[mirror_ubuntu-jammy-kernel.git] / lib / stackdepot.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Generic stack depot for storing stack traces.
 *
 * Some debugging tools need to save stack traces of certain events which can
 * be later presented to the user. For example, KASAN needs to safe alloc and
 * free stacks for each object, but storing two stack traces per object
 * requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
 * that).
 *
 * Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
 * and free stacks repeat a lot, we save about 100x space.
 * Stacks are never removed from depot, so we store them contiguously one after
 * another in a contiguos memory allocation.
 *
 * Author: Alexander Potapenko <glider@google.com>
 * Copyright (C) 2016 Google, Inc.
 *
 * Based on code by Dmitry Chernenkov.
 */

#include <linux/gfp.h>
#include <linux/jhash.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/percpu.h>
#include <linux/printk.h>
#include <linux/slab.h>
#include <linux/stacktrace.h>
#include <linux/stackdepot.h>
#include <linux/string.h>
#include <linux/types.h>

#define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)

#define STACK_ALLOC_NULL_PROTECTION_BITS 1
#define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
#define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
#define STACK_ALLOC_ALIGN 4
#define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
					STACK_ALLOC_ALIGN)
#define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
		STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
#define STACK_ALLOC_SLABS_CAP 8192
#define STACK_ALLOC_MAX_SLABS \
	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
	 (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)

/* The compact structure to store the reference to stacks. */
union handle_parts {
	depot_stack_handle_t handle;
	struct {
		u32 slabindex : STACK_ALLOC_INDEX_BITS;
		u32 offset : STACK_ALLOC_OFFSET_BITS;
		u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
	};
};

struct stack_record {
	struct stack_record *next;	/* Link in the hashtable */
	u32 hash;			/* Hash in the hastable */
	u32 size;			/* Number of frames in the stack */
	union handle_parts handle;
	unsigned long entries[1];	/* Variable-sized array of entries. */
};

static void *stack_slabs[STACK_ALLOC_MAX_SLABS];

static int depot_index;
static int next_slab_inited;
static size_t depot_offset;
static DEFINE_SPINLOCK(depot_lock);

static bool init_stack_slab(void **prealloc)
{
	if (!*prealloc)
		return false;
	/*
	 * This smp_load_acquire() pairs with smp_store_release() to
	 * |next_slab_inited| below and in depot_alloc_stack().
	 */
	if (smp_load_acquire(&next_slab_inited))
		return true;
	if (stack_slabs[depot_index] == NULL) {
		stack_slabs[depot_index] = *prealloc;
		*prealloc = NULL;
	} else {
		/* If this is the last depot slab, do not touch the next one. */
		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS) {
			stack_slabs[depot_index + 1] = *prealloc;
			*prealloc = NULL;
		}
		/*
		 * This smp_store_release pairs with smp_load_acquire() from
		 * |next_slab_inited| above and in stack_depot_save().
		 */
		smp_store_release(&next_slab_inited, 1);
	}
	return true;
}

/* Allocation of a new stack in raw storage */
static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
		u32 hash, void **prealloc, gfp_t alloc_flags)
{
	int required_size = offsetof(struct stack_record, entries) +
		sizeof(unsigned long) * size;
	struct stack_record *stack;

	required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);

	if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
		if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
			WARN_ONCE(1, "Stack depot reached limit capacity");
			return NULL;
		}
		depot_index++;
		depot_offset = 0;
		/*
		 * smp_store_release() here pairs with smp_load_acquire() from
		 * |next_slab_inited| in stack_depot_save() and
		 * init_stack_slab().
		 */
		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
			smp_store_release(&next_slab_inited, 0);
	}
	init_stack_slab(prealloc);
	if (stack_slabs[depot_index] == NULL)
		return NULL;

	stack = stack_slabs[depot_index] + depot_offset;

	stack->hash = hash;
	stack->size = size;
	stack->handle.slabindex = depot_index;
	stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
	stack->handle.valid = 1;
	memcpy(stack->entries, entries, size * sizeof(unsigned long));
	depot_offset += required_size;

	return stack;
}

#define STACK_HASH_ORDER 20
#define STACK_HASH_SIZE (1L << STACK_HASH_ORDER)
#define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
#define STACK_HASH_SEED 0x9747b28c

static struct stack_record *stack_table[STACK_HASH_SIZE] = {
	[0 ...	STACK_HASH_SIZE - 1] = NULL
};

