2007-01-23 17:25:00

by Cordenner jean noel

[permalink] [raw]
Subject: [RFC] [patch 3/3] i_version update for ext4: user interface

This patch adds a ``st_i_version'' field in the stat
structure, and modifies the stat(2) syscall accordingly. Currently the
change is only visible on i386 and x86_64 archs.



Signed-off-by: Jean Noel Cordenner <[email protected]>

Index: linux-2.6.20-rc5/fs/stat.c
===================================================================
--- linux-2.6.20-rc5.orig/fs/stat.c 2007-01-19 16:59:15.000000000 +0100
+++ linux-2.6.20-rc5/fs/stat.c 2007-01-23 16:21:17.000000000 +0100
@@ -31,6 +31,7 @@
stat->atime = inode->i_atime;
stat->mtime = inode->i_mtime;
stat->ctime = inode->i_ctime;
+ stat->i_version = inode->i_version;
stat->size = i_size_read(inode);
stat->blocks = inode->i_blocks;
stat->blksize = (1 << inode->i_blkbits);
@@ -226,6 +227,7 @@
tmp.st_atime = stat->atime.tv_sec;
tmp.st_mtime = stat->mtime.tv_sec;
tmp.st_ctime = stat->ctime.tv_sec;
+ tmp.st_i_version = stat->i_version;
#ifdef STAT_HAVE_NSEC
tmp.st_atime_nsec = stat->atime.tv_nsec;
tmp.st_mtime_nsec = stat->mtime.tv_nsec;
@@ -359,6 +361,7 @@
tmp.st_mtime_nsec = stat->mtime.tv_nsec;
tmp.st_ctime = stat->ctime.tv_sec;
tmp.st_ctime_nsec = stat->ctime.tv_nsec;
+ tmp.st_i_version = stat->i_version;
tmp.st_size = stat->size;
tmp.st_blocks = stat->blocks;
tmp.st_blksize = stat->blksize;
Index: linux-2.6.20-rc5/include/linux/stat.h
===================================================================
--- linux-2.6.20-rc5.orig/include/linux/stat.h 2006-11-29
22:57:37.000000000 +0100
+++ linux-2.6.20-rc5/include/linux/stat.h 2007-01-23
16:22:18.000000000 +0100
@@ -68,6 +68,7 @@
struct timespec atime;
struct timespec mtime;
struct timespec ctime;
+ unsigned long i_version;
unsigned long blksize;
unsigned long long blocks;
};
Index: linux-2.6.20-rc5/include/asm-i386/stat.h
===================================================================
--- linux-2.6.20-rc5.orig/include/asm-i386/stat.h 2006-11-29
22:57:37.000000000 +0100
+++ linux-2.6.20-rc5/include/asm-i386/stat.h 2007-01-23
17:48:36.000000000 +0100
@@ -32,7 +32,7 @@
unsigned long st_mtime_nsec;
unsigned long st_ctime;
unsigned long st_ctime_nsec;
- unsigned long __unused4;
+ unsigned long st_i_version;
unsigned long __unused5;
};

@@ -41,7 +41,7 @@
*/
struct stat64 {
unsigned long long st_dev;
- unsigned char __pad0[4];
+ unsigned int st_i_version;

#define STAT64_HAS_BROKEN_ST_INO 1
unsigned long __st_ino;
Index: linux-2.6.20-rc5/include/asm-x86_64/stat.h
===================================================================
--- linux-2.6.20-rc5.orig/include/asm-x86_64/stat.h 2006-11-29
22:57:37.000000000 +0100
+++ linux-2.6.20-rc5/include/asm-x86_64/stat.h 2007-01-23
17:57:03.000000000 +0100
@@ -23,7 +23,8 @@
unsigned long st_mtime_nsec;
unsigned long st_ctime;
unsigned long st_ctime_nsec;
- long __unused[3];
+ unsigned long st_i_version;
+ long __unused[2];
};

/* For 32bit emulation */


2007-01-23 18:52:08

by Andreas Dilger

[permalink] [raw]
Subject: Re: [RFC] [patch 3/3] i_version update for ext4: user interface

On Jan 23, 2007 18:24 +0100, Cordenner jean noel wrote:
> This patch adds a ``st_i_version'' field in the stat
> structure, and modifies the stat(2) syscall accordingly. Currently the
> change is only visible on i386 and x86_64 archs.

What is the need for exporting i_version to userspace?

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

2007-01-24 18:14:17

by Bruce Fields

[permalink] [raw]
Subject: Re: [RFC] [patch 3/3] i_version update for ext4: user interface

On Tue, Jan 23, 2007 at 11:52:08AM -0700, Andreas Dilger wrote:
> On Jan 23, 2007 18:24 +0100, Cordenner jean noel wrote:
> > This patch adds a ``st_i_version'' field in the stat
> > structure, and modifies the stat(2) syscall accordingly. Currently the
> > change is only visible on i386 and x86_64 archs.
>
> What is the need for exporting i_version to userspace?

Maybe none, but it seems potentially really useful to me; in practice
aren't most uses of mtime/ctime exactly in order to find out if the file
has changed recently?

Who else is asking for finer time resolutions? Is it because they need
the precision (or the inter-file comparison) or because they need to
know whether a file has changed?

Isn't anybody that currently maintains some cached information about a
filesystem currently living with the same races as NFS?

--b.

2007-01-24 18:17:27

by Peter Staubach

[permalink] [raw]
Subject: Re: [RFC] [patch 3/3] i_version update for ext4: user interface

J. Bruce Fields wrote:
> On Tue, Jan 23, 2007 at 11:52:08AM -0700, Andreas Dilger wrote:
>
>> On Jan 23, 2007 18:24 +0100, Cordenner jean noel wrote:
>>
>>> This patch adds a ``st_i_version'' field in the stat
>>> structure, and modifies the stat(2) syscall accordingly. Currently the
>>> change is only visible on i386 and x86_64 archs.
>>>
>> What is the need for exporting i_version to userspace?
>>
>
> Maybe none, but it seems potentially really useful to me; in practice
> aren't most uses of mtime/ctime exactly in order to find out if the file
> has changed recently?
>
> Who else is asking for finer time resolutions? Is it because they need
> the precision (or the inter-file comparison) or because they need to
> know whether a file has changed?
>
> Isn't anybody that currently maintains some cached information about a
> filesystem currently living with the same races as NFS?

Doesn't the BSD stat struct have a field like this? If so, perhaps
they can remember why it was deemed to be interesting?

Thanx...

ps