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/module.h>
  73#include <linux/kthread.h>
  74#include <linux/prio_tree.h>
  75#include <linux/fs.h>
  76#include <linux/debugfs.h>
  77#include <linux/seq_file.h>
  78#include <linux/cpumask.h>
  79#include <linux/spinlock.h>
  80#include <linux/mutex.h>
  81#include <linux/rcupdate.h>
  82#include <linux/stacktrace.h>
  83#include <linux/cache.h>
  84#include <linux/percpu.h>
  85#include <linux/hardirq.h>
  86#include <linux/mmzone.h>
  87#include <linux/slab.h>
  88#include <linux/thread_info.h>
  89#include <linux/err.h>
  90#include <linux/uaccess.h>
  91#include <linux/string.h>
  92#include <linux/nodemask.h>
  93#include <linux/mm.h>
  94#include <linux/workqueue.h>
  95#include <linux/crc32.h>
  96
  97#include <asm/sections.h>
  98#include <asm/processor.h>
  99#include <linux/atomic.h>
 100
 101#include <linux/kmemcheck.h>
 102#include <linux/kmemleak.h>
 103
 104/*
 105 * Kmemleak configuration and common defines.
 106 */
 107#define MAX_TRACE               16      /* stack trace length */
 108#define MSECS_MIN_AGE           5000    /* minimum object age for reporting */
 109#define SECS_FIRST_SCAN         60      /* delay before the first scan */
 110#define SECS_SCAN_WAIT          600     /* subsequent auto scanning delay */
 111#define MAX_SCAN_SIZE           4096    /* maximum size of a scanned block */
 112
 113#define BYTES_PER_POINTER       sizeof(void *)
 114
 115/* GFP bitmask for kmemleak internal allocations */
 116#define gfp_kmemleak_mask(gfp)  (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
 117                                 __GFP_NORETRY | __GFP_NOMEMALLOC | \
 118                                 __GFP_NOWARN)
 119
 120/* scanning area inside a memory block */
 121struct kmemleak_scan_area {
 122        struct hlist_node node;
 123        unsigned long start;
 124        size_t size;
 125};
 126
 127#define KMEMLEAK_GREY   0
 128#define KMEMLEAK_BLACK  -1
 129
 130/*
 131 * Structure holding the metadata for each allocated memory block.
 132 * Modifications to such objects should be made while holding the
 133 * object->lock. Insertions or deletions from object_list, gray_list or
 134 * tree_node are already protected by the corresponding locks or mutex (see
 135 * the notes on locking above). These objects are reference-counted
 136 * (use_count) and freed using the RCU mechanism.
 137 */
 138struct kmemleak_object {
 139        spinlock_t lock;
 140        unsigned long flags;            /* object status flags */
 141        struct list_head object_list;
 142        struct list_head gray_list;
 143        struct prio_tree_node tree_node;
 144        struct rcu_head rcu;            /* object_list lockless traversal */
 145        /* object usage count; object freed when use_count == 0 */
 146        atomic_t use_count;
 147        unsigned long pointer;
 148        size_t size;
 149        /* minimum number of a pointers found before it is considered leak */
 150        int min_count;
 151        /* the total number of pointers found pointing to this object */
 152        int count;
 153        /* checksum for detecting modified objects */
 154        u32 checksum;
 155        /* memory ranges to be scanned inside an object (empty for all) */
 156        struct hlist_head area_list;
 157        unsigned long trace[MAX_TRACE];
 158        unsigned int trace_len;
 159        unsigned long jiffies;          /* creation timestamp */
 160        pid_t pid;                      /* pid of the current task */
 161        char comm[TASK_COMM_LEN];       /* executable name */
 162};
 163
 164/* flag representing the memory block allocation status */
 165#define OBJECT_ALLOCATED        (1 << 0)
 166/* flag set after the first reporting of an unreference object */
 167#define OBJECT_REPORTED         (1 << 1)
 168/* flag set to not scan the object */
 169#define OBJECT_NO_SCAN          (1 << 2)
 170
 171/* number of bytes to print per line; must be 16 or 32 */
 172#define HEX_ROW_SIZE            16
 173/* number of bytes to print at a time (1, 2, 4, 8) */
 174#define HEX_GROUP_SIZE          1
 175/* include ASCII after the hex output */
 176#define HEX_ASCII               1
 177/* max number of lines to be printed */
 178#define HEX_MAX_LINES           2
 179
 180/* the list of all allocated objects */
 181static LIST_HEAD(object_list);
 182/* the list of gray-colored objects (see color_gray comment below) */
 183static LIST_HEAD(gray_list);
 184/* prio search tree for object boundaries */
 185static struct prio_tree_root object_tree_root;
 186/* rw_lock protecting the access to object_list and prio_tree_root */
 187static DEFINE_RWLOCK(kmemleak_lock);
 188
 189/* allocation caches for kmemleak internal data */
 190static struct kmem_cache *object_cache;
 191static struct kmem_cache *scan_area_cache;
 192
 193/* set if tracing memory operations is enabled */
 194static atomic_t kmemleak_enabled = ATOMIC_INIT(0);
 195/* set in the late_initcall if there were no errors */
 196static atomic_t kmemleak_initialized = ATOMIC_INIT(0);
 197/* enables or disables early logging of the memory operations */
 198static atomic_t kmemleak_early_log = ATOMIC_INIT(1);
 199/* set if a fata kmemleak error has occurred */
 200static atomic_t kmemleak_error = ATOMIC_INIT(0);
 201
 202/* minimum and maximum address that may be valid pointers */
 203static unsigned long min_addr = ULONG_MAX;
 204static unsigned long max_addr;
 205
 206static struct task_struct *scan_thread;
 207/* used to avoid reporting of recently allocated objects */
 208static unsigned long jiffies_min_age;
 209static unsigned long jiffies_last_scan;
 210/* delay between automatic memory scannings */
 211static signed long jiffies_scan_wait;
 212/* enables or disables the task stacks scanning */
 213static int kmemleak_stack_scan = 1;
 214/* protects the memory scanning, parameters and debug/kmemleak file access */
 215static DEFINE_MUTEX(scan_mutex);
 216/* setting kmemleak=on, will set this var, skipping the disable */
 217static int kmemleak_skip_disable;
 218
 219
 220/*
 221 * Early object allocation/freeing logging. Kmemleak is initialized after the
 222 * kernel allocator. However, both the kernel allocator and kmemleak may
 223 * allocate memory blocks which need to be tracked. Kmemleak defines an
 224 * arbitrary buffer to hold the allocation/freeing information before it is
 225 * fully initialized.
 226 */
 227
 228/* kmemleak operation type for early logging */
 229enum {
 230        KMEMLEAK_ALLOC,
 231        KMEMLEAK_FREE,
 232        KMEMLEAK_FREE_PART,
 233        KMEMLEAK_NOT_LEAK,
 234        KMEMLEAK_IGNORE,
 235        KMEMLEAK_SCAN_AREA,
 236        KMEMLEAK_NO_SCAN
 237};
 238
 239/*
 240 * Structure holding the information passed to kmemleak callbacks during the
 241 * early logging.
 242 */
 243struct early_log {
 244        int op_type;                    /* kmemleak operation type */
 245        const void *ptr;                /* allocated/freed memory block */
 246        size_t size;                    /* memory block size */
 247        int min_count;                  /* minimum reference count */
 248        unsigned long trace[MAX_TRACE]; /* stack trace */
 249        unsigned int trace_len;         /* stack trace length */
 250};
 251
 252/* early logging buffer and current position */
 253static struct early_log
 254        early_log[CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE] __initdata;
 255static int crt_early_log __initdata;
 256
 257static void kmemleak_disable(void);
 258
 259/*
 260 * Print a warning and dump the stack trace.
