nginx location的管理以及查找

关于nginx代码解析，我师兄雕梁的博客（http://simohayha.javaeye.com）有一系列的文章可以阅读。我这里将只介绍他博客里没有关注到的或者讲述不详细的，但是我个人又认为是nginx里面比较重要的东西。在这一篇文章里，我将介绍nginx关于location的处理，大家都知道Nginx配置文件里面会有很多的location，nginx的配置指令的作用域可以分为 main，server，location这3个种，实际上这3者不是依次包含的关系，而是相互独立的关系，比如一个只具有main级别作用域的指令，是不能写在某个server或者location内的，模块的某个指令可以同时具有main，server，location这3种作用域，另外每个模块有 main,srv,loc这3个级别的配置，一个模块的main级别的配置对所有的server和location都是共享的，srv级别的配置对所有 location都是共享的，location只有自己独立的loc级别的配置，这就是为什么一个模块的srv和loc级别的配置需要merge，而 main级别的配置不需要merge的原因。这里看起来有点绕，区分一下main，server，location分别作为一种作用域级别和一个主体，类似于形容词和名字的区别，nginx的配置关系还是不难理解的。

一般来说一个请求url过来，nginx会将它解析到某一个location来处理。这个解析的过程实际上根据location的配置基本可以分为字符串匹配和正则表达式匹配这2种。对于location的组织方式，最简单的就是直接将它们保存为一个链表，解析url的时候一个一个遍历即可找到相应location，但是这样效率太低，对像nginx这种高性能的服务器来说是完全不可取的，nginx将字符串匹配的location组织成了一个三叉的字符串排序树，而且建立的时候也考虑了树的平衡性。文章后面我讲详细介绍源码的实现。

首先我来大概的介绍一下location的种类和匹配规则，以nginx wiki(http://wiki.nginx.org/HttpCoreModule#location)的例子做说明：

[cpp] view plain copy print ?

location = / {
# matches the query / only.
[ configuration A ]
}
location / {
# matches any query, since all queries begin with /, but regular
# expressions and any longer conventional blocks will be
# matched first.
[ configuration B ]
}
location ^~ /images/ {
# matches any query beginning with /images/ and halts searching,
# so regular expressions will not be checked.
[ configuration C ]
}
location ~* \.(gif|jpg|jpeg)$ {
# matches any request ending in gif, jpg, or jpeg. However, all
# requests to the /images/ directory will be handled by
# Configuration C.
[ configuration D ]
}
location @named {
# Such locations are not used during normal processing of requests,
# they are intended only to process internally redirected requests (for example error_page, try_files).
[ configuration E ]
}

location  = / {
  # matches the query / only.
  [ configuration A ] 
}
location  / {
  # matches any query, since all queries begin with /, but regular
  # expressions and any longer conventional blocks will be
  # matched first.
  [ configuration B ] 
}
location ^~ /images/ {
  # matches any query beginning with /images/ and halts searching,
  # so regular expressions will not be checked.
  [ configuration C ] 
}
location ~* \.(gif|jpg|jpeg)$ {
  # matches any request ending in gif, jpg, or jpeg. However, all
  # requests to the /images/ directory will be handled by
  # Configuration C.   
  [ configuration D ] 
}

location  @named {
  # Such locations are not used during normal processing of requests, 
  # they are intended only to process internally redirected requests (for example error_page, try_files).
  [ configuration E ] 
}

可以看到上面的例子中有5种不同类型的location，其中第4个带 “~” 号前缀的为需要正则匹配的location，nginx在进行url解析时对这5种不同类型的location具有不同的优先级规则，大致的规则如下：

1，字符串精确匹配到一个带 “=” 号前缀的location，则停止，且使用这个location的配置；

2，字符串匹配剩下的非正则和非特殊location，如果匹配到某个带 "^~" 前缀的location，则停止；

3，正则匹配，匹配顺序为location在配置文件中出现的顺序。如果匹配到某个正则location，则停止，并使用这个location的配置；否则，使用步骤2中得到的具有最大字符串匹配的location配置。

