缘起
因项目需要，自行设计一套通用的文件读写锁，要求该机制能用于本地文件系统和NFS文件系统。
内核的文件数据结构
内核中有3个数据结构和文件直接相关，分别是：file descriptor table, file table and i-node table。其中file descriptor table是进程私有的；file table和inode table是独立于进程的。进程内部的文件描述符fd只在该进程有才有意义。File table可在进程内共享，也可在进程间共享。I-node table则在整个系统中共享。
文件描述符fd可由Open，dup和fork等操作创建；file table只能有open创建；i-node table在创建真实文件是创建。
摘自Apue-3.12
摘自Apue-14.3
摘自TLPI-5.4
对TLPI-5.4的图作如下说明：
1) A中的fd 1和20共享file table记录，可认为fd 20复制于fd 1(可也反过来)
2) A和B的fd 2共享file table记录，可认为B由A fork产生，B重定向fd 0和1。
3) File table中的0和86指向相同的文件，这意味着0和86由open相同的文件。
理解这个3个数据结构的关系大体可知道linux是如何管理文件。Stat和fcntl可获取或者修改相关的文件信息，结合这两个接口将有助于理解这3个文件结构。
文件记录锁fcntl
文件记录锁位于i-node table这个结构中，使用PID作为锁拥有者的标识。这使其拥有如下特点：
1） 记录锁采用(PID, start,end)三元组作为锁标识，一个文件可拥有多个记录锁，同一区域只允许有一个记录锁。
2） 当进程终止(正常/不正常)，该进程拥有的所有记录锁都将释放。
3） 同一个进程中，指向同一文件(i-node)的fd都可以操作该文件上的记录锁：如释放、修改等。显式调用F_UNLCK和close(fd)都将释放锁，close将释放整个文件中该进程拥有的所有记录锁。
4） 记录锁不被fork的子进程继承(PID不同)。
5） 记录锁的类型转换、改变锁范围等操作均为原子的。
6） 未设置FD_CLOEXEC时，记录锁将被exec后的进程继承(PID相同)。
7） 记录锁对文件打开mode有要求：加读锁要求fd有读权限；加写锁要求fd有写权限。
fcntl的使用示例和测试代码：
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h>
int lock_reg(int, int, int, off_t, int, off_t);
#define read_lock(fd, offset, whence, len) \
lock_reg((fd), F_SETLK, F_RDLCK, (offset), (whence), (len))
#define readw_lock(fd, offset, whence, len) \
lock_reg((fd), F_SETLKW, F_RDLCK, (offset), (whence), (len))
#define write_lock(fd, offset, whence, len) \
lock_reg((fd), F_SETLK, F_WRLCK, (offset), (whence), (len))
#define writew_lock(fd, offset, whence, len) \
lock_reg((fd), F_SETLKW, F_WRLCK, (offset), (whence), (len))
#define un_lock(fd, offset, whence, len) \
lock_reg((fd), F_SETLK, F_UNLCK, (offset), (whence), (len))
pid_t lock_test(int, int, off_t, int, off_t);
#define is_read_lockable(fd, offset, whence, len) \
(lock_test((fd), F_RDLCK, (offset), (whence), (len)) == 0)
#define is_write_lockable(fd, offset, whence, len) \
(lock_test((fd), F_WRLCK, (offset), (whence), (len)) == 0)
int lock_reg(int fd, int cmd, int type, off_t offset, int whence, off_t len)
{
struct flock lock;
lock.l_type = type; /* F_RDLCK, F_WRLCK, F_UNLCK */
lock.l_start = offset; /* byte offset, relative to l_whence */
lock.l_whence = whence; /* SEEK_SET, SEEK_CUR, SEEK_END */
lock.l_len = len; /* #bytes (0 means to EOF) */
return(fcntl(fd, cmd, &lock));
}
/* Note: lock_test always success with the obtained lock process */
pid_t lock_test(int fd, int type, off_t offset, int whence, off_t len)
{
struct flock lock;
lock.l_type = type; /* F_RDLCK or F_WRLCK */
lock.l_start = offset; /* byte offset, relative to l_whence */
lock.l_whence = whence; /* SEEK_SET, SEEK_CUR, SEEK_END */
lock.l_len = len; /* #bytes (0 means to EOF) */
if (fcntl(fd, F_GETLK, &lock) < 0)
perror("fcntl error");
/* printf("F_RDLCK=%d, F_WRLCK=%d, F_UNLCK=%d\n", F_RDLCK, F_WRLCK, F_UNLCK); */
printf(" l_type=%d, l_start=%lu, l_where=%d, l_len=%lu, l_pid=%d\n"
, lock.l_type, lock.l_start, lock.l_whence, lock.l_len, lock.l_type);
if (lock.l_type == F_UNLCK)
return(0); /* false, region isn't locked by another proc */
return(lock.l_pid); /* true, return pid of lock owner */
