monotonic time for mount.cifs timeouts
Subsystems (2)
Hi,
attached are patches to spot the clock_gettime function and to use that and the monotonic clock for timeout handling if possible.
If there are no objections, can you please apply them, Jeff?
Thanks
Björn
From 250d5b5df23d93ebed66a4c39bccf3a3614c54ad Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Jacke?= <[email protected]>
Date: Tue, 24 Aug 2010 09:39:32 +0200
Subject: [PATCH 1/2] autoconf: add checks for clock_gettime
--- configure.ac | 10 ++++++++++ 1 files changed, 10 insertions(+), 0 deletions(-)
configure.ac +10 -0
@@ -110,6 +110,16 @@ AC_FUNC_STRNLEN # check for required functions AC_CHECK_FUNCS([alarm atexit endpwent getmntent getpass gettimeofday inet_ntop memset realpath setenv strchr strcmp strdup strerror strncasecmp strndup strpbrk strrchr strstr strtol strtoul tolower uname], , [AC_MSG_ERROR([necessary functions(s) not found])]) +AC_CHECK_FUNCS(clock_gettime, [], [ + AC_CHECK_LIB(rt, clock_gettime, [ + AC_DEFINE(HAVE_CLOCK_GETTIME, 1) + AC_DEFINE(HAVE_CLOCK_GETTIME,1, + [Whether the clock_gettime func is there]) + LIBS="$LIBS -lrt" + ]) + ]) + + # ugly, but I'm not sure how to check for functions in a library that's not in $LIBS cu_saved_libs=$LIBS LIBS="$LIBS $KRB5_LDADD"
Signature · 1 line
1.7.1
From 4e8d243b9b0a5eb7bf4619c4cddbd9d3a8f4eb2d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Jacke?= <[email protected]>
Date: Tue, 24 Aug 2010 09:41:01 +0200
Subject: [PATCH 2/2] mount.cifs: use monotonic time for timeouts
this is especially important during the boot process, where the clock is often being set initially and clock jumps are more common.
--- mtab.c | 23 ++++++++++++++++++++--- 1 files changed, 20 insertions(+), 3 deletions(-)
mtab.c +20 -3
@@ -39,6 +39,7 @@#include <stdlib.h> #include <signal.h> #include "mount.h" +#include "config.h" /* Updating mtab ----------------------------------------------*/
@@ -60,6 +61,22 @@ setlkw_timeout (int sig __attribute__((unused))) {/* nothing, fcntl will fail anyway */ } +/* use monotonic time for timeouts */ +struct timeval +mono_time (void ) { +struct timeval ret; +#if defined(HAVE_CLOCK_GETTIME ) && defined(CLOCK_MONOTONIC ) +struct timespec ts; +if (clock_gettime (CLOCK_MONOTONIC , &ts) ==0 ) { + ret.tv_sec = ts.tv_sec; + ret.tv_usec = ts.tv_nsec/1000 ; +return ret; + } +#endif +gettimeofday (&ret,NULL ); +return ret; +} +/* Remove lock file. */ void unlock_mtab (void ) {
@@ -150,7 +167,7 @@ lock_mtab (void) { }close (i); -gettimeofday (&maxtime,NULL ); + maxtime =mono_time (); maxtime.tv_sec +=MOUNTLOCK_MAXTIME ; waittime.tv_sec =0 ;
@@ -177,7 +194,7 @@ lock_mtab (void) {if (lockfile_fd <0 ) {/* Strange... Maybe the file was just deleted? */ -gettimeofday (&now,NULL ); + now =mono_time ();if (errno ==ENOENT && now.tv_sec < maxtime.tv_sec) { we_created_lockfile =0 ;continue ;
@@ -199,7 +216,7 @@ lock_mtab (void) { (void )unlink (linktargetfile); }else {/* Someone else made the link. Wait. */ -gettimeofday (&now,NULL ); + now =mono_time ();if (now.tv_sec < maxtime.tv_sec) {alarm (maxtime.tv_sec - now.tv_sec);if (fcntl (lockfile_fd,F_SETLKW , &flock) ==-1 ) {
Signature · 1 line
1.7.1
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 9 lines · On Tue, 24 Aug 2010 18:16:19 +0200 Björn JACKE…
Hi,
attached are patches to spot the
clock_gettimefunction and to use that and the monotonic clock for timeout handling if possible.If there are no objections, can you please apply them, Jeff?
Thanks
Björn
Sure, they look reasonable to me. If no one chimes in the next couple of days, I'll plan to apply them.
Thanks for the patches,
Signature · 1 line
Jeff Layton <[email protected]>
A message that is not in the archive (sent off-list or not yet received): 20100824133418.6319ead7-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org
Quoted 9 lines · Jeff Layton
Quoted 6 lines · On Tue, 24 Aug 2010 18:16:19 +0200 Björn JACKE…
Hi,
attached are patches to spot the
clock_gettimefunction and to use that and the monotonic clock for timeout handling if possible.If there are no objections, can you please apply them, Jeff?
Any pointers to background on gettimeofday vs clock_gettime(CLOCK_MONOTONIC) and why the latter is better?
Signature · 3 lines
Thanks, Steve
A message that is not in the archive (sent off-list or not yet received): AANLkTi==5ZbC7Fb8wnCea50OndXZ=QwV5xPwv19K1Fyz-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org
Quoted 2 lines · On Tue, 24 Aug 2010 13:07:55 -0500 Steve French…
Any pointers to background on gettimeofday vs
clock_gettime(CLOCK_MONOTONIC)and why the latter is better?
