linux 的 ping 原理
ping命令的工作原理是:
ping命令是用来查看网络上另一个主机系统的网络连接是否正常的一个工具。
他向网络上的另一个主机系统发送ICMP报文,如果指定系统得到了报文,它将把报文原样传回给发送者,这有点像潜水艇的声纳系统中使用的发声装置。
例如,在Linux终端上执行ping localhost命令将会看到以下结果:
[root@Oracle9i work]# ping localhost
PING Oracle9i (127.0.0.1) 56(84) bytes of data.
64 bytes from Oracle9i (127.0.0.1): icmp_seq=0 ttl=64 time=4.04 ms
64 bytes from Oracle9i (127.0.0.1): icmp_seq=1 ttl=64 time=0.137 ms
64 bytes from Oracle9i (127.0.0.1): icmp_seq=2 ttl=64 time=0.055 ms
64 bytes from Oracle9i (127.0.0.1): icmp_seq=3 ttl=64 time=0.051 ms
--- Oracle9i ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3001ms
rtt min/avg/max/mdev = 0.051/1.070/4.040/1.715 ms, pipe 2
[root@Oracle9i work]#
由上面的执行结果可以看到,ping命令执行后显示出:
1>. 被测试系统主机名和相应IP地址
2>. 返回给当前主机的ICMP报文顺序号(icmp_seq)
3>. ttl生存时间
4>. 往返时间rtt(time, 单位是毫秒,即千分之一秒)。
要写一个模拟ping命令,这些信息有启示作用。
要真正了解ping命令实现原理,就要了解ping命令所使用到的TCP/IP协议。
ICMP(Internet Control Message, 网际控制报文协议)是为网关和目标主机而提供的一种差错控制机制,使它们在遇到差错时能把错误报告给报文源发方。
ICMP协议是IP层的一个协议,但是由于差错报告在发送给报文源发方时可能也要经过若干子网,因此牵涉到路由选择等问题,所以ICMP报文需通过IP协议来发送。
ICMP数据报的数据在发送前需要两级封装:
首先添加ICMP报头形成ICMP报文,再添加IP报头形成IP数据报。如下图所示:
<图1_ICMP数据报的数据在发送前需要两级封装.jpg>
IP报头格式
由于IP层协议是一种点对点的协议,而非端对端的协议,它提供无连接的数据报服务,没有端口的概念,因此很少使用bind()和connect()函数,若有使用也只是用于设置IP地址。
发送数据使用sendto()函数,接收数据使用recvfrom()函数。IP报头格式如下图:
<图2_IP报头格式.jpg>
在Linux中,IP报头格式数据结构(</usr/include/netinet/ip.h>)定义如下:
struct ip
{
#if __BYTE_ORDER == __LITTLE_ENDIAN
unsigned int ip_hl:4; /* header length */
unsigned int ip_v:4; /* version */
#endif
#if __BYTE_ORDER == __BIG_ENDIAN
unsigned int ip_v:4; /* version */
unsigned int ip_hl:4; /* header length */
#endif
u_int8_t ip_tos; /* type of service */
u_short ip_len; /* total length */
u_short ip_id; /* identification */
u_short ip_off; /* fragment offset field */
#define IP_RF 0x8000 /* reserved fragment flag */
#define IP_DF 0x4000 /* dont fragment flag */
#define IP_MF 0x2000 /* more fragments flag */
#define IP_OFFMASK 0x1fff /* mask for fragmenting bits */
u_int8_t ip_ttl; /* time to live */
u_int8_t ip_p; /* protocol */
u_short ip_sum; /* checksum */
struct in_addr ip_src, ip_dst; /* source and dest address */
};
其中ping程序只使用以下数据:
a. IP报头长度IHL(Internet Header Length)
以4字节为一个单位来记录IP报头的长度,是上述IP数据结构的ip_hl变量。
b. 生存时间TTL(Time To Live)
以秒为单位,指出IP数据报能在网络上停留的最长时间,其值由发送方设定,并在经过路由的每一个节点时减一。
当该值为0时,数据报将被丢弃,TTL是上述IP数据结构的ip_ttl变量。
ICMP报头格式
ICMP报文分为两种,一是错误报告报文,二是查询报文。
每个ICMP报头均包含类型、编码和校验和这三项内容,长度为8位,8位和16位,其余选项则随ICMP的功能不同而不同。
Ping命令只使用众多ICMP报文中的两种:"请求回送'(ICMP_ECHO)和"请求回应'(ICMP_ECHOREPLY)。在Linux中定义如下:
#define ICMP_ECHO 0
#define ICMP_ECHOREPLY 8
这两种ICMP类型报头格式如下:
<图3_ICMP类型报头格式.jpg>
在Linux中ICMP数据结构(</usr/include/netinet/ip_icmp.h>)定义如下:
struct icmp
{
u_int8_t icmp_type; /* type of message, see below */
u_int8_t icmp_code; /* type sub code */
