관리-도구

편집 파일: mbuf.cpp

/*
 * twemproxy - A fast and lightweight proxy for memcached protocol.
 * Copyright (C) 2011 Twitter, Inc.
 * Copyright (C) 2014-2018 Phusion Holding B.V.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <cassert>
#include <cstdlib>
#include <cstring>
#include <oxt/macros.hpp>
#include <oxt/thread.hpp>
#include <oxt/backtrace.hpp>
#include <algorithm>
#include <ostream>
#include <MemoryKit/mbuf.h>
#include <LoggingKit/LoggingKit.h>
#include <StaticString.h>
#include <StrIntTools/StrIntUtils.h>

namespace Passenger {
namespace MemoryKit {

//#define MBUF_DEBUG_REFCOUNTS

#define ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, expr) \
	do { \
		if (OXT_UNLIKELY(!(expr))) { \
			P_BUG_WITH_FORMATTER_CODE(stream, \
				stream << "Assertion failed: " #expr "\n"; \
				mbuf_block_print(mbuf_block, stream); \
			); \
		} \
	} while (false)

static void mbuf_block_print(struct mbuf_block *mbuf_block, std::ostream &stream);

static void
_mbuf_block_mark_as_active(struct mbuf_pool *pool, struct mbuf_block *mbuf_block)
{
	STAILQ_NEXT(mbuf_block, next) = NULL;
	#ifdef MBUF_ENABLE_DEBUGGING
		TAILQ_INSERT_HEAD(&pool->active_mbuf_blockq, mbuf_block, active_q);
	#endif
	#ifdef MBUF_ENABLE_BACKTRACES
		mbuf_block->backtrace = strdup(oxt::thread::current_backtrace().c_str());
	#endif
	mbuf_block->refcount = 1;
	pool->nactive_mbuf_blockq++;
}

static struct mbuf_block *
_mbuf_block_init(struct mbuf_pool *pool, char *buf, size_t block_offset)
{
	struct mbuf_block *mbuf_block;

/*
	 * There are two types of mbuf_blocks: normal ones and standalone ones.
	 *
	 * ## Normal mbuf_blocks
	 *
	 * mbuf_block header is at the tail end of the mbuf_block. The data
	 * precedes the header. This enables us to catch buffer overrun early
	 * by asserting on the magic value during get or put operations.
	 * All normal mbuf_blocks in a pool have the same mbuf_block_offset,
	 * allowing them to be reused through a freelist.
	 *
	 *   <------------ pool->mbuf_block_chunk_size -------------->
	 *   +-------------------------------------------------------+
	 *   |       mbuf_block data          |  mbuf_block header   |
	 *   |                                |                      |
	 *   |  (pool->mbuf_block_offset)     | (struct mbuf_block)  |
	 *   +-------------------------------------------------------+
	 *   ^                                ^
	 *   |                                |
	 *   \                                |\
	 * block->start                       | block->end (one byte past valid bound)
	 *                                    \
	 *                                    block
	 *
	 * ## Standalone mbuf_blocks
	 *
	 * Standalone mbuf_blocks are like normal ones, but can contain
	 * arbitrarily-sized data. Different standalone mbuf_blocks in a pool
	 * can have different data sizes. They cannot be reused through the
	 * freelist. The fact that the 'offset' field in the header is set to
	 * a non-zero is an indication that it is standalone.
	 *
	 *   <------------- offset + MBUF_BLOCK_HSIZE --------------->
	 *   +-------------------------------------------------------+
	 *   |       mbuf_block data          |  mbuf_block header   |
	 *   |                                |                      |
	 *   |           (offset)             | (struct mbuf_block)  |
	 *   +-------------------------------------------------------+
	 *   ^                                ^
	 *   |                                |
	 *   \                                |\
	 * block->start                       | block->end (one byte past valid bound)
	 *                                    \
	 *                                    block
	 */
	uintptr_t ptr = reinterpret_cast<uintptr_t>(buf + block_offset);
	assert(ptr % alignof(struct mbuf_block) == 0);
	mbuf_block = reinterpret_cast<struct mbuf_block*>(ptr);
	mbuf_block->magic = MBUF_BLOCK_MAGIC;
	mbuf_block->pool  = pool;
	mbuf_block->offset = 0;

