
982
983    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
984        for_each_possible_cpu(cpu)
985            delete_object_full((unsigned long)per_cpu_ptr(ptr,
986                                      cpu));
987    else if (kmemleak_early_log)
988        log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
989}
990EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
991
992/**
993 * kmemleak_update_trace - update object allocation stack trace
994 * @ptr: pointer to beginning of the object
995 *
996 * Override the object allocation stack trace for cases where the actual
997 * allocation place is not always useful.
998 */
999void __ref kmemleak_update_trace(const void *ptr)
1000{
1001    struct kmemleak_object *object;
1002    unsigned long flags;
1003
1004    pr_debug("%s(0x%p)\n", __func__, ptr);
1005
1006    if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
1007        return;
1008
1009    object = find_and_get_object((unsigned long)ptr, 1);
1010    if (!object) {
1011#ifdef DEBUG
1012        kmemleak_warn("Updating stack trace for unknown object at %p\n",
1013                  ptr);
1014#endif
1015        return;
1016    }
1017
1018    spin_lock_irqsave(&object->lock, flags);
1019    object->trace_len = __save_stack_trace(object->trace);
1020    spin_unlock_irqrestore(&object->lock, flags);
1021
1022    put_object(object);
1023}
1024EXPORT_SYMBOL(kmemleak_update_trace);
1025
1026/**
1027 * kmemleak_not_leak - mark an allocated object as false positive
1028 * @ptr: pointer to beginning of the object
1029 *
1030 * Calling this function on an object will cause the memory block to no longer
1031 * be reported as leak and always be scanned.
1032 */
1033void __ref kmemleak_not_leak(const void *ptr)
1034{
1035    pr_debug("%s(0x%p)\n", __func__, ptr);
1036
1037    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1038        make_gray_object((unsigned long)ptr);
1039    else if (kmemleak_early_log)
1040        log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
1041}
1042EXPORT_SYMBOL(kmemleak_not_leak);
1043
1044/**
1045 * kmemleak_ignore - ignore an allocated object
1046 * @ptr: pointer to beginning of the object
1047 *
1048 * Calling this function on an object will cause the memory block to be
1049 * ignored (not scanned and not reported as a leak). This is usually done when
1050 * it is known that the corresponding block is not a leak and does not contain
1051 * any references to other allocated memory blocks.
1052 */
1053void __ref kmemleak_ignore(const void *ptr)
1054{
1055    pr_debug("%s(0x%p)\n", __func__, ptr);
1056
1057    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1058        make_black_object((unsigned long)ptr);
1059    else if (kmemleak_early_log)
1060        log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
1061}
1062EXPORT_SYMBOL(kmemleak_ignore);
1063
1064/**
1065 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1066 * @ptr: pointer to beginning or inside the object. This also
1067 * represents the start of the scan area
1068 * @size: size of the scan area
1069 * @gfp: kmalloc() flags used for kmemleak internal memory allocations
1070 *
1071 * This function is used when it is known that only certain parts of an object
1072 * contain references to other objects. Kmemleak will only scan these areas
1073 * reducing the number false negatives.
1074 */
1075void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1076{
1077    pr_debug("%s(0x%p)\n", __func__, ptr);
1078
1079    if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
1080        add_scan_area((unsigned long)ptr, size, gfp);
1081    else if (kmemleak_early_log)
1082        log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
1083}
1084EXPORT_SYMBOL(kmemleak_scan_area);
1085
1086/**
1087 * kmemleak_no_scan - do not scan an allocated object
1088 * @ptr: pointer to beginning of the object
1089 *
1090 * This function notifies kmemleak not to scan the given memory block. Useful
1091 * in situations where it is known that the given object does not contain any
1092 * references to other objects. Kmemleak will not scan such objects reducing
1093 * the number of false negatives.
1094 */
1095void __ref kmemleak_no_scan(const void *ptr)
1096{
1097    pr_debug("%s(0x%p)\n", __func__, ptr);
1098
1099    if (kmemleak_enabled && ptr && !IS_ERR(ptr))
1100        object_no_scan((unsigned long)ptr);
1101    else if (kmemleak_early_log)
1102        log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
1103}
1104EXPORT_SYMBOL(kmemleak_no_scan);
1105
1106/*
1107 * Update an object's checksum and return true if it was modified.
