Skip to contentSkip to conversation

BUG: Is it a bug in Linux time() function or Linux OS calls?

DISCUSSION Srinivas Nayak 3 participants 5 messages Mar 10, 2010 – Mar 11, 2010

lore
Share & tools
lore
#0 lore raw

Dear All,

I wrote a small program, that creates files at an interval of 1 minute. But the time at which the file is created and last written and the last modification time of the file as shown by ls command differs by 1 second. The code and the output is presented below. please let me know where could be the bug?

	root@new:/home/srinivas# cat b.c 
	#include <time.h> 
	#include <stdio.h> 
	#include <sys/stat.h> 
	#include <dirent.h> 
	#include <fcntl.h> 
	int main () 
	{ 
	    int fd; 
	    int i=0; 
	    time_t initial_time = time(NULL); 
	    time_t interval = 60; 
	    time_t curr_time = time(NULL); 
	    fd=open ("test1.txt", O_WRONLY | O_CREAT | O_TRUNC, 0666); 
	    write(fd,"abcd1",5); 
	    while(1) 
	    { 
		curr_time = time(NULL); 
		if(curr_time >= initial_time) 
		{ 
		    if(i==0) 
		    { 
			close(fd); 
			printf("\ntime before test2.txt fileopen= %d\n", time(NULL)); 
			fd=open ("test2.txt", O_WRONLY | O_CREAT | O_TRUNC, 0666); 
			write(fd,"abcd2",5); 
			printf("time after test2.txt filewrite= %d\n", time(NULL)); 
			system("ls -l --time-style=+%s test2.txt"); 
			initial_time += interval; 
			i=1; 
		    } 
		    else 
		    { 
			close(fd); 
			printf("\ntime before test1.txt fileopen= %d\n", time(NULL)); 
			fd=open ("test1.txt", O_WRONLY | O_CREAT | O_TRUNC, 0666); 
			write(fd,"abcd1",5); 
			printf("time after test1.txt filewrite= %d\n", time(NULL)); 
			system("ls -l --time-style=+%s test1.txt"); 
			initial_time += interval; 
			i=0; 
		    } 
		} 
		usleep(1000); 
	    } 
	    return 0; 
	} 
	root@new:/home/srinivas# gcc b.c 
	root@new:/home/srinivas# ./a.out 

	time before test2.txt fileopen= 1268203133 
	time after test2.txt filewrite= 1268203133 
	-rw-r--r-- 1 root root 5 1268203133 test2.txt 

	time before test1.txt fileopen= 1268203193 
	time after test1.txt filewrite= 1268203193 
	-rw-r--r-- 1 root root 5 1268203192 test1.txt 

	time before test2.txt fileopen= 1268203253 
	time after test2.txt filewrite= 1268203253 
	-rw-r--r-- 1 root root 5 1268203252 test2.txt 

	time before test1.txt fileopen= 1268203313 
	time after test1.txt filewrite= 1268203313 
	-rw-r--r-- 1 root root 5 1268203312 test1.txt 

	time before test2.txt fileopen= 1268203373 
	time after test2.txt filewrite= 1268203373 
	-rw-r--r-- 1 root root 5 1268203372 test2.txt 

	root@new:/home/srinivas# ls -ltr --time-style=+%s 
	total 40 
	-rwxrwxrwx  1 root     root      1095 1268202457 b.c 
	-rwxr-xr-x  1 root     root     10300 1268202459 a.out 
	-rw-r--r--  1 root     root         5 1268203312 test1.txt 
	-rw-r--r--  1 root     root         5 1268203372 test2.txt 
	root@new:/home/srinivas# 

Some system info:
===========
Linux new 2.6.24-19-server #1 SMP Wed Jun 18 14:44:47 UTC 2008 x86_64 GNU/Linux

Gnu C                  4.2.4
Linux C Library        2.7
binutils               2.18.0.20080103
util-linux             2.13.1
Console-tools          0.2.3
Sh-utils               6.10

===========

With best regards,

Yours sincerely,
Srinivas Nayak

My Home page: http://www.mathmeth.com/sn/
My Blog: http://srinivas-nayak.blogspot.com

The INTERNET now has a personality. YOURS! See your Yahoo! Homepage. http://in.yahoo.com/

#1 lore raw

Dear Srinivas Nayak,

I don't suppose you're writing to a network mounted disk, with the network device's clock about half a second behind your test machine? Otherwise -- and I'm just guessing here, and will leave it to you to UTSL -- this could be explained by use of two different algorithms for converting a higher-resolution time source to the one-second resolution. A truncation algorithm versus a rounding algorithm could produce the result you demonstrate.