例如，对下面的请求有:

1， / -> 精确匹配到第1个location，匹配停止，使用configuration A
2，/some/other/url -> 首先前缀部分字符串匹配到了第2个location，然后进行正则匹配，显然没有匹配上，则使用第2个location的配置configurationB
3，/images /1.jpg -> 首先前缀部分字符串匹配到了第2个location，但是接着对第3个location也前缀匹配上了，而且这时已经是配置文件里面对这个url的最大字符串匹配了，并且location带有 "^~" 前缀，则不再进行正则匹配，最终使用configuration C
4，/some/other/path/to/1.jpg -> 首先前缀部分同样字符串匹配到了第2个location，然后进行正则匹配，这时正则匹配成功，则使用congifuration D

nginx的url匹配规则实际上有点不妥，大部分情况下一个url必须先进行字符串匹配，然后再做正则匹配，但是实际上如果先做正则匹配，没有匹配上再做字符串匹配，在很多情况下可以节省掉做字符串匹配的时间。不管怎样，先来看一下nginx源码里面的实现，在介绍匹配location过程之前，先来介绍一下nginx里面对location的组织方式，实际上在配置解析阶段，nginx将字符串匹配的location和正则匹配的location分别存储在http core模块的loc配置ngx_http_core_loc_conf_t结构的下面2个字段：

[cpp] view plain copy print ?

ngx_http_location_tree_node_t *static_locations;
(NGX_PCRE)
ngx_http_core_loc_conf_t **regex_locations;
if

    ngx_http_location_tree_node_t   *static_locations;
#if (NGX_PCRE)
    ngx_http_core_loc_conf_t       **regex_locations;
#endif

从这2个字段的类型可以看出，字符串匹配的location被组织成了一个location tree，而正则匹配的location只是一个数组，location tree和regex_locations数组建立过程在ngx_http_block中：

[cpp] view plain copy print ?

/* create location trees */
for (s = 0; s < cmcf->servers.nelts; s++) {
clcf = cscfp[s]->ctx->loc_conf[ngx_http_core_module.ctx_index];
if (ngx_http_init_locations(cf, cscfp[s], clcf) != NGX_OK) {
return NGX_CONF_ERROR;
}
if (ngx_http_init_static_location_trees(cf, clcf) != NGX_OK) {
return NGX_CONF_ERROR;
}
}

/* create location trees */

    for (s = 0; s < cmcf->servers.nelts; s++) {

        clcf = cscfp[s]->ctx->loc_conf[ngx_http_core_module.ctx_index];

        if (ngx_http_init_locations(cf, cscfp[s], clcf) != NGX_OK) {
            return NGX_CONF_ERROR;
        }

        if (ngx_http_init_static_location_trees(cf, clcf) != NGX_OK) {
            return NGX_CONF_ERROR;
        }
    }

经过配置的读取之后，所有server都被保存在http core模块的main配置中的servers数组中，而每个server里面的location都被按配置中出现的顺序保存在http core模块的loc配置的locations队列中，上面的代码中先对每个server的location进行排序和分类处理，这一步发生在 ngx_http_init_location()函数中：

[cpp] view plain copy print ?

