PHP manages global status

                                                                           

Manage global state

Some global space is always needed in imperative languages. When programming PHP or extensions, we will make a clear distinction between what we call request-bound global variables and true global variables.

Request global variables are global variables that need to carry and remember information during request processing. A simple example is when you ask the user to provide a value in a function parameter and want to be able to use it in other functions. Except that this information "holds its value" across several PHP function calls, it only retains the value for the current request. The next request that comes should know nothing. PHP provides a mechanism to manage request global variables regardless of the multiprocessing model chosen, which we will cover in detail later in this chapter.

Real global variables are pieces of information that are retained across requests. This information is usually read-only. If you need to write to such a global variable as part of request handling, PHP can't help you. If you use threads as the multiprocessing model, you need to implement memory locking yourself. If you use processes as the multiprocessing model, you need to use your own IPC (inter-process communication). However, this should not happen in PHP extension programming.

Related learning recommendations: PHP programming from entry to proficiency

Manage request global variables

The following is a request global variable Simple extension example:

/* 真正的 C 全局 */
static zend_long rnd = 0;

static void pib_rnd_init(void)
{
    /* 在 0 到 100 之间随机一个数字 */
    php_random_int(0, 100, &rnd, 0);
}

PHP_RINIT_FUNCTION(pib)
{
    pib_rnd_init();

    return SUCCESS;
}

PHP_FUNCTION(pib_guess)
{
    zend_long r;

    if (zend_parse_parameters(ZEND_NUM_ARGS(), "l", &r) == FAILURE) {
        return;
    }

    if (r == rnd) {
        /* 将数字重置以进行猜测 */
        pib_rnd_init();
        RETURN_TRUE;
    }

    if (r < rnd) {
        RETURN_STRING("more");
    }

    RETURN_STRING("less");
}

PHP_FUNCTION(pib_reset)
{
    if (zend_parse_parameters_none() == FAILURE) {
        return;
    }

    pib_rnd_init();
}

As you can see, this extension picks a random integer at the beginning of the request, and then tries to guess the array through pib_guess(). Once guessed, the number will reset. If the user wants to manually reset the number, it can also manually call pib_reset() to reset the value.
The random number is implemented as a C global variable. If PHP is used in-process as part of a multi-process model it is no longer an issue, if threads are used later, this is not okay.

NOTE

As a reminder, you do not need to know which multi-process model you are going to use. When you design an extension, you must be prepared for both models.

When using threads, a C global variable is shared for each thread in the server. For example, in our example above, each concurrent user of the network server will share the same value. Some may reset the value initially, while others try to guess it. In short, you clearly understand the key issue with threads.

We must persist data to the same request. Even if running PHP's multi-process model will utilize threads, it must be bound to the current request.

Use TSRM macros to protect global space

PHP is designed with a layer that helps extension and kernel developers handle global requests. This layer is called TSRM (Thread-Safe Resource Management) and is exposed as a set of macros that you must use any time you need to access the request-bound globals (read and write).

In the case of a multi-process model using processes, behind the scenes, these macros will be parsed into code similar to the one we showed above. As we can see, the above code is completely valid if threading is not used. So, when using processes, these macros will be expanded to similar macros.

The first thing you have to do is declare a structure that will be the root of all your global variables:

ZEND_BEGIN_MODULE_GLOBALS(pib)
    zend_long rnd;
ZEND_END_MODULE_GLOBALS(pib)

/* 解析为 :
*
* typedef struct _zend_pib_globals {
*    zend_long rnd;
* } zend_pib_globals;
*/

Then, create a global variable like this:

ZEND_DECLARE_MODULE_GLOBALS(pib)

/* 解析为 zend_pib_globals pib_globals; */

Now, you can access the data using global macro accessors. This macro is created by the framework and should be defined in your php_pib.h header file. This looks like this:

#ifdef ZTS
#define PIB_G(v) ZEND_MODULE_GLOBALS_ACCESSOR(pib, v)
#else
#define PIB_G(v) (pib_globals.v)
#endif

As you can see, if ZTS mode is not enabled, i.e. compiling non-thread-safe PHP and extensions (we call it NTS mode: non-threaded safe), the macro simply resolves to the data declared in the structure. Therefore, there are the following changes:

static void pib_rnd_init(void)
{
    php_random_int(0, 100, &PIB_G(rnd), 0);
}

PHP_FUNCTION(pib_guess)
{
    zend_long r;

    if (zend_parse_parameters(ZEND_NUM_ARGS(), "l", &r) == FAILURE) {
        return;
    }

    if (r == PIB_G(rnd)) {
        pib_rnd_init();
        RETURN_TRUE;
    }

    if (r < PIB_G(rnd)) {
        RETURN_STRING("more");
    }

    RETURN_STRING("less");
}

Note

When using a process model, TSRM macros resolve to accesses to C global variables.

When using threads, i.e. when you compile ZTS PHP, things get more complicated. Then, all the macros we see resolve to something completely different, which is hard to explain here. Basically, TSRM does a hard job using TLS (Thread Local Storage) when compiled with ZTS.

