[RFC][PATCH] XFS: memory leak in xfs_inactive() - is xfs_trans_free() enough or do we need xfs_trans_cancel() ?

[RFC][PATCH] XFS: memory leak in xfs_inactive() - is xfs_trans_free() enough or do we need xfs_trans_cancel() ?

[Jesper Juhl]

Hi,

The Coverity checker found a memory leak in xfs_inactive().

The offending code is this bit : 

1671 		tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);

At conditional (1): "truncate != 0" taking true path

1672 		if (truncate) {
1673 			/*
1674 			 * Do the xfs_itruncate_start() call before
1675 			 * reserving any log space because itruncate_start
1676 			 * will call into the buffer cache and we can't
1677 			 * do that within a transaction.
1678 			 */
1679 			xfs_ilock(ip, XFS_IOLOCK_EXCL);
1680 	
1681 			error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);

At conditional (2): "error != 0" taking true path

1682 			if (error) {
1683 				xfs_iunlock(ip, XFS_IOLOCK_EXCL);

Event leaked_storage: Returned without freeing storage "tp"
Also see events: [alloc_fn][var_assign]

1684 				return VN_INACTIVE_CACHE;
1685 			}

So, the code allocates a transaction, but in the case where 'truncate' is !=0 and xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); happens to return an error, we'll just return from the function without dealing with the memory allocated byxfs_trans_alloc() and assigned to 'tp', thus it'll be orphaned/leaked - not good.

What I'm wondering is this; is it enough, at this point, to call xfs_trans_free(tp); (it would seem to me that would be OK, but I'm not intimite with this code) or do we need a full xfs_trans_cancel(tp, 0);  ???

In case I'm right and xfs_trans_free(tp); is all we need, then please consider the patch below. Otherwise please NACK the patch and I'll cook up another one :-)


Fix memory leak on error in xfs_inactive().

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
--- 
 fs/xfs/xfs_vnodeops.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index de17aed..e0d3d51 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -1681,6 +1681,7 @@ xfs_inactive(
 		error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
 		if (error) {
 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+			xfs_trans_free(tp);
 			return VN_INACTIVE_CACHE;
 		}
 


-- 
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html

Re: [RFC][PATCH] XFS: memory leak in xfs_inactive() - is xfs_trans_free() enough or do we need xfs_trans_cancel() ?

[David Chinner]

On Wed, May 16, 2007 at 11:31:16PM +0200, Jesper Juhl wrote:
> Hi,
> 
> The Coverity checker found a memory leak in xfs_inactive().
....
> So, the code allocates a transaction, but in the case where 'truncate' is
> !=0 and xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); happens to return
> an error, we'll just return from the function without dealing with the
> memory allocated byxfs_trans_alloc() and assigned to 'tp', thus it'll be
> orphaned/leaked - not good.

Yeah, introduced by:

http://git2.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d3cf209476b72c83907a412b6708c5e498410aa7

Thanks for reporting the problem, Jesper.

> What I'm wondering is this; is it enough, at this point, to call
> xfs_trans_free(tp); (it would seem to me that would be OK, but I'm not
> intimite with this code) or do we need a full xfs_trans_cancel(tp, 0);  ???

xfs_trans_free() is not supposed to be called by anything but the transaction
code (it's static). So a xfs_trans_cancel() would need to be issued.

> In case I'm right and xfs_trans_free(tp); is all we need, then please
> consider the patch below. Otherwise please NACK the patch and I'll cook up
> another one :-)

NACK ;)

xfs_trans_cancel() is needed. Patch below.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

---
 fs/xfs/xfs_vnodeops.c |    1 +
 1 file changed, 1 insertion(+)

Index: 2.6.x-xfs-new/fs/xfs/xfs_vnodeops.c
===================================================================
--- 2.6.x-xfs-new.orig/fs/xfs/xfs_vnodeops.c	2007-05-11 16:04:03.000000000 +1000
+++ 2.6.x-xfs-new/fs/xfs/xfs_vnodeops.c	2007-05-17 12:37:25.671399078 +1000
@@ -1710,6 +1710,7 @@ xfs_inactive(
 
 		error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
 		if (error) {
+			xfs_trans_cancel(tp, 0);
 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 			return VN_INACTIVE_CACHE;
 		}

Re: [RFC][PATCH] XFS: memory leak in xfs_inactive() - is xfs_trans_free() enough or do we need xfs_trans_cancel() ?