1108 */
1109static bool update_checksum(struct kmemleak_object *object)
1110{
1111    u32 old_csum = object->checksum;
1112
1113    if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
1114        return false;
1115
1116    object->checksum = crc32(0, (void *)object->pointer, object->size);
1117    return object->checksum != old_csum;
1118}
1119
1120/*
1121 * Memory scanning is a long process and it needs to be interruptable. This
1122 * function checks whether such interrupt condition occurred.
1123 */
1124static int scan_should_stop(void)
1125{
1126    if (!kmemleak_enabled)
1127        return 1;
1128
1129    /*
1130     * This function may be called from either process or kthread context,
1131     * hence the need to check for both stop conditions.
1132     */
1133    if (current->mm)
1134        return signal_pending(current);
1135    else
1136        return kthread_should_stop();
1137
1138    return 0;
1139}
1140
1141/*
1142 * Scan a memory block (exclusive range) for valid pointers and add those
1143 * found to the gray list.
1144 */
1145static void scan_block(void *_start, void *_end,
1146               struct kmemleak_object *scanned, int allow_resched)
1147{
1148    unsigned long *ptr;
1149    unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1150    unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1151
1152    for (ptr = start; ptr < end; ptr++) {
1153        struct kmemleak_object *object;
1154        unsigned long flags;
1155        unsigned long pointer;
1156
1157        if (allow_resched)
1158            cond_resched();
1159        if (scan_should_stop())
1160            break;
1161
1162        /* don't scan uninitialized memory */
1163        if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
1164                          BYTES_PER_POINTER))
1165            continue;
1166
1167        pointer = *ptr;
1168
1169        object = find_and_get_object(pointer, 1);
1170        if (!object)
1171            continue;
1172        if (object == scanned) {
1173            /* self referenced, ignore */
1174            put_object(object);
1175            continue;
1176        }
1177
1178        /*
1179         * Avoid the lockdep recursive warning on object->lock being
1180         * previously acquired in scan_object(). These locks are
1181         * enclosed by scan_mutex.
1182         */
1183        spin_lock_irqsave_nested(&object->lock, flags,
1184                     SINGLE_DEPTH_NESTING);
1185        if (!color_white(object)) {
1186            /* non-orphan, ignored or new */
1187            spin_unlock_irqrestore(&object->lock, flags);
1188            put_object(object);
1189            continue;
1190        }
1191
1192        /*
1193         * Increase the object's reference count (number of pointers
1194         * to the memory block). If this count reaches the required
1195         * minimum, the object's color will become gray and it will be
1196         * added to the gray_list.
1197         */
1198        object->count++;
1199        if (color_gray(object)) {
1200            list_add_tail(&object->gray_list, &gray_list);
1201            spin_unlock_irqrestore(&object->lock, flags);
1202            continue;
1203        }
1204
1205        spin_unlock_irqrestore(&object->lock, flags);
1206        put_object(object);
1207    }
1208}
1209
1210/*
1211 * Scan a memory block corresponding to a kmemleak_object. A condition is
1212 * that object->use_count >= 1.
1213 */
1214static void scan_object(struct kmemleak_object *object)
1215{
1216    struct kmemleak_scan_area *area;
1217    unsigned long flags;
1218
1219    /*
1220     * Once the object->lock is acquired, the corresponding memory block
1221     * cannot be freed (the same lock is acquired in delete_object).
1222     */
1223    spin_lock_irqsave(&object->lock, flags);
1224    if (object->flags & OBJECT_NO_SCAN)
1225        goto out;
1226    if (!(object->flags & OBJECT_ALLOCATED))
1227        /* already freed object */
1228        goto out;
1229    if (hlist_empty(&object->area_list)) {
1230        void *start = (void *)object->pointer;
1231        void *end = (void *)(object->pointer + object->size);
1232
1233        while (start < end && (object->flags & OBJECT_ALLOCATED) &&
1234               !(object->flags & OBJECT_NO_SCAN)) {
1235            scan_block(start, min(start + MAX_SCAN_SIZE, end),
1236                   object, 0);
1237            start += MAX_SCAN_SIZE;
1238
1239            spin_unlock_irqrestore(&object->lock, flags);
1240            cond_resched();
1241            spin_lock_irqsave(&object->lock, flags);
1242        }
1243    } else
1244        hlist_for_each_entry(area, &object->area_list, node)
1245            scan_block((void *)area->start,
1246                   (void *)(area->start + area->size),
1247                   object, 0);
1248out:
1249    spin_unlock_irqrestore(&object->lock, flags);
1250}
1251
1252/*
1253 * Scan the objects already referenced (gray objects). More objects will be
1254 * referenced and, if there are no memory leaks, all the objects are scanned.