u_int16_t icmp_cksum; /* ones complement checksum of struct */
union
{
u_char ih_pptr; /* ICMP_PARAMPROB */
struct in_addr ih_gwaddr; /* gateway address */
struct ih_idseq /* echo datagram */
{
u_int16_t icd_id;
u_int16_t icd_seq;
} ih_idseq;
u_int32_t ih_void;
/* ICMP_UNREACH_NEEDFRAG -- Path MTU Discovery (RFC1191) */
struct ih_pmtu
{
u_int16_t ipm_void;
u_int16_t ipm_nextmtu;
} ih_pmtu;
struct ih_rtradv
{
u_int8_t irt_num_addrs;
u_int8_t irt_wpa;
u_int16_t irt_lifetime;
} ih_rtradv;
} icmp_hun;
#define icmp_pptr icmp_hun.ih_pptr
#define icmp_gwaddr icmp_hun.ih_gwaddr
#define icmp_id icmp_hun.ih_idseq.icd_id
#define icmp_seq icmp_hun.ih_idseq.icd_seq
#define icmp_void icmp_hun.ih_void
#define icmp_pmvoid icmp_hun.ih_pmtu.ipm_void
#define icmp_nextmtu icmp_hun.ih_pmtu.ipm_nextmtu
#define icmp_num_addrs icmp_hun.ih_rtradv.irt_num_addrs
#define icmp_wpa icmp_hun.ih_rtradv.irt_wpa
#define icmp_lifetime icmp_hun.ih_rtradv.irt_lifetime
union
{
struct
{
u_int32_t its_otime;
u_int32_t its_rtime;
u_int32_t its_ttime;
} id_ts;
struct
{
struct ip idi_ip;
/* options and then 64 bits of data */
} id_ip;
struct icmp_ra_addr id_radv;
u_int32_t id_mask;
u_int8_t id_data[1];
} icmp_dun;
#define icmp_otime icmp_dun.id_ts.its_otime
#define icmp_rtime icmp_dun.id_ts.its_rtime
#define icmp_ttime icmp_dun.id_ts.its_ttime
#define icmp_ip icmp_dun.id_ip.idi_ip
#define icmp_radv icmp_dun.id_radv
#define icmp_mask icmp_dun.id_mask
#define icmp_data icmp_dun.id_data
};
使用宏定义令表达更简洁, 其中ICMP报头为8字节, 数据报长度最大为64K字节。
1. 校验和算法
这一算法称为网际校验和算法,把被校验的数据16位进行累加,然后取反码,若数据字节长度为奇数,则数据尾部补一个字节的0以凑成偶数。
此算法适用于IPv4、ICMPv4、IGMPV4、ICMPv6、UDP和TCP校验和,更详细的信息请参考RFC1071,校验和字段为上述ICMP数据结构的icmp_cksum变量。
2. 标识符
用于唯一标识ICMP报文, 为上述ICMP数据结构的icmp_id宏所指的变量。
3. 顺序号
ping命令的icmp_seq便由这里读出,代表ICMP报文的发送顺序,为上述ICMP数据结构的icmp_seq宏所指的变量。
ICMP数据报
Ping命令中需要显示的信息,包括icmp_seq和ttl都已有实现的办法,但还缺rtt往返时间。
为了实现这一功能,可利用ICMP数据报携带一个时间戳。使用以下函数生成时间戳:
#include
int gettimeofday(struct timeval *tp, void *tzp);
其中timeval结构如下:
struct timeval
{
long tv_sec;
long tv_usec;
};
其中tv_sec为秒数,tv_usec微秒数。在发送和接收报文时由gettimeofday分别生成两个timeval结构,两者之差即为往返时间,即ICMP报文发送与接收的时间差,
而timeval结构由ICMP数据报携带,tzp指针表示时区,一般都不使用,赋NULL值。
数据统计
系统自带的ping命令当它接送完所有ICMP报文后,会对所有发送和所有接收的ICMP报文进行统计,从而计算ICMP报文丢失的比率。
为达此目的,定义两个全局变量:接收计数器和发送计数器,用于记录ICMP报文接受和发送数目。丢失数目=发送总数-接收总数,丢失比率=丢失数目/发送总数。
现给出模拟Ping程序功能的代码如下:
/***********************************************************
* 作者:梁俊辉 *
* 时间:2001年10月 *
* 名称:myping.c *
* 说明:本程序用于演示ping命令的实现原理 *
***********************************************************/
#include <stdio.h>
#include <signal.h>
#include <arpa/inet.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <unistd.h>
#include <netinet/in.h>
#include <netinet/ip_icmp.h>
#include <netdb.h>
#include <setjmp.h>
#include <errno.h>
