1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * mm/kmemleak.c |
4 | * |
5 | * Copyright (C) 2008 ARM Limited |
6 | * Written by Catalin Marinas <catalin.marinas@arm.com> |
7 | * |
8 | * For more information on the algorithm and kmemleak usage, please see |
9 | * Documentation/dev-tools/kmemleak.rst. |
10 | * |
11 | * Notes on locking |
12 | * ---------------- |
13 | * |
14 | * The following locks and mutexes are used by kmemleak: |
15 | * |
16 | * - kmemleak_lock (raw_spinlock_t): protects the object_list as well as |
17 | * del_state modifications and accesses to the object_tree_root (or |
18 | * object_phys_tree_root). The object_list is the main list holding the |
19 | * metadata (struct kmemleak_object) for the allocated memory blocks. |
20 | * The object_tree_root and object_phys_tree_root are red |
21 | * black trees used to look-up metadata based on a pointer to the |
22 | * corresponding memory block. The object_phys_tree_root is for objects |
23 | * allocated with physical address. The kmemleak_object structures are |
24 | * added to the object_list and object_tree_root (or object_phys_tree_root) |
25 | * in the create_object() function called from the kmemleak_alloc() (or |
26 | * kmemleak_alloc_phys()) callback and removed in delete_object() called from |
27 | * the kmemleak_free() callback |
28 | * - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object. |
29 | * Accesses to the metadata (e.g. count) are protected by this lock. Note |
30 | * that some members of this structure may be protected by other means |
31 | * (atomic or kmemleak_lock). This lock is also held when scanning the |
32 | * corresponding memory block to avoid the kernel freeing it via the |
33 | * kmemleak_free() callback. This is less heavyweight than holding a global |
34 | * lock like kmemleak_lock during scanning. |
35 | * - scan_mutex (mutex): ensures that only one thread may scan the memory for |
36 | * unreferenced objects at a time. The gray_list contains the objects which |
37 | * are already referenced or marked as false positives and need to be |
38 | * scanned. This list is only modified during a scanning episode when the |
39 | * scan_mutex is held. At the end of a scan, the gray_list is always empty. |
40 | * Note that the kmemleak_object.use_count is incremented when an object is |
41 | * added to the gray_list and therefore cannot be freed. This mutex also |
42 | * prevents multiple users of the "kmemleak" debugfs file together with |
43 | * modifications to the memory scanning parameters including the scan_thread |
44 | * pointer |
45 | * |
46 | * Locks and mutexes are acquired/nested in the following order: |
47 | * |
48 | * scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING) |
49 | * |
50 | * No kmemleak_lock and object->lock nesting is allowed outside scan_mutex |
51 | * regions. |
52 | * |
53 | * The kmemleak_object structures have a use_count incremented or decremented |
54 | * using the get_object()/put_object() functions. When the use_count becomes |
55 | * 0, this count can no longer be incremented and put_object() schedules the |
56 | * kmemleak_object freeing via an RCU callback. All calls to the get_object() |
57 | * function must be protected by rcu_read_lock() to avoid accessing a freed |
58 | * structure. |
59 | */ |
60 | |
61 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
62 | |
63 | #include <linux/init.h> |
64 | #include <linux/kernel.h> |
65 | #include <linux/list.h> |
66 | #include <linux/sched/signal.h> |
67 | #include <linux/sched/task.h> |
68 | #include <linux/sched/task_stack.h> |
69 | #include <linux/jiffies.h> |
70 | #include <linux/delay.h> |
71 | #include <linux/export.h> |
72 | #include <linux/kthread.h> |
73 | #include <linux/rbtree.h> |
74 | #include <linux/fs.h> |
75 | #include <linux/debugfs.h> |
76 | #include <linux/seq_file.h> |
77 | #include <linux/cpumask.h> |
78 | #include <linux/spinlock.h> |
79 | #include <linux/module.h> |
80 | #include <linux/mutex.h> |
81 | #include <linux/rcupdate.h> |
82 | #include <linux/stacktrace.h> |
83 | #include <linux/stackdepot.h> |
84 | #include <linux/cache.h> |
85 | #include <linux/percpu.h> |
86 | #include <linux/memblock.h> |
87 | #include <linux/pfn.h> |
88 | #include <linux/mmzone.h> |
89 | #include <linux/slab.h> |
90 | #include <linux/thread_info.h> |
91 | #include <linux/err.h> |
92 | #include <linux/uaccess.h> |
93 | #include <linux/string.h> |
94 | #include <linux/nodemask.h> |
95 | #include <linux/mm.h> |
96 | #include <linux/workqueue.h> |
97 | #include <linux/crc32.h> |
98 | |
99 | #include <asm/sections.h> |
100 | #include <asm/processor.h> |
101 | #include <linux/atomic.h> |
102 | |
103 | #include <linux/kasan.h> |
104 | #include <linux/kfence.h> |
105 | #include <linux/kmemleak.h> |
106 | #include <linux/memory_hotplug.h> |
107 | |
108 | /* |
109 | * Kmemleak configuration and common defines. |
110 | */ |
111 | #define MAX_TRACE 16 /* stack trace length */ |
112 | #define MSECS_MIN_AGE 5000 /* minimum object age for reporting */ |
113 | #define SECS_FIRST_SCAN 60 /* delay before the first scan */ |
114 | #define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */ |
115 | #define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */ |
116 | |
117 | #define BYTES_PER_POINTER sizeof(void *) |
118 | |
119 | /* GFP bitmask for kmemleak internal allocations */ |
120 | #define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC | \ |
121 | __GFP_NOLOCKDEP)) | \ |
122 | __GFP_NORETRY | __GFP_NOMEMALLOC | \ |
123 | __GFP_NOWARN) |
124 | |
125 | /* scanning area inside a memory block */ |
126 | struct kmemleak_scan_area { |
127 | struct hlist_node node; |
128 | unsigned long start; |
129 | size_t size; |
130 | }; |
131 | |
132 | #define KMEMLEAK_GREY 0 |
133 | #define KMEMLEAK_BLACK -1 |
134 | |
135 | /* |
136 | * Structure holding the metadata for each allocated memory block. |
137 | * Modifications to such objects should be made while holding the |
138 | * object->lock. Insertions or deletions from object_list, gray_list or |
139 | * rb_node are already protected by the corresponding locks or mutex (see |
140 | * the notes on locking above). These objects are reference-counted |
141 | * (use_count) and freed using the RCU mechanism. |
142 | */ |
143 | struct kmemleak_object { |
144 | raw_spinlock_t lock; |
145 | unsigned int flags; /* object status flags */ |
146 | struct list_head object_list; |
147 | struct list_head gray_list; |
148 | struct rb_node rb_node; |
149 | struct rcu_head rcu; /* object_list lockless traversal */ |
150 | /* object usage count; object freed when use_count == 0 */ |
151 | atomic_t use_count; |
152 | unsigned int del_state; /* deletion state */ |
153 | unsigned long pointer; |
154 | size_t size; |
155 | /* pass surplus references to this pointer */ |
156 | unsigned long excess_ref; |
157 | /* minimum number of a pointers found before it is considered leak */ |
158 | int min_count; |
159 | /* the total number of pointers found pointing to this object */ |
160 | int count; |
161 | /* checksum for detecting modified objects */ |
162 | u32 checksum; |
163 | /* memory ranges to be scanned inside an object (empty for all) */ |
164 | struct hlist_head area_list; |
165 | depot_stack_handle_t trace_handle; |
166 | unsigned long jiffies; /* creation timestamp */ |
167 | pid_t pid; /* pid of the current task */ |
168 | char comm[TASK_COMM_LEN]; /* executable name */ |
169 | }; |
170 | |
171 | /* flag representing the memory block allocation status */ |
172 | #define OBJECT_ALLOCATED (1 << 0) |
173 | /* flag set after the first reporting of an unreference object */ |
174 | #define OBJECT_REPORTED (1 << 1) |
175 | /* flag set to not scan the object */ |
176 | #define OBJECT_NO_SCAN (1 << 2) |
177 | /* flag set to fully scan the object when scan_area allocation failed */ |
178 | #define OBJECT_FULL_SCAN (1 << 3) |
179 | /* flag set for object allocated with physical address */ |
180 | #define OBJECT_PHYS (1 << 4) |
181 | |
182 | /* set when __remove_object() called */ |
183 | #define DELSTATE_REMOVED (1 << 0) |
184 | /* set to temporarily prevent deletion from object_list */ |
185 | #define DELSTATE_NO_DELETE (1 << 1) |
186 | |
187 | #define HEX_PREFIX " " |
188 | /* number of bytes to print per line; must be 16 or 32 */ |
189 | #define HEX_ROW_SIZE 16 |
190 | /* number of bytes to print at a time (1, 2, 4, 8) */ |
191 | #define HEX_GROUP_SIZE 1 |
192 | /* include ASCII after the hex output */ |
193 | #define HEX_ASCII 1 |
194 | /* max number of lines to be printed */ |
195 | #define HEX_MAX_LINES 2 |
196 | |
197 | /* the list of all allocated objects */ |
198 | static LIST_HEAD(object_list); |
199 | /* the list of gray-colored objects (see color_gray comment below) */ |
200 | static LIST_HEAD(gray_list); |
201 | /* memory pool allocation */ |
202 | static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE]; |
203 | static int mem_pool_free_count = ARRAY_SIZE(mem_pool); |
204 | static LIST_HEAD(mem_pool_free_list); |
205 | /* search tree for object boundaries */ |
206 | static struct rb_root object_tree_root = RB_ROOT; |
207 | /* search tree for object (with OBJECT_PHYS flag) boundaries */ |
208 | static struct rb_root object_phys_tree_root = RB_ROOT; |
209 | /* protecting the access to object_list, object_tree_root (or object_phys_tree_root) */ |
210 | static DEFINE_RAW_SPINLOCK(kmemleak_lock); |
211 | |
212 | /* allocation caches for kmemleak internal data */ |
213 | static struct kmem_cache *object_cache; |
214 | static struct kmem_cache *scan_area_cache; |
215 | |
216 | /* set if tracing memory operations is enabled */ |
217 | static int kmemleak_enabled = 1; |
218 | /* same as above but only for the kmemleak_free() callback */ |
219 | static int kmemleak_free_enabled = 1; |
220 | /* set in the late_initcall if there were no errors */ |
221 | static int kmemleak_late_initialized; |
222 | /* set if a kmemleak warning was issued */ |
223 | static int kmemleak_warning; |