1255 */
1256static void scan_gray_list(void)
1257{
1258    struct kmemleak_object *object, *tmp;
1259
1260    /*
1261     * The list traversal is safe for both tail additions and removals
1262     * from inside the loop. The kmemleak objects cannot be freed from
1263     * outside the loop because their use_count was incremented.
1264     */
1265    object = list_entry(gray_list.next, typeof(*object), gray_list);
1266    while (&object->gray_list != &gray_list) {
1267        cond_resched();
1268
1269        /* may add new objects to the list */
1270        if (!scan_should_stop())
1271            scan_object(object);
1272
1273        tmp = list_entry(object->gray_list.next, typeof(*object),
1274                 gray_list);
1275
1276        /* remove the object from the list and release it */
1277        list_del(&object->gray_list);
1278        put_object(object);
1279
1280        object = tmp;
1281    }
1282    WARN_ON(!list_empty(&gray_list));
1283}
1284
1285/*
1286 * Scan data sections and all the referenced memory blocks allocated via the
1287 * kernel's standard allocators. This function must be called with the
1288 * scan_mutex held.
1289 */
1290static void kmemleak_scan(void)
1291{
1292    unsigned long flags;
1293    struct kmemleak_object *object;
1294    int i;
1295    int new_leaks = 0;
1296
1297    jiffies_last_scan = jiffies;
1298
1299    /* prepare the kmemleak_object's */
1300    rcu_read_lock();
1301    list_for_each_entry_rcu(object, &object_list, object_list) {
1302        spin_lock_irqsave(&object->lock, flags);
1303#ifdef DEBUG
1304        /*
1305         * With a few exceptions there should be a maximum of
1306         * 1 reference to any object at this point.
1307         */
1308        if (atomic_read(&object->use_count) > 1) {
1309            pr_debug("object->use_count = %d\n",
1310                 atomic_read(&object->use_count));
1311            dump_object_info(object);
1312        }
1313#endif
1314        /* reset the reference count (whiten the object) */
1315        object->count = 0;
1316        if (color_gray(object) && get_object(object))
1317            list_add_tail(&object->gray_list, &gray_list);
1318
1319        spin_unlock_irqrestore(&object->lock, flags);
1320    }
1321    rcu_read_unlock();
1322
1323    /* data/bss scanning */
1324    scan_block(_sdata, _edata, NULL, 1);
1325    scan_block(__bss_start, __bss_stop, NULL, 1);
1326
1327#ifdef CONFIG_SMP
1328    /* per-cpu sections scanning */
1329    for_each_possible_cpu(i)
1330        scan_block(__per_cpu_start + per_cpu_offset(i),
1331               __per_cpu_end + per_cpu_offset(i), NULL, 1);
1332#endif
1333
1334    /*
1335     * Struct page scanning for each node.
1336     */
1337    get_online_mems();
1338    for_each_online_node(i) {
1339        unsigned long start_pfn = node_start_pfn(i);
1340        unsigned long end_pfn = node_end_pfn(i);
1341        unsigned long pfn;
1342
1343        for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1344            struct page *page;
1345
1346            if (!pfn_valid(pfn))
1347                continue;
1348            page = pfn_to_page(pfn);
1349            /* only scan if page is in use */
1350            if (page_count(page) == 0)
1351                continue;
1352            scan_block(page, page + 1, NULL, 1);
1353        }
1354    }
1355    put_online_mems();
1356
1357    /*
1358     * Scanning the task stacks (may introduce false negatives).
1359     */
1360    if (kmemleak_stack_scan) {
1361        struct task_struct *p, *g;
1362
1363        read_lock(&tasklist_lock);
1364        do_each_thread(g, p) {
1365            scan_block(task_stack_page(p), task_stack_page(p) +
1366                   THREAD_SIZE, NULL, 0);
1367        } while_each_thread(g, p);
1368        read_unlock(&tasklist_lock);
1369    }
1370
1371    /*
1372     * Scan the objects already referenced from the sections scanned
1373     * above.
