linux/mm/kmemleak.c
<<
>>
Prefs
   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/export.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 <linux/atomic.h>
 100
 101#include <linux/kmemcheck.h>
 102#include <linux/kmemleak.h>
 103#include <linux/memory_hotplug.h>
 104
 105/*
 106 * Kmemleak configuration and common defines.
 107 */
 108#define MAX_TRACE               16      /* stack trace length */
 109#define MSECS_MIN_AGE           5000    /* minimum object age for reporting */
 110#define SECS_FIRST_SCAN         60      /* delay before the first scan */
 111#define SECS_SCAN_WAIT          600     /* subsequent auto scanning delay */
 112#define MAX_SCAN_SIZE           4096    /* maximum size of a scanned block */
 113
 114#define BYTES_PER_POINTER       sizeof(void *)
 115
 116/* GFP bitmask for kmemleak internal allocations */
 117#define gfp_kmemleak_mask(gfp)  (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
 118                                 __GFP_NORETRY | __GFP_NOMEMALLOC | \
 119                                 __GFP_NOWARN)
 120
 121/* scanning area inside a memory block */
 122struct kmemleak_scan_area {
 123        struct hlist_node node;
 124        unsigned long start;
 125        size_t size;
 126};
 127
 128#define KMEMLEAK_GREY   0
 129#define KMEMLEAK_BLACK  -1
 130
 131/*
 132 * Structure holding the metadata for each allocated memory block.
 133 * Modifications to such objects should be made while holding the
 134 * object->lock. Insertions or deletions from object_list, gray_list or
 135 * tree_node are already protected by the corresponding locks or mutex (see
 136 * the notes on locking above). These objects are reference-counted
 137 * (use_count) and freed using the RCU mechanism.
 138 */
 139struct kmemleak_object {
 140        spinlock_t lock;
 141        unsigned long flags;            /* object status flags */
 142        struct list_head object_list;
 143        struct list_head gray_list;
 144        struct prio_tree_node tree_node;
 145        struct rcu_head rcu;            /* object_list lockless traversal */
 146        /* object usage count; object freed when use_count == 0 */
 147        atomic_t use_count;
 148        unsigned long pointer;
 149        size_t size;
 150        /* minimum number of a pointers found before it is considered leak */
 151        int min_count;
 152        /* the total number of pointers found pointing to this object */
 153        int count;
 154        /* checksum for detecting modified objects */
 155        u32 checksum;
 156        /* memory ranges to be scanned inside an object (empty for all) */
 157        struct hlist_head area_list;
 158        unsigned long trace[MAX_TRACE];
 159        unsigned int trace_len;
 160        unsigned long jiffies;          /* creation timestamp */
 161        pid_t pid;                      /* pid of the current task */
 162        char comm[TASK_COMM_LEN];       /* executable name */
 163};
 164
 165/* flag representing the memory block allocation status */
 166#define OBJECT_ALLOCATED        (1 << 0)
 167/* flag set after the first reporting of an unreference object */
 168#define OBJECT_REPORTED         (1 << 1)
 169/* flag set to not scan the object */
 170#define OBJECT_NO_SCAN          (1 << 2)
 171
 172/* number of bytes to print per line; must be 16 or 32 */
 173#define HEX_ROW_SIZE            16
 174/* number of bytes to print at a time (1, 2, 4, 8) */
 175#define HEX_GROUP_SIZE          1
 176/* include ASCII after the hex output */
 177#define HEX_ASCII               1
 178/* max number of lines to be printed */
 179#define HEX_MAX_LINES           2
 180
 181/* the list of all allocated objects */
 182static LIST_HEAD(object_list);
 183/* the list of gray-colored objects (see color_gray comment below) */
 184static LIST_HEAD(gray_list);
 185/* prio search tree for object boundaries */
 186static struct prio_tree_root object_tree_root;
 187/* rw_lock protecting the access to object_list and prio_tree_root */
 188static DEFINE_RWLOCK(kmemleak_lock);
 189
 190/* allocation caches for kmemleak internal data */
 191static struct kmem_cache *object_cache;
 192static struct kmem_cache *scan_area_cache;
 193
 194/* set if tracing memory operations is enabled */
 195static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
 196/* set in the late_initcall if there were no errors */
 197static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
 198/* enables or disables early logging of the memory operations */
 199static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
 200/* set if a kmemleak warning was issued */
 201static atomic_t kmemleak_warning = ATOMIC_INIT(0);
 202/* set if a fatal kmemleak error has occurred */
 203static atomic_t kmemleak_error = ATOMIC_INIT(0);
 204
 205/* minimum and maximum address that may be valid pointers */
 206static unsigned long min_addr = ULONG_MAX;
 207static unsigned long max_addr;
 208
 209static struct task_struct *scan_thread;
 210/* used to avoid reporting of recently allocated objects */
 211static unsigned long jiffies_min_age;
 212static unsigned long jiffies_last_scan;
 213/* delay between automatic memory scannings */
 214static signed long jiffies_scan_wait;
 215/* enables or disables the task stacks scanning */
 216static int kmemleak_stack_scan = 1;
 217/* protects the memory scanning, parameters and debug/kmemleak file access */
 218static DEFINE_MUTEX(scan_mutex);
 219/* setting kmemleak=on, will set this var, skipping the disable */
 220static int kmemleak_skip_disable;
 221
 222
 223/*
 224 * Early object allocation/freeing logging. Kmemleak is initialized after the
 225 * kernel allocator. However, both the kernel allocator and kmemleak may
 226 * allocate memory blocks which need to be tracked. Kmemleak defines an
 227 * arbitrary buffer to hold the allocation/freeing information before it is
 228 * fully initialized.
 229 */
 230
 231/* kmemleak operation type for early logging */
 232enum {
 233        KMEMLEAK_ALLOC,
 234        KMEMLEAK_ALLOC_PERCPU,
 235        KMEMLEAK_FREE,
 236        KMEMLEAK_FREE_PART,
 237        KMEMLEAK_FREE_PERCPU,
 238        KMEMLEAK_NOT_LEAK,
 239        KMEMLEAK_IGNORE,
 240        KMEMLEAK_SCAN_AREA,
 241        KMEMLEAK_NO_SCAN
 242};
 243
 244/*
 245 * Structure holding the information passed to kmemleak callbacks during the
 246 * early logging.
 247 */
 248struct early_log {
 249        int op_type;                    /* kmemleak operation type */
 250        const void *ptr;                /* allocated/freed memory block */
 251        size_t size;                    /* memory block size */
 252        int min_count;                  /* minimum reference count */
 253        unsigned long trace[MAX_TRACE]; /* stack trace */
 254        unsigned int trace_len;         /* stack trace length */
 255};
 256
 257/* early logging buffer and current position */
 258static struct early_log
 259        early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
 260static int crt_early_log __initdata;
 261
 262static void kmemleak_disable(void);
 263
 264/*
 265 * Print a warning and dump the stack trace.
 266 */
 267#define kmemleak_warn(x...)     do {            \
 268        pr_warning(x);                          \
 269        dump_stack();                           \
 270        atomic_set(&kmemleak_warning, 1);       \
 271} while (0)
 272
 273/*
 274 * Macro invoked when a serious kmemleak condition occurred and cannot be
 275 * recovered from. Kmemleak will be disabled and further allocation/freeing
 276 * tracing no longer available.
 277 */
 278#define kmemleak_stop(x...)     do {    \
 279        kmemleak_warn(x);               \
 280        kmemleak_disable();             \
 281} while (0)
 282
 283/*
 284 * Printing of the objects hex dump to the seq file. The number of lines to be
 285 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
 286 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
 287 * with the object->lock held.