224 | /* set if a fatal kmemleak error has occurred */ |
225 | static int kmemleak_error; |
226 | |
227 | /* minimum and maximum address that may be valid pointers */ |
228 | static unsigned long min_addr = ULONG_MAX; |
229 | static unsigned long max_addr; |
230 | |
231 | static struct task_struct *scan_thread; |
232 | /* used to avoid reporting of recently allocated objects */ |
233 | static unsigned long jiffies_min_age; |
234 | static unsigned long jiffies_last_scan; |
235 | /* delay between automatic memory scannings */ |
236 | static unsigned long jiffies_scan_wait; |
237 | /* enables or disables the task stacks scanning */ |
238 | static int kmemleak_stack_scan = 1; |
239 | /* protects the memory scanning, parameters and debug/kmemleak file access */ |
240 | static DEFINE_MUTEX(scan_mutex); |
241 | /* setting kmemleak=on, will set this var, skipping the disable */ |
242 | static int kmemleak_skip_disable; |
243 | /* If there are leaks that can be reported */ |
244 | static bool kmemleak_found_leaks; |
245 | |
246 | static bool kmemleak_verbose; |
247 | module_param_named(verbose, kmemleak_verbose, bool, 0600); |
248 | |
249 | static void kmemleak_disable(void); |
250 | |
251 | /* |
252 | * Print a warning and dump the stack trace. |
253 | */ |
254 | #define kmemleak_warn(x...) do { \ |
255 | pr_warn(x); \ |
256 | dump_stack(); \ |
257 | kmemleak_warning = 1; \ |
258 | } while (0) |
259 | |
260 | /* |
261 | * Macro invoked when a serious kmemleak condition occurred and cannot be |
262 | * recovered from. Kmemleak will be disabled and further allocation/freeing |
263 | * tracing no longer available. |
264 | */ |
265 | #define kmemleak_stop(x...) do { \ |
266 | kmemleak_warn(x); \ |
267 | kmemleak_disable(); \ |
268 | } while (0) |
269 | |
270 | #define warn_or_seq_printf(seq, fmt, ...) do { \ |
271 | if (seq) \ |
272 | seq_printf(seq, fmt, ##__VA_ARGS__); \ |
273 | else \ |
274 | pr_warn(fmt, ##__VA_ARGS__); \ |
275 | } while (0) |
276 | |
277 | static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type, |
278 | int rowsize, int groupsize, const void *buf, |
279 | size_t len, bool ascii) |
280 | { |
281 | if (seq) |
282 | seq_hex_dump(m: seq, HEX_PREFIX, prefix_type, rowsize, groupsize, |
283 | buf, len, ascii); |
284 | else |
285 | print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type, |
286 | rowsize, groupsize, buf, len, ascii); |
287 | } |
288 | |
289 | /* |
290 | * Printing of the objects hex dump to the seq file. The number of lines to be |
291 | * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The |
292 | * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called |
293 | * with the object->lock held. |
294 | */ |
295 | static void hex_dump_object(struct seq_file *seq, |
296 | struct kmemleak_object *object) |
297 | { |
298 | const u8 *ptr = (const u8 *)object->pointer; |
299 | size_t len; |
300 | |
301 | if (WARN_ON_ONCE(object->flags & OBJECT_PHYS)) |
302 | return; |
303 | |
304 | /* limit the number of lines to HEX_MAX_LINES */ |
305 | len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE); |
306 | |
307 | warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n" , len); |
308 | kasan_disable_current(); |
309 | warn_or_seq_hex_dump(seq, prefix_type: DUMP_PREFIX_NONE, HEX_ROW_SIZE, |
310 | HEX_GROUP_SIZE, buf: kasan_reset_tag(addr: (void *)ptr), len, HEX_ASCII); |
311 | kasan_enable_current(); |
312 | } |
313 | |
314 | /* |
315 | * Object colors, encoded with count and min_count: |
316 | * - white - orphan object, not enough references to it (count < min_count) |
317 | * - gray - not orphan, not marked as false positive (min_count == 0) or |
318 | * sufficient references to it (count >= min_count) |
319 | * - black - ignore, it doesn't contain references (e.g. text section) |
320 | * (min_count == -1). No function defined for this color. |
321 | * Newly created objects don't have any color assigned (object->count == -1) |
322 | * before the next memory scan when they become white. |
323 | */ |
324 | static bool color_white(const struct kmemleak_object *object) |
325 | { |
326 | return object->count != KMEMLEAK_BLACK && |
327 | object->count < object->min_count; |
328 | } |
329 | |
330 | static bool color_gray(const struct kmemleak_object *object) |
331 | { |
332 | return object->min_count != KMEMLEAK_BLACK && |
333 | object->count >= object->min_count; |
334 | } |
335 | |
336 | /* |
337 | * Objects are considered unreferenced only if their color is white, they have |
338 | * not be deleted and have a minimum age to avoid false positives caused by |
339 | * pointers temporarily stored in CPU registers. |
340 | */ |
341 | static bool unreferenced_object(struct kmemleak_object *object) |
342 | { |
343 | return (color_white(object) && object->flags & OBJECT_ALLOCATED) && |
344 | time_before_eq(object->jiffies + jiffies_min_age, |
345 | jiffies_last_scan); |
346 | } |
347 | |
348 | /* |
349 | * Printing of the unreferenced objects information to the seq file. The |
350 | * print_unreferenced function must be called with the object->lock held. |
351 | */ |
352 | static void print_unreferenced(struct seq_file *seq, |
353 | struct kmemleak_object *object) |
354 | { |
355 | int i; |
356 | unsigned long *entries; |
357 | unsigned int nr_entries; |
358 | unsigned int msecs_age = jiffies_to_msecs(j: jiffies - object->jiffies); |
359 | |
360 | nr_entries = stack_depot_fetch(handle: object->trace_handle, entries: &entries); |
361 | warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n" , |
362 | object->pointer, object->size); |
363 | warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n" , |
364 | object->comm, object->pid, object->jiffies, |
365 | msecs_age / 1000, msecs_age % 1000); |
366 | hex_dump_object(seq, object); |
367 | warn_or_seq_printf(seq, " backtrace:\n" ); |
368 | |
369 | for (i = 0; i < nr_entries; i++) { |
370 | void *ptr = (void *)entries[i]; |
371 | warn_or_seq_printf(seq, " [<%pK>] %pS\n" , ptr, ptr); |
372 | } |
373 | } |
374 | |
375 | /* |
376 | * Print the kmemleak_object information. This function is used mainly for |
377 | * debugging special cases when kmemleak operations. It must be called with |
378 | * the object->lock held. |
379 | */ |
380 | static void dump_object_info(struct kmemleak_object *object) |
381 | { |
382 | pr_notice("Object 0x%08lx (size %zu):\n" , |
383 | object->pointer, object->size); |
384 | pr_notice(" comm \"%s\", pid %d, jiffies %lu\n" , |
385 | object->comm, object->pid, object->jiffies); |
386 | pr_notice(" min_count = %d\n" , object->min_count); |
387 | pr_notice(" count = %d\n" , object->count); |
388 | pr_notice(" flags = 0x%x\n" , object->flags); |
389 | pr_notice(" checksum = %u\n" , object->checksum); |
390 | pr_notice(" backtrace:\n" ); |
391 | if (object->trace_handle) |
392 | stack_depot_print(stack: object->trace_handle); |
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 | */ |
401 | static struct kmemleak_object *__lookup_object(unsigned long ptr, int alias, |
402 | bool is_phys) |
403 | { |
404 | struct rb_node *rb = is_phys ? object_phys_tree_root.rb_node : |
405 | object_tree_root.rb_node; |
406 | unsigned long untagged_ptr = (unsigned long)kasan_reset_tag(addr: (void *)ptr); |
407 | |
408 | while (rb) { |
409 | struct kmemleak_object *object; |
410 | unsigned long untagged_objp; |
411 | |
412 | object = rb_entry(rb, struct kmemleak_object, rb_node); |
413 | untagged_objp = (unsigned long)kasan_reset_tag(addr: (void *)object->pointer); |
414 | |
415 | if (untagged_ptr < untagged_objp) |
416 | rb = object->rb_node.rb_left; |
417 | else if (untagged_objp + object->size <= untagged_ptr) |
418 | rb = object->rb_node.rb_right; |
419 | else if (untagged_objp == untagged_ptr || alias) |
420 | return object; |
421 | else { |
422 | kmemleak_warn("Found object by alias at 0x%08lx\n" , |
423 | ptr); |
424 | dump_object_info(object); |
425 | break; |
426 | } |
427 | } |
428 | return NULL; |
429 | } |
430 | |
431 | /* Look-up a kmemleak object which allocated with virtual address. */ |
432 | static struct kmemleak_object *lookup_object(unsigned long ptr, int alias) |
433 | { |
434 | return __lookup_object(ptr, alias, is_phys: false); |
435 | } |
436 | |
437 | /* |
438 | * Increment the object use_count. Return 1 if successful or 0 otherwise. Note |
439 | * that once an object's use_count reached 0, the RCU freeing was already |
440 | * registered and the object should no longer be used. This function must be |
441 | * called under the protection of rcu_read_lock(). |
442 | */ |
443 | static int get_object(struct kmemleak_object *object) |
444 | { |
445 | return atomic_inc_not_zero(v: &object->use_count); |
446 | } |
447 | |
448 | /* |
449 | * Memory pool allocation and freeing. kmemleak_lock must not be held. |
450 | */ |
451 | static struct kmemleak_object *mem_pool_alloc(gfp_t gfp) |
452 | { |
453 | unsigned long flags; |
454 | struct kmemleak_object *object; |
455 | |
456 | /* try the slab allocator first */ |
457 | if (object_cache) { |
458 | object = kmem_cache_alloc(cachep: object_cache, gfp_kmemleak_mask(gfp)); |
459 | if (object) |
460 | return object; |
461 | } |
462 | |
463 | /* slab allocation failed, try the memory pool */ |
464 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
465 | object = list_first_entry_or_null(&mem_pool_free_list, |
466 | typeof(*object), object_list); |
467 | if (object) |
468 | list_del(entry: &object->object_list); |
469 | else if (mem_pool_free_count) |
470 | object = &mem_pool[--mem_pool_free_count]; |
471 | else |
472 | pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n" ); |
473 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
474 | |
475 | return object; |
476 | } |
477 | |
478 | /* |
479 | * Return the object to either the slab allocator or the memory pool. |
480 | */ |
481 | static void mem_pool_free(struct kmemleak_object *object) |
482 | { |
483 | unsigned long flags; |
484 | |
485 | if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) { |