#define PACKET_SIZE 4096
#define MAX_WAIT_TIME 5
#define MAX_NO_PACKETS 3
char sendpacket[PACKET_SIZE];
char recvpacket[PACKET_SIZE];
int sockfd, datalen=56;
int nsend=0, nreceived=0;
struct sockaddr_in dest_addr;/*结构体sockaddr_in在*/
pid_t pid;
struct sockaddr_in from;
struct timeval tvrecv;
void statistics(int signo);
unsigned short cal_chksum(unsigned short *addr, int len);
int pack(int pack_no);
void send_packet(void);
void recv_packet(void);
int unpack(char *buf, int len);
void tv_sub(struct timeval *out, struct timeval *in);
void statistics(int signo)
{
printf("\n--------------------PING statistics-------------------\n");
printf("%d packets transmitted, %d received , %%%d lost\n", nsend, nreceived, (nsend-nreceived)/nsend*100);
close(sockfd);
exit(1);
}
/*校验和算法*/
unsigned short cal_chksum(unsigned short *addr, int len)
{
int nleft=len;
int sum=0;
unsigned short *w=addr;
unsigned short answer=0;
/*把ICMP报头二进制数据以2字节为单位累加起来*/
while(nleft>1)
{
sum+=*w++;
nleft-=2;
}
/*若ICMP报头为奇数个字节,会剩下最后一字节。把最后一个字节视为一个2字节数据的高字节,这个2字节数据的低字节为0,继续累加*/
if( nleft==1)
{
*(unsigned char *)(&answer)=*(unsigned char *)w;
sum+=answer;
}
sum=(sum>>16)+(sum&0xffff);
sum+=(sum>>16);
answer=~sum;
return answer;
}
/*设置ICMP报头*/
int pack(int pack_no)
{
int i, packsize;
struct icmp *icmp;
struct timeval *tval;
icmp=(struct icmp*)sendpacket;
icmp->icmp_type=ICMP_ECHO;
icmp->icmp_code=0;
icmp->icmp_cksum=0;
icmp->icmp_seq=pack_no;
icmp->icmp_id=pid;
packsize=8+datalen;
tval= (struct timeval *)icmp->icmp_data;
gettimeofday(tval, NULL); /*记录发送时间*/
icmp->icmp_cksum=cal_chksum((unsigned short *)icmp, packsize); /*校验算法*/
return packsize;
}
/*发送三个ICMP报文*/
void send_packet()
{
int packetsize;
while( nsend<MAX_NO_PACKETS)
{
nsend++;
packetsize=pack(nsend); /*设置ICMP报头*/
if( sendto(sockfd, sendpacket, packetsize, 0, (struct sockaddr *)&dest_addr, sizeof(dest_addr) )<0 )
{
perror("sendto error");
continue;
}
sleep(1); /*每隔一秒发送一个ICMP报文*/
}
}
/*接收所有ICMP报文*/
void recv_packet()
{
int n, fromlen;
extern int errno;
signal(SIGALRM, statistics);
fromlen=sizeof(from);
while( nreceived<nsend)
{
alarm(MAX_WAIT_TIME);
if( (n=recvfrom(sockfd, recvpacket, sizeof(recvpacket), 0, (struct sockaddr *)&from, &fromlen)) <0)
{
if(errno==EINTR)continue;
perror("recvfrom error");
continue;
}
gettimeofday(&tvrecv, NULL); /*记录接收时间*/
if(unpack(recvpacket, n)==-1)
{
continue;
}
nreceived++;
}
}
/*剥去ICMP报头*/
int unpack(char *buf, int len)
{
int i, iphdrlen;
struct ip *ip;
struct icmp *icmp;
struct timeval *tvsend;
double rtt;
ip=(struct ip *)buf;
iphdrlen=ip->ip_hl<<2; /*求ip报头长度,即ip报头的长度标志乘4*/
icmp=(struct icmp *)(buf+iphdrlen); /*越过ip报头,指向ICMP报头*/
len-=iphdrlen; /*ICMP报头及ICMP数据报的总长度*/
if( len<8 ) /*小于ICMP报头长度则不合理*/
{
printf("ICMP packets\'s length is less than 8\n");