_mbuf_block_mark_as_active(pool, mbuf_block);
	return mbuf_block;
}

static struct mbuf_block *
_mbuf_block_get(struct mbuf_pool *pool)
{
	struct mbuf_block *mbuf_block;
	char *buf;

if (!STAILQ_EMPTY(&pool->free_mbuf_blockq)) {
		assert(pool->nfree_mbuf_blockq > 0);

mbuf_block = STAILQ_FIRST(&pool->free_mbuf_blockq);
		ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->magic == MBUF_BLOCK_MAGIC);
		ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount == 0);

pool->nfree_mbuf_blockq--;
		STAILQ_REMOVE_HEAD(&pool->free_mbuf_blockq, next);
		_mbuf_block_mark_as_active(pool, mbuf_block);
		return mbuf_block;
	}

buf = (char *) aligned_alloc(alignof(struct mbuf_block), pool->mbuf_block_chunk_size);
	if (OXT_UNLIKELY(buf == NULL)) {
		return NULL;
	}

return _mbuf_block_init(pool, buf, pool->mbuf_block_offset);
}

struct mbuf_block *
mbuf_block_get(struct mbuf_pool *pool)
{
	struct mbuf_block *mbuf_block;
	char *buf;

mbuf_block = _mbuf_block_get(pool);
	if (OXT_UNLIKELY(mbuf_block == NULL)) {
		return NULL;
	}

buf = (char *)mbuf_block - pool->mbuf_block_offset;
	mbuf_block->start = buf;
	mbuf_block->end = buf + pool->mbuf_block_offset;

ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block,
		mbuf_block->end - mbuf_block->start == (int) pool->mbuf_block_offset);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->start < mbuf_block->end);

#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block get %p\n", oxt::thread_signature, mbuf_block);
	#endif

return mbuf_block;
}

struct mbuf_block *
mbuf_block_new_standalone(struct mbuf_pool *pool, size_t size)
{
	struct mbuf_block *mbuf_block;
	size_t block_offset;
	char *buf;

block_offset = std::max<size_t>(size, mbuf_pool_data_size(pool));
	if (OXT_UNLIKELY(block_offset % alignof(struct mbuf_block) != 0)) {
		return NULL;
	}
	buf = (char *) aligned_alloc(alignof(struct mbuf_block), MBUF_BLOCK_HSIZE + block_offset);
	if (OXT_UNLIKELY(buf == NULL)) {
		return NULL;
	}

mbuf_block = _mbuf_block_init(pool, buf, block_offset);
	mbuf_block->start = buf;
	mbuf_block->end = buf + size;
	mbuf_block->offset = block_offset;

ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block,
		mbuf_block->end - mbuf_block->start == (int) size);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block,
		mbuf_block->start < mbuf_block->end);

#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block new standalone %p\n", oxt::thread_signature, mbuf_block);
	#endif

return mbuf_block;
}

static void
mbuf_block_free(struct mbuf_block *mbuf_block)
{
	char *buf;

#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block free %p\n", oxt::thread_signature, mbuf_block);
	#endif

ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, STAILQ_NEXT(mbuf_block, next) == NULL);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->magic == MBUF_BLOCK_MAGIC);

#ifdef MBUF_ENABLE_DEBUGGING
		TAILQ_REMOVE(&mbuf_block->pool->active_mbuf_blockq, mbuf_block, active_q);
	#endif
	#ifdef MBUF_ENABLE_BACKTRACES
		free(mbuf_block->backtrace);
	#endif

if (mbuf_block->offset > 0) {
		buf = (char *) mbuf_block - mbuf_block->offset;
	} else {
		buf = (char *) mbuf_block - mbuf_block->pool->mbuf_block_offset;
	}
	free(buf);
}

void
mbuf_block_put(struct mbuf_block *mbuf_block)
{
	#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block put %p\n", oxt::thread_signature, mbuf_block);
	#endif

ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, STAILQ_NEXT(mbuf_block, next) == NULL);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->magic == MBUF_BLOCK_MAGIC);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount == 0);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->pool->nactive_mbuf_blockq > 0);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->offset == 0);

mbuf_block->pool->nfree_mbuf_blockq++;
	mbuf_block->pool->nactive_mbuf_blockq--;
	STAILQ_INSERT_HEAD(&mbuf_block->pool->free_mbuf_blockq, mbuf_block, next);

#ifdef MBUF_ENABLE_DEBUGGING
		TAILQ_REMOVE(&mbuf_block->pool->active_mbuf_blockq, mbuf_block, active_q);
	#endif
}

/*
 * Remove mbuf_block from the mhdr Q
 */
static void
mbuf_block_remove(struct mhdr *mhdr, struct mbuf_block *mbuf_block)
{
	//log_debug(LOG_VVERB, "remove mbuf_block %p", mbuf_block);

STAILQ_REMOVE(mhdr, mbuf_block, struct mbuf_block, next);
	STAILQ_NEXT(mbuf_block, next) = NULL;
}

void
_mbuf_block_assert_refcount_at_least_two(struct mbuf_block *mbuf_block) {
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount >= 2);
}

void
mbuf_pool_init(struct mbuf_pool *pool)
{
	pool->nfree_mbuf_blockq = 0;
	pool->nactive_mbuf_blockq = 0;
	STAILQ_INIT(&pool->free_mbuf_blockq);

#ifdef MBUF_ENABLE_DEBUGGING
		TAILQ_INIT(&pool->active_mbuf_blockq);
	#endif

size_t block_offset = pool->mbuf_block_chunk_size - MBUF_BLOCK_HSIZE;
	assert((block_offset % alignof(struct mbuf_block)) == 0);
	pool->mbuf_block_offset = block_offset;
}

void
mbuf_pool_deinit(struct mbuf_pool *pool)
{
	mbuf_pool_compact(pool);
}

/*
 * Return the maximum available space size for data in any mbuf_block. Mbuf cannot
 * contain more than 2^32 bytes (4G).
 */
size_t
mbuf_pool_data_size(struct mbuf_pool *pool)
{
	return pool->mbuf_block_offset;
}

unsigned int
mbuf_pool_compact(struct mbuf_pool *pool)
{
	unsigned int count = pool->nfree_mbuf_blockq;

while (!STAILQ_EMPTY(&pool->free_mbuf_blockq)) {
		struct mbuf_block *mbuf_block = STAILQ_FIRST(&pool->free_mbuf_blockq);
		mbuf_block_remove(&pool->free_mbuf_blockq, mbuf_block);
		mbuf_block_free(mbuf_block);
		pool->nfree_mbuf_blockq--;
	}
	assert(pool->nfree_mbuf_blockq == 0);

return count;
}

void
mbuf_block_ref(struct mbuf_block *mbuf_block)
{
	#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block ref %p: %u -> %u\n",
			oxt::thread_signature, mbuf_block,
			mbuf_block->refcount, mbuf_block->refcount + 1);
	#endif
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount > 0);
	#ifdef MBUF_ENABLE_BACKTRACES
		mbuf_block->backtrace = strdup(oxt::thread::current_backtrace().c_str());
	#endif

mbuf_block->refcount++;
}

void
mbuf_block_unref(struct mbuf_block *mbuf_block)
{
	#ifdef MBUF_DEBUG_REFCOUNTS
		printf("[%p] mbuf_block unref %p: %u -> %u\n",
			oxt::thread_signature, mbuf_block,
			mbuf_block->refcount, mbuf_block->refcount - 1);
	#endif

ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, STAILQ_NEXT(mbuf_block, next) == NULL);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->magic == MBUF_BLOCK_MAGIC);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount > 0);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->pool->nactive_mbuf_blockq > 0);

mbuf_block->refcount--;
	if (mbuf_block->refcount == 0) {
		if (mbuf_block->offset > 0) {
			ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->pool->nactive_mbuf_blockq > 0);
			mbuf_block->pool->nactive_mbuf_blockq--;
			mbuf_block_free(mbuf_block);
		} else {
			mbuf_block_put(mbuf_block);
		}
	}
}

mbuf
mbuf_block_subset(struct mbuf_block *mbuf_block, unsigned int start, unsigned int len)
{
	return mbuf(mbuf_block, start, len);
}

mbuf
mbuf_get(struct mbuf_pool *pool)
{
	struct mbuf_block *block = mbuf_block_get(pool);
	if (OXT_UNLIKELY(block == NULL)) {
		return mbuf();
	}