/* Calculate hash for a stack */
static inline u32 hash_stack(unsigned long *entries, unsigned int size)
{
	return jhash2((u32 *)entries,
			       size * sizeof(unsigned long) / sizeof(u32),
			       STACK_HASH_SEED);
}

/* Use our own, non-instrumented version of memcmp().
 *
 * We actually don't care about the order, just the equality.
 */
static inline
int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
			unsigned int n)
{
	for ( ; n-- ; u1++, u2++) {
		if (*u1 != *u2)
			return 1;
	}
	return 0;
}

/* Find a stack that is equal to the one stored in entries in the hash */
static inline struct stack_record *find_stack(struct stack_record *bucket,
					     unsigned long *entries, int size,
					     u32 hash)
{
	struct stack_record *found;

	for (found = bucket; found; found = found->next) {
		if (found->hash == hash &&
		    found->size == size &&
		    !stackdepot_memcmp(entries, found->entries, size))
			return found;
	}
	return NULL;
}

/**
 * stack_depot_fetch - Fetch stack entries from a depot
 *
 * @handle:		Stack depot handle which was returned from
 *			stack_depot_save().
 * @entries:		Pointer to store the entries address
 *
 * Return: The number of trace entries for this depot.
 */
unsigned int stack_depot_fetch(depot_stack_handle_t handle,
			       unsigned long **entries)
{
	union handle_parts parts = { .handle = handle };
	void *slab = stack_slabs[parts.slabindex];
	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
	struct stack_record *stack = slab + offset;

	*entries = stack->entries;
	return stack->size;
}
EXPORT_SYMBOL_GPL(stack_depot_fetch);

/**
 * stack_depot_save - Save a stack trace from an array
 *
 * @entries:		Pointer to storage array
 * @nr_entries:		Size of the storage array
 * @alloc_flags:	Allocation gfp flags
 *
 * Return: The handle of the stack struct stored in depot
 */
depot_stack_handle_t stack_depot_save(unsigned long *entries,
				      unsigned int nr_entries,
				      gfp_t alloc_flags)
{
	struct stack_record *found = NULL, **bucket;
	depot_stack_handle_t retval = 0;
	struct page *page = NULL;
	void *prealloc = NULL;
	unsigned long flags;
	u32 hash;

	if (unlikely(nr_entries == 0))
		goto fast_exit;

	hash = hash_stack(entries, nr_entries);
	bucket = &stack_table[hash & STACK_HASH_MASK];

	/*
	 * Fast path: look the stack trace up without locking.
	 * The smp_load_acquire() here pairs with smp_store_release() to
	 * |bucket| below.
	 */
	found = find_stack(smp_load_acquire(bucket), entries,
			   nr_entries, hash);
	if (found)
		goto exit;

	/*
	 * Check if the current or the next stack slab need to be initialized.
	 * If so, allocate the memory - we won't be able to do that under the
	 * lock.
	 *
	 * The smp_load_acquire() here pairs with smp_store_release() to
	 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
	 */
	if (unlikely(!smp_load_acquire(&next_slab_inited))) {
		/*
		 * Zero out zone modifiers, as we don't have specific zone
		 * requirements. Keep the flags related to allocation in atomic
		 * contexts and I/O.
		 */
		alloc_flags &= ~GFP_ZONEMASK;
		alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
		alloc_flags |= __GFP_NOWARN;
		page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
		if (page)
			prealloc = page_address(page);
	}

	spin_lock_irqsave(&depot_lock, flags);

	found = find_stack(*bucket, entries, nr_entries, hash);
	if (!found) {
		struct stack_record *new =
			depot_alloc_stack(entries, nr_entries,
					  hash, &prealloc, alloc_flags);
		if (new) {
			new->next = *bucket;
			/*
			 * This smp_store_release() pairs with
			 * smp_load_acquire() from |bucket| above.
			 */
			smp_store_release(bucket, new);
			found = new;
		}
	} else if (prealloc) {
		/*
		 * We didn't need to store this stack trace, but let's keep
		 * the preallocated memory for the future.
		 */
		WARN_ON(!init_stack_slab(&prealloc));
	}

