每天一个 PHP 语法五引用计数与写时复制的实现
说明
这里基于php7.2.5进行测试,php7之后内部结构变化应该不是太大,但与php5.X有差别。
引用计数
我们之前说过存放字符串的结构为zend_stirng, 忘了的看着里
struct _zend_string {
// 这里是引用计数
zend_refcounted_h gc;
zend_ulong h; /* hash value */
size_t len; // 长度
char val[1]; // 内容
};
来看zend_refcounted_h的结构
typedef struct _zend_refcounted_h {
// 我们只关注这里 整型的 引用计数 1/2/3...
uint32_t refcount; /* reference counter 32-bit */
union {
struct {
ZEND_ENDIAN_LOHI_3(
zend_uchar type,
zend_uchar flags, /* used for strings & objects */
uint16_t gc_info) /* keeps GC root number (or 0) and color */
} v;
uint32_t type_info;
} u;
} zend_refcounted_h;
引用计数是什么
因为引用计数存在zend_value的具体类型中的,如zend_string, zend_array等,所以,引用计数是指当前这个zend_value被多少个zval指向。
引用计数如何产生作用
$a = "hi".time();
$b = $a;
$a = time()."hello";
echo $a;
$b = $a;
echo $b;
当赋值$a时,$a指向”hi”.time()所在的zend_value, 所以此时引用计数为1
当$b赋值时,$b也指向了”hi”.time(), 这时引用计数更新为2
$a = time()."hello";
echo $a;
$b = $a;
echo $b;
unset($b);
echo $a;
这里我们通过gdb调试程序可以看到引用计数的变化
gdb /home/php7.2.5/debug/bin/php
// 设置断点 ZEND_ECHO_SPEC_CV_HANDLER 是echo 的处理程序
(gdb) b ZEND_ECHO_SPEC_CV_HANDLER
Breakpoint 1 at 0x973289: file /root/php-7.2.5/Zend/zend_vm_execute.h, line 33086.
// 运行脚本
(gdb) run hello.php
// 第一个echo断点
Breakpoint 1, ZEND_ECHO_SPEC_CV_HANDLER () at /root/php-7.2.5/Zend/zend_vm_execute.h:33086
33086 SAVE_OPLINE();
Missing separate debuginfos, use: debuginfo-install glibc-2.17-222.el7.x86_64 libxml2-2.9.1-6.el7_2.3.x86_64 nss-softokn-freebl-3.36.0-5.el7_5.x86_64 xz-libs-5.2.2-1.el7.x86_64 zlib-1.2.7-17.el7.x86_64
(gdb) n
33087 z = _get_zval_ptr_cv_undef(opline->op1.var EXECUTE_DATA_CC);
(gdb) n
33089 if (Z_TYPE_P(z) == IS_STRING) {
// 打印当前zval, zval.u1.v.type=6说明是字符串
(gdb) p *z
$1 = {value = {lval = 140737318919936, dval = 6.9533474366143666e-310, counted = 0x7ffff5e69f00, str = 0x7ffff5e69f00, arr = 0x7ffff5e69f00, obj = 0x7ffff5e69f00,
res = 0x7ffff5e69f00, ref = 0x7ffff5e69f00, ast = 0x7ffff5e69f00, zv = 0x7ffff5e69f00, ptr = 0x7ffff5e69f00, ce = 0x7ffff5e69f00, func = 0x7ffff5e69f00, ww = {
w1 = 4125531904, w2 = 32767}}, u1 = {v = {type = 6 '\006', type_flags = 20 '\024', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 5126}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p *$1.value.str
// 可以看到 目前的refcount=1
$2 = {gc = {refcount = 1, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
(gdb) p *$2.val@15
$3 = "1587044278hello"
(gdb) n
33090 zend_string *str = Z_STR_P(z);
(gdb) n
33092 if (ZSTR_LEN(str) != 0) {
(gdb) c
Continuing.
