[PATCH v3 1/1] xdp: avoid calling kfree twice

[PATCH v3 1/1] xdp: avoid calling kfree twice

[Zhu Yanjun]

In the function xdp_umem_pin_pages, if npgs != umem->npgs and
npgs >= 0, the function xdp_umem_unpin_pages is called. In this
function, kfree is called to handle umem->pgs, and then in the
function xdp_umem_pin_pages, kfree is called again to handle
umem->pgs. Eventually, to umem->pgs, kfree is called twice.

Since umem->pgs is set to NULL after the first kfree, the second
kfree would not trigger call trace.

Fixes: c0c77d8fb787 ("xsk: add user memory registration support sockopt")
CC: Ye Dong <dong.ye@intel.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@intel.com>
---
 net/xdp/xdp_umem.c | 17 +++++------------
 1 file changed, 5 insertions(+), 12 deletions(-)

diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index 56a28a686988..01b31c56cead 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -97,7 +97,6 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
 {
 	unsigned int gup_flags = FOLL_WRITE;
 	long npgs;
-	int err;
 
 	umem->pgs = kcalloc(umem->npgs, sizeof(*umem->pgs),
 			    GFP_KERNEL | __GFP_NOWARN);
@@ -112,20 +111,14 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
 	if (npgs != umem->npgs) {
 		if (npgs >= 0) {
 			umem->npgs = npgs;
-			err = -ENOMEM;
-			goto out_pin;
+			xdp_umem_unpin_pages(umem);
+			return -ENOMEM;
 		}
-		err = npgs;
-		goto out_pgs;
+		kfree(umem->pgs);
+		umem->pgs = NULL;
+		return (int)npgs;
 	}
 	return 0;
-
-out_pin:
-	xdp_umem_unpin_pages(umem);
-out_pgs:
-	kfree(umem->pgs);
-	umem->pgs = NULL;
-	return err;
 }
 
 static int xdp_umem_account_pages(struct xdp_umem *umem)
-- 
2.18.4

Re: [PATCH v3 1/1] xdp: avoid calling kfree twice

[Daniel Borkmann]

On 12/11/20 5:26 AM, Zhu Yanjun wrote:
> In the function xdp_umem_pin_pages, if npgs != umem->npgs and
> npgs >= 0, the function xdp_umem_unpin_pages is called. In this
> function, kfree is called to handle umem->pgs, and then in the
> function xdp_umem_pin_pages, kfree is called again to handle
> umem->pgs. Eventually, to umem->pgs, kfree is called twice.
> 
> Since umem->pgs is set to NULL after the first kfree, the second
> kfree would not trigger call trace.

This can still be misinterpreted imho; maybe lets simplify, for example:

   [bpf-next] xdp: avoid unnecessary second call to kfree

   For the case when in xdp_umem_pin_pages() the call to pin_user_pages()
   wasn't able to pin all the requested number of pages in memory (but some)
   then we error out by cleaning up the ones that got pinned through a call
   to xdp_umem_unpin_pages() and later on we free kfree(umem->pgs) itself.

   This is unneeded since xdp_umem_unpin_pages() itself already does the
   kfree(umem->pgs) internally with subsequent setting umem->pgs to NULL, so
   that in xdp_umem_pin_pages() the second kfree(umem->pgs) becomes entirely
   unnecessary for this case. Therefore, clean the error handling up.

> Fixes: c0c77d8fb787 ("xsk: add user memory registration support sockopt")

Why do we need a Fixes tag here? It's a _cleanup_, not a bug/fix technically.