 288 */
 289static void hex_dump_object(struct seq_file *seq,
 290                            struct kmemleak_object *object)
 291{
 292        const u8 *ptr = (const u8 *)object->pointer;
 293        int i, len, remaining;
 294        unsigned char linebuf[HEX_ROW_SIZE * 5];
 295
 296        /* limit the number of lines to HEX_MAX_LINES */
 297        remaining = len =
 298                min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
 299
 300        seq_printf(seq, "  hex dump (first %d bytes):\n", len);
 301        for (i = 0; i < len; i += HEX_ROW_SIZE) {
 302                int linelen = min(remaining, HEX_ROW_SIZE);
 303
 304                remaining -= HEX_ROW_SIZE;
 305                hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
 306                                   HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
 307                                   HEX_ASCII);
 308                seq_printf(seq, "    %s\n", linebuf);
 309        }
 310}
 311
 312/*
 313 * Object colors, encoded with count and min_count:
 314 * - white - orphan object, not enough references to it (count < min_count)
 315 * - gray  - not orphan, not marked as false positive (min_count == 0) or
 316 *              sufficient references to it (count >= min_count)
 317 * - black - ignore, it doesn't contain references (e.g. text section)
 318 *              (min_count == -1). No function defined for this color.
 319 * Newly created objects don't have any color assigned (object->count == -1)
 320 * before the next memory scan when they become white.
 321 */
 322static bool color_white(const struct kmemleak_object *object)
 323{
 324        return object->count != KMEMLEAK_BLACK &&
 325                object->count < object->min_count;
 326}
 327
 328static bool color_gray(const struct kmemleak_object *object)
 329{
 330        return object->min_count != KMEMLEAK_BLACK &&
 331                object->count >= object->min_count;
 332}
 333
 334/*
 335 * Objects are considered unreferenced only if their color is white, they have
 336 * not be deleted and have a minimum age to avoid false positives caused by
 337 * pointers temporarily stored in CPU registers.
 338 */
 339static bool unreferenced_object(struct kmemleak_object *object)
 340{
 341        return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
 342                time_before_eq(object->jiffies + jiffies_min_age,
 343                               jiffies_last_scan);
 344}
 345
 346/*
 347 * Printing of the unreferenced objects information to the seq file. The
 348 * print_unreferenced function must be called with the object->lock held.
 349 */
 350static void print_unreferenced(struct seq_file *seq,
 351                               struct kmemleak_object *object)
 352{
 353        int i;
 354        unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
 355
 356        seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
 357                   object->pointer, object->size);
 358        seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
 359                   object->comm, object->pid, object->jiffies,
 360                   msecs_age / 1000, msecs_age % 1000);
 361        hex_dump_object(seq, object);
 362        seq_printf(seq, "  backtrace:\n");
 363
 364        for (i = 0; i < object->trace_len; i++) {
 365                void *ptr = (void *)object->trace[i];
 366                seq_printf(seq, "    [<%p>] %pS\n", ptr, ptr);
 367        }
 368}
 369
 370/*
 371 * Print the kmemleak_object information. This function is used mainly for
 372 * debugging special cases when kmemleak operations. It must be called with
 373 * the object->lock held.
 374 */
 375static void dump_object_info(struct kmemleak_object *object)
 376{
 377        struct stack_trace trace;
 378
 379        trace.nr_entries = object->trace_len;
 380        trace.entries = object->trace;
 381
 382        pr_notice("Object 0x%08lx (size %zu):\n",
 383                  object->tree_node.start, 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%lx\n", object->flags);
 389        pr_notice("  checksum = %d\n", object->checksum);
 390        pr_notice("  backtrace:\n");
 391        print_stack_trace(&trace, 4);
 392}
 393
 394/*
 395 * Look-up a memory block metadata (kmemleak_object) in the priority search
 396 * tree based on a pointer value. If alias is 0, only values pointing to the
 397 * beginning of the memory block are allowed. The kmemleak_lock must be held
 398 * when calling this function.
 399 */
 400static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
 401{
 402        struct prio_tree_node *node;
 403        struct prio_tree_iter iter;
 404        struct kmemleak_object *object;
 405
 406        prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
 407        node = prio_tree_next(&iter);
 408        if (node) {
 409                object = prio_tree_entry(node, struct kmemleak_object,
 410                                         tree_node);
 411                if (!alias && object->pointer != ptr) {
 412                        kmemleak_warn("Found object by alias at 0x%08lx\n",
 413                                      ptr);
 414                        dump_object_info(object);
 415                        object = NULL;
 416                }
 417        } else
 418                object = NULL;
 419
 420        return object;
 421}
 422
 423/*
 424 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
 425 * that once an object's use_count reached 0, the RCU freeing was already
 426 * registered and the object should no longer be used. This function must be
 427 * called under the protection of rcu_read_lock().
 428 */
 429static int get_object(struct kmemleak_object *object)
 430{
 431        return atomic_inc_not_zero(&object->use_count);
 432}
 433
 434/*
 435 * RCU callback to free a kmemleak_object.
 436 */
 437static void free_object_rcu(struct rcu_head *rcu)
 438{
 439        struct hlist_node *elem, *tmp;
 440        struct kmemleak_scan_area *area;
 441        struct kmemleak_object *object =
 442                container_of(rcu, struct kmemleak_object, rcu);
 443
 444        /*
 445         * Once use_count is 0 (guaranteed by put_object), there is no other
 446         * code accessing this object, hence no need for locking.
 447         */
 448        hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
 449                hlist_del(elem);
 450                kmem_cache_free(scan_area_cache, area);
 451        }
 452        kmem_cache_free(object_cache, object);
 453}
 454
 455/*
 456 * Decrement the object use_count. Once the count is 0, free the object using
 457 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
 458 * delete_object() path, the delayed RCU freeing ensures that there is no
 459 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
 460 * is also possible.
 461 */
 462static void put_object(struct kmemleak_object *object)
 463{
 464        if (!atomic_dec_and_test(&object->use_count))
 465                return;
 466
 467        /* should only get here after delete_object was called */
 468        WARN_ON(object->flags & OBJECT_ALLOCATED);
 469
 470        call_rcu(&object->rcu, free_object_rcu);
 471}
 472
 473/*
 474 * Look up an object in the prio search tree and increase its use_count.
 475 */
 476static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
 477{
 478        unsigned long flags;
 479        struct kmemleak_object *object = NULL;
 480
 481        rcu_read_lock();
 482        read_lock_irqsave(&kmemleak_lock, flags);
 483        if (ptr >= min_addr && ptr < max_addr)
 484                object = lookup_object(ptr, alias);
 485        read_unlock_irqrestore(&kmemleak_lock, flags);
 486
 487        /* check whether the object is still available */
 488        if (object && !get_object(object))
 489                object = NULL;
 490        rcu_read_unlock();
 491
 492        return object;
 493}
 494
 495/*
 496 * Save stack trace to the given array of MAX_TRACE size.
 497 */
 498static int __save_stack_trace(unsigned long *trace)
 499{
 500        struct stack_trace stack_trace;
 501
 502        stack_trace.max_entries = MAX_TRACE;
 503        stack_trace.nr_entries = 0;
 504        stack_trace.entries = trace;
 505        stack_trace.skip = 2;
 506        save_stack_trace(&stack_trace);
 507
 508        return stack_trace.nr_entries;
 509}
 510
 511/*
 512 * Create the metadata (struct kmemleak_object) corresponding to an allocated
 513 * memory block and add it to the object_list and object_tree_root.