	spin_unlock_irqrestore(&depot_lock, flags);
exit:
	if (prealloc) {
		/* Nobody used this memory, ok to free it. */
		free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
	}
	if (found)
		retval = found->handle.handle;
fast_exit:
	return retval;
}
EXPORT_SYMBOL_GPL(stack_depot_save);
Commit	Line	Data
1802d0be	1	// SPDX-License-Identifier: GPL-2.0-only
cd11016e AP	2	/*
	3	* Generic stack depot for storing stack traces.
	4	*
	5	* Some debugging tools need to save stack traces of certain events which can
	6	* be later presented to the user. For example, KASAN needs to safe alloc and
	7	* free stacks for each object, but storing two stack traces per object
	8	* requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
	9	* that).
	10	*
	11	* Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
	12	* and free stacks repeat a lot, we save about 100x space.
	13	* Stacks are never removed from depot, so we store them contiguously one after
	14	* another in a contiguos memory allocation.
	15	*
	16	* Author: Alexander Potapenko <glider@google.com>
	17	* Copyright (C) 2016 Google, Inc.
	18	*
	19	* Based on code by Dmitry Chernenkov.
cd11016e AP	20	*/
	21
	22	#include <linux/gfp.h>
	23	#include <linux/jhash.h>
	24	#include <linux/kernel.h>
	25	#include <linux/mm.h>
	26	#include <linux/percpu.h>
	27	#include <linux/printk.h>
	28	#include <linux/slab.h>
	29	#include <linux/stacktrace.h>
	30	#include <linux/stackdepot.h>
	31	#include <linux/string.h>
	32	#include <linux/types.h>
	33
	34	#define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
	35
7c31190b	36	#define STACK_ALLOC_NULL_PROTECTION_BITS 1
cd11016e AP	37	#define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
	38	#define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
	39	#define STACK_ALLOC_ALIGN 4
	40	#define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
	41	STACK_ALLOC_ALIGN)
7c31190b JK	42	#define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
7c31190b JK	43	STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
02754e0a	44	#define STACK_ALLOC_SLABS_CAP 8192
cd11016e AP	45	#define STACK_ALLOC_MAX_SLABS \
	46	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
	47	(1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)
	48
	49	/* The compact structure to store the reference to stacks. */
	50	union handle_parts {
	51	depot_stack_handle_t handle;
	52	struct {
	53	u32 slabindex : STACK_ALLOC_INDEX_BITS;
	54	u32 offset : STACK_ALLOC_OFFSET_BITS;
7c31190b	55	u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
cd11016e AP	56	};
	57	};
	58
	59	struct stack_record {
	60	struct stack_record next; / Link in the hashtable */
	61	u32 hash; /* Hash in the hastable */
	62	u32 size; /* Number of frames in the stack */
	63	union handle_parts handle;
	64	unsigned long entries[1]; /* Variable-sized array of entries. */
	65	};
	66
	67	static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
	68
	69	static int depot_index;
	70	static int next_slab_inited;
	71	static size_t depot_offset;
	72	static DEFINE_SPINLOCK(depot_lock);
	73
	74	static bool init_stack_slab(void **prealloc)
	75	{
	76	if (!*prealloc)
	77	return false;
	78	/*
	79	* This smp_load_acquire() pairs with smp_store_release() to
	80	* \|next_slab_inited\| below and in depot_alloc_stack().
	81	*/
	82	if (smp_load_acquire(&next_slab_inited))
	83	return true;
	84	if (stack_slabs[depot_index] == NULL) {
	85	stack_slabs[depot_index] = *prealloc;
305e519c	86	*prealloc = NULL;
cd11016e	87	} else {
305e519c AP	88	/* If this is the last depot slab, do not touch the next one. */