 261 */
 262#define kmemleak_warn(x...)     do {    \
 263        pr_warning(x);                  \
 264        dump_stack();                   \
 265} while (0)
 266
 267/*
 268 * Macro invoked when a serious kmemleak condition occurred and cannot be
 269 * recovered from. Kmemleak will be disabled and further allocation/freeing
 270 * tracing no longer available.
 271 */
 272#define kmemleak_stop(x...)     do {    \
 273        kmemleak_warn(x);               \
 274        kmemleak_disable();             \
 275} while (0)
 276
 277/*
 278 * Printing of the objects hex dump to the seq file. The number of lines to be
 279 * printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
 280 * actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
 281 * with the object->lock held.
 282 */
 283static void hex_dump_object(struct seq_file *seq,
 284                            struct kmemleak_object *object)
 285{
 286        const u8 *ptr = (const u8 *)object->pointer;
 287        int i, len, remaining;
 288        unsigned char linebuf[HEX_ROW_SIZE * 5];
 289
 290        /* limit the number of lines to HEX_MAX_LINES */
 291        remaining = len =
 292                min(object->size, (size_t)(HEX_MAX_LINES * HEX_ROW_SIZE));
 293
 294        seq_printf(seq, "  hex dump (first %d bytes):\n", len);
 295        for (i = 0; i < len; i += HEX_ROW_SIZE) {
 296                int linelen = min(remaining, HEX_ROW_SIZE);
 297
 298                remaining -= HEX_ROW_SIZE;
 299                hex_dump_to_buffer(ptr + i, linelen, HEX_ROW_SIZE,
 300                                   HEX_GROUP_SIZE, linebuf, sizeof(linebuf),
 301                                   HEX_ASCII);
 302                seq_printf(seq, "    %s\n", linebuf);
 303        }
 304}
 305
 306/*
 307 * Object colors, encoded with count and min_count:
 308 * - white - orphan object, not enough references to it (count < min_count)
 309 * - gray  - not orphan, not marked as false positive (min_count == 0) or
 310 *              sufficient references to it (count >= min_count)
 311 * - black - ignore, it doesn't contain references (e.g. text section)
 312 *              (min_count == -1). No function defined for this color.
 313 * Newly created objects don't have any color assigned (object->count == -1)
 314 * before the next memory scan when they become white.
 315 */
 316static bool color_white(const struct kmemleak_object *object)
 317{
 318        return object->count != KMEMLEAK_BLACK &&
 319                object->count < object->min_count;
 320}
 321
 322static bool color_gray(const struct kmemleak_object *object)
 323{
 324        return object->min_count != KMEMLEAK_BLACK &&
 325                object->count >= object->min_count;
 326}
 327
 328/*
 329 * Objects are considered unreferenced only if their color is white, they have
 330 * not be deleted and have a minimum age to avoid false positives caused by
 331 * pointers temporarily stored in CPU registers.
 332 */
 333static bool unreferenced_object(struct kmemleak_object *object)
 334{
 335        return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
 336                time_before_eq(object->jiffies + jiffies_min_age,
 337                               jiffies_last_scan);
 338}
 339
 340/*
 341 * Printing of the unreferenced objects information to the seq file. The
 342 * print_unreferenced function must be called with the object->lock held.
 343 */
 344static void print_unreferenced(struct seq_file *seq,
 345                               struct kmemleak_object *object)
 346{
 347        int i;
 348        unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
 349
 350        seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
 351                   object->pointer, object->size);
 352        seq_printf(seq, "  comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
 353                   object->comm, object->pid, object->jiffies,
 354                   msecs_age / 1000, msecs_age % 1000);
 355        hex_dump_object(seq, object);
 356        seq_printf(seq, "  backtrace:\n");
 357
 358        for (i = 0; i < object->trace_len; i++) {
 359                void *ptr = (void *)object->trace[i];
 360                seq_printf(seq, "    [<%p>] %pS\n", ptr, ptr);
 361        }
 362}
 363
 364/*
 365 * Print the kmemleak_object information. This function is used mainly for
 366 * debugging special cases when kmemleak operations. It must be called with
 367 * the object->lock held.
 368 */
 369static void dump_object_info(struct kmemleak_object *object)
 370{
 371        struct stack_trace trace;
 372
 373        trace.nr_entries = object->trace_len;
 374        trace.entries = object->trace;
 375
 376        pr_notice("Object 0x%08lx (size %zu):\n",
 377                  object->tree_node.start, object->size);
 378        pr_notice("  comm \"%s\", pid %d, jiffies %lu\n",
 379                  object->comm, object->pid, object->jiffies);
 380        pr_notice("  min_count = %d\n", object->min_count);
 381        pr_notice("  count = %d\n", object->count);
 382        pr_notice("  flags = 0x%lx\n", object->flags);
 383        pr_notice("  checksum = %d\n", object->checksum);
 384        pr_notice("  backtrace:\n");
 385        print_stack_trace(&trace, 4);
 386}
 387
 388/*
 389 * Look-up a memory block metadata (kmemleak_object) in the priority search
 390 * tree based on a pointer value. If alias is 0, only values pointing to the
 391 * beginning of the memory block are allowed. The kmemleak_lock must be held
 392 * when calling this function.
 393 */
 394static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
 395{
 396        struct prio_tree_node *node;
 397        struct prio_tree_iter iter;
 398        struct kmemleak_object *object;
 399
 400        prio_tree_iter_init(&iter, &object_tree_root, ptr, ptr);
 401        node = prio_tree_next(&iter);
 402        if (node) {
 403                object = prio_tree_entry(node, struct kmemleak_object,
 404                                         tree_node);
 405                if (!alias && object->pointer != ptr) {
 406                        pr_warning("Found object by alias at 0x%08lx\n", ptr);
 407                        dump_stack();
 408                        dump_object_info(object);
 409                        object = NULL;
 410                }
 411        } else
 412                object = NULL;
 413
 414        return object;
 415}
 416
 417/*
 418 * Increment the object use_count. Return 1 if successful or 0 otherwise. Note
 419 * that once an object's use_count reached 0, the RCU freeing was already
 420 * registered and the object should no longer be used. This function must be
 421 * called under the protection of rcu_read_lock().
 422 */
 423static int get_object(struct kmemleak_object *object)
 424{
 425        return atomic_inc_not_zero(&object->use_count);
 426}
 427
 428/*
 429 * RCU callback to free a kmemleak_object.
 430 */
 431static void free_object_rcu(struct rcu_head *rcu)
 432{
 433        struct hlist_node *elem, *tmp;
 434        struct kmemleak_scan_area *area;
 435        struct kmemleak_object *object =
 436                container_of(rcu, struct kmemleak_object, rcu);
 437
 438        /*
 439         * Once use_count is 0 (guaranteed by put_object), there is no other
 440         * code accessing this object, hence no need for locking.
 441         */
 442        hlist_for_each_entry_safe(area, elem, tmp, &object->area_list, node) {
 443                hlist_del(elem);
 444                kmem_cache_free(scan_area_cache, area);
 445        }
 446        kmem_cache_free(object_cache, object);
 447}
 448
 449/*
 450 * Decrement the object use_count. Once the count is 0, free the object using
 451 * an RCU callback. Since put_object() may be called via the kmemleak_free() ->
 452 * delete_object() path, the delayed RCU freeing ensures that there is no
 453 * recursive call to the kernel allocator. Lock-less RCU object_list traversal
 454 * is also possible.