 514 */
 515static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
 516                                             int min_count, gfp_t gfp)
 517{
 518        unsigned long flags;
 519        struct kmemleak_object *object;
 520        struct prio_tree_node *node;
 521
 522        object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
 523        if (!object) {
 524                pr_warning("Cannot allocate a kmemleak_object structure\n");
 525                kmemleak_disable();
 526                return NULL;
 527        }
 528
 529        INIT_LIST_HEAD(&object->object_list);
 530        INIT_LIST_HEAD(&object->gray_list);
 531        INIT_HLIST_HEAD(&object->area_list);
 532        spin_lock_init(&object->lock);
 533        atomic_set(&object->use_count, 1);
 534        object->flags = OBJECT_ALLOCATED;
 535        object->pointer = ptr;
 536        object->size = size;
 537        object->min_count = min_count;
 538        object->count = 0;                      /* white color initially */
 539        object->jiffies = jiffies;
 540        object->checksum = 0;
 541
 542        /* task information */
 543        if (in_irq()) {
 544                object->pid = 0;
 545                strncpy(object->comm, "hardirq", sizeof(object->comm));
 546        } else if (in_softirq()) {
 547                object->pid = 0;
 548                strncpy(object->comm, "softirq", sizeof(object->comm));
 549        } else {
 550                object->pid = current->pid;
 551                /*
 552                 * There is a small chance of a race with set_task_comm(),
 553                 * however using get_task_comm() here may cause locking
 554                 * dependency issues with current->alloc_lock. In the worst
 555                 * case, the command line is not correct.
 556                 */
 557                strncpy(object->comm, current->comm, sizeof(object->comm));
 558        }
 559
 560        /* kernel backtrace */
 561        object->trace_len = __save_stack_trace(object->trace);
 562
 563        INIT_PRIO_TREE_NODE(&object->tree_node);
 564        object->tree_node.start = ptr;
 565        object->tree_node.last = ptr + size - 1;
 566
 567        write_lock_irqsave(&kmemleak_lock, flags);
 568
 569        min_addr = min(min_addr, ptr);
 570        max_addr = max(max_addr, ptr + size);
 571        node = prio_tree_insert(&object_tree_root, &object->tree_node);
 572        /*
 573         * The code calling the kernel does not yet have the pointer to the
 574         * memory block to be able to free it.  However, we still hold the
 575         * kmemleak_lock here in case parts of the kernel started freeing
 576         * random memory blocks.
 577         */
 578        if (node != &object->tree_node) {
 579                kmemleak_stop("Cannot insert 0x%lx into the object search tree "
 580                              "(already existing)\n", ptr);
 581                object = lookup_object(ptr, 1);
 582                spin_lock(&object->lock);
 583                dump_object_info(object);
 584                spin_unlock(&object->lock);
 585
 586                goto out;
 587        }
 588        list_add_tail_rcu(&object->object_list, &object_list);
 589out:
 590        write_unlock_irqrestore(&kmemleak_lock, flags);
 591        return object;
 592}
 593
 594/*
 595 * Remove the metadata (struct kmemleak_object) for a memory block from the
 596 * object_list and object_tree_root and decrement its use_count.
 597 */
 598static void __delete_object(struct kmemleak_object *object)
 599{
 600        unsigned long flags;
 601
 602        write_lock_irqsave(&kmemleak_lock, flags);
 603        prio_tree_remove(&object_tree_root, &object->tree_node);
 604        list_del_rcu(&object->object_list);
 605        write_unlock_irqrestore(&kmemleak_lock, flags);
 606
 607        WARN_ON(!(object->flags & OBJECT_ALLOCATED));
 608        WARN_ON(atomic_read(&object->use_count) < 2);
 609
 610        /*
 611         * Locking here also ensures that the corresponding memory block
 612         * cannot be freed when it is being scanned.
 613         */
 614        spin_lock_irqsave(&object->lock, flags);
 615        object->flags &= ~OBJECT_ALLOCATED;
 616        spin_unlock_irqrestore(&object->lock, flags);
 617        put_object(object);
 618}
 619
 620/*
 621 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
 622 * delete it.
 623 */
 624static void delete_object_full(unsigned long ptr)
 625{
 626        struct kmemleak_object *object;
 627
 628        object = find_and_get_object(ptr, 0);
 629        if (!object) {
 630#ifdef DEBUG
 631                kmemleak_warn("Freeing unknown object at 0x%08lx\n",
 632                              ptr);
 633#endif
 634                return;
 635        }
 636        __delete_object(object);
 637        put_object(object);
 638}
 639
 640/*
 641 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
 642 * delete it. If the memory block is partially freed, the function may create
 643 * additional metadata for the remaining parts of the block.
 644 */
 645static void delete_object_part(unsigned long ptr, size_t size)
 646{
 647        struct kmemleak_object *object;
 648        unsigned long start, end;
 649
 650        object = find_and_get_object(ptr, 1);
 651        if (!object) {
 652#ifdef DEBUG
 653                kmemleak_warn("Partially freeing unknown object at 0x%08lx "
 654                              "(size %zu)\n", ptr, size);
 655#endif
 656                return;
 657        }
 658        __delete_object(object);
 659
 660        /*
 661         * Create one or two objects that may result from the memory block
 662         * split. Note that partial freeing is only done by free_bootmem() and
 663         * this happens before kmemleak_init() is called. The path below is
 664         * only executed during early log recording in kmemleak_init(), so
 665         * GFP_KERNEL is enough.
 666         */
 667        start = object->pointer;
 668        end = object->pointer + object->size;
 669        if (ptr > start)
 670                create_object(start, ptr - start, object->min_count,
 671                              GFP_KERNEL);
 672        if (ptr + size < end)
 673                create_object(ptr + size, end - ptr - size, object->min_count,
 674                              GFP_KERNEL);
 675
 676        put_object(object);
 677}
 678
 679static void __paint_it(struct kmemleak_object *object, int color)
 680{
 681        object->min_count = color;
 682        if (color == KMEMLEAK_BLACK)
 683                object->flags |= OBJECT_NO_SCAN;
 684}
 685
 686static void paint_it(struct kmemleak_object *object, int color)
 687{
 688        unsigned long flags;
 689
 690        spin_lock_irqsave(&object->lock, flags);
 691        __paint_it(object, color);
 692        spin_unlock_irqrestore(&object->lock, flags);
 693}
 694
 695static void paint_ptr(unsigned long ptr, int color)
 696{
 697        struct kmemleak_object *object;
 698
 699        object = find_and_get_object(ptr, 0);
 700        if (!object) {
 701                kmemleak_warn("Trying to color unknown object "
 702                              "at 0x%08lx as %s\n", ptr,
 703                              (color == KMEMLEAK_GREY) ? "Grey" :
 704                              (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
 705                return;
 706        }
 707        paint_it(object, color);
 708        put_object(object);
 709}
 710
 711/*
 712 * Mark an object permanently as gray-colored so that it can no longer be
 713 * reported as a leak. This is used in general to mark a false positive.
 714 */
 715static void make_gray_object(unsigned long ptr)
 716{
 717        paint_ptr(ptr, KMEMLEAK_GREY);
 718}
 719
 720/*
 721 * Mark the object as black-colored so that it is ignored from scans and
 722 * reporting.
 723 */
 724static void make_black_object(unsigned long ptr)
 725{
 726        paint_ptr(ptr, KMEMLEAK_BLACK);
 727}
 728
 729/*
 730 * Add a scanning area to the object. If at least one such area is added,
 731 * kmemleak will only scan these ranges rather than the whole memory block.
