Re: [patch 3/5] fuse: fix race in llseek

[Miklos Szeredi]

> > From: Miklos Szeredi <mszeredi@xxxxxxx>
> > 
> > Fuse doesn't use i_mutex to protect setting i_size, and so
> > generic_file_llseek() can be racy: it doesn't use i_size_read().
> > 
> > So do a fuse specific llseek method, which does use i_size_read().
> > 
> 
> Is there any specific reason why we don't use i_size_read() in 
> generic_file_llseek()?

Just to save some cycles on 32bit archs.  Considering that
lseek(SEEK_END) isn't a very performance sensitive operation...

> If no, why don't you fix it instead?

...we could do that.  Does anybody object?

Miklos
----


Some filesystems (e.g. fuse) don't use i_mutex to protect setting
i_size, and so generic_file_llseek() can be racy.

Since lseek(SEEK_END) isn't a very performance sensitive operation, do
this in generic_file_llseek(), instead of requiring such filesystems
to duplicate this function.

Signed-off-by: Miklos Szeredi <mszeredi@xxxxxxx>
---
 fs/read_write.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Index: linux/fs/read_write.c
===================================================================
--- linux.orig/fs/read_write.c	2008-04-17 04:49:44.000000000 +0200
+++ linux/fs/read_write.c	2008-04-28 17:08:25.000000000 +0200
@@ -39,7 +39,11 @@ loff_t generic_file_llseek(struct file *
 	mutex_lock(&inode->i_mutex);
 	switch (origin) {
 		case SEEK_END:
-			offset += inode->i_size;
+			/*
+			 * Some filesystems use a different lock as i_mutex
+			 * to protect writing i_size
+			 */
+			offset += i_size_read(inode);
 			break;
 		case SEEK_CUR:
 			offset += file->f_pos;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/