}
/* ======================================================== */
void Usage(const char* program)
{
fprintf(stderr, "Usage: %s read/read2/write/rd-wr/wr-rd/fork/dup file time\n", program);
exit(-1);
}
int main(int argc, char *argv[])
{
if(argc < 3){
Usage(argv[0]);
}
int interval = 10;
if(argc > 3) {
interval = atoi(argv[3]);
}
printf("PID=%d\n", getpid());
char* path = argv[2];
if( strcmp(argv[1], "read") == 0 ) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
printf("read lock try\n");
/* read lock entire file */
if( read_lock(fd, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Read Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "read2") == 0 ){
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
printf("read lock try\n");
/* read lock entire file */
if( read_lock(fd, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Read Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
if( read_lock(fd, 0, SEEK_SET, 1) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock 2 success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "write") == 0) {
int fd = open(path, O_WRONLY );
printf("file: %s, fd=%d\n", path, fd);
if(fd == -1) {
perror("open");
exit(0);
}
printf("write lock try\n");
/* write lock entire file */
if( write_lock(fd, 0, SEEK_SET, 0) < 0){
perror("write_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
printf("Read Lock by pid=%d\n", lock_test(fd, F_RDLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("write lock success.\n");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
int ret = write(fd, path, strlen(path) );
if(ret != strlen(path)){
perror("write");
}
sleep(interval);
close(fd);
exit(0);
}
else if( strcmp(argv[1], "rd-wr") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* read lock entire file */
if( read_lock(fd, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
/* if( un_lock(fd, 0, SEEK_SET, 0) < 0){
perror("un-lock:");
} */
if( write_lock(fd, 0, SEEK_SET, 1) < 0){
perror("write_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("write lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "wr-rd") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd, 0, SEEK_SET, 0) < 0){
perror("write_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("write lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
/* if( un_lock(fd, 0, SEEK_SET, 0) < 0){
perror("un-lock:");
} */
if( read_lock(fd, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "fork") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd, 0, SEEK_SET, 0) < 0){
perror("write_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("write lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
pid_t pid;
if ((pid = fork()) < 0) {
perror("un-lock:");
exit(1);
}
else if(pid == 0){
sleep(1);
printf("New PID=%d\n", getpid());
if( read_lock(fd, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
sleep(interval);
exit(0);
}
else{
exit(1);
}
}
else if( strcmp(argv[1], "dup") == 0) {
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd, 0, SEEK_SET, 0) < 0){
