linux/mm/kmemleak.c

1997 lines
56 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* mm/kmemleak.c
*
* Copyright (C) 2008 ARM Limited
* Written by Catalin Marinas <catalin.marinas@arm.com>
*
* For more information on the algorithm and kmemleak usage, please see
* Documentation/dev-tools/kmemleak.rst.
*
* Notes on locking
* ----------------
*
* The following locks and mutexes are used by kmemleak:
*
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
* - kmemleak_lock (raw_spinlock_t): protects the object_list modifications and
* accesses to the object_tree_root. The object_list is the main list
* holding the metadata (struct kmemleak_object) for the allocated memory
* blocks. The object_tree_root is a red black tree used to look-up
* metadata based on a pointer to the corresponding memory block. The
* kmemleak_object structures are added to the object_list and
* object_tree_root in the create_object() function called from the
* kmemleak_alloc() callback and removed in delete_object() called from the
* kmemleak_free() callback
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
* - kmemleak_object.lock (raw_spinlock_t): protects a kmemleak_object.
* Accesses to the metadata (e.g. count) are protected by this lock. Note
* that some members of this structure may be protected by other means
* (atomic or kmemleak_lock). This lock is also held when scanning the
* corresponding memory block to avoid the kernel freeing it via the
* kmemleak_free() callback. This is less heavyweight than holding a global
* lock like kmemleak_lock during scanning.
* - scan_mutex (mutex): ensures that only one thread may scan the memory for
* unreferenced objects at a time. The gray_list contains the objects which
* are already referenced or marked as false positives and need to be
* scanned. This list is only modified during a scanning episode when the
* scan_mutex is held. At the end of a scan, the gray_list is always empty.
* Note that the kmemleak_object.use_count is incremented when an object is
* added to the gray_list and therefore cannot be freed. This mutex also
* prevents multiple users of the "kmemleak" debugfs file together with
* modifications to the memory scanning parameters including the scan_thread
* pointer
*
* Locks and mutexes are acquired/nested in the following order:
*
* scan_mutex [-> object->lock] -> kmemleak_lock -> other_object->lock (SINGLE_DEPTH_NESTING)
*
* No kmemleak_lock and object->lock nesting is allowed outside scan_mutex
* regions.
*
* The kmemleak_object structures have a use_count incremented or decremented
* using the get_object()/put_object() functions. When the use_count becomes
* 0, this count can no longer be incremented and put_object() schedules the
* kmemleak_object freeing via an RCU callback. All calls to the get_object()
* function must be protected by rcu_read_lock() to avoid accessing a freed
* structure.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/sched/signal.h>
#include <linux/sched/task.h>
#include <linux/sched/task_stack.h>
#include <linux/jiffies.h>
#include <linux/delay.h>
#include <linux/export.h>
#include <linux/kthread.h>
#include <linux/rbtree.h>
#include <linux/fs.h>
#include <linux/debugfs.h>
#include <linux/seq_file.h>
#include <linux/cpumask.h>
#include <linux/spinlock.h>
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/rcupdate.h>
#include <linux/stacktrace.h>
#include <linux/cache.h>
#include <linux/percpu.h>
mm: remove include/linux/bootmem.h Move remaining definitions and declarations from include/linux/bootmem.h into include/linux/memblock.h and remove the redundant header. The includes were replaced with the semantic patch below and then semi-automated removal of duplicated '#include <linux/memblock.h> @@ @@ - #include <linux/bootmem.h> + #include <linux/memblock.h> [sfr@canb.auug.org.au: dma-direct: fix up for the removal of linux/bootmem.h] Link: http://lkml.kernel.org/r/20181002185342.133d1680@canb.auug.org.au [sfr@canb.auug.org.au: powerpc: fix up for removal of linux/bootmem.h] Link: http://lkml.kernel.org/r/20181005161406.73ef8727@canb.auug.org.au [sfr@canb.auug.org.au: x86/kaslr, ACPI/NUMA: fix for linux/bootmem.h removal] Link: http://lkml.kernel.org/r/20181008190341.5e396491@canb.auug.org.au Link: http://lkml.kernel.org/r/1536927045-23536-30-git-send-email-rppt@linux.vnet.ibm.com Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Chris Zankel <chris@zankel.net> Cc: "David S. Miller" <davem@davemloft.net> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Greentime Hu <green.hu@gmail.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Guan Xuetao <gxt@pku.edu.cn> Cc: Ingo Molnar <mingo@redhat.com> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Ley Foon Tan <lftan@altera.com> Cc: Mark Salter <msalter@redhat.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Simek <monstr@monstr.eu> Cc: Palmer Dabbelt <palmer@sifive.com> Cc: Paul Burton <paul.burton@mips.com> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Richard Weinberger <richard@nod.at> Cc: Rich Felker <dalias@libc.org> Cc: Russell King <linux@armlinux.org.uk> Cc: Serge Semin <fancer.lancer@gmail.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Tony Luck <tony.luck@intel.com> Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-30 22:09:49 +00:00
#include <linux/memblock.h>
#include <linux/pfn.h>
#include <linux/mmzone.h>
#include <linux/slab.h>
#include <linux/thread_info.h>
#include <linux/err.h>
#include <linux/uaccess.h>
#include <linux/string.h>
#include <linux/nodemask.h>
#include <linux/mm.h>
#include <linux/workqueue.h>
#include <linux/crc32.h>
#include <asm/sections.h>
#include <asm/processor.h>
#include <linux/atomic.h>
#include <linux/kasan.h>
2021-03-25 04:37:47 +00:00
#include <linux/kfence.h>
#include <linux/kmemleak.h>
#include <linux/memory_hotplug.h>
/*
* Kmemleak configuration and common defines.
*/
#define MAX_TRACE 16 /* stack trace length */
#define MSECS_MIN_AGE 5000 /* minimum object age for reporting */
#define SECS_FIRST_SCAN 60 /* delay before the first scan */
#define SECS_SCAN_WAIT 600 /* subsequent auto scanning delay */
#define MAX_SCAN_SIZE 4096 /* maximum size of a scanned block */
#define BYTES_PER_POINTER sizeof(void *)
/* GFP bitmask for kmemleak internal allocations */
#define gfp_kmemleak_mask(gfp) (((gfp) & (GFP_KERNEL | GFP_ATOMIC)) | \
__GFP_NORETRY | __GFP_NOMEMALLOC | \
Revert "kmemleak: allow to coexist with fault injection" When running ltp's oom test with kmemleak enabled, the below warning was triggerred since kernel detects __GFP_NOFAIL & ~__GFP_DIRECT_RECLAIM is passed in: WARNING: CPU: 105 PID: 2138 at mm/page_alloc.c:4608 __alloc_pages_nodemask+0x1c31/0x1d50 Modules linked in: loop dax_pmem dax_pmem_core ip_tables x_tables xfs virtio_net net_failover virtio_blk failover ata_generic virtio_pci virtio_ring virtio libata CPU: 105 PID: 2138 Comm: oom01 Not tainted 5.2.0-next-20190710+ #7 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.10.2-0-g5f4c7b1-prebuilt.qemu-project.org 04/01/2014 RIP: 0010:__alloc_pages_nodemask+0x1c31/0x1d50 ... kmemleak_alloc+0x4e/0xb0 kmem_cache_alloc+0x2a7/0x3e0 mempool_alloc_slab+0x2d/0x40 mempool_alloc+0x118/0x2b0 bio_alloc_bioset+0x19d/0x350 get_swap_bio+0x80/0x230 __swap_writepage+0x5ff/0xb20 The mempool_alloc_slab() clears __GFP_DIRECT_RECLAIM, however kmemleak has __GFP_NOFAIL set all the time due to d9570ee3bd1d4f2 ("kmemleak: allow to coexist with fault injection"). But, it doesn't make any sense to have __GFP_NOFAIL and ~__GFP_DIRECT_RECLAIM specified at the same time. According to the discussion on the mailing list, the commit should be reverted for short term solution. Catalin Marinas would follow up with a better solution for longer term. The failure rate of kmemleak metadata allocation may increase in some circumstances, but this should be expected side effect. Link: http://lkml.kernel.org/r/1563299431-111710-1-git-send-email-yang.shi@linux.alibaba.com Fixes: d9570ee3bd1d4f2 ("kmemleak: allow to coexist with fault injection") Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> Suggested-by: Catalin Marinas <catalin.marinas@arm.com> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: David Rientjes <rientjes@google.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-08-03 04:48:37 +00:00
__GFP_NOWARN)
/* scanning area inside a memory block */
struct kmemleak_scan_area {
struct hlist_node node;
unsigned long start;
size_t size;
};
#define KMEMLEAK_GREY 0
#define KMEMLEAK_BLACK -1
/*
* Structure holding the metadata for each allocated memory block.
* Modifications to such objects should be made while holding the
* object->lock. Insertions or deletions from object_list, gray_list or
* rb_node are already protected by the corresponding locks or mutex (see
* the notes on locking above). These objects are reference-counted
* (use_count) and freed using the RCU mechanism.
*/
struct kmemleak_object {
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spinlock_t lock;
unsigned int flags; /* object status flags */
struct list_head object_list;
struct list_head gray_list;
struct rb_node rb_node;
struct rcu_head rcu; /* object_list lockless traversal */
/* object usage count; object freed when use_count == 0 */
atomic_t use_count;
unsigned long pointer;
size_t size;
/* pass surplus references to this pointer */
unsigned long excess_ref;
/* minimum number of a pointers found before it is considered leak */
int min_count;
/* the total number of pointers found pointing to this object */
int count;
/* checksum for detecting modified objects */
u32 checksum;
/* memory ranges to be scanned inside an object (empty for all) */
struct hlist_head area_list;
unsigned long trace[MAX_TRACE];
unsigned int trace_len;
unsigned long jiffies; /* creation timestamp */
pid_t pid; /* pid of the current task */
char comm[TASK_COMM_LEN]; /* executable name */
};
/* flag representing the memory block allocation status */
#define OBJECT_ALLOCATED (1 << 0)
/* flag set after the first reporting of an unreference object */
#define OBJECT_REPORTED (1 << 1)
/* flag set to not scan the object */
#define OBJECT_NO_SCAN (1 << 2)
mm: kmemleak: make the tool tolerant to struct scan_area allocation failures Patch series "mm: kmemleak: Use a memory pool for kmemleak object allocations", v3. Following the discussions on v2 of this patch(set) [1], this series takes slightly different approach: - it implements its own simple memory pool that does not rely on the slab allocator - drops the early log buffer logic entirely since it can now allocate metadata from the memory pool directly before kmemleak is fully initialised - CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE option is renamed to CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE - moves the kmemleak_init() call earlier (mm_init()) - to avoid a separate memory pool for struct scan_area, it makes the tool robust when such allocations fail as scan areas are rather an optimisation [1] http://lkml.kernel.org/r/20190727132334.9184-1-catalin.marinas@arm.com This patch (of 3): Object scan areas are an optimisation aimed to decrease the false positives and slightly improve the scanning time of large objects known to only have a few specific pointers. If a struct scan_area fails to allocate, kmemleak can still function normally by scanning the full object. Introduce an OBJECT_FULL_SCAN flag and mark objects as such when scan_area allocation fails. Link: http://lkml.kernel.org/r/20190812160642.52134-2-catalin.marinas@arm.com Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-09-23 22:33:59 +00:00
/* flag set to fully scan the object when scan_area allocation failed */
#define OBJECT_FULL_SCAN (1 << 3)
#define HEX_PREFIX " "
/* number of bytes to print per line; must be 16 or 32 */
#define HEX_ROW_SIZE 16
/* number of bytes to print at a time (1, 2, 4, 8) */
#define HEX_GROUP_SIZE 1
/* include ASCII after the hex output */
#define HEX_ASCII 1
/* max number of lines to be printed */
#define HEX_MAX_LINES 2
/* the list of all allocated objects */
static LIST_HEAD(object_list);
/* the list of gray-colored objects (see color_gray comment below) */
static LIST_HEAD(gray_list);
/* memory pool allocation */
static struct kmemleak_object mem_pool[CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE];
static int mem_pool_free_count = ARRAY_SIZE(mem_pool);
static LIST_HEAD(mem_pool_free_list);
/* search tree for object boundaries */
static struct rb_root object_tree_root = RB_ROOT;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
/* protecting the access to object_list and object_tree_root */
static DEFINE_RAW_SPINLOCK(kmemleak_lock);
/* allocation caches for kmemleak internal data */
static struct kmem_cache *object_cache;
static struct kmem_cache *scan_area_cache;
/* set if tracing memory operations is enabled */
static int kmemleak_enabled = 1;
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
/* same as above but only for the kmemleak_free() callback */
static int kmemleak_free_enabled = 1;
/* set in the late_initcall if there were no errors */
static int kmemleak_initialized;
/* set if a kmemleak warning was issued */
static int kmemleak_warning;
/* set if a fatal kmemleak error has occurred */
static int kmemleak_error;
/* minimum and maximum address that may be valid pointers */
static unsigned long min_addr = ULONG_MAX;
static unsigned long max_addr;
static struct task_struct *scan_thread;
/* used to avoid reporting of recently allocated objects */
static unsigned long jiffies_min_age;
static unsigned long jiffies_last_scan;
/* delay between automatic memory scannings */
static unsigned long jiffies_scan_wait;
/* enables or disables the task stacks scanning */
static int kmemleak_stack_scan = 1;
/* protects the memory scanning, parameters and debug/kmemleak file access */
static DEFINE_MUTEX(scan_mutex);
/* setting kmemleak=on, will set this var, skipping the disable */
static int kmemleak_skip_disable;
/* If there are leaks that can be reported */
static bool kmemleak_found_leaks;
static bool kmemleak_verbose;
module_param_named(verbose, kmemleak_verbose, bool, 0600);
static void kmemleak_disable(void);
/*
* Print a warning and dump the stack trace.
