使用GDB调试PHP代码，解决PHP代码死循环

最近在帮同事解决Swoole Server问题时，发现有1个worker进程一直处于R的状态，而且CPU耗时非常高。初步断定是PHP代码中发生死循环。

下面通过一段代码展示如何解决PHP死循环问题。

#dead_loop.php
$array = array();
for($i = 0; $i < 10000; $i++)
{
	$array[] = $i;
}
include __DIR__."/include.php";

#include.php 
while(1)
{
	usleep(10);
	$keys = array_flip($array);
	$index = array_search(rand(1500, 9999), $array);
	$str = str_repeat('A', $index);
	$strb = test($index, $str);
}

function test($index, $str)
{
	return str_replace('A', 'B', $str);
}

通过ps aux得到进程ID和状态如下，使用gdb -p 进程ptrace跟踪，通过bt命令得到调用栈

htf 3834 2.6 0.2 166676 22060 pts/12 R+ 10:50 0:12 php dead_loop.php
gdb -p 3834
 (gdb) bt
#0  0x00000000008cc03f in zend_mm_check_ptr (heap=0x1eaa2c0, ptr=0x2584910, silent=1, __zend_filename=0xee3d40 "/home/htf/workspace/php-5.4.27/Zend/zend_variables.c", 
    __zend_lineno=182, __zend_orig_filename=0xee1888 "/home/htf/workspace/php-5.4.27/Zend/zend_execute_API.c", __zend_orig_lineno=437)
    at /home/htf/workspace/php-5.4.27/Zend/zend_alloc.c:1485
#1  0x00000000008cd643 in _zend_mm_free_int (heap=0x1eaa2c0, p=0x2584910, __zend_filename=0xee3d40 "/home/htf/workspace/php-5.4.27/Zend/zend_variables.c", __zend_lineno=182, 
    __zend_orig_filename=0xee1888 "/home/htf/workspace/php-5.4.27/Zend/zend_execute_API.c", __zend_orig_lineno=437) at /home/htf/workspace/php-5.4.27/Zend/zend_alloc.c:2064
#2  0x00000000008cebf7 in _efree (ptr=0x2584910, __zend_filename=0xee3d40 "/home/htf/workspace/php-5.4.27/Zend/zend_variables.c", __zend_lineno=182, 
    __zend_orig_filename=0xee1888 "/home/htf/workspace/php-5.4.27/Zend/zend_execute_API.c", __zend_orig_lineno=437) at /home/htf/workspace/php-5.4.27/Zend/zend_alloc.c:2436
#3  0x00000000008eda0a in _zval_ptr_dtor (zval_ptr=0x25849a0, __zend_filename=0xee3d40 "/home/htf/workspace/php-5.4.27/Zend/zend_variables.c", __zend_lineno=182)
    at /home/htf/workspace/php-5.4.27/Zend/zend_execute_API.c:437
#4  0x00000000008fe687 in _zval_ptr_dtor_wrapper (zval_ptr=0x25849a0) at /home/htf/workspace/php-5.4.27/Zend/zend_variables.c:182
#5  0x000000000091259f in zend_hash_destroy (ht=0x7f7263f6e380) at /home/htf/workspace/php-5.4.27/Zend/zend_hash.c:560
#6  0x00000000008fe2c5 in _zval_dtor_func (zvalue=0x7f726426fe50, __zend_filename=0xeea290 "/home/htf/workspace/php-5.4.27/Zend/zend_execute.c", __zend_lineno=901)
    at /home/htf/workspace/php-5.4.27/Zend/zend_variables.c:45
#7  0x0000000000936656 in _zval_dtor (zvalue=0x7f726426fe50, __zend_filename=0xeea290 "/home/htf/workspace/php-5.4.27/Zend/zend_execute.c", __zend_lineno=901)
    at /home/htf/workspace/php-5.4.27/Zend/zend_variables.h:35
#8  0x0000000000939747 in zend_assign_to_variable (variable_ptr_ptr=0x7f7263f8e738, value=0x7f726426f6a8) at /home/htf/workspace/php-5.4.27/Zend/zend_execute.c:901
#9  0x0000000000997ee5 in ZEND_ASSIGN_SPEC_CV_VAR_HANDLER (execute_data=0x7f726d04b2a8) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:33168
#10 0x000000000093b5fd in execute (op_array=0x21d58b0) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:410
#11 0x0000000000901692 in zend_execute_scripts (type=8, retval=0x0, file_count=3) at /home/htf/workspace/php-5.4.27/Zend/zend.c:1315
#12 0x000000000087926a in php_execute_script (primary_file=0x7ffffe0038d0) at /home/htf/workspace/php-5.4.27/main/main.c:2502
#13 0x00000000009a32e3 in do_cli (argc=2, argv=0x7ffffe004d18) at /home/htf/workspace/php-5.4.27/sapi/cli/php_cli.c:989
#14 0x00000000009a4491 in main (argc=2, argv=0x7ffffe004d18) at /home/htf/workspace/php-5.4.27/sapi/cli/php_cli.c:1365