	89	if (depot_index + 1 < STACK_ALLOC_MAX_SLABS) {
	90	stack_slabs[depot_index + 1] = *prealloc;
	91	*prealloc = NULL;
	92	}
cd11016e AP	93	/*
cd11016e AP	94	* This smp_store_release pairs with smp_load_acquire() from
ee050dc8	95	* \|next_slab_inited\| above and in stack_depot_save().
cd11016e AP	96	*/
	97	smp_store_release(&next_slab_inited, 1);
	98	}
cd11016e AP	99	return true;
	100	}
	101
	102	/* Allocation of a new stack in raw storage */
	103	static struct stack_record depot_alloc_stack(unsigned long entries, int size,
	104	u32 hash, void **prealloc, gfp_t alloc_flags)
	105	{
	106	int required_size = offsetof(struct stack_record, entries) +
	107	sizeof(unsigned long) * size;
	108	struct stack_record *stack;
	109
	110	required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);
	111
	112	if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
	113	if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
	114	WARN_ONCE(1, "Stack depot reached limit capacity");
	115	return NULL;
	116	}
	117	depot_index++;
	118	depot_offset = 0;
	119	/*
	120	* smp_store_release() here pairs with smp_load_acquire() from
ee050dc8	121	* \|next_slab_inited\| in stack_depot_save() and
cd11016e AP	122	* init_stack_slab().
	123	*/
	124	if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
	125	smp_store_release(&next_slab_inited, 0);
	126	}
	127	init_stack_slab(prealloc);
	128	if (stack_slabs[depot_index] == NULL)
	129	return NULL;
	130
	131	stack = stack_slabs[depot_index] + depot_offset;
	132
	133	stack->hash = hash;
	134	stack->size = size;
	135	stack->handle.slabindex = depot_index;
	136	stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
7c31190b	137	stack->handle.valid = 1;
cd11016e AP	138	memcpy(stack->entries, entries, size * sizeof(unsigned long));
	139	depot_offset += required_size;
	140
	141	return stack;
	142	}
	143
	144	#define STACK_HASH_ORDER 20
	145	#define STACK_HASH_SIZE (1L << STACK_HASH_ORDER)
	146	#define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
	147	#define STACK_HASH_SEED 0x9747b28c
	148
	149	static struct stack_record *stack_table[STACK_HASH_SIZE] = {
	150	[0 ... STACK_HASH_SIZE - 1] = NULL
	151	};
	152
	153	/* Calculate hash for a stack */
	154	static inline u32 hash_stack(unsigned long *entries, unsigned int size)
	155	{
	156	return jhash2((u32 *)entries,
	157	size * sizeof(unsigned long) / sizeof(u32),
	158	STACK_HASH_SEED);
	159	}
	160
a571b272 AP	161	/* Use our own, non-instrumented version of memcmp().
	162	*
	163	* We actually don't care about the order, just the equality.
	164	*/
	165	static inline
	166	int stackdepot_memcmp(const unsigned long u1, const unsigned long u2,
	167	unsigned int n)
	168	{
	169	for ( ; n-- ; u1++, u2++) {
	170	if (u1 != u2)
	171	return 1;
	172	}
	173	return 0;
	174	}
	175
cd11016e AP	176	/* Find a stack that is equal to the one stored in entries in the hash */
	177	static inline struct stack_record find_stack(struct stack_record bucket,
	178	unsigned long *entries, int size,
	179	u32 hash)
	180	{
	181	struct stack_record *found;
	182
	183	for (found = bucket; found; found = found->next) {
	184	if (found->hash == hash &&
	185	found->size == size &&
a571b272	186	!stackdepot_memcmp(entries, found->entries, size))
cd11016e	187	return found;
cd11016e AP	188	}
	189	return NULL;
	190	}
	191
c0cfc337 TG	192	/**
	193	* stack_depot_fetch - Fetch stack entries from a depot
	194	*
	195	* @handle: Stack depot handle which was returned from
	196	* stack_depot_save().
	197	* @entries: Pointer to store the entries address
	198	*
	199	* Return: The number of trace entries for this depot.
	200	*/