1374     */
1375    scan_gray_list();
1376
1377    /*
1378     * Check for new or unreferenced objects modified since the previous
1379     * scan and color them gray until the next scan.
1380     */
1381    rcu_read_lock();
1382    list_for_each_entry_rcu(object, &object_list, object_list) {
1383        spin_lock_irqsave(&object->lock, flags);
1384        if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1385            && update_checksum(object) && get_object(object)) {
1386            /* color it gray temporarily */
1387            object->count = object->min_count;
1388            list_add_tail(&object->gray_list, &gray_list);
1389        }
1390        spin_unlock_irqrestore(&object->lock, flags);
1391    }
1392    rcu_read_unlock();
1393
1394    /*
1395     * Re-scan the gray list for modified unreferenced objects.
1396     */
1397    scan_gray_list();
1398
1399    /*
1400     * If scanning was stopped do not report any new unreferenced objects.
1401     */
1402    if (scan_should_stop())
1403        return;
1404
1405    /*
1406     * Scanning result reporting.
1407     */
1408    rcu_read_lock();
1409    list_for_each_entry_rcu(object, &object_list, object_list) {
1410        spin_lock_irqsave(&object->lock, flags);
1411        if (unreferenced_object(object) &&
1412            !(object->flags & OBJECT_REPORTED)) {
1413            object->flags |= OBJECT_REPORTED;
1414            new_leaks++;
1415        }
1416        spin_unlock_irqrestore(&object->lock, flags);
1417    }
1418    rcu_read_unlock();
1419
1420    if (new_leaks) {
1421        kmemleak_found_leaks = true;
1422
1423        pr_info("%d new suspected memory leaks (see "
1424            "/sys/kernel/debug/kmemleak)\n", new_leaks);
1425    }
1426
1427}
1428
1429/*
1430 * Thread function performing automatic memory scanning. Unreferenced objects
1431 * at the end of a memory scan are reported but only the first time.
1432 */
1433static int kmemleak_scan_thread(void *arg)
1434{
1435    static int first_run = 1;
1436
1437    pr_info("Automatic memory scanning thread started\n");
1438    set_user_nice(current, 10);
1439
1440    /*
1441     * Wait before the first scan to allow the system to fully initialize.
1442     */
1443    if (first_run) {
1444        first_run = 0;
1445        ssleep(SECS_FIRST_SCAN);
1446    }
1447
1448    while (!kthread_should_stop()) {
1449        signed long timeout = jiffies_scan_wait;
1450
1451        mutex_lock(&scan_mutex);
1452        kmemleak_scan();
1453        mutex_unlock(&scan_mutex);
1454
1455        /* wait before the next scan */
1456        while (timeout && !kthread_should_stop())
1457            timeout = schedule_timeout_interruptible(timeout);
1458    }
1459
1460    pr_info("Automatic memory scanning thread ended\n");
1461
1462    return 0;
1463}
1464
1465/*
1466 * Start the automatic memory scanning thread. This function must be called
1467 * with the scan_mutex held.
1468 */
1469static void start_scan_thread(void)
1470{
1471    if (scan_thread)
1472        return;
1473    scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1474    if (IS_ERR(scan_thread)) {
1475        pr_warning("Failed to create the scan thread\n");
1476        scan_thread = NULL;
1477    }
1478}
1479
1480/*
1481 * Stop the automatic memory scanning thread. This function must be called
1482 * with the scan_mutex held.
1483 */
1484static void stop_scan_thread(void)
1485{
1486    if (scan_thread) {
1487        kthread_stop(scan_thread);
1488        scan_thread = NULL;
1489    }
1490}
1491
1492/*
1493 * Iterate over the object_list and return the first valid object at or after
1494 * the required position with its use_count incremented. The function triggers
1495 * a memory scanning when the pos argument points to the first position.
1496 */
1497static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1498{
1499    struct kmemleak_object *object;
1500    loff_t n = *pos;
1501    int err;
1502
1503    err = mutex_lock_interruptible(&scan_mutex);
1504    if (err < 0)
1505        return ERR_PTR(err);
1506
1507    rcu_read_lock();
1508    list_for_each_entry_rcu(object, &object_list, object_list) {
1509        if (n-- > 0)
1510            continue;
1511        if (get_object(object))
1512            goto out;
1513    }
1514    object = NULL;
1515out:
1516    return object;
1517}
1518
1519/*
1520 * Return the next object in the object_list. The function decrements the
1521 * use_count of the previous object and increases that of the next one.