 732 */
 733static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
 734{
 735        unsigned long flags;
 736        struct kmemleak_object *object;
 737        struct kmemleak_scan_area *area;
 738
 739        object = find_and_get_object(ptr, 1);
 740        if (!object) {
 741                kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
 742                              ptr);
 743                return;
 744        }
 745
 746        area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
 747        if (!area) {
 748                pr_warning("Cannot allocate a scan area\n");
 749                goto out;
 750        }
 751
 752        spin_lock_irqsave(&object->lock, flags);
 753        if (ptr + size > object->pointer + object->size) {
 754                kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
 755                dump_object_info(object);
 756                kmem_cache_free(scan_area_cache, area);
 757                goto out_unlock;
 758        }
 759
 760        INIT_HLIST_NODE(&area->node);
 761        area->start = ptr;
 762        area->size = size;
 763
 764        hlist_add_head(&area->node, &object->area_list);
 765out_unlock:
 766        spin_unlock_irqrestore(&object->lock, flags);
 767out:
 768        put_object(object);
 769}
 770
 771/*
 772 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
 773 * pointer. Such object will not be scanned by kmemleak but references to it
 774 * are searched.
 775 */
 776static void object_no_scan(unsigned long ptr)
 777{
 778        unsigned long flags;
 779        struct kmemleak_object *object;
 780
 781        object = find_and_get_object(ptr, 0);
 782        if (!object) {
 783                kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
 784                return;
 785        }
 786
 787        spin_lock_irqsave(&object->lock, flags);
 788        object->flags |= OBJECT_NO_SCAN;
 789        spin_unlock_irqrestore(&object->lock, flags);
 790        put_object(object);
 791}
 792
 793/*
 794 * Log an early kmemleak_* call to the early_log buffer. These calls will be
 795 * processed later once kmemleak is fully initialized.
 796 */
 797static void __init log_early(int op_type, const void *ptr, size_t size,
 798                             int min_count)
 799{
 800        unsigned long flags;
 801        struct early_log *log;
 802
 803        if (atomic_read(&kmemleak_error)) {
 804                /* kmemleak stopped recording, just count the requests */
 805                crt_early_log++;
 806                return;
 807        }
 808
 809        if (crt_early_log >= ARRAY_SIZE(early_log)) {
 810                kmemleak_disable();
 811                return;
 812        }
 813
 814        /*
 815         * There is no need for locking since the kernel is still in UP mode
 816         * at this stage. Disabling the IRQs is enough.
 817         */
 818        local_irq_save(flags);
 819        log = &early_log[crt_early_log];
 820        log->op_type = op_type;
 821        log->ptr = ptr;
 822        log->size = size;
 823        log->min_count = min_count;
 824        log->trace_len = __save_stack_trace(log->trace);
 825        crt_early_log++;
 826        local_irq_restore(flags);
 827}
 828
 829/*
 830 * Log an early allocated block and populate the stack trace.
 831 */
 832static void early_alloc(struct early_log *log)
 833{
 834        struct kmemleak_object *object;
 835        unsigned long flags;
 836        int i;
 837
 838        if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
 839                return;
 840
 841        /*
 842         * RCU locking needed to ensure object is not freed via put_object().
 843         */
 844        rcu_read_lock();
 845        object = create_object((unsigned long)log->ptr, log->size,
 846                               log->min_count, GFP_ATOMIC);
 847        if (!object)
 848                goto out;
 849        spin_lock_irqsave(&object->lock, flags);
 850        for (i = 0; i < log->trace_len; i++)
 851                object->trace[i] = log->trace[i];
 852        object->trace_len = log->trace_len;
 853        spin_unlock_irqrestore(&object->lock, flags);
 854out:
 855        rcu_read_unlock();
 856}
 857
 858/*
 859 * Log an early allocated block and populate the stack trace.
 860 */
 861static void early_alloc_percpu(struct early_log *log)
 862{
 863        unsigned int cpu;
 864        const void __percpu *ptr = log->ptr;
 865
 866        for_each_possible_cpu(cpu) {
 867                log->ptr = per_cpu_ptr(ptr, cpu);
 868                early_alloc(log);
 869        }
 870}
 871
 872/**
 873 * kmemleak_alloc - register a newly allocated object
 874 * @ptr:        pointer to beginning of the object
 875 * @size:       size of the object
 876 * @min_count:  minimum number of references to this object. If during memory
 877 *              scanning a number of references less than @min_count is found,
 878 *              the object is reported as a memory leak. If @min_count is 0,
 879 *              the object is never reported as a leak. If @min_count is -1,
 880 *              the object is ignored (not scanned and not reported as a leak)
 881 * @gfp:        kmalloc() flags used for kmemleak internal memory allocations
 882 *
 883 * This function is called from the kernel allocators when a new object
 884 * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.).
 885 */
 886void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
 887                          gfp_t gfp)
 888{
 889        pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
 890
 891        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 892                create_object((unsigned long)ptr, size, min_count, gfp);
 893        else if (atomic_read(&kmemleak_early_log))
 894                log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
 895}
 896EXPORT_SYMBOL_GPL(kmemleak_alloc);
 897
 898/**
 899 * kmemleak_alloc_percpu - register a newly allocated __percpu object
 900 * @ptr:        __percpu pointer to beginning of the object
 901 * @size:       size of the object
 902 *
 903 * This function is called from the kernel percpu allocator when a new object
 904 * (memory block) is allocated (alloc_percpu). It assumes GFP_KERNEL
 905 * allocation.
 906 */
 907void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size)
 908{
 909        unsigned int cpu;
 910
 911        pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
 912
 913        /*
 914         * Percpu allocations are only scanned and not reported as leaks
 915         * (min_count is set to 0).
 916         */
 917        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 918                for_each_possible_cpu(cpu)
 919                        create_object((unsigned long)per_cpu_ptr(ptr, cpu),
 920                                      size, 0, GFP_KERNEL);
 921        else if (atomic_read(&kmemleak_early_log))
 922                log_early(KMEMLEAK_ALLOC_PERCPU, ptr, size, 0);
 923}
 924EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
 925
 926/**
 927 * kmemleak_free - unregister a previously registered object
 928 * @ptr:        pointer to beginning of the object
 929 *
 930 * This function is called from the kernel allocators when an object (memory
 931 * block) is freed (kmem_cache_free, kfree, vfree etc.).
 932 */
 933void __ref kmemleak_free(const void *ptr)
 934{
 935        pr_debug("%s(0x%p)\n", __func__, ptr);
 936
 937        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 938                delete_object_full((unsigned long)ptr);
 939        else if (atomic_read(&kmemleak_early_log))
 940                log_early(KMEMLEAK_FREE, ptr, 0, 0);
 941}
 942EXPORT_SYMBOL_GPL(kmemleak_free);
 943
 944/**
 945 * kmemleak_free_part - partially unregister a previously registered object
 946 * @ptr:        pointer to the beginning or inside the object. This also
 947 *              represents the start of the range to be freed
 948 * @size:       size to be unregistered
 949 *
 950 * This function is called when only a part of a memory block is freed
 951 * (usually from the bootmem allocator).
