sk_buff结构

Posted 2020-10-07 Alex

sk_buff结构用来描述已接收或者待发送的数据报文信息；skb在不同网络协议层之间传递，可被用于不同网络协议，如二层的mac或其他链路层协议，三层的ip，四层的tcp或者udp协议，其中某些成员变量会在该结构从一层向另一层传递时发生改变，从上层向下层传递需要添加首部，从下层向上层传递需要移除首部；

 

 1 /* skb头结构 */
 2 struct sk_buff_head {
 3     /* These two members must be first. */
 4     /* 通过下面两个指针成员将skb连接成双向链表 */
 5     struct sk_buff    *next; /* 指向后一个skb */
 6     struct sk_buff    *prev; /* 指向前一个skb */
 7 
 8     __u32        qlen; /* 链表中元素个数 */
 9     spinlock_t    lock; /* 自旋锁 */
10 };

 

  1 /** 
  2  *    struct sk_buff - socket buffer
  3  *    @next: Next buffer in list
  4  *    @prev: Previous buffer in list
  5  *    @tstamp: Time we arrived/left
  6  *    @rbnode: RB tree node, alternative to next/prev for netem/tcp
  7  *    @sk: Socket we are owned by
  8  *    @dev: Device we arrived on/are leaving by
  9  *    @cb: Control buffer. Free for use by every layer. Put private vars here
 10  *    @_skb_refdst: destination entry (with norefcount bit)
 11  *    @sp: the security path, used for xfrm
 12  *    @len: Length of actual data
 13  *    @data_len: Data length
 14  *    @mac_len: Length of link layer header
 15  *    @hdr_len: writable header length of cloned skb
 16  *    @csum: Checksum (must include start/offset pair)
 17  *    @csum_start: Offset from skb->head where checksumming should start
 18  *    @csum_offset: Offset from csum_start where checksum should be stored
 19  *    @priority: Packet queueing priority
 20  *    @ignore_df: allow local fragmentation
 21  *    @cloned: Head may be cloned (check refcnt to be sure)
 22  *    @ip_summed: Driver fed us an IP checksum
 23  *    @nohdr: Payload reference only, must not modify header
 24  *    @pkt_type: Packet class
 25  *    @fclone: skbuff clone status
 26  *    @ipvs_property: skbuff is owned by ipvs
 27  *    @tc_skip_classify: do not classify packet. set by IFB device
 28  *    @tc_at_ingress: used within tc_classify to distinguish in/egress
 29  *    @tc_redirected: packet was redirected by a tc action
 30  *    @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
 31  *    @peeked: this packet has been seen already, so stats have been
 32  *        done for it, don‘t do them again
 33  *    @nf_trace: netfilter packet trace flag
 34  *    @protocol: Packet protocol from driver
 35  *    @destructor: Destruct function
 36  *    @_nfct: Associated connection, if any (with nfctinfo bits)
 37  *    @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
 38  *    @skb_iif: ifindex of device we arrived on
 39  *    @tc_index: Traffic control index
 40  *    @hash: the packet hash
 41  *    @queue_mapping: Queue mapping for multiqueue devices
 42  *    @xmit_more: More SKBs are pending for this queue
 43  *    @ndisc_nodetype: router type (from link layer)
 44  *    @ooo_okay: allow the mapping of a socket to a queue to be changed
 45  *    @l4_hash: indicate hash is a canonical 4-tuple hash over transport
 46  *        ports.