[Jesper Juhl]

On Thursday 17 May 2007 04:40:24 David Chinner wrote:
> On Wed, May 16, 2007 at 11:31:16PM +0200, Jesper Juhl wrote:
> > Hi,
> > 
> > The Coverity checker found a memory leak in xfs_inactive().
> ....
> > So, the code allocates a transaction, but in the case where 'truncate' is
> > !=0 and xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); happens to return
> > an error, we'll just return from the function without dealing with the
> > memory allocated byxfs_trans_alloc() and assigned to 'tp', thus it'll be
> > orphaned/leaked - not good.
> 
> Yeah, introduced by:
> 
> http://git2.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d3cf209476b72c83907a412b6708c5e498410aa7
> 
> Thanks for reporting the problem, Jesper.
> 
You are welcome.

That commit introduces the same problem in xfs_inactive_free_eofblocks(). 
Patch to fix it below.

> > What I'm wondering is this; is it enough, at this point, to call
> > xfs_trans_free(tp); (it would seem to me that would be OK, but I'm not
> > intimite with this code) or do we need a full xfs_trans_cancel(tp, 0);  ???
> 
> xfs_trans_free() is not supposed to be called by anything but the transaction
> code (it's static). So a xfs_trans_cancel() would need to be issued.
> 
Makes sense. Thanks. I completely missed the static nature :-/



Fix XFS memory leak; allocated transaction not freed in xfs_inactive_free_eofblocks() in failure case.

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
--- 
 fs/xfs/xfs_vnodeops.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index de17aed..32519cf 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -1260,6 +1260,7 @@ xfs_inactive_free_eofblocks(
 		error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
 				    ip->i_size);
 		if (error) {
+			xfs_trans_cancel(tp, 0);
 			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 			return error;
 		}

Re: [RFC][PATCH] XFS: memory leak in xfs_inactive() - is xfs_trans_free() enough or do we need xfs_trans_cancel() ?

[Jesper Juhl]

Any chance the patches below that fix two mem leaks in XFS will make
it in in time for 2.6.22? I believe they should...

On 18/05/07, Jesper Juhl <jesper.juhl@gmail.com> wrote:
> On Thursday 17 May 2007 04:40:24 David Chinner wrote:
> > On Wed, May 16, 2007 at 11:31:16PM +0200, Jesper Juhl wrote:
> > > Hi,
> > >
> > > The Coverity checker found a memory leak in xfs_inactive().
> > ....
> > > So, the code allocates a transaction, but in the case where 'truncate' is
> > > !=0 and xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); happens to return
> > > an error, we'll just return from the function without dealing with the
> > > memory allocated byxfs_trans_alloc() and assigned to 'tp', thus it'll be
> > > orphaned/leaked - not good.
> >
> > Yeah, introduced by:
> >
> > http://git2.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d3cf209476b72c83907a412b6708c5e498410aa7
> >
> > Thanks for reporting the problem, Jesper.
> >
> You are welcome.
>
> That commit introduces the same problem in xfs_inactive_free_eofblocks().
> Patch to fix it below.
>
> > > What I'm wondering is this; is it enough, at this point, to call
> > > xfs_trans_free(tp); (it would seem to me that would be OK, but I'm not
> > > intimite with this code) or do we need a full xfs_trans_cancel(tp, 0);  ???
> >
> > xfs_trans_free() is not supposed to be called by anything but the transaction
> > code (it's static). So a xfs_trans_cancel() would need to be issued.
> >
> Makes sense. Thanks. I completely missed the static nature :-/
>
>
>
> Fix XFS memory leak; allocated transaction not freed in xfs_inactive_free_eofblocks() in failure case.
>
> Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
> ---
>  fs/xfs/xfs_vnodeops.c |    1 +
>  1 files changed, 1 insertions(+), 0 deletions(-)
>
> diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
> index de17aed..32519cf 100644
> --- a/fs/xfs/xfs_vnodeops.c
> +++ b/fs/xfs/xfs_vnodeops.c
> @@ -1260,6 +1260,7 @@ xfs_inactive_free_eofblocks(
>                 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
>                                     ip->i_size);
>                 if (error) {
> +                       xfs_trans_cancel(tp, 0);
>                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
>                         return error;
>                 }
>
>

-- 
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html