*/
#define kmemleak_warn(x...) do { \
pr_warn(x); \
dump_stack(); \
kmemleak_warning = 1; \
} while (0)
/*
* Macro invoked when a serious kmemleak condition occurred and cannot be
* recovered from. Kmemleak will be disabled and further allocation/freeing
* tracing no longer available.
*/
#define kmemleak_stop(x...) do { \
kmemleak_warn(x); \
kmemleak_disable(); \
} while (0)
#define warn_or_seq_printf(seq, fmt, ...) do { \
if (seq) \
seq_printf(seq, fmt, ##__VA_ARGS__); \
else \
pr_warn(fmt, ##__VA_ARGS__); \
} while (0)
static void warn_or_seq_hex_dump(struct seq_file *seq, int prefix_type,
int rowsize, int groupsize, const void *buf,
size_t len, bool ascii)
{
if (seq)
seq_hex_dump(seq, HEX_PREFIX, prefix_type, rowsize, groupsize,
buf, len, ascii);
else
print_hex_dump(KERN_WARNING, pr_fmt(HEX_PREFIX), prefix_type,
rowsize, groupsize, buf, len, ascii);
}
/*
* Printing of the objects hex dump to the seq file. The number of lines to be
* printed is limited to HEX_MAX_LINES to prevent seq file spamming. The
* actual number of printed bytes depends on HEX_ROW_SIZE. It must be called
* with the object->lock held.
*/
static void hex_dump_object(struct seq_file *seq,
struct kmemleak_object *object)
{
const u8 *ptr = (const u8 *)object->pointer;
size_t len;
/* limit the number of lines to HEX_MAX_LINES */
len = min_t(size_t, object->size, HEX_MAX_LINES * HEX_ROW_SIZE);
warn_or_seq_printf(seq, " hex dump (first %zu bytes):\n", len);
kasan_disable_current();
warn_or_seq_hex_dump(seq, DUMP_PREFIX_NONE, HEX_ROW_SIZE,
kasan, kmemleak: reset tags when scanning block Patch series "kasan, slub: reset tag when printing address", v3. With hardware tag-based kasan enabled, we reset the tag when we access metadata to avoid from false alarm. This patch (of 2): Kmemleak needs to scan kernel memory to check memory leak. With hardware tag-based kasan enabled, when it scans on the invalid slab and dereference, the issue will occur as below. Hardware tag-based KASAN doesn't use compiler instrumentation, we can not use kasan_disable_current() to ignore tag check. Based on the below report, there are 11 0xf7 granules, which amounts to 176 bytes, and the object is allocated from the kmalloc-256 cache. So when kmemleak accesses the last 256-176 bytes, it causes faults, as those are marked with KASAN_KMALLOC_REDZONE == KASAN_TAG_INVALID == 0xfe. Thus, we reset tags before accessing metadata to avoid from false positives. BUG: KASAN: out-of-bounds in scan_block+0x58/0x170 Read at addr f7ff0000c0074eb0 by task kmemleak/138 Pointer tag: [f7], memory tag: [fe] CPU: 7 PID: 138 Comm: kmemleak Not tainted 5.14.0-rc2-00001-g8cae8cd89f05-dirty #134 Hardware name: linux,dummy-virt (DT) Call trace: dump_backtrace+0x0/0x1b0 show_stack+0x1c/0x30 dump_stack_lvl+0x68/0x84 print_address_description+0x7c/0x2b4 kasan_report+0x138/0x38c __do_kernel_fault+0x190/0x1c4 do_tag_check_fault+0x78/0x90 do_mem_abort+0x44/0xb4 el1_abort+0x40/0x60 el1h_64_sync_handler+0xb4/0xd0 el1h_64_sync+0x78/0x7c scan_block+0x58/0x170 scan_gray_list+0xdc/0x1a0 kmemleak_scan+0x2ac/0x560 kmemleak_scan_thread+0xb0/0xe0 kthread+0x154/0x160 ret_from_fork+0x10/0x18 Allocated by task 0: kasan_save_stack+0x2c/0x60 __kasan_kmalloc+0xec/0x104 __kmalloc+0x224/0x3c4 __register_sysctl_paths+0x200/0x290 register_sysctl_table+0x2c/0x40 sysctl_init+0x20/0x34 proc_sys_init+0x3c/0x48 proc_root_init+0x80/0x9c start_kernel+0x648/0x6a4 __primary_switched+0xc0/0xc8 Freed by task 0: kasan_save_stack+0x2c/0x60 kasan_set_track+0x2c/0x40 kasan_set_free_info+0x44/0x54 ____kasan_slab_free.constprop.0+0x150/0x1b0 __kasan_slab_free+0x14/0x20 slab_free_freelist_hook+0xa4/0x1fc kfree+0x1e8/0x30c put_fs_context+0x124/0x220 vfs_kern_mount.part.0+0x60/0xd4 kern_mount+0x24/0x4c bdev_cache_init+0x70/0x9c vfs_caches_init+0xdc/0xf4 start_kernel+0x638/0x6a4 __primary_switched+0xc0/0xc8 The buggy address belongs to the object at ffff0000c0074e00 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 176 bytes inside of 256-byte region [ffff0000c0074e00, ffff0000c0074f00) The buggy address belongs to the page: page:(____ptrval____) refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x100074 head:(____ptrval____) order:2 compound_mapcount:0 compound_pincount:0 flags: 0xbfffc0000010200(slab|head|node=0|zone=2|lastcpupid=0xffff|kasantag=0x0) raw: 0bfffc0000010200 0000000000000000 dead000000000122 f5ff0000c0002300 raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff0000c0074c00: f0 f0 f0 f0 f0 f0 f0 f0 f0 fe fe fe fe fe fe fe ffff0000c0074d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe >ffff0000c0074e00: f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 fe fe fe fe fe ^ ffff0000c0074f00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe ffff0000c0075000: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== Disabling lock debugging due to kernel taint kmemleak: 181 new suspected memory leaks (see /sys/kernel/debug/kmemleak) Link: https://lkml.kernel.org/r/20210804090957.12393-1-Kuan-Ying.Lee@mediatek.com Link: https://lkml.kernel.org/r/20210804090957.12393-2-Kuan-Ying.Lee@mediatek.com Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@mediatek.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com> Cc: Marco Elver <elver@google.com> Cc: Nicholas Tang <nicholas.tang@mediatek.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Alexander Potapenko <glider@google.com> Cc: Chinwen Chang <chinwen.chang@mediatek.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-08-13 23:54:24 +00:00
HEX_GROUP_SIZE, kasan_reset_tag((void *)ptr), len, HEX_ASCII);
kasan_enable_current();
}
/*
* Object colors, encoded with count and min_count:
* - white - orphan object, not enough references to it (count < min_count)
* - gray - not orphan, not marked as false positive (min_count == 0) or
* sufficient references to it (count >= min_count)
* - black - ignore, it doesn't contain references (e.g. text section)
* (min_count == -1). No function defined for this color.
* Newly created objects don't have any color assigned (object->count == -1)
* before the next memory scan when they become white.
*/
static bool color_white(const struct kmemleak_object *object)
{
return object->count != KMEMLEAK_BLACK &&
object->count < object->min_count;
}
static bool color_gray(const struct kmemleak_object *object)
{
return object->min_count != KMEMLEAK_BLACK &&
object->count >= object->min_count;
}
/*
* Objects are considered unreferenced only if their color is white, they have
* not be deleted and have a minimum age to avoid false positives caused by
* pointers temporarily stored in CPU registers.
*/
static bool unreferenced_object(struct kmemleak_object *object)
{
return (color_white(object) && object->flags & OBJECT_ALLOCATED) &&
time_before_eq(object->jiffies + jiffies_min_age,
jiffies_last_scan);
}
/*
* Printing of the unreferenced objects information to the seq file. The
* print_unreferenced function must be called with the object->lock held.
*/
static void print_unreferenced(struct seq_file *seq,
struct kmemleak_object *object)
{
int i;
unsigned int msecs_age = jiffies_to_msecs(jiffies - object->jiffies);
warn_or_seq_printf(seq, "unreferenced object 0x%08lx (size %zu):\n",
object->pointer, object->size);
warn_or_seq_printf(seq, " comm \"%s\", pid %d, jiffies %lu (age %d.%03ds)\n",
object->comm, object->pid, object->jiffies,
msecs_age / 1000, msecs_age % 1000);
hex_dump_object(seq, object);
warn_or_seq_printf(seq, " backtrace:\n");
for (i = 0; i < object->trace_len; i++) {
void *ptr = (void *)object->trace[i];
warn_or_seq_printf(seq, " [<%p>] %pS\n", ptr, ptr);
}
}
/*
* Print the kmemleak_object information. This function is used mainly for
* debugging special cases when kmemleak operations. It must be called with
* the object->lock held.
*/
static void dump_object_info(struct kmemleak_object *object)
{
pr_notice("Object 0x%08lx (size %zu):\n",
object->pointer, object->size);
pr_notice(" comm \"%s\", pid %d, jiffies %lu\n",
object->comm, object->pid, object->jiffies);
pr_notice(" min_count = %d\n", object->min_count);
pr_notice(" count = %d\n", object->count);
pr_notice(" flags = 0x%x\n", object->flags);
pr_notice(" checksum = %u\n", object->checksum);
pr_notice(" backtrace:\n");
mm/kmemleak: Simplify stacktrace handling Replace the indirection through struct stack_trace by using the storage array based interfaces. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: linux-mm@kvack.org Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Alexander Potapenko <glider@google.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Andrey Ryabinin <aryabinin@virtuozzo.com> Cc: kasan-dev@googlegroups.com Cc: Mike Rapoport <rppt@linux.vnet.ibm.com> Cc: Akinobu Mita <akinobu.mita@gmail.com> Cc: Christoph Hellwig <hch@lst.de> Cc: iommu@lists.linux-foundation.org Cc: Robin Murphy <robin.murphy@arm.com> Cc: Marek Szyprowski <m.szyprowski@samsung.com> Cc: Johannes Thumshirn <jthumshirn@suse.de> Cc: David Sterba <dsterba@suse.com> Cc: Chris Mason <clm@fb.com> Cc: Josef Bacik <josef@toxicpanda.com> Cc: linux-btrfs@vger.kernel.org Cc: dm-devel@redhat.com Cc: Mike Snitzer <snitzer@redhat.com> Cc: Alasdair Kergon <agk@redhat.com> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: intel-gfx@lists.freedesktop.org Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: dri-devel@lists.freedesktop.org Cc: David Airlie <airlied@linux.ie> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: Tom Zanussi <tom.zanussi@linux.intel.com> Cc: Miroslav Benes <mbenes@suse.cz> Cc: linux-arch@vger.kernel.org Link: https://lkml.kernel.org/r/20190425094801.863716911@linutronix.de
2019-04-25 09:45:01 +00:00
stack_trace_print(object->trace, object->trace_len, 4);
}
/*
* Look-up a memory block metadata (kmemleak_object) in the object search
* tree based on a pointer value. If alias is 0, only values pointing to the
* beginning of the memory block are allowed. The kmemleak_lock must be held
* when calling this function.
*/
static struct kmemleak_object *lookup_object(unsigned long ptr, int alias)
{
struct rb_node *rb = object_tree_root.rb_node;
while (rb) {
struct kmemleak_object *object =
rb_entry(rb, struct kmemleak_object, rb_node);
if (ptr < object->pointer)
rb = object->rb_node.rb_left;
else if (object->pointer + object->size <= ptr)
rb = object->rb_node.rb_right;
else if (object->pointer == ptr || alias)
return object;
else {
kmemleak_warn("Found object by alias at 0x%08lx\n",
ptr);
dump_object_info(object);
break;
}
}
return NULL;
}
/*
* Increment the object use_count. Return 1 if successful or 0 otherwise. Note
* that once an object's use_count reached 0, the RCU freeing was already
* registered and the object should no longer be used. This function must be
* called under the protection of rcu_read_lock().
*/
static int get_object(struct kmemleak_object *object)
{
return atomic_inc_not_zero(&object->use_count);
}
/*
* Memory pool allocation and freeing. kmemleak_lock must not be held.
*/
static struct kmemleak_object *mem_pool_alloc(gfp_t gfp)
{
unsigned long flags;
struct kmemleak_object *object;
/* try the slab allocator first */
if (object_cache) {
object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
if (object)
return object;
}
/* slab allocation failed, try the memory pool */
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
object = list_first_entry_or_null(&mem_pool_free_list,
typeof(*object), object_list);
if (object)
list_del(&object->object_list);
else if (mem_pool_free_count)
object = &mem_pool[--mem_pool_free_count];
else
pr_warn_once("Memory pool empty, consider increasing CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE\n");
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
return object;
}
/*
* Return the object to either the slab allocator or the memory pool.
*/
static void mem_pool_free(struct kmemleak_object *object)
{
unsigned long flags;
if (object < mem_pool || object >= mem_pool + ARRAY_SIZE(mem_pool)) {
kmem_cache_free(object_cache, object);
return;
}
/* add the object to the memory pool free list */
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
list_add(&object->object_list, &mem_pool_free_list);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
}
/*
* RCU callback to free a kmemleak_object.
*/
static void free_object_rcu(struct rcu_head *rcu)
{
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
struct hlist_node *tmp;
struct kmemleak_scan_area *area;
struct kmemleak_object *object =
container_of(rcu, struct kmemleak_object, rcu);
/*
* Once use_count is 0 (guaranteed by put_object), there is no other
* code accessing this object, hence no need for locking.
*/
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
hlist_for_each_entry_safe(area, tmp, &object->area_list, node) {
hlist_del(&area->node);
kmem_cache_free(scan_area_cache, area);
}
mem_pool_free(object);
}
/*
* Decrement the object use_count. Once the count is 0, free the object using
* an RCU callback. Since put_object() may be called via the kmemleak_free() ->
* delete_object() path, the delayed RCU freeing ensures that there is no
* recursive call to the kernel allocator. Lock-less RCU object_list traversal
* is also possible.