return -1;
}
/*确保所接收的是我所发的的ICMP的回应*/
if( (icmp->icmp_type==ICMP_ECHOREPLY) && (icmp->icmp_id==pid) )
{
tvsend=(struct timeval *)icmp->icmp_data;
tv_sub(&tvrecv, tvsend); /*接收和发送的时间差*/
rtt=tvrecv.tv_sec*1000+tvrecv.tv_usec/1000; /*以毫秒为单位计算rtt*/
/*显示相关信息*/
printf("%d byte from %s: icmp_seq=%u ttl=%d rtt=%.3f ms\n",
len,
inet_ntoa(from.sin_addr),
icmp->icmp_seq,
ip->ip_ttl,
rtt);
}
else return -1;
}
/*两个timeval结构相减*/
void tv_sub(struct timeval *out, struct timeval *in)
{
if( (out->tv_usec-=in->tv_usec)<0)
{
--out->tv_sec;
out->tv_usec+=1000000;
}
out->tv_sec-=in->tv_sec;
}
int main(int argc, char *argv[])
{
struct hostent *host;/*结构体hostent在/usr/include/netdb.h中定义,前面已经#include <netdb.h>*/
struct protoent *protocol; /*结构体protoent在/usr/include/netdb.h中定义,前面已经#include <netdb.h>*/
unsigned long inaddr=0l;/*0l中的l什么意思?*/
int waittime=MAX_WAIT_TIME;/*宏定义#define MAX_WAIT_TIME 5*/
int size=50*1024;
if( argc<2 )
{
printf("usage:%s hostname/IP address\n", argv[0]);
exit(1);
}
printf("here is ok");/*调试*/
sleep(3); /*调试*/
if( (protocol=getprotobyname("icmp") )==NULL)/*函数getprotobyname在netdb.h定义,struct protoent *getprotobyname(const char *name); 可man查看*/
{
perror("getprotobyname error");
exit(1);
}
/*生成使用ICMP的原始套接字,这种套接字只有root才能生成*/
if((sockfd=socket(AF_INET, SOCK_RAW, protocol->p_proto))<0) /*函数socket()在<sys/socket.h>中定义,int socket(int domain, int type, int protocol);*/
{
perror("socket error");
exit(1);
}
printf("here is ok");/*调试*/
sleep(3); /*调试*/
/* 回收root权限,设置当前用户权限*/
setuid(getuid());
/*扩大套接字接收缓冲区到50K这样做主要为了减小接收缓冲区溢出的可能性,若无意中ping一个广播地址或多播地址,将会引来大量应答*/
setsockopt(sockfd, SOL_SOCKET, SO_RCVBUF, &size, sizeof(size));
bzero(&dest_addr, sizeof(dest_addr));
dest_addr.sin_family=AF_INET;
/*判断是主机名还是ip地址*/
if( (inaddr=inet_addr(argv[1]))==INADDR_NONE ) /*函数inet_addr在<arpa/inet.h>中定义,in_addr_t inet_addr(const char *cp);*/
{
if( (host=gethostbyname(argv[1]))==NULL ) /*是主机名*/
{
perror("gethostbyname error");
exit(1);
}
memcpy((char *)&dest_addr.sin_addr, host->h_addr, host->h_length);
}
else /*是ip地址*/
memcpy((char *)&dest_addr, (char *)&inaddr, host->h_length);
/*获取main的进程id,用于设置ICMP的标志符*/
pid=getpid();
sleep(3); /*调试*/
printf("here is ok");/*调试*/
printf("PING %s(%s): %d bytes data in ICMP packets.\n", argv[1], inet_ntoa(dest_addr.sin_addr), datalen); /*全局定义int sockfd, datalen=56;*/
send_packet(); /*发送所有ICMP报文*/
recv_packet(); /*接收所有ICMP报文*/
statistics(SIGALRM); /*进行统计*/
return 0;
}
/*------------- The End -----------*/
特别注意:
只有root用户才能利用socket()函数生成原始套接字,要让Linux的一般用户能执行以上程序,需进行如下的特别操作:
用root登陆,编译以上程序:gcc -o myping myping.c,其目的有二:一是编译,二是让myping属于root用户。
再执行chmod u+s myping,目的是把myping程序设成SUID的属性。
退出root,用一般用户登陆,执行./myping www.cn.ibm.com,有以下执行结果:
PING www.cn.ibm.com(202.95.2.148): 56 bytes data in ICMP packets.
