quagga源碼分析--路由信息處理zebra-rib

對于各個協議生成的路由信息的處理屬于quagga中非常重要的一個功能，如何在內核進行路由增加，更新，刪除是一個復雜的過程。

quagga在thread任務調度中加入了一種工作隊列，work_queue，與內核的工作隊列類似，是一種相對而言，低優先級的任務，這里的任務看成類似的系統進程。

1、隊列初始化：

/* initialise zebra rib work queue */
static void
rib_queue_init(struct zebra_t *zebra)
{
    assert(zebra);

    if (!(zebra->ribq = work_queue_new(zebra->master,
                                       "route_node processing")))
    {
        zlog_err("%s: could not initialise work queue!", __func__);
        return;
    }

    /* fill in the work queue spec */
    zebra->ribq->spec.workfunc = &meta_queue_process;
    zebra->ribq->spec.errorfunc = NULL;
    /* XXX: TODO: These should be runtime configurable via vty */
    zebra->ribq->spec.max_retries = 3;
    zebra->ribq->spec.hold = rib_process_hold_time;

    if (!(zebra->mq = meta_queue_new()))
    {
        zlog_err("%s: could not initialise meta queue!", __func__);
        return;
    }
    return;
}

第19行，zebra->ribq->spec.hold = rib_process_hold_time; 指定了rib工作隊列在thread_fetch的時候會等待10毫秒

?

/* Hold time for RIB process, should be very minimal.
 * it is useful to able to set it otherwise for testing, hence exported
 * as global here for test-rig code.
 */
int rib_process_hold_time = 10;

?

在添加thread任務的時候進行了時間單位換算：

/* Add a background thread, with an optional millisec delay */
struct thread*
funcname_thread_add_background(struct thread_master *m,
                               int (*func)(struct thread *),
                               void *arg, long delay,
                               debugargdef) {
    struct timeval trel;

    assert(m != NULL);

    if (delay) {
        trel.tv_sec = delay / 1000;
        trel.tv_usec = 1000 * (delay % 1000);
    } else {
        trel.tv_sec = 0;
        trel.tv_usec = 0;
    }

    return funcname_thread_add_timer_timeval(m, func, THREAD_BACKGROUND,
                                             arg, &trel, debugargpass);
}

OK，meta_queue_process，就指定了工作隊列在調度執行的處理函數，由此guagga就會一直同步更新路由了。

2、每個子網的下一跳路由表項的描述：

quagga使用了雙向鏈表來管理表項，定義了路由表現的詳細信息，但比如 status 這個字段是用來在更新路由時來做比較的關鍵字段。如下宏定義了3種狀態：

#define RIB_ENTRY_REMOVED?? ?(1 << 0)
#define RIB_ENTRY_CHANGED?? ?(1 << 1)
#define RIB_ENTRY_SELECTED_FIB?? ?(1 << 2)

struct rib { 
  struct rib *next;         /* Link list. */
  struct rib *prev;
  struct nexthop *nexthop; /* Nexthop structure */
  unsigned long refcnt;    /* Refrence count. */
  time_t uptime;           /* Uptime. */
  int type;                /* Type fo this route. */
  vrf_id_t vrf_id;         /* VRF identifier. */
  int table;               /* Which routing table */
  u_int32_t metric;        /* Metric */
  u_int32_t mtu;           /* MTU */
  u_int32_t nexthop_mtu;
  u_char distance;         /* Distance. */
  u_char flags;             /* Flags of this route. in lib/zebra.h ZEBRA_FLAG_* */
  u_char status;            /* RIB internal status */
#define RIB_ENTRY_REMOVED    (1 << 0)
#define RIB_ENTRY_CHANGED    (1 << 1)
#define RIB_ENTRY_SELECTED_FIB    (1 << 2)
  u_char nexthop_num;        /* Nexthop information. */
  u_char nexthop_active_num;
  u_char nexthop_fib_num;
};

?3、整個路由表的描述：

/* Routing table top structure. */
struct route_table {struct route_node *top;/** Delegate that performs certain functions for this table.*/route_table_delegate_t *delegate;unsigned long count;void *info;     /* User data. */
};

route_table包含了一個二叉樹結構來保存所有的路由前綴和下一跳路由表項，prefix結構保持了路由前綴的長度和值，用來做最長前綴匹配：

/* Each routing entry. */
struct route_node {
  struct prefix p;   /* Actual prefix of this radix. */
  struct route_table *table;   /* Tree link. */
  struct route_node *parent;
  struct route_node *link[2];
  unsigned int lock; /* Lock of this radix */
  void *info;        /* Each node of route. */
  void *aggregate;   /* Aggregation. */

#define l_left   link[0]
#define l_right  link[1]
};

呃，說好的mtire樹呢? 好吧，我們不太可能把成千上萬的路由表項塞給linux內核，夠用就行。

?

?