【转载】DPDK编程开发（5）—mbuf

原文

1、知识百科

rte_mbuf的结构与linux内核协议栈的skb_buf相似，在保存报文的内存块前后分别保留headroom和tailroom，以方便应用解封报文，headroom默认128字节，可以通过宏RTE_PKTMBUF_HEADROOM调整。

返回值	操作函数	函数功能
宏	rte_pktmbuf_mtod(m, t)	将指针指到数据部分的首部
宏	rte_pktmbuf_pkt_len(m)	获取所有mbuf的数据长度
宏	rte_pktmbuf_data_len(m)	获取本mbuf的数据长度
char *	rte_get_rx_ol_flag_name
char *	rte_get_tx_ol_flag_name
uint16_t	rte_mbuf_refcnt_update
uint16_t	rte_mbuf_refcnt_read
void	rte_mbuf_refcnt_set
void	rte_mbuf_sanity_check
void	rte_ctrlmbuf_init
int	rte_is_ctrlmbuf
void	rte_pktmbuf_init
void	rte_pktmbuf_pool_init
void	rte_pktmbuf_reset
struct rte_mbuf *	rte_pktmbuf_alloc	从rte_mempool获取一个mbuf
void	rte_pktmbuf_attach
void	rte_pktmbuf_detach
void	rte_pktmbuf_free_seg
void	rte_pktmbuf_free	将mbuf归还到rte_mempool中
struct rte_mbuf *	rte_pktmbuf_clone
void	rte_pktmbuf_refcnt_update
uint16_t	rte_pktmbuf_headroom	计算出mbuf的headroom大小
uint16_t	rte_pktmbuf_tailroom	计算出mbuf的tailroom大小
struct rte_mbuf *	rte_pktmbuf_lastseg
char *	rte_pktmbuf_prepend
char *	rte_pktmbuf_append	向mbuf追加数据，修改pkt_len和data_len
char *	rte_pktmbuf_adj	向mbuf删减数据，修改pkt_len和data_len
int	rte_pktmbuf_trim
int	rte_pktmbuf_is_contiguous
void	rte_pktmbuf_dump

2、数据结构

rte_vlan_macip

union rte_vlan_macip {

                   uint32_t data;

                   struct {

                                      uint16_t l3_len:9;  //L3层首部= sizeof(struct ipv4_hdr)=20

                                      uint16_t l2_len:7;  // L3层首部= sizeof(struct ether_hdr)=14

                                      uint16_t vlan_tci;

                   } f;

};

rte_pktmbuf

struct rte_pktmbuf {

                   struct rte_mbuf *next;  /**< Next segment of scattered packet. */

                   void* data;            //mbuf数据部分的起始位置

                   uint16_t data_len;      //本mbuf的数据长度

 

                   uint8_t nb_segs;        /**< Number of segments. */

                   uint8_t in_port;        /**< Input port. */

                   uint32_t pkt_len;       //所有mbuf的数据长度

 

                   union rte_vlan_macip vlan_macip;

                   union {

                                      uint32_t rss;       /**< RSS hash result if RSS enabled */

                                      struct {

                                                         uint16_t hash;

                                                         uint16_t id;

                                      } fdir;             /**< Filter identifier if FDIR enabled */

                                      uint32_t sched;     /**< Hierarchical scheduler */

                   } hash;                 /**< hash information */

};

rte_mbuf

struct rte_mbuf {

                   struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */

                   void *buf_addr;         //mbuf的起始地址

                   phys_addr_t buf_physaddr; /**< Physical address of segment buffer. */

                   uint16_t buf_len;        //mbuf的长度=mp->elt_size - sizeof(struct rte_mbuf)

 

                   union {

                                      rte_atomic16_t refcnt_atomic;   /**< Atomically accessed refcnt */

                                      uint16_t refcnt;               /**< Non-atomically accessed refcnt */

                   };

                   uint8_t type;           /**< Type of mbuf. */

                   uint8_t reserved;        /**< Unused field. Required for padding. */

                   uint16_t ol_flags;        // PKT_TX_IP_CKSUM=0x1000，让网卡计算checksum

 

                   union {

                                      struct rte_ctrlmbuf ctrl;

                                      struct rte_pktmbuf pkt;

                   };

} __rte_cache_aligned;

3、操作函数

mbuf_init->rte_mempool_create

函数功能：mbuf由缓冲池rte_mempool管理，rte_mempool在初始化时一次申请多个mbuf，申请的mbuf个数和长度都由用户指定，使用rte_pktmbuf_alloc申请一个mbuf。

app.lcore_params[lcore].pool = rte_mempool_create(

                name,

                DEFAULT_MEMPOOL_BUFFERS,   // 8192 * 32=256K个

                DEFAULT_MBUF_SIZE,           // (4096 + sizeof(struct rte_mbuf) + 128)

                DEFAULT_MEMPOOL_CACHE_SIZE, //32

                sizeof(struct rte_pktmbuf_pool_private),

                rte_pktmbuf_pool_init, NULL,

                rte_pktmbuf_init, NULL,

                0,

                0);