 47  *    @sw_hash: indicates hash was computed in software stack
 48  *    @wifi_acked_valid: wifi_acked was set
 49  *    @wifi_acked: whether frame was acked on wifi or not
 50  *    @no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
 51  *    @dst_pending_confirm: need to confirm neighbour
 52   *    @napi_id: id of the NAPI struct this skb came from
 53  *    @secmark: security marking
 54  *    @mark: Generic packet mark
 55  *    @vlan_proto: vlan encapsulation protocol
 56  *    @vlan_tci: vlan tag control information
 57  *    @inner_protocol: Protocol (encapsulation)
 58  *    @inner_transport_header: Inner transport layer header (encapsulation)
 59  *    @inner_network_header: Network layer header (encapsulation)
 60  *    @inner_mac_header: Link layer header (encapsulation)
 61  *    @transport_header: Transport layer header
 62  *    @network_header: Network layer header
 63  *    @mac_header: Link layer header
 64  *    @tail: Tail pointer
 65  *    @end: End pointer
 66  *    @head: Head of buffer
 67  *    @data: Data head pointer
 68  *    @truesize: Buffer size
 69  *    @users: User count - see {datagram,tcp}.c
 70  */
 71 /* skb结构 */
 72 struct sk_buff {
 73     union {
 74         struct {
 75             /* These two members must be first. */
 76             struct sk_buff        *next;
 77             struct sk_buff        *prev;
 78 
 79             /* 报文到达或者离开的时间戳 */
 80             union {
 81                 ktime_t        tstamp;
 82                 struct skb_mstamp skb_mstamp;
 83             };
 84         };
 85         struct rb_node    rbnode; /* used in netem & tcp stack */
 86     };
 87     
 88     /* 
 89         指向缓冲区的套接字sock数据结构。当数据在本地产生或者正由本地进程接收时，
 90         该数据以及套接字相关信息会被L4(tcp或者udp)以及用户应用程序使用
 91         当缓冲区只是被转发时(本地机器不是来源也不是目的地)，该指针为NULL
 92     */
 93     struct sock        *sk;
 94 
 95     union {
 96         /* 报文到达或者离开时的网络设备 */
 97         struct net_device    *dev;
 98         /* Some protocols might use this space to store information,
 99          * while device pointer would be NULL.
100          * UDP receive path is one user.
101          */
102         unsigned long        dev_scratch;
103     };
104     /*
105      * This is the control buffer. It is free to use for every
106      * layer. Please put your private variables there. If you
107      * want to keep them across layers you have to do a skb_clone()
108      * first. This is owned by whoever has the skb queued ATM.
109      */
110     /*
111         控制缓冲区，用于存储私有信息，每层协议自己维护并使用，
112         并且只在本层有有效
113     */
114     char            cb[48] __aligned(8);
115 
116     /* 路由缓存，输入或者输出报文都要查询到目的路由缓存项，才能确定流向 */
117     unsigned long        _skb_refdst;
118     
119     /* 
120         当缓冲区被删除时，可以完成某些清理工作
121         当缓冲区不属于一个套接字时，该函数通常不被初始化
122         属于一个套接字时，通常设置为sock_rfree或sock_wfree
123         sock_xxx函数用于更新套接字队列中所持有的内存
124     */
125     void            (*destructor)(struct sk_buff *skb);
126 #ifdef CONFIG_XFRM
127     struct    sec_path    *sp;
128 #endif
129 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
130     unsigned long         _nfct;
131 #endif
132 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
133     struct nf_bridge_info    *nf_bridge;
134 #endif
135     /* 
136         缓冲区的数据区块大小，该长度包括主缓冲区(head指针指向)的数据
137         以及一些片段(fragment)的数据，当缓冲区从一个网络分层移动到下一个
138         网络分层时，该值会发生变化，因为在协议栈中向上层移动时报头会被丢弃
139         向下层移动时报头会添加，len也会把协议报头算在内，与"数据预留和对齐"操作
140     */
141     unsigned int        len,
142     /* 片段(fragment)中的数据大小 */
143                 data_len;
144     /* mac报头大小 */
145     __u16            mac_len,
146     /* 克隆skb时可写报文头部长度 */
147                 hdr_len;
148 
149     /* Following fields are _not_ copied in __copy_skb_header()
150      * Note that queue_mapping is here mostly to fill a hole.
