2010-10-19 06:34:36

by Namhyung Kim

Subject: [PATCH] ext3: Check return value of sb_getblk()

Check return value of sb_getblk() is NULL. unlikely is addded here
since it is called from a loop and we've been OK without the check
until now.

Signed-off-by: Namhyung Kim <[email protected]>
---
fs/ext3/inode.c | 6 ++++++
1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/fs/ext3/inode.c b/fs/ext3/inode.c
index ef1c23a..a7ac778 100644
--- a/fs/ext3/inode.c
+++ b/fs/ext3/inode.c
@@ -655,6 +655,12 @@ static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
* parent to disk.
*/
bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
+ if (unlikely(!bh)) {
+ n--;
+ err = -ENOMEM;
+ goto failed;
+ }
+
branch[n].bh = bh;
lock_buffer(bh);
BUFFER_TRACE(bh, "call get_create_access");
--
1.7.0.4

2010-10-19 07:57:36

by Namhyung Kim

Subject: Re: [PATCH] ext3: Check return value of sb_getblk()

2010-10-19 (화), 15:34 +0900, Namhyung Kim:
> Check return value of sb_getblk() is NULL. unlikely is addded here
> since it is called from a loop and we've been OK without the check
> until now.
>
> Signed-off-by: Namhyung Kim <[email protected]>
> ---
> fs/ext3/inode.c | 6 ++++++
> 1 files changed, 6 insertions(+), 0 deletions(-)
>
> diff --git a/fs/ext3/inode.c b/fs/ext3/inode.c
> index ef1c23a..a7ac778 100644
> --- a/fs/ext3/inode.c
> +++ b/fs/ext3/inode.c
> @@ -655,6 +655,12 @@ static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
> * parent to disk.
> */
> bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
> + if (unlikely(!bh)) {
> + n--;
> + err = -ENOMEM;
> + goto failed;
> + }
> +
> branch[n].bh = bh;
> lock_buffer(bh);
> BUFFER_TRACE(bh, "call get_create_access");

Maybe EIO would be more proper error code, I guess.


--
Regards,
Namhyung Kim

2010-10-19 20:42:27

by Jan Kara

Subject: Re: [PATCH] ext3: Check return value of sb_getblk()

On Tue 19-10-10 16:57:29, Namhyung Kim wrote:
> 2010-10-19 (화), 15:34 +0900, Namhyung Kim:
> > Check return value of sb_getblk() is NULL. unlikely is addded here
> > since it is called from a loop and we've been OK without the check
> > until now.
> >
> > Signed-off-by: Namhyung Kim <[email protected]>
> > ---
> > fs/ext3/inode.c | 6 ++++++
> > 1 files changed, 6 insertions(+), 0 deletions(-)
> >
> > diff --git a/fs/ext3/inode.c b/fs/ext3/inode.c
> > index ef1c23a..a7ac778 100644
> > --- a/fs/ext3/inode.c
> > +++ b/fs/ext3/inode.c
> > @@ -655,6 +655,12 @@ static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
> > * parent to disk.
> > */
> > bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
> > + if (unlikely(!bh)) {
> > + n--;
> > + err = -ENOMEM;
> > + goto failed;
> > + }
> > +
> > branch[n].bh = bh;
> > lock_buffer(bh);
> > BUFFER_TRACE(bh, "call get_create_access");
>
> Maybe EIO would be more proper error code, I guess.
Hmm, for ext3, sb_getblk() cannot really realistically fail. Block
numbers are 32-bit so they can never be big enough to overflow pagecache
index and grow_buffers() loops indefinitely if it cannot allocate buffers.
But OK, EIO might be a reasonable return value and we can have the check
there just in case sb_getblk() grows some other possiblility to fail.

Honza

--
Jan Kara <[email protected]>
SUSE Labs, CR