486 | kmem_cache_free(s: object_cache, objp: object); |
487 | return; |
488 | } |
489 | |
490 | /* add the object to the memory pool free list */ |
491 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
492 | list_add(new: &object->object_list, head: &mem_pool_free_list); |
493 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
494 | } |
495 | |
496 | /* |
497 | * RCU callback to free a kmemleak_object. |
498 | */ |
499 | static void free_object_rcu(struct rcu_head *rcu) |
500 | { |
501 | struct hlist_node *tmp; |
502 | struct kmemleak_scan_area *area; |
503 | struct kmemleak_object *object = |
504 | container_of(rcu, struct kmemleak_object, rcu); |
505 | |
506 | /* |
507 | * Once use_count is 0 (guaranteed by put_object), there is no other |
508 | * code accessing this object, hence no need for locking. |
509 | */ |
510 | hlist_for_each_entry_safe(area, tmp, &object->area_list, node) { |
511 | hlist_del(n: &area->node); |
512 | kmem_cache_free(s: scan_area_cache, objp: area); |
513 | } |
514 | mem_pool_free(object); |
515 | } |
516 | |
517 | /* |
518 | * Decrement the object use_count. Once the count is 0, free the object using |
519 | * an RCU callback. Since put_object() may be called via the kmemleak_free() -> |
520 | * delete_object() path, the delayed RCU freeing ensures that there is no |
521 | * recursive call to the kernel allocator. Lock-less RCU object_list traversal |
522 | * is also possible. |
523 | */ |
524 | static void put_object(struct kmemleak_object *object) |
525 | { |
526 | if (!atomic_dec_and_test(v: &object->use_count)) |
527 | return; |
528 | |
529 | /* should only get here after delete_object was called */ |
530 | WARN_ON(object->flags & OBJECT_ALLOCATED); |
531 | |
532 | /* |
533 | * It may be too early for the RCU callbacks, however, there is no |
534 | * concurrent object_list traversal when !object_cache and all objects |
535 | * came from the memory pool. Free the object directly. |
536 | */ |
537 | if (object_cache) |
538 | call_rcu(head: &object->rcu, func: free_object_rcu); |
539 | else |
540 | free_object_rcu(rcu: &object->rcu); |
541 | } |
542 | |
543 | /* |
544 | * Look up an object in the object search tree and increase its use_count. |
545 | */ |
546 | static struct kmemleak_object *__find_and_get_object(unsigned long ptr, int alias, |
547 | bool is_phys) |
548 | { |
549 | unsigned long flags; |
550 | struct kmemleak_object *object; |
551 | |
552 | rcu_read_lock(); |
553 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
554 | object = __lookup_object(ptr, alias, is_phys); |
555 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
556 | |
557 | /* check whether the object is still available */ |
558 | if (object && !get_object(object)) |
559 | object = NULL; |
560 | rcu_read_unlock(); |
561 | |
562 | return object; |
563 | } |
564 | |
565 | /* Look up and get an object which allocated with virtual address. */ |
566 | static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias) |
567 | { |
568 | return __find_and_get_object(ptr, alias, is_phys: false); |
569 | } |
570 | |
571 | /* |
572 | * Remove an object from the object_tree_root (or object_phys_tree_root) |
573 | * and object_list. Must be called with the kmemleak_lock held _if_ kmemleak |
574 | * is still enabled. |
575 | */ |
576 | static void __remove_object(struct kmemleak_object *object) |
577 | { |
578 | rb_erase(&object->rb_node, object->flags & OBJECT_PHYS ? |
579 | &object_phys_tree_root : |
580 | &object_tree_root); |
581 | if (!(object->del_state & DELSTATE_NO_DELETE)) |
582 | list_del_rcu(entry: &object->object_list); |
583 | object->del_state |= DELSTATE_REMOVED; |
584 | } |
585 | |
586 | static struct kmemleak_object *__find_and_remove_object(unsigned long ptr, |
587 | int alias, |
588 | bool is_phys) |
589 | { |
590 | struct kmemleak_object *object; |
591 | |
592 | object = __lookup_object(ptr, alias, is_phys); |
593 | if (object) |
594 | __remove_object(object); |
595 | |
596 | return object; |
597 | } |
598 | |
599 | /* |
600 | * Look up an object in the object search tree and remove it from both |
601 | * object_tree_root (or object_phys_tree_root) and object_list. The |
602 | * returned object's use_count should be at least 1, as initially set |
603 | * by create_object(). |
604 | */ |
605 | static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias, |
606 | bool is_phys) |
607 | { |
608 | unsigned long flags; |
609 | struct kmemleak_object *object; |
610 | |
611 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
612 | object = __find_and_remove_object(ptr, alias, is_phys); |
613 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
614 | |
615 | return object; |
616 | } |
617 | |
618 | static noinline depot_stack_handle_t set_track_prepare(void) |
619 | { |
620 | depot_stack_handle_t trace_handle; |
621 | unsigned long entries[MAX_TRACE]; |
622 | unsigned int nr_entries; |
623 | |
624 | /* |
625 | * Use object_cache to determine whether kmemleak_init() has |
626 | * been invoked. stack_depot_early_init() is called before |
627 | * kmemleak_init() in mm_core_init(). |
628 | */ |
629 | if (!object_cache) |
630 | return 0; |
631 | nr_entries = stack_trace_save(store: entries, ARRAY_SIZE(entries), skipnr: 3); |
632 | trace_handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT); |
633 | |
634 | return trace_handle; |
635 | } |
636 | |
637 | static struct kmemleak_object *__alloc_object(gfp_t gfp) |
638 | { |
639 | struct kmemleak_object *object; |
640 | |
641 | object = mem_pool_alloc(gfp); |
642 | if (!object) { |
643 | pr_warn("Cannot allocate a kmemleak_object structure\n" ); |
644 | kmemleak_disable(); |
645 | } |
646 | |
647 | return object; |
648 | } |
649 | |
650 | static int __link_object(struct kmemleak_object *object, unsigned long ptr, |
651 | size_t size, int min_count, bool is_phys) |
652 | { |
653 | |
654 | struct kmemleak_object *parent; |
655 | struct rb_node **link, *rb_parent; |
656 | unsigned long untagged_ptr; |
657 | unsigned long untagged_objp; |
658 | |
659 | INIT_LIST_HEAD(list: &object->object_list); |
660 | INIT_LIST_HEAD(list: &object->gray_list); |
661 | INIT_HLIST_HEAD(&object->area_list); |
662 | raw_spin_lock_init(&object->lock); |
663 | atomic_set(v: &object->use_count, i: 1); |
664 | object->flags = OBJECT_ALLOCATED | (is_phys ? OBJECT_PHYS : 0); |
665 | object->pointer = ptr; |
666 | object->size = kfence_ksize(addr: (void *)ptr) ?: size; |
667 | object->excess_ref = 0; |
668 | object->min_count = min_count; |
669 | object->count = 0; /* white color initially */ |
670 | object->jiffies = jiffies; |
671 | object->checksum = 0; |
672 | object->del_state = 0; |
673 | |
674 | /* task information */ |
675 | if (in_hardirq()) { |
676 | object->pid = 0; |
677 | strncpy(p: object->comm, q: "hardirq" , size: sizeof(object->comm)); |
678 | } else if (in_serving_softirq()) { |
679 | object->pid = 0; |
680 | strncpy(p: object->comm, q: "softirq" , size: sizeof(object->comm)); |
681 | } else { |
682 | object->pid = current->pid; |
683 | /* |
684 | * There is a small chance of a race with set_task_comm(), |
685 | * however using get_task_comm() here may cause locking |
686 | * dependency issues with current->alloc_lock. In the worst |
687 | * case, the command line is not correct. |
688 | */ |
689 | strncpy(p: object->comm, current->comm, size: sizeof(object->comm)); |
690 | } |
691 | |
692 | /* kernel backtrace */ |
693 | object->trace_handle = set_track_prepare(); |
694 | |
695 | untagged_ptr = (unsigned long)kasan_reset_tag(addr: (void *)ptr); |
696 | /* |
697 | * Only update min_addr and max_addr with object |
698 | * storing virtual address. |
699 | */ |
700 | if (!is_phys) { |
701 | min_addr = min(min_addr, untagged_ptr); |
702 | max_addr = max(max_addr, untagged_ptr + size); |
703 | } |
704 | link = is_phys ? &object_phys_tree_root.rb_node : |
705 | &object_tree_root.rb_node; |
706 | rb_parent = NULL; |
707 | while (*link) { |
708 | rb_parent = *link; |
709 | parent = rb_entry(rb_parent, struct kmemleak_object, rb_node); |
710 | untagged_objp = (unsigned long)kasan_reset_tag(addr: (void *)parent->pointer); |
711 | if (untagged_ptr + size <= untagged_objp) |
712 | link = &parent->rb_node.rb_left; |
713 | else if (untagged_objp + parent->size <= untagged_ptr) |
714 | link = &parent->rb_node.rb_right; |
715 | else { |
716 | kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n" , |
717 | ptr); |
718 | /* |
719 | * No need for parent->lock here since "parent" cannot |
720 | * be freed while the kmemleak_lock is held. |
721 | */ |
722 | dump_object_info(object: parent); |
723 | return -EEXIST; |
724 | } |
725 | } |
726 | rb_link_node(node: &object->rb_node, parent: rb_parent, rb_link: link); |
727 | rb_insert_color(&object->rb_node, is_phys ? &object_phys_tree_root : |
728 | &object_tree_root); |
729 | list_add_tail_rcu(new: &object->object_list, head: &object_list); |
730 | |
731 | return 0; |
732 | } |
733 | |
734 | /* |
735 | * Create the metadata (struct kmemleak_object) corresponding to an allocated |
736 | * memory block and add it to the object_list and object_tree_root (or |
737 | * object_phys_tree_root). |
738 | */ |
739 | static void __create_object(unsigned long ptr, size_t size, |
740 | int min_count, gfp_t gfp, bool is_phys) |
741 | { |
742 | struct kmemleak_object *object; |
743 | unsigned long flags; |
744 | int ret; |
745 | |
746 | object = __alloc_object(gfp); |
747 | if (!object) |
748 | return; |
749 | |
750 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
751 | ret = __link_object(object, ptr, size, min_count, is_phys); |
752 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
753 | if (ret) |
754 | mem_pool_free(object); |
755 | } |
756 | |
757 | /* Create kmemleak object which allocated with virtual address. */ |
758 | static void create_object(unsigned long ptr, size_t size, |
759 | int min_count, gfp_t gfp) |
760 | { |
761 | __create_object(ptr, size, min_count, gfp, is_phys: false); |