ASSERT_MBUF_BLOCK_PROPERTY(block, block->refcount == 1);
	return mbuf(block, 0, block->end - block->start, mbuf::just_created_t());
}

mbuf
mbuf_get_with_size(struct mbuf_pool *pool, size_t size)
{
	struct mbuf_block *block;
	if (size <= mbuf_pool_data_size(pool)) {
		block = mbuf_block_get(pool);
	} else {
		block = mbuf_block_new_standalone(pool, size);
	}
	if (OXT_UNLIKELY(block == NULL)) {
		return mbuf();
	}

ASSERT_MBUF_BLOCK_PROPERTY(block, block->refcount == 1);
	return mbuf(block, 0, size, mbuf::just_created_t());
}

static void
mbuf_block_print(struct mbuf_block *mbuf_block, std::ostream &stream)
{
	stream << "mbuf_block: " << (void *) mbuf_block << "\n"
		"mbuf_block.magic: " << mbuf_block->magic << "\n"
		"mbuf_block.next: " << (void *) STAILQ_NEXT(mbuf_block, next) << "\n"
		"mbuf_block.start: " << (void *) mbuf_block->start << "\n"
		"mbuf_block.end: " << (void *) mbuf_block->end << "\n"
		"mbuf_block.contents: \"" << cEscapeString(StaticString(mbuf_block->start,
			mbuf_block->end - mbuf_block->start)) << "\"\n"
		"mbuf_block.refcount: " << mbuf_block->refcount << "\n"
		"mbuf_block.offset: " << mbuf_block->offset << "\n"
		"mbuf_block.pool: " << (void *) mbuf_block->pool << "\n"
		"mbuf_block.pool.nfree_mbuf_blockq: " << mbuf_block->pool->nfree_mbuf_blockq << "\n"
		"mbuf_block.pool.nactive_mbuf_blockq: " << mbuf_block->pool->nactive_mbuf_blockq << "\n"
		"mbuf_block.pool.mbuf_block_chunk_size: " << mbuf_block->pool->mbuf_block_chunk_size << "\n"
		"mbuf_block.pool.mbuf_block_offset: " << mbuf_block->pool->mbuf_block_offset << "\n";
}

template<typename Address>
static Address clamp(Address value, Address min, Address max) {
	return std::max(std::min(value, max), min);
}

void
mbuf::initialize_with_block(unsigned int start, unsigned int len) {
	this->start = clamp<char *>(
		mbuf_block->start + start,
		mbuf_block->start,
		mbuf_block->end);
	this->end = clamp<char *>(
		mbuf_block->start + start + len,
		mbuf_block->start,
		mbuf_block->end);
	if (mbuf_block != NULL) {
		mbuf_block_ref(mbuf_block);
	}
}

void
mbuf::initialize_with_block_just_created(unsigned int start, unsigned int len) {
	this->start = clamp<char *>(
		mbuf_block->start + start,
		mbuf_block->start,
		mbuf_block->end);
	this->end = clamp<char *>(
		mbuf_block->start + start + len,
		mbuf_block->start,
		mbuf_block->end);
	ASSERT_MBUF_BLOCK_PROPERTY(mbuf_block, mbuf_block->refcount == 1);
	#ifdef MBUF_ENABLE_BACKTRACES
		mbuf_block->backtrace = strdup(oxt::thread::current_backtrace().c_str());
	#endif
}

void
mbuf::initialize_with_mbuf(const mbuf &other, unsigned int start, unsigned int len) {
	mbuf_block = other.mbuf_block;
	this->start = clamp<char *>(
		other.start + start,
		other.start,
		other.end);
	this->end = clamp<char *>(
		other.start + start + len,
		other.start,
		other.end);
	if (other.mbuf_block != NULL) {
		mbuf_block_ref(other.mbuf_block);
	}
}

} // namespace MemoryKit
} // namespace Passenger