perror("write_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("write lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd, F_WRLCK, 0, SEEK_SET, 0 ));
int fd2 = 0;
if( (fd2=dup(fd)) < 0){
perror("dup:");
exit(1);
}
close(fd);
if( read_lock(fd2, 0, SEEK_SET, 0) < 0){
perror("read_lock: ");
printf("Write Lock by pid=%d\n", lock_test(fd2, F_WRLCK, 0, SEEK_SET, 0 ));
exit(0);
}
printf("read lock success.\n");
printf("Lock by pid=%d\n", lock_test(fd2, F_WRLCK, 0, SEEK_SET, 0 ));
sleep(interval);
exit(0);
} else {
printf("Unknown action!\n");
}
return 0;
}
文件锁flock
文件锁位于file table这个数据结构中，一个文件可有多个file table entry(多次open)，一个entry有且只有一个文件锁(共享/独占)。文件锁有如下特点：
1）每个file table entry有且只有一个文件锁，fork和dup后的fd指向同一个file table entry，故拥有同一个文件锁。这意味着文件锁可被fork后的子进程继承。
2）显式调用LOCK_UN将释放该file table entry的文件锁；当所有指向同一file table entry的fd关闭后，该file table entry将被释放，其上的文件锁也将随之释放。
3）一个进程中多次打开同一文件将创建多个file table entry，这意味着这些entry的文件锁是独立的。
4）一个文件只能有一个文件锁，锁类型可转换，flock不保证锁类型转换是原子操作。
5）未设置FD_CLOEXEC时，记录锁将被exec后的进程继承(file entry未释放)。
6）文件锁对文件的读写权限没有要求。
flock的使用示例和测试代码：
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <stdlib.h>
/* #include <fcntl.h> */
#include <sys/file.h>
#define read_lock(fd) \
flock((fd), LOCK_SH|LOCK_NB)
#define readw_lock(fd) \
flock((fd), LOCK_SH)
#define write_lock(fd) \
flock((fd), LOCK_EX|LOCK_NB)
#define writew_lock(fd, offset, whence, len) \
flock((fd), LOCK_EX)
#define un_lock(fd) \
flock((fd), LOCK_UN)
/* ======================================================== */
void Usage(const char* program)
{
fprintf(stderr, "Usage: %s read/read2/write/rd-wr/wr-rd/fork/dup/unlink file time\n", program);
exit(-1);
}
int main(int argc, char *argv[])
{
if(argc < 3){
Usage(argv[0]);
}
int interval = 10;
if(argc > 3) {
interval = atoi(argv[3]);
}
printf("PID=%d\n", getpid());
char* path = argv[2];
if( strcmp(argv[1], "read") == 0 ) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
printf("read lock try...\n");
/* read lock entire file */
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "read2") == 0 ){
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
printf("read lock try...\n");
/* read lock entire file */
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock 2 success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "write") == 0) {
int fd = open(path, O_WRONLY );
printf("file: %s, fd=%d\n", path, fd);
if(fd == -1) {
perror("open");
exit(0);
}
printf("write lock try...\n");
/* write lock entire file */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
int ret = write(fd, path, strlen(path) );
if(ret != strlen(path)){
perror("write");
}
sleep(interval);
close(fd);
exit(0);
}
else if( strcmp(argv[1], "rd-wr") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* read lock entire file */
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
/* if( un_lock(fd) < 0){
perror("un-lock:");
} */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "wr-rd") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
/* if( un_lock(fd, 0, SEEK_SET, 0) < 0){
perror("un-lock:");
} */
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "fork") == 0){
