[PATCH] fat: Fix non-atomic NFS i_pos read

[PATCH] fat: Fix non-atomic NFS i_pos read

[Steven J. Magnani]

fat_encode_fh() can fetch an invalid i_pos value on systems
where 64-bit accesses are not atomic. Make it use the same
accessor as the rest of the FAT code.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
---
diff -uprN linux-3.5-rc4/fs/fat/inode.c new/fs/fat/inode.c
--- linux-3.5-rc4/fs/fat/inode.c	2012-06-29 11:20:12.781348652 -0500
+++ new/fs/fat/inode.c	2012-06-29 11:25:29.484713183 -0500
@@ -738,22 +738,22 @@ static int
 fat_encode_fh(struct inode *inode, __u32 *fh, int *lenp, struct inode *parent)
 {
 	int len = *lenp;
-	u32 ipos_h, ipos_m, ipos_l;
+	struct super_block *sb = inode->i_sb;
+	struct msdos_sb_info *sbi = MSDOS_SB(sb);
+	loff_t i_pos;
 
 	if (len < 5) {
 		*lenp = 5;
 		return 255; /* no room */
 	}
 
-	ipos_h = MSDOS_I(inode)->i_pos >> 8;
-	ipos_m = (MSDOS_I(inode)->i_pos & 0xf0) << 24;
-	ipos_l = (MSDOS_I(inode)->i_pos & 0x0f) << 28;
+	i_pos = fat_i_pos_read(sbi, inode);
 	*lenp = 5;
 	fh[0] = inode->i_ino;
 	fh[1] = inode->i_generation;
-	fh[2] = ipos_h;
-	fh[3] = ipos_m | MSDOS_I(inode)->i_logstart;
-	fh[4] = ipos_l;
+	fh[2] = i_pos >> 8;
+	fh[3] = ((i_pos & 0xf0) << 24) | MSDOS_I(inode)->i_logstart;
+	fh[4] = (i_pos & 0x0f) << 28;
 	if (parent)
 		fh[4] |= MSDOS_I(parent)->i_logstart;
 	return 3;

Re: [PATCH] fat: Fix non-atomic NFS i_pos read

[OGAWA Hirofumi]

"Steven J. Magnani" <steve@digidescorp.com> writes:

> fat_encode_fh() can fetch an invalid i_pos value on systems
> where 64-bit accesses are not atomic. Make it use the same
> accessor as the rest of the FAT code.
>
> Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
> ---
> diff -uprN linux-3.5-rc4/fs/fat/inode.c new/fs/fat/inode.c
> --- linux-3.5-rc4/fs/fat/inode.c	2012-06-29 11:20:12.781348652 -0500
> +++ new/fs/fat/inode.c	2012-06-29 11:25:29.484713183 -0500
> @@ -738,22 +738,22 @@ static int
>  fat_encode_fh(struct inode *inode, __u32 *fh, int *lenp, struct inode *parent)
>  {
>  	int len = *lenp;
> -	u32 ipos_h, ipos_m, ipos_l;
> +	struct super_block *sb = inode->i_sb;
> +	struct msdos_sb_info *sbi = MSDOS_SB(sb);

sb seems to be unused. So, we can just

	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);

true? Otherwise, ack.

Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Just curious, this happened on the real system? I recently heard about
NFS issue.

Thanks.

> +	loff_t i_pos;
>  
>  	if (len < 5) {
>  		*lenp = 5;
>  		return 255; /* no room */
>  	}
>  
> -	ipos_h = MSDOS_I(inode)->i_pos >> 8;
> -	ipos_m = (MSDOS_I(inode)->i_pos & 0xf0) << 24;
> -	ipos_l = (MSDOS_I(inode)->i_pos & 0x0f) << 28;
> +	i_pos = fat_i_pos_read(sbi, inode);
>  	*lenp = 5;
>  	fh[0] = inode->i_ino;
>  	fh[1] = inode->i_generation;
> -	fh[2] = ipos_h;
> -	fh[3] = ipos_m | MSDOS_I(inode)->i_logstart;
> -	fh[4] = ipos_l;
> +	fh[2] = i_pos >> 8;
> +	fh[3] = ((i_pos & 0xf0) << 24) | MSDOS_I(inode)->i_logstart;
> +	fh[4] = (i_pos & 0x0f) << 28;
>  	if (parent)
>  		fh[4] |= MSDOS_I(parent)->i_logstart;
>  	return 3;
>

-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Re: [PATCH] fat: Fix non-atomic NFS i_pos read

[Steven J. Magnani]

On Sat, 2012-06-30 at 03:06 +0900, OGAWA Hirofumi wrote: 
> "Steven J. Magnani" <steve@digidescorp.com> writes:
> 
> > fat_encode_fh() can fetch an invalid i_pos value on systems
> > where 64-bit accesses are not atomic. Make it use the same
> > accessor as the rest of the FAT code.
> >
> > Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
> > ---
> > diff -uprN linux-3.5-rc4/fs/fat/inode.c new/fs/fat/inode.c
> > --- linux-3.5-rc4/fs/fat/inode.c	2012-06-29 11:20:12.781348652 -0500
> > +++ new/fs/fat/inode.c	2012-06-29 11:25:29.484713183 -0500
> > @@ -738,22 +738,22 @@ static int
> >  fat_encode_fh(struct inode *inode, __u32 *fh, int *lenp, struct inode *parent)
> >  {
> >  	int len = *lenp;
> > -	u32 ipos_h, ipos_m, ipos_l;
> > +	struct super_block *sb = inode->i_sb;
> > +	struct msdos_sb_info *sbi = MSDOS_SB(sb);
> 
> sb seems to be unused. So, we can just
> 
> 	struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
> 
> true? Otherwise, ack.
OK, I'll repost with that change.

> 
> Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
> 
> Just curious, this happened on the real system? I recently heard about
> NFS issue.
Not that I'm aware of. Can you point me to any discussion of the NFS
issue? I've spent a lot of time recently studying the FAT NFS
implementation and come to the conclusion that it is extremely
vulnerable to inode eviction.  More to follow...

Regards,
Steve

Re: [PATCH] fat: Fix non-atomic NFS i_pos read

[OGAWA Hirofumi]

"Steven J. Magnani" <steve@digidescorp.com> writes:

>> Just curious, this happened on the real system? I recently heard about
>> NFS issue.
> Not that I'm aware of. Can you point me to any discussion of the NFS
> issue? I've spent a lot of time recently studying the FAT NFS
> implementation and come to the conclusion that it is extremely
> vulnerable to inode eviction.  More to follow...

It is true, I also know NFS support of FAT is broken, but I'm not sure
if it was read-only.  Sorry, recent report was by private email, not
public.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>