NOTE

In short, when compiling in ZTS, global variables are bound to the current thread. When compiling NTS, global variables will be bound to the current process. The TSRM macro handles the hard work. You may be interested in how this works, and browse the /TSRM directory of the PHP source code to learn more about PHP thread safety.

在扩展中使用全局钩子

有时，可能需要将全局变量初始化为一些默认值，通常为零。引擎帮助下的TSRM系统提供了一个钩子来为您的全局变量提供默认值，我们称之为GINIT。

注意

关于 PHP 挂钩的完整信息，请参考 PHP 生命周期章节。

让我们将随机值设为零：

PHP_GSHUTDOWN_FUNCTION(pib)
{ }

PHP_GINIT_FUNCTION(pib)
{
    pib_globals->rnd = 0;
}

zend_module_entry pib_module_entry = {
    STANDARD_MODULE_HEADER,
    "pib",
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    NULL,
    "0.1",
    PHP_MODULE_GLOBALS(pib),
    PHP_GINIT(pib),
    PHP_GSHUTDOWN(pib),
    NULL, /* PRSHUTDOWN() */
    STANDARD_MODULE_PROPERTIES_EX
};

我们选择仅显示 zend_module_entry（和其他 NULL）的相关部分。如你所见，全局管理挂钩发生在结构的中间。首先是PHP_MODULE_GLOBALS()来确定全局变量的大小，然后是我们的 GINIT和 GSHUTDOWN钩子。然后我们使用了STANDARD_MODULE_PROPERTIES_EX关闭结构，而不是STANDARD_MODULE_PROPERTIES。只需以正确的方式完成结构即可，请参阅？：

#define STANDARD_MODULE_PROPERTIES
    NO_MODULE_GLOBALS, NULL, STANDARD_MODULE_PROPERTIES_EX

在GINIT 函数中，你传递了一个指向全局变量当前存储位置的指针。你可以使用它来初始化全局变量。在这里，我们将零放入随机值（虽然不是很有用，但我们接受它）。

警告

不要在 GINIT 中使用PIB_G()宏。使用你得到的指针。

注意

对于当前进程，在MINIT()之前启动了GINIT()。如果是 NTS，就这样而已。 如果是 ZTS，线程库产生的每个新线程都会额外调用GINIT（）。

警告

GINIT()不作为RINIT()的一部分被调用。如果你需要在每次新请求时清除全局变量，则需要像在本章所示示例中所做的那样手动进行。

完整的例子

这是一个更高级的完整示例。如果玩家获胜，则将其得分（尝试次数）添加到可以从用户区获取的得分数组中。没什么难的，得分数组在请求启动时初始化，然后在玩家获胜时使用，并在当前请求结束时清除：

ZEND_BEGIN_MODULE_GLOBALS(pib)
    zend_long rnd;
    zend_ulong cur_score;
    zval scores;
ZEND_END_MODULE_GLOBALS(pib)

ZEND_DECLARE_MODULE_GLOBALS(pib)

static void pib_rnd_init(void)
{
    /* 重置当前分数 */
    PIB_G(cur_score) = 0;
    php_random_int(0, 100, &PIB_G(rnd), 0);
}

PHP_GINIT_FUNCTION(pib)
{
    /* ZEND_SECURE_ZERO 是 memset(0)。也可以解析为 bzero() */
    ZEND_SECURE_ZERO(pib_globals, sizeof(*pib_globals));
}

ZEND_BEGIN_ARG_INFO_EX(arginfo_guess, 0, 0, 1)
    ZEND_ARG_INFO(0, num)
ZEND_END_ARG_INFO()

PHP_RINIT_FUNCTION(pib)
{
    array_init(&PIB_G(scores));
    pib_rnd_init();

    return SUCCESS;
}

PHP_RSHUTDOWN_FUNCTION(pib)
{
    zval_dtor(&PIB_G(scores));

    return SUCCESS;
}

PHP_FUNCTION(pib_guess)
{
    zend_long r;

    if (zend_parse_parameters(ZEND_NUM_ARGS(), "l", &r) == FAILURE) {
        return;
    }

    if (r == PIB_G(rnd)) {
        add_next_index_long(&PIB_G(scores), PIB_G(cur_score));
        pib_rnd_init();
        RETURN_TRUE;
    }

    PIB_G(cur_score)++;

    if (r < PIB_G(rnd)) {
        RETURN_STRING("more");
    }

    RETURN_STRING("less");
}

PHP_FUNCTION(pib_get_scores)
{
    if (zend_parse_parameters_none() == FAILURE) {
        return;
    }

