新秀nginx源代码分析数据结构篇（两）双链表ngx_queue

nginx源代码分析数据结构篇（两）双链表ngx_queue_t

Author：Echo Chen（陈斌）

Email：chenb19870707@gmail.com

Blog：Blog.csdn.net/chen19870707

Date：October 20h, 2014

1.ngx_queue优势和特点

ngx_queue作为顺序容器链表。它优势在于其能够高效地运行插入、删除、合并操作，在插入删除的过程中，仅仅须要改动指针指向。而不须要拷贝数据，因此。对于频繁改动的容器非常适合。

此外，相对于STL list，它还具有下面特点：

自身实现了排序功能
轻量级，不负责内存的分配
自身支持两个链表的合并

2.源码位置

头文件：http://trac.nginx.org/nginx/browser/nginx/src/core/ngx_queue.h

源文件：http://trac.nginx.org/nginx/browser/nginx/src/core/ngx_queue.c

3.数据结构定义

   1: typedef struct ngx_queue_s  ngx_queue_t;

2:

   3: struct ngx_queue_s {

   4:     ngx_queue_t  *prev;

   5:     ngx_queue_t  *next;

   6: };

能够看到，它的结构很easy，仅有两个成员：prev、next。这样对于链表中元素来说，空间上仅仅添加了两个指针的消耗。

4.初始化ngx_queue_init

   1: //q 为链表容器结构体ngx_queue_t的指针，将头尾指针指向自己

   2: #define ngx_queue_init(q)                                                     \

   3:     (q)->prev = q;                                                            \

   4:     (q)->next = q

初始状态的链表如图所看到的：

5.推断链表容器是否为空ngx_queue_empty

推断方法很easy，即推断链表的prev指针是否指向自己。如上图所看到的

   1: #define ngx_queue_empty(h)                                                    \

   2:     (h == (h)->prev)

6.头部插入ngx_queue_insert_head

   1: //h为链表指针，x为要插入的元素

   2: #define ngx_queue_insert_head(h, x)                                           \

   3:     (x)->next = (h)->next;                                                    \

   4:     (x)->next->prev = x;                                                      \

   5:     (x)->prev = h;                                                            \

   6:     (h)->next = x

标准的双链表插入四步操作，如图所看到的：

7.尾部插入ngx_queue_insert_tail

与头部插入类似，仅仅是第一步给的h->prev 。即为最后一个结点：

   1: #define ngx_queue_insert_tail(h, x)                                           \

   2:     (x)->prev = (h)->prev;                                                    \

   3:     (x)->prev->next = x;                                                      \

   4:     (x)->next = h;                                                            \

   5:     (h)->prev = x

8.链表删除ngx_queue_remove

x为要删除的结点，将x的下一个的结点的prev指针指向x的上一个结点，再将x的前一个结点的next指针指向x的下一个结点。常规链表双链表结点删除操作，不处理内存释放

   1: #define ngx_queue_remove(x)                                                   \

   2:     (x)->next->prev = (x)->prev;                                              \

   3:     (x)->prev->next = (x)->next

4:

9.链表拆分ngx_queue_split

   1: #define ngx_queue_split(h, q, n)                                              \

   2:     (n)->prev = (h)->prev;                                                    \

   3:     (n)->prev->next = n;                                                      \

   4:     (n)->next = q;                                                            \

   5:     (h)->prev = (q)->prev;                                                    \

   6:     (h)->prev->next = h;                                                      \

   7:     (q)->prev = n;

h为链表容器，q为链表h中的一个元素。这种方法能够将链表h以元素q为界拆分为两个链表h和n，当中h由原链表的前半部分组成（不包括q）。而n由后半部分组成。q为首元素，操作也非常easy，如图所看到的：

10.链表合并ngx_queue_add

   1: #define ngx_queue_add(h, n)                                                   \

   2:     (h)->prev->next = (n)->next;                                              \

   3:     (n)->next->prev = (h)->prev;                                              \

   4:     (h)->prev = (n)->prev;                                                    \

   5:     (h)->prev->next = h;

将链表n 合并到链表h的尾部，如图所看到的：

11. 链表中心元素ngx_queue_middle

   1: ngx_queue_t *ngx_queue_middle(ngx_queue_t *queue)

   2: {

   3:     ngx_queue_t  *middle, *next;

4:

   5:     middle = ngx_queue_head(queue);

6:

   7:     //头尾相等。空链表，返回头就可以

   8:     if (middle == ngx_queue_last(queue)) {

   9:         return middle;

  10:     }

11:

  12:     next = ngx_queue_head(queue);

13:

  14:     for ( ;; ) {

  15:         middle = ngx_queue_next(middle);

  16:         next = ngx_queue_next(next);

17:

  18:         if (next == ngx_queue_last(queue)) {

  19:             return middle;

  20:         }

21:

  22:         next = ngx_queue_next(next);

23:

  24:         if (next == ngx_queue_last(queue)) {

  25:             return middle;