1587044278hello
// 到了第二个echo
Breakpoint 1, ZEND_ECHO_SPEC_CV_HANDLER () at /root/php-7.2.5/Zend/zend_vm_execute.h:33086
33086 SAVE_OPLINE();
(gdb) p z
$4 = (zval *) 0x7ffff5e1e090
// 打印当前的zval
(gdb) p *z
$5 = {value = {lval = 140737318919936, dval = 6.9533474366143666e-310, counted = 0x7ffff5e69f00, str = 0x7ffff5e69f00, arr = 0x7ffff5e69f00, obj = 0x7ffff5e69f00,
res = 0x7ffff5e69f00, ref = 0x7ffff5e69f00, ast = 0x7ffff5e69f00, zv = 0x7ffff5e69f00, ptr = 0x7ffff5e69f00, ce = 0x7ffff5e69f00, func = 0x7ffff5e69f00, ww = {
w1 = 4125531904, w2 = 32767}}, u1 = {v = {type = 6 '\006', type_flags = 20 '\024', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 5126}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p *$5.value.str
// 可以看到refount增加为2
$6 = {gc = {refcount = 2, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
可以看到当给$a赋值时,值的引用计数为1,当$a赋值给$b,引用计数再次加1变为2.
// unset 操作的处理
Breakpoint 2, ZEND_UNSET_CV_SPEC_CV_UNUSED_HANDLER () at /root/php-7.2.5/Zend/zend_vm_execute.h:40511
40511 zval *var = EX_VAR(opline->op1.var);
(gdb) n
40513 if (Z_REFCOUNTED_P(var)) {
(gdb) n
40514 zend_refcounted *garbage = Z_COUNTED_P(var);
(gdb) n
40516 ZVAL_UNDEF(var);
(gdb) n
40517 SAVE_OPLINE();
(gdb) n
// --GC_REFCOUNT 引用计数减一
40518 if (!--GC_REFCOUNT(garbage)) {
(gdb) n
40521 gc_check_possible_root(garbage);
(gdb) n
40523 ZEND_VM_NEXT_OPCODE_CHECK_EXCEPTION();
(gdb) n
40528 }
(gdb) p *var
$5 = {value = {lval = 140737318919936, dval = 6.9533474366143666e-310, counted = 0x7ffff5e69f00, str = 0x7ffff5e69f00, arr = 0x7ffff5e69f00, obj = 0x7ffff5e69f00,
res = 0x7ffff5e69f00, ref = 0x7ffff5e69f00, ast = 0x7ffff5e69f00, zv = 0x7ffff5e69f00, ptr = 0x7ffff5e69f00, ce = 0x7ffff5e69f00, func = 0x7ffff5e69f00, ww = {
w1 = 4125531904, w2 = 32767}}, u1 = {v = {type = 0 '\000', type_flags = 0 '\000', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 0}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p *$5.value.str
// 再次打印value的ref
$6 = {gc = {refcount = 1, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
(gdb)
也就是在执行unset($b)之后引用计数减一
写时复制
意思就是变量发生变化时再复制一份。
如上所示$b=$a时,并没有把$a复制一份给$b,而是$b、$a指向同一个zend_value, 并更新zend_value的refcount,这样是节省内存的。
那如果接着操作$b重新赋值会发生什么呢
$a = time()."hello";
echo $a;
$b = $a;
$b = "hi".time();
echo $b;
看调试结果
(gdb) p *z
$1 = {value = {lval = 140737318919856, dval = 6.9533474366104141e-310, counted = 0x7ffff5e69eb0, str = 0x7ffff5e69eb0, arr = 0x7ffff5e69eb0, obj = 0x7ffff5e69eb0,
res = 0x7ffff5e69eb0, ref = 0x7ffff5e69eb0, ast = 0x7ffff5e69eb0, zv = 0x7ffff5e69eb0, ptr = 0x7ffff5e69eb0, ce = 0x7ffff5e69eb0, func = 0x7ffff5e69eb0, ww = {
w1 = 4125531824, w2 = 32767}}, u1 = {v = {type = 6 '\006', type_flags = 20 '\024', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 5126}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p z
$2 = (zval *) 0x7ffff5e1e080
(gdb) p *$1.value.str
// $a = time()."hello" 之后 refcount=1
$3 = {gc = {refcount = 1, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
(gdb) p *$3.val@15
$4 = "1587215850hello"
......
