[PATCH] hostfs: Use noop_fsync for directories

[PATCH] hostfs: Use noop_fsync for directories

[Richard Weinberger]

Daniel reported that dpkg(1) dies if the root filesystem is a hostfs
because it does not expect fsync(2) to fail with EINVAL on directories.
While fsync(2) is allowed to fail with EINVAL if the filesystem does not
support it we can do better and use noop_fsync() to not confuse userspace
further.

Cc: stable@vger.kernel.org
Reported-and-tested-by: Daniel Gröber <dxld@darkboxed.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/hostfs/hostfs_kern.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/hostfs/hostfs_kern.c b/fs/hostfs/hostfs_kern.c
index fd62cae..a7ac856 100644
--- a/fs/hostfs/hostfs_kern.c
+++ b/fs/hostfs/hostfs_kern.c
@@ -393,6 +393,7 @@ static const struct file_operations hostfs_dir_fops = {
 	.llseek		= generic_file_llseek,
 	.iterate	= hostfs_readdir,
 	.read		= generic_read_dir,
+	.fsync		= noop_fsync,
 };
 
 static int hostfs_writepage(struct page *page, struct writeback_control *wbc)
-- 
1.8.4.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[PATCH] hostfs: Use noop_fsync for directories

[Richard Weinberger]

Daniel reported that dpkg(1) dies if the root filesystem is a hostfs
because it does not expect fsync(2) to fail with EINVAL on directories.
While fsync(2) is allowed to fail with EINVAL if the filesystem does not
support it we can do better and use noop_fsync() to not confuse userspace
further.

Cc: stable@vger.kernel.org
Reported-and-tested-by: Daniel Gröber <dxld@darkboxed.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/hostfs/hostfs_kern.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/hostfs/hostfs_kern.c b/fs/hostfs/hostfs_kern.c
index fd62cae..a7ac856 100644
--- a/fs/hostfs/hostfs_kern.c
+++ b/fs/hostfs/hostfs_kern.c
@@ -393,6 +393,7 @@ static const struct file_operations hostfs_dir_fops = {
 	.llseek		= generic_file_llseek,
 	.iterate	= hostfs_readdir,
 	.read		= generic_read_dir,
+	.fsync		= noop_fsync,
 };
 
 static int hostfs_writepage(struct page *page, struct writeback_control *wbc)
-- 
1.8.4.5

Re: [uml-devel] [PATCH] hostfs: Use noop_fsync for directories

[Christoph Hellwig]

On Tue, Jan 13, 2015 at 11:15:58PM +0100, Richard Weinberger wrote:
> Daniel reported that dpkg(1) dies if the root filesystem is a hostfs
> because it does not expect fsync(2) to fail with EINVAL on directories.
> While fsync(2) is allowed to fail with EINVAL if the filesystem does not
> support it we can do better and use noop_fsync() to not confuse userspace
> further.

Shouldn't hostfs pass the fsync through to the host filesystem?

------------------------------------------------------------------------------
New Year. New Location. New Benefits. New Data Center in Ashburn, VA.
GigeNET is offering a free month of service with a new server in Ashburn.
Choose from 2 high performing configs, both with 100TB of bandwidth.
Higher redundancy.Lower latency.Increased capacity.Completely compliant.
http://p.sf.net/sfu/gigenet
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [PATCH] hostfs: Use noop_fsync for directories

[Christoph Hellwig]

On Tue, Jan 13, 2015 at 11:15:58PM +0100, Richard Weinberger wrote:
> Daniel reported that dpkg(1) dies if the root filesystem is a hostfs
> because it does not expect fsync(2) to fail with EINVAL on directories.
> While fsync(2) is allowed to fail with EINVAL if the filesystem does not
> support it we can do better and use noop_fsync() to not confuse userspace
> further.

Shouldn't hostfs pass the fsync through to the host filesystem?