  26:         }

  27:     }

  28: }

这里用到的技巧是每次middle向后移动一步，next向后移动两步，这样next指到队尾的时候，middle就指到了中间，时间复杂度就是O(N)，这是一道经典的面试题。今天在这里看到了源代码，似成相识啊，果然经典面试题目都不是凭空而来。

12.链表排序ngx_queue_sort

能够看到，这里採用的是插入排序算法。时间复杂度为O(n)。整个代码很简洁。

   1: void ngx_queue_sort(ngx_queue_t *queue, ngx_int_t (*cmp)(const ngx_queue_t *, const ngx_queue_t *))

   2: {

   3:     ngx_queue_t  *q, *prev, *next;

4:

   5:     q = ngx_queue_head(queue);

6:

   7:     //假设是空链表。直接返回

   8:     if (q == ngx_queue_last(queue)) {

   9:         return;

  10:     }

11:

  12:     for (q = ngx_queue_next(q); q != ngx_queue_sentinel(queue); q = next) {

13:

  14:         prev = ngx_queue_prev(q);

  15:         next = ngx_queue_next(q);

16:

  17:         ngx_queue_remove(q);

18:

  19:         //找到插入位置

  20:         do {

  21:             if (cmp(prev, q) <= 0) {

  22:                 break;

  23:             }

24:

  25:             prev = ngx_queue_prev(prev);

26:

  27:         } while (prev != ngx_queue_sentinel(queue));

28:

  29:         //插入

  30:         ngx_queue_insert_after(prev, q);

  31:     }

  32: }

13.依据ngx_queue_t 找到链表元素

   1: #define ngx_queue_data(q, type, link)                                         \

   2:     (type *) ((u_char *) q - offsetof(type, link))

当中q为ngx_queue_t* 类型。函数作用为依据q算出。算出链表元素的地址，当中linux接口offsetof是算出link在type中的偏移。

14.其他方法

   1: #define ngx_queue_head(h)                                                     \

   2:     (h)->next

3:

4:

   5: #define ngx_queue_last(h)                                                     \

   6:     (h)->prev

7:

8:

   9: #define ngx_queue_sentinel(h)                                                 \

  10:     (h)

11:

12:

  13: #define ngx_queue_next(q)                                                     \

  14:     (q)->next

15:

16:

  17: #define ngx_queue_prev(q)                                                     \

  18:     (q)->prev

15.实战

   1: #include <iostream>

   2: #include <algorithm>

   3: #include <pthread.h>

   4: #include <time.h>

   5: #include <stdio.h>

   6: #include <errno.h>

   7: #include <string.h>

   8: #include "ngx_queue.h"

9:

  10: struct student_info

  11: {

  12:    long stu_id;

  13:    unsigned int age;

  14:    unsigned int score;

  15:    ngx_queue_t qEle;

  16: };

17:

  18: ngx_int_t compareStudent(const ngx_queue_t *a, const ngx_queue_t *b)

  19: {

  20:     //分别取得a b 对象指针

  21:     student_info *ainfo = ngx_queue_data(a,student_info,qEle);

  22:     student_info *binfo = ngx_queue_data(b,student_info,qEle);

23:

  24:     return ainfo->score >binfo->score;

  25: }

26:

  27: void print_ngx_queue(ngx_queue_t *queue)

  28: {

  29:     //遍历输出

  30:     for(ngx_queue_t *q = ngx_queue_head(queue);q != ngx_queue_sentinel(queue);q = ngx_queue_next(q))

  31:     {

  32:         student_info *info = ngx_queue_data(q,student_info,qEle);

  33:         if(info != NULL)

  34:         {

  35:             std::cout <<info->score << "  ";

  36:         }

  37:     }

38:

  39:     std::cout << std::endl;

  40: }

41:

  42: int main()

  43: {

44:

  45:     ngx_queue_t queue;

  46:     ngx_queue_init(&queue);

47:

  48:     student_info info[5];

  49:     for(int i = 0;i < 5;i++)

  50:     {

  51:         info[i].stu_id = i;

  52:         info[i].age = i;

  53:         info[i].score = i;

54:

  55:         if(i%2)

  56:         {

  57:             ngx_queue_insert_tail(&queue,&info[i].qEle);

  58:         }

  59:         else

  60:         {

  61:             ngx_queue_insert_head(&queue,&info[i].qEle);

  62:         }

  63:     }

64:

  65:     print_ngx_queue(&queue);

66:

  67:     ngx_queue_sort(&queue,compareStudent);

68:

  69:     print_ngx_queue(&queue);

70:

  71:     return 0;

  72: }

输出结果：

16.总结

ngx_queue设计非常静止，基本涵盖了双链表的全部操作，建议须要面试的童鞋看一看，非常多链表的题目都迎刃而解。此外，ngx_queue与其他nginx 代码耦合度低，有须要这样的双向链表的实现时最好还是直接拿过来使用。

Echo Chen：Blog.csdn.net/chen19870707