*/
static void put_object(struct kmemleak_object *object)
{
if (!atomic_dec_and_test(&object->use_count))
return;
/* should only get here after delete_object was called */
WARN_ON(object->flags & OBJECT_ALLOCATED);
/*
* It may be too early for the RCU callbacks, however, there is no
* concurrent object_list traversal when !object_cache and all objects
* came from the memory pool. Free the object directly.
*/
if (object_cache)
call_rcu(&object->rcu, free_object_rcu);
else
free_object_rcu(&object->rcu);
}
/*
* Look up an object in the object search tree and increase its use_count.
*/
static struct kmemleak_object *find_and_get_object(unsigned long ptr, int alias)
{
unsigned long flags;
struct kmemleak_object *object;
rcu_read_lock();
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
object = lookup_object(ptr, alias);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
/* check whether the object is still available */
if (object && !get_object(object))
object = NULL;
rcu_read_unlock();
return object;
}
/*
* Remove an object from the object_tree_root and object_list. Must be called
* with the kmemleak_lock held _if_ kmemleak is still enabled.
*/
static void __remove_object(struct kmemleak_object *object)
{
rb_erase(&object->rb_node, &object_tree_root);
list_del_rcu(&object->object_list);
}
/*
* Look up an object in the object search tree and remove it from both
* object_tree_root and object_list. The returned object's use_count should be
* at least 1, as initially set by create_object().
*/
static struct kmemleak_object *find_and_remove_object(unsigned long ptr, int alias)
{
unsigned long flags;
struct kmemleak_object *object;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
object = lookup_object(ptr, alias);
if (object)
__remove_object(object);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
return object;
}
/*
* Save stack trace to the given array of MAX_TRACE size.
*/
static int __save_stack_trace(unsigned long *trace)
{
mm/kmemleak: Simplify stacktrace handling Replace the indirection through struct stack_trace by using the storage array based interfaces. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: linux-mm@kvack.org Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Alexander Potapenko <glider@google.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Andrey Ryabinin <aryabinin@virtuozzo.com> Cc: kasan-dev@googlegroups.com Cc: Mike Rapoport <rppt@linux.vnet.ibm.com> Cc: Akinobu Mita <akinobu.mita@gmail.com> Cc: Christoph Hellwig <hch@lst.de> Cc: iommu@lists.linux-foundation.org Cc: Robin Murphy <robin.murphy@arm.com> Cc: Marek Szyprowski <m.szyprowski@samsung.com> Cc: Johannes Thumshirn <jthumshirn@suse.de> Cc: David Sterba <dsterba@suse.com> Cc: Chris Mason <clm@fb.com> Cc: Josef Bacik <josef@toxicpanda.com> Cc: linux-btrfs@vger.kernel.org Cc: dm-devel@redhat.com Cc: Mike Snitzer <snitzer@redhat.com> Cc: Alasdair Kergon <agk@redhat.com> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: intel-gfx@lists.freedesktop.org Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: dri-devel@lists.freedesktop.org Cc: David Airlie <airlied@linux.ie> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Cc: Tom Zanussi <tom.zanussi@linux.intel.com> Cc: Miroslav Benes <mbenes@suse.cz> Cc: linux-arch@vger.kernel.org Link: https://lkml.kernel.org/r/20190425094801.863716911@linutronix.de
2019-04-25 09:45:01 +00:00
return stack_trace_save(trace, MAX_TRACE, 2);
}
/*
* Create the metadata (struct kmemleak_object) corresponding to an allocated
* memory block and add it to the object_list and object_tree_root.
*/
static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
int min_count, gfp_t gfp)
{
unsigned long flags;
struct kmemleak_object *object, *parent;
struct rb_node **link, *rb_parent;
unsigned long untagged_ptr;
object = mem_pool_alloc(gfp);
if (!object) {
pr_warn("Cannot allocate a kmemleak_object structure\n");
kmemleak_disable();
return NULL;
}
INIT_LIST_HEAD(&object->object_list);
INIT_LIST_HEAD(&object->gray_list);
INIT_HLIST_HEAD(&object->area_list);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_init(&object->lock);
atomic_set(&object->use_count, 1);
object->flags = OBJECT_ALLOCATED;
object->pointer = ptr;
2021-03-25 04:37:47 +00:00
object->size = kfence_ksize((void *)ptr) ?: size;
object->excess_ref = 0;
object->min_count = min_count;
object->count = 0; /* white color initially */
object->jiffies = jiffies;
object->checksum = 0;
/* task information */
if (in_irq()) {
object->pid = 0;
strncpy(object->comm, "hardirq", sizeof(object->comm));
mm/kmemleak.c: fix check for softirq context in_softirq() is a wrong predicate to check if we are in a softirq context. It also returns true if we have BH disabled, so objects are falsely stamped with "softirq" comm. The correct predicate is in_serving_softirq(). If user does cat from /sys/kernel/debug/kmemleak previously they would see this, which is clearly wrong, this is system call context (see the comm): unreferenced object 0xffff88805bd661c0 (size 64): comm "softirq", pid 0, jiffies 4294942959 (age 12.400s) hex dump (first 32 bytes): 00 00 00 00 00 00 00 00 ff ff ff ff 00 00 00 00 ................ 00 00 00 00 00 00 00 00 01 00 00 00 00 00 00 00 ................ backtrace: [<0000000007dcb30c>] kmemleak_alloc_recursive include/linux/kmemleak.h:55 [inline] [<0000000007dcb30c>] slab_post_alloc_hook mm/slab.h:439 [inline] [<0000000007dcb30c>] slab_alloc mm/slab.c:3326 [inline] [<0000000007dcb30c>] kmem_cache_alloc_trace+0x13d/0x280 mm/slab.c:3553 [<00000000969722b7>] kmalloc include/linux/slab.h:547 [inline] [<00000000969722b7>] kzalloc include/linux/slab.h:742 [inline] [<00000000969722b7>] ip_mc_add1_src net/ipv4/igmp.c:1961 [inline] [<00000000969722b7>] ip_mc_add_src+0x36b/0x400 net/ipv4/igmp.c:2085 [<00000000a4134b5f>] ip_mc_msfilter+0x22d/0x310 net/ipv4/igmp.c:2475 [<00000000d20248ad>] do_ip_setsockopt.isra.0+0x19fe/0x1c00 net/ipv4/ip_sockglue.c:957 [<000000003d367be7>] ip_setsockopt+0x3b/0xb0 net/ipv4/ip_sockglue.c:1246 [<000000003c7c76af>] udp_setsockopt+0x4e/0x90 net/ipv4/udp.c:2616 [<000000000c1aeb23>] sock_common_setsockopt+0x3e/0x50 net/core/sock.c:3130 [<000000000157b92b>] __sys_setsockopt+0x9e/0x120 net/socket.c:2078 [<00000000a9f3d058>] __do_sys_setsockopt net/socket.c:2089 [inline] [<00000000a9f3d058>] __se_sys_setsockopt net/socket.c:2086 [inline] [<00000000a9f3d058>] __x64_sys_setsockopt+0x26/0x30 net/socket.c:2086 [<000000001b8da885>] do_syscall_64+0x7c/0x1a0 arch/x86/entry/common.c:301 [<00000000ba770c62>] entry_SYSCALL_64_after_hwframe+0x44/0xa9 now they will see this: unreferenced object 0xffff88805413c800 (size 64): comm "syz-executor.4", pid 8960, jiffies 4294994003 (age 14.350s) hex dump (first 32 bytes): 00 7a 8a 57 80 88 ff ff e0 00 00 01 00 00 00 00 .z.W............ 00 00 00 00 00 00 00 00 01 00 00 00 00 00 00 00 ................ backtrace: [<00000000c5d3be64>] kmemleak_alloc_recursive include/linux/kmemleak.h:55 [inline] [<00000000c5d3be64>] slab_post_alloc_hook mm/slab.h:439 [inline] [<00000000c5d3be64>] slab_alloc mm/slab.c:3326 [inline] [<00000000c5d3be64>] kmem_cache_alloc_trace+0x13d/0x280 mm/slab.c:3553 [<0000000023865be2>] kmalloc include/linux/slab.h:547 [inline] [<0000000023865be2>] kzalloc include/linux/slab.h:742 [inline] [<0000000023865be2>] ip_mc_add1_src net/ipv4/igmp.c:1961 [inline] [<0000000023865be2>] ip_mc_add_src+0x36b/0x400 net/ipv4/igmp.c:2085 [<000000003029a9d4>] ip_mc_msfilter+0x22d/0x310 net/ipv4/igmp.c:2475 [<00000000ccd0a87c>] do_ip_setsockopt.isra.0+0x19fe/0x1c00 net/ipv4/ip_sockglue.c:957 [<00000000a85a3785>] ip_setsockopt+0x3b/0xb0 net/ipv4/ip_sockglue.c:1246 [<00000000ec13c18d>] udp_setsockopt+0x4e/0x90 net/ipv4/udp.c:2616 [<0000000052d748e3>] sock_common_setsockopt+0x3e/0x50 net/core/sock.c:3130 [<00000000512f1014>] __sys_setsockopt+0x9e/0x120 net/socket.c:2078 [<00000000181758bc>] __do_sys_setsockopt net/socket.c:2089 [inline] [<00000000181758bc>] __se_sys_setsockopt net/socket.c:2086 [inline] [<00000000181758bc>] __x64_sys_setsockopt+0x26/0x30 net/socket.c:2086 [<00000000d4b73623>] do_syscall_64+0x7c/0x1a0 arch/x86/entry/common.c:301 [<00000000c1098bec>] entry_SYSCALL_64_after_hwframe+0x44/0xa9 Link: http://lkml.kernel.org/r/20190517171507.96046-1-dvyukov@gmail.com Signed-off-by: Dmitry Vyukov <dvyukov@google.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 03:53:39 +00:00
} else if (in_serving_softirq()) {
object->pid = 0;
strncpy(object->comm, "softirq", sizeof(object->comm));
} else {
object->pid = current->pid;
/*
* There is a small chance of a race with set_task_comm(),
* however using get_task_comm() here may cause locking
* dependency issues with current->alloc_lock. In the worst
* case, the command line is not correct.
*/
strncpy(object->comm, current->comm, sizeof(object->comm));
}
/* kernel backtrace */
object->trace_len = __save_stack_trace(object->trace);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
untagged_ptr = (unsigned long)kasan_reset_tag((void *)ptr);
min_addr = min(min_addr, untagged_ptr);
max_addr = max(max_addr, untagged_ptr + size);
link = &object_tree_root.rb_node;
rb_parent = NULL;
while (*link) {
rb_parent = *link;
parent = rb_entry(rb_parent, struct kmemleak_object, rb_node);
if (ptr + size <= parent->pointer)
link = &parent->rb_node.rb_left;
else if (parent->pointer + parent->size <= ptr)
link = &parent->rb_node.rb_right;
else {
kmemleak_stop("Cannot insert 0x%lx into the object search tree (overlaps existing)\n",
ptr);
/*
* No need for parent->lock here since "parent" cannot
* be freed while the kmemleak_lock is held.
*/
dump_object_info(parent);
kmem_cache_free(object_cache, object);
object = NULL;
goto out;
}
}
rb_link_node(&object->rb_node, rb_parent, link);
rb_insert_color(&object->rb_node, &object_tree_root);
list_add_tail_rcu(&object->object_list, &object_list);
out:
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
return object;
}
/*
* Mark the object as not allocated and schedule RCU freeing via put_object().
*/
static void __delete_object(struct kmemleak_object *object)
{
unsigned long flags;
WARN_ON(!(object->flags & OBJECT_ALLOCATED));
WARN_ON(atomic_read(&object->use_count) < 1);
/*
* Locking here also ensures that the corresponding memory block
* cannot be freed when it is being scanned.
*/
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
object->flags &= ~OBJECT_ALLOCATED;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
}
/*
* Look up the metadata (struct kmemleak_object) corresponding to ptr and
* delete it.
*/
static void delete_object_full(unsigned long ptr)
{
struct kmemleak_object *object;
object = find_and_remove_object(ptr, 0);
if (!object) {
#ifdef DEBUG
kmemleak_warn("Freeing unknown object at 0x%08lx\n",
ptr);
#endif
return;
}
__delete_object(object);
}
/*
* Look up the metadata (struct kmemleak_object) corresponding to ptr and
* delete it. If the memory block is partially freed, the function may create
* additional metadata for the remaining parts of the block.
*/
static void delete_object_part(unsigned long ptr, size_t size)
{
struct kmemleak_object *object;
unsigned long start, end;
object = find_and_remove_object(ptr, 1);
if (!object) {
#ifdef DEBUG
kmemleak_warn("Partially freeing unknown object at 0x%08lx (size %zu)\n",
ptr, size);
#endif
return;
}
/*
* Create one or two objects that may result from the memory block
* split. Note that partial freeing is only done by free_bootmem() and
* this happens before kmemleak_init() is called.
*/
start = object->pointer;
end = object->pointer + object->size;
if (ptr > start)
create_object(start, ptr - start, object->min_count,
GFP_KERNEL);
if (ptr + size < end)
create_object(ptr + size, end - ptr - size, object->min_count,
GFP_KERNEL);
__delete_object(object);
}
static void __paint_it(struct kmemleak_object *object, int color)
{
object->min_count = color;
if (color == KMEMLEAK_BLACK)
object->flags |= OBJECT_NO_SCAN;
}
static void paint_it(struct kmemleak_object *object, int color)
{
unsigned long flags;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
__paint_it(object, color);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
static void paint_ptr(unsigned long ptr, int color)
{
struct kmemleak_object *object;
object = find_and_get_object(ptr, 0);
if (!object) {
kmemleak_warn("Trying to color unknown object at 0x%08lx as %s\n",
ptr,
(color == KMEMLEAK_GREY) ? "Grey" :
(color == KMEMLEAK_BLACK) ? "Black" : "Unknown");
return;
}
paint_it(object, color);
put_object(object);
}
/*
* Mark an object permanently as gray-colored so that it can no longer be
* reported as a leak. This is used in general to mark a false positive.