 455 */
 456static void put_object(struct kmemleak_object *object)
 457{
 458        if (!atomic_dec_and_test(&object->use_count))
 459                return;
 460
 461        /* should only get here after delete_object was called */
 462        WARN_ON(object->flags & OBJECT_ALLOCATED);
 463
 464        call_rcu(&object->rcu, free_object_rcu);
 465}
 466
 467/*
 468 * Look up an object in the prio search tree and increase its use_count.
 469 */
 470static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
 471{
 472        unsigned long flags;
 473        struct kmemleak_object *object = NULL;
 474
 475        rcu_read_lock();
 476        read_lock_irqsave(&kmemleak_lock, flags);
 477        if (ptr >= min_addr && ptr < max_addr)
 478                object = lookup_object(ptr, alias);
 479        read_unlock_irqrestore(&kmemleak_lock, flags);
 480
 481        /* check whether the object is still available */
 482        if (object && !get_object(object))
 483                object = NULL;
 484        rcu_read_unlock();
 485
 486        return object;
 487}
 488
 489/*
 490 * Save stack trace to the given array of MAX_TRACE size.
 491 */
 492static int __save_stack_trace(unsigned long *trace)
 493{
 494        struct stack_trace stack_trace;
 495
 496        stack_trace.max_entries = MAX_TRACE;
 497        stack_trace.nr_entries = 0;
 498        stack_trace.entries = trace;
 499        stack_trace.skip = 2;
 500        save_stack_trace(&stack_trace);
 501
 502        return stack_trace.nr_entries;
 503}
 504
 505/*
 506 * Create the metadata (struct kmemleak_object) corresponding to an allocated
 507 * memory block and add it to the object_list and object_tree_root.
 508 */
 509static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
 510                                             int min_count, gfp_t gfp)
 511{
 512        unsigned long flags;
 513        struct kmemleak_object *object;
 514        struct prio_tree_node *node;
 515
 516        object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
 517        if (!object) {
 518                pr_warning("Cannot allocate a kmemleak_object structure\n");
 519                kmemleak_disable();
 520                return NULL;
 521        }
 522
 523        INIT_LIST_HEAD(&object->object_list);
 524        INIT_LIST_HEAD(&object->gray_list);
 525        INIT_HLIST_HEAD(&object->area_list);
 526        spin_lock_init(&object->lock);
 527        atomic_set(&object->use_count, 1);
 528        object->flags = OBJECT_ALLOCATED;
 529        object->pointer = ptr;
 530        object->size = size;
 531        object->min_count = min_count;
 532        object->count = 0;                      /* white color initially */
 533        object->jiffies = jiffies;
 534        object->checksum = 0;
 535
 536        /* task information */
 537        if (in_irq()) {
 538                object->pid = 0;
 539                strncpy(object->comm, "hardirq", sizeof(object->comm));
 540        } else if (in_softirq()) {
 541                object->pid = 0;
 542                strncpy(object->comm, "softirq", sizeof(object->comm));
 543        } else {
 544                object->pid = current->pid;
 545                /*
 546                 * There is a small chance of a race with set_task_comm(),
 547                 * however using get_task_comm() here may cause locking
 548                 * dependency issues with current->alloc_lock. In the worst
 549                 * case, the command line is not correct.
 550                 */
 551                strncpy(object->comm, current->comm, sizeof(object->comm));
 552        }
 553
 554        /* kernel backtrace */
 555        object->trace_len = __save_stack_trace(object->trace);
 556
 557        INIT_PRIO_TREE_NODE(&object->tree_node);
 558        object->tree_node.start = ptr;
 559        object->tree_node.last = ptr + size - 1;
 560
 561        write_lock_irqsave(&kmemleak_lock, flags);
 562
 563        min_addr = min(min_addr, ptr);
 564        max_addr = max(max_addr, ptr + size);
 565        node = prio_tree_insert(&object_tree_root, &object->tree_node);
 566        /*
 567         * The code calling the kernel does not yet have the pointer to the
 568         * memory block to be able to free it.  However, we still hold the
 569         * kmemleak_lock here in case parts of the kernel started freeing
 570         * random memory blocks.
 571         */
 572        if (node != &object->tree_node) {
 573                kmemleak_stop("Cannot insert 0x%lx into the object search tree "
 574                              "(already existing)\n", ptr);
 575                object = lookup_object(ptr, 1);
 576                spin_lock(&object->lock);
 577                dump_object_info(object);
 578                spin_unlock(&object->lock);
 579
 580                goto out;
 581        }
 582        list_add_tail_rcu(&object->object_list, &object_list);
 583out:
 584        write_unlock_irqrestore(&kmemleak_lock, flags);
 585        return object;
 586}
 587
 588/*
 589 * Remove the metadata (struct kmemleak_object) for a memory block from the
 590 * object_list and object_tree_root and decrement its use_count.
 591 */
 592static void __delete_object(struct kmemleak_object *object)
 593{
 594        unsigned long flags;
 595
 596        write_lock_irqsave(&kmemleak_lock, flags);
 597        prio_tree_remove(&object_tree_root, &object->tree_node);
 598        list_del_rcu(&object->object_list);
 599        write_unlock_irqrestore(&kmemleak_lock, flags);
 600
 601        WARN_ON(!(object->flags & OBJECT_ALLOCATED));
 602        WARN_ON(atomic_read(&object->use_count) < 2);
 603
 604        /*
 605         * Locking here also ensures that the corresponding memory block
 606         * cannot be freed when it is being scanned.
 607         */
 608        spin_lock_irqsave(&object->lock, flags);
 609        object->flags &= ~OBJECT_ALLOCATED;
 610        spin_unlock_irqrestore(&object->lock, flags);
 611        put_object(object);
 612}
 613
 614/*
 615 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
 616 * delete it.
 617 */
 618static void delete_object_full(unsigned long ptr)
 619{
 620        struct kmemleak_object *object;
 621
 622        object = find_and_get_object(ptr, 0);
 623        if (!object) {
 624#ifdef DEBUG
 625                kmemleak_warn("Freeing unknown object at 0x%08lx\n",
 626                              ptr);
 627#endif
 628                return;
 629        }
 630        __delete_object(object);
 631        put_object(object);
 632}
 633
 634/*
 635 * Look up the metadata (struct kmemleak_object) corresponding to ptr and
 636 * delete it. If the memory block is partially freed, the function may create
 637 * additional metadata for the remaining parts of the block.
 638 */
 639static void delete_object_part(unsigned long ptr, size_t size)
 640{
 641        struct kmemleak_object *object;
 642        unsigned long start, end;
 643
 644        object = find_and_get_object(ptr, 1);
 645        if (!object) {
 646#ifdef DEBUG
 647                kmemleak_warn("Partially freeing unknown object at 0x%08lx "
 648                              "(size %zu)\n", ptr, size);
 649#endif
 650                return;
 651        }
 652        __delete_object(object);
 653
 654        /*
 655         * Create one or two objects that may result from the memory block
 656         * split. Note that partial freeing is only done by free_bootmem() and
 657         * this happens before kmemleak_init() is called. The path below is
 658         * only executed during early log recording in kmemleak_init(), so
 659         * GFP_KERNEL is enough.
