PHP内核探索：预定义变量

大家都知道PHP脚本在执行的时候用户全局变量(在用户空间显式定义的变量)会保存在一个HashTable数据类型的符号表(symbol_table)中，在PHP中有一些比较特殊的全局变量例如: $_GET，$_POST，$_SERVER等变量，我们并没有在程序中定义这些变量，并且这些变量也同样保存在符号表中，从这些表象我们不难得出结论：PHP是在脚本运行之前就将这些特殊的变量加入到了符号表中了。

$GLOBALS的初始化

我们以cgi模式为例说明$GLOBALS的初始化。从cgi_main.c文件main函数开始。整个调用顺序如下所示：

[main() -> php_request_startup() -> zend_activate() -> init_executor() ]

01... //  省略
02zend_hash_init(&EG(symbol_table), 50, NULL, ZVAL_PTR_DTOR, 0);
03{
04    zval *globals;
05  
06    ALLOC_ZVAL(globals);
07    Z_SET_REFCOUNT_P(globals, 1);
08    Z_SET_ISREF_P(globals);
09    Z_TYPE_P(globals) = IS_ARRAY;
10    Z_ARRVAL_P(globals) = &EG(symbol_table);
11    zend_hash_update(&EG(symbol_table), "GLOBALS", sizeof("GLOBALS"),
12        &globals, sizeof(zval *), NULL);      //  添加全局变量GLOBALS
13}
14... //  省略

上面的代码的关键点zend_hash_update函数的调用，它将变量名为GLOBALS的变量注册到EG(symbol_table)中， EG(symbol_table)是一个HashTable的结构，用来存放所有的全局变量。这在下面将要提到的$_GET等变量初始化时也会用到。

$_GET、$_POST等变量的初始化

$_GET、$_COOKIE、$_SERVER、$_ENV、$_FILES、$_REQUEST这六个变量都是通过如下的调用序列进行初始化。 [main() -> php_request_startup() -> php_hash_environment() ]

在请求初始化时，通过调用 php_hash_environment 函数初始化以上的六个预定义的变量。如下所示为php_hash_environment函数的代码。在代码之后我们以$_POST为例说明整个初始化的过程。