 

#define RTE_PKTMBUF_HEADROOM 128

备注：调用rte_mempool_create()函数创建rte_mempool的时候，指定申请多少个rte_mbuf以及每个rte_mbuf中elt_size大小，elt_size是为网卡接收的数据包预先分配的内存的大小，该内存块就是rte_mbuf->pkt.data的实际存储区域。

rte_pktmbuf_alloc

函数功能：从rte_mempool获取一个mbuf。

static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)

{

                   struct rte_mbuf *m;

                   if ((m = __rte_mbuf_raw_alloc(mp)) != NULL)

                                      rte_pktmbuf_reset(m);

                   return (m);

}

rte_pktmbuf_free

函数功能：将mbuf归还到rte_mempool中。

static inline void rte_pktmbuf_free(struct rte_mbuf *m)

{

                   struct rte_mbuf *m_next;

 

                   __rte_mbuf_sanity_check(m, RTE_MBUF_PKT, 1);

 

                   while (m != NULL) {

                                      m_next = m->pkt.next;

                                      rte_pktmbuf_free_seg(m);

                                      m = m_next;

                   }

}

rte_pktmbuf_mtod

函数功能：将指针指到数据部分的首部。

#define rte_pktmbuf_mtod(m, t) ((t)((m)->pkt.data))

rte_pktmbuf_pkt_len

函数功能：获取所有mbuf的数据长度。

#define rte_pktmbuf_pkt_len(m) ((m)->pkt.pkt_len)

rte_pktmbuf_data_len

函数功能：获取本mbuf的数据长度。

#define rte_pktmbuf_data_len(m) ((m)->pkt.data_len)

rte_pktmbuf_headroom

函数功能：计算出mbuf的headroom大小。

static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)

{

                   __rte_mbuf_sanity_check(m, RTE_MBUF_PKT, 1);

                   return (uint16_t) ((char*) m->pkt.data - (char*) m->buf_addr);

}

rte_pktmbuf_tailroom

函数功能：计算出mbuf的tailroom大小。

static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)

{

                   __rte_mbuf_sanity_check(m, RTE_MBUF_PKT, 1);

                   return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -

                                                           m->pkt.data_len);

}

rte_pktmbuf_append

函数功能：向mbuf追加数据，修改pkt_len和data_len。

static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)

{

                   void *tail;

                   struct rte_mbuf *m_last;

 

                   __rte_mbuf_sanity_check(m, RTE_MBUF_PKT, 1);

 

                   m_last = rte_pktmbuf_lastseg(m);

                   if (unlikely(len > rte_pktmbuf_tailroom(m_last)))

                                      return NULL;

 

                   tail = (char*) m_last->pkt.data + m_last->pkt.data_len;       //tail的首地址

                   m_last->pkt.data_len = (uint16_t)(m_last->pkt.data_len + len); //修改last mbuf的data_len

                   m->pkt.pkt_len  = (m->pkt.pkt_len + len);                 //修改所有mbuf的pkt_len

                   return (char*) tail;

}

 

rte_pktmbuf_adj

函数功能：向mbuf删减数据，修改pkt_len和data_len。

static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)

{

                   __rte_mbuf_sanity_check(m, RTE_MBUF_PKT, 1);

 

                   if (unlikely(len > m->pkt.data_len))

                                      return NULL;

 

                   m->pkt.data_len = (uint16_t)(m->pkt.data_len - len);

                   m->pkt.data = ((char*) m->pkt.data + len);

                   m->pkt.pkt_len  = (m->pkt.pkt_len - len);

                   return (char*) m->pkt.data;

}

4、应用实例

网卡接收包

struct rte_mbuf *m = _m;

uint32_t buf_len = mp->elt_size - sizeof(struct rte_mbuf);

RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct rte_mbuf));

memset(m, 0, mp->elt_size);

 

m->buf_addr = (char *)m + sizeof(struct rte_mbuf);

m->buf_len = (uint16_t)buf_len;

m->pkt.data = (char*) m->buf_addr + RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len);

 

m->type = RTE_MBUF_PKT;

….

填充新数据

如果需要在tailroom 中加入N个字节数据，我们可以通过以下操作完成：

tail = m->pkt.data + m->pkt.data_len;     // tail记录tailroom首地址

m->pkt.data_len += N;

m->pkt.pkt_len += N;

这些操作由rte_pktmbuf_append()实现，函数原型如下：

char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)

剥离数据

假设m->pkt.data指向报文的二层首地址，我们可以通过以下一系列操作剥去报文的二层头部：

m->pkt.data += 14;

m->pkt.data_len -= 14;

m->pkt.pkt_len -= 14;

这些操作已经由rte_pktmbuf_adj()实现，函数原型如下：

char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)

5、参考资料

rte_mbuf

http://dpdk.org/doc/api/rte__mbuf_8h.html

http://www.tuicool.com/articles/rYfqMb

http://www.cnblogs.com/ziding/p/4214499.html

 

dpdk-mempool

http://dpdk.org/doc/api/rte__mempool_8h.html