 660         */
 661        start = object->pointer;
 662        end = object->pointer + object->size;
 663        if (ptr > start)
 664                create_object(start, ptr - start, object->min_count,
 665                              GFP_KERNEL);
 666        if (ptr + size < end)
 667                create_object(ptr + size, end - ptr - size, object->min_count,
 668                              GFP_KERNEL);
 669
 670        put_object(object);
 671}
 672
 673static void __paint_it(struct kmemleak_object *object, int color)
 674{
 675        object->min_count = color;
 676        if (color == KMEMLEAK_BLACK)
 677                object->flags |= OBJECT_NO_SCAN;
 678}
 679
 680static void paint_it(struct kmemleak_object *object, int color)
 681{
 682        unsigned long flags;
 683
 684        spin_lock_irqsave(&object->lock, flags);
 685        __paint_it(object, color);
 686        spin_unlock_irqrestore(&object->lock, flags);
 687}
 688
 689static void paint_ptr(unsigned long ptr, int color)
 690{
 691        struct kmemleak_object *object;
 692
 693        object = find_and_get_object(ptr, 0);
 694        if (!object) {
 695                kmemleak_warn("Trying to color unknown object "
 696                              "at 0x%08lx as %s\n", ptr,
 697                              (color == KMEMLEAK_GREY) ? "Grey" :
 698                              (color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
 699                return;
 700        }
 701        paint_it(object, color);
 702        put_object(object);
 703}
 704
 705/*
 706 * Mark an object permanently as gray-colored so that it can no longer be
 707 * reported as a leak. This is used in general to mark a false positive.
 708 */
 709static void make_gray_object(unsigned long ptr)
 710{
 711        paint_ptr(ptr, KMEMLEAK_GREY);
 712}
 713
 714/*
 715 * Mark the object as black-colored so that it is ignored from scans and
 716 * reporting.
 717 */
 718static void make_black_object(unsigned long ptr)
 719{
 720        paint_ptr(ptr, KMEMLEAK_BLACK);
 721}
 722
 723/*
 724 * Add a scanning area to the object. If at least one such area is added,
 725 * kmemleak will only scan these ranges rather than the whole memory block.
 726 */
 727static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
 728{
 729        unsigned long flags;
 730        struct kmemleak_object *object;
 731        struct kmemleak_scan_area *area;
 732
 733        object = find_and_get_object(ptr, 1);
 734        if (!object) {
 735                kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
 736                              ptr);
 737                return;
 738        }
 739
 740        area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
 741        if (!area) {
 742                pr_warning("Cannot allocate a scan area\n");
 743                goto out;
 744        }
 745
 746        spin_lock_irqsave(&object->lock, flags);
 747        if (ptr + size > object->pointer + object->size) {
 748                kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
 749                dump_object_info(object);
 750                kmem_cache_free(scan_area_cache, area);
 751                goto out_unlock;
 752        }
 753
 754        INIT_HLIST_NODE(&area->node);
 755        area->start = ptr;
 756        area->size = size;
 757
 758        hlist_add_head(&area->node, &object->area_list);
 759out_unlock:
 760        spin_unlock_irqrestore(&object->lock, flags);
 761out:
 762        put_object(object);
 763}
 764
 765/*
 766 * Set the OBJECT_NO_SCAN flag for the object corresponding to the give
 767 * pointer. Such object will not be scanned by kmemleak but references to it
 768 * are searched.
 769 */
 770static void object_no_scan(unsigned long ptr)
 771{
 772        unsigned long flags;
 773        struct kmemleak_object *object;
 774
 775        object = find_and_get_object(ptr, 0);
 776        if (!object) {
 777                kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
 778                return;
 779        }
 780
 781        spin_lock_irqsave(&object->lock, flags);
 782        object->flags |= OBJECT_NO_SCAN;
 783        spin_unlock_irqrestore(&object->lock, flags);
 784        put_object(object);
 785}
 786
 787/*
 788 * Log an early kmemleak_* call to the early_log buffer. These calls will be
 789 * processed later once kmemleak is fully initialized.
 790 */
 791static void __init log_early(int op_type, const void *ptr, size_t size,
 792                             int min_count)
 793{
 794        unsigned long flags;
 795        struct early_log *log;
 796
 797        if (crt_early_log >= ARRAY_SIZE(early_log)) {
 798                pr_warning("Early log buffer exceeded, "
 799                           "please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE\n");
 800                kmemleak_disable();
 801                return;
 802        }
 803
 804        /*
 805         * There is no need for locking since the kernel is still in UP mode
 806         * at this stage. Disabling the IRQs is enough.
 807         */
 808        local_irq_save(flags);
 809        log = &early_log[crt_early_log];
 810        log->op_type = op_type;
 811        log->ptr = ptr;
 812        log->size = size;
 813        log->min_count = min_count;
 814        if (op_type == KMEMLEAK_ALLOC)
 815                log->trace_len = __save_stack_trace(log->trace);
 816        crt_early_log++;
 817        local_irq_restore(flags);
 818}
 819
 820/*
 821 * Log an early allocated block and populate the stack trace.
 822 */
 823static void early_alloc(struct early_log *log)
 824{
 825        struct kmemleak_object *object;
 826        unsigned long flags;
 827        int i;
 828
 829        if (!atomic_read(&kmemleak_enabled) || !log->ptr || IS_ERR(log->ptr))
 830                return;
 831
 832        /*
 833         * RCU locking needed to ensure object is not freed via put_object().
 834         */
 835        rcu_read_lock();
 836        object = create_object((unsigned long)log->ptr, log->size,
 837                               log->min_count, GFP_ATOMIC);
 838        if (!object)
 839                goto out;
 840        spin_lock_irqsave(&object->lock, flags);
 841        for (i = 0; i < log->trace_len; i++)
 842                object->trace[i] = log->trace[i];
 843        object->trace_len = log->trace_len;
 844        spin_unlock_irqrestore(&object->lock, flags);
 845out:
 846        rcu_read_unlock();
 847}
 848
 849/**
 850 * kmemleak_alloc - register a newly allocated object
 851 * @ptr:        pointer to beginning of the object
 852 * @size:       size of the object
 853 * @min_count:  minimum number of references to this object. If during memory
 854 *              scanning a number of references less than @min_count is found,
 855 *              the object is reported as a memory leak. If @min_count is 0,
 856 *              the object is never reported as a leak. If @min_count is -1,
 857 *              the object is ignored (not scanned and not reported as a leak)
 858 * @gfp:        kmalloc() flags used for kmemleak internal memory allocations
 859 *
 860 * This function is called from the kernel allocators when a new object
 861 * (memory block) is allocated (kmem_cache_alloc, kmalloc, vmalloc etc.).
 862 */
 863void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
 864                          gfp_t gfp)
 865{
 866        pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
 867
 868        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 869                create_object((unsigned long)ptr, size, min_count, gfp);
 870        else if (atomic_read(&kmemleak_early_log))
 871                log_early(KMEMLEAK_ALLOC, ptr, size, min_count);
 872}
 873EXPORT_SYMBOL_GPL(kmemleak_alloc);
 874
 875/**
 876 * kmemleak_free - unregister a previously registered object
 877 * @ptr:        pointer to beginning of the object
 878 *
 879 * This function is called from the kernel allocators when an object (memory
 880 * block) is freed (kmem_cache_free, kfree, vfree etc.).