*/
static void make_gray_object(unsigned long ptr)
{
paint_ptr(ptr, KMEMLEAK_GREY);
}
/*
* Mark the object as black-colored so that it is ignored from scans and
* reporting.
*/
static void make_black_object(unsigned long ptr)
{
paint_ptr(ptr, KMEMLEAK_BLACK);
}
/*
* Add a scanning area to the object. If at least one such area is added,
* kmemleak will only scan these ranges rather than the whole memory block.
*/
static void add_scan_area(unsigned long ptr, size_t size, gfp_t gfp)
{
unsigned long flags;
struct kmemleak_object *object;
struct kmemleak_scan_area *area = NULL;
object = find_and_get_object(ptr, 1);
if (!object) {
kmemleak_warn("Adding scan area to unknown object at 0x%08lx\n",
ptr);
return;
}
if (scan_area_cache)
area = kmem_cache_alloc(scan_area_cache, gfp_kmemleak_mask(gfp));
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
mm: kmemleak: make the tool tolerant to struct scan_area allocation failures Patch series "mm: kmemleak: Use a memory pool for kmemleak object allocations", v3. Following the discussions on v2 of this patch(set) [1], this series takes slightly different approach: - it implements its own simple memory pool that does not rely on the slab allocator - drops the early log buffer logic entirely since it can now allocate metadata from the memory pool directly before kmemleak is fully initialised - CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE option is renamed to CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE - moves the kmemleak_init() call earlier (mm_init()) - to avoid a separate memory pool for struct scan_area, it makes the tool robust when such allocations fail as scan areas are rather an optimisation [1] http://lkml.kernel.org/r/20190727132334.9184-1-catalin.marinas@arm.com This patch (of 3): Object scan areas are an optimisation aimed to decrease the false positives and slightly improve the scanning time of large objects known to only have a few specific pointers. If a struct scan_area fails to allocate, kmemleak can still function normally by scanning the full object. Introduce an OBJECT_FULL_SCAN flag and mark objects as such when scan_area allocation fails. Link: http://lkml.kernel.org/r/20190812160642.52134-2-catalin.marinas@arm.com Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-09-23 22:33:59 +00:00
if (!area) {
pr_warn_once("Cannot allocate a scan area, scanning the full object\n");
/* mark the object for full scan to avoid false positives */
object->flags |= OBJECT_FULL_SCAN;
goto out_unlock;
}
if (size == SIZE_MAX) {
size = object->pointer + object->size - ptr;
} else if (ptr + size > object->pointer + object->size) {
kmemleak_warn("Scan area larger than object 0x%08lx\n", ptr);
dump_object_info(object);
kmem_cache_free(scan_area_cache, area);
goto out_unlock;
}
INIT_HLIST_NODE(&area->node);
area->start = ptr;
area->size = size;
hlist_add_head(&area->node, &object->area_list);
out_unlock:
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
}
/*
* Any surplus references (object already gray) to 'ptr' are passed to
* 'excess_ref'. This is used in the vmalloc() case where a pointer to
* vm_struct may be used as an alternative reference to the vmalloc'ed object
* (see free_thread_stack()).
*/
static void object_set_excess_ref(unsigned long ptr, unsigned long excess_ref)
{
unsigned long flags;
struct kmemleak_object *object;
object = find_and_get_object(ptr, 0);
if (!object) {
kmemleak_warn("Setting excess_ref on unknown object at 0x%08lx\n",
ptr);
return;
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
object->excess_ref = excess_ref;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
}
/*
* Set the OBJECT_NO_SCAN flag for the object corresponding to the give
* pointer. Such object will not be scanned by kmemleak but references to it
* are searched.
*/
static void object_no_scan(unsigned long ptr)
{
unsigned long flags;
struct kmemleak_object *object;
object = find_and_get_object(ptr, 0);
if (!object) {
kmemleak_warn("Not scanning unknown object at 0x%08lx\n", ptr);
return;
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
object->flags |= OBJECT_NO_SCAN;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
}
/**
* kmemleak_alloc - register a newly allocated object
* @ptr: pointer to beginning of the object
* @size: size of the object
* @min_count: minimum number of references to this object. If during memory
* scanning a number of references less than @min_count is found,
* the object is reported as a memory leak. If @min_count is 0,
* the object is never reported as a leak. If @min_count is -1,
* the object is ignored (not scanned and not reported as a leak)
* @gfp: kmalloc() flags used for kmemleak internal memory allocations
*
* This function is called from the kernel allocators when a new object
* (memory block) is allocated (kmem_cache_alloc, kmalloc etc.).
*/
void __ref kmemleak_alloc(const void *ptr, size_t size, int min_count,
gfp_t gfp)
{
pr_debug("%s(0x%p, %zu, %d)\n", __func__, ptr, size, min_count);
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
create_object((unsigned long)ptr, size, min_count, gfp);
}
EXPORT_SYMBOL_GPL(kmemleak_alloc);
/**
* kmemleak_alloc_percpu - register a newly allocated __percpu object
* @ptr: __percpu pointer to beginning of the object
* @size: size of the object
mm: kmemleak_alloc_percpu() should follow the gfp from per_alloc() Beginning at commit d52d3997f843 ("ipv6: Create percpu rt6_info"), the following INFO splat is logged: =============================== [ INFO: suspicious RCU usage. ] 4.1.0-rc7-next-20150612 #1 Not tainted ------------------------------- kernel/sched/core.c:7318 Illegal context switch in RCU-bh read-side critical section! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 0 3 locks held by systemd/1: #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff815f0c8f>] rtnetlink_rcv+0x1f/0x40 #1: (rcu_read_lock_bh){......}, at: [<ffffffff816a34e2>] ipv6_add_addr+0x62/0x540 #2: (addrconf_hash_lock){+...+.}, at: [<ffffffff816a3604>] ipv6_add_addr+0x184/0x540 stack backtrace: CPU: 0 PID: 1 Comm: systemd Not tainted 4.1.0-rc7-next-20150612 #1 Hardware name: TOSHIBA TECRA A50-A/TECRA A50-A, BIOS Version 4.20 04/17/2014 Call Trace: dump_stack+0x4c/0x6e lockdep_rcu_suspicious+0xe7/0x120 ___might_sleep+0x1d5/0x1f0 __might_sleep+0x4d/0x90 kmem_cache_alloc+0x47/0x250 create_object+0x39/0x2e0 kmemleak_alloc_percpu+0x61/0xe0 pcpu_alloc+0x370/0x630 Additional backtrace lines are truncated. In addition, the above splat is followed by several "BUG: sleeping function called from invalid context at mm/slub.c:1268" outputs. As suggested by Martin KaFai Lau, these are the clue to the fix. Routine kmemleak_alloc_percpu() always uses GFP_KERNEL for its allocations, whereas it should follow the gfp from its callers. Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: <stable@vger.kernel.org> [3.18+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:51 +00:00
* @gfp: flags used for kmemleak internal memory allocations
*
* This function is called from the kernel percpu allocator when a new object
mm: kmemleak_alloc_percpu() should follow the gfp from per_alloc() Beginning at commit d52d3997f843 ("ipv6: Create percpu rt6_info"), the following INFO splat is logged: =============================== [ INFO: suspicious RCU usage. ] 4.1.0-rc7-next-20150612 #1 Not tainted ------------------------------- kernel/sched/core.c:7318 Illegal context switch in RCU-bh read-side critical section! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 0 3 locks held by systemd/1: #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff815f0c8f>] rtnetlink_rcv+0x1f/0x40 #1: (rcu_read_lock_bh){......}, at: [<ffffffff816a34e2>] ipv6_add_addr+0x62/0x540 #2: (addrconf_hash_lock){+...+.}, at: [<ffffffff816a3604>] ipv6_add_addr+0x184/0x540 stack backtrace: CPU: 0 PID: 1 Comm: systemd Not tainted 4.1.0-rc7-next-20150612 #1 Hardware name: TOSHIBA TECRA A50-A/TECRA A50-A, BIOS Version 4.20 04/17/2014 Call Trace: dump_stack+0x4c/0x6e lockdep_rcu_suspicious+0xe7/0x120 ___might_sleep+0x1d5/0x1f0 __might_sleep+0x4d/0x90 kmem_cache_alloc+0x47/0x250 create_object+0x39/0x2e0 kmemleak_alloc_percpu+0x61/0xe0 pcpu_alloc+0x370/0x630 Additional backtrace lines are truncated. In addition, the above splat is followed by several "BUG: sleeping function called from invalid context at mm/slub.c:1268" outputs. As suggested by Martin KaFai Lau, these are the clue to the fix. Routine kmemleak_alloc_percpu() always uses GFP_KERNEL for its allocations, whereas it should follow the gfp from its callers. Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: <stable@vger.kernel.org> [3.18+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:51 +00:00
* (memory block) is allocated (alloc_percpu).
*/
mm: kmemleak_alloc_percpu() should follow the gfp from per_alloc() Beginning at commit d52d3997f843 ("ipv6: Create percpu rt6_info"), the following INFO splat is logged: =============================== [ INFO: suspicious RCU usage. ] 4.1.0-rc7-next-20150612 #1 Not tainted ------------------------------- kernel/sched/core.c:7318 Illegal context switch in RCU-bh read-side critical section! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 0 3 locks held by systemd/1: #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff815f0c8f>] rtnetlink_rcv+0x1f/0x40 #1: (rcu_read_lock_bh){......}, at: [<ffffffff816a34e2>] ipv6_add_addr+0x62/0x540 #2: (addrconf_hash_lock){+...+.}, at: [<ffffffff816a3604>] ipv6_add_addr+0x184/0x540 stack backtrace: CPU: 0 PID: 1 Comm: systemd Not tainted 4.1.0-rc7-next-20150612 #1 Hardware name: TOSHIBA TECRA A50-A/TECRA A50-A, BIOS Version 4.20 04/17/2014 Call Trace: dump_stack+0x4c/0x6e lockdep_rcu_suspicious+0xe7/0x120 ___might_sleep+0x1d5/0x1f0 __might_sleep+0x4d/0x90 kmem_cache_alloc+0x47/0x250 create_object+0x39/0x2e0 kmemleak_alloc_percpu+0x61/0xe0 pcpu_alloc+0x370/0x630 Additional backtrace lines are truncated. In addition, the above splat is followed by several "BUG: sleeping function called from invalid context at mm/slub.c:1268" outputs. As suggested by Martin KaFai Lau, these are the clue to the fix. Routine kmemleak_alloc_percpu() always uses GFP_KERNEL for its allocations, whereas it should follow the gfp from its callers. Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: <stable@vger.kernel.org> [3.18+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:51 +00:00
void __ref kmemleak_alloc_percpu(const void __percpu *ptr, size_t size,
gfp_t gfp)
{
unsigned int cpu;
pr_debug("%s(0x%p, %zu)\n", __func__, ptr, size);
/*
* Percpu allocations are only scanned and not reported as leaks
* (min_count is set to 0).
*/
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
for_each_possible_cpu(cpu)
create_object((unsigned long)per_cpu_ptr(ptr, cpu),
mm: kmemleak_alloc_percpu() should follow the gfp from per_alloc() Beginning at commit d52d3997f843 ("ipv6: Create percpu rt6_info"), the following INFO splat is logged: =============================== [ INFO: suspicious RCU usage. ] 4.1.0-rc7-next-20150612 #1 Not tainted ------------------------------- kernel/sched/core.c:7318 Illegal context switch in RCU-bh read-side critical section! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 0 3 locks held by systemd/1: #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff815f0c8f>] rtnetlink_rcv+0x1f/0x40 #1: (rcu_read_lock_bh){......}, at: [<ffffffff816a34e2>] ipv6_add_addr+0x62/0x540 #2: (addrconf_hash_lock){+...+.}, at: [<ffffffff816a3604>] ipv6_add_addr+0x184/0x540 stack backtrace: CPU: 0 PID: 1 Comm: systemd Not tainted 4.1.0-rc7-next-20150612 #1 Hardware name: TOSHIBA TECRA A50-A/TECRA A50-A, BIOS Version 4.20 04/17/2014 Call Trace: dump_stack+0x4c/0x6e lockdep_rcu_suspicious+0xe7/0x120 ___might_sleep+0x1d5/0x1f0 __might_sleep+0x4d/0x90 kmem_cache_alloc+0x47/0x250 create_object+0x39/0x2e0 kmemleak_alloc_percpu+0x61/0xe0 pcpu_alloc+0x370/0x630 Additional backtrace lines are truncated. In addition, the above splat is followed by several "BUG: sleeping function called from invalid context at mm/slub.c:1268" outputs. As suggested by Martin KaFai Lau, these are the clue to the fix. Routine kmemleak_alloc_percpu() always uses GFP_KERNEL for its allocations, whereas it should follow the gfp from its callers. Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Cc: Martin KaFai Lau <kafai@fb.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: <stable@vger.kernel.org> [3.18+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:51 +00:00
size, 0, gfp);
}
EXPORT_SYMBOL_GPL(kmemleak_alloc_percpu);
/**
* kmemleak_vmalloc - register a newly vmalloc'ed object
* @area: pointer to vm_struct
* @size: size of the object
* @gfp: __vmalloc() flags used for kmemleak internal memory allocations
*
* This function is called from the vmalloc() kernel allocator when a new
* object (memory block) is allocated.