IIUC, the reason for this is that gettimeofday is affected by wallclock changes (think NTP):
CLOCK_MONOTONIC
Clock that cannot be set and represents monotonic time since some unspecified starting point.
...so by preferring CLOCK_MONOTONIC, the mtab locking code isn't affected by clock jumps or skew. Bjorn, is this correct?
Signature · 1 line
Jeff Layton <[email protected]>
A message that is not in the archive (sent off-list or not yet received): 20100824142248.164caebe-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org
Quoted 9 lines · On Tue, 24 Aug 2010 14:22:48 -0400 Jeff Layton…
Quoted 2 lines · On Tue, 24 Aug 2010 13:07:55 -0500 Steve French…
Any pointers to background on gettimeofday vs
clock_gettime(CLOCK_MONOTONIC)and why the latter is better?IIUC, the reason for this is that gettimeofday is affected by wallclock changes (think NTP):
Quoted 7 lines
CLOCK_MONOTONIC
Clock that cannot be set and represents monotonic time since some unspecified starting point....so by preferring
CLOCK_MONOTONIC, the mtab locking code isn't affected by clock jumps or skew. Bjorn, is this correct?
Correction...CLOCK_MONOTONIC is affected by adjtimex(). Over a 30s period though, that shouldn't mean much of a delta. The big danger is large clock jumps and this should take care of that.
Out of curiousity though...did you or someone you know hit this problem in practice, Bjorn? Or did you just notice it via inspection?
Signature · 1 line
Jeff Layton <[email protected]>
A message that is not in the archive (sent off-list or not yet received): 20100824143433.5049ebe1-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org
On 2010-08-24 at 14:34 -0400 Jeff Layton sent off:
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
Any pointers to background on gettimeofday vs
clock_gettime(CLOCK_MONOTONIC)and why the latter is better?IIUC, the reason for this is that gettimeofday is affected by wallclock changes (think NTP):
Quoted 6 lines
Quoted 6 lines
CLOCK_MONOTONIC
Clock that cannot be set and represents monotonic time since some unspecified starting point....so by preferring
CLOCK_MONOTONIC, the mtab locking code isn't affected by clock jumps or skew. Bjorn, is this correct?
yes, correct. In the worst case you start a timeout period while the clock is set to 2020, during the timeout period the clock is corrected by ntpdate. In that case the timeout using the RTC clock is a bit longer than expected :-)
Quoted 6 lines
Correction...CLOCK_MONOTONIC is affected by
adjtimex(). Over a 30s period though, that shouldn't mean much of a delta. The big danger is large clock jumps and this should take care of that.Out of curiousity though...did you or someone you know hit this problem in practice, Bjorn? Or did you just notice it via inspection?
you see all kinds of strange timing effects if you have a VM that has an unstable clock which you need to regulary adjust had via ntpdate for example. Programs hang or for a hile or forever in the worst case. I didn't see a problem in this particular case with mount.cifs but as this is something which is called at boot time where ntpdate often runs at the sme time to initally hard reset the clock this is something I thought should to be adjusted to use the monotonic clock.
Cheers
Björn
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 26 lines · On Tue, 24 Aug 2010 22:08:09 +0200 Björn JACKE…
On 2010-08-24 at 14:34 -0400 Jeff Layton sent off:
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
Any pointers to background on gettimeofday vs
clock_gettime(CLOCK_MONOTONIC)and why the latter is better?IIUC, the reason for this is that gettimeofday is affected by wallclock changes (think NTP):
Quoted 6 lines
Quoted 6 lines
CLOCK_MONOTONIC
Clock that cannot be set and represents monotonic time since some unspecified starting point....so by preferring
CLOCK_MONOTONIC, the mtab locking code isn't affected by clock jumps or skew. Bjorn, is this correct?yes, correct. In the worst case you start a timeout period while the clock is set to 2020, during the timeout period the clock is corrected by ntpdate. In that case the timeout using the RTC clock is a bit longer than expected :-)
Quoted 6 lines
Correction...CLOCK_MONOTONIC is affected by
adjtimex(). Over a 30s period though, that shouldn't mean much of a delta. The big danger is large clock jumps and this should take care of that.Out of curiousity though...did you or someone you know hit this problem in practice, Bjorn? Or did you just notice it via inspection?
Quoted 8 lines
you see all kinds of strange timing effects if you have a VM that has an unstable clock which you need to regulary adjust had via ntpdate for example. Programs hang or for a hile or forever in the worst case. I didn't see a problem in this particular case with mount.cifs but as this is something which is called at boot time where ntpdate often runs at the sme time to initally hard reset the clock this is something I thought should to be adjusted to use the monotonic clock.
Ok, thanks for the clarification. Committed...
This should make the 4.7 release.
Thanks!
Signature · 1 line
Jeff Layton <[email protected]>
A message that is not in the archive (sent off-list or not yet received): 20100824193025.24b0f8c4-4QP7MXygkU+dMjc06nkz3ljfA9RmPOcC@public.gmane.org
Quoted 8 lines · Jeff Layton
On Tue, 24 Aug 2010 22:08:09 +0200
Björn JACKE <[email protected]> wrote:Ok, thanks for the clarification. Committed...
This should make the 4.7 release.
Thanks!
Yes - makes sense - thanks
Signature · 3 lines
Thanks, Steve