 881 */
 882void __ref kmemleak_free(const void *ptr)
 883{
 884        pr_debug("%s(0x%p)\n", __func__, ptr);
 885
 886        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 887                delete_object_full((unsigned long)ptr);
 888        else if (atomic_read(&kmemleak_early_log))
 889                log_early(KMEMLEAK_FREE, ptr, 0, 0);
 890}
 891EXPORT_SYMBOL_GPL(kmemleak_free);
 892
 893/**
 894 * kmemleak_free_part - partially unregister a previously registered object
 895 * @ptr:        pointer to the beginning or inside the object. This also
 896 *              represents the start of the range to be freed
 897 * @size:       size to be unregistered
 898 *
 899 * This function is called when only a part of a memory block is freed
 900 * (usually from the bootmem allocator).
 901 */
 902void __ref kmemleak_free_part(const void *ptr, size_t size)
 903{
 904        pr_debug("%s(0x%p)\n", __func__, ptr);
 905
 906        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 907                delete_object_part((unsigned long)ptr, size);
 908        else if (atomic_read(&kmemleak_early_log))
 909                log_early(KMEMLEAK_FREE_PART, ptr, size, 0);
 910}
 911EXPORT_SYMBOL_GPL(kmemleak_free_part);
 912
 913/**
 914 * kmemleak_not_leak - mark an allocated object as false positive
 915 * @ptr:        pointer to beginning of the object
 916 *
 917 * Calling this function on an object will cause the memory block to no longer
 918 * be reported as leak and always be scanned.
 919 */
 920void __ref kmemleak_not_leak(const void *ptr)
 921{
 922        pr_debug("%s(0x%p)\n", __func__, ptr);
 923
 924        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 925                make_gray_object((unsigned long)ptr);
 926        else if (atomic_read(&kmemleak_early_log))
 927                log_early(KMEMLEAK_NOT_LEAK, ptr, 0, 0);
 928}
 929EXPORT_SYMBOL(kmemleak_not_leak);
 930
 931/**
 932 * kmemleak_ignore - ignore an allocated object
 933 * @ptr:        pointer to beginning of the object
 934 *
 935 * Calling this function on an object will cause the memory block to be
 936 * ignored (not scanned and not reported as a leak). This is usually done when
 937 * it is known that the corresponding block is not a leak and does not contain
 938 * any references to other allocated memory blocks.
 939 */
 940void __ref kmemleak_ignore(const void *ptr)
 941{
 942        pr_debug("%s(0x%p)\n", __func__, ptr);
 943
 944        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 945                make_black_object((unsigned long)ptr);
 946        else if (atomic_read(&kmemleak_early_log))
 947                log_early(KMEMLEAK_IGNORE, ptr, 0, 0);
 948}
 949EXPORT_SYMBOL(kmemleak_ignore);
 950
 951/**
 952 * kmemleak_scan_area - limit the range to be scanned in an allocated object
 953 * @ptr:        pointer to beginning or inside the object. This also
 954 *              represents the start of the scan area
 955 * @size:       size of the scan area
 956 * @gfp:        kmalloc() flags used for kmemleak internal memory allocations
 957 *
 958 * This function is used when it is known that only certain parts of an object
 959 * contain references to other objects. Kmemleak will only scan these areas
 960 * reducing the number false negatives.
 961 */
 962void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
 963{
 964        pr_debug("%s(0x%p)\n", __func__, ptr);
 965
 966        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 967                add_scan_area((unsigned long)ptr, size, gfp);
 968        else if (atomic_read(&kmemleak_early_log))
 969                log_early(KMEMLEAK_SCAN_AREA, ptr, size, 0);
 970}
 971EXPORT_SYMBOL(kmemleak_scan_area);
 972
 973/**
 974 * kmemleak_no_scan - do not scan an allocated object
 975 * @ptr:        pointer to beginning of the object
 976 *
 977 * This function notifies kmemleak not to scan the given memory block. Useful
 978 * in situations where it is known that the given object does not contain any
 979 * references to other objects. Kmemleak will not scan such objects reducing
 980 * the number of false negatives.
 981 */
 982void __ref kmemleak_no_scan(const void *ptr)
 983{
 984        pr_debug("%s(0x%p)\n", __func__, ptr);
 985
 986        if (atomic_read(&kmemleak_enabled) && ptr && !IS_ERR(ptr))
 987                object_no_scan((unsigned long)ptr);
 988        else if (atomic_read(&kmemleak_early_log))
 989                log_early(KMEMLEAK_NO_SCAN, ptr, 0, 0);
 990}
 991EXPORT_SYMBOL(kmemleak_no_scan);
 992
 993/*
 994 * Update an object's checksum and return true if it was modified.
 995 */
 996static bool update_checksum(struct kmemleak_object *object)
 997{
 998        u32 old_csum = object->checksum;
 999
1000        if (!kmemcheck_is_obj_initialized(object->pointer, object->size))
1001                return false;
1002
1003        object->checksum = crc32(0, (void *)object->pointer, object->size);
1004        return object->checksum != old_csum;
1005}
1006
1007/*
1008 * Memory scanning is a long process and it needs to be interruptable. This
1009 * function checks whether such interrupt condition occurred.
1010 */
1011static int scan_should_stop(void)
1012{
1013        if (!atomic_read(&kmemleak_enabled))
1014                return 1;
1015
1016        /*
1017         * This function may be called from either process or kthread context,
1018         * hence the need to check for both stop conditions.
1019         */
1020        if (current->mm)
1021                return signal_pending(current);
1022        else
1023                return kthread_should_stop();
1024
1025        return 0;
1026}
1027
1028/*
1029 * Scan a memory block (exclusive range) for valid pointers and add those
1030 * found to the gray list.
1031 */
1032static void scan_block(void *_start, void *_end,
1033                       struct kmemleak_object *scanned, int allow_resched)
1034{
1035        unsigned long *ptr;
1036        unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
1037        unsigned long *end = _end - (BYTES_PER_POINTER - 1);
1038
1039        for (ptr = start; ptr < end; ptr++) {
1040                struct kmemleak_object *object;
1041                unsigned long flags;
1042                unsigned long pointer;
1043
1044                if (allow_resched)
1045                        cond_resched();
1046                if (scan_should_stop())
1047                        break;
1048
1049                /* don't scan uninitialized memory */
1050                if (!kmemcheck_is_obj_initialized((unsigned long)ptr,
1051                                                  BYTES_PER_POINTER))
1052                        continue;
1053
1054                pointer = *ptr;
1055
1056                object = find_and_get_object(pointer, 1);
1057                if (!object)
1058                        continue;
1059                if (object == scanned) {
1060                        /* self referenced, ignore */
1061                        put_object(object);
1062                        continue;
1063                }
1064
1065                /*
1066                 * Avoid the lockdep recursive warning on object->lock being
1067                 * previously acquired in scan_object(). These locks are
1068                 * enclosed by scan_mutex.
1069                 */
1070                spin_lock_irqsave_nested(&object->lock, flags,
1071                                         SINGLE_DEPTH_NESTING);
1072                if (!color_white(object)) {
1073                        /* non-orphan, ignored or new */
1074                        spin_unlock_irqrestore(&object->lock, flags);
1075                        put_object(object);
1076                        continue;
1077                }
1078
1079                /*
1080                 * Increase the object's reference count (number of pointers
1081                 * to the memory block). If this count reaches the required
1082                 * minimum, the object's color will become gray and it will be
1083                 * added to the gray_list.
1084                 */
1085                object->count++;
1086                if (color_gray(object)) {
1087                        list_add_tail(&object->gray_list, &gray_list);
1088                        spin_unlock_irqrestore(&object->lock, flags);
1089                        continue;
1090                }
1091
1092                spin_unlock_irqrestore(&object->lock, flags);
1093                put_object(object);
1094        }
1095}
1096
1097/*
1098 * Scan a memory block corresponding to a kmemleak_object. A condition is
1099 * that object->use_count >= 1.