 952 */
 953void __ref kmemleak_free_part(const void *ptr, size_t size)
 954{
 955        pr_debug("%s(0x%p)\n", __func__, ptr);
 956
 957        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 958                delete_object_part((unsigned long)ptr, size);
 959        else if (atomic_read(&kmemleak_early_log))
 960                log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
 961}
 962EXPORT_SYMBOL_GPL(kmemleak_free_part);
 963
 964/**
 965 * kmemleak_free_percpu - unregister a previously registered __percpu object
 966 * @ptr:        __percpu pointer to beginning of the object
 967 *
 968 * This function is called from the kernel percpu allocator when an object
 969 * (memory block) is freed (free_percpu).
 970 */
 971void __ref kmemleak_free_percpu(const void __percpu *ptr)
 972{
 973        unsigned int cpu;
 974
 975        pr_debug("%s(0x%p)\n", __func__, ptr);
 976
 977        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 978                for_each_possible_cpu(cpu)
 979                        delete_object_full((unsigned long)per_cpu_ptr(ptr,
 980                                                                      cpu));
 981        else if (atomic_read(&kmemleak_early_log))
 982                log_early(KMEMLEAK_FREE_PERCPU, ptr, 0, 0);
 983}
 984EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
 985
 986/**
 987 * kmemleak_not_leak - mark an allocated object as false positive
 988 * @ptr:        pointer to beginning of the object
 989 *
 990 * Calling this function on an object will cause the memory block to no longer
 991 * be reported as leak and always be scanned.
 992 */
 993void __ref kmemleak_not_leak(const void *ptr)
 994{
 995        pr_debug("%s(0x%p)\n", __func__, ptr);
 996
 997        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 998                make_gray_object((unsigned long)ptr);
 999        else if (atomic_read(&kmemleak_early_log))
1000                log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
1001}
1002EXPORT_SYMBOL(kmemleak_not_leak);
1003
1004/**
1005 * kmemleak_ignore - ignore an allocated object
1006 * @ptr:        pointer to beginning of the object
1007 *
1008 * Calling this function on an object will cause the memory block to be
1009 * ignored (not scanned and not reported as a leak). This is usually done when
1010 * it is known that the corresponding block is not a leak and does not contain
1011 * any references to other allocated memory blocks.
1012 */
1013void __ref kmemleak_ignore(const void *ptr)
1014{
1015        pr_debug("%s(0x%p)\n", __func__, ptr);
1016
1017        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
1018                make_black_object((unsigned long)ptr);
1019        else if (atomic_read(&kmemleak_early_log))
1020                log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
1021}
1022EXPORT_SYMBOL(kmemleak_ignore);
1023
1024/**
1025 * kmemleak_scan_area - limit the range to be scanned in an allocated object
1026 * @ptr:        pointer to beginning or inside the object. This also
1027 *              represents the start of the scan area
1028 * @size:       size of the scan area
1029 * @gfp:        kmalloc() flags used for kmemleak internal memory allocations
1030 *
1031 * This function is used when it is known that only certain parts of an object
1032 * contain references to other objects. Kmemleak will only scan these areas
1033 * reducing the number false negatives.
1034 */
1035void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
1036{
1037        pr_debug("%s(0x%p)\n", __func__, ptr);
1038
1039        if (atomic_read(&kmemleak_enabled) && ptr && size && !IS_ERR(ptr))
1040                add_scan_area((unsigned long)ptr, size, gfp);
1041        else if (atomic_read(&kmemleak_early_log))
1042                log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
1043}
1044EXPORT_SYMBOL(kmemleak_scan_area);
1045
1046/**
1047 * kmemleak_no_scan - do not scan an allocated object
1048 * @ptr:        pointer to beginning of the object
1049 *
1050 * This function notifies kmemleak not to scan the given memory block. Useful
1051 * in situations where it is known that the given object does not contain any
1052 * references to other objects. Kmemleak will not scan such objects reducing
1053 * the number of false negatives.
1054 */
1055void __ref kmemleak_no_scan(const void *ptr)
1056{
1057        pr_debug("%s(0x%p)\n", __func__, ptr);
1058
1059        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
1060                object_no_scan((unsigned long)ptr);
1061        else if (atomic_read(&kmemleak_early_log))
1062                log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
1063}
1064EXPORT_SYMBOL(kmemleak_no_scan);
1065
1066/*
1067 * Update an object's checksum and return true if it was modified.
1068 */
1069static bool update_checksum(struct kmemleak_object *object)
1070{
1071        u32 old_csum = object->checksum;
1072
1073        if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
1074                return false;
1075
1076        object->checksum = crc32(0, (void *)object->pointer, object->size);
1077        return object->checksum != old_csum;
1078}
1079
1080/*
1081 * Memory scanning is a long process and it needs to be interruptable. This
1082 * function checks whether such interrupt condition occurred.
1083 */
1084static int scan_should_stop(void)
1085{
1086        if (!atomic_read(&kmemleak_enabled))
1087                return 1;
1088
1089        /*
1090         * This function may be called from either process or kthread context,
1091         * hence the need to check for both stop conditions.
1092         */
1093        if (current->mm)
1094                return signal_pending(current);
1095        else
1096                return kthread_should_stop();
1097
1098        return 0;
1099}
1100
1101/*
1102 * Scan a memory block (exclusive range) for valid pointers and add those
1103 * found to the gray list.
1104 */
1105static void scan_block(void *_start, void *_end,
1106                       struct kmemleak_object *scanned, int allow_resched)
1107{
1108        unsigned long *ptr;
1109        unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1110        unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1111
1112        for (ptr = start; ptr < end; ptr++) {
1113                struct kmemleak_object *object;
1114                unsigned long flags;
1115                unsigned long pointer;
1116
1117                if (allow_resched)
1118                        cond_resched();
1119                if (scan_should_stop())
1120                        break;
1121
1122                /* don't scan uninitialized memory */
1123                if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
1124                                                  BYTES_PER_POINTER))
1125                        continue;
1126
1127                pointer = *ptr;
1128
1129                object = find_and_get_object(pointer, 1);
1130                if (!object)
1131                        continue;
1132                if (object == scanned) {
1133                        /* self referenced, ignore */
1134                        put_object(object);
1135                        continue;
1136                }
1137
1138                /*
1139                 * Avoid the lockdep recursive warning on object->lock being
1140                 * previously acquired in scan_object(). These locks are
1141                 * enclosed by scan_mutex.
1142                 */
1143                spin_lock_irqsave_nested(&object->lock, flags,
1144                                         SINGLE_DEPTH_NESTING);
1145                if (!color_white(object)) {
1146                        /* non-orphan, ignored or new */
1147                        spin_unlock_irqrestore(&object->lock, flags);
1148                        put_object(object);
1149                        continue;
1150                }
1151
1152                /*
1153                 * Increase the object's reference count (number of pointers
1154                 * to the memory block). If this count reaches the required
1155                 * minimum, the object's color will become gray and it will be
1156                 * added to the gray_list.
1157                 */
1158                object->count++;
1159                if (color_gray(object)) {
1160                        list_add_tail(&object->gray_list, &gray_list);
1161                        spin_unlock_irqrestore(&object->lock, flags);
1162                        continue;
1163                }
1164
1165                spin_unlock_irqrestore(&object->lock, flags);
1166                put_object(object);
1167        }
1168}
1169
1170/*
1171 * Scan a memory block corresponding to a kmemleak_object. A condition is
1172 * that object->use_count >= 1.
1173 */
1174static void scan_object(struct kmemleak_object *object)
1175{
1176        struct kmemleak_scan_area *area;
1177        struct hlist_node *elem;
1178        unsigned long flags;
1179
1180        /*
1181         * Once the object->lock is acquired, the corresponding memory block
1182         * cannot be freed (the same lock is acquired in delete_object).