1522 */
1523static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1524{
1525    struct kmemleak_object *prev_obj = v;
1526    struct kmemleak_object *next_obj = NULL;
1527    struct kmemleak_object *obj = prev_obj;
1528
1529    ++(*pos);
1530
1531    list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
1532        if (get_object(obj)) {
1533            next_obj = obj;
1534            break;
1535        }
1536    }
1537
1538    put_object(prev_obj);
1539    return next_obj;
1540}
1541
1542/*
1543 * Decrement the use_count of the last object required, if any.
1544 */
1545static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1546{
1547    if (!IS_ERR(v)) {
1548        /*
1549         * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1550         * waiting was interrupted, so only release it if !IS_ERR.
1551         */
1552        rcu_read_unlock();
1553        mutex_unlock(&scan_mutex);
1554        if (v)
1555            put_object(v);
1556    }
1557}
1558
1559/*
1560 * Print the information for an unreferenced object to the seq file.
1561 */
1562static int kmemleak_seq_show(struct seq_file *seq, void *v)
1563{
1564    struct kmemleak_object *object = v;
1565    unsigned long flags;
1566
1567    spin_lock_irqsave(&object->lock, flags);
1568    if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1569        print_unreferenced(seq, object);
1570    spin_unlock_irqrestore(&object->lock, flags);
1571    return 0;
1572}
1573
1574static const struct seq_operations kmemleak_seq_ops = {
1575    .start = kmemleak_seq_start,
1576    .next = kmemleak_seq_next,
1577    .stop = kmemleak_seq_stop,
1578    .show = kmemleak_seq_show,
1579};
1580
1581static int kmemleak_open(struct inode *inode, struct file *file)
1582{
1583    return seq_open(file, &kmemleak_seq_ops);
1584}
1585
1586static int dump_str_object_info(const char *str)
1587{
1588    unsigned long flags;
1589    struct kmemleak_object *object;
1590    unsigned long addr;
1591
1592    if (kstrtoul(str, 0, &addr))
1593        return -EINVAL;
1594    object = find_and_get_object(addr, 0);
1595    if (!object) {
1596        pr_info("Unknown object at 0x%08lx\n", addr);
1597        return -EINVAL;
1598    }
1599
1600    spin_lock_irqsave(&object->lock, flags);
1601    dump_object_info(object);
1602    spin_unlock_irqrestore(&object->lock, flags);
1603
1604    put_object(object);
1605    return 0;
1606}
1607
1608/*
1609 * We use grey instead of black to ensure we can do future scans on the same
1610 * objects. If we did not do future scans these black objects could
1611 * potentially contain references to newly allocated objects in the future and
1612 * we'd end up with false positives.
1613 */
1614static void kmemleak_clear(void)
1615{
1616    struct kmemleak_object *object;
1617    unsigned long flags;
1618
1619    rcu_read_lock();
1620    list_for_each_entry_rcu(object, &object_list, object_list) {
1621        spin_lock_irqsave(&object->lock, flags);
1622        if ((object->flags & OBJECT_REPORTED) &&
1623            unreferenced_object(object))
1624            __paint_it(object, KMEMLEAK_GREY);
1625        spin_unlock_irqrestore(&object->lock, flags);
1626    }
1627    rcu_read_unlock();
1628
1629    kmemleak_found_leaks = false;
1630}
1631
1632static void __kmemleak_do_cleanup(void);
1633
1634/*
1635 * File write operation to configure kmemleak at run-time. The following
1636 * commands can be written to the /sys/kernel/debug/kmemleak file:
1637 * off - disable kmemleak (irreversible)
1638 * stack=on - enable the task stacks scanning
1639 * stack=off - disable the tasks stacks scanning
1640 * scan=on - start the automatic memory scanning thread
1641 * scan=off - stop the automatic memory scanning thread
1642 * scan=... - set the automatic memory scanning period in seconds (0 to
1643 * disable it)
1644 * scan - trigger a memory scan
1645 * clear - mark all current reported unreferenced kmemleak objects as
1646 * grey to ignore printing them, or free all kmemleak objects
1647 * if kmemleak has been disabled.