> CC: Ye Dong <dong.ye@intel.com>
> Acked-by: Björn Töpel <bjorn.topel@intel.com>
> Signed-off-by: Zhu Yanjun <yanjun.zhu@intel.com>
> ---
>   net/xdp/xdp_umem.c | 17 +++++------------
>   1 file changed, 5 insertions(+), 12 deletions(-)
> 
> diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
> index 56a28a686988..01b31c56cead 100644
> --- a/net/xdp/xdp_umem.c
> +++ b/net/xdp/xdp_umem.c
> @@ -97,7 +97,6 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
>   {
>   	unsigned int gup_flags = FOLL_WRITE;
>   	long npgs;
> -	int err;
>   
>   	umem->pgs = kcalloc(umem->npgs, sizeof(*umem->pgs),
>   			    GFP_KERNEL | __GFP_NOWARN);
> @@ -112,20 +111,14 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
>   	if (npgs != umem->npgs) {
>   		if (npgs >= 0) {
>   			umem->npgs = npgs;
> -			err = -ENOMEM;
> -			goto out_pin;
> +			xdp_umem_unpin_pages(umem);
> +			return -ENOMEM;
>   		}
> -		err = npgs;
> -		goto out_pgs;
> +		kfree(umem->pgs);
> +		umem->pgs = NULL;
> +		return (int)npgs;
>   	}
>   	return 0;
> -
> -out_pin:
> -	xdp_umem_unpin_pages(umem);
> -out_pgs:
> -	kfree(umem->pgs);
> -	umem->pgs = NULL;
> -	return err;
>   }
>   
>   static int xdp_umem_account_pages(struct xdp_umem *umem)
> 

While at it, maybe we could also simplify the if (npgs != umem->npgs) a bit to
get rid of the indent, something like:

diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index 56a28a686988..fa4dd16cced5 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -97,7 +97,6 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
  {
  	unsigned int gup_flags = FOLL_WRITE;
  	long npgs;
-	int err;

  	umem->pgs = kcalloc(umem->npgs, sizeof(*umem->pgs),
  			    GFP_KERNEL | __GFP_NOWARN);
@@ -108,24 +107,17 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
  	npgs = pin_user_pages(address, umem->npgs,
  			      gup_flags | FOLL_LONGTERM, &umem->pgs[0], NULL);
  	mmap_read_unlock(current->mm);
-
-	if (npgs != umem->npgs) {
-		if (npgs >= 0) {
-			umem->npgs = npgs;
-			err = -ENOMEM;
-			goto out_pin;
-		}
-		err = npgs;
-		goto out_pgs;
+	if (npgs == umem->npgs)
+		return 0;
+	if (npgs >= 0) {
+		umem->npgs = npgs;
+		xdp_umem_unpin_pages(umem);
+		return -ENOMEM;
  	}
-	return 0;

-out_pin:
-	xdp_umem_unpin_pages(umem);
-out_pgs:
  	kfree(umem->pgs);
  	umem->pgs = NULL;
-	return err;
+	return npgs;
  }

  static int xdp_umem_account_pages(struct xdp_umem *umem)

Re: [PATCH v3 1/1] xdp: avoid calling kfree twice

[Rain River]