Re: [PATCH] hostfs: Use noop_fsync for directories

[Richard Weinberger]

Am 13.01.2015 um 23:19 schrieb Christoph Hellwig:
> On Tue, Jan 13, 2015 at 11:15:58PM +0100, Richard Weinberger wrote:
>> Daniel reported that dpkg(1) dies if the root filesystem is a hostfs
>> because it does not expect fsync(2) to fail with EINVAL on directories.
>> While fsync(2) is allowed to fail with EINVAL if the filesystem does not
>> support it we can do better and use noop_fsync() to not confuse userspace
>> further.
> 
> Shouldn't hostfs pass the fsync through to the host filesystem?

hostfs tries do reduce the amount of syscall between guest and host as much
as possible. For file operations it passes everything down to the host but
for directory operations only ->iterate() does.

It is already horrible slow, if we add an ->open() for directory too it would
get even more slower. :-(

Thanks,
//richard

Re: [uml-devel] [PATCH] hostfs: Use noop_fsync for directories

[Christoph Hellwig]

On Tue, Jan 13, 2015 at 11:26:38PM +0100, Richard Weinberger wrote:
> hostfs tries do reduce the amount of syscall between guest and host as much
> as possible. For file operations it passes everything down to the host but
> for directory operations only ->iterate() does.
> 
> It is already horrible slow, if we add an ->open() for directory too it would
> get even more slower. :-(

This sounds fairlt dangerous.  At least add some good documentation
explaining these semantics.


------------------------------------------------------------------------------
New Year. New Location. New Benefits. New Data Center in Ashburn, VA.
GigeNET is offering a free month of service with a new server in Ashburn.
Choose from 2 high performing configs, both with 100TB of bandwidth.
Higher redundancy.Lower latency.Increased capacity.Completely compliant.
http://p.sf.net/sfu/gigenet
_______________________________________________
User-mode-linux-devel mailing list
User-mode-linux-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel

Re: [PATCH] hostfs: Use noop_fsync for directories

[Christoph Hellwig]

On Tue, Jan 13, 2015 at 11:26:38PM +0100, Richard Weinberger wrote:
> hostfs tries do reduce the amount of syscall between guest and host as much
> as possible. For file operations it passes everything down to the host but
> for directory operations only ->iterate() does.
> 
> It is already horrible slow, if we add an ->open() for directory too it would
> get even more slower. :-(

This sounds fairlt dangerous.  At least add some good documentation
explaining these semantics.

Re: [PATCH] hostfs: Use noop_fsync for directories

[Richard Weinberger]

Am 14.01.2015 um 09:39 schrieb Christoph Hellwig:
> On Tue, Jan 13, 2015 at 11:26:38PM +0100, Richard Weinberger wrote:
>> hostfs tries do reduce the amount of syscall between guest and host as much
>> as possible. For file operations it passes everything down to the host but
>> for directory operations only ->iterate() does.
>>
>> It is already horrible slow, if we add an ->open() for directory too it would
>> get even more slower. :-(
> 
> This sounds fairlt dangerous.  At least add some good documentation
> explaining these semantics.

Understood. Maybe it is time to rebenchmark hostfs with full directory pass-through support.
Daniel, are you interested in a small kernel project?
As explained on IRC adding real support for directory fsync() should be an easy task.
The only interesting point is how much overhead it will add.

Thanks,
//richard

Re: [PATCH] hostfs: Use noop_fsync for directories

[Daniel Gröber]

[-- Attachment #1: Type: Text/Plain, Size: 337 bytes --]

From: Richard Weinberger <richard@nod.at>
Subject: Re: [PATCH] hostfs: Use noop_fsync for directories
Date: Wed, 14 Jan 2015 09:43:53 +0100

> Daniel, are you interested in a small kernel project?

Sure, I'm just pretty busy with university right now, I'll probably
pick this back up after exams are finished in a month or so.

--Daniel

[-- Attachment #2: Type: application/pgp-signature, Size: 819 bytes --]