1183         */
1184        spin_lock_irqsave(&object->lock, flags);
1185        if (object->flags & OBJECT_NO_SCAN)
1186                goto out;
1187        if (!(object->flags & OBJECT_ALLOCATED))
1188                /* already freed object */
1189                goto out;
1190        if (hlist_empty(&object->area_list)) {
1191                void *start = (void *)object->pointer;
1192                void *end = (void *)(object->pointer + object->size);
1193
1194                while (start < end && (object->flags & OBJECT_ALLOCATED) &&
1195                       !(object->flags & OBJECT_NO_SCAN)) {
1196                        scan_block(start, min(start + MAX_SCAN_SIZE, end),
1197                                   object, 0);
1198                        start += MAX_SCAN_SIZE;
1199
1200                        spin_unlock_irqrestore(&object->lock, flags);
1201                        cond_resched();
1202                        spin_lock_irqsave(&object->lock, flags);
1203                }
1204        } else
1205                hlist_for_each_entry(area, elem, &object->area_list, node)
1206                        scan_block((void *)area->start,
1207                                   (void *)(area->start + area->size),
1208                                   object, 0);
1209out:
1210        spin_unlock_irqrestore(&object->lock, flags);
1211}
1212
1213/*
1214 * Scan the objects already referenced (gray objects). More objects will be
1215 * referenced and, if there are no memory leaks, all the objects are scanned.
1216 */
1217static void scan_gray_list(void)
1218{
1219        struct kmemleak_object *object, *tmp;
1220
1221        /*
1222         * The list traversal is safe for both tail additions and removals
1223         * from inside the loop. The kmemleak objects cannot be freed from
1224         * outside the loop because their use_count was incremented.
1225         */
1226        object = list_entry(gray_list.next, typeof(*object), gray_list);
1227        while (&object->gray_list != &gray_list) {
1228                cond_resched();
1229
1230                /* may add new objects to the list */
1231                if (!scan_should_stop())
1232                        scan_object(object);
1233
1234                tmp = list_entry(object->gray_list.next, typeof(*object),
1235                                 gray_list);
1236
1237                /* remove the object from the list and release it */
1238                list_del(&object->gray_list);
1239                put_object(object);
1240
1241                object = tmp;
1242        }
1243        WARN_ON(!list_empty(&gray_list));
1244}
1245
1246/*
1247 * Scan data sections and all the referenced memory blocks allocated via the
1248 * kernel's standard allocators. This function must be called with the
1249 * scan_mutex held.
1250 */
1251static void kmemleak_scan(void)
1252{
1253        unsigned long flags;
1254        struct kmemleak_object *object;
1255        int i;
1256        int new_leaks = 0;
1257
1258        jiffies_last_scan = jiffies;
1259
1260        /* prepare the kmemleak_object's */
1261        rcu_read_lock();
1262        list_for_each_entry_rcu(object, &object_list, object_list) {
1263                spin_lock_irqsave(&object->lock, flags);
1264#ifdef DEBUG
1265                /*
1266                 * With a few exceptions there should be a maximum of
1267                 * 1 reference to any object at this point.
1268                 */
1269                if (atomic_read(&object->use_count) > 1) {
1270                        pr_debug("object->use_count = %d\n",
1271                                 atomic_read(&object->use_count));
1272                        dump_object_info(object);
1273                }
1274#endif
1275                /* reset the reference count (whiten the object) */
1276                object->count = 0;
1277                if (color_gray(object) && get_object(object))
1278                        list_add_tail(&object->gray_list, &gray_list);
1279
1280                spin_unlock_irqrestore(&object->lock, flags);
1281        }
1282        rcu_read_unlock();
1283
1284        /* data/bss scanning */
1285        scan_block(_sdata, _edata, NULL, 1);
1286        scan_block(__bss_start, __bss_stop, NULL, 1);
1287
1288#ifdef CONFIG_SMP
1289        /* per-cpu sections scanning */
1290        for_each_possible_cpu(i)
1291                scan_block(__per_cpu_start + per_cpu_offset(i),
1292                           __per_cpu_end + per_cpu_offset(i), NULL, 1);
1293#endif
1294
1295        /*
1296         * Struct page scanning for each node.
1297         */
1298        lock_memory_hotplug();
1299        for_each_online_node(i) {
1300                pg_data_t *pgdat = NODE_DATA(i);
1301                unsigned long start_pfn = pgdat->node_start_pfn;
1302                unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1303                unsigned long pfn;
1304
1305                for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1306                        struct page *page;
1307
1308                        if (!pfn_valid(pfn))
1309                                continue;
1310                        page = pfn_to_page(pfn);
1311                        /* only scan if page is in use */
1312                        if (page_count(page) == 0)
1313                                continue;
1314                        scan_block(page, page + 1, NULL, 1);
1315                }
1316        }
1317        unlock_memory_hotplug();
1318
1319        /*
1320         * Scanning the task stacks (may introduce false negatives).
1321         */
1322        if (kmemleak_stack_scan) {
1323                struct task_struct *p, *g;
1324
1325                read_lock(&tasklist_lock);
1326                do_each_thread(g, p) {
1327                        scan_block(task_stack_page(p), task_stack_page(p) +
1328                                   THREAD_SIZE, NULL, 0);
1329                } while_each_thread(g, p);
1330                read_unlock(&tasklist_lock);
1331        }
1332
1333        /*
1334         * Scan the objects already referenced from the sections scanned
1335         * above.
1336         */
1337        scan_gray_list();
1338
1339        /*
1340         * Check for new or unreferenced objects modified since the previous
1341         * scan and color them gray until the next scan.
1342         */
1343        rcu_read_lock();
1344        list_for_each_entry_rcu(object, &object_list, object_list) {
1345                spin_lock_irqsave(&object->lock, flags);
1346                if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1347                    && update_checksum(object) && get_object(object)) {
1348                        /* color it gray temporarily */
1349                        object->count = object->min_count;
1350                        list_add_tail(&object->gray_list, &gray_list);
1351                }
1352                spin_unlock_irqrestore(&object->lock, flags);
1353        }
1354        rcu_read_unlock();
1355
1356        /*
1357         * Re-scan the gray list for modified unreferenced objects.
1358         */
1359        scan_gray_list();
1360
1361        /*
1362         * If scanning was stopped do not report any new unreferenced objects.
1363         */
1364        if (scan_should_stop())
1365                return;
1366
1367        /*
1368         * Scanning result reporting.
1369         */
1370        rcu_read_lock();
1371        list_for_each_entry_rcu(object, &object_list, object_list) {
1372                spin_lock_irqsave(&object->lock, flags);
1373                if (unreferenced_object(object) &&
1374                    !(object->flags & OBJECT_REPORTED)) {
1375                        object->flags |= OBJECT_REPORTED;
1376                        new_leaks++;
1377                }
1378                spin_unlock_irqrestore(&object->lock, flags);
1379        }
1380        rcu_read_unlock();
1381
1382        if (new_leaks)
1383                pr_info("%d new suspected memory leaks (see "
1384                        "/sys/kernel/debug/kmemleak)\n", new_leaks);
1385
1386}
1387
1388/*
1389 * Thread function performing automatic memory scanning. Unreferenced objects
1390 * at the end of a memory scan are reported but only the first time.
1391 */
1392static int kmemleak_scan_thread(void *arg)
1393{
1394        static int first_run = 1;
1395
1396        pr_info("Automatic memory scanning thread started\n");
1397        set_user_nice(current, 10);
1398
1399        /*
1400         * Wait before the first scan to allow the system to fully initialize.