762 | } |
763 | |
764 | /* Create kmemleak object which allocated with physical address. */ |
765 | static void create_object_phys(unsigned long ptr, size_t size, |
766 | int min_count, gfp_t gfp) |
767 | { |
768 | __create_object(ptr, size, min_count, gfp, is_phys: true); |
769 | } |
770 | |
771 | /* |
772 | * Mark the object as not allocated and schedule RCU freeing via put_object(). |
773 | */ |
774 | static void __delete_object(struct kmemleak_object *object) |
775 | { |
776 | unsigned long flags; |
777 | |
778 | WARN_ON(!(object->flags & OBJECT_ALLOCATED)); |
779 | WARN_ON(atomic_read(&object->use_count) < 1); |
780 | |
781 | /* |
782 | * Locking here also ensures that the corresponding memory block |
783 | * cannot be freed when it is being scanned. |
784 | */ |
785 | raw_spin_lock_irqsave(&object->lock, flags); |
786 | object->flags &= ~OBJECT_ALLOCATED; |
787 | raw_spin_unlock_irqrestore(&object->lock, flags); |
788 | put_object(object); |
789 | } |
790 | |
791 | /* |
792 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and |
793 | * delete it. |
794 | */ |
795 | static void delete_object_full(unsigned long ptr) |
796 | { |
797 | struct kmemleak_object *object; |
798 | |
799 | object = find_and_remove_object(ptr, alias: 0, is_phys: false); |
800 | if (!object) { |
801 | #ifdef DEBUG |
802 | kmemleak_warn("Freeing unknown object at 0x%08lx\n" , |
803 | ptr); |
804 | #endif |
805 | return; |
806 | } |
807 | __delete_object(object); |
808 | } |
809 | |
810 | /* |
811 | * Look up the metadata (struct kmemleak_object) corresponding to ptr and |
812 | * delete it. If the memory block is partially freed, the function may create |
813 | * additional metadata for the remaining parts of the block. |
814 | */ |
815 | static void delete_object_part(unsigned long ptr, size_t size, bool is_phys) |
816 | { |
817 | struct kmemleak_object *object, *object_l, *object_r; |
818 | unsigned long start, end, flags; |
819 | |
820 | object_l = __alloc_object(GFP_KERNEL); |
821 | if (!object_l) |
822 | return; |
823 | |
824 | object_r = __alloc_object(GFP_KERNEL); |
825 | if (!object_r) |
826 | goto out; |
827 | |
828 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
829 | object = __find_and_remove_object(ptr, alias: 1, is_phys); |
830 | if (!object) { |
831 | #ifdef DEBUG |
832 | kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n" , |
833 | ptr, size); |
834 | #endif |
835 | goto unlock; |
836 | } |
837 | |
838 | /* |
839 | * Create one or two objects that may result from the memory block |
840 | * split. Note that partial freeing is only done by free_bootmem() and |
841 | * this happens before kmemleak_init() is called. |
842 | */ |
843 | start = object->pointer; |
844 | end = object->pointer + object->size; |
845 | if ((ptr > start) && |
846 | !__link_object(object: object_l, ptr: start, size: ptr - start, |
847 | min_count: object->min_count, is_phys)) |
848 | object_l = NULL; |
849 | if ((ptr + size < end) && |
850 | !__link_object(object: object_r, ptr: ptr + size, size: end - ptr - size, |
851 | min_count: object->min_count, is_phys)) |
852 | object_r = NULL; |
853 | |
854 | unlock: |
855 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
856 | if (object) |
857 | __delete_object(object); |
858 | |
859 | out: |
860 | if (object_l) |
861 | mem_pool_free(object: object_l); |
862 | if (object_r) |
863 | mem_pool_free(object: object_r); |
864 | } |
865 | |
866 | static void __paint_it(struct kmemleak_object *object, int color) |
867 | { |
868 | object->min_count = color; |
869 | if (color == KMEMLEAK_BLACK) |
870 | object->flags |= OBJECT_NO_SCAN; |
871 | } |
872 | |
873 | static void paint_it(struct kmemleak_object *object, int color) |
874 | { |
875 | unsigned long flags; |
876 | |
877 | raw_spin_lock_irqsave(&object->lock, flags); |
878 | __paint_it(object, color); |
879 | raw_spin_unlock_irqrestore(&object->lock, flags); |
880 | } |
881 | |
882 | static void paint_ptr(unsigned long ptr, int color, bool is_phys) |
883 | { |
884 | struct kmemleak_object *object; |
885 | |
886 | object = __find_and_get_object(ptr, alias: 0, is_phys); |
887 | if (!object) { |
888 | kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n" , |
889 | ptr, |
890 | (color == KMEMLEAK_GREY) ? "Grey" : |
891 | (color == KMEMLEAK_BLACK) ? "Black" : "Unknown" ); |
892 | return; |
893 | } |
894 | paint_it(object, color); |
895 | put_object(object); |
896 | } |
897 | |
898 | /* |
899 | * Mark an object permanently as gray-colored so that it can no longer be |
900 | * reported as a leak. This is used in general to mark a false positive. |
901 | */ |
902 | static void make_gray_object(unsigned long ptr) |
903 | { |
904 | paint_ptr(ptr, KMEMLEAK_GREY, is_phys: false); |
905 | } |
906 | |
907 | /* |
908 | * Mark the object as black-colored so that it is ignored from scans and |
909 | * reporting. |
910 | */ |
911 | static void make_black_object(unsigned long ptr, bool is_phys) |
912 | { |
913 | paint_ptr(ptr, KMEMLEAK_BLACK, is_phys); |
914 | } |
915 | |
916 | /* |
917 | * Add a scanning area to the object. If at least one such area is added, |
918 | * kmemleak will only scan these ranges rather than the whole memory block. |
919 | */ |
920 | static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp) |
921 | { |
922 | unsigned long flags; |
923 | struct kmemleak_object *object; |
924 | struct kmemleak_scan_area *area = NULL; |
925 | unsigned long untagged_ptr; |
926 | unsigned long untagged_objp; |
927 | |
928 | object = find_and_get_object(ptr, alias: 1); |
929 | if (!object) { |
930 | kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n" , |
931 | ptr); |
932 | return; |
933 | } |
934 | |
935 | untagged_ptr = (unsigned long)kasan_reset_tag(addr: (void *)ptr); |
936 | untagged_objp = (unsigned long)kasan_reset_tag(addr: (void *)object->pointer); |
937 | |
938 | if (scan_area_cache) |
939 | area = kmem_cache_alloc(cachep: scan_area_cache, gfp_kmemleak_mask(gfp)); |
940 | |
941 | raw_spin_lock_irqsave(&object->lock, flags); |
942 | if (!area) { |
943 | pr_warn_once("Cannot allocate a scan area, scanning the full object\n" ); |
944 | /* mark the object for full scan to avoid false positives */ |
945 | object->flags |= OBJECT_FULL_SCAN; |
946 | goto out_unlock; |
947 | } |
948 | if (size == SIZE_MAX) { |
949 | size = untagged_objp + object->size - untagged_ptr; |
950 | } else if (untagged_ptr + size > untagged_objp + object->size) { |
951 | kmemleak_warn("Scan area larger than object 0x%08lx\n" , ptr); |
952 | dump_object_info(object); |
953 | kmem_cache_free(s: scan_area_cache, objp: area); |
954 | goto out_unlock; |
955 | } |
956 | |
957 | INIT_HLIST_NODE(h: &area->node); |
958 | area->start = ptr; |
959 | area->size = size; |
960 | |
961 | hlist_add_head(n: &area->node, h: &object->area_list); |
962 | out_unlock: |
963 | raw_spin_unlock_irqrestore(&object->lock, flags); |
964 | put_object(object); |
965 | } |
966 | |
967 | /* |
968 | * Any surplus references (object already gray) to 'ptr' are passed to |
969 | * 'excess_ref'. This is used in the vmalloc() case where a pointer to |
970 | * vm_struct may be used as an alternative reference to the vmalloc'ed object |
971 | * (see free_thread_stack()). |
972 | */ |
973 | static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref) |
974 | { |
975 | unsigned long flags; |
976 | struct kmemleak_object *object; |
977 | |
978 | object = find_and_get_object(ptr, alias: 0); |
979 | if (!object) { |
980 | kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n" , |
981 | ptr); |
982 | return; |
983 | } |
984 | |
985 | raw_spin_lock_irqsave(&object->lock, flags); |
986 | object->excess_ref = excess_ref; |
987 | raw_spin_unlock_irqrestore(&object->lock, flags); |
988 | put_object(object); |
989 | } |
990 | |
991 | /* |
992 | * Set the OBJECT_NO_SCAN flag for the object corresponding to the give |
993 | * pointer. Such object will not be scanned by kmemleak but references to it |
994 | * are searched. |
995 | */ |
996 | static void object_no_scan(unsigned long ptr) |
997 | { |
998 | unsigned long flags; |
999 | struct kmemleak_object *object; |
1000 | |
1001 | object = find_and_get_object(ptr, alias: 0); |
1002 | if (!object) { |
1003 | kmemleak_warn("Not scanning unknown object at 0x%08lx\n" , ptr); |
1004 | return; |
1005 | } |
1006 | |
1007 | raw_spin_lock_irqsave(&object->lock, flags); |
1008 | object->flags |= OBJECT_NO_SCAN; |
1009 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1010 | put_object(object); |
1011 | } |
1012 | |
1013 | /** |
1014 | * kmemleak_alloc - register a newly allocated object |
1015 | * @ptr: pointer to beginning of the object |
1016 | * @size: size of the object |
1017 | * @min_count: minimum number of references to this object. If during memory |
1018 | * scanning a number of references less than @min_count is found, |
1019 | * the object is reported as a memory leak. If @min_count is 0, |
1020 | * the object is never reported as a leak. If @min_count is -1, |
1021 | * the object is ignored (not scanned and not reported as a leak) |
1022 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations |
1023 | * |
1024 | * This function is called from the kernel allocators when a new object |
1025 | * (memory block) is allocated (kmem_cache_alloc, kmalloc etc.). |
1026 | */ |
1027 | void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count, |
1028 | gfp_t gfp) |
1029 | { |
1030 | pr_debug("%s(0x%px, %zu, %d)\n" , __func__, ptr, size, min_count); |
1031 | |
1032 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1033 | create_object(ptr: (unsigned long)ptr, size, min_count, gfp); |
1034 | } |
1035 | EXPORT_SYMBOL_GPL(kmemleak_alloc); |
1036 | |
1037 | /** |
1038 | * kmemleak_alloc_percpu - register a newly allocated __percpu object |
1039 | * @ptr: __percpu pointer to beginning of the object |
1040 | * @size: size of the object |
1041 | * @gfp: flags used for kmemleak internal memory allocations |
1042 | * |