1100 */
1101static void scan_object(struct kmemleak_object *object)
1102{
1103        struct kmemleak_scan_area *area;
1104        struct hlist_node *elem;
1105        unsigned long flags;
1106
1107        /*
1108         * Once the object->lock is acquired, the corresponding memory block
1109         * cannot be freed (the same lock is acquired in delete_object).
1110         */
1111        spin_lock_irqsave(&object->lock, flags);
1112        if (object->flags & OBJECT_NO_SCAN)
1113                goto out;
1114        if (!(object->flags & OBJECT_ALLOCATED))
1115                /* already freed object */
1116                goto out;
1117        if (hlist_empty(&object->area_list)) {
1118                void *start = (void *)object->pointer;
1119                void *end = (void *)(object->pointer + object->size);
1120
1121                while (start < end && (object->flags & OBJECT_ALLOCATED) &&
1122                       !(object->flags & OBJECT_NO_SCAN)) {
1123                        scan_block(start, min(start + MAX_SCAN_SIZE, end),
1124                                   object, 0);
1125                        start += MAX_SCAN_SIZE;
1126
1127                        spin_unlock_irqrestore(&object->lock, flags);
1128                        cond_resched();
1129                        spin_lock_irqsave(&object->lock, flags);
1130                }
1131        } else
1132                hlist_for_each_entry(area, elem, &object->area_list, node)
1133                        scan_block((void *)area->start,
1134                                   (void *)(area->start + area->size),
1135                                   object, 0);
1136out:
1137        spin_unlock_irqrestore(&object->lock, flags);
1138}
1139
1140/*
1141 * Scan the objects already referenced (gray objects). More objects will be
1142 * referenced and, if there are no memory leaks, all the objects are scanned.
1143 */
1144static void scan_gray_list(void)
1145{
1146        struct kmemleak_object *object, *tmp;
1147
1148        /*
1149         * The list traversal is safe for both tail additions and removals
1150         * from inside the loop. The kmemleak objects cannot be freed from
1151         * outside the loop because their use_count was incremented.
1152         */
1153        object = list_entry(gray_list.next, typeof(*object), gray_list);
1154        while (&object->gray_list != &gray_list) {
1155                cond_resched();
1156
1157                /* may add new objects to the list */
1158                if (!scan_should_stop())
1159                        scan_object(object);
1160
1161                tmp = list_entry(object->gray_list.next, typeof(*object),
1162                                 gray_list);
1163
1164                /* remove the object from the list and release it */
1165                list_del(&object->gray_list);
1166                put_object(object);
1167
1168                object = tmp;
1169        }
1170        WARN_ON(!list_empty(&gray_list));
1171}
1172
1173/*
1174 * Scan data sections and all the referenced memory blocks allocated via the
1175 * kernel's standard allocators. This function must be called with the
1176 * scan_mutex held.
1177 */
1178static void kmemleak_scan(void)
1179{
1180        unsigned long flags;
1181        struct kmemleak_object *object;
1182        int i;
1183        int new_leaks = 0;
1184
1185        jiffies_last_scan = jiffies;
1186
1187        /* prepare the kmemleak_object's */
1188        rcu_read_lock();
1189        list_for_each_entry_rcu(object, &object_list, object_list) {
1190                spin_lock_irqsave(&object->lock, flags);
1191#ifdef DEBUG
1192                /*
1193                 * With a few exceptions there should be a maximum of
1194                 * 1 reference to any object at this point.
1195                 */
1196                if (atomic_read(&object->use_count) > 1) {
1197                        pr_debug("object->use_count = %d\n",
1198                                 atomic_read(&object->use_count));
1199                        dump_object_info(object);
1200                }
1201#endif
1202                /* reset the reference count (whiten the object) */
1203                object->count = 0;
1204                if (color_gray(object) && get_object(object))
1205                        list_add_tail(&object->gray_list, &gray_list);
1206
1207                spin_unlock_irqrestore(&object->lock, flags);
1208        }
1209        rcu_read_unlock();
1210
1211        /* data/bss scanning */
1212        scan_block(_sdata, _edata, NULL, 1);
1213        scan_block(__bss_start, __bss_stop, NULL, 1);
1214
1215#ifdef CONFIG_SMP
1216        /* per-cpu sections scanning */
1217        for_each_possible_cpu(i)
1218                scan_block(__per_cpu_start + per_cpu_offset(i),
1219                           __per_cpu_end + per_cpu_offset(i), NULL, 1);
1220#endif
1221
1222        /*
1223         * Struct page scanning for each node. The code below is not yet safe
1224         * with MEMORY_HOTPLUG.
1225         */
1226        for_each_online_node(i) {
1227                pg_data_t *pgdat = NODE_DATA(i);
1228                unsigned long start_pfn = pgdat->node_start_pfn;
1229                unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1230                unsigned long pfn;
1231
1232                for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1233                        struct page *page;
1234
1235                        if (!pfn_valid(pfn))
1236                                continue;
1237                        page = pfn_to_page(pfn);
1238                        /* only scan if page is in use */
1239                        if (page_count(page) == 0)
1240                                continue;
1241                        scan_block(page, page + 1, NULL, 1);
1242                }
1243        }
1244
1245        /*
1246         * Scanning the task stacks (may introduce false negatives).
1247         */
1248        if (kmemleak_stack_scan) {
1249                struct task_struct *p, *g;
1250
1251                read_lock(&tasklist_lock);
1252                do_each_thread(g, p) {
1253                        scan_block(task_stack_page(p), task_stack_page(p) +
1254                                   THREAD_SIZE, NULL, 0);
1255                } while_each_thread(g, p);
1256                read_unlock(&tasklist_lock);
1257        }
1258
1259        /*
1260         * Scan the objects already referenced from the sections scanned
1261         * above.
1262         */
1263        scan_gray_list();
1264
1265        /*
1266         * Check for new or unreferenced objects modified since the previous
1267         * scan and color them gray until the next scan.
1268         */
1269        rcu_read_lock();
1270        list_for_each_entry_rcu(object, &object_list, object_list) {
1271                spin_lock_irqsave(&object->lock, flags);
1272                if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
1273                    && update_checksum(object) && get_object(object)) {
1274                        /* color it gray temporarily */
1275                        object->count = object->min_count;
1276                        list_add_tail(&object->gray_list, &gray_list);
1277                }
1278                spin_unlock_irqrestore(&object->lock, flags);
1279        }
1280        rcu_read_unlock();
1281
1282        /*
1283         * Re-scan the gray list for modified unreferenced objects.
1284         */
1285        scan_gray_list();
1286
1287        /*
1288         * If scanning was stopped do not report any new unreferenced objects.
1289         */
1290        if (scan_should_stop())
1291                return;
1292
1293        /*
1294         * Scanning result reporting.
1295         */
1296        rcu_read_lock();
1297        list_for_each_entry_rcu(object, &object_list, object_list) {
1298                spin_lock_irqsave(&object->lock, flags);
1299                if (unreferenced_object(object) &&
1300                    !(object->flags & OBJECT_REPORTED)) {
1301                        object->flags |= OBJECT_REPORTED;
1302                        new_leaks++;
1303                }
1304                spin_unlock_irqrestore(&object->lock, flags);
1305        }
1306        rcu_read_unlock();
1307
1308        if (new_leaks)
1309                pr_info("%d new suspected memory leaks (see "
1310                        "/sys/kernel/debug/kmemleak)\n", new_leaks);
1311
1312}
1313
1314/*
1315 * Thread function performing automatic memory scanning. Unreferenced objects
1316 * at the end of a memory scan are reported but only the first time.