*/
void __ref kmemleak_vmalloc(const struct vm_struct *area, size_t size, gfp_t gfp)
{
pr_debug("%s(0x%p, %zu)\n", __func__, area, size);
/*
* A min_count = 2 is needed because vm_struct contains a reference to
* the virtual address of the vmalloc'ed block.
*/
if (kmemleak_enabled) {
create_object((unsigned long)area->addr, size, 2, gfp);
object_set_excess_ref((unsigned long)area,
(unsigned long)area->addr);
}
}
EXPORT_SYMBOL_GPL(kmemleak_vmalloc);
/**
* kmemleak_free - unregister a previously registered object
* @ptr: pointer to beginning of the object
*
* This function is called from the kernel allocators when an object (memory
* block) is freed (kmem_cache_free, kfree, vfree etc.).
*/
void __ref kmemleak_free(const void *ptr)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
delete_object_full((unsigned long)ptr);
}
EXPORT_SYMBOL_GPL(kmemleak_free);
/**
* kmemleak_free_part - partially unregister a previously registered object
* @ptr: pointer to the beginning or inside the object. This also
* represents the start of the range to be freed
* @size: size to be unregistered
*
* This function is called when only a part of a memory block is freed
* (usually from the bootmem allocator).
*/
void __ref kmemleak_free_part(const void *ptr, size_t size)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
delete_object_part((unsigned long)ptr, size);
}
EXPORT_SYMBOL_GPL(kmemleak_free_part);
/**
* kmemleak_free_percpu - unregister a previously registered __percpu object
* @ptr: __percpu pointer to beginning of the object
*
* This function is called from the kernel percpu allocator when an object
* (memory block) is freed (free_percpu).
*/
void __ref kmemleak_free_percpu(const void __percpu *ptr)
{
unsigned int cpu;
pr_debug("%s(0x%p)\n", __func__, ptr);
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
if (kmemleak_free_enabled && ptr && !IS_ERR(ptr))
for_each_possible_cpu(cpu)
delete_object_full((unsigned long)per_cpu_ptr(ptr,
cpu));
}
EXPORT_SYMBOL_GPL(kmemleak_free_percpu);
/**
* kmemleak_update_trace - update object allocation stack trace
* @ptr: pointer to beginning of the object
*
* Override the object allocation stack trace for cases where the actual
* allocation place is not always useful.
*/
void __ref kmemleak_update_trace(const void *ptr)
{
struct kmemleak_object *object;
unsigned long flags;
pr_debug("%s(0x%p)\n", __func__, ptr);
if (!kmemleak_enabled || IS_ERR_OR_NULL(ptr))
return;
object = find_and_get_object((unsigned long)ptr, 1);
if (!object) {
#ifdef DEBUG
kmemleak_warn("Updating stack trace for unknown object at %p\n",
ptr);
#endif
return;
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
object->trace_len = __save_stack_trace(object->trace);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
}
EXPORT_SYMBOL(kmemleak_update_trace);
/**
* kmemleak_not_leak - mark an allocated object as false positive
* @ptr: pointer to beginning of the object
*
* Calling this function on an object will cause the memory block to no longer
* be reported as leak and always be scanned.
*/
void __ref kmemleak_not_leak(const void *ptr)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
make_gray_object((unsigned long)ptr);
}
EXPORT_SYMBOL(kmemleak_not_leak);
/**
* kmemleak_ignore - ignore an allocated object
* @ptr: pointer to beginning of the object
*
* Calling this function on an object will cause the memory block to be
* ignored (not scanned and not reported as a leak). This is usually done when
* it is known that the corresponding block is not a leak and does not contain
* any references to other allocated memory blocks.
*/
void __ref kmemleak_ignore(const void *ptr)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
make_black_object((unsigned long)ptr);
}
EXPORT_SYMBOL(kmemleak_ignore);
/**
* kmemleak_scan_area - limit the range to be scanned in an allocated object
* @ptr: pointer to beginning or inside the object. This also
* represents the start of the scan area
* @size: size of the scan area
* @gfp: kmalloc() flags used for kmemleak internal memory allocations
*
* This function is used when it is known that only certain parts of an object
* contain references to other objects. Kmemleak will only scan these areas
* reducing the number false negatives.
*/
void __ref kmemleak_scan_area(const void *ptr, size_t size, gfp_t gfp)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
if (kmemleak_enabled && ptr && size && !IS_ERR(ptr))
add_scan_area((unsigned long)ptr, size, gfp);
}
EXPORT_SYMBOL(kmemleak_scan_area);
/**
* kmemleak_no_scan - do not scan an allocated object
* @ptr: pointer to beginning of the object
*
* This function notifies kmemleak not to scan the given memory block. Useful
* in situations where it is known that the given object does not contain any
* references to other objects. Kmemleak will not scan such objects reducing
* the number of false negatives.
*/
void __ref kmemleak_no_scan(const void *ptr)
{
pr_debug("%s(0x%p)\n", __func__, ptr);
if (kmemleak_enabled && ptr && !IS_ERR(ptr))
object_no_scan((unsigned long)ptr);
}
EXPORT_SYMBOL(kmemleak_no_scan);
/**
* kmemleak_alloc_phys - similar to kmemleak_alloc but taking a physical
* address argument
* @phys: physical address of the object
* @size: size of the object
* @min_count: minimum number of references to this object.
* See kmemleak_alloc()
* @gfp: kmalloc() flags used for kmemleak internal memory allocations
*/
void __ref kmemleak_alloc_phys(phys_addr_t phys, size_t size, int min_count,
gfp_t gfp)
{
if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
kmemleak_alloc(__va(phys), size, min_count, gfp);
}
EXPORT_SYMBOL(kmemleak_alloc_phys);
/**
* kmemleak_free_part_phys - similar to kmemleak_free_part but taking a
* physical address argument
* @phys: physical address if the beginning or inside an object. This
* also represents the start of the range to be freed
* @size: size to be unregistered
*/
void __ref kmemleak_free_part_phys(phys_addr_t phys, size_t size)
{
if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
kmemleak_free_part(__va(phys), size);
}
EXPORT_SYMBOL(kmemleak_free_part_phys);
/**
* kmemleak_not_leak_phys - similar to kmemleak_not_leak but taking a physical
* address argument
* @phys: physical address of the object
*/
void __ref kmemleak_not_leak_phys(phys_addr_t phys)
{
if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
kmemleak_not_leak(__va(phys));
}
EXPORT_SYMBOL(kmemleak_not_leak_phys);
/**
* kmemleak_ignore_phys - similar to kmemleak_ignore but taking a physical
* address argument
* @phys: physical address of the object
*/
void __ref kmemleak_ignore_phys(phys_addr_t phys)
{
if (!IS_ENABLED(CONFIG_HIGHMEM) || PHYS_PFN(phys) < max_low_pfn)
kmemleak_ignore(__va(phys));
}
EXPORT_SYMBOL(kmemleak_ignore_phys);
/*
* Update an object's checksum and return true if it was modified.
*/
static bool update_checksum(struct kmemleak_object *object)
{
u32 old_csum = object->checksum;
kasan_disable_current();
mm/kmemleak: silence KCSAN splats in checksum Even if KCSAN is disabled for kmemleak, update_checksum() could still call crc32() (which is outside of kmemleak.c) to dereference object->pointer. Thus, the value of object->pointer could be accessed concurrently as noticed by KCSAN, BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12: do_raw_spin_lock+0x114/0x200 debug_spin_lock_after at kernel/locking/spinlock_debug.c:91 (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115 _raw_spin_lock+0x40/0x50 __handle_mm_fault+0xa9e/0xd00 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x6f9 page_fault+0x34/0x40 read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60: crc32_le_base+0x67/0x350 crc32_le_base+0x67/0x350: crc32_body at lib/crc32.c:106 (inlined by) crc32_le_generic at lib/crc32.c:179 (inlined by) crc32_le at lib/crc32.c:197 kmemleak_scan+0x528/0xd90 update_checksum at mm/kmemleak.c:1172 (inlined by) kmemleak_scan at mm/kmemleak.c:1497 kmemleak_scan_thread+0xcc/0xfa kthread+0x1e0/0x200 ret_from_fork+0x27/0x50 If a shattered value was returned due to a data race, it will be corrected in the next scan. Thus, let KCSAN ignore all reads in the region to silence KCSAN in case the write side is non-atomic. Suggested-by: Marco Elver <elver@google.com> Signed-off-by: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Acked-by: Marco Elver <elver@google.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Link: http://lkml.kernel.org/r/20200317182754.2180-1-cai@lca.pw Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-08-15 00:31:14 +00:00
kcsan_disable_current();
kasan, kmemleak: reset tags when scanning block Patch series "kasan, slub: reset tag when printing address", v3. With hardware tag-based kasan enabled, we reset the tag when we access metadata to avoid from false alarm. This patch (of 2): Kmemleak needs to scan kernel memory to check memory leak. With hardware tag-based kasan enabled, when it scans on the invalid slab and dereference, the issue will occur as below. Hardware tag-based KASAN doesn't use compiler instrumentation, we can not use kasan_disable_current() to ignore tag check. Based on the below report, there are 11 0xf7 granules, which amounts to 176 bytes, and the object is allocated from the kmalloc-256 cache. So when kmemleak accesses the last 256-176 bytes, it causes faults, as those are marked with KASAN_KMALLOC_REDZONE == KASAN_TAG_INVALID == 0xfe. Thus, we reset tags before accessing metadata to avoid from false positives. BUG: KASAN: out-of-bounds in scan_block+0x58/0x170 Read at addr f7ff0000c0074eb0 by task kmemleak/138 Pointer tag: [f7], memory tag: [fe] CPU: 7 PID: 138 Comm: kmemleak Not tainted 5.14.0-rc2-00001-g8cae8cd89f05-dirty #134 Hardware name: linux,dummy-virt (DT) Call trace: dump_backtrace+0x0/0x1b0 show_stack+0x1c/0x30 dump_stack_lvl+0x68/0x84 print_address_description+0x7c/0x2b4 kasan_report+0x138/0x38c __do_kernel_fault+0x190/0x1c4 do_tag_check_fault+0x78/0x90 do_mem_abort+0x44/0xb4 el1_abort+0x40/0x60 el1h_64_sync_handler+0xb4/0xd0 el1h_64_sync+0x78/0x7c scan_block+0x58/0x170 scan_gray_list+0xdc/0x1a0 kmemleak_scan+0x2ac/0x560 kmemleak_scan_thread+0xb0/0xe0 kthread+0x154/0x160 ret_from_fork+0x10/0x18 Allocated by task 0: kasan_save_stack+0x2c/0x60 __kasan_kmalloc+0xec/0x104 __kmalloc+0x224/0x3c4 __register_sysctl_paths+0x200/0x290 register_sysctl_table+0x2c/0x40 sysctl_init+0x20/0x34 proc_sys_init+0x3c/0x48 proc_root_init+0x80/0x9c start_kernel+0x648/0x6a4 __primary_switched+0xc0/0xc8 Freed by task 0: kasan_save_stack+0x2c/0x60 kasan_set_track+0x2c/0x40 kasan_set_free_info+0x44/0x54 ____kasan_slab_free.constprop.0+0x150/0x1b0 __kasan_slab_free+0x14/0x20 slab_free_freelist_hook+0xa4/0x1fc kfree+0x1e8/0x30c put_fs_context+0x124/0x220 vfs_kern_mount.part.0+0x60/0xd4 kern_mount+0x24/0x4c bdev_cache_init+0x70/0x9c vfs_caches_init+0xdc/0xf4 start_kernel+0x638/0x6a4 __primary_switched+0xc0/0xc8 The buggy address belongs to the object at ffff0000c0074e00 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 176 bytes inside of 256-byte region [ffff0000c0074e00, ffff0000c0074f00) The buggy address belongs to the page: page:(____ptrval____) refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x100074 head:(____ptrval____) order:2 compound_mapcount:0 compound_pincount:0 flags: 0xbfffc0000010200(slab|head|node=0|zone=2|lastcpupid=0xffff|kasantag=0x0) raw: 0bfffc0000010200 0000000000000000 dead000000000122 f5ff0000c0002300 raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff0000c0074c00: f0 f0 f0 f0 f0 f0 f0 f0 f0 fe fe fe fe fe fe fe ffff0000c0074d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe >ffff0000c0074e00: f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 fe fe fe fe fe ^ ffff0000c0074f00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe ffff0000c0075000: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== Disabling lock debugging due to kernel taint kmemleak: 181 new suspected memory leaks (see /sys/kernel/debug/kmemleak) Link: https://lkml.kernel.org/r/20210804090957.12393-1-Kuan-Ying.Lee@mediatek.com Link: https://lkml.kernel.org/r/20210804090957.12393-2-Kuan-Ying.Lee@mediatek.com Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@mediatek.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com> Cc: Marco Elver <elver@google.com> Cc: Nicholas Tang <nicholas.tang@mediatek.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Alexander Potapenko <glider@google.com> Cc: Chinwen Chang <chinwen.chang@mediatek.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-08-13 23:54:24 +00:00
object->checksum = crc32(0, kasan_reset_tag((void *)object->pointer), object->size);
kasan_enable_current();
mm/kmemleak: silence KCSAN splats in checksum Even if KCSAN is disabled for kmemleak, update_checksum() could still call crc32() (which is outside of kmemleak.c) to dereference object->pointer. Thus, the value of object->pointer could be accessed concurrently as noticed by KCSAN, BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12: do_raw_spin_lock+0x114/0x200 debug_spin_lock_after at kernel/locking/spinlock_debug.c:91 (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115 _raw_spin_lock+0x40/0x50 __handle_mm_fault+0xa9e/0xd00 handle_mm_fault+0xfc/0x2f0 do_page_fault+0x263/0x6f9 page_fault+0x34/0x40 read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60: crc32_le_base+0x67/0x350 crc32_le_base+0x67/0x350: crc32_body at lib/crc32.c:106 (inlined by) crc32_le_generic at lib/crc32.c:179 (inlined by) crc32_le at lib/crc32.c:197 kmemleak_scan+0x528/0xd90 update_checksum at mm/kmemleak.c:1172 (inlined by) kmemleak_scan at mm/kmemleak.c:1497 kmemleak_scan_thread+0xcc/0xfa kthread+0x1e0/0x200 ret_from_fork+0x27/0x50 If a shattered value was returned due to a data race, it will be corrected in the next scan. Thus, let KCSAN ignore all reads in the region to silence KCSAN in case the write side is non-atomic. Suggested-by: Marco Elver <elver@google.com> Signed-off-by: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Acked-by: Marco Elver <elver@google.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Link: http://lkml.kernel.org/r/20200317182754.2180-1-cai@lca.pw Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-08-15 00:31:14 +00:00
kcsan_enable_current();
return object->checksum != old_csum;
}
/*
* Update an object's references. object->lock must be held by the caller.