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
pid_t pid;
if ((pid = fork()) < 0) {
perror("fork:");
exit(1);
}
else if(pid == 0){
sleep(1);
printf("New PID=%d\n", getpid());
if( read_lock(fd) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
fd = open(path, O_RDWR); /* O_RDWR */
sleep(interval);
exit(0);
}
else{
sleep(interval);
exit(1);
}
}
else if( strcmp(argv[1], "dup") == 0) {
int fd = open(path, O_RDWR); /* O_RDWR */
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
int fd2 = 0;
if( (fd2=dup(fd)) < 0){
perror("dup:");
exit(1);
}
close(fd);
if( read_lock(fd2) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "open") == 0) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( write_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
int fd2 = open(path, O_WRONLY);
if(fd2 < 0){
perror("open:");
exit(1);
}
if( read_lock(fd2) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "open2") == 0) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( read_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("write lock success.\n");
int fd2 = open(path, O_WRONLY);
if(fd2 < 0){
perror("open:");
exit(1);
}
if( write_lock(fd2) < 0){
perror("read_lock: ");
exit(0);
}
printf("read lock success.\n");
sleep(interval);
exit(0);
}
else if( strcmp(argv[1], "unlink") == 0) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
exit(0);
}
/* write lock entire file */
if( read_lock(fd) < 0){
perror("write_lock: ");
exit(0);
}
printf("read lock success.\n");
int fd2 = open(path, O_WRONLY);
if(fd2 < 0){
perror("open:");
exit(1);
}
unlink(path);
sleep(interval);
exit(0);
} else {
printf("Unknown action!\n");
}
return 0;
}
NFS文件锁-fcntl
NFS3支持文件记录fcntl和O_EXCL创建文件，但不支持文件锁。NFS的文件锁是通过Network Lock Manager(lockd后台进程)这个独立程序实现的，NFS server本身依旧是无状态的，所有的锁状态均由NLM维护。从NFS4开始，锁协议融入NFS自身的协议中，锁状态由NFS server维护，这意味这NFS4是有状态的，未简化锁设计，采用租赁锁。
文件读写锁
双文件读写锁机制
双文件读写锁机制采用目标文件(datafile)和“众所周知”的临时写文件(datafile.lock)，依赖于新文件的原子创建来保证进程间的互斥写操作。其本质：新文件的原子创建模拟写锁，保证有且只有一个进程能够进行写操作；多进程直接读数据，并依赖操作系统保证打开后的文件在fd关闭之前一直可读的技巧保证读不会因文件删除而中断。
缺点：1) 不支持部分修改；2）某个进程拥有“读锁”，无法保证后续进程也将拿到“读锁”(文件被移除)。3）临时文件(锁文件)遗留问题。如果临时文件(datafile.lock)未能在写操作终止后(正常/不正常)被移除，那么将阻塞其他写进程。这个问题很棘手。可通过atexit()或者顶层脚本部分解决，但两者都很难避免删除的文件不是另一个写进程新创建的临时文件。
双文件读写锁机制的程序流程如下：
改进的双文件读写锁机制
改进的双文件读写锁机制采用文件硬链接数来模拟“读锁”，硬链接大于1表示有进程拥有目标文件的读锁。
改进的双文件机制和实际的“读写锁”吻合，只有当“读锁”全部被释放后，进程才能拿到“写锁”。但是仍然无法解决“锁文件遗留问题”。
上面的流程可进一步简化，既然建立link必不可少，可让读进程直接读link文件，通过link()的原子操作保证获取“读锁”的原子性，从而简化读流程。
替代的类文件锁技术
open(file, O_CREAT | O_EXCL,...) plus unlink(file)
注释：
open(file, O_CREAT | O_EXCL,...)获取锁；unlink(file)释放锁。
处理遗留锁文件的一个较可靠的办法：将PID写入锁文件；原子创建失败后，读取PID，kill(pid,0)检查PID是否有效，若无效，删除后，重新原子创建。(删除和创建存在“竞态”，无法避免！)
link(file, lockfile) plus unlink(lockfile)
用法：link和unlink均为原子的。每个需要加锁的进程，先创建一个临时文件，并link一个“众所周知”的“锁文件”到该目标文件。Link成功获取锁；unlink释放锁。
open(file, O_CREAT | O_TRUNC | O_WRONLY, 0) plus unlink(file)
用法：当使用O_TRUNC | O_WRONLY去打开一个已存在的文件，若该进程不具有写权限时，open将失败。
总结
依赖于“锁文件”的替代文件锁机制都无法真正可靠地解决“锁文件”遗留问题。这在实际工程项目会成为一个很棘手的问题。在未找到可靠解决“锁文件”遗留问题的方案之前，还是尽可能地使用系统自带的文件锁机制(flock/fcntl)。
参考书籍：
高级UNIX环境编程
The Linux Programming Interface
相关文章：