1648 * dump=... - dump information about the object found at the given address
1649 */
1650static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1651                  size_t size, loff_t *ppos)
1652{
1653    char buf[64];
1654    int buf_size;
1655    int ret;
1656
1657    buf_size = min(size, (sizeof(buf) - 1));
1658    if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1659        return -EFAULT;
1660    buf[buf_size] = 0;
1661
1662    ret = mutex_lock_interruptible(&scan_mutex);
1663    if (ret < 0)
1664        return ret;
1665
1666    if (strncmp(buf, "clear", 5) == 0) {
1667        if (kmemleak_enabled)
1668            kmemleak_clear();
1669        else
1670            __kmemleak_do_cleanup();
1671        goto out;
1672    }
1673
1674    if (!kmemleak_enabled) {
1675        ret = -EBUSY;
1676        goto out;
1677    }
1678
1679    if (strncmp(buf, "off", 3) == 0)
1680        kmemleak_disable();
1681    else if (strncmp(buf, "stack=on", 8) == 0)
1682        kmemleak_stack_scan = 1;
1683    else if (strncmp(buf, "stack=off", 9) == 0)
1684        kmemleak_stack_scan = 0;
1685    else if (strncmp(buf, "scan=on", 7) == 0)
1686        start_scan_thread();
1687    else if (strncmp(buf, "scan=off", 8) == 0)
1688        stop_scan_thread();
1689    else if (strncmp(buf, "scan=", 5) == 0) {
1690        unsigned long secs;
1691
1692        ret = kstrtoul(buf + 5, 0, &secs);
1693        if (ret < 0)
1694            goto out;
1695        stop_scan_thread();
1696        if (secs) {
1697            jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1698            start_scan_thread();
1699        }
1700    } else if (strncmp(buf, "scan", 4) == 0)
1701        kmemleak_scan();
1702    else if (strncmp(buf, "dump=", 5) == 0)
1703        ret = dump_str_object_info(buf + 5);
1704    else
1705        ret = -EINVAL;
1706
1707out:
1708    mutex_unlock(&scan_mutex);
1709    if (ret < 0)
1710        return ret;
1711
1712    /* ignore the rest of the buffer, only one command at a time */
1713    *ppos += size;
1714    return size;
1715}
1716
1717static const struct file_operations kmemleak_fops = {
1718    .owner = THIS_MODULE,
1719    .open = kmemleak_open,
1720    .read = seq_read,
1721    .write = kmemleak_write,
1722    .llseek = seq_lseek,
1723    .release = seq_release,
1724};
1725
1726static void __kmemleak_do_cleanup(void)
1727{
1728    struct kmemleak_object *object;
1729
1730    rcu_read_lock();
1731    list_for_each_entry_rcu(object, &object_list, object_list)
1732        delete_object_full(object->pointer);
1733    rcu_read_unlock();
1734}
1735
1736/*
1737 * Stop the memory scanning thread and free the kmemleak internal objects if
1738 * no previous scan thread (otherwise, kmemleak may still have some useful
1739 * information on memory leaks).
1740 */
1741static void kmemleak_do_cleanup(struct work_struct *work)
1742{
1743    mutex_lock(&scan_mutex);
1744    stop_scan_thread();
1745
1746    if (!kmemleak_found_leaks)
1747        __kmemleak_do_cleanup();
1748    else
1749        pr_info("Kmemleak disabled without freeing internal data. "
1750            "Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\"\n");
1751    mutex_unlock(&scan_mutex);
1752}
1753
1754static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1755
1756/*
1757 * Disable kmemleak. No memory allocation/freeing will be traced once this
1758 * function is called. Disabling kmemleak is an irreversible operation.
1759 */
1760static void kmemleak_disable(void)
1761{
1762    /* atomically check whether it was already invoked */
1763    if (cmpxchg(&kmemleak_error, 0, 1))
1764        return;
1765
1766    /* stop any memory operation tracing */
1767    kmemleak_enabled = 0;
1768
1769    /* check whether it is too early for a kernel thread */
1770    if (kmemleak_initialized)
1771        schedule_work(&cleanup_work);
1772
1773    pr_info("Kernel memory leak detector disabled\n");
1774}
1775
1776/*
1777 * Allow boot-time kmemleak disabling (enabled by default).