*/
static void update_refs(struct kmemleak_object *object)
{
if (!color_white(object)) {
/* non-orphan, ignored or new */
return;
}
/*
* Increase the object's reference count (number of pointers to the
* memory block). If this count reaches the required minimum, the
* object's color will become gray and it will be added to the
* gray_list.
*/
object->count++;
if (color_gray(object)) {
/* put_object() called when removing from gray_list */
WARN_ON(!get_object(object));
list_add_tail(&object->gray_list, &gray_list);
}
}
/*
* Memory scanning is a long process and it needs to be interruptible. This
* function checks whether such interrupt condition occurred.
*/
static int scan_should_stop(void)
{
if (!kmemleak_enabled)
return 1;
/*
* This function may be called from either process or kthread context,
* hence the need to check for both stop conditions.
*/
if (current->mm)
return signal_pending(current);
else
return kthread_should_stop();
return 0;
}
/*
* Scan a memory block (exclusive range) for valid pointers and add those
* found to the gray list.
*/
static void scan_block(void *_start, void *_end,
struct kmemleak_object *scanned)
{
unsigned long *ptr;
unsigned long *start = PTR_ALIGN(_start, BYTES_PER_POINTER);
unsigned long *end = _end - (BYTES_PER_POINTER - 1);
unsigned long flags;
unsigned long untagged_ptr;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&kmemleak_lock, flags);
for (ptr = start; ptr < end; ptr++) {
struct kmemleak_object *object;
kmemleak: Don't scan uninitialized memory when kmemcheck is enabled Ingo Molnar reported the following kmemcheck warning when running both kmemleak and kmemcheck enabled: PM: Adding info for No Bus:vcsa7 WARNING: kmemcheck: Caught 32-bit read from uninitialized memory (f6f6e1a4) d873f9f600000000c42ae4c1005c87f70000000070665f666978656400000000 i i i i u u u u i i i i i i i i i i i i i i i i i i i i i u u u ^ Pid: 3091, comm: kmemleak Not tainted (2.6.31-rc7-tip #1303) P4DC6 EIP: 0060:[<c110301f>] EFLAGS: 00010006 CPU: 0 EIP is at scan_block+0x3f/0xe0 EAX: f40bd700 EBX: f40bd780 ECX: f16b46c0 EDX: 00000001 ESI: f6f6e1a4 EDI: 00000000 EBP: f10f3f4c ESP: c2605fcc DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 CR0: 8005003b CR2: e89a4844 CR3: 30ff1000 CR4: 000006f0 DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000 DR6: ffff4ff0 DR7: 00000400 [<c110313c>] scan_object+0x7c/0xf0 [<c1103389>] kmemleak_scan+0x1d9/0x400 [<c1103a3c>] kmemleak_scan_thread+0x4c/0xb0 [<c10819d4>] kthread+0x74/0x80 [<c10257db>] kernel_thread_helper+0x7/0x3c [<ffffffff>] 0xffffffff kmemleak: 515 new suspected memory leaks (see /sys/kernel/debug/kmemleak) kmemleak: 42 new suspected memory leaks (see /sys/kernel/debug/kmemleak) The problem here is that kmemleak will scan partially initialized objects that makes kmemcheck complain. Fix that up by skipping uninitialized memory regions when kmemcheck is enabled. Reported-by: Ingo Molnar <mingo@elte.hu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-08-27 13:50:00 +00:00
unsigned long pointer;
unsigned long excess_ref;
if (scan_should_stop())
break;
kasan_disable_current();
kasan, kmemleak: reset tags when scanning block Patch series "kasan, slub: reset tag when printing address", v3. With hardware tag-based kasan enabled, we reset the tag when we access metadata to avoid from false alarm. This patch (of 2): Kmemleak needs to scan kernel memory to check memory leak. With hardware tag-based kasan enabled, when it scans on the invalid slab and dereference, the issue will occur as below. Hardware tag-based KASAN doesn't use compiler instrumentation, we can not use kasan_disable_current() to ignore tag check. Based on the below report, there are 11 0xf7 granules, which amounts to 176 bytes, and the object is allocated from the kmalloc-256 cache. So when kmemleak accesses the last 256-176 bytes, it causes faults, as those are marked with KASAN_KMALLOC_REDZONE == KASAN_TAG_INVALID == 0xfe. Thus, we reset tags before accessing metadata to avoid from false positives. BUG: KASAN: out-of-bounds in scan_block+0x58/0x170 Read at addr f7ff0000c0074eb0 by task kmemleak/138 Pointer tag: [f7], memory tag: [fe] CPU: 7 PID: 138 Comm: kmemleak Not tainted 5.14.0-rc2-00001-g8cae8cd89f05-dirty #134 Hardware name: linux,dummy-virt (DT) Call trace: dump_backtrace+0x0/0x1b0 show_stack+0x1c/0x30 dump_stack_lvl+0x68/0x84 print_address_description+0x7c/0x2b4 kasan_report+0x138/0x38c __do_kernel_fault+0x190/0x1c4 do_tag_check_fault+0x78/0x90 do_mem_abort+0x44/0xb4 el1_abort+0x40/0x60 el1h_64_sync_handler+0xb4/0xd0 el1h_64_sync+0x78/0x7c scan_block+0x58/0x170 scan_gray_list+0xdc/0x1a0 kmemleak_scan+0x2ac/0x560 kmemleak_scan_thread+0xb0/0xe0 kthread+0x154/0x160 ret_from_fork+0x10/0x18 Allocated by task 0: kasan_save_stack+0x2c/0x60 __kasan_kmalloc+0xec/0x104 __kmalloc+0x224/0x3c4 __register_sysctl_paths+0x200/0x290 register_sysctl_table+0x2c/0x40 sysctl_init+0x20/0x34 proc_sys_init+0x3c/0x48 proc_root_init+0x80/0x9c start_kernel+0x648/0x6a4 __primary_switched+0xc0/0xc8 Freed by task 0: kasan_save_stack+0x2c/0x60 kasan_set_track+0x2c/0x40 kasan_set_free_info+0x44/0x54 ____kasan_slab_free.constprop.0+0x150/0x1b0 __kasan_slab_free+0x14/0x20 slab_free_freelist_hook+0xa4/0x1fc kfree+0x1e8/0x30c put_fs_context+0x124/0x220 vfs_kern_mount.part.0+0x60/0xd4 kern_mount+0x24/0x4c bdev_cache_init+0x70/0x9c vfs_caches_init+0xdc/0xf4 start_kernel+0x638/0x6a4 __primary_switched+0xc0/0xc8 The buggy address belongs to the object at ffff0000c0074e00 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 176 bytes inside of 256-byte region [ffff0000c0074e00, ffff0000c0074f00) The buggy address belongs to the page: page:(____ptrval____) refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x100074 head:(____ptrval____) order:2 compound_mapcount:0 compound_pincount:0 flags: 0xbfffc0000010200(slab|head|node=0|zone=2|lastcpupid=0xffff|kasantag=0x0) raw: 0bfffc0000010200 0000000000000000 dead000000000122 f5ff0000c0002300 raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff0000c0074c00: f0 f0 f0 f0 f0 f0 f0 f0 f0 fe fe fe fe fe fe fe ffff0000c0074d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe >ffff0000c0074e00: f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 fe fe fe fe fe ^ ffff0000c0074f00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe ffff0000c0075000: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ================================================================== Disabling lock debugging due to kernel taint kmemleak: 181 new suspected memory leaks (see /sys/kernel/debug/kmemleak) Link: https://lkml.kernel.org/r/20210804090957.12393-1-Kuan-Ying.Lee@mediatek.com Link: https://lkml.kernel.org/r/20210804090957.12393-2-Kuan-Ying.Lee@mediatek.com Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@mediatek.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com> Cc: Marco Elver <elver@google.com> Cc: Nicholas Tang <nicholas.tang@mediatek.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Alexander Potapenko <glider@google.com> Cc: Chinwen Chang <chinwen.chang@mediatek.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-08-13 23:54:24 +00:00
pointer = *(unsigned long *)kasan_reset_tag((void *)ptr);
kasan_enable_current();
kmemleak: Don't scan uninitialized memory when kmemcheck is enabled Ingo Molnar reported the following kmemcheck warning when running both kmemleak and kmemcheck enabled: PM: Adding info for No Bus:vcsa7 WARNING: kmemcheck: Caught 32-bit read from uninitialized memory (f6f6e1a4) d873f9f600000000c42ae4c1005c87f70000000070665f666978656400000000 i i i i u u u u i i i i i i i i i i i i i i i i i i i i i u u u ^ Pid: 3091, comm: kmemleak Not tainted (2.6.31-rc7-tip #1303) P4DC6 EIP: 0060:[<c110301f>] EFLAGS: 00010006 CPU: 0 EIP is at scan_block+0x3f/0xe0 EAX: f40bd700 EBX: f40bd780 ECX: f16b46c0 EDX: 00000001 ESI: f6f6e1a4 EDI: 00000000 EBP: f10f3f4c ESP: c2605fcc DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 CR0: 8005003b CR2: e89a4844 CR3: 30ff1000 CR4: 000006f0 DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000 DR6: ffff4ff0 DR7: 00000400 [<c110313c>] scan_object+0x7c/0xf0 [<c1103389>] kmemleak_scan+0x1d9/0x400 [<c1103a3c>] kmemleak_scan_thread+0x4c/0xb0 [<c10819d4>] kthread+0x74/0x80 [<c10257db>] kernel_thread_helper+0x7/0x3c [<ffffffff>] 0xffffffff kmemleak: 515 new suspected memory leaks (see /sys/kernel/debug/kmemleak) kmemleak: 42 new suspected memory leaks (see /sys/kernel/debug/kmemleak) The problem here is that kmemleak will scan partially initialized objects that makes kmemcheck complain. Fix that up by skipping uninitialized memory regions when kmemcheck is enabled. Reported-by: Ingo Molnar <mingo@elte.hu> Acked-by: Ingo Molnar <mingo@elte.hu> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-08-27 13:50:00 +00:00
untagged_ptr = (unsigned long)kasan_reset_tag((void *)pointer);
if (untagged_ptr < min_addr || untagged_ptr >= max_addr)
continue;
/*
* No need for get_object() here since we hold kmemleak_lock.
* object->use_count cannot be dropped to 0 while the object
* is still present in object_tree_root and object_list
* (with updates protected by kmemleak_lock).
*/
object = lookup_object(pointer, 1);
if (!object)
continue;
if (object == scanned)
/* self referenced, ignore */
continue;
/*
* Avoid the lockdep recursive warning on object->lock being
* previously acquired in scan_object(). These locks are
* enclosed by scan_mutex.
*/
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
/* only pass surplus references (object already gray) */
if (color_gray(object)) {
excess_ref = object->excess_ref;
/* no need for update_refs() if object already gray */
} else {
excess_ref = 0;
update_refs(object);
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock(&object->lock);
if (excess_ref) {
object = lookup_object(excess_ref, 0);
if (!object)
continue;
if (object == scanned)
/* circular reference, ignore */
continue;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_nested(&object->lock, SINGLE_DEPTH_NESTING);
update_refs(object);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock(&object->lock);
}
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&kmemleak_lock, flags);
}
/*
* Scan a large memory block in MAX_SCAN_SIZE chunks to reduce the latency.
*/
#ifdef CONFIG_SMP
static void scan_large_block(void *start, void *end)
{
void *next;
while (start < end) {
next = min(start + MAX_SCAN_SIZE, end);
scan_block(start, next, NULL);
start = next;
cond_resched();
}
}
#endif
/*
* Scan a memory block corresponding to a kmemleak_object. A condition is
* that object->use_count >= 1.
*/
static void scan_object(struct kmemleak_object *object)
{
struct kmemleak_scan_area *area;
unsigned long flags;
/*
* Once the object->lock is acquired, the corresponding memory block
* cannot be freed (the same lock is acquired in delete_object).
*/
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
if (object->flags & OBJECT_NO_SCAN)
goto out;
if (!(object->flags & OBJECT_ALLOCATED))
/* already freed object */
goto out;
mm: kmemleak: make the tool tolerant to struct scan_area allocation failures Patch series "mm: kmemleak: Use a memory pool for kmemleak object allocations", v3. Following the discussions on v2 of this patch(set) [1], this series takes slightly different approach: - it implements its own simple memory pool that does not rely on the slab allocator - drops the early log buffer logic entirely since it can now allocate metadata from the memory pool directly before kmemleak is fully initialised - CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE option is renamed to CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE - moves the kmemleak_init() call earlier (mm_init()) - to avoid a separate memory pool for struct scan_area, it makes the tool robust when such allocations fail as scan areas are rather an optimisation [1] http://lkml.kernel.org/r/20190727132334.9184-1-catalin.marinas@arm.com This patch (of 3): Object scan areas are an optimisation aimed to decrease the false positives and slightly improve the scanning time of large objects known to only have a few specific pointers. If a struct scan_area fails to allocate, kmemleak can still function normally by scanning the full object. Introduce an OBJECT_FULL_SCAN flag and mark objects as such when scan_area allocation fails. Link: http://lkml.kernel.org/r/20190812160642.52134-2-catalin.marinas@arm.com Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-09-23 22:33:59 +00:00
if (hlist_empty(&object->area_list) ||
object->flags & OBJECT_FULL_SCAN) {
void *start = (void *)object->pointer;
void *end = (void *)(object->pointer + object->size);
void *next;
do {
next = min(start + MAX_SCAN_SIZE, end);
scan_block(start, next, object);
start = next;
if (start >= end)
break;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
cond_resched();
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
} while (object->flags & OBJECT_ALLOCATED);
} else
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 01:06:00 +00:00
hlist_for_each_entry(area, &object->area_list, node)
scan_block((void *)area->start,
(void *)(area->start + area->size),
object);
out:
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
/*
* Scan the objects already referenced (gray objects). More objects will be
* referenced and, if there are no memory leaks, all the objects are scanned.