static ngx_int_t
ngx_http_init_locations(ngx_conf_t *cf, ngx_http_core_srv_conf_t *cscf,
ngx_http_core_loc_conf_t *pclcf)
{
...
locations = pclcf->locations;
...
/* 按照类型排序location，排序完后的队列： (exact_match 或 inclusive) (排序好的，如果某个exact_match名字和inclusive location相同，exact_match排在前面)
| regex（未排序）| named(排序好的) | noname（未排序）*/
ngx_queue_sort(locations, ngx_http_cmp_locations);
named = NULL;
n = 0;
#if (NGX_PCRE)
regex = NULL;
r = 0;
#endif
for (q = ngx_queue_head(locations);
q != ngx_queue_sentinel(locations);
q = ngx_queue_next(q))
{
lq = (ngx_http_location_queue_t *) q;
clcf = lq->exact ? lq->exact : lq->inclusive;
/* 由于可能存在nested location，也就是location里面嵌套的location，这里需要递归的处理一下当前location下面的nested location */
if (ngx_http_init_locations(cf, NULL, clcf) != NGX_OK) {
return NGX_ERROR;
}
#if (NGX_PCRE)
if (clcf->regex) {
r++;
if (regex == NULL) {
regex = q;
}
continue;
}
#endif
if (clcf->named) {
n++;
if (named == NULL) {
named = q;
}
continue;
}
if (clcf->noname) {
break;
}
}
if (q != ngx_queue_sentinel(locations)) {
ngx_queue_split(locations, q, &tail);
}
/* 如果有named location，将它们保存在所属server的named_locations数组中 */
if (named) {
clcfp = ngx_palloc(cf->pool,
(n + 1) * sizeof(ngx_http_core_loc_conf_t **));
if (clcfp == NULL) {
return NGX_ERROR;
}
cscf->named_locations = clcfp;
for (q = named;
q != ngx_queue_sentinel(locations);
q = ngx_queue_next(q))
{
lq = (ngx_http_location_queue_t *) q;
*(clcfp++) = lq->exact;
}
*clcfp = NULL;
ngx_queue_split(locations, named, &tail);
}
#if (NGX_PCRE)
/* 如果有正则匹配location，将它们保存在所属server的http core模块的loc配置的regex_locations 数组中，
这里和named location保存位置不同的原因是由于named location只能存在server里面，而regex location可以作为nested location */
if (regex) {
clcfp = ngx_palloc(cf->pool,
(r + 1) * sizeof(ngx_http_core_loc_conf_t **));
if (clcfp == NULL) {
return NGX_ERROR;
}
pclcf->regex_locations = clcfp;
for (q = regex;
q != ngx_queue_sentinel(locations);
q = ngx_queue_next(q))
{
lq = (ngx_http_location_queue_t *) q;
*(clcfp++) = lq->exact;
}
*clcfp = NULL;
ngx_queue_split(locations, regex, &tail);
}
#endif
return NGX_OK;
}

static ngx_int_t
ngx_http_init_locations(ngx_conf_t *cf, ngx_http_core_srv_conf_t *cscf,
    ngx_http_core_loc_conf_t *pclcf)
{
  ...
    locations = pclcf->locations;

  ...
    /* 按照类型排序location，排序完后的队列：  (exact_match 或 inclusive) (排序好的，如果某个exact_match名字和inclusive location相同，exact_match排在前面)
       |  regex（未排序）| named(排序好的)  |  noname（未排序）*/
    ngx_queue_sort(locations, ngx_http_cmp_locations);

    named = NULL;
    n = 0;
#if (NGX_PCRE)
    regex = NULL;
    r = 0;
#endif

    for (q = ngx_queue_head(locations);
         q != ngx_queue_sentinel(locations);
         q = ngx_queue_next(q))
    {
        lq = (ngx_http_location_queue_t *) q;

        clcf = lq->exact ? lq->exact : lq->inclusive;
        /* 由于可能存在nested location，也就是location里面嵌套的location，这里需要递归的处理一下当前location下面的nested location */
        if (ngx_http_init_locations(cf, NULL, clcf) != NGX_OK) {
            return NGX_ERROR;
        }

#if (NGX_PCRE)

        if (clcf->regex) {
            r++;

            if (regex == NULL) {
                regex = q;
            }

            continue;
        }

#endif

        if (clcf->named) {
            n++;

            if (named == NULL) {
                named = q;
            }

            continue;
        }

        if (clcf->noname) {
            break;
        }
    }

    if (q != ngx_queue_sentinel(locations)) {
        ngx_queue_split(locations, q, &tail);
    }
    /* 如果有named location，将它们保存在所属server的named_locations数组中 */
    if (named) {
        clcfp = ngx_palloc(cf->pool,
                           (n + 1) * sizeof(ngx_http_core_loc_conf_t **));
        if (clcfp == NULL) {
            return NGX_ERROR;
        }

        cscf->named_locations = clcfp;

        for (q = named;
             q != ngx_queue_sentinel(locations);
             q = ngx_queue_next(q))
        {
            lq = (ngx_http_location_queue_t *) q;

            *(clcfp++) = lq->exact;
        }

        *clcfp = NULL;

        ngx_queue_split(locations, named, &tail);
    }

#if (NGX_PCRE)
    /* 如果有正则匹配location，将它们保存在所属server的http core模块的loc配置的regex_locations 数组中，
       这里和named location保存位置不同的原因是由于named location只能存在server里面，而regex location可以作为nested location */
    if (regex) {

        clcfp = ngx_palloc(cf->pool,
                           (r + 1) * sizeof(ngx_http_core_loc_conf_t **));
        if (clcfp == NULL) {
            return NGX_ERROR;
        }

        pclcf->regex_locations = clcfp;

        for (q = regex;
             q != ngx_queue_sentinel(locations);
             q = ngx_queue_next(q))
        {
            lq = (ngx_http_location_queue_t *) q;

            *(clcfp++) = lq->exact;
        }

        *clcfp = NULL;

        ngx_queue_split(locations, regex, &tail);
    }

#endif

    return NGX_OK;
}

上面的步骤将正则匹配的location保存好了，location tree的建立在ngx_http_init_static_location_trees中进行：

[cpp] view plain copy print ?

static ngx_int_t
ngx_http_init_static_location_trees(ngx_conf_t *cf,
ngx_http_core_loc_conf_t *pclcf)
{
ngx_queue_t *q, *locations;
ngx_http_core_loc_conf_t *clcf;
ngx_http_location_queue_t *lq;
locations = pclcf->locations;
if (locations == NULL) {
return NGX_OK;
}
if (ngx_queue_empty(locations)) {
return NGX_OK;
}
/* 这里也是由于nested location，需要递归一下 */
for (q = ngx_queue_head(locations);
q != ngx_queue_sentinel(locations);
q = ngx_queue_next(q))
{
lq = (ngx_http_location_queue_t *) q;
clcf = lq->exact ? lq->exact : lq->inclusive;
if (ngx_http_init_static_location_trees(cf, clcf) != NGX_OK) {
return NGX_ERROR;
}
}
/* join队列中名字相同的inclusive和exact类型location，也就是如果某个exact_match的location名字和普通字符串匹配的location名字相同的话，
就将它们合到一个节点中，分别保存在节点的exact和inclusive下，这一步的目的实际是去重，为后面的建立排序树做准备 */
if (ngx_http_join_exact_locations(cf, locations) != NGX_OK) {
return NGX_ERROR;
}
/* 递归每个location节点，得到当前节点的名字为其前缀的location的列表，保存在当前节点的list字段下 */
ngx_http_create_locations_list(locations, ngx_queue_head(locations));
/* 递归建立location三叉排序树 */
pclcf->static_locations = ngx_http_create_locations_tree(cf, locations, 0);
if (pclcf->static_locations == NULL) {
return NGX_ERROR;
}
return NGX_OK;
}

static ngx_int_t
ngx_http_init_static_location_trees(ngx_conf_t *cf,
    ngx_http_core_loc_conf_t *pclcf)
{
    ngx_queue_t                *q, *locations;
    ngx_http_core_loc_conf_t   *clcf;
    ngx_http_location_queue_t  *lq;

    locations = pclcf->locations;

    if (locations == NULL) {
        return NGX_OK;
    }

    if (ngx_queue_empty(locations)) {
        return NGX_OK;
    }
    /* 这里也是由于nested location，需要递归一下 */
    for (q = ngx_queue_head(locations);
         q != ngx_queue_sentinel(locations);
         q = ngx_queue_next(q))
    {
        lq = (ngx_http_location_queue_t *) q;

        clcf = lq->exact ? lq->exact : lq->inclusive;

        if (ngx_http_init_static_location_trees(cf, clcf) != NGX_OK) {
            return NGX_ERROR;
        }
    }
    /* join队列中名字相同的inclusive和exact类型location，也就是如果某个exact_match的location名字和普通字符串匹配的location名字相同的话，
       就将它们合到一个节点中，分别保存在节点的exact和inclusive下，这一步的目的实际是去重，为后面的建立排序树做准备 */
    if (ngx_http_join_exact_locations(cf, locations) != NGX_OK) {
        return NGX_ERROR;
    }
    /* 递归每个location节点，得到当前节点的名字为其前缀的location的列表，保存在当前节点的list字段下 */
    ngx_http_create_locations_list(locations, ngx_queue_head(locations));

