| Message ID | 1289381008-5484-1-git-send-email-xiaohui.xin@intel.com |
|---|---|
| State | RFC, archived |
| Delegated to: | David Miller |
| Headers | show |
From: xiaohui.xin@intel.com Date: Wed, 10 Nov 2010 17:23:28 +0800 > From: Xin Xiaohui <xiaohui.xin@intel.com> > >>2) The idea to key off of skb->dev in skb_release_data() is >> fundamentally flawed since many actions can change skb->dev on you, >> which will end up causing a leak of your external data areas. > > How about this one? If the destructor_arg is not a good candidate, > then I have to add an apparent field in shinfo. If destructor_arg is actually a net_device pointer or similar, you will need to take a reference count on it or similar. Which means --> good bye performance especially on SMP. You're going to be adding new serialization points and at least two new atomics per packet. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
>-----Original Message----- >From: David Miller [mailto:davem@davemloft.net] >Sent: Thursday, November 11, 2010 1:47 AM >To: Xin, Xiaohui >Cc: netdev@vger.kernel.org; kvm@vger.kernel.org; linux-kernel@vger.kernel.org; >mst@redhat.com; mingo@elte.hu; herbert@gondor.apana.org.au; jdike@linux.intel.com >Subject: Re: [PATCH v15 00/17] Provide a zero-copy method on KVM virtio-net. > >From: xiaohui.xin@intel.com >Date: Wed, 10 Nov 2010 17:23:28 +0800 > >> From: Xin Xiaohui <xiaohui.xin@intel.com> >> >>>2) The idea to key off of skb->dev in skb_release_data() is >>> fundamentally flawed since many actions can change skb->dev on you, >>> which will end up causing a leak of your external data areas. >> >> How about this one? If the destructor_arg is not a good candidate, >> then I have to add an apparent field in shinfo. > >If destructor_arg is actually a net_device pointer or similar, >you will need to take a reference count on it or similar. > Do you mean destructor_arg will be consumed by other user? If that case, may I add a new structure member in shinