Debian Bug report logs - #839071
zfsutils-linux: Issues based on the fact that ZFS relies on mtab.

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Eric Desrochers <eric.desrochers@canonical.com>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: zfsutils-linux: Issues based on the fact that ZFS relies on mtab.

Date: Wed, 28 Sep 2016 09:39:36 -0400

Package: zfsutils-linux
Severity: normal

Dear Maintainer,

ZFS relies on mtab which is a broken concept (putting info about mounted file systems on a mounted file system).
The upstream commit "79251738" addresses the situation by relying on /proc/self/mounts instead. 

Ideally, ZFS should be integrated with libmount in a second phase (ZoL maintainers are aware that eventually a 2nd is needed to integrate libmount), until then /proc/self/mounts is a viable solution to address issues cause by mtab.

Related issues
---
Ubuntu bug      : https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/1607920
Upstream bug    : https://github.com/zfsonlinux/zfs/issues/4680

Upstream commit 
---
https://github.com/zfsonlinux/zfs/commit/792517389fad5c495a2738b61c2e9c65dedaaa9a

Eric

-- System Information:
Debian Release: 8.4
  APT prefers stable
  APT policy: (500, 'stable')
Architecture: amd64 (x86_64)

Kernel: Linux 4.2.0-17-generic (SMP w/4 CPU cores)
Locale: LANG=en_CA.UTF-8, LC_CTYPE=en_CA.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
Init: systemd (via /run/systemd/system)

Message #10 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Turbo Fredriksson <turbo@bayour.com>

To: Eric Desrochers <eric.desrochers@canonical.com>, 839071@bugs.debian.org

Cc: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: Re: [Pkg-zfsonlinux-devel] Bug#839071: zfsutils-linux: Issues based on the fact that ZFS relies on mtab.

Date: Wed, 28 Sep 2016 14:48:13 +0100

> ZFS relies on mtab which is a broken concept (putting info about mounted file systems on a mounted file system).
> The upstream commit "79251738" addresses the situation by relying on /proc/self/mounts instead. 

On Debian GNU/Linux almost since the "dawn of time", /etc/mtab is a symlink to /proc/self/mounts so this is less of a problem on Debian GNU/Linux.

Message #15 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Petter Reinholdtsen <pere@hungry.com>

To: 839071@bugs.debian.org, Eric Desrochers <eric.desrochers@canonical.com>

Subject: Re: Bug#839071: zfsutils-linux: Issues based on the fact that ZFS relies on mtab.

Date: Wed, 28 Sep 2016 16:15:37 +0200

[Turbo Fredriksson]
> On Debian GNU/Linux almost since the "dawn of time", /etc/mtab is a
> symlink to /proc/self/mounts so this is less of a problem on Debian
> GNU/Linux.

Note, the "dawn of time" happened in sysvinit version 2.88dsf-14
released 2011-12-13 to experimental, and version 2.88dsf-16 released
2011-12-17 to unstable.

-- 
Happy hacking
Petter Reinholdtsen

Message #20 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Eric Desrochers <eric.desrochers@canonical.com>

To: 839071@bugs.debian.org

Subject: Patch for bug #839071

Date: Wed, 28 Sep 2016 10:15:45 -0400

[Message part 1 (text/plain, inline)]

[0001-Change-etc-mtab-to-proc-self-mounts.patch (text/x-patch, attachment)]

Message #25 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Petter Reinholdtsen <pere@hungry.com>

To: Eric Desrochers <eric.desrochers@canonical.com>, 839071@bugs.debian.org

Subject: Re: [Pkg-zfsonlinux-devel] Bug#839071: Patch for bug #839071

Date: Thu, 29 Sep 2016 10:17:00 +0200

Hi.

I got a quick question.  The patch seem to set MNTTAB to
"/proc/self/mounts", but include changes like this:

[Eric Desrochers]
> --- a/cmd/mount_zfs/mount_zfs.c
> +++ b/cmd/mount_zfs/mount_zfs.c
> @@ -292,11 +292,11 @@ mtab_is_writeable(void)
>  	struct stat st;
>  	int error, fd;
>  
> -	error = lstat(MNTTAB, &st);
> +	error = lstat("/etc/mtab", &st);
>  	if (error || S_ISLNK(st.st_mode))
>  		return (0);
>  
> -	fd = open(MNTTAB, O_RDWR | O_CREAT, 0644);
> +	fd = open("/etc/mtab", O_RDWR | O_CREAT, 0644);
>  	if (fd < 0)
>  		return (0);
>  
> @@ -318,21 +318,21 @@ mtab_update(char *dataset, char *mntpoint, char *type, char *mntopts)
>  	mnt.mnt_freq = 0;
>  	mnt.mnt_passno = 0;
>  
> -	fp = setmntent(MNTTAB, "a+");
> +	fp = setmntent("/etc/mtab", "a+");
>  	if (!fp) {
>  		(void) fprintf(stderr, gettext(
> -		    "filesystem '%s' was mounted, but %s "
> +		    "filesystem '%s' was mounted, but /etc/mtab "
>  		    "could not be opened due to error %d\n"),
> -		    dataset, MNTTAB, errno);
> +		    dataset, errno);
>  		return (MOUNT_FILEIO);
>  	}
>  
>  	error = addmntent(fp, &mnt);
>  	if (error) {
>  		(void) fprintf(stderr, gettext(
> -		    "filesystem '%s' was mounted, but %s "
> +		    "filesystem '%s' was mounted, but /etc/mtab "
>  		    "could not be updated due to error %d\n"),
> -		    dataset, MNTTAB, errno);
> +		    dataset, errno);
>  		return (MOUNT_FILEIO);
>  	}
>  

Why do you hardcode /etc/mtab here?  I fail to understand how your patch
is intended to work, and hope you can reduce my confusion.

-- 
Happy hacking
Petter Reinholdtsen

Message #30 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Petter Reinholdtsen <pere@hungry.com>

To: Richard Laager <rlaager@wiktel.com>, 839071@bugs.debian.org, Eric Desrochers <eric.desrochers@canonical.com>

Subject: Re: Bug#839071: Bug#839071: Patch for bug #839071

Date: Thu, 29 Sep 2016 10:28:44 +0200

[Richard Laager]
> The *writing* is hard-coded to /etc/mtab, but if it doesn't exist,
> that's fine. ZFS just won't update /etc/mtab.

Right.  Perhaps better to use two defines, one MNTTAB_READ and one
MNTTAB_WRITE to make this more clear, instead of introducing the
/etc/mtab hardcoding?

-- 
Happy hacking
Petter Reinholdtsen

Message #35 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Richard Laager <rlaager@wiktel.com>

To: Petter Reinholdtsen <pere@hungry.com>, 839071@bugs.debian.org, Eric Desrochers <eric.desrochers@canonical.com>

Subject: Re: [Pkg-zfsonlinux-devel] Bug#839071: Bug#839071: Patch for bug #839071

Date: Thu, 29 Sep 2016 03:24:53 -0500

On 09/29/2016 03:17 AM, Petter Reinholdtsen wrote:
> Why do you hardcode /etc/mtab here?  I fail to understand how your patch
> is intended to work, and hope you can reduce my confusion.

All *reading* is hard-coded to /proc/self/mounts. This removes the
dependency on /etc/mtab.

Ubuntu is specifically concerned about a race on first boot of cloud
images where the /etc/mtab -> /proc/self/mounts has not been created in
time.

The *writing* is hard-coded to /etc/mtab, but if it doesn't exist,
that's fine. ZFS just won't update /etc/mtab.

-- 
Richard

Message #40 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Richard Laager <rlaager@wiktel.com>

To: Petter Reinholdtsen <pere@hungry.com>, 839071@bugs.debian.org, Eric Desrochers <eric.desrochers@canonical.com>

Subject: Re: Bug#839071: Bug#839071: Patch for bug #839071

Date: Thu, 29 Sep 2016 03:34:32 -0500

On 09/29/2016 03:28 AM, Petter Reinholdtsen wrote:
> [Richard Laager]
>> The *writing* is hard-coded to /etc/mtab, but if it doesn't exist,
>> that's fine. ZFS just won't update /etc/mtab.
> 
> Right.  Perhaps better to use two defines, one MNTTAB_READ and one
> MNTTAB_WRITE to make this more clear, instead of introducing the
> /etc/mtab hardcoding?

Maybe. I had originally hoped to simply remove all the writing, but it
seems that some distros still need that. The current hope is to replace
the writing (and maybe reading) with libmount, but that's a bigger patch.

Strictly speaking, I'm not sure it was necessary in the final patch to
replace MNTTAB with hard-coded /etc/mtab, but I'd have to look at the
whole thing again to be sure.

-- 
Richard

Message #45 received at 839071@bugs.debian.org (full text, mbox, reply):

From: Antonio Russo <antonio.e.russo@gmail.com>

To: 839071@bugs.debian.org

Subject: Patch

Date: Tue, 5 Dec 2017 22:06:20 -0500

The proposed patch has been included upstream, and should be reflected in the current package.

Because mtab has "always" been linked to /proc/self/mounts, I'm not sure if
this report was "theoretical" or represented an observed problem that may
have a different root cause.

Could you please clarify if upstream's response is adequate?

Message #50 received at 839071-done@bugs.debian.org (full text, mbox, reply):

From: Aron Xu <aron@debian.org>

To: 839071-done@bugs.debian.org

Subject: Re: Bug #839071: zfsutils-linux: Issues based on the fact that ZFS relies on mtab.

Date: Mon, 5 Mar 2018 16:45:05 +0800

Source-Version: 0.7.3-1

This issue was addressed in upstream 0.7.0-rc2 release, closing now.

Regards,
Aron

Send a report that this bug log contains spam.