151      */
152     kmemcheck_bitfield_begin(flags1);
153     __u16            queue_mapping;
154 
155 /* if you move cloned around you also must adapt those constants */
156 #ifdef __BIG_ENDIAN_BITFIELD
157 #define CLONED_MASK    (1 << 7)
158 #else
159 #define CLONED_MASK    1
160 #endif
161 #define CLONED_OFFSET()        offsetof(struct sk_buff, __cloned_offset)
162 
163     __u8            __cloned_offset[0];
164     /* 表示该skb是另外一个skb的克隆 */
165     __u8            cloned:1,
166                 /* 
167     payload是否被单独引用，不存在协议首部，如果被引用，则不能修改协议首部，也不能通过skb->data来访问协议首部 
168                 */
169                 nohdr:1,
170                 /*
171                     当前克隆状态
172                     SKB_FCLONE_UNAVAILABLE-skb未被克隆
173                     SKB_FCLONE_ORIG-在skbuff_fclone_cache分配的父skb，可以被克隆
174     SKB_FCLONE_CLONE-在skbuff_fclone_cache分配的子skb，从父skb克隆得到
175                 */
176                 fclone:2,
177                 peeked:1,
178                 head_frag:1,
179                 xmit_more:1,
180                 __unused:1; /* one bit hole */
181     kmemcheck_bitfield_end(flags1);
182 
183     /* fields enclosed in headers_start/headers_end are copied
184      * using a single memcpy() in __copy_skb_header()
185      */
186     /* private: */
187     __u32            headers_start[0];
188     /* public: */
189 
190 /* if you move pkt_type around you also must adapt those constants */
191 #ifdef __BIG_ENDIAN_BITFIELD
192 #define PKT_TYPE_MAX    (7 << 5)
193 #else
194 #define PKT_TYPE_MAX    7
195 #endif
196 #define PKT_TYPE_OFFSET()    offsetof(struct sk_buff, __pkt_type_offset)
197 
198     __u8            __pkt_type_offset[0];
199     /*
200         此字段根据l2的目的地址进行划分
201         PACKET_HOST-mac地址与接收设备mac地址相等，说明是发给该主机的
202         PACKET_BROADCAST-mac地址是接收设备的广播地址
203         PACKET_MULTICAST-mac地址接收改设备注册的多播地址之一
204         PACKET_OTHERHOST-mac地址不属于接收设备的地址，启用转发则转发，否则丢弃
205         PACKET_OUTGOING-数据包将被发出，用到这个标记的功能包括decnet，
206 或者为每个网络tab都复制一份发出包的函数
207         PACKET_LOOPBACK-数据包发往回环设备，有此标识，处理回环设备时，
208         可以跳过一些真实设备所需的操作
209         PACKET_USER-发送到用户空间，netlink使用
210         PACKET_KERNEL-发送到内核空间，netlink使用
211         PACKET_FASTROUTE-未使用
212     */
213     __u8            pkt_type:3;
214     __u8            pfmemalloc:1;
215     __u8            ignore_df:1;
216 
217     __u8            nf_trace:1;
218     /*
219         CHECKSUM_NONE-硬件不支持，完全由软件执行校验和
220         CHECKSUM_PARTIAL-由硬件来执行校验和
221         CHECKSUM_UNNECESSARY-没必要执行校验和
222         CHECKSUM_COMPLETE-已完成执行校验和
223     */
224     __u8            ip_summed:2;
225     __u8            ooo_okay:1;
226     __u8            l4_hash:1;
227     __u8            sw_hash:1;
228     __u8            wifi_acked_valid:1;
229     __u8            wifi_acked:1;
230 
231     __u8            no_fcs:1;
232     /* Indicates the inner headers are valid in the skbuff. */
233     __u8            encapsulation:1;
234     __u8            encap_hdr_csum:1;
235     __u8            csum_valid:1;
236     __u8            csum_complete_sw:1;
237     __u8            csum_level:2;
238     __u8            csum_bad:1;
239 
240     __u8            dst_pending_confirm:1;