1778 */
1779static int kmemleak_boot_config(char *str)
1780{
1781    if (!str)
1782        return -EINVAL;
1783    if (strcmp(str, "off") == 0)
1784        kmemleak_disable();
1785    else if (strcmp(str, "on") == 0)
1786        kmemleak_skip_disable = 1;
1787    else
1788        return -EINVAL;
1789    return 0;
1790}
1791early_param("kmemleak", kmemleak_boot_config);
1792
1793static void __init print_log_trace(struct early_log *log)
1794{
1795    struct stack_trace trace;
1796
1797    trace.nr_entries = log->trace_len;
1798    trace.entries = log->trace;
1799
1800    pr_notice("Early log backtrace:\n");
1801    print_stack_trace(&trace, 2);
1802}
1803
1804/*
1805 * Kmemleak initialization.
1806 */
1807void __init kmemleak_init(void)
1808{
1809    int i;
1810    unsigned long flags;
1811
1812#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1813    if (!kmemleak_skip_disable) {
1814        kmemleak_early_log = 0;
1815        kmemleak_disable();
1816        return;
1817    }
1818#endif
1819
1820    jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1821    jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1822
1823    object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1824    scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1825
1826    if (crt_early_log >= ARRAY_SIZE(early_log))
1827        pr_warning("Early log buffer exceeded (%d), please increase "
1828               "DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n", crt_early_log);
1829
1830    /* the kernel is still in UP mode, so disabling the IRQs is enough */
1831    local_irq_save(flags);
1832    kmemleak_early_log = 0;
1833    if (kmemleak_error) {
1834        local_irq_restore(flags);
1835        return;
1836    } else
1837        kmemleak_enabled = 1;
1838    local_irq_restore(flags);
1839
1840    /*
1841     * This is the point where tracking allocations is safe. Automatic
1842     * scanning is started during the late initcall. Add the early logged
1843     * callbacks to the kmemleak infrastructure.
1844     */
1845    for (i = 0; i < crt_early_log; i++) {
1846        struct early_log *log = &early_log[i];
1847
1848        switch (log->op_type) {
1849        case KMEMLEAK_ALLOC:
1850            early_alloc(log);
1851            break;
1852        case KMEMLEAK_ALLOC_PERCPU:
1853            early_alloc_percpu(log);
1854            break;
1855        case KMEMLEAK_FREE:
1856            kmemleak_free(log->ptr);
1857            break;
1858        case KMEMLEAK_FREE_PART:
1859            kmemleak_free_part(log->ptr, log->size);
1860            break;
1861        case KMEMLEAK_FREE_PERCPU:
1862            kmemleak_free_percpu(log->ptr);
1863            break;
1864        case KMEMLEAK_NOT_LEAK:
1865            kmemleak_not_leak(log->ptr);
1866            break;
1867        case KMEMLEAK_IGNORE:
1868            kmemleak_ignore(log->ptr);
1869            break;
1870        case KMEMLEAK_SCAN_AREA:
1871            kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
1872            break;
1873        case KMEMLEAK_NO_SCAN:
1874            kmemleak_no_scan(log->ptr);
1875            break;
1876        default:
1877            kmemleak_warn("Unknown early log operation: %d\n",
1878                      log->op_type);
1879        }
1880
1881        if (kmemleak_warning) {
1882            print_log_trace(log);
1883            kmemleak_warning = 0;
1884        }
1885    }
1886}
1887
1888/*
1889 * Late initialization function.
1890 */
1891static int __init kmemleak_late_init(void)
1892{
1893    struct dentry *dentry;
1894
1895    kmemleak_initialized = 1;
1896
1897    if (kmemleak_error) {
1898        /*
1899         * Some error occurred and kmemleak was disabled. There is a
1900         * small chance that kmemleak_disable() was called immediately
1901         * after setting kmemleak_initialized and we may end up with
1902         * two clean-up threads but serialized by scan_mutex.
1903         */
1904        schedule_work(&cleanup_work);
1905        return -ENOMEM;
1906    }
1907
1908    dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1909                     &kmemleak_fops);
1910    if (!dentry)
1911        pr_warning("Failed to create the debugfs kmemleak file\n");
1912    mutex_lock(&scan_mutex);
1913    start_scan_thread();
1914    mutex_unlock(&scan_mutex);
1915
1916    pr_info("Kernel memory leak detector initialized\n");
1917
1918    return 0;
1919}
1920late_initcall(kmemleak_late_init);
1921

Archive Download this file



interactive