1317 */
1318static int kmemleak_scan_thread(void *arg)
1319{
1320        static int first_run = 1;
1321
1322        pr_info("Automatic memory scanning thread started\n");
1323        set_user_nice(current, 10);
1324
1325        /*
1326         * Wait before the first scan to allow the system to fully initialize.
1327         */
1328        if (first_run) {
1329                first_run = 0;
1330                ssleep(SECS_FIRST_SCAN);
1331        }
1332
1333        while (!kthread_should_stop()) {
1334                signed long timeout = jiffies_scan_wait;
1335
1336                mutex_lock(&scan_mutex);
1337                kmemleak_scan();
1338                mutex_unlock(&scan_mutex);
1339
1340                /* wait before the next scan */
1341                while (timeout && !kthread_should_stop())
1342                        timeout = schedule_timeout_interruptible(timeout);
1343        }
1344
1345        pr_info("Automatic memory scanning thread ended\n");
1346
1347        return 0;
1348}
1349
1350/*
1351 * Start the automatic memory scanning thread. This function must be called
1352 * with the scan_mutex held.
1353 */
1354static void start_scan_thread(void)
1355{
1356        if (scan_thread)
1357                return;
1358        scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
1359        if (IS_ERR(scan_thread)) {
1360                pr_warning("Failed to create the scan thread\n");
1361                scan_thread = NULL;
1362        }
1363}
1364
1365/*
1366 * Stop the automatic memory scanning thread. This function must be called
1367 * with the scan_mutex held.
1368 */
1369static void stop_scan_thread(void)
1370{
1371        if (scan_thread) {
1372                kthread_stop(scan_thread);
1373                scan_thread = NULL;
1374        }
1375}
1376
1377/*
1378 * Iterate over the object_list and return the first valid object at or after
1379 * the required position with its use_count incremented. The function triggers
1380 * a memory scanning when the pos argument points to the first position.
1381 */
1382static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
1383{
1384        struct kmemleak_object *object;
1385        loff_t n = *pos;
1386        int err;
1387
1388        err = mutex_lock_interruptible(&scan_mutex);
1389        if (err < 0)
1390                return ERR_PTR(err);
1391
1392        rcu_read_lock();
1393        list_for_each_entry_rcu(object, &object_list, object_list) {
1394                if (n-- > 0)
1395                        continue;
1396                if (get_object(object))
1397                        goto out;
1398        }
1399        object = NULL;
1400out:
1401        return object;
1402}
1403
1404/*
1405 * Return the next object in the object_list. The function decrements the
1406 * use_count of the previous object and increases that of the next one.
1407 */
1408static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1409{
1410        struct kmemleak_object *prev_obj = v;
1411        struct kmemleak_object *next_obj = NULL;
1412        struct list_head *n = &prev_obj->object_list;
1413
1414        ++(*pos);
1415
1416        list_for_each_continue_rcu(n, &object_list) {
1417                struct kmemleak_object *obj =
1418                        list_entry(n, struct kmemleak_object, object_list);
1419                if (get_object(obj)) {
1420                        next_obj = obj;
1421                        break;
1422                }
1423        }
1424
1425        put_object(prev_obj);
1426        return next_obj;
1427}
1428
1429/*
1430 * Decrement the use_count of the last object required, if any.
1431 */
1432static void kmemleak_seq_stop(struct seq_file *seq, void *v)
1433{
1434        if (!IS_ERR(v)) {
1435                /*
1436                 * kmemleak_seq_start may return ERR_PTR if the scan_mutex
1437                 * waiting was interrupted, so only release it if !IS_ERR.
1438                 */
1439                rcu_read_unlock();
1440                mutex_unlock(&scan_mutex);
1441                if (v)
1442                        put_object(v);
1443        }
1444}
1445
1446/*
1447 * Print the information for an unreferenced object to the seq file.
1448 */
1449static int kmemleak_seq_show(struct seq_file *seq, void *v)
1450{
1451        struct kmemleak_object *object = v;
1452        unsigned long flags;
1453
1454        spin_lock_irqsave(&object->lock, flags);
1455        if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
1456                print_unreferenced(seq, object);
1457        spin_unlock_irqrestore(&object->lock, flags);
1458        return 0;
1459}
1460
1461static const struct seq_operations kmemleak_seq_ops = {
1462        .start = kmemleak_seq_start,
1463        .next  = kmemleak_seq_next,
1464        .stop  = kmemleak_seq_stop,
1465        .show  = kmemleak_seq_show,
1466};
1467
1468static int kmemleak_open(struct inode *inode, struct file *file)
1469{
1470        if (!atomic_read(&kmemleak_enabled))
1471                return -EBUSY;
1472
1473        return seq_open(file, &kmemleak_seq_ops);
1474}
1475
1476static int kmemleak_release(struct inode *inode, struct file *file)
1477{
1478        return seq_release(inode, file);
1479}
1480
1481static int dump_str_object_info(const char *str)
1482{
1483        unsigned long flags;
1484        struct kmemleak_object *object;
1485        unsigned long addr;
1486
1487        addr= simple_strtoul(str, NULL, 0);
1488        object = find_and_get_object(addr, 0);
1489        if (!object) {
1490                pr_info("Unknown object at 0x%08lx\n", addr);
1491                return -EINVAL;
1492        }
1493
1494        spin_lock_irqsave(&object->lock, flags);
1495        dump_object_info(object);
1496        spin_unlock_irqrestore(&object->lock, flags);
1497
1498        put_object(object);
1499        return 0;
1500}
1501
1502/*
1503 * We use grey instead of black to ensure we can do future scans on the same
1504 * objects. If we did not do future scans these black objects could
1505 * potentially contain references to newly allocated objects in the future and
1506 * we'd end up with false positives.