001/* {{{ php_hash_environment
002 */
003int php_hash_environment(TSRMLS_D)
004{
005        char *p;
006        unsigned char _gpc_flags[5] = {0, 0, 0, 0, 0};
007        zend_bool jit_initialization = (PG(auto_globals_jit) && !PG(register_globals) && !PG(register_long_arrays));
008        struct auto_global_record {
009                char *name;
010                uint name_len;
011                char *long_name;
012                uint long_name_len;
013                zend_bool jit_initialization;
014        } auto_global_records[] = {
015                { "_POST", sizeof("_POST"), "HTTP_POST_VARS", sizeof("HTTP_POST_VARS"), 0 },
016                { "_GET", sizeof("_GET"), "HTTP_GET_VARS", sizeof("HTTP_GET_VARS"), 0 },
017                { "_COOKIE", sizeof("_COOKIE"), "HTTP_COOKIE_VARS", sizeof("HTTP_COOKIE_VARS"), 0 },
018                { "_SERVER", sizeof("_SERVER"), "HTTP_SERVER_VARS", sizeof("HTTP_SERVER_VARS"), 1 },
019                { "_ENV", sizeof("_ENV"), "HTTP_ENV_VARS", sizeof("HTTP_ENV_VARS"), 1 },
020                { "_FILES", sizeof("_FILES"), "HTTP_POST_FILES", sizeof("HTTP_POST_FILES"), 0 },
021        };
022        size_t num_track_vars = sizeof(auto_global_records)/sizeof(struct auto_global_record);
023        size_t i;
024  
025        /* jit_initialization = 0; */
026        for (i=0; i< num_track_vars; i++) {
027                PG(http_globals)[i] = NULL;
028        }
029  
030        for (p=PG(variables_order); p && *p; p++) {
031                switch(*p) {
032                        case 'p':
033                        case 'P':
034                                if (!_gpc_flags[0] && !SG(headers_sent) && SG(request_info).request_method && !strcasecmp(SG(request_info).request_method, "POST")) {
035                                        sapi_module.treat_data(PARSE_POST, NULL, NULL TSRMLS_CC);   /* POST Data */
036                                        _gpc_flags[0] = 1;
037                                        if (PG(register_globals)) {
038                                                php_autoglobal_merge(&EG(symbol_table), Z_ARRVAL_P(PG(http_globals)[TRACK_VARS_POST]) TSRMLS_CC);
039                                        }
040                                }
041                                break;
042                        case 'c':
043                        case 'C':
044                                if (!_gpc_flags[1]) {
045                                        sapi_module.treat_data(PARSE_COOKIE, NULL, NULL TSRMLS_CC); /* Cookie Data */
046                                        _gpc_flags[1] = 1;
047                                        if (PG(register_globals)) {
048                                                php_autoglobal_merge(&EG(symbol_table), Z_ARRVAL_P(PG(http_globals)[TRACK_VARS_COOKIE]) TSRMLS_CC);
049                                        }
050                                }
051                                break;
052                        case 'g':
053                        case 'G':
054                                if (!_gpc_flags[2]) {
055                                        sapi_module.treat_data(PARSE_GET, NULL, NULL TSRMLS_CC);    /* GET Data */
056                                        _gpc_flags[2] = 1;
057                                        if (PG(register_globals)) {
058                                                php_autoglobal_merge(&EG(symbol_table), Z_ARRVAL_P(PG(http_globals)[TRACK_VARS_GET]) TSRMLS_CC);
059                                        }
060                                }
061                                break;
062                        case 'e':
063                        case 'E':
064                                if (!jit_initialization && !_gpc_flags[3]) {
065                                        zend_auto_global_disable_jit("_ENV", sizeof("_ENV")-1 TSRMLS_CC);
066                                        php_auto_globals_create_env("_ENV", sizeof("_ENV")-1 TSRMLS_CC);
067                                        _gpc_flags[3] = 1;
068                                        if (PG(register_globals)) {
069                                                php_autoglobal_merge(&EG(symbol_table), Z_ARRVAL_P(PG(http_globals)[TRACK_VARS_ENV]) TSRMLS_CC);
070                                        }
071                                }
072                                break;
073                        case 's':
074                        case 'S':
075                                if (!jit_initialization && !_gpc_flags[4]) {
076                                        zend_auto_global_disable_jit("_SERVER", sizeof("_SERVER")-1 TSRMLS_CC);
077                                        php_register_server_variables(TSRMLS_C);
078                                        _gpc_flags[4] = 1;
079                                        if (PG(register_globals)) {
080                                                php_autoglobal_merge(&EG(symbol_table), Z_ARRVAL_P(PG(http_globals)[TRACK_VARS_SERVER]) TSRMLS_CC);
081                                        }
082                                }
083                                break;
084                }
085        }
086  
087        /* argv/argc support */
088        if (PG(register_argc_argv)) {
089                php_build_argv(SG(request_info).query_string, PG(http_globals)[TRACK_VARS_SERVER] TSRMLS_CC);
090        }
091  
092        for (i=0; i < num_track_vars; i++) {
093                if (jit_initialization && auto_global_records[i].jit_initialization) {
094                        continue;
095                }
096                if (!PG(http_globals)[i]) {
097                        ALLOC_ZVAL(PG(http_globals)[i]);
098                        array_init(PG(http_globals)[i]);
099                        INIT_PZVAL(PG(http_globals)[i]);
100                }
101  
102                Z_ADDREF_P(PG(http_globals)[i]);
103                zend_hash_update(&EG(symbol_table), auto_global_records[i].name, auto_global_records[i].name_len, &PG(http_globals)[i], sizeof(zval *), NULL);
104                if (PG(register_long_arrays)) {
105                        zend_hash_update(&EG(symbol_table), auto_global_records[i].long_name, auto_global_records[i].long_name_len, &PG(http_globals)[i], sizeof(zval *), NULL);
106                        Z_ADDREF_P(PG(http_globals)[i]);
107                }
108        }
109  
110        /* Create _REQUEST */
111        if (!jit_initialization) {
112                zend_auto_global_disable_jit("_REQUEST", sizeof("_REQUEST")-1 TSRMLS_CC);
113                php_auto_globals_create_request("_REQUEST", sizeof("_REQUEST")-1 TSRMLS_CC);
114        }
115  
116        return SUCCESS;
117}

以$_POST为例，首先以 auto_global_record 数组形式定义好将要初始化的变量的相关信息。在变量初始化完成后，按照PG(variables_order)指定的顺序（在php.ini中指定），通过调用sapi_module.treat_data处理数据。

从PHP实现的架构设计看，treat_data函数在SAPI目录下不同的服务器应该有不同的实现，只是现在大部分都是使用的默认实现。

在treat_data后，如果打开了PG(register_globals)，则会调用php_autoglobal_merge将相关变量的值写到符号表。

以上的所有数据处理是一个赋值前的初始化行为。在此之后，通过遍历之前定义的结构体，调用zend_hash_update，将相关变量的值赋值给&EG(symbol_table)。另外对于$_REQUEST有独立的处理方法。

Arthur's Blog

$GLOBALS的初始化

$_GET、$_POST等变量的初始化

导航

最近发布

友情链接