执行gdb后，死循环的进程会变成T的状态，表示正在Trace。这个是独占的，所以不能再使用strace/gdb或者其他ptrace工具对此 进程进行调试。另外此进程会中断执行。gdb输入c后，程序继续向下运行。然后再次按下ctrl + c中断程序。 通过bt命令查看进程的调用栈。

 
(gdb) bt
#0  _zend_mm_alloc_int (heap=0x1eaa2c0, size=72, __zend_filename=0xe43410 "/home/htf/workspace/php-5.4.27/ext/standard/array.c", __zend_lineno=2719, 
    __zend_orig_filename=0xee5a38 "/home/htf/workspace/php-5.4.27/Zend/zend_hash.c", __zend_orig_lineno=412) at /home/htf/workspace/php-5.4.27/Zend/zend_alloc.c:1895
#1  0x00000000008ceb86 in _emalloc (size=72, __zend_filename=0xe43410 "/home/htf/workspace/php-5.4.27/ext/standard/array.c", __zend_lineno=2719, 
    __zend_orig_filename=0xee5a38 "/home/htf/workspace/php-5.4.27/Zend/zend_hash.c", __zend_orig_lineno=412) at /home/htf/workspace/php-5.4.27/Zend/zend_alloc.c:2425
#2  0x0000000000911d85 in _zend_hash_index_update_or_next_insert (ht=0x2257a10, h=3972, pData=0x7ffffe0012b0, nDataSize=8, pDest=0x0, flag=1, 
    __zend_filename=0xe43410 "/home/htf/workspace/php-5.4.27/ext/standard/array.c", __zend_lineno=2719) at /home/htf/workspace/php-5.4.27/Zend/zend_hash.c:412
#3  0x00000000007767e1 in zif_array_flip (ht=1, return_value=0x7f726424ea68, return_value_ptr=0x0, this_ptr=0x0, return_value_used=1)
    at /home/htf/workspace/php-5.4.27/ext/standard/array.c:2719
#4  0x000000000093c03e in zend_do_fcall_common_helper_SPEC (execute_data=0x7f726d04b2a8) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:643
#5  0x00000000009400e6 in ZEND_DO_FCALL_SPEC_CONST_HANDLER (execute_data=0x7f726d04b2a8) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:2233
#6  0x000000000093b5fd in execute (op_array=0x21d58b0) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:410

两次的BT信息不一样，这是因为程序在不同的位置中断。看到execute (op_array=0x21d58b0) 这一行，这里就是PHP执行op_array的入口了。gdb下输入f 6，(通过调用栈编号可得)。

(gdb) f 6
#6  0x000000000093b5fd in execute (op_array=0x21d58b0) at /home/htf/workspace/php-5.4.27/Zend/zend_vm_execute.h:410
410			if ((ret = OPLINE->handler(execute_data TSRMLS_CC)) > 0) {
(gdb)  p *op_array
$2 = {type = 2 '02', function_name = 0x7f726d086540 "test", scope = 0x0, fn_flags = 134217728, prototype = 0x0, num_args = 2, required_num_args = 2, arg_info = 0x7f726d086bd8, 
  refcount = 0x7f726d0870f0, opcodes = 0x7f726424d600, last = 8, vars = 0x7f726424e890, last_var = 2, T = 1, brk_cont_array = 0x0, last_brk_cont = 0, try_catch_array = 0x0, 
  last_try_catch = 0, static_variables = 0x0, this_var = 4294967295, filename = 0x7f726424ba38 "/home/htf/wwwroot/include.php", line_start = 12, line_end = 15, doc_comment = 0x0, 
  doc_comment_len = 0, early_binding = 4294967295, literals = 0x7f726424eae0, last_literal = 4, run_time_cache = 0x7f726450bfb0, last_cache_slot = 1, reserved = {0x0, 0x0, 0x0, 0x0}}

这里的filename就能看到op_array是哪个PHP文件的。然后输入f 0进入当前位置。

(gdb) p **executor_globals.opline_ptr
$4 = {handler = 0x93ff9c <zend_do_fcall_spec_const_handler>, op1 = {constant = 1680133296, var = 1680133296, num = 1680133296, hash = 140129283132592, opline_num = 1680133296, 
    jmp_addr = 0x7f726424ccb0, zv = 0x7f726424ccb0, literal = 0x7f726424ccb0, ptr = 0x7f726424ccb0}, op2 = {constant = 0, var = 0, num = 0, hash = 0, opline_num = 0, jmp_addr = 0x0, 
    zv = 0x0, literal = 0x0, ptr = 0x0}, result = {constant = 32, var = 32, num = 32, hash = 32, opline_num = 32, jmp_addr = 0x20, zv = 0x20, literal = 0x20, ptr = 0x20}, 
  extended_value = 1, lineno = 5, opcode = 60 '<', op1_type = 1 '01', op2_type = 8 '', result_type = 4 '04'}
</zend_do_fcall_spec_const_handler>

这里的lineno表示OPCODE所在的代码行数，可以到对应文件里去看下是哪行代码。

使用GDB可以查看到更多的信息，这里就不再一一介绍了，有兴趣各位可以自行尝试。