    /* 递归建立location三叉排序树 */
    pclcf->static_locations = ngx_http_create_locations_tree(cf, locations, 0);
    if (pclcf->static_locations == NULL) {
        return NGX_ERROR;
    }

    return NGX_OK;
}

经过ngx_http_init_location（）函数处理之后，locations队列已经是排好序的了，建立三叉树的过程的主要工作都在ngx_http_create_locations_list()和ngx_http_create_locations_tree()中完成，这2个函数都是递归函数，第1个函数递归locations队列中的每个节点，得到以当前节点的名字为前缀的location，并保存在当前节点的list字段下，例如，对下列location：

[cpp] view plain copy print ?

location /xyz {
}
location = /xyz {
}
location /xyza {
}
location /xyzab {
}
location /xyzb {
}
location /abc {
}
location /efg {
}
location /efgaa {
}

location  /xyz {

}

location  = /xyz {

}
location  /xyza {

}

location  /xyzab {

}
location  /xyzb {

}
location  /abc {

}
location  /efg {

}
location  /efgaa {

}

排序的结果为/abc /efg /efgaa =/xyz /xyz /xyza /xyzab /xyzb，去重后结果为 /abc /efg /efgaa /xyz /xyza /xyzab/xyzb，ngx_http_create_locations_list()执行后的结果为：

最后，来看下ngx_http_create_locations_tree函数：

[cpp] view plain copy print ?

static ngx_http_location_tree_node_t *
ngx_http_create_locations_tree(ngx_conf_t *cf, ngx_queue_t *locations,
size_t prefix)
{
...
/* 根节点为locations队列的中间节点 */
q = ngx_queue_middle(locations);
lq = (ngx_http_location_queue_t *) q;
len = lq->name->len - prefix;
node = ngx_palloc(cf->pool,
offsetof(ngx_http_location_tree_node_t, name) + len);
if (node == NULL) {
return NULL;
}
node->left = NULL;
node->right = NULL;
node->tree = NULL;
node->exact = lq->exact;
node->inclusive = lq->inclusive;
node->auto_redirect = (u_char) ((lq->exact && lq->exact->auto_redirect)
|| (lq->inclusive && lq->inclusive->auto_redirect));
node->len = (u_char) len;
ngx_memcpy(node->name, &lq->name->data[prefix], len);
/* 从中间节点开始断开 */
ngx_queue_split(locations, q, &tail);
if (ngx_queue_empty(locations)) {
/*
* ngx_queue_split() insures that if left part is empty,
* then right one is empty too
*/
goto inclusive;
}
/* 从locations左半部分得到左子树 */
node->left = ngx_http_create_locations_tree(cf, locations, prefix);
if (node->left == NULL) {
return NULL;
}
ngx_queue_remove(q);
if (ngx_queue_empty(&tail)) {
goto inclusive;
}
/* 从locations右半部分得到右子树 */
node->right = ngx_http_create_locations_tree(cf, &tail, prefix);
if (node->right == NULL) {
return NULL;
}
inclusive:
if (ngx_queue_empty(&lq->list)) {
return node;
}
/* 从list队列得到tree子树 */
node->tree = ngx_http_create_locations_tree(cf, &lq->list, prefix + len);
if (node->tree == NULL) {
return NULL;
}
return node;
}

static ngx_http_location_tree_node_t *
ngx_http_create_locations_tree(ngx_conf_t *cf, ngx_queue_t *locations,
    size_t prefix)
{
   ...
    /* 根节点为locations队列的中间节点 */
    q = ngx_queue_middle(locations);

