Message-ID: <4AD34EDB.3050201@redhat.com>
Date:	Mon, 12 Oct 2009 10:44:27 -0500
From:	Eric Sandeen <sandeen@...hat.com>
To:	Scott James Remnant <scott@...ntu.com>
CC:	linux-ext4@...r.kernel.org, Theodore Tso <tytso@....edu>
Subject: Re: [PATCH] e2fsck: Always fix last mount/write/check time with -p

Scott James Remnant wrote:
> After much debugging, we've come to the conclusion that the problems
> with a last mount, write and check time in the future by up to 24
> hours are not caused by buggy init scripts.  In fact, in every report,
> the "now" time was always correct and it was the last mount/write/check
> time that was wrong.
> 
> Investigation has led us to the following common scenarios:
> 
>  - Overwriting a system that previously had Windows installed, and
>    thus a localtime ticking hardware clock.  The installer may not
>    even be aware that the hardware clock is in localtime, and the
>    user fails to notice that the clock is wrong while they are
>    installing.
> 
>    The system is (properly) configured to assume the hardware clock
>    is in UTC, but until the user connects to a network (usually after
>    the first reboot) there is no external source to actually configure
>    this.
> 
>    Thus the first mount of the filesystem is in the future, a reboot
>    will cause the fsck error.
> 
>  - Installing onto a system with an externally synchronised hardware
>    clock, where the hardware clock is (without the user's knowledge
>    and beyond the capabilities of the Installer to detect) being set
>    to local time.
> 
>    Again, each time NTP will fix the problem, but after each reboot
>    the mount time will again be in the future.
> 
>    (This is a common case when using virtualisation, where the hardware
>     clock is in fact being emulated based off something's system clock.)
> 
> Ironically this error is in of itself counter-productive, since if you
> fix the underlying configuration problems or hardware clock problem,
> it doesn't solve the fact that the last *mount* time is still in the
> future.  So rebooting after fixing the problem actually gives you the
> fsck inconsistency!  This leads people to both the wrong assumptions
> and the wrong fixes!
> 
> This kind of filesystem issue is exactly what the -p ("preen") option
> is for, it can be safely fixed without human intervention.  Indeed,
> the very people that tend to run into this are the ones who are not
> equipped to fix it themselves.
> 
> Remove the previous "buggy_init_scripts" option and make its behaviour
> the default.  As with that option, fsck without -p will still treat
> this as an error.
> 
> Signed-off-by: Scott James Remnant <scott@...ntu.com>


This has recently started showing up on Fedora as well, and I agree that 
the 24h "grace period" seems like the best approach.  Ted, if you bless 
this I'd like to pull it into Fedora too.  I might even go so far as to 
say a times outside 24h don't necessarily warrant a full fs check 
either, but I'm not hung up on that.  :)

Minor nitpick, a

#define 24H_IN_SECONDS 86400

might be nicer than having the sorta-magic number sprinkled around 
(though I see that was just inherited from the current code)

Thanks,
-Eric

> ---
>  debian/rules              |    6 ------
>  e2fsck/e2fsck.conf.5.in   |   14 --------------
>  e2fsck/e2fsck.conf.ubuntu |    2 --
>  e2fsck/e2fsck.h           |    1 -
>  e2fsck/problem.c          |    4 ++--
>  e2fsck/super.c            |   39 +++++++++++++++------------------------
>  e2fsck/unix.c             |    2 +-
>  7 files changed, 18 insertions(+), 50 deletions(-)
>  delete mode 100644 e2fsck/e2fsck.conf.ubuntu
> 
> diff --git a/debian/rules b/debian/rules
> index f62e86f..f658bd1 100755
> --- a/debian/rules
> +++ b/debian/rules
> @@ -445,12 +445,6 @@ endif
>  	$(INSTALL) -p -m 0644 debugfs/debug_cmds.ct \
>  		${debdir}/ss-dev/usr/share/doc/libss${SS_SOVERSION}/examples
>  
> -	if test -f /etc/lsb-release && \
> -		grep -q DISTRIB_ID=Ubuntu /etc/lsb-release; then \
> -	$(INSTALL) -p -m 0644 e2fsck/e2fsck.conf.ubuntu \
> -		${debdir}/e2fsprogs/etc/e2fsck.conf; \
> -	fi
> -
>  	dh_installinfo -pcomerr-dev ${stdbuilddir}/lib/et/com_err.info
>  	dh_installinfo -pe2fslibs-dev ${stdbuilddir}/doc/libext2fs.info
>  
> diff --git a/e2fsck/e2fsck.conf.5.in b/e2fsck/e2fsck.conf.5.in
> index 6638a39..daa4e9f 100644
> --- a/e2fsck/e2fsck.conf.5.in
> +++ b/e2fsck/e2fsck.conf.5.in
> @@ -87,20 +87,6 @@ interrupts e2fsck using ^C, and the filesystem is not explicitly flagged
>  as containing errors, e2fsck will exit with an exit status of 0 instead
>  of 32.  This setting defaults to false.
>  .TP
> -.I buggy_init_scripts
> -Some buggy distributions (such as Ubuntu) have init scripts and/or
> -installers which fail to correctly set the system clock before running
> -e2fsck and/or formatting the filesystem initially.  Normally this
> -happens because the hardware clock is ticking localtime, instead of the
> -more proper and less error-prone UTC time.  So while the kernel is
> -booting, the system time (which in Linux systems always ticks in UTC
> -time) is set from the hardware clock, but since the hardware clock is
> -ticking localtime, the system time is incorrect.  Unfortunately, some
> -buggy distributions do not correct this before running e2fsck.  If this
> -option is set to a boolean value of true, we attempt to work around this
> -situation by allowing the superblock last write time, last mount time,
> -and last check time to be in the future by up to 24 hours.
> -.TP
>  .I clear_test_fs_flag
>  This boolean relation controls whether or not 
>  .BR e2fsck (8)
> diff --git a/e2fsck/e2fsck.conf.ubuntu b/e2fsck/e2fsck.conf.ubuntu
> deleted file mode 100644
> index 49d6d19..0000000
> --- a/e2fsck/e2fsck.conf.ubuntu
> +++ /dev/null
> @@ -1,2 +0,0 @@
> -[options]
> -	buggy_init_scripts = 1
> diff --git a/e2fsck/e2fsck.h b/e2fsck/e2fsck.h
> index ff73444..a5a811f 100644
> --- a/e2fsck/e2fsck.h
> +++ b/e2fsck/e2fsck.h
> @@ -339,7 +339,6 @@ struct e2fsck_struct {
>  
>  	/* misc fields */
>  	time_t now;
> -	time_t time_fudge;	/* For working around buggy init scripts */
>  	int ext_attr_ver;
>  	profile_t	profile;
>  	int blocks_per_page;
> diff --git a/e2fsck/problem.c b/e2fsck/problem.c
> index 540ac91..7e3cafb 100644
> --- a/e2fsck/problem.c
> +++ b/e2fsck/problem.c
> @@ -388,13 +388,13 @@ static struct e2fsck_problem problem_table[] = {
>  	/* Last mount time is in the future (fudged) */
>  	{ PR_0_FUTURE_SB_LAST_MOUNT_FUDGED,
>  	  N_("@S last mount time is in the future.\n\t(by less than a day, "
> -	     "probably due to buggy init scripts)  "),
> +	     "probably due to bad system clock last boot)  "),
>  	  PROMPT_FIX, PR_PREEN_OK | PR_NO_OK },
>  
>  	/* Last write time is in the future (fudged) */
>  	{ PR_0_FUTURE_SB_LAST_WRITE_FUDGED,
>  	  N_("@S last write time is in the future.\n\t(by less than a day, "
> -	     "probably due to buggy init scripts).  "),
> +	     "probably due to bad system clock last boot).  "),