1043 | * This function is called from the kernel percpu allocator when a new object |
1044 | * (memory block) is allocated (alloc_percpu). |
1045 | */ |
1046 | void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size, |
1047 | gfp_t gfp) |
1048 | { |
1049 | unsigned int cpu; |
1050 | |
1051 | pr_debug("%s(0x%px, %zu)\n" , __func__, ptr, size); |
1052 | |
1053 | /* |
1054 | * Percpu allocations are only scanned and not reported as leaks |
1055 | * (min_count is set to 0). |
1056 | */ |
1057 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1058 | for_each_possible_cpu(cpu) |
1059 | create_object(ptr: (unsigned long)per_cpu_ptr(ptr, cpu), |
1060 | size, min_count: 0, gfp); |
1061 | } |
1062 | EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu); |
1063 | |
1064 | /** |
1065 | * kmemleak_vmalloc - register a newly vmalloc'ed object |
1066 | * @area: pointer to vm_struct |
1067 | * @size: size of the object |
1068 | * @gfp: __vmalloc() flags used for kmemleak internal memory allocations |
1069 | * |
1070 | * This function is called from the vmalloc() kernel allocator when a new |
1071 | * object (memory block) is allocated. |
1072 | */ |
1073 | void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp) |
1074 | { |
1075 | pr_debug("%s(0x%px, %zu)\n" , __func__, area, size); |
1076 | |
1077 | /* |
1078 | * A min_count = 2 is needed because vm_struct contains a reference to |
1079 | * the virtual address of the vmalloc'ed block. |
1080 | */ |
1081 | if (kmemleak_enabled) { |
1082 | create_object(ptr: (unsigned long)area->addr, size, min_count: 2, gfp); |
1083 | object_set_excess_ref(ptr: (unsigned long)area, |
1084 | excess_ref: (unsigned long)area->addr); |
1085 | } |
1086 | } |
1087 | EXPORT_SYMBOL_GPL(kmemleak_vmalloc); |
1088 | |
1089 | /** |
1090 | * kmemleak_free - unregister a previously registered object |
1091 | * @ptr: pointer to beginning of the object |
1092 | * |
1093 | * This function is called from the kernel allocators when an object (memory |
1094 | * block) is freed (kmem_cache_free, kfree, vfree etc.). |
1095 | */ |
1096 | void __ref kmemleak_free(const void *ptr) |
1097 | { |
1098 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1099 | |
1100 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
1101 | delete_object_full(ptr: (unsigned long)ptr); |
1102 | } |
1103 | EXPORT_SYMBOL_GPL(kmemleak_free); |
1104 | |
1105 | /** |
1106 | * kmemleak_free_part - partially unregister a previously registered object |
1107 | * @ptr: pointer to the beginning or inside the object. This also |
1108 | * represents the start of the range to be freed |
1109 | * @size: size to be unregistered |
1110 | * |
1111 | * This function is called when only a part of a memory block is freed |
1112 | * (usually from the bootmem allocator). |
1113 | */ |
1114 | void __ref kmemleak_free_part(const void *ptr, size_t size) |
1115 | { |
1116 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1117 | |
1118 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1119 | delete_object_part(ptr: (unsigned long)ptr, size, is_phys: false); |
1120 | } |
1121 | EXPORT_SYMBOL_GPL(kmemleak_free_part); |
1122 | |
1123 | /** |
1124 | * kmemleak_free_percpu - unregister a previously registered __percpu object |
1125 | * @ptr: __percpu pointer to beginning of the object |
1126 | * |
1127 | * This function is called from the kernel percpu allocator when an object |
1128 | * (memory block) is freed (free_percpu). |
1129 | */ |
1130 | void __ref kmemleak_free_percpu(const void __percpu *ptr) |
1131 | { |
1132 | unsigned int cpu; |
1133 | |
1134 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1135 | |
1136 | if (kmemleak_free_enabled && ptr && !IS_ERR(ptr)) |
1137 | for_each_possible_cpu(cpu) |
1138 | delete_object_full(ptr: (unsigned long)per_cpu_ptr(ptr, |
1139 | cpu)); |
1140 | } |
1141 | EXPORT_SYMBOL_GPL(kmemleak_free_percpu); |
1142 | |
1143 | /** |
1144 | * kmemleak_update_trace - update object allocation stack trace |
1145 | * @ptr: pointer to beginning of the object |
1146 | * |
1147 | * Override the object allocation stack trace for cases where the actual |
1148 | * allocation place is not always useful. |
1149 | */ |
1150 | void __ref kmemleak_update_trace(const void *ptr) |
1151 | { |
1152 | struct kmemleak_object *object; |
1153 | unsigned long flags; |
1154 | |
1155 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1156 | |
1157 | if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr)) |
1158 | return; |
1159 | |
1160 | object = find_and_get_object(ptr: (unsigned long)ptr, alias: 1); |
1161 | if (!object) { |
1162 | #ifdef DEBUG |
1163 | kmemleak_warn("Updating stack trace for unknown object at %p\n" , |
1164 | ptr); |
1165 | #endif |
1166 | return; |
1167 | } |
1168 | |
1169 | raw_spin_lock_irqsave(&object->lock, flags); |
1170 | object->trace_handle = set_track_prepare(); |
1171 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1172 | |
1173 | put_object(object); |
1174 | } |
1175 | EXPORT_SYMBOL(kmemleak_update_trace); |
1176 | |
1177 | /** |
1178 | * kmemleak_not_leak - mark an allocated object as false positive |
1179 | * @ptr: pointer to beginning of the object |
1180 | * |
1181 | * Calling this function on an object will cause the memory block to no longer |
1182 | * be reported as leak and always be scanned. |
1183 | */ |
1184 | void __ref kmemleak_not_leak(const void *ptr) |
1185 | { |
1186 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1187 | |
1188 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1189 | make_gray_object(ptr: (unsigned long)ptr); |
1190 | } |
1191 | EXPORT_SYMBOL(kmemleak_not_leak); |
1192 | |
1193 | /** |
1194 | * kmemleak_ignore - ignore an allocated object |
1195 | * @ptr: pointer to beginning of the object |
1196 | * |
1197 | * Calling this function on an object will cause the memory block to be |
1198 | * ignored (not scanned and not reported as a leak). This is usually done when |
1199 | * it is known that the corresponding block is not a leak and does not contain |
1200 | * any references to other allocated memory blocks. |
1201 | */ |
1202 | void __ref kmemleak_ignore(const void *ptr) |
1203 | { |
1204 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1205 | |
1206 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1207 | make_black_object(ptr: (unsigned long)ptr, is_phys: false); |
1208 | } |
1209 | EXPORT_SYMBOL(kmemleak_ignore); |
1210 | |
1211 | /** |
1212 | * kmemleak_scan_area - limit the range to be scanned in an allocated object |
1213 | * @ptr: pointer to beginning or inside the object. This also |
1214 | * represents the start of the scan area |
1215 | * @size: size of the scan area |
1216 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations |
1217 | * |
1218 | * This function is used when it is known that only certain parts of an object |
1219 | * contain references to other objects. Kmemleak will only scan these areas |
1220 | * reducing the number false negatives. |
1221 | */ |
1222 | void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp) |
1223 | { |
1224 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1225 | |
1226 | if (kmemleak_enabled && ptr && size && !IS_ERR(ptr)) |
1227 | add_scan_area(ptr: (unsigned long)ptr, size, gfp); |
1228 | } |
1229 | EXPORT_SYMBOL(kmemleak_scan_area); |
1230 | |
1231 | /** |
1232 | * kmemleak_no_scan - do not scan an allocated object |
1233 | * @ptr: pointer to beginning of the object |
1234 | * |
1235 | * This function notifies kmemleak not to scan the given memory block. Useful |
1236 | * in situations where it is known that the given object does not contain any |
1237 | * references to other objects. Kmemleak will not scan such objects reducing |
1238 | * the number of false negatives. |
1239 | */ |
1240 | void __ref kmemleak_no_scan(const void *ptr) |
1241 | { |
1242 | pr_debug("%s(0x%px)\n" , __func__, ptr); |
1243 | |
1244 | if (kmemleak_enabled && ptr && !IS_ERR(ptr)) |
1245 | object_no_scan(ptr: (unsigned long)ptr); |
1246 | } |
1247 | EXPORT_SYMBOL(kmemleak_no_scan); |
1248 | |
1249 | /** |
1250 | * kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical |
1251 | * address argument |
1252 | * @phys: physical address of the object |
1253 | * @size: size of the object |
1254 | * @gfp: kmalloc() flags used for kmemleak internal memory allocations |
1255 | */ |
1256 | void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, gfp_t gfp) |
1257 | { |
1258 | pr_debug("%s(0x%px, %zu)\n" , __func__, &phys, size); |
1259 | |
1260 | if (kmemleak_enabled) |
1261 | /* |
1262 | * Create object with OBJECT_PHYS flag and |
1263 | * assume min_count 0. |
1264 | */ |
1265 | create_object_phys(ptr: (unsigned long)phys, size, min_count: 0, gfp); |
1266 | } |
1267 | EXPORT_SYMBOL(kmemleak_alloc_phys); |
1268 | |
1269 | /** |
1270 | * kmemleak_free_part_phys - similar to kmemleak_free_part but taking a |
1271 | * physical address argument |
1272 | * @phys: physical address if the beginning or inside an object. This |
1273 | * also represents the start of the range to be freed |
1274 | * @size: size to be unregistered |
1275 | */ |
1276 | void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size) |
1277 | { |
1278 | pr_debug("%s(0x%px)\n" , __func__, &phys); |
1279 | |
1280 | if (kmemleak_enabled) |
1281 | delete_object_part(ptr: (unsigned long)phys, size, is_phys: true); |
1282 | } |
1283 | EXPORT_SYMBOL(kmemleak_free_part_phys); |
1284 | |
1285 | /** |
1286 | * kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical |
1287 | * address argument |
1288 | * @phys: physical address of the object |
1289 | */ |
1290 | void __ref kmemleak_ignore_phys(phys_addr_t phys) |
1291 | { |
1292 | pr_debug("%s(0x%px)\n" , __func__, &phys); |
1293 | |
1294 | if (kmemleak_enabled) |
1295 | make_black_object(ptr: (unsigned long)phys, is_phys: true); |
1296 | } |
1297 | EXPORT_SYMBOL(kmemleak_ignore_phys); |
1298 | |
1299 | /* |
1300 | * Update an object's checksum and return true if it was modified. |
1301 | */ |
1302 | static bool update_checksum(struct kmemleak_object *object) |
1303 | { |
1304 | u32 old_csum = object->checksum; |
1305 | |
1306 | if (WARN_ON_ONCE(object->flags & OBJECT_PHYS)) |
1307 | return false; |
1308 | |
1309 | kasan_disable_current(); |
1310 | kcsan_disable_current(); |
1311 | object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size); |
1312 | kasan_enable_current(); |
1313 | kcsan_enable_current(); |
1314 | |
1315 | return object->checksum != old_csum; |
1316 | } |
1317 | |
1318 | /* |
1319 | * Update an object's references. object->lock must be held by the caller. |
1320 | */ |
1321 | static void update_refs(struct kmemleak_object *object) |
1322 | { |
1323 | if (!color_white(object)) { |
1324 | /* non-orphan, ignored or new */ |
1325 | return; |
1326 | } |
1327 | |
1328 | /* |
1329 | * Increase the object's reference count (number of pointers to the |
1330 | * memory block). If this count reaches the required minimum, the |
1331 | * object's color will become gray and it will be added to the |
1332 | * gray_list. |
1333 | */ |
1334 | object->count++; |
1335 | if (color_gray(object)) { |
1336 | /* put_object() called when removing from gray_list */ |
1337 | WARN_ON(!get_object(object)); |
1338 | list_add_tail(new: &object->gray_list, head: &gray_list); |
1339 | } |
1340 | } |
1341 | |
1342 | /* |
1343 | * Memory scanning is a long process and it needs to be interruptible. This |
1344 | * function checks whether such interrupt condition occurred. |
1345 | */ |
1346 | static int scan_should_stop(void) |
1347 | { |
1348 | if (!kmemleak_enabled) |
1349 | return 1; |
1350 | |
1351 | /* |
1352 | * This function may be called from either process or kthread context, |
1353 | * hence the need to check for both stop conditions. |
1354 | */ |
1355 | if (current->mm) |
1356 | return signal_pending(current); |
1357 | else |
1358 | return kthread_should_stop(); |
1359 | |
1360 | return 0; |
1361 | } |
1362 | |
1363 | /* |
1364 | * Scan a memory block (exclusive range) for valid pointers and add those |
1365 | * found to the gray list. |
1366 | */ |
1367 | static void scan_block(void *_start, void *_end, |
1368 | struct kmemleak_object *scanned) |
1369 | { |
1370 | unsigned long *ptr; |
1371 | unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER); |
1372 | unsigned long *end = _end - (BYTES_PER_POINTER - 1); |
1373 | unsigned long flags; |
1374 | unsigned long untagged_ptr; |
1375 | |
1376 | raw_spin_lock_irqsave(&kmemleak_lock, flags); |
1377 | for (ptr = start; ptr < end; ptr++) { |
1378 | struct kmemleak_object *object; |
1379 | unsigned long pointer; |
1380 | unsigned long excess_ref; |
1381 | |
1382 | if (scan_should_stop()) |
1383 | break; |
1384 | |
1385 | kasan_disable_current(); |
1386 | pointer = *(unsigned long *)kasan_reset_tag(addr: (void *)ptr); |
1387 | kasan_enable_current(); |
1388 | |
1389 | untagged_ptr = (unsigned long)kasan_reset_tag(addr: (void *)pointer); |
1390 | if (untagged_ptr < min_addr || untagged_ptr >= max_addr) |
1391 | continue; |
1392 | |
1393 | /* |
1394 | * No need for get_object() here since we hold kmemleak_lock. |
1395 | * object->use_count cannot be dropped to 0 while the object |
1396 | * is still present in object_tree_root and object_list |
1397 | * (with updates protected by kmemleak_lock). |
1398 | */ |
1399 | object = lookup_object(ptr: pointer, alias: 1); |
1400 | if (!object) |
1401 | continue; |
1402 | if (object == scanned) |
1403 | /* self referenced, ignore */ |
1404 | continue; |
1405 | |
1406 | /* |
1407 | * Avoid the lockdep recursive warning on object->lock being |
1408 | * previously acquired in scan_object(). These locks are |
1409 | * enclosed by scan_mutex. |
1410 | */ |
1411 | raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); |
1412 | /* only pass surplus references (object already gray) */ |
1413 | if (color_gray(object)) { |
1414 | excess_ref = object->excess_ref; |
1415 | /* no need for update_refs() if object already gray */ |
1416 | } else { |
1417 | excess_ref = 0; |
1418 | update_refs(object); |
1419 | } |
1420 | raw_spin_unlock(&object->lock); |
1421 | |
1422 | if (excess_ref) { |
1423 | object = lookup_object(ptr: excess_ref, alias: 0); |
1424 | if (!object) |
1425 | continue; |
1426 | if (object == scanned) |
1427 | /* circular reference, ignore */ |
1428 | continue; |
1429 | raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING); |
1430 | update_refs(object); |
1431 | raw_spin_unlock(&object->lock); |
1432 | } |
1433 | } |
1434 | raw_spin_unlock_irqrestore(&kmemleak_lock, flags); |
1435 | } |
1436 | |
1437 | /* |
1438 | * Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency. |
1439 | */ |
1440 | #ifdef CONFIG_SMP |
1441 | static void scan_large_block(void *start, void *end) |
1442 | { |
1443 | void *next; |
1444 | |
1445 | while (start < end) { |
1446 | next = min(start + MAX_SCAN_SIZE, end); |
1447 | scan_block(start: start, end: next, NULL); |
1448 | start = next; |
1449 | cond_resched(); |
1450 | } |
1451 | } |
1452 | #endif |
1453 | |
1454 | /* |
1455 | * Scan a memory block corresponding to a kmemleak_object. A condition is |
1456 | * that object->use_count >= 1. |
1457 | */ |
1458 | static void scan_object(struct kmemleak_object *object) |
1459 | { |
1460 | struct kmemleak_scan_area *area; |
1461 | unsigned long flags; |
1462 | void *obj_ptr; |
1463 | |
1464 | /* |
1465 | * Once the object->lock is acquired, the corresponding memory block |
1466 | * cannot be freed (the same lock is acquired in delete_object). |
1467 | */ |
1468 | raw_spin_lock_irqsave(&object->lock, flags); |
1469 | if (object->flags & OBJECT_NO_SCAN) |
1470 | goto out; |
1471 | if (!(object->flags & OBJECT_ALLOCATED)) |
1472 | /* already freed object */ |
1473 | goto out; |
1474 | |
1475 | obj_ptr = object->flags & OBJECT_PHYS ? |
1476 | __va((phys_addr_t)object->pointer) : |
1477 | (void *)object->pointer; |
1478 | |
1479 | if (hlist_empty(h: &object->area_list) || |
1480 | object->flags & OBJECT_FULL_SCAN) { |
1481 | void *start = obj_ptr; |
1482 | void *end = obj_ptr + object->size; |
1483 | void *next; |
1484 | |
1485 | do { |
1486 | next = min(start + MAX_SCAN_SIZE, end); |
1487 | scan_block(start: start, end: next, scanned: object); |
1488 | |
1489 | start = next; |
1490 | if (start >= end) |
1491 | break; |
1492 | |
1493 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1494 | cond_resched(); |
1495 | raw_spin_lock_irqsave(&object->lock, flags); |
1496 | } while (object->flags & OBJECT_ALLOCATED); |
1497 | } else |
1498 | hlist_for_each_entry(area, &object->area_list, node) |
1499 | scan_block(start: (void *)area->start, |
1500 | end: (void *)(area->start + area->size), |
1501 | scanned: object); |
1502 | out: |
1503 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1504 | } |
1505 | |
1506 | /* |
1507 | * Scan the objects already referenced (gray objects). More objects will be |
1508 | * referenced and, if there are no memory leaks, all the objects are scanned. |
1509 | */ |
1510 | static void scan_gray_list(void) |
1511 | { |
1512 | struct kmemleak_object *object, *tmp; |
1513 | |
1514 | /* |
1515 | * The list traversal is safe for both tail additions and removals |
1516 | * from inside the loop. The kmemleak objects cannot be freed from |
1517 | * outside the loop because their use_count was incremented. |
1518 | */ |
1519 | object = list_entry(gray_list.next, typeof(*object), gray_list); |
1520 | while (&object->gray_list != &gray_list) { |
1521 | cond_resched(); |
1522 | |
1523 | /* may add new objects to the list */ |
1524 | if (!scan_should_stop()) |
1525 | scan_object(object); |
1526 | |
1527 | tmp = list_entry(object->gray_list.next, typeof(*object), |
1528 | gray_list); |
1529 | |
1530 | /* remove the object from the list and release it */ |
1531 | list_del(entry: &object->gray_list); |
1532 | put_object(object); |
1533 | |
1534 | object = tmp; |
1535 | } |
1536 | WARN_ON(!list_empty(&gray_list)); |
1537 | } |
1538 | |
1539 | /* |
1540 | * Conditionally call resched() in an object iteration loop while making sure |
1541 | * that the given object won't go away without RCU read lock by performing a |
1542 | * get_object() if necessaary. |
1543 | */ |
1544 | static void kmemleak_cond_resched(struct kmemleak_object *object) |
1545 | { |
1546 | if (!get_object(object)) |
1547 | return; /* Try next object */ |
1548 | |
1549 | raw_spin_lock_irq(&kmemleak_lock); |
1550 | if (object->del_state & DELSTATE_REMOVED) |
1551 | goto unlock_put; /* Object removed */ |
1552 | object->del_state |= DELSTATE_NO_DELETE; |
1553 | raw_spin_unlock_irq(&kmemleak_lock); |
1554 | |
1555 | rcu_read_unlock(); |
1556 | cond_resched(); |
1557 | rcu_read_lock(); |
1558 | |
1559 | raw_spin_lock_irq(&kmemleak_lock); |
1560 | if (object->del_state & DELSTATE_REMOVED) |
1561 | list_del_rcu(entry: &object->object_list); |
1562 | object->del_state &= ~DELSTATE_NO_DELETE; |
1563 | unlock_put: |
1564 | raw_spin_unlock_irq(&kmemleak_lock); |
1565 | put_object(object); |
1566 | } |
1567 | |
1568 | /* |
1569 | * Scan data sections and all the referenced memory blocks allocated via the |
1570 | * kernel's standard allocators. This function must be called with the |
1571 | * scan_mutex held. |
1572 | */ |
1573 | static void kmemleak_scan(void) |
1574 | { |
1575 | struct kmemleak_object *object; |
1576 | struct zone *zone; |
1577 | int __maybe_unused i; |
1578 | int new_leaks = 0; |
1579 | |
1580 | jiffies_last_scan = jiffies; |
1581 | |
1582 | /* prepare the kmemleak_object's */ |
1583 | rcu_read_lock(); |
1584 | list_for_each_entry_rcu(object, &object_list, object_list) { |
1585 | raw_spin_lock_irq(&object->lock); |
1586 | #ifdef DEBUG |
1587 | /* |
1588 | * With a few exceptions there should be a maximum of |
1589 | * 1 reference to any object at this point. |
1590 | */ |
1591 | if (atomic_read(&object->use_count) > 1) { |
1592 | pr_debug("object->use_count = %d\n" , |
1593 | atomic_read(&object->use_count)); |