1507 */
1508static void kmemleak_clear(void)
1509{
1510        struct kmemleak_object *object;
1511        unsigned long flags;
1512
1513        rcu_read_lock();
1514        list_for_each_entry_rcu(object, &object_list, object_list) {
1515                spin_lock_irqsave(&object->lock, flags);
1516                if ((object->flags & OBJECT_REPORTED) &&
1517                    unreferenced_object(object))
1518                        __paint_it(object, KMEMLEAK_GREY);
1519                spin_unlock_irqrestore(&object->lock, flags);
1520        }
1521        rcu_read_unlock();
1522}
1523
1524/*
1525 * File write operation to configure kmemleak at run-time. The following
1526 * commands can be written to the /sys/kernel/debug/kmemleak file:
1527 *   off        - disable kmemleak (irreversible)
1528 *   stack=on   - enable the task stacks scanning
1529 *   stack=off  - disable the tasks stacks scanning
1530 *   scan=on    - start the automatic memory scanning thread
1531 *   scan=off   - stop the automatic memory scanning thread
1532 *   scan=...   - set the automatic memory scanning period in seconds (0 to
1533 *                disable it)
1534 *   scan       - trigger a memory scan
1535 *   clear      - mark all current reported unreferenced kmemleak objects as
1536 *                grey to ignore printing them
1537 *   dump=...   - dump information about the object found at the given address
1538 */
1539static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
1540                              size_t size, loff_t *ppos)
1541{
1542        char buf[64];
1543        int buf_size;
1544        int ret;
1545
1546        buf_size = min(size, (sizeof(buf) - 1));
1547        if (strncpy_from_user(buf, user_buf, buf_size) < 0)
1548                return -EFAULT;
1549        buf[buf_size] = 0;
1550
1551        ret = mutex_lock_interruptible(&scan_mutex);
1552        if (ret < 0)
1553                return ret;
1554
1555        if (strncmp(buf, "off", 3) == 0)
1556                kmemleak_disable();
1557        else if (strncmp(buf, "stack=on", 8) == 0)
1558                kmemleak_stack_scan = 1;
1559        else if (strncmp(buf, "stack=off", 9) == 0)
1560                kmemleak_stack_scan = 0;
1561        else if (strncmp(buf, "scan=on", 7) == 0)
1562                start_scan_thread();
1563        else if (strncmp(buf, "scan=off", 8) == 0)
1564                stop_scan_thread();
1565        else if (strncmp(buf, "scan=", 5) == 0) {
1566                unsigned long secs;
1567
1568                ret = strict_strtoul(buf + 5, 0, &secs);
1569                if (ret < 0)
1570                        goto out;
1571                stop_scan_thread();
1572                if (secs) {
1573                        jiffies_scan_wait = msecs_to_jiffies(secs * 1000);
1574                        start_scan_thread();
1575                }
1576        } else if (strncmp(buf, "scan", 4) == 0)
1577                kmemleak_scan();
1578        else if (strncmp(buf, "clear", 5) == 0)
1579                kmemleak_clear();
1580        else if (strncmp(buf, "dump=", 5) == 0)
1581                ret = dump_str_object_info(buf + 5);
1582        else
1583                ret = -EINVAL;
1584
1585out:
1586        mutex_unlock(&scan_mutex);
1587        if (ret < 0)
1588                return ret;
1589
1590        /* ignore the rest of the buffer, only one command at a time */
1591        *ppos += size;
1592        return size;
1593}
1594
1595static const struct file_operations kmemleak_fops = {
1596        .owner          = THIS_MODULE,
1597        .open           = kmemleak_open,
1598        .read           = seq_read,
1599        .write          = kmemleak_write,
1600        .llseek         = seq_lseek,
1601        .release        = kmemleak_release,
1602};
1603
1604/*
1605 * Perform the freeing of the kmemleak internal objects after waiting for any
1606 * current memory scan to complete.
1607 */
1608static void kmemleak_do_cleanup(struct work_struct *work)
1609{
1610        struct kmemleak_object *object;
1611
1612        mutex_lock(&scan_mutex);
1613        stop_scan_thread();
1614
1615        rcu_read_lock();
1616        list_for_each_entry_rcu(object, &object_list, object_list)
1617                delete_object_full(object->pointer);
1618        rcu_read_unlock();
1619        mutex_unlock(&scan_mutex);
1620}
1621
1622static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
1623
1624/*
1625 * Disable kmemleak. No memory allocation/freeing will be traced once this
1626 * function is called. Disabling kmemleak is an irreversible operation.
1627 */
1628static void kmemleak_disable(void)
1629{
1630        /* atomically check whether it was already invoked */
1631        if (atomic_cmpxchg(&kmemleak_error, 0, 1))
1632                return;
1633
1634        /* stop any memory operation tracing */
1635        atomic_set(&kmemleak_early_log, 0);
1636        atomic_set(&kmemleak_enabled, 0);
1637
1638        /* check whether it is too early for a kernel thread */
1639        if (atomic_read(&kmemleak_initialized))
1640                schedule_work(&cleanup_work);
1641
1642        pr_info("Kernel memory leak detector disabled\n");
1643}
1644
1645/*
1646 * Allow boot-time kmemleak disabling (enabled by default).
1647 */
1648static int kmemleak_boot_config(char *str)
1649{
1650        if (!str)
1651                return -EINVAL;
1652        if (strcmp(str, "off") == 0)
1653                kmemleak_disable();
1654        else if (strcmp(str, "on") == 0)
1655                kmemleak_skip_disable = 1;
1656        else
1657                return -EINVAL;
1658        return 0;
1659}
1660early_param("kmemleak", kmemleak_boot_config);
1661
1662/*
1663 * Kmemleak initialization.
1664 */
1665void __init kmemleak_init(void)
1666{
1667        int i;
1668        unsigned long flags;
1669
1670#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
1671        if (!kmemleak_skip_disable) {
1672                kmemleak_disable();
1673                return;
1674        }
1675#endif
1676
1677        jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
1678        jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
1679
1680        object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
1681        scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
1682        INIT_PRIO_TREE_ROOT(&object_tree_root);
1683
1684        /* the kernel is still in UP mode, so disabling the IRQs is enough */
1685        local_irq_save(flags);
1686        if (!atomic_read(&kmemleak_error)) {
1687                atomic_set(&kmemleak_enabled, 1);
1688                atomic_set(&kmemleak_early_log, 0);
1689        }
1690        local_irq_restore(flags);
1691
1692        /*
1693         * This is the point where tracking allocations is safe. Automatic
1694         * scanning is started during the late initcall. Add the early logged
1695         * callbacks to the kmemleak infrastructure.
1696         */
1697        for (i = 0; i < crt_early_log; i++) {
1698                struct early_log *log = &early_log[i];
1699
1700                switch (log->op_type) {
1701                case KMEMLEAK_ALLOC:
1702                        early_alloc(log);
1703                        break;
1704                case KMEMLEAK_FREE:
1705                        kmemleak_free(log->ptr);
1706                        break;
1707                case KMEMLEAK_FREE_PART:
1708                        kmemleak_free_part(log->ptr, log->size);
1709                        break;
1710                case KMEMLEAK_NOT_LEAK:
1711                        kmemleak_not_leak(log->ptr);
1712                        break;
1713                case KMEMLEAK_IGNORE:
1714                        kmemleak_ignore(log->ptr);
1715                        break;
1716                case KMEMLEAK_SCAN_AREA:
1717                        kmemleak_scan_area(log->ptr, log->size, GFP_KERNEL);
1718                        break;
1719                case KMEMLEAK_NO_SCAN:
1720                        kmemleak_no_scan(log->ptr);
1721                        break;
1722                default:
1723                        WARN_ON(1);
1724                }
1725        }
1726}
1727
1728/*
1729 * Late initialization function.
1730 */
1731static int __init kmemleak_late_init(void)
1732{
1733        struct dentry *dentry;
1734
1735        atomic_set(&kmemleak_initialized, 1);
1736
1737        if (atomic_read(&kmemleak_error)) {
1738                /*
1739                 * Some error occurred and kmemleak was disabled. There is a
1740                 * small chance that kmemleak_disable() was called immediately
1741                 * after setting kmemleak_initialized and we may end up with
1742                 * two clean-up threads but serialized by scan_mutex.
1743                 */
1744                schedule_work(&cleanup_work);
1745                return -ENOMEM;
1746        }
1747
1748        dentry = debugfs_create_file("kmemleak", S_IRUGO, NULL, NULL,
1749                                     &kmemleak_fops);
1750        if (!dentry)
1751                pr_warning("Failed to create the debugfs kmemleak file\n");
1752        mutex_lock(&scan_mutex);
1753        start_scan_thread();
1754        mutex_unlock(&scan_mutex);
1755
1756        pr_info("Kernel memory leak detector initialized\n");
1757
1758        return 0;
1759}
1760late_initcall(kmemleak_late_init);
1761
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.