    lq = (ngx_http_location_queue_t *) q;
    len = lq->name->len - prefix;
    
    node = ngx_palloc(cf->pool,
                      offsetof(ngx_http_location_tree_node_t, name) + len);
    if (node == NULL) {
        return NULL;
    }

    node->left = NULL;
    node->right = NULL;
    node->tree = NULL;
    node->exact = lq->exact;
    node->inclusive = lq->inclusive;

    node->auto_redirect = (u_char) ((lq->exact && lq->exact->auto_redirect)
                           || (lq->inclusive && lq->inclusive->auto_redirect));

    node->len = (u_char) len;
    ngx_memcpy(node->name, &lq->name->data[prefix], len);

    /* 从中间节点开始断开 */
    ngx_queue_split(locations, q, &tail);

    if (ngx_queue_empty(locations)) {
        /*
         * ngx_queue_split() insures that if left part is empty,
         * then right one is empty too
         */
        goto inclusive;
    }

    /* 从locations左半部分得到左子树 */
    node->left = ngx_http_create_locations_tree(cf, locations, prefix);
    if (node->left == NULL) {
        return NULL;
    }

    ngx_queue_remove(q);

    if (ngx_queue_empty(&tail)) {
        goto inclusive;
    }
 

    /* 从locations右半部分得到右子树 */
    node->right = ngx_http_create_locations_tree(cf, &tail, prefix);
    if (node->right == NULL) {
        return NULL;
    }

inclusive:

    if (ngx_queue_empty(&lq->list)) {
        return node;
    }

    /* 从list队列得到tree子树 */
    node->tree = ngx_http_create_locations_tree(cf, &lq->list, prefix + len);
    if (node->tree == NULL) {
        return NULL;
    }

    return node;
}

location tree节点的ngx_http_location_tree_node_s结构：

[cpp] view plain copy print ?

struct ngx_http_location_tree_node_s {
ngx_http_location_tree_node_t *left;
ngx_http_location_tree_node_t *right;
ngx_http_location_tree_node_t *tree;
ngx_http_core_loc_conf_t *exact;
ngx_http_core_loc_conf_t *inclusive;
u_char auto_redirect;
u_char len;
u_char name[1];
};

struct ngx_http_location_tree_node_s {
    ngx_http_location_tree_node_t   *left;
    ngx_http_location_tree_node_t   *right;
    ngx_http_location_tree_node_t   *tree;

    ngx_http_core_loc_conf_t        *exact;
    ngx_http_core_loc_conf_t        *inclusive;

    u_char                           auto_redirect;
    u_char                           len;
    u_char                           name[1];
};

location tree结构用到的是left，right，tree 这3个字段， location tree实际上是一个三叉的字符串排序树，而且这里如果某个节点只考虑左，右子树，它是一颗平衡树，它的建立过程有点类似于一颗平衡排序二叉树的建立过程，先排序再用二分查找找到的节点顺序插入，ngx_http_location_tree_node_s的tree节点也是一颗平衡排序树，它是用该节点由ngx_http_create_locations_list()得到的list建立的，也就是该节点的名字是它的tree子树里面的所有节点名字的前缀，所以tree子树里面的所有节点的名字不用保存公共前缀，而且查找的时候，如果是转向tree节点的话，也是不需要再比较父节点的那段字符串了。
ngx_http_create_locations_tree()函数写的很清晰，它有一个参数是队列locations，它返回一颗三叉树，根节点为locations的中间节点，其左子树为locations队列的左半部分建立的location tree，右子树为location队列的右半部分建立的tree，tree节点为该根节点的list队列建立的tree。

最终建立的location tree如下（为了方便阅读，图中列出了tree节点的完整名字）：

location的匹配发生在nginx请求处理的NGX_HTTP_FIND_CONFIG_PHASE阶段(请求的处理所有PHASE可以看这篇博客http://simohayha.iteye.com/blog/670326)，匹配的规则文章前面已经做过介绍，源码的实现也很简单，这里不再做介绍了。
(责任编辑：IT)

搜索

热门标签:

nginx location的管理以及查找