1401         */
1402        if (first_run) {
1403                first_run = 0;
1404                ssleep(SECS_FIRST_SCAN);
1405        }
1406
1407        while (!kthread_should_stop()) {
1408                signed long timeout = jiffies_scan_wait;
1409
1410                mutex_lock(&scan_mutex);
1411                kmemleak_scan();
1412                mutex_unlock(&scan_mutex);
1413
1414                /* wait before the next scan */
1415                while (timeout && !kthread_should_stop())
1416                        timeout = schedule_timeout_interruptible(timeout);
1417        }
1418
1419        pr_info("Automatic memory scanning thread ended\n");
1420
1421        return 0;
1422}
1423
1424/*
1425 * Start the automatic memory scanning thread. This function must be called
1426 * with the scan_mutex held.
1427 */
1428static void start_scan_thread(void)
1429{
1430        if (scan_thread)
1431                return;
1432        scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1433        if (IS_ERR(scan_thread)) {
1434                pr_warning("Failed to create the scan thread\n");
1435                scan_thread = NULL;
1436        }
1437}
1438
1439/*
1440 * Stop the automatic memory scanning thread. This function must be called
1441 * with the scan_mutex held.
1442 */
1443static void stop_scan_thread(void)
1444{
1445        if (scan_thread) {
1446                kthread_stop(scan_thread);
1447                scan_thread = NULL;
1448        }
1449}
1450
1451/*
1452 * Iterate over the object_list and return the first valid object at or after
1453 * the required position with its use_count incremented. The function triggers
1454 * a memory scanning when the pos argument points to the first position.
1455 */
1456static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1457{
1458        struct kmemleak_object *object;
1459        loff_t n = *pos;
1460        int err;
1461
1462        err = mutex_lock_interruptible(&scan_mutex);
1463        if (err < 0)
1464                return ERR_PTR(err);
1465
1466        rcu_read_lock();
1467        list_for_each_entry_rcu(object, &object_list, object_list) {
1468                if (n-- > 0)
1469                        continue;
1470                if (get_object(object))
1471                        goto out;
1472        }
1473        object = NULL;
1474out:
1475        return object;
1476}
1477
1478/*
1479 * Return the next object in the object_list. The function decrements the
1480 * use_count of the previous object and increases that of the next one.
1481 */
1482static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1483{
1484        struct kmemleak_object *prev_obj = v;
1485        struct kmemleak_object *next_obj = NULL;
1486        struct list_head *n = &prev_obj->object_list;
1487
1488        ++(*pos);
1489
1490        list_for_each_continue_rcu(n, &object_list) {
1491                struct kmemleak_object *obj =
1492                        list_entry(n, struct kmemleak_object, object_list);
1493                if (get_object(obj)) {
1494                        next_obj = obj;
1495                        break;
1496                }
1497        }
1498
1499        put_object(prev_obj);
1500        return next_obj;
1501}
1502
1503/*
1504 * Decrement the use_count of the last object required, if any.
1505 */
1506static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1507{
1508        if (!IS_ERR(v)) {
1509                /*
1510                 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1511                 * waiting was interrupted, so only release it if !IS_ERR.
1512                 */
1513                rcu_read_unlock();
1514                mutex_unlock(&scan_mutex);
1515                if (v)
1516                        put_object(v);
1517        }
1518}
1519
1520/*
1521 * Print the information for an unreferenced object to the seq file.
1522 */
1523static int kmemleak_seq_show(struct seq_file *seq, void *v)
1524{
1525        struct kmemleak_object *object = v;
1526        unsigned long flags;
1527
1528        spin_lock_irqsave(&object->lock, flags);
1529        if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1530                print_unreferenced(seq, object);
1531        spin_unlock_irqrestore(&object->lock, flags);
1532        return 0;
1533}
1534
1535static const struct seq_operations kmemleak_seq_ops = {
1536        .start = kmemleak_seq_start,
1537        .next  = kmemleak_seq_next,
1538        .stop  = kmemleak_seq_stop,
1539        .show  = kmemleak_seq_show,
1540};
1541
1542static int kmemleak_open(struct inode *inode, struct file *file)
1543{
1544        return seq_open(file, &kmemleak_seq_ops);
1545}
1546
1547static int kmemleak_release(struct inode *inode, struct file *file)
1548{
1549        return seq_release(inode, file);
1550}
1551
1552static int dump_str_object_info(const char *str)
1553{
1554        unsigned long flags;
1555        struct kmemleak_object *object;
1556        unsigned long addr;
1557
1558        addr= simple_strtoul(str, NULL, 0);
1559        object = find_and_get_object(addr, 0);
1560        if (!object) {
1561                pr_info("Unknown object at 0x%08lx\n", addr);
1562                return -EINVAL;
1563        }
1564
1565        spin_lock_irqsave(&object->lock, flags);
1566        dump_object_info(object);
1567        spin_unlock_irqrestore(&object->lock, flags);
1568
1569        put_object(object);
1570        return 0;
1571}
1572
1573/*
1574 * We use grey instead of black to ensure we can do future scans on the same
1575 * objects. If we did not do future scans these black objects could
1576 * potentially contain references to newly allocated objects in the future and
1577 * we'd end up with false positives.
1578 */
1579static void kmemleak_clear(void)
1580{
1581        struct kmemleak_object *object;
1582        unsigned long flags;
1583
1584        rcu_read_lock();
1585        list_for_each_entry_rcu(object, &object_list, object_list) {
1586                spin_lock_irqsave(&object->lock, flags);
1587                if ((object->flags & OBJECT_REPORTED) &&
1588                    unreferenced_object(object))
1589                        __paint_it(object, KMEMLEAK_GREY);
1590                spin_unlock_irqrestore(&object->lock, flags);
1591        }
1592        rcu_read_unlock();
1593}
1594
1595/*
1596 * File write operation to configure kmemleak at run-time. The following
1597 * commands can be written to the /sys/kernel/debug/kmemleak file:
1598 *   off        - disable kmemleak (irreversible)
1599 *   stack=on   - enable the task stacks scanning
1600 *   stack=off  - disable the tasks stacks scanning
1601 *   scan=on    - start the automatic memory scanning thread
1602 *   scan=off   - stop the automatic memory scanning thread
1603 *   scan=...   - set the automatic memory scanning period in seconds (0 to
1604 *                disable it)
1605 *   scan       - trigger a memory scan
1606 *   clear      - mark all current reported unreferenced kmemleak objects as
1607 *                grey to ignore printing them
1608 *   dump=...   - dump information about the object found at the given address
1609 */
1610static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1611                              size_t size, loff_t *ppos)
1612{
1613        char buf[64];
1614        int buf_size;
1615        int ret;
1616
1617        if (!atomic_read(&kmemleak_enabled))
1618                return -EBUSY;
1619
1620        buf_size = min(size, (sizeof(buf) - 1));
1621        if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1622                return -EFAULT;
1623        buf[buf_size] = 0;
1624
1625        ret = mutex_lock_interruptible(&scan_mutex);
1626        if (ret < 0)
1627                return ret;
1628
1629        if (strncmp(buf, "off", 3) == 0)
1630                kmemleak_disable();
1631        else if (strncmp(buf, "stack=on", 8) == 0)
1632                kmemleak_stack_scan = 1;
1633        else if (strncmp(buf, "stack=off", 9) == 0)
1634                kmemleak_stack_scan = 0;
1635        else if (strncmp(buf, "scan=on", 7) == 0)
1636                start_scan_thread();
1637        else if (strncmp(buf, "scan=off", 8) == 0)
1638                stop_scan_thread();
1639        else if (strncmp(buf, "scan=", 5) == 0) {
1640                unsigned long secs;
1641
1642                ret = strict_strtoul(buf + 5, 0, &secs);
1643                if (ret < 0)
1644                        goto out;
1645                stop_scan_thread();
1646                if (secs) {
1647                        jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1648                        start_scan_thread();
1649                }
1650        } else if (strncmp(buf, "scan", 4) == 0)
1651                kmemleak_scan();
1652        else if (strncmp(buf, "clear", 5) == 0)
1653                kmemleak_clear();
1654        else if (strncmp(buf, "dump=", 5) == 0)
1655                ret = dump_str_object_info(buf + 5);
1656        else
1657                ret = -EINVAL;
1658
1659out:
1660        mutex_unlock(&scan_mutex);
1661        if (ret < 0)
1662                return ret;
1663
1664        /* ignore the rest of the buffer, only one command at a time */
1665        *ppos += size;
1666        return size;
1667}
1668
1669static const struct file_operations kmemleak_fops = {
1670        .owner          = THIS_MODULE,
1671        .open           = kmemleak_open,
1672        .read           = seq_read,
1673        .write          = kmemleak_write,
1674        .llseek         = seq_lseek,
1675        .release        = kmemleak_release,
1676};
1677
1678/*
1679 * Stop the memory scanning thread and free the kmemleak internal objects if
1680 * no previous scan thread (otherwise, kmemleak may still have some useful
1681 * information on memory leaks).
