对于各个协议生成的路由信息的处理属于quagga中非常重要的一个功能,如何在内核进行路由增加,更新,删除是一个复杂的过程。
quagga在thread任务调度中加入了一种工作队列,work_queue,与内核的工作队列类似,是一种相对而言,低优先级的任务,这里的任务看成类似的系统进程。
1、队列初始化:
1 /*initialise zebra rib work queue*/
2 static void
3 rib_queue_init(struct zebra_t *zebra)4 {5 assert(zebra);6
7 if (!(zebra->ribq = work_queue_new(zebra->master,8 "route_node processing")))9 {10 zlog_err("%s: could not initialise work queue!", __func__);11 return;12 }13
14 /*fill in the work queue spec*/
15 zebra->ribq->spec.workfunc = &meta_queue_process;16 zebra->ribq->spec.errorfunc =NULL;17 /*XXX: TODO: These should be runtime configurable via vty*/
18 zebra->ribq->spec.max_retries = 3;19 zebra->ribq->spec.hold =rib_process_hold_time;20
21 if (!(zebra->mq =meta_queue_new()))22 {23 zlog_err("%s: could not initialise meta queue!", __func__);24 return;25 }26 return;27 }
第19行,zebra->ribq->spec.hold =rib_process_hold_time; 指定了rib工作队列在thread_fetch的时候会等待10毫秒
1 /*Hold time for RIB process, should be very minimal.2 * it is useful to able to set it otherwise for testing, hence exported3 * as global here for test-rig code.4 */
5 int rib_process_hold_time = 10;
在添加thread任务的时候进行了时间单位换算:
1 /*Add a background thread, with an optional millisec delay*/
2 struct thread*
3 funcname_thread_add_background(struct thread_master *m,4 int (*func)(struct thread *),5 void *arg, longdelay,6 debugargdef) {7 structtimeval trel;8
9 assert(m !=NULL);10
11 if(delay) {12 trel.tv_sec = delay / 1000;13 trel.tv_usec = 1000 * (delay % 1000);14 } else{15 trel.tv_sec = 0;16 trel.tv_usec = 0;17 }18
19 returnfuncname_thread_add_timer_timeval(m, func, THREAD_BACKGROUND,20 arg, &trel, debugargpass);21 }
OK,meta_queue_process,就指定了工作队列在调度执行的处理函数,由此guagga就会一直同步更新路由了。
2、每个子网的下一跳路由表项的描述:
quagga使用了双向链表来管理表项,定义了路由表现的详细信息,但比如 status 这个字段是用来在更新路由时来做比较的关键字段。如下宏定义了3种状态:
#define RIB_ENTRY_REMOVED (1 << 0)
#define RIB_ENTRY_CHANGED (1 << 1)
#define RIB_ENTRY_SELECTED_FIB (1 << 2)
1 structrib {2 struct rib *next; /*Link list.*/
3 struct rib *prev;4 struct nexthop *nexthop; /*Nexthop structure*/
5 unsigned long refcnt; /*Refrence count.*/
6 time_t uptime; /*Uptime.*/
7 int type; /*Type fo this route.*/
8 vrf_id_t vrf_id; /*VRF identifier.*/
9 int table; /*Which routing table*/
10 u_int32_t metric; /*Metric*/
11 u_int32_t mtu; /*MTU*/
12 u_int32_t nexthop_mtu;13 u_char distance; /*Distance.*/
14 u_char flags; /*Flags of this route. in lib/zebra.h ZEBRA_FLAG_**/
15 u_char status; /*RIB internal status*/
16 #define RIB_ENTRY_REMOVED (1 << 0)
17 #define RIB_ENTRY_CHANGED (1 << 1)
18 #define RIB_ENTRY_SELECTED_FIB (1 << 2)
19 u_char nexthop_num; /*Nexthop information.*/
20 u_char nexthop_active_num;21 u_char nexthop_fib_num;22 };
3、整个路由表的描述:
/*Routing table top structure.*/
structroute_table {struct route_node *top;/** Delegate that performs certain functions for this table.*/route_table_delegate_t*delegate;
unsignedlongcount;void *info; /*User data.*/};
route_table包含了一个二叉树结构来保存所有的路由前缀和下一跳路由表项,prefix结构保持了路由前缀的长度和值,用来做最长前缀匹配:
1 /*Each routing entry.*/
2 structroute_node {3 struct prefix p; /*Actual prefix of this radix.*/
4 struct route_table *table; /*Tree link.*/
5 struct route_node *parent;6 struct route_node *link[2];7 unsigned int lock; /*Lock of this radix*/
8 void *info; /*Each node of route.*/
9 void *aggregate; /*Aggregation.*/
10
11 #define l_left link[0]
12 #define l_right link[1]
13 };
呃,说好的mtire树呢? 好吧,我们不太可能把成千上万的路由表项塞给linux内核,够用就行。