Usage of get_fs and set_fs

  最近在看 drbd的源代码，内核模块 socket通信中有下面一段程序：

1.  mm_segment_t oldfs;
2.     struct kvec iov = {
3.         .iov_base = buf,
4.         .iov_len = size,
5.     };
6.     struct msghdr msg = {
7.         .msg_iovlen = 1,
8.         .msg_iov = (struct iovec *)&iov,
9.         .msg_flags = (flags ? flags : MSG_WAITALL | MSG_NOSIGNAL)
10.     };
11.     int rv;
13.     oldfs = get_fs();
14.     set_fs(KERNEL_DS);
15.     rv = sock_recvmsg(sock, &msg, size, msg.msg_flags);
16.     set_fs(oldfs);

再看看内核中的定义：

1. #define KERNEL_DS       MAKE_MM_SEG(0xFFFFFFFFUL)
2. #define USER_DS         MAKE_MM_SEG(PAGE_OFFSET)
3. #define get_ds()        (KERNEL_DS)
4. #define get_fs()        (current_thread_info()->addr_limit)
5. #define set_fs(x)       (current_thread_info()->addr_limit = (x))

其作用是用来修改当前进程的内存访问权限。

      When you make a system call from user space, the first thing that is checked is if the address of the parameter is well within the legal virtual address space (i.e. 0 to 3 GB for the user space). If this is not so, the call will fail. If you want to make the same system call from the Kernel Space( Virtual Address 3 - 4 GB) however, this address checking has to be avoided so that the call will not fail. Now, every process has a tak_struct associated with it and this structure contains the legal virtual address boundaries for that process( Virtual Address space represented by mm_segment_t). The get_fs() macro will retrieve this boundary and the set_fs() will set it with a value. So, when you want to access a memory region which is beyond the User Space Virtual Address limit( i.e. falling in the Kernel Space Virtual Address region)。
set KERNEL_DS when user space address is used