    RETVAL_ZVAL(&PIB_G(scores), 1, 0);
}

PHP_FUNCTION(pib_reset)
{
    if (zend_parse_parameters_none() == FAILURE) {
        return;
    }

    pib_rnd_init();
}

static const zend_function_entry func[] = {
    PHP_FE(pib_reset, NULL)
    PHP_FE(pib_get_scores, NULL)
    PHP_FE(pib_guess, arginfo_guess)
    PHP_FE_END
};

zend_module_entry pib_module_entry = {
    STANDARD_MODULE_HEADER,
    "pib",
    func, /* 函数入口 */
    NULL, /* 模块初始化 */
    NULL, /* 模块关闭 */
    PHP_RINIT(pib), /* 请求初始化 */
    PHP_RSHUTDOWN(pib), /* 请求关闭 */
    NULL, /* 模块信息 */
    "0.1", /* 替换为扩展的版本号 */
    PHP_MODULE_GLOBALS(pib),
    PHP_GINIT(pib),
    NULL,
    NULL,
    STANDARD_MODULE_PROPERTIES_EX
};

这里必须要注意的是，如果你希望在请求之间保持分数，PHP 不提供任何便利。而是需要一个持久的共享存储，例如文件，数据库，某些内存区域等。PHP 的设计目的不是将信息持久存储在其内部的请求，因此它不提供这么做，但它提供了实用程序来访问请求绑定的全局空间，如我们所示。

然后，很容易地在RINIT()中初始化一个数组，然后在RSHUTDOWN()中销毁它。请记住，array_init创建一个zend_array 并放入一个 zval。但这是免分配的，不要担心分配用户无法使用的数组（因此浪费分配），array_init()非常廉价 (阅读源代码)。

当我们将这样的数组返回给用户时，我们不会忘记增加其引用计数（在 RETVAL_ZVAL中），因为我们在扩展中保留了对此类数组的引用。

使用真实的全局变量

真实全局变量是非线程保护的真实C全局变量。有时可能会需要它们。但是请记住主要规则：在处理请求时，不能安全地写入此类全局变量。因此，通常在 PHP 中，我们需要此类变量并将其用作只读变量。

请记住，在 PHP 生命周期的MINIT()或MSHUTDOWN()步骤中编写真实全局变量是绝对安全的。但是不能在处理请求时给他们写入值（但可以从他们那里读取）。

因此，一个简单的示例是你想要读取环境值以对其进行处理。此外，初始化持久性的 zend_string并在之后处理某些请求时加以利用是很常见的。

这是介绍真实全局变量的修补示例，我们仅显示与先前代码的差异，而不显示完整代码：

static zend_string *more, *less;
static zend_ulong max = 100;

static void register_persistent_string(char *str, zend_string **result)
{
    *result = zend_string_init(str, strlen(str), 1);
    zend_string_hash_val(*result);

    GC_FLAGS(*result) |= IS_INTERNED;
}

static void pib_rnd_init(void)
{
    /* 重置当前分数 */
    PIB_G(cur_score) = 0;
    php_random_int(0, max, &PIB_G(rnd), 0);
}

PHP_MINIT_FUNCTION(pib)
{
    char *pib_max;

    register_persistent_string("more", &more);
    register_persistent_string("less", &less);

    if (pib_max = getenv("PIB_RAND_MAX")) {
        if (!strchr(pib_max, '-')) {
            max = ZEND_STRTOUL(pib_max, NULL, 10);
        }
    }

    return SUCCESS;
}

PHP_MSHUTDOWN_FUNCTION(pib)
{
    zend_string_release(more);
    zend_string_release(less);

    return SUCCESS;
}

PHP_FUNCTION(pib_guess)
{
    zend_long r;

    if (zend_parse_parameters(ZEND_NUM_ARGS(), "l", &r) == FAILURE) {
        return;
    }

    if (r == PIB_G(rnd)) {
        add_next_index_long(&PIB_G(scores), PIB_G(cur_score));
        pib_rnd_init();
        RETURN_TRUE;
    }

    PIB_G(cur_score)++;

    if (r < PIB_G(rnd)) {
        RETURN_STR(more);
    }

    RETURN_STR(less);
}

在这里我们创建了两个 zend_string 变量 more 和 less。这些字符串不需要像以前一样在使用时立即创建和销毁。这些是不可变的字符串，只要保持不变，就可以分配一次并在需要的任何时间重复使用（即只读）。我们在zend_string_init()中使用持久分配，在MINIT()中初始化这两个字符串，我们现在预先计算其哈希值（而不是先执行第一个请求），并且我们告诉 zval 垃圾收集器，这些字符串已被扣留，因此它将永远不会尝试销毁它们（但是，如果将它们用作写操作（例如连接）的一部分，则可能需要复制它们）。显然我们不会忘记在MSHUTDOWN()中销毁这些字符串。

Then in MINIT() we probe for a PIB_RAND_MAX environment and use that as the maximum range value for random number selection. Since we use unsigned integers, and we know that strtoull() won't complain about negative numbers (thus wrapping the integer range as a sign mismatch), we just avoid using negative numbers (classic libc workaround).

The above is the detailed content of PHP manages global status. For more information, please follow other related articles on the PHP Chinese website!