64 byte from 202.95.2.148: icmp_seq=1 ttl=242 rtt=3029.000 ms
64 byte from 202.95.2.148: icmp_seq=2 ttl=242 rtt=2020.000 ms
64 byte from 202.95.2.148: icmp_seq=3 ttl=242 rtt=1010.000 ms
--------------------PING statistics-------------------
3 packets transmitted, 3 received , %0 lost
由于myping.c是发送完所有的ICMP报文才去接收,因此第一、第二和第三个ICMP报文的往返时间依此是3秒,2秒,1秒,上述结果中rtt信息正反映这一事实。
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <signal.h>
#include <arpa/inet.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <unistd.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/ip_icmp.h>
#include <netdb.h>
#include <setjmp.h>
#include <errno.h>
#include <sys/time.h>
#define PACKET_SIZE 4096
#define MAX_WAIT_TIME 5
#define MAX_NO_PACKETS 3
#define _DEBUG_ 1
char sendpacket[PACKET_SIZE];
char recvpacket[PACKET_SIZE];
int sockfd, datalen = 8;
int nsend = 0, nreceived = 0;
struct sockaddr_in dest_addr;
pid_t pid;
struct sockaddr_in from;
struct timeval tvrecv;
void statistics(int signo);
unsigned short cal_chksum(unsigned short *addr,int len);
int pack(int pack_no);
void send_packet(void);
void recv_packet(void);
int unpack(char *buf,int len);
void tv_sub(struct timeval *out,struct timeval *in);
void statistics(int signo)
{
printf("\n--------------------PING statistics-------------------\n");
printf("%d packets transmitted, %d received , %%%d lost\n", nsend, nreceived, (nsend-nreceived)/nsend*100);
close(sockfd);
printf("catch signal\n");
exit(1);
}
unsigned short cal_chksum(unsigned short *addr,int len)
{
int nleft = len;
int sum = 0;
unsigned short *w = addr;
unsigned short answer = 0;
while (nleft > 1)
{
sum += *w++;
nleft -= 2;
}
if (nleft == 1)
{
/
int pack(int pack_no)
{
int packsize;
struct icmp *icmp;
struct timeval *tval;
icmp = (struct icmp*)sendpacket;
icmp->icmp_type = ICMP_ECHO;
icmp->icmp_code = 0;
icmp->icmp_cksum = 0;
icmp->icmp_seq = pack_no;
icmp->icmp_id = pid;
packsize = 8 + datalen;
tval = (struct timeval *)icmp->icmp_data;
gettimeofday(tval, NULL);
#if _DEBUG_
//printf("sec = %ld, usec = %ld\n", tval->tv_sec, tval->tv_usec);
#endif
icmp->icmp_cksum = cal_chksum( (unsigned short *)icmp,packsize);
#if _DEBUG_
printf("cksum = 0x%x\n", icmp->icmp_cksum);
#endif
return packsize;
}
void send_packet()
{
int packetsize;
{
nsend++;
packetsize = pack(nsend);
#if _DEBUG_
//printf("nsend = %d, packetsize = %d\n", nsend, packetsize);
#endif
if (sendto(sockfd, sendpacket, packetsize, 0,
(struct sockaddr *)&dest_addr, sizeof(dest_addr)) < 0)
{
perror("sendto error");
return;
}
//usleep(1000);
}
}
void recv_packet()
{
int n, count = 0;
socklen_t fromlen;
extern int errno;
//fromlen = sizeof(from);
while ( count < 2)
{