	201	unsigned int stack_depot_fetch(depot_stack_handle_t handle,
	202	unsigned long **entries)
cd11016e AP	203	{
	204	union handle_parts parts = { .handle = handle };
	205	void *slab = stack_slabs[parts.slabindex];
	206	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
	207	struct stack_record *stack = slab + offset;
	208
c0cfc337 TG	209	*entries = stack->entries;
	210	return stack->size;
	211	}
	212	EXPORT_SYMBOL_GPL(stack_depot_fetch);
	213
cd11016e	214	/**
c0cfc337 TG	215	* stack_depot_save - Save a stack trace from an array
	216	*
	217	* @entries: Pointer to storage array
	218	* @nr_entries: Size of the storage array
	219	* @alloc_flags: Allocation gfp flags
cd11016e	220	*
c0cfc337	221	* Return: The handle of the stack struct stored in depot
cd11016e	222	*/
c0cfc337 TG	223	depot_stack_handle_t stack_depot_save(unsigned long *entries,
	224	unsigned int nr_entries,
	225	gfp_t alloc_flags)
cd11016e	226	{
cd11016e	227	struct stack_record found = NULL, *bucket;
c0cfc337	228	depot_stack_handle_t retval = 0;
cd11016e AP	229	struct page *page = NULL;
cd11016e AP	230	void *prealloc = NULL;
c0cfc337 TG	231	unsigned long flags;
c0cfc337 TG	232	u32 hash;
cd11016e	233
c0cfc337	234	if (unlikely(nr_entries == 0))
cd11016e AP	235	goto fast_exit;
cd11016e AP	236
c0cfc337	237	hash = hash_stack(entries, nr_entries);
cd11016e AP	238	bucket = &stack_table[hash & STACK_HASH_MASK];
	239
	240	/*
	241	* Fast path: look the stack trace up without locking.
	242	* The smp_load_acquire() here pairs with smp_store_release() to
	243	* \|bucket\| below.
	244	*/
c0cfc337 TG	245	found = find_stack(smp_load_acquire(bucket), entries,
c0cfc337 TG	246	nr_entries, hash);
cd11016e AP	247	if (found)
	248	goto exit;
	249
	250	/*
	251	* Check if the current or the next stack slab need to be initialized.
	252	* If so, allocate the memory - we won't be able to do that under the
	253	* lock.
	254	*
	255	* The smp_load_acquire() here pairs with smp_store_release() to
	256	* \|next_slab_inited\| in depot_alloc_stack() and init_stack_slab().
	257	*/
	258	if (unlikely(!smp_load_acquire(&next_slab_inited))) {
	259	/*
	260	* Zero out zone modifiers, as we don't have specific zone
	261	* requirements. Keep the flags related to allocation in atomic
	262	* contexts and I/O.
	263	*/
	264	alloc_flags &= ~GFP_ZONEMASK;
	265	alloc_flags &= (GFP_ATOMIC \| GFP_KERNEL);
87cc271d	266	alloc_flags \|= __GFP_NOWARN;
cd11016e AP	267	page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
	268	if (page)
	269	prealloc = page_address(page);
	270	}
	271
	272	spin_lock_irqsave(&depot_lock, flags);
	273
c0cfc337	274	found = find_stack(*bucket, entries, nr_entries, hash);
cd11016e AP	275	if (!found) {
cd11016e AP	276	struct stack_record *new =
c0cfc337	277	depot_alloc_stack(entries, nr_entries,
cd11016e AP	278	hash, &prealloc, alloc_flags);
	279	if (new) {
	280	new->next = *bucket;
	281	/*
	282	* This smp_store_release() pairs with
	283	* smp_load_acquire() from \|bucket\| above.
	284	*/
	285	smp_store_release(bucket, new);
	286	found = new;
	287	}
	288	} else if (prealloc) {
	289	/*
	290	* We didn't need to store this stack trace, but let's keep
	291	* the preallocated memory for the future.
	292	*/
	293	WARN_ON(!init_stack_slab(&prealloc));
	294	}
	295
	296	spin_unlock_irqrestore(&depot_lock, flags);
	297	exit:
	298	if (prealloc) {
	299	/* Nobody used this memory, ok to free it. */
	300	free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
	301	}
	302	if (found)
	303	retval = found->handle.handle;
	304	fast_exit:
	305	return retval;
	306	}
c0cfc337	307	EXPORT_SYMBOL_GPL(stack_depot_save);