1594 | dump_object_info(object); |
1595 | } |
1596 | #endif |
1597 | |
1598 | /* ignore objects outside lowmem (paint them black) */ |
1599 | if ((object->flags & OBJECT_PHYS) && |
1600 | !(object->flags & OBJECT_NO_SCAN)) { |
1601 | unsigned long phys = object->pointer; |
1602 | |
1603 | if (PHYS_PFN(phys) < min_low_pfn || |
1604 | PHYS_PFN(phys + object->size) >= max_low_pfn) |
1605 | __paint_it(object, KMEMLEAK_BLACK); |
1606 | } |
1607 | |
1608 | /* reset the reference count (whiten the object) */ |
1609 | object->count = 0; |
1610 | if (color_gray(object) && get_object(object)) |
1611 | list_add_tail(new: &object->gray_list, head: &gray_list); |
1612 | |
1613 | raw_spin_unlock_irq(&object->lock); |
1614 | |
1615 | if (need_resched()) |
1616 | kmemleak_cond_resched(object); |
1617 | } |
1618 | rcu_read_unlock(); |
1619 | |
1620 | #ifdef CONFIG_SMP |
1621 | /* per-cpu sections scanning */ |
1622 | for_each_possible_cpu(i) |
1623 | scan_large_block(start: __per_cpu_start + per_cpu_offset(i), |
1624 | end: __per_cpu_end + per_cpu_offset(i)); |
1625 | #endif |
1626 | |
1627 | /* |
1628 | * Struct page scanning for each node. |
1629 | */ |
1630 | get_online_mems(); |
1631 | for_each_populated_zone(zone) { |
1632 | unsigned long start_pfn = zone->zone_start_pfn; |
1633 | unsigned long end_pfn = zone_end_pfn(zone); |
1634 | unsigned long pfn; |
1635 | |
1636 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
1637 | struct page *page = pfn_to_online_page(pfn); |
1638 | |
1639 | if (!(pfn & 63)) |
1640 | cond_resched(); |
1641 | |
1642 | if (!page) |
1643 | continue; |
1644 | |
1645 | /* only scan pages belonging to this zone */ |
1646 | if (page_zone(page) != zone) |
1647 | continue; |
1648 | /* only scan if page is in use */ |
1649 | if (page_count(page) == 0) |
1650 | continue; |
1651 | scan_block(start: page, end: page + 1, NULL); |
1652 | } |
1653 | } |
1654 | put_online_mems(); |
1655 | |
1656 | /* |
1657 | * Scanning the task stacks (may introduce false negatives). |
1658 | */ |
1659 | if (kmemleak_stack_scan) { |
1660 | struct task_struct *p, *g; |
1661 | |
1662 | rcu_read_lock(); |
1663 | for_each_process_thread(g, p) { |
1664 | void *stack = try_get_task_stack(tsk: p); |
1665 | if (stack) { |
1666 | scan_block(start: stack, end: stack + THREAD_SIZE, NULL); |
1667 | put_task_stack(tsk: p); |
1668 | } |
1669 | } |
1670 | rcu_read_unlock(); |
1671 | } |
1672 | |
1673 | /* |
1674 | * Scan the objects already referenced from the sections scanned |
1675 | * above. |
1676 | */ |
1677 | scan_gray_list(); |
1678 | |
1679 | /* |
1680 | * Check for new or unreferenced objects modified since the previous |
1681 | * scan and color them gray until the next scan. |
1682 | */ |
1683 | rcu_read_lock(); |
1684 | list_for_each_entry_rcu(object, &object_list, object_list) { |
1685 | if (need_resched()) |
1686 | kmemleak_cond_resched(object); |
1687 | |
1688 | /* |
1689 | * This is racy but we can save the overhead of lock/unlock |
1690 | * calls. The missed objects, if any, should be caught in |
1691 | * the next scan. |
1692 | */ |
1693 | if (!color_white(object)) |
1694 | continue; |
1695 | raw_spin_lock_irq(&object->lock); |
1696 | if (color_white(object) && (object->flags & OBJECT_ALLOCATED) |
1697 | && update_checksum(object) && get_object(object)) { |
1698 | /* color it gray temporarily */ |
1699 | object->count = object->min_count; |
1700 | list_add_tail(new: &object->gray_list, head: &gray_list); |
1701 | } |
1702 | raw_spin_unlock_irq(&object->lock); |
1703 | } |
1704 | rcu_read_unlock(); |
1705 | |
1706 | /* |
1707 | * Re-scan the gray list for modified unreferenced objects. |
1708 | */ |
1709 | scan_gray_list(); |
1710 | |
1711 | /* |
1712 | * If scanning was stopped do not report any new unreferenced objects. |
1713 | */ |
1714 | if (scan_should_stop()) |
1715 | return; |
1716 | |
1717 | /* |
1718 | * Scanning result reporting. |
1719 | */ |
1720 | rcu_read_lock(); |
1721 | list_for_each_entry_rcu(object, &object_list, object_list) { |
1722 | if (need_resched()) |
1723 | kmemleak_cond_resched(object); |
1724 | |
1725 | /* |
1726 | * This is racy but we can save the overhead of lock/unlock |
1727 | * calls. The missed objects, if any, should be caught in |
1728 | * the next scan. |
1729 | */ |
1730 | if (!color_white(object)) |
1731 | continue; |
1732 | raw_spin_lock_irq(&object->lock); |
1733 | if (unreferenced_object(object) && |
1734 | !(object->flags & OBJECT_REPORTED)) { |
1735 | object->flags |= OBJECT_REPORTED; |
1736 | |
1737 | if (kmemleak_verbose) |
1738 | print_unreferenced(NULL, object); |
1739 | |
1740 | new_leaks++; |
1741 | } |
1742 | raw_spin_unlock_irq(&object->lock); |
1743 | } |
1744 | rcu_read_unlock(); |
1745 | |
1746 | if (new_leaks) { |
1747 | kmemleak_found_leaks = true; |
1748 | |
1749 | pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n" , |
1750 | new_leaks); |
1751 | } |
1752 | |
1753 | } |
1754 | |
1755 | /* |
1756 | * Thread function performing automatic memory scanning. Unreferenced objects |
1757 | * at the end of a memory scan are reported but only the first time. |
1758 | */ |
1759 | static int kmemleak_scan_thread(void *arg) |
1760 | { |
1761 | static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN); |
1762 | |
1763 | pr_info("Automatic memory scanning thread started\n" ); |
1764 | set_user_nice(current, nice: 10); |
1765 | |
1766 | /* |
1767 | * Wait before the first scan to allow the system to fully initialize. |
1768 | */ |
1769 | if (first_run) { |
1770 | signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000); |
1771 | first_run = 0; |
1772 | while (timeout && !kthread_should_stop()) |
1773 | timeout = schedule_timeout_interruptible(timeout); |
1774 | } |
1775 | |
1776 | while (!kthread_should_stop()) { |
1777 | signed long timeout = READ_ONCE(jiffies_scan_wait); |
1778 | |
1779 | mutex_lock(&scan_mutex); |
1780 | kmemleak_scan(); |
1781 | mutex_unlock(lock: &scan_mutex); |
1782 | |
1783 | /* wait before the next scan */ |
1784 | while (timeout && !kthread_should_stop()) |
1785 | timeout = schedule_timeout_interruptible(timeout); |
1786 | } |
1787 | |
1788 | pr_info("Automatic memory scanning thread ended\n" ); |
1789 | |
1790 | return 0; |
1791 | } |
1792 | |
1793 | /* |
1794 | * Start the automatic memory scanning thread. This function must be called |
1795 | * with the scan_mutex held. |
1796 | */ |
1797 | static void start_scan_thread(void) |
1798 | { |
1799 | if (scan_thread) |
1800 | return; |
1801 | scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak" ); |
1802 | if (IS_ERR(ptr: scan_thread)) { |
1803 | pr_warn("Failed to create the scan thread\n" ); |
1804 | scan_thread = NULL; |
1805 | } |
1806 | } |
1807 | |
1808 | /* |
1809 | * Stop the automatic memory scanning thread. |
1810 | */ |
1811 | static void stop_scan_thread(void) |
1812 | { |
1813 | if (scan_thread) { |
1814 | kthread_stop(k: scan_thread); |
1815 | scan_thread = NULL; |
1816 | } |
1817 | } |
1818 | |
1819 | /* |
1820 | * Iterate over the object_list and return the first valid object at or after |
1821 | * the required position with its use_count incremented. The function triggers |
1822 | * a memory scanning when the pos argument points to the first position. |
1823 | */ |
1824 | static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos) |
1825 | { |
1826 | struct kmemleak_object *object; |
1827 | loff_t n = *pos; |
1828 | int err; |
1829 | |
1830 | err = mutex_lock_interruptible(&scan_mutex); |
1831 | if (err < 0) |
1832 | return ERR_PTR(error: err); |
1833 | |
1834 | rcu_read_lock(); |
1835 | list_for_each_entry_rcu(object, &object_list, object_list) { |
1836 | if (n-- > 0) |
1837 | continue; |
1838 | if (get_object(object)) |
1839 | goto out; |
1840 | } |
1841 | object = NULL; |
1842 | out: |
1843 | return object; |
1844 | } |
1845 | |
1846 | /* |
1847 | * Return the next object in the object_list. The function decrements the |
1848 | * use_count of the previous object and increases that of the next one. |
1849 | */ |
1850 | static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1851 | { |
1852 | struct kmemleak_object *prev_obj = v; |
1853 | struct kmemleak_object *next_obj = NULL; |
1854 | struct kmemleak_object *obj = prev_obj; |
1855 | |
1856 | ++(*pos); |
1857 | |
1858 | list_for_each_entry_continue_rcu(obj, &object_list, object_list) { |
1859 | if (get_object(object: obj)) { |
1860 | next_obj = obj; |
1861 | break; |
1862 | } |
1863 | } |
1864 | |
1865 | put_object(object: prev_obj); |
1866 | return next_obj; |
1867 | } |
1868 | |
1869 | /* |
1870 | * Decrement the use_count of the last object required, if any. |
1871 | */ |
1872 | static void kmemleak_seq_stop(struct seq_file *seq, void *v) |
1873 | { |
1874 | if (!IS_ERR(ptr: v)) { |
1875 | /* |
1876 | * kmemleak_seq_start may return ERR_PTR if the scan_mutex |
1877 | * waiting was interrupted, so only release it if !IS_ERR. |
1878 | */ |
1879 | rcu_read_unlock(); |
1880 | mutex_unlock(lock: &scan_mutex); |
1881 | if (v) |
1882 | put_object(object: v); |
1883 | } |
1884 | } |
1885 | |
1886 | /* |
1887 | * Print the information for an unreferenced object to the seq file. |
1888 | */ |
1889 | static int kmemleak_seq_show(struct seq_file *seq, void *v) |
1890 | { |
1891 | struct kmemleak_object *object = v; |
1892 | unsigned long flags; |
1893 | |
1894 | raw_spin_lock_irqsave(&object->lock, flags); |
1895 | if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object)) |
1896 | print_unreferenced(seq, object); |
1897 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1898 | return 0; |
1899 | } |
1900 | |
1901 | static const struct seq_operations kmemleak_seq_ops = { |
1902 | .start = kmemleak_seq_start, |
1903 | .next = kmemleak_seq_next, |
1904 | .stop = kmemleak_seq_stop, |
1905 | .show = kmemleak_seq_show, |
1906 | }; |
1907 | |
1908 | static int kmemleak_open(struct inode *inode, struct file *file) |
1909 | { |
1910 | return seq_open(file, &kmemleak_seq_ops); |