1682 */
1683static void kmemleak_do_cleanup(struct work_struct *work)
1684{
1685        struct kmemleak_object *object;
1686        bool cleanup = scan_thread == NULL;
1687
1688        mutex_lock(&scan_mutex);
1689        stop_scan_thread();
1690
1691        if (cleanup) {
1692                rcu_read_lock();
1693                list_for_each_entry_rcu(object, &object_list, object_list)
1694                        delete_object_full(object->pointer);
1695                rcu_read_unlock();
1696        }
1697        mutex_unlock(&scan_mutex);
1698}
1699
1700static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1701
1702/*
1703 * Disable kmemleak. No memory allocation/freeing will be traced once this
1704 * function is called. Disabling kmemleak is an irreversible operation.
1705 */
1706static void kmemleak_disable(void)
1707{
1708        /* atomically check whether it was already invoked */
1709        if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1710                return;
1711
1712        /* stop any memory operation tracing */
1713        atomic_set(&kmemleak_enabled, 0);
1714
1715        /* check whether it is too early for a kernel thread */
1716        if (atomic_read(&kmemleak_initialized))
1717                schedule_work(&cleanup_work);
1718
1719        pr_info("Kernel memory leak detector disabled\n");
1720}
1721
1722/*
1723 * Allow boot-time kmemleak disabling (enabled by default).
1724 */
1725static int kmemleak_boot_config(char *str)
1726{
1727        if (!str)
1728                return -EINVAL;
1729        if (strcmp(str, "off") == 0)
1730                kmemleak_disable();
1731        else if (strcmp(str, "on") == 0)
1732                kmemleak_skip_disable = 1;
1733        else
1734                return -EINVAL;
1735        return 0;
1736}
1737early_param("kmemleak", kmemleak_boot_config);
1738
1739static void __init print_log_trace(struct early_log *log)
1740{
1741        struct stack_trace trace;
1742
1743        trace.nr_entries = log->trace_len;
1744        trace.entries = log->trace;
1745
1746        pr_notice("Early log backtrace:\n");
1747        print_stack_trace(&trace, 2);
1748}
1749
1750/*
1751 * Kmemleak initialization.
1752 */
1753void __init kmemleak_init(void)
1754{
1755        int i;
1756        unsigned long flags;
1757
1758#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1759        if (!kmemleak_skip_disable) {
1760                atomic_set(&kmemleak_early_log, 0);
1761                kmemleak_disable();
1762                return;
1763        }
1764#endif
1765
1766        jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1767        jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1768
1769        object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1770        scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1771        INIT_PRIO_TREE_ROOT(&object_tree_root);
1772
1773        if (crt_early_log >= ARRAY_SIZE(early_log))
1774                pr_warning("Early log buffer exceeded (%d), please increase "
1775                           "DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n", crt_early_log);
1776
1777        /* the kernel is still in UP mode, so disabling the IRQs is enough */
1778        local_irq_save(flags);
1779        atomic_set(&kmemleak_early_log, 0);
1780        if (atomic_read(&kmemleak_error)) {
1781                local_irq_restore(flags);
1782                return;
1783        } else
1784                atomic_set(&kmemleak_enabled, 1);
1785        local_irq_restore(flags);
1786
1787        /*
1788         * This is the point where tracking allocations is safe. Automatic
1789         * scanning is started during the late initcall. Add the early logged
1790         * callbacks to the kmemleak infrastructure.
1791         */
1792        for (i = 0; i < crt_early_log; i++) {
1793                struct early_log *log = &early_log[i];
1794
1795                switch (log->op_type) {
1796                case KMEMLEAK_ALLOC:
1797                        early_alloc(log);
1798                        break;
1799                case KMEMLEAK_ALLOC_PERCPU:
1800                        early_alloc_percpu(log);
1801                        break;
1802                case KMEMLEAK_FREE:
1803                        kmemleak_free(log->ptr);
1804                        break;
1805                case KMEMLEAK_FREE_PART:
1806                        kmemleak_free_part(log->ptr, log->size);
1807                        break;
1808                case KMEMLEAK_FREE_PERCPU:
1809                        kmemleak_free_percpu(log->ptr);
1810                        break;
1811                case KMEMLEAK_NOT_LEAK:
1812                        kmemleak_not_leak(log->ptr);
1813                        break;
1814                case KMEMLEAK_IGNORE:
1815                        kmemleak_ignore(log->ptr);
1816                        break;
1817                case KMEMLEAK_SCAN_AREA:
1818                        kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
1819                        break;
1820                case KMEMLEAK_NO_SCAN:
1821                        kmemleak_no_scan(log->ptr);
1822                        break;
1823                default:
1824                        kmemleak_warn("Unknown early log operation: %d\n",
1825                                      log->op_type);
1826                }
1827
1828                if (atomic_read(&kmemleak_warning)) {
1829                        print_log_trace(log);
1830                        atomic_set(&kmemleak_warning, 0);
1831                }
1832        }
1833}
1834
1835/*
1836 * Late initialization function.
1837 */
1838static int __init kmemleak_late_init(void)
1839{
1840        struct dentry *dentry;
1841
1842        atomic_set(&kmemleak_initialized, 1);
1843
1844        if (atomic_read(&kmemleak_error)) {
1845                /*
1846                 * Some error occurred and kmemleak was disabled. There is a
1847                 * small chance that kmemleak_disable() was called immediately
1848                 * after setting kmemleak_initialized and we may end up with
1849                 * two clean-up threads but serialized by scan_mutex.
1850                 */
1851                schedule_work(&cleanup_work);
1852                return -ENOMEM;
1853        }
1854
1855        dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1856                                     &kmemleak_fops);
1857        if (!dentry)
1858                pr_warning("Failed to create the debugfs kmemleak file\n");
1859        mutex_lock(&scan_mutex);
1860        start_scan_thread();
1861        mutex_unlock(&scan_mutex);
1862
1863        pr_info("Kernel memory leak detector initialized\n");
1864
1865        return 0;
1866}
1867late_initcall(kmemleak_late_init);
1868
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.