%PDF- %PDF-
Direktori : /opt/cpanel/ea-ruby27/src/passenger-release-6.0.23/src/cxx_supportlib/MemoryKit/ |
Current File : //opt/cpanel/ea-ruby27/src/passenger-release-6.0.23/src/cxx_supportlib/MemoryKit/mbuf.h |
/* * twemproxy - A fast and lightweight proxy for memcached protocol. * Copyright (C) 2011 Twitter, Inc. * Copyright (C) 2014-2017 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. */ #ifndef _PSG_MBUF_BLOCK_H_ #define _PSG_MBUF_BLOCK_H_ #include <psg_sysqueue.h> #include <algorithm> #include <cstddef> #include <cassert> #include <cstring> #include <oxt/macros.hpp> #include <boost/cstdint.hpp> #include <boost/move/core.hpp> /** A memory buffer allocator system taken from twemproxy and modified to * suit our needs. * * In twemproxy, mbufs enables zero-copy because the same buffer on which a * request was received from the client is used for forwarding it to the * server. Similarly the same mbuf_block on which a response was received from * the server is used for forwarding it to the client. * * Furthermore, memory for mbufs is managed using a reuse pool. This means * that once mbuf_block is allocated, it is not deallocated, but just put back * into the reuse pool. By default each mbuf_block chunk is set to 16K bytes in * size. There is a trade-off between the mbuf_block size and number of concurrent * connections twemproxy can support. A large mbuf_block size reduces the number * of read syscalls made by twemproxy when reading requests or responses. * However, with large mbuf_block size, every active connection would use up 16K * bytes of buffer which might be an issue when twemproxy is handling large * number of concurrent connections from clients. When twemproxy is meant * to handle a large number of concurrent client connections, you should * set chunk size to a small value like 512 bytes. * * In Phusion Passenger, we modified the mbuf system so that we can take subsets * of an mbuf. The actual mbuf (now called mbuf_block in this modification) is * not actually put back on the freelist until all subsets are destroyed too. * This approach is similar to how Node.js manages buffer slices. * We also got rid of the global variables, and put them in an mbuf_pool * struct, which acts like a context structure. */ //#define MBUF_ENABLE_DEBUGGING //#define MBUF_ENABLE_BACKTRACES namespace Passenger { namespace MemoryKit { using namespace std; using namespace boost; struct mbuf_block; struct mhdr; typedef void (*mbuf_block_copy_t)(struct mbuf_block *, void *); /* See _mbuf_block_init() for format description */ struct mbuf_block { boost::uint32_t magic; /* mbuf_block magic (const) */ STAILQ_ENTRY(struct mbuf_block) next; /* next free mbuf_block */ #ifdef MBUF_ENABLE_DEBUGGING TAILQ_ENTRY(struct mbuf_block) active_q; /* prev and next active mbuf_block */ #endif #ifdef MBUF_ENABLE_BACKTRACES char *backtrace; #endif char *start; /* start of buffer (const) */ char *end; /* end of buffer (const) */ struct mbuf_pool *pool; /* containing pool (const) */ boost::uint32_t refcount; /* number of references by mbuf subsets */ boost::uint32_t offset; /* standalone mbuf_block data size */ }; STAILQ_HEAD(mhdr, struct mbuf_block); #ifdef MBUF_ENABLE_DEBUGGING TAILQ_HEAD(active_mbuf_block_list, struct mbuf_block); #endif struct mbuf_pool { boost::uint32_t nfree_mbuf_blockq; /* # free mbuf_block */ boost::uint32_t nactive_mbuf_blockq; /* # active (non-free) mbuf_block */ struct mhdr free_mbuf_blockq; /* free mbuf_block q */ #ifdef MBUF_ENABLE_DEBUGGING struct active_mbuf_block_list active_mbuf_blockq; /* active mbuf_block q */ #endif size_t mbuf_block_chunk_size; /* mbuf_block chunk size - header + data (const) */ size_t mbuf_block_offset; /* mbuf_block offset in chunk (const) */ }; #define MBUF_BLOCK_MAGIC 0xdeadbeef #define MBUF_BLOCK_MIN_SIZE 512 #define MBUF_BLOCK_MAX_SIZE 16777216 #define MBUF_BLOCK_HSIZE sizeof(struct mbuf_block) #define MBUF_BLOCK_EMPTY(mbuf_block) ((mbuf_block)->pos == (mbuf_block)->last) #define MBUF_BLOCK_FULL(mbuf_block) ((mbuf_block)->last == (mbuf_block)->end) void mbuf_pool_init(struct mbuf_pool *pool); void mbuf_pool_deinit(struct mbuf_pool *pool); size_t mbuf_pool_data_size(struct mbuf_pool *pool); unsigned int mbuf_pool_compact(struct mbuf_pool *pool); struct mbuf_block *mbuf_block_get(struct mbuf_pool *pool); struct mbuf_block *mbuf_block_new_standalone(struct mbuf_pool *pool, size_t size); void mbuf_block_put(struct mbuf_block *mbuf_block); void mbuf_block_ref(struct mbuf_block *mbuf_block); void mbuf_block_unref(struct mbuf_block *mbuf_block); void _mbuf_block_assert_refcount_at_least_two(struct mbuf_block *mbuf_block); /* A subset of an mbuf_block. */ class mbuf { private: BOOST_COPYABLE_AND_MOVABLE(mbuf) void initialize_with_block(unsigned int start, unsigned int len); void initialize_with_block_just_created(unsigned int start, unsigned int len); void initialize_with_mbuf(const mbuf &other, unsigned int start, unsigned int len); public: struct just_created_t { }; struct mbuf_block *mbuf_block; /* container block */ char *start; /* start of subset (const) */ char *end; /* end of subset (const) */ mbuf() : mbuf_block(NULL), start(NULL), end(NULL) { } explicit mbuf(struct mbuf_block *block, unsigned int start = 0) : mbuf_block(block) { initialize_with_block(start, block->end - block->start); } explicit mbuf(struct mbuf_block *block, unsigned int start, unsigned int len) : mbuf_block(block) { initialize_with_block(start, len); } explicit mbuf(struct mbuf_block *block, unsigned int start, unsigned int len, const just_created_t &t) : mbuf_block(block) { initialize_with_block_just_created(start, len); } // Create an mbuf as a dumb wrapper around a memory buffer. explicit mbuf(const char *data, unsigned int len) : mbuf_block(NULL), start(const_cast<char *>(data)), end(const_cast<char *>(data) + len) { } explicit mbuf(const char *data) : mbuf_block(NULL), start(const_cast<char *>(data)), end(const_cast<char *>(data) + strlen(data)) { } // Copy constructor. mbuf(const mbuf &mbuf, unsigned int start = 0) { initialize_with_mbuf(mbuf, start, mbuf.end - mbuf.start); } // Take a subset of another mbuf. mbuf(const mbuf &mbuf, unsigned int start, unsigned int len) { initialize_with_mbuf(mbuf, start, len); } // Move constructor. explicit mbuf(BOOST_RV_REF(mbuf) mbuf) : mbuf_block(mbuf.mbuf_block), start(mbuf.start), end(mbuf.end) { mbuf.mbuf_block = NULL; mbuf.start = NULL; mbuf.end = NULL; } ~mbuf() { if (mbuf_block != NULL) { mbuf_block_unref(mbuf_block); } } // Copy assignment. mbuf &operator=(BOOST_COPY_ASSIGN_REF(mbuf) other) { if (&other != this) { #ifndef NDEBUG if (mbuf_block != NULL && mbuf_block == other.mbuf_block) { _mbuf_block_assert_refcount_at_least_two(mbuf_block); } #endif if (mbuf_block != NULL) { mbuf_block_unref(mbuf_block); } mbuf_block = other.mbuf_block; start = other.start; end = other.end; // We reference 'other.mbuf_block' instead of 'this->mbuf_block' as // a micro-optimization. This should decrease the number of data // dependencies and allow the CPU to reorder the instructions better. if (other.mbuf_block != NULL) { mbuf_block_ref(other.mbuf_block); } } return *this; } // Move assignment. mbuf &operator=(BOOST_RV_REF(mbuf) other) { if (&other != this) { #ifndef NDEBUG if (mbuf_block != NULL && mbuf_block == other.mbuf_block) { _mbuf_block_assert_refcount_at_least_two(mbuf_block); } #endif if (mbuf_block != NULL) { mbuf_block_unref(mbuf_block); } mbuf_block = other.mbuf_block; start = other.start; end = other.end; other.mbuf_block = NULL; other.start = NULL; other.end = NULL; } return *this; } OXT_FORCE_INLINE size_t size() const { return end - start; } OXT_FORCE_INLINE bool empty() const { return start == end; } OXT_FORCE_INLINE bool is_null() const { return start == NULL; } }; mbuf mbuf_block_subset(struct mbuf_block *mbuf_block, unsigned int start, unsigned int len); mbuf mbuf_get(struct mbuf_pool *pool); mbuf mbuf_get_with_size(struct mbuf_pool *pool, size_t size); } // namespace MemoryKit } // namespace Passenger #endif /* _PSG_MBUF_BLOCK_H_ */