1911 | } |
1912 | |
1913 | static int dump_str_object_info(const char *str) |
1914 | { |
1915 | unsigned long flags; |
1916 | struct kmemleak_object *object; |
1917 | unsigned long addr; |
1918 | |
1919 | if (kstrtoul(s: str, base: 0, res: &addr)) |
1920 | return -EINVAL; |
1921 | object = find_and_get_object(ptr: addr, alias: 0); |
1922 | if (!object) { |
1923 | pr_info("Unknown object at 0x%08lx\n" , addr); |
1924 | return -EINVAL; |
1925 | } |
1926 | |
1927 | raw_spin_lock_irqsave(&object->lock, flags); |
1928 | dump_object_info(object); |
1929 | raw_spin_unlock_irqrestore(&object->lock, flags); |
1930 | |
1931 | put_object(object); |
1932 | return 0; |
1933 | } |
1934 | |
1935 | /* |
1936 | * We use grey instead of black to ensure we can do future scans on the same |
1937 | * objects. If we did not do future scans these black objects could |
1938 | * potentially contain references to newly allocated objects in the future and |
1939 | * we'd end up with false positives. |
1940 | */ |
1941 | static void kmemleak_clear(void) |
1942 | { |
1943 | struct kmemleak_object *object; |
1944 | |
1945 | rcu_read_lock(); |
1946 | list_for_each_entry_rcu(object, &object_list, object_list) { |
1947 | raw_spin_lock_irq(&object->lock); |
1948 | if ((object->flags & OBJECT_REPORTED) && |
1949 | unreferenced_object(object)) |
1950 | __paint_it(object, KMEMLEAK_GREY); |
1951 | raw_spin_unlock_irq(&object->lock); |
1952 | } |
1953 | rcu_read_unlock(); |
1954 | |
1955 | kmemleak_found_leaks = false; |
1956 | } |
1957 | |
1958 | static void __kmemleak_do_cleanup(void); |
1959 | |
1960 | /* |
1961 | * File write operation to configure kmemleak at run-time. The following |
1962 | * commands can be written to the /sys/kernel/debug/kmemleak file: |
1963 | * off - disable kmemleak (irreversible) |
1964 | * stack=on - enable the task stacks scanning |
1965 | * stack=off - disable the tasks stacks scanning |
1966 | * scan=on - start the automatic memory scanning thread |
1967 | * scan=off - stop the automatic memory scanning thread |
1968 | * scan=... - set the automatic memory scanning period in seconds (0 to |
1969 | * disable it) |
1970 | * scan - trigger a memory scan |
1971 | * clear - mark all current reported unreferenced kmemleak objects as |
1972 | * grey to ignore printing them, or free all kmemleak objects |
1973 | * if kmemleak has been disabled. |
1974 | * dump=... - dump information about the object found at the given address |
1975 | */ |
1976 | static ssize_t kmemleak_write(struct file *file, const char __user *user_buf, |
1977 | size_t size, loff_t *ppos) |
1978 | { |
1979 | char buf[64]; |
1980 | int buf_size; |
1981 | int ret; |
1982 | |
1983 | buf_size = min(size, (sizeof(buf) - 1)); |
1984 | if (strncpy_from_user(dst: buf, src: user_buf, count: buf_size) < 0) |
1985 | return -EFAULT; |
1986 | buf[buf_size] = 0; |
1987 | |
1988 | ret = mutex_lock_interruptible(&scan_mutex); |
1989 | if (ret < 0) |
1990 | return ret; |
1991 | |
1992 | if (strncmp(buf, "clear" , 5) == 0) { |
1993 | if (kmemleak_enabled) |
1994 | kmemleak_clear(); |
1995 | else |
1996 | __kmemleak_do_cleanup(); |
1997 | goto out; |
1998 | } |
1999 | |
2000 | if (!kmemleak_enabled) { |
2001 | ret = -EPERM; |
2002 | goto out; |
2003 | } |
2004 | |
2005 | if (strncmp(buf, "off" , 3) == 0) |
2006 | kmemleak_disable(); |
2007 | else if (strncmp(buf, "stack=on" , 8) == 0) |
2008 | kmemleak_stack_scan = 1; |
2009 | else if (strncmp(buf, "stack=off" , 9) == 0) |
2010 | kmemleak_stack_scan = 0; |
2011 | else if (strncmp(buf, "scan=on" , 7) == 0) |
2012 | start_scan_thread(); |
2013 | else if (strncmp(buf, "scan=off" , 8) == 0) |
2014 | stop_scan_thread(); |
2015 | else if (strncmp(buf, "scan=" , 5) == 0) { |
2016 | unsigned secs; |
2017 | unsigned long msecs; |
2018 | |
2019 | ret = kstrtouint(s: buf + 5, base: 0, res: &secs); |
2020 | if (ret < 0) |
2021 | goto out; |
2022 | |
2023 | msecs = secs * MSEC_PER_SEC; |
2024 | if (msecs > UINT_MAX) |
2025 | msecs = UINT_MAX; |
2026 | |
2027 | stop_scan_thread(); |
2028 | if (msecs) { |
2029 | WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs)); |
2030 | start_scan_thread(); |
2031 | } |
2032 | } else if (strncmp(buf, "scan" , 4) == 0) |
2033 | kmemleak_scan(); |
2034 | else if (strncmp(buf, "dump=" , 5) == 0) |
2035 | ret = dump_str_object_info(str: buf + 5); |
2036 | else |
2037 | ret = -EINVAL; |
2038 | |
2039 | out: |
2040 | mutex_unlock(lock: &scan_mutex); |
2041 | if (ret < 0) |
2042 | return ret; |
2043 | |
2044 | /* ignore the rest of the buffer, only one command at a time */ |
2045 | *ppos += size; |
2046 | return size; |
2047 | } |
2048 | |
2049 | static const struct file_operations kmemleak_fops = { |
2050 | .owner = THIS_MODULE, |
2051 | .open = kmemleak_open, |
2052 | .read = seq_read, |
2053 | .write = kmemleak_write, |
2054 | .llseek = seq_lseek, |
2055 | .release = seq_release, |
2056 | }; |
2057 | |
2058 | static void __kmemleak_do_cleanup(void) |
2059 | { |
2060 | struct kmemleak_object *object, *tmp; |
2061 | |
2062 | /* |
2063 | * Kmemleak has already been disabled, no need for RCU list traversal |
2064 | * or kmemleak_lock held. |
2065 | */ |
2066 | list_for_each_entry_safe(object, tmp, &object_list, object_list) { |
2067 | __remove_object(object); |
2068 | __delete_object(object); |
2069 | } |
2070 | } |
2071 | |
2072 | /* |
2073 | * Stop the memory scanning thread and free the kmemleak internal objects if |
2074 | * no previous scan thread (otherwise, kmemleak may still have some useful |
2075 | * information on memory leaks). |
2076 | */ |
2077 | static void kmemleak_do_cleanup(struct work_struct *work) |
2078 | { |
2079 | stop_scan_thread(); |
2080 | |
2081 | mutex_lock(&scan_mutex); |
2082 | /* |
2083 | * Once it is made sure that kmemleak_scan has stopped, it is safe to no |
2084 | * longer track object freeing. Ordering of the scan thread stopping and |
2085 | * the memory accesses below is guaranteed by the kthread_stop() |
2086 | * function. |
2087 | */ |
2088 | kmemleak_free_enabled = 0; |
2089 | mutex_unlock(lock: &scan_mutex); |
2090 | |
2091 | if (!kmemleak_found_leaks) |
2092 | __kmemleak_do_cleanup(); |
2093 | else |
2094 | pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n" ); |
2095 | } |
2096 | |
2097 | static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup); |
2098 | |
2099 | /* |
2100 | * Disable kmemleak. No memory allocation/freeing will be traced once this |
2101 | * function is called. Disabling kmemleak is an irreversible operation. |
2102 | */ |
2103 | static void kmemleak_disable(void) |
2104 | { |
2105 | /* atomically check whether it was already invoked */ |
2106 | if (cmpxchg(&kmemleak_error, 0, 1)) |
2107 | return; |
2108 | |
2109 | /* stop any memory operation tracing */ |
2110 | kmemleak_enabled = 0; |
2111 | |
2112 | /* check whether it is too early for a kernel thread */ |
2113 | if (kmemleak_late_initialized) |
2114 | schedule_work(work: &cleanup_work); |
2115 | else |
2116 | kmemleak_free_enabled = 0; |
2117 | |
2118 | pr_info("Kernel memory leak detector disabled\n" ); |
2119 | } |
2120 | |
2121 | /* |
2122 | * Allow boot-time kmemleak disabling (enabled by default). |
2123 | */ |
2124 | static int __init kmemleak_boot_config(char *str) |
2125 | { |
2126 | if (!str) |
2127 | return -EINVAL; |
2128 | if (strcmp(str, "off" ) == 0) |
2129 | kmemleak_disable(); |
2130 | else if (strcmp(str, "on" ) == 0) { |
2131 | kmemleak_skip_disable = 1; |
2132 | stack_depot_request_early_init(); |
2133 | } |
2134 | else |
2135 | return -EINVAL; |
2136 | return 0; |
2137 | } |
2138 | early_param("kmemleak" , kmemleak_boot_config); |
2139 | |
2140 | /* |
2141 | * Kmemleak initialization. |
2142 | */ |
2143 | void __init kmemleak_init(void) |
2144 | { |
2145 | #ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF |
2146 | if (!kmemleak_skip_disable) { |
2147 | kmemleak_disable(); |
2148 | return; |
2149 | } |
2150 | #endif |
2151 | |
2152 | if (kmemleak_error) |
2153 | return; |
2154 | |
2155 | jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE); |
2156 | jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000); |
2157 | |
2158 | object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE); |
2159 | scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE); |
2160 | |
2161 | /* register the data/bss sections */ |
2162 | create_object(ptr: (unsigned long)_sdata, size: _edata - _sdata, |
2163 | KMEMLEAK_GREY, GFP_ATOMIC); |
2164 | create_object(ptr: (unsigned long)__bss_start, size: __bss_stop - __bss_start, |
2165 | KMEMLEAK_GREY, GFP_ATOMIC); |
2166 | /* only register .data..ro_after_init if not within .data */ |
2167 | if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata) |
2168 | create_object(ptr: (unsigned long)__start_ro_after_init, |
2169 | size: __end_ro_after_init - __start_ro_after_init, |
2170 | KMEMLEAK_GREY, GFP_ATOMIC); |
2171 | } |
2172 | |
2173 | /* |
2174 | * Late initialization function. |
2175 | */ |
2176 | static int __init kmemleak_late_init(void) |
2177 | { |
2178 | kmemleak_late_initialized = 1; |
2179 | |
2180 | debugfs_create_file(name: "kmemleak" , mode: 0644, NULL, NULL, fops: &kmemleak_fops); |
2181 | |
2182 | if (kmemleak_error) { |
2183 | /* |
2184 | * Some error occurred and kmemleak was disabled. There is a |
2185 | * small chance that kmemleak_disable() was called immediately |
2186 | * after setting kmemleak_late_initialized and we may end up with |
2187 | * two clean-up threads but serialized by scan_mutex. |
2188 | */ |
2189 | schedule_work(work: &cleanup_work); |
2190 | return -ENOMEM; |
2191 | } |
2192 | |
2193 | if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) { |
2194 | mutex_lock(&scan_mutex); |
2195 | start_scan_thread(); |
2196 | mutex_unlock(lock: &scan_mutex); |
2197 | } |
2198 | |
2199 | pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n" , |
2200 | mem_pool_free_count); |
2201 | |
2202 | return 0; |
2203 | } |
2204 | late_initcall(kmemleak_late_init); |
2205 | |