>  	  PROMPT_FIX, PR_PREEN_OK | PR_NO_OK },
>  
>  	/* Block group checksum (latch question) is invalid. */
> diff --git a/e2fsck/super.c b/e2fsck/super.c
> index 097c87a..3fa0441 100644
> --- a/e2fsck/super.c
> +++ b/e2fsck/super.c
> @@ -465,7 +465,6 @@ void check_super_block(e2fsck_t ctx)
>  	int	inodes_per_block;
>  	int	ipg_max;
>  	int	inode_size;
> -	int	buggy_init_scripts;
>  	dgrp_t	i;
>  	blk_t	should_be;
>  	struct problem_context	pctx;
> @@ -801,34 +800,26 @@ void check_super_block(e2fsck_t ctx)
>  	}
>  
>  	/*
> -	 * Some buggy distributions (such as Ubuntu) have init scripts
> -	 * and/or installers which fail to correctly set the system
> -	 * clock before running e2fsck and/or formatting the
> -	 * filesystem initially.  Normally this happens because the
> -	 * hardware clock is ticking localtime, instead of the more
> -	 * proper and less error-prone UTC time.  So while the kernel
> -	 * is booting, the system time (which in Linux systems always
> -	 * ticks in UTC time) is set from the hardware clock, but
> -	 * since the hardware clock is ticking localtime, the system
> -	 * time is incorrect.  Unfortunately, some buggy distributions
> -	 * do not correct this before running e2fsck.  If this option
> -	 * is set to a boolean value of true, we attempt to work
> -	 * around this situation by allowing the superblock last write
> -	 * time, last mount time, and last check time to be in the
> -	 * future by up to 24 hours.
> -	 */
> -	profile_get_boolean(ctx->profile, "options", "buggy_init_scripts",
> -			    0, 0, &buggy_init_scripts);
> -	ctx->time_fudge = buggy_init_scripts ? 86400 : 0;
> -
> -	/*
>  	 * Check to see if the superblock last mount time or last
>  	 * write time is in the future.
> +	 *
> +	 * Unfortunately it's remarkably easy for users to get
> +	 * themselves into a problem where they've mounted the
> +	 * filesystem at least once before the hardware clock has
> +	 * been corrected by NTP (especially when overwriting or
> +	 * dual booting with Windows).  Sadly these are the very
> +	 * users who have no idea what to do when presented with a
> +	 * root shell and invited to fix their filesystem.
> +	 *
> +	 * Work around this situation by always allowing the
> +	 * superblock last write time, last mount time, and last
> +	 * check time to be in the future by up to 24 hours (when
> +	 * fsck called with -a/-p).
>  	 */
>  	if (fs->super->s_mtime > (__u32) ctx->now) {
>  		pctx.num = fs->super->s_mtime;
>  		problem = PR_0_FUTURE_SB_LAST_MOUNT;
> -		if (fs->super->s_mtime <= (__u32) ctx->now + ctx->time_fudge)
> +		if (fs->super->s_mtime <= (__u32) ctx->now + 86400)
>  			problem = PR_0_FUTURE_SB_LAST_MOUNT_FUDGED;
>  		if (fix_problem(ctx, problem, &pctx)) {
>  			fs->super->s_mtime = ctx->now;
> @@ -838,7 +829,7 @@ void check_super_block(e2fsck_t ctx)
>  	if (fs->super->s_wtime > (__u32) ctx->now) {
>  		pctx.num = fs->super->s_wtime;
>  		problem = PR_0_FUTURE_SB_LAST_WRITE;
> -		if (fs->super->s_wtime <= (__u32) ctx->now + ctx->time_fudge)
> +		if (fs->super->s_wtime <= (__u32) ctx->now + 86400)
>  			problem = PR_0_FUTURE_SB_LAST_WRITE_FUDGED;
>  		if (fix_problem(ctx, problem, &pctx)) {
>  			fs->super->s_wtime = ctx->now;
> diff --git a/e2fsck/unix.c b/e2fsck/unix.c
> index e629602..fe7093b 100644
> --- a/e2fsck/unix.c
> +++ b/e2fsck/unix.c
> @@ -304,7 +304,7 @@ static void check_if_skip(e2fsck_t ctx)
>  
>  	lastcheck = fs->super->s_lastcheck;
>  	if (lastcheck > ctx->now)
> -		lastcheck -= ctx->time_fudge;
> +		lastcheck -= 86400; /* see comment in super.c */
>  	if ((fs->super->s_state & EXT2_ERROR_FS) ||
>  	    !ext2fs_test_valid(fs))
>  		reason = _(" contains a file system with errors");

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists