typeof, offsetof 和container_of

要理解Linux中实现的双向循环链表（"侵入式"链表），首先得弄明白宏container_of。 本文尝试从gcc的关键字typeof和宏offsetof入手，循序渐进地剖析宏container_of之实现原理。

1. typeof (from: https://en.wikipedia.org/wiki/Typeof)

typeof is an operator provided by several programming languages to
determine the data type of a variable. This is useful when
constructing programs that must accept multiple types of data
without explicitly specifying the type.

The GNU compiler (GCC) extensions for the C programming language provide
typeof:

#define max(a, b)        ({ typeof (a) _a = (a);           typeof (b) _b = (b);          _a > _b ? _a : _b; })

typeof和sizeof一样，都是关键字。只不过typeof不是标准的c语言关键字，而是gcc的c语言的扩展关键字。 typeof的作用是取得某个变量的数据类型。例如：

unsigned int a = 1; // typeof (a) is unsigned int
short b = 2;        // typeof (b) is short

2. offsetof (from: include/linux/stddef.h)

1 #define offsetof(TYPE, MEMBER)  ((size_t)&((TYPE *)0)->MEMBER)

/*  * @TYPE: The type of the structure  * @MEMBER: The member within the structure to get the offset of  */

宏offsetof的作用是获取某个成员变量(MEMBER)在其所在的结构体(TYPE)里的偏移。 上面的宏实现得非常巧妙（如果对汇编熟悉就更容易理解），剖析如下：

(1) X = (TYPE *)0  // 将地址0x0强制转化为结构体类型TYPE的首地址
(2) Y = X->MEMBER  // 访问成员变量MEMBER
(3) Z = &Y         // 取得成员变量MEMBER的内存地址
(4) N = (size_t)Z  // 将成员变量MEMBER的内存地址强制转换成偏移量，
      = (size_t)&Y // 由于内存地址也是整数，所以成员变量MEMBER相对于结构体首地址的偏移等于&Y
      = (size_t)&(X->MEMBER) = (size_t)&X->MEMBER  // 注意： -> 比 & 优先级高
      = (size_t)&(X)->MEMBER = (size_t)&((TYPE *)0)->MEMBER

3. container_of (from: include/linux/kernel.h)

 1 /**
 2  * container_of - cast a member of a structure out to the containing structure
 3  * @ptr:        the pointer to the member.
 4  * @type:       the type of the container struct this is embedded in.
 5  * @member:     the name of the member within the struct.
 6  *
 7  */
 8 #define container_of(ptr, type, member) ({                       9         const typeof( ((type *)0)->member ) *__mptr = (ptr);    10         (type *)( (char *)__mptr - offsetof(type,member) );})

• L9:  用一个临时变量__mptr保存成员变量的指针ptr
• L10: offsetof(type, member): 计算出成员变量相对于其所在的结构体的偏移，不妨记为OFFSET
• L10: __mptr - OFFSET, 就是成员变量所在的结构体的首地址

因此， 宏container_of的作用就是根据某个成员变量的内存地址，反推出其所在的结构体变量的首地址。

示例代码： foo.c

 1 #include <stdio.h>
 2
 3 #define offsetof(TYPE, MEMBER)  ((size_t)&(((TYPE *)0)->MEMBER))
 4
 5 #define container_of(ptr, type, member) ({                               6                 const typeof( ((type *)0)->member ) *__mptr = (ptr);     7                 (type *)( (char *)__mptr - offsetof(type,member) );})
 8
 9 typedef struct foo_s {
10         int             m_int;
11         short           m_short;
12         char            m_char;
13         long long       m_longlong;
14 } foo_t;
15
16 int
17 main(int argc, char *argv[])
18 {
19         foo_t ox = {0x12345678, 0x1234, ‘A‘, 0xfedcba9876543210};
20         char *p3 = &ox.m_char;
21         foo_t *p = container_of(p3, foo_t, m_char);
22
23         printf("foo_t ox (%p) sizeof(foo_t) = %d\n", &ox, sizeof (foo_t));
24         printf("foo_t *p (%p) p3(%p)\n", p, p3);
25         printf("foo_t->m_int      = %#x\t\t(%d)(%p)\n",
26             p->m_int, offsetof(foo_t, m_int), &ox.m_int);
27         printf("foo_t->m_short    = %#x\t\t(%d)(%p)\n",
28             p->m_short, offsetof(foo_t, m_short), &ox.m_short);
29         printf("foo_t->m_char     = %c\t\t\t(%d)(%p)\n",
30             p->m_char, offsetof(foo_t, m_char), &ox.m_char);
31         printf("foo_t->m_longlong = %#llx\t(%d)(%p)\n",
32             p->m_longlong, offsetof(foo_t, m_longlong), &ox.m_longlong);
33
34         return 0;
35 }

编译并运行

$ gcc -g -Wall -m32 -o foo foo.c
$ ./foo
foo_t ox (0xbfdfe990) sizeof(foo_t) = 16
foo_t *p (0xbfdfe990) p3(0xbfdfe996)
foo_t->m_int      = 0x12345678          (0)(0xbfdfe990)
foo_t->m_short    = 0x1234              (4)(0xbfdfe994)
foo_t->m_char     = A                   (6)(0xbfdfe996)
foo_t->m_longlong = 0xfedcba9876543210  (8)(0xbfdfe998)

反汇编并结合gcc -E foo.c

(gdb) set disassembly-flavor intel
(gdb) disas /m main
Dump of assembler code for function main:
18      {
   0x0804841d <+0>:     push   ebp
   0x0804841e <+1>:     mov    ebp,esp
   0x08048420 <+3>:     and    esp,0xfffffff0
   0x08048423 <+6>:     sub    esp,0x40

19              foo_t ox = {0x12345678, 0x1234, ‘A‘, 0xfedcba9876543210};
   0x08048426 <+9>:     mov    DWORD PTR [esp+0x30],0x12345678
   0x0804842e <+17>:    mov    WORD PTR [esp+0x34],0x1234
   0x08048435 <+24>:    mov    BYTE PTR [esp+0x36],0x41
   0x0804843a <+29>:    mov    DWORD PTR [esp+0x38],0x76543210
   0x08048442 <+37>:    mov    DWORD PTR [esp+0x3c],0xfedcba98

20              char *p3 = &ox.m_char;
   0x0804844a <+45>:    lea    eax,[esp+0x30]
   0x0804844e <+49>:    add    eax,0x6
   0x08048451 <+52>:    mov    DWORD PTR [esp+0x24],eax

21              foo_t *p = container_of(p3, foo_t, m_char);
   0x08048455 <+56>:    mov    eax,DWORD PTR [esp+0x24]
   0x08048459 <+60>:    mov    DWORD PTR [esp+0x28],eax
   0x0804845d <+64>:    mov    eax,DWORD PTR [esp+0x28]
   0x08048461 <+68>:    sub    eax,0x6
   0x08048464 <+71>:    mov    DWORD PTR [esp+0x2c],eax
#
# --- L21‘s output from "gcc -E foo.c" ---
# 001               foo_t *p = ({
# 002    const typeof( ((foo_t *)0)->m_char ) *__mptr = (p3);
# 003    (foo_t *)( (char *)__mptr - ((size_t)&(((foo_t *)0)->m_char)) );
# 004               });
#

分析(TBD)

小结(TBD)