Regards,

David

#2 lore raw

Dear David,

I am not writing to network mounted disk. This is locally mounted and filesystem is ext3.

With best regards,

Yours sincerely,
Srinivas Nayak

My Home page: http://www.mathmeth.com/sn/
My Blog: http://srinivas-nayak.blogspot.com

Quoted 16 lines · David Newall

From: David Newall <[email protected]>
Subject: Re: BUG: Is it a bug in Linux time() function or Linux OS calls?
To: "Srinivas Nayak" <[email protected]>
Cc: [email protected]
Date: Wednesday, 10 March, 2010, 2:17 PM
Dear Srinivas Nayak,

I don't suppose you're writing to a network mounted disk, with the network device's clock about half a second behind your test machine?  Otherwise -- and I'm just guessing here, and will leave it to you to UTSL -- this could be explained by use of two different algorithms for converting a higher-resolution time source to the one-second resolution.  A truncation algorithm versus a rounding algorithm could produce the result you demonstrate.

Quoted 4 lines

Regards,

David

The INTERNET now has a personality. YOURS! See your Yahoo! Homepage. http://in.yahoo.com/

#3 lore raw
In reply to Srinivas Nayak
Quoted 2 lines · Srinivas Nayak

I wrote a small program, that creates files at an interval of 1 minute. But the time at which the file is created and last written and the last modification time of the file as shown by ls command differs by 1 second. The code and the output is presented below. please let me know where could be the bug?

[snip]

Quoted 10 lines

Some system info:
===========
Linux new 2.6.24-19-server #1 SMP Wed Jun 18 14:44:47 UTC 2008 x86_64 GNU/Linux Gnu C                  4.2.4
Linux C Library        2.7
binutils               2.18.0.20080103
util-linux             2.13.1
Console-tools          0.2.3
Sh-utils               6.10
===========

Hrm. Is CONFIG_NO_HZ enabled? If so this is a known issue. The filesystem timestamps use xtime, which is normally updated every HZ. However to reduce overhead with NO_HZ we changed the update to be a half-second, which can cause issues like what you describe.

The issue has since been resolved, so I'd be interested to hear if you can still reproduce it on a more recent kernel.

thanks
-john

#4 lore raw

Dear John,

Thanks for your answer. This is informative.

With best regards,

Yours sincerely,
Srinivas Nayak

My Home page: http://www.mathmeth.com/sn/
My Blog: http://srinivas-nayak.blogspot.com

Quoted 38 lines · john stultz

From: john stultz <[email protected]>
Subject: Re: BUG: Is it a bug in Linux time() function or Linux OS calls?
To: "Srinivas Nayak" <[email protected]>
Cc: [email protected]
Date: Wednesday, 10 March, 2010, 10:49 PM
On Tue, Mar 9, 2010 at 11:59 PM,
Srinivas Nayak
<[email protected]>

Quoted 1 line · wrote:

I wrote a small program, that creates files at an

interval of 1 minute. But the time at which the file is created and last written and the last modification time of the file as shown by ls command differs by 1 second. The code and the output is presented below. please let me know where could be the bug?

Quoted 1 line

[snip]

Quoted 3 lines

Some system info:
===========
Linux new 2.6.24-19-server #1 SMP Wed Jun 18 14:44:47

UTC 2008 x86_64 GNU/Linux

Quoted 7 lines

Gnu C                  4.2.4
Linux C Library        2.7
binutils               2.18.0.20080103
util-linux             2.13.1
Console-tools          0.2.3
Sh-utils               6.10
===========

Hrm. Is CONFIG_NO_HZ enabled?  If so this is a known issue. The
filesystem timestamps use xtime, which is normally updated every HZ.
However to reduce overhead with NO_HZ we changed the update to be a
half-second, which can cause issues like what you describe.

Quoted 7 lines

The issue has since been resolved, so I'd be interested to hear if you
can still reproduce it on a more recent kernel.

thanks
-john

Your Mail works best with the New Yahoo Optimized IE8. Get it NOW! http://downloads.yahoo.com/in/internetexplorer/