241 #ifdef CONFIG_IPV6_NDISC_NODETYPE
242     __u8            ndisc_nodetype:2;
243 #endif
244     __u8            ipvs_property:1;
245     __u8            inner_protocol_type:1;
246     __u8            remcsum_offload:1;
247 #ifdef CONFIG_NET_SWITCHDEV
248     __u8            offload_fwd_mark:1;
249 #endif
250 #ifdef CONFIG_NET_CLS_ACT
251     __u8            tc_skip_classify:1;
252     __u8            tc_at_ingress:1;
253     __u8            tc_redirected:1;
254     __u8            tc_from_ingress:1;
255 #endif
256 
257 #ifdef CONFIG_NET_SCHED
258     __u16            tc_index;    /* traffic control index */
259 #endif
260 
261     union {
262         /* 校验和，必须包含csum_start和csum_offset */
263         __wsum        csum;
264         struct {
265             /* 校验开始位置，相对于header */
266             __u16    csum_start;
267             /* 校验和存储位置，相对于csum_start */
268             __u16    csum_offset;
269         };
270     };
271     /* 
272         正在被传输的数据包QoS等级
273         数据包由本地产生，套接字会定义优先级的值
274         数据包在被转发，则在调用ip_forward函数时，会根据
275         ip头本身的ToS字段定义该值
276     */
277     __u32            priority;
278     int            skb_iif;
279     __u32            hash;
280     __be16            vlan_proto;
281     __u16            vlan_tci;
282 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
283     union {
284         unsigned int    napi_id;
285         unsigned int    sender_cpu;
286     };
287 #endif
288 #ifdef CONFIG_NETWORK_SECMARK
289     __u32        secmark;
290 #endif
291 
292     union {
293         __u32        mark;
294         __u32        reserved_tailroom;
295     };
296 
297     /* 封装的协议 */
298     union {
299         __be16        inner_protocol;
300         __u8        inner_ipproto;
301     };
302     /* 封装的传输层头部相对于head的偏移 */
303     __u16            inner_transport_header;
304     /* 封装的网络层头部相对于head的偏移 */
305     __u16            inner_network_header;
306     /* 封装的链路层头部相对于head的偏移 */
307     __u16            inner_mac_header;
308 
309     /* 
310         l3层协议值
311         如ETH_P_IP-ipv4报文
312         ETH_P_ARP-arp报文等
313     */
314     __be16            protocol;
315     /* 传输层头部相对于head的偏移 */
316     __u16            transport_header;
317     /* 网络层头部相对于head的偏移 */
318     __u16            network_header;
319     /* 链路层头部相对于head的偏移 */
320     __u16            mac_header;
321 
322     /* private: */
323     __u32            headers_end[0];
324     /* public: */
325 
326     /* These elements must be at the end, see alloc_skb() for details.  */
327     /* 实际数据的尾部 */
328     sk_buff_data_t        tail;
329     /* 缓冲区的尾部 */
330     sk_buff_data_t        end;
331     /* 缓冲区的头部 */
332     unsigned char        *head,
333     /* 实际数据的头部 */
334                 *data;
335     /*
336         缓冲区的总大小，包括skb本身和实际数据len大小，alloc_skb函数将
337         该字段设置为len+sizeof(sk_buff)
338         每当len值更新，该值也要对应更新
339     */
340     unsigned int        truesize;
341     
342     /* 
343         引用计数，在使用该skb缓冲区的实例个数，当引用计数为0时，skb才能被释放
344         skb_get()获取操作中会增加引用计数，kfree_skb释放过程中检查引用计数，
345         引用计数为0时，才真正释放skb
346         该计数器只计算sk_buff结构引用计数，缓冲区包含的实际数据由
347         skb_shared_info->dataref字段记录
348     */
349     atomic_t        users;
350 };