*/
static void scan_gray_list(void)
{
struct kmemleak_object *object, *tmp;
/*
* The list traversal is safe for both tail additions and removals
* from inside the loop. The kmemleak objects cannot be freed from
* outside the loop because their use_count was incremented.
*/
object = list_entry(gray_list.next, typeof(*object), gray_list);
while (&object->gray_list != &gray_list) {
cond_resched();
/* may add new objects to the list */
if (!scan_should_stop())
scan_object(object);
tmp = list_entry(object->gray_list.next, typeof(*object),
gray_list);
/* remove the object from the list and release it */
list_del(&object->gray_list);
put_object(object);
object = tmp;
}
WARN_ON(!list_empty(&gray_list));
}
/*
* Scan data sections and all the referenced memory blocks allocated via the
* kernel's standard allocators. This function must be called with the
* scan_mutex held.
*/
static void kmemleak_scan(void)
{
unsigned long flags;
struct kmemleak_object *object;
int i;
int new_leaks = 0;
jiffies_last_scan = jiffies;
/* prepare the kmemleak_object's */
rcu_read_lock();
list_for_each_entry_rcu(object, &object_list, object_list) {
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
#ifdef DEBUG
/*
* With a few exceptions there should be a maximum of
* 1 reference to any object at this point.
*/
if (atomic_read(&object->use_count) > 1) {
pr_debug("object->use_count = %d\n",
atomic_read(&object->use_count));
dump_object_info(object);
}
#endif
/* reset the reference count (whiten the object) */
object->count = 0;
if (color_gray(object) && get_object(object))
list_add_tail(&object->gray_list, &gray_list);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
rcu_read_unlock();
#ifdef CONFIG_SMP
/* per-cpu sections scanning */
for_each_possible_cpu(i)
scan_large_block(__per_cpu_start + per_cpu_offset(i),
__per_cpu_end + per_cpu_offset(i));
#endif
/*
* Struct page scanning for each node.
*/
mem-hotplug: implement get/put_online_mems kmem_cache_{create,destroy,shrink} need to get a stable value of cpu/node online mask, because they init/destroy/access per-cpu/node kmem_cache parts, which can be allocated or destroyed on cpu/mem hotplug. To protect against cpu hotplug, these functions use {get,put}_online_cpus. However, they do nothing to synchronize with memory hotplug - taking the slab_mutex does not eliminate the possibility of race as described in patch 2. What we need there is something like get_online_cpus, but for memory. We already have lock_memory_hotplug, which serves for the purpose, but it's a bit of a hammer right now, because it's backed by a mutex. As a result, it imposes some limitations to locking order, which are not desirable, and can't be used just like get_online_cpus. That's why in patch 1 I substitute it with get/put_online_mems, which work exactly like get/put_online_cpus except they block not cpu, but memory hotplug. [ v1 can be found at https://lkml.org/lkml/2014/4/6/68. I NAK'ed it by myself, because it used an rw semaphore for get/put_online_mems, making them dead lock prune. ] This patch (of 2): {un}lock_memory_hotplug, which is used to synchronize against memory hotplug, is currently backed by a mutex, which makes it a bit of a hammer - threads that only want to get a stable value of online nodes mask won't be able to proceed concurrently. Also, it imposes some strong locking ordering rules on it, which narrows down the set of its usage scenarios. This patch introduces get/put_online_mems, which are the same as get/put_online_cpus, but for memory hotplug, i.e. executing a code inside a get/put_online_mems section will guarantee a stable value of online nodes, present pages, etc. lock_memory_hotplug()/unlock_memory_hotplug() are removed altogether. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Tang Chen <tangchen@cn.fujitsu.com> Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com> Cc: Toshi Kani <toshi.kani@hp.com> Cc: Xishi Qiu <qiuxishi@huawei.com> Cc: Jiang Liu <liuj97@gmail.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: David Rientjes <rientjes@google.com> Cc: Wen Congyang <wency@cn.fujitsu.com> Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-04 23:07:18 +00:00
get_online_mems();
for_each_online_node(i) {
unsigned long start_pfn = node_start_pfn(i);
unsigned long end_pfn = node_end_pfn(i);
unsigned long pfn;
for (pfn = start_pfn; pfn < end_pfn; pfn++) {
mm, kmemleak: little optimization while scanning kmemleak_scan() goes through all online nodes and tries to scan all used pages. We can do better and use pfn_to_online_page(), so in case we have CONFIG_MEMORY_HOTPLUG, offlined pages will be skiped automatically. For boxes where CONFIG_MEMORY_HOTPLUG is not present, pfn_to_online_page() will fallback to pfn_valid(). Another little optimization is to check if the page belongs to the node we are currently checking, so in case we have nodes interleaved we will not check the same pfn multiple times. I ran some tests: Add some memory to node1 and node2 making it interleaved: (qemu) object_add memory-backend-ram,id=ram0,size=1G (qemu) device_add pc-dimm,id=dimm0,memdev=ram0,node=1 (qemu) object_add memory-backend-ram,id=ram1,size=1G (qemu) device_add pc-dimm,id=dimm1,memdev=ram1,node=2 (qemu) object_add memory-backend-ram,id=ram2,size=1G (qemu) device_add pc-dimm,id=dimm2,memdev=ram2,node=1 Then, we offline that memory: # for i in {32..39} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;done # for i in {48..55} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;don # for i in {40..47} ; do echo "offline" > /sys/devices/system/node/node2/memory$i/state;done And we run kmemleak_scan: # echo "scan" > /sys/kernel/debug/kmemleak before the patch: kmemleak: time spend: 41596 us after the patch: kmemleak: time spend: 34899 us [akpm@linux-foundation.org: remove stray newline, per Oscar] Link: http://lkml.kernel.org/r/20181206131918.25099-1-osalvador@suse.de Signed-off-by: Oscar Salvador <osalvador@suse.de> Reviewed-by: Wei Yang <richard.weiyang@gmail.com> Suggested-by: Michal Hocko <mhocko@suse.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 08:37:57 +00:00
struct page *page = pfn_to_online_page(pfn);
mm, kmemleak: little optimization while scanning kmemleak_scan() goes through all online nodes and tries to scan all used pages. We can do better and use pfn_to_online_page(), so in case we have CONFIG_MEMORY_HOTPLUG, offlined pages will be skiped automatically. For boxes where CONFIG_MEMORY_HOTPLUG is not present, pfn_to_online_page() will fallback to pfn_valid(). Another little optimization is to check if the page belongs to the node we are currently checking, so in case we have nodes interleaved we will not check the same pfn multiple times. I ran some tests: Add some memory to node1 and node2 making it interleaved: (qemu) object_add memory-backend-ram,id=ram0,size=1G (qemu) device_add pc-dimm,id=dimm0,memdev=ram0,node=1 (qemu) object_add memory-backend-ram,id=ram1,size=1G (qemu) device_add pc-dimm,id=dimm1,memdev=ram1,node=2 (qemu) object_add memory-backend-ram,id=ram2,size=1G (qemu) device_add pc-dimm,id=dimm2,memdev=ram2,node=1 Then, we offline that memory: # for i in {32..39} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;done # for i in {48..55} ; do echo "offline" > /sys/devices/system/node/node1/memory$i/state;don # for i in {40..47} ; do echo "offline" > /sys/devices/system/node/node2/memory$i/state;done And we run kmemleak_scan: # echo "scan" > /sys/kernel/debug/kmemleak before the patch: kmemleak: time spend: 41596 us after the patch: kmemleak: time spend: 34899 us [akpm@linux-foundation.org: remove stray newline, per Oscar] Link: http://lkml.kernel.org/r/20181206131918.25099-1-osalvador@suse.de Signed-off-by: Oscar Salvador <osalvador@suse.de> Reviewed-by: Wei Yang <richard.weiyang@gmail.com> Suggested-by: Michal Hocko <mhocko@suse.com> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 08:37:57 +00:00
if (!page)
continue;
/* only scan pages belonging to this node */
if (page_to_nid(page) != i)
continue;
/* only scan if page is in use */
if (page_count(page) == 0)
continue;
scan_block(page, page + 1, NULL);
if (!(pfn & 63))
cond_resched();
}
}
mem-hotplug: implement get/put_online_mems kmem_cache_{create,destroy,shrink} need to get a stable value of cpu/node online mask, because they init/destroy/access per-cpu/node kmem_cache parts, which can be allocated or destroyed on cpu/mem hotplug. To protect against cpu hotplug, these functions use {get,put}_online_cpus. However, they do nothing to synchronize with memory hotplug - taking the slab_mutex does not eliminate the possibility of race as described in patch 2. What we need there is something like get_online_cpus, but for memory. We already have lock_memory_hotplug, which serves for the purpose, but it's a bit of a hammer right now, because it's backed by a mutex. As a result, it imposes some limitations to locking order, which are not desirable, and can't be used just like get_online_cpus. That's why in patch 1 I substitute it with get/put_online_mems, which work exactly like get/put_online_cpus except they block not cpu, but memory hotplug. [ v1 can be found at https://lkml.org/lkml/2014/4/6/68. I NAK'ed it by myself, because it used an rw semaphore for get/put_online_mems, making them dead lock prune. ] This patch (of 2): {un}lock_memory_hotplug, which is used to synchronize against memory hotplug, is currently backed by a mutex, which makes it a bit of a hammer - threads that only want to get a stable value of online nodes mask won't be able to proceed concurrently. Also, it imposes some strong locking ordering rules on it, which narrows down the set of its usage scenarios. This patch introduces get/put_online_mems, which are the same as get/put_online_cpus, but for memory hotplug, i.e. executing a code inside a get/put_online_mems section will guarantee a stable value of online nodes, present pages, etc. lock_memory_hotplug()/unlock_memory_hotplug() are removed altogether. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Tang Chen <tangchen@cn.fujitsu.com> Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com> Cc: Toshi Kani <toshi.kani@hp.com> Cc: Xishi Qiu <qiuxishi@huawei.com> Cc: Jiang Liu <liuj97@gmail.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: David Rientjes <rientjes@google.com> Cc: Wen Congyang <wency@cn.fujitsu.com> Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-04 23:07:18 +00:00
put_online_mems();
/*
* Scanning the task stacks (may introduce false negatives).
*/
if (kmemleak_stack_scan) {
struct task_struct *p, *g;
2020-10-13 23:48:50 +00:00
rcu_read_lock();
for_each_process_thread(g, p) {
void *stack = try_get_task_stack(p);
if (stack) {
scan_block(stack, stack + THREAD_SIZE, NULL);
put_task_stack(p);
}
2020-10-13 23:48:50 +00:00
}
rcu_read_unlock();
}
/*
* Scan the objects already referenced from the sections scanned
* above.
*/
scan_gray_list();
/*
* Check for new or unreferenced objects modified since the previous
* scan and color them gray until the next scan.
*/
rcu_read_lock();
list_for_each_entry_rcu(object, &object_list, object_list) {
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
if (color_white(object) && (object->flags & OBJECT_ALLOCATED)
&& update_checksum(object) && get_object(object)) {
/* color it gray temporarily */
object->count = object->min_count;
list_add_tail(&object->gray_list, &gray_list);
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
rcu_read_unlock();
/*
* Re-scan the gray list for modified unreferenced objects.
*/
scan_gray_list();
/*
* If scanning was stopped do not report any new unreferenced objects.
*/
if (scan_should_stop())
return;
/*
* Scanning result reporting.
*/
rcu_read_lock();
list_for_each_entry_rcu(object, &object_list, object_list) {
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
if (unreferenced_object(object) &&
!(object->flags & OBJECT_REPORTED)) {
object->flags |= OBJECT_REPORTED;
if (kmemleak_verbose)
print_unreferenced(NULL, object);
new_leaks++;
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
rcu_read_unlock();
if (new_leaks) {
kmemleak_found_leaks = true;
pr_info("%d new suspected memory leaks (see /sys/kernel/debug/kmemleak)\n",
new_leaks);
}
}
/*
* Thread function performing automatic memory scanning. Unreferenced objects
* at the end of a memory scan are reported but only the first time.
*/
static int kmemleak_scan_thread(void *arg)
{
static int first_run = IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN);
pr_info("Automatic memory scanning thread started\n");
set_user_nice(current, 10);
/*
* Wait before the first scan to allow the system to fully initialize.
*/
if (first_run) {
signed long timeout = msecs_to_jiffies(SECS_FIRST_SCAN * 1000);
first_run = 0;
while (timeout && !kthread_should_stop())
timeout = schedule_timeout_interruptible(timeout);
}
while (!kthread_should_stop()) {
signed long timeout = READ_ONCE(jiffies_scan_wait);
mutex_lock(&scan_mutex);
kmemleak_scan();
mutex_unlock(&scan_mutex);
/* wait before the next scan */
while (timeout && !kthread_should_stop())
timeout = schedule_timeout_interruptible(timeout);
}
pr_info("Automatic memory scanning thread ended\n");
return 0;
}
/*
* Start the automatic memory scanning thread. This function must be called
* with the scan_mutex held.