count++;
fromlen = sizeof(from);
alarm(MAX_WAIT_TIME);
if ((n = recvfrom(sockfd, recvpacket, sizeof(recvpacket), 0,
(struct sockaddr *)&from, &fromlen)) < 0)
{
if (errno == EINTR)
{
printf("errno == EINTR\n");
continue;
}
perror("recvfrom error");
continue;
}
alarm(0);
printf("n = %d\n", n);
//gettimeofday(&tvrecv, NULL);
if (unpack(recvpacket, n) == -1) continue;
nreceived++;
break;
}
return;
}
int unpack(char *buf,int len)
{
int iphdrlen;
struct ip *ip;
struct icmp *icmp;
struct timeval *tvsend;
double rtt;
ip = (struct ip *)buf;
iphdrlen = ip->ip_hl<<2;
icmp = (struct icmp *)(buf+iphdrlen);
len -= iphdrlen;
if (len < 8)
{
printf("ICMP packets\'s length is less than 8\n");
return -1;
}
//printf("%d, icmp_id = %d\n", ICMP_ECHOREPLY, icmp->icmp_id);
if ((icmp->icmp_type == ICMP_ECHOREPLY) && (icmp->icmp_id == pid))
{
gettimeofday(&tvrecv, NULL);
#if _DEBUG_
//printf("sec = %ld, usec = %ld\n", tvrecv.tv_sec, tvrecv.tv_usec);
#endif
tvsend = (struct timeval *)icmp->icmp_data;
tv_sub(&tvrecv, tvsend);
rtt = tvrecv.tv_sec*1000 + (double)tvrecv.tv_usec/1000;
printf("%d byte from %s: icmp_seq=%u ttl=%d rtt=%.3f ms\n",
len, inet_ntoa(from.sin_addr), icmp->icmp_seq, ip->ip_ttl, rtt);
return 1;
}
else return -1;
}
int main(int argc,char *argv[])
{
struct hostent *host;
struct protoent *protocol;
unsigned long inaddr = 0l;
//int waittime = MAX_WAIT_TIME;
memset(sendpacket, 'a', sizeof(sendpacket));
int size = 50*1024;
if (argc < 2)
{
printf("usage:%s hostname/IP address\n",argv[0]);
exit(1);
}
if ((protocol = getprotobyname("icmp")) == NULL)
{
perror("getprotobyname");
exit(1);
}
#if _DEBUG_
printf("the protocol number used is %d\n", protocol->p_proto);
#endif
if ((sockfd = socket(AF_INET, SOCK_RAW, protocol->p_proto)) < 0)
{
perror("socket error");
exit(1);
}
//setuid(getuid());
setsockopt(sockfd, SOL_SOCKET, SO_RCVBUF, &size, sizeof(size));
bzero(&dest_addr,sizeof(dest_addr));
dest_addr.sin_family = AF_INET;
if ((inaddr = inet_addr(argv[1])) == INADDR_NONE)
{
if ((host = gethostbyname(argv[1])) == NULL)
{
perror("gethostbyname error");
exit(1);
}
memcpy( (char *)&dest_addr.sin_addr, host->h_addr, host->h_length);
}
else
{
memcpy( (char *)&dest_addr.sin_addr.s_addr, (char *)&inaddr, sizeof(inaddr));
}
pid = getpid();
printf("PING %s(%s): %d bytes data in ICMP packets.\n",argv[1],
inet_ntoa(dest_addr.sin_addr),datalen);
signal(SIGALRM, statistics);
while (nsend < MAX_NO_PACKETS)
{
send_packet();
recv_packet();
sleep(1);
}
statistics(SIGALRM);
return 0;
}
void tv_sub(struct timeval *out, struct timeval *in)
{
if ((out->tv_usec -= in->tv_usec) < 0)
{
--out->tv_sec;
out->tv_usec += 1000000;
}
out->tv_sec -= in->tv_sec;
}