(gdb) c
Continuing.
1587215850hello
// $b = $a 操作
Breakpoint 2, ZEND_ASSIGN_SPEC_CV_CV_RETVAL_UNUSED_HANDLER () at /root/php-7.2.5/Zend/zend_vm_execute.h:43779
43779 SAVE_OPLINE();
(gdb) n
43780 value = _get_zval_ptr_cv_BP_VAR_R(opline->op2.var EXECUTE_DATA_CC);
(gdb) n
43781 variable_ptr = _get_zval_ptr_cv_undef_BP_VAR_W(opline->op1.var EXECUTE_DATA_CC);
(gdb) n
43789 value = zend_assign_to_variable(variable_ptr, value, IS_CV);
(gdb) n
43797 ZEND_VM_NEXT_OPCODE_CHECK_EXCEPTION();
(gdb) n
43798 }
(gdb) p *value
$5 = {value = {lval = 140737318919856, dval = 6.9533474366104141e-310, counted = 0x7ffff5e69eb0, str = 0x7ffff5e69eb0, arr = 0x7ffff5e69eb0, obj = 0x7ffff5e69eb0,
res = 0x7ffff5e69eb0, ref = 0x7ffff5e69eb0, ast = 0x7ffff5e69eb0, zv = 0x7ffff5e69eb0, ptr = 0x7ffff5e69eb0, ce = 0x7ffff5e69eb0, func = 0x7ffff5e69eb0, ww = {
w1 = 4125531824, w2 = 32767}}, u1 = {v = {type = 6 '\006', type_flags = 20 '\024', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 5126}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p *$5.value.str
// $b = $a之后 refcount=2
$6 = {gc = {refcount = 2, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
....
// $b = "hi".time()
Breakpoint 1, ZEND_ECHO_SPEC_CV_HANDLER () at /root/php-7.2.5/Zend/zend_vm_execute.h:33086
33086 SAVE_OPLINE();
(gdb) n
33087 z = _get_zval_ptr_cv_undef(opline->op1.var EXECUTE_DATA_CC);
(gdb) n
33089 if (Z_TYPE_P(z) == IS_STRING) {
(gdb) p *z
// 注意str的地址 跟$a不是一个了
$7 = {value = {lval = 140737318919936, dval = 6.9533474366143666e-310, counted = 0x7ffff5e69f00, str = 0x7ffff5e69f00, arr = 0x7ffff5e69f00, obj = 0x7ffff5e69f00,
res = 0x7ffff5e69f00, ref = 0x7ffff5e69f00, ast = 0x7ffff5e69f00, zv = 0x7ffff5e69f00, ptr = 0x7ffff5e69f00, ce = 0x7ffff5e69f00, func = 0x7ffff5e69f00, ww = {
w1 = 4125531904, w2 = 32767}}, u1 = {v = {type = 6 '\006', type_flags = 20 '\024', const_flags = 0 '\000', reserved = 0 '\000'}, type_info = 5126}, u2 = {next = 0,
cache_slot = 0, lineno = 0, num_args = 0, fe_pos = 0, fe_iter_idx = 0, access_flags = 0, property_guard = 0, extra = 0}}
(gdb) p *$7.value.str
// 新值的refcount=1
$8 = {gc = {refcount = 1, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 12, val = "h"}
(gdb) p *$8.val@12
$12 = "hi1587216014"
// 再来看 原值的refcount 变为了1
(gdb) p *$1.value.str
$11 = {gc = {refcount = 1, u = {v = {type = 6 '\006', flags = 0 '\000', gc_info = 0}, type_info = 6}}, h = 0, len = 15, val = "1"}
总结
引用计数与写时复制是PHP自动垃圾回收的基础。
通过对zval的引用计数的变化监测判断是否可以回收变量,而写时复制则会节省变量所占内存。
注意事项:
gdb调试时,PHP必须要打开debug模式,在编译的时候加上–enable-debug即可
参考资料:
《PHP7内核剖析》
本作品采用《CC 协议》,转载必须注明作者和本文链接