On Fri, Dec 11, 2020 at 11:53 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
>
> On 12/11/20 5:26 AM, Zhu Yanjun wrote:
> > In the function xdp_umem_pin_pages, if npgs != umem->npgs and
> > npgs >= 0, the function xdp_umem_unpin_pages is called. In this
> > function, kfree is called to handle umem->pgs, and then in the
> > function xdp_umem_pin_pages, kfree is called again to handle
> > umem->pgs. Eventually, to umem->pgs, kfree is called twice.
> >
> > Since umem->pgs is set to NULL after the first kfree, the second
> > kfree would not trigger call trace.
>
> This can still be misinterpreted imho; maybe lets simplify, for example:
>
>    [bpf-next] xdp: avoid unnecessary second call to kfree
>
>    For the case when in xdp_umem_pin_pages() the call to pin_user_pages()
>    wasn't able to pin all the requested number of pages in memory (but some)
>    then we error out by cleaning up the ones that got pinned through a call
>    to xdp_umem_unpin_pages() and later on we free kfree(umem->pgs) itself.
>
>    This is unneeded since xdp_umem_unpin_pages() itself already does the
>    kfree(umem->pgs) internally with subsequent setting umem->pgs to NULL, so
>    that in xdp_umem_pin_pages() the second kfree(umem->pgs) becomes entirely
>    unnecessary for this case. Therefore, clean the error handling up.
>
> > Fixes: c0c77d8fb787 ("xsk: add user memory registration support sockopt")
>
> Why do we need a Fixes tag here? It's a _cleanup_, not a bug/fix technically.
>
> > CC: Ye Dong <dong.ye@intel.com>
> > Acked-by: Björn Töpel <bjorn.topel@intel.com>
> > Signed-off-by: Zhu Yanjun <yanjun.zhu@intel.com>
> > ---
> >   net/xdp/xdp_umem.c | 17 +++++------------
> >   1 file changed, 5 insertions(+), 12 deletions(-)
> >
> > diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
> > index 56a28a686988..01b31c56cead 100644
> > --- a/net/xdp/xdp_umem.c
> > +++ b/net/xdp/xdp_umem.c
> > @@ -97,7 +97,6 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
> >   {
> >       unsigned int gup_flags = FOLL_WRITE;
> >       long npgs;
> > -     int err;
> >
> >       umem->pgs = kcalloc(umem->npgs, sizeof(*umem->pgs),
> >                           GFP_KERNEL | __GFP_NOWARN);
> > @@ -112,20 +111,14 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
> >       if (npgs != umem->npgs) {
> >               if (npgs >= 0) {
> >                       umem->npgs = npgs;
> > -                     err = -ENOMEM;
> > -                     goto out_pin;
> > +                     xdp_umem_unpin_pages(umem);
> > +                     return -ENOMEM;
> >               }
> > -             err = npgs;
> > -             goto out_pgs;
> > +             kfree(umem->pgs);
> > +             umem->pgs = NULL;
> > +             return (int)npgs;
> >       }
> >       return 0;
> > -
> > -out_pin:
> > -     xdp_umem_unpin_pages(umem);
> > -out_pgs:
> > -     kfree(umem->pgs);
> > -     umem->pgs = NULL;
> > -     return err;
> >   }
> >
> >   static int xdp_umem_account_pages(struct xdp_umem *umem)
> >
>
> While at it, maybe we could also simplify the if (npgs != umem->npgs) a bit to
> get rid of the indent, something like:

The original patch is just to make some cleanups. Please do not make
it so complicated.
If you like it, please file an official patch for code review.

>
> diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
> index 56a28a686988..fa4dd16cced5 100644
> --- a/net/xdp/xdp_umem.c
> +++ b/net/xdp/xdp_umem.c
> @@ -97,7 +97,6 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
>   {
>         unsigned int gup_flags = FOLL_WRITE;
>         long npgs;
> -       int err;
>
>         umem->pgs = kcalloc(umem->npgs, sizeof(*umem->pgs),
>                             GFP_KERNEL | __GFP_NOWARN);
> @@ -108,24 +107,17 @@ static int xdp_umem_pin_pages(struct xdp_umem *umem, unsigned long address)
>         npgs = pin_user_pages(address, umem->npgs,
>                               gup_flags | FOLL_LONGTERM, &umem->pgs[0], NULL);
>         mmap_read_unlock(current->mm);
> -
> -       if (npgs != umem->npgs) {
> -               if (npgs >= 0) {
> -                       umem->npgs = npgs;
> -                       err = -ENOMEM;
> -                       goto out_pin;
> -               }
> -               err = npgs;
> -               goto out_pgs;
> +       if (npgs == umem->npgs)
> +               return 0;
> +       if (npgs >= 0) {
> +               umem->npgs = npgs;
> +               xdp_umem_unpin_pages(umem);
> +               return -ENOMEM;
>         }
> -       return 0;
>
> -out_pin:
> -       xdp_umem_unpin_pages(umem);
> -out_pgs:
>         kfree(umem->pgs);
>         umem->pgs = NULL;
> -       return err;
> +       return npgs;
>   }
>
>   static int xdp_umem_account_pages(struct xdp_umem *umem)