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

Archive Download this file



interactive