*/
static void start_scan_thread(void)
{
if (scan_thread)
return;
scan_thread = kthread_run(kmemleak_scan_thread, NULL, "kmemleak");
if (IS_ERR(scan_thread)) {
pr_warn("Failed to create the scan thread\n");
scan_thread = NULL;
}
}
/*
* Stop the automatic memory scanning thread.
*/
static void stop_scan_thread(void)
{
if (scan_thread) {
kthread_stop(scan_thread);
scan_thread = NULL;
}
}
/*
* Iterate over the object_list and return the first valid object at or after
* the required position with its use_count incremented. The function triggers
* a memory scanning when the pos argument points to the first position.
*/
static void *kmemleak_seq_start(struct seq_file *seq, loff_t *pos)
{
struct kmemleak_object *object;
loff_t n = *pos;
int err;
err = mutex_lock_interruptible(&scan_mutex);
if (err < 0)
return ERR_PTR(err);
rcu_read_lock();
list_for_each_entry_rcu(object, &object_list, object_list) {
if (n-- > 0)
continue;
if (get_object(object))
goto out;
}
object = NULL;
out:
return object;
}
/*
* Return the next object in the object_list. The function decrements the
* use_count of the previous object and increases that of the next one.
*/
static void *kmemleak_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
struct kmemleak_object *prev_obj = v;
struct kmemleak_object *next_obj = NULL;
struct kmemleak_object *obj = prev_obj;
++(*pos);
list_for_each_entry_continue_rcu(obj, &object_list, object_list) {
if (get_object(obj)) {
next_obj = obj;
break;
}
}
put_object(prev_obj);
return next_obj;
}
/*
* Decrement the use_count of the last object required, if any.
*/
static void kmemleak_seq_stop(struct seq_file *seq, void *v)
{
if (!IS_ERR(v)) {
/*
* kmemleak_seq_start may return ERR_PTR if the scan_mutex
* waiting was interrupted, so only release it if !IS_ERR.
*/
rcu_read_unlock();
mutex_unlock(&scan_mutex);
if (v)
put_object(v);
}
}
/*
* Print the information for an unreferenced object to the seq file.
*/
static int kmemleak_seq_show(struct seq_file *seq, void *v)
{
struct kmemleak_object *object = v;
unsigned long flags;
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
if ((object->flags & OBJECT_REPORTED) && unreferenced_object(object))
print_unreferenced(seq, object);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
return 0;
}
static const struct seq_operations kmemleak_seq_ops = {
.start = kmemleak_seq_start,
.next = kmemleak_seq_next,
.stop = kmemleak_seq_stop,
.show = kmemleak_seq_show,
};
static int kmemleak_open(struct inode *inode, struct file *file)
{
return seq_open(file, &kmemleak_seq_ops);
}
static int dump_str_object_info(const char *str)
{
unsigned long flags;
struct kmemleak_object *object;
unsigned long addr;
if (kstrtoul(str, 0, &addr))
return -EINVAL;
object = find_and_get_object(addr, 0);
if (!object) {
pr_info("Unknown object at 0x%08lx\n", addr);
return -EINVAL;
}
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
dump_object_info(object);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
put_object(object);
return 0;
}
/*
* We use grey instead of black to ensure we can do future scans on the same
* objects. If we did not do future scans these black objects could
* potentially contain references to newly allocated objects in the future and
* we'd end up with false positives.
*/
static void kmemleak_clear(void)
{
struct kmemleak_object *object;
unsigned long flags;
rcu_read_lock();
list_for_each_entry_rcu(object, &object_list, object_list) {
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_lock_irqsave(&object->lock, flags);
if ((object->flags & OBJECT_REPORTED) &&
unreferenced_object(object))
__paint_it(object, KMEMLEAK_GREY);
mm/kmemleak: turn kmemleak_lock and object->lock to raw_spinlock_t kmemleak_lock as a rwlock on RT can possibly be acquired in atomic context which does work. Since the kmemleak operation is performed in atomic context make it a raw_spinlock_t so it can also be acquired on RT. This is used for debugging and is not enabled by default in a production like environment (where performance/latency matters) so it makes sense to make it a raw_spinlock_t instead trying to get rid of the atomic context. Turn also the kmemleak_object->lock into raw_spinlock_t which is acquired (nested) while the kmemleak_lock is held. The time spent in "echo scan > kmemleak" slightly improved on 64core box with this patch applied after boot. [bigeasy@linutronix.de: redo the description, update comments. Merge the individual bits: He Zhe did the kmemleak_lock, Liu Haitao the ->lock and Yongxin Liu forwarded Liu's patch.] Link: http://lkml.kernel.org/r/20191219170834.4tah3prf2gdothz4@linutronix.de Link: https://lkml.kernel.org/r/20181218150744.GB20197@arrakis.emea.arm.com Link: https://lkml.kernel.org/r/1542877459-144382-1-git-send-email-zhe.he@windriver.com Link: https://lkml.kernel.org/r/20190927082230.34152-1-yongxin.liu@windriver.com Signed-off-by: He Zhe <zhe.he@windriver.com> Signed-off-by: Liu Haitao <haitao.liu@windriver.com> Signed-off-by: Yongxin Liu <yongxin.liu@windriver.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-31 06:12:00 +00:00
raw_spin_unlock_irqrestore(&object->lock, flags);
}
rcu_read_unlock();
kmemleak_found_leaks = false;
}
static void __kmemleak_do_cleanup(void);
/*
* File write operation to configure kmemleak at run-time. The following
* commands can be written to the /sys/kernel/debug/kmemleak file:
* off - disable kmemleak (irreversible)
* stack=on - enable the task stacks scanning
* stack=off - disable the tasks stacks scanning
* scan=on - start the automatic memory scanning thread
* scan=off - stop the automatic memory scanning thread
* scan=... - set the automatic memory scanning period in seconds (0 to
* disable it)
* scan - trigger a memory scan
* clear - mark all current reported unreferenced kmemleak objects as
* grey to ignore printing them, or free all kmemleak objects
* if kmemleak has been disabled.
* dump=... - dump information about the object found at the given address
*/
static ssize_t kmemleak_write(struct file *file, const char __user *user_buf,
size_t size, loff_t *ppos)
{
char buf[64];
int buf_size;
int ret;
buf_size = min(size, (sizeof(buf) - 1));
if (strncpy_from_user(buf, user_buf, buf_size) < 0)
return -EFAULT;
buf[buf_size] = 0;
ret = mutex_lock_interruptible(&scan_mutex);
if (ret < 0)
return ret;
if (strncmp(buf, "clear", 5) == 0) {
if (kmemleak_enabled)
kmemleak_clear();
else
__kmemleak_do_cleanup();
goto out;
}
if (!kmemleak_enabled) {
ret = -EPERM;
goto out;
}
if (strncmp(buf, "off", 3) == 0)
kmemleak_disable();
else if (strncmp(buf, "stack=on", 8) == 0)
kmemleak_stack_scan = 1;
else if (strncmp(buf, "stack=off", 9) == 0)
kmemleak_stack_scan = 0;
else if (strncmp(buf, "scan=on", 7) == 0)
start_scan_thread();
else if (strncmp(buf, "scan=off", 8) == 0)
stop_scan_thread();
else if (strncmp(buf, "scan=", 5) == 0) {
unsigned secs;
unsigned long msecs;
ret = kstrtouint(buf + 5, 0, &secs);
if (ret < 0)
goto out;
msecs = secs * MSEC_PER_SEC;
if (msecs > UINT_MAX)
msecs = UINT_MAX;
stop_scan_thread();
if (msecs) {
WRITE_ONCE(jiffies_scan_wait, msecs_to_jiffies(msecs));
start_scan_thread();
}
} else if (strncmp(buf, "scan", 4) == 0)
kmemleak_scan();
else if (strncmp(buf, "dump=", 5) == 0)
ret = dump_str_object_info(buf + 5);
else
ret = -EINVAL;
out:
mutex_unlock(&scan_mutex);
if (ret < 0)
return ret;
/* ignore the rest of the buffer, only one command at a time */
*ppos += size;
return size;
}
static const struct file_operations kmemleak_fops = {
.owner = THIS_MODULE,
.open = kmemleak_open,
.read = seq_read,
.write = kmemleak_write,
.llseek = seq_lseek,
.release = seq_release,
};
static void __kmemleak_do_cleanup(void)
{
struct kmemleak_object *object, *tmp;
/*
* Kmemleak has already been disabled, no need for RCU list traversal
* or kmemleak_lock held.
*/
list_for_each_entry_safe(object, tmp, &object_list, object_list) {
__remove_object(object);
__delete_object(object);
}
}
/*
* Stop the memory scanning thread and free the kmemleak internal objects if
* no previous scan thread (otherwise, kmemleak may still have some useful
* information on memory leaks).
*/
static void kmemleak_do_cleanup(struct work_struct *work)
{
stop_scan_thread();
mutex_lock(&scan_mutex);
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
/*
* Once it is made sure that kmemleak_scan has stopped, it is safe to no
* longer track object freeing. Ordering of the scan thread stopping and
* the memory accesses below is guaranteed by the kthread_stop()
* function.
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
*/
kmemleak_free_enabled = 0;
mutex_unlock(&scan_mutex);
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
if (!kmemleak_found_leaks)
__kmemleak_do_cleanup();
else
pr_info("Kmemleak disabled without freeing internal data. Reclaim the memory with \"echo clear > /sys/kernel/debug/kmemleak\".\n");
}
static DECLARE_WORK(cleanup_work, kmemleak_do_cleanup);
/*
* Disable kmemleak. No memory allocation/freeing will be traced once this
* function is called. Disabling kmemleak is an irreversible operation.
*/
static void kmemleak_disable(void)
{
/* atomically check whether it was already invoked */
if (cmpxchg(&kmemleak_error, 0, 1))
return;
/* stop any memory operation tracing */
kmemleak_enabled = 0;
/* check whether it is too early for a kernel thread */
if (kmemleak_initialized)
schedule_work(&cleanup_work);
mm: kmemleak: allow safe memory scanning during kmemleak disabling The kmemleak scanning thread can run for minutes. Callbacks like kmemleak_free() are allowed during this time, the race being taken care of by the object->lock spinlock. Such lock also prevents a memory block from being freed or unmapped while it is being scanned by blocking the kmemleak_free() -> ... -> __delete_object() function until the lock is released in scan_object(). When a kmemleak error occurs (e.g. it fails to allocate its metadata), kmemleak_enabled is set and __delete_object() is no longer called on freed objects. If kmemleak_scan is running at the same time, kmemleak_free() no longer waits for the object scanning to complete, allowing the corresponding memory block to be freed or unmapped (in the case of vfree()). This leads to kmemleak_scan potentially triggering a page fault. This patch separates the kmemleak_free() enabling/disabling from the overall kmemleak_enabled nob so that we can defer the disabling of the object freeing tracking until the scanning thread completed. The kmemleak_free_part() is deliberately ignored by this patch since this is only called during boot before the scanning thread started. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Reported-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Tested-by: Vignesh Radhakrishnan <vigneshr@codeaurora.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-24 23:58:26 +00:00
else
kmemleak_free_enabled = 0;
pr_info("Kernel memory leak detector disabled\n");
}
/*
* Allow boot-time kmemleak disabling (enabled by default).
*/
static int __init kmemleak_boot_config(char *str)
{
if (!str)
return -EINVAL;
if (strcmp(str, "off") == 0)
kmemleak_disable();
else if (strcmp(str, "on") == 0)
kmemleak_skip_disable = 1;
else
return -EINVAL;
return 0;
}
early_param("kmemleak", kmemleak_boot_config);
/*
* Kmemleak initialization.
*/
void __init kmemleak_init(void)
{
#ifdef CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF
if (!kmemleak_skip_disable) {
kmemleak_disable();
return;
}
#endif
if (kmemleak_error)
return;
jiffies_min_age = msecs_to_jiffies(MSECS_MIN_AGE);
jiffies_scan_wait = msecs_to_jiffies(SECS_SCAN_WAIT * 1000);
object_cache = KMEM_CACHE(kmemleak_object, SLAB_NOLEAKTRACE);
scan_area_cache = KMEM_CACHE(kmemleak_scan_area, SLAB_NOLEAKTRACE);
/* register the data/bss sections */
create_object((unsigned long)_sdata, _edata - _sdata,
KMEMLEAK_GREY, GFP_ATOMIC);
create_object((unsigned long)__bss_start, __bss_stop - __bss_start,
KMEMLEAK_GREY, GFP_ATOMIC);
/* only register .data..ro_after_init if not within .data */
if (&__start_ro_after_init < &_sdata || &__end_ro_after_init > &_edata)
create_object((unsigned long)__start_ro_after_init,
__end_ro_after_init - __start_ro_after_init,
KMEMLEAK_GREY, GFP_ATOMIC);
}
/*
* Late initialization function.
*/
static int __init kmemleak_late_init(void)
{
kmemleak_initialized = 1;
debugfs_create_file("kmemleak", 0644, NULL, NULL, &kmemleak_fops);
if (kmemleak_error) {
/*
* Some error occurred and kmemleak was disabled. There is a
* small chance that kmemleak_disable() was called immediately
* after setting kmemleak_initialized and we may end up with
* two clean-up threads but serialized by scan_mutex.
*/
schedule_work(&cleanup_work);
return -ENOMEM;
}
if (IS_ENABLED(CONFIG_DEBUG_KMEMLEAK_AUTO_SCAN)) {
mutex_lock(&scan_mutex);
start_scan_thread();
mutex_unlock(&scan_mutex);
}
pr_info("Kernel memory leak detector initialized (mem pool available: %d)\n",
mem_pool_free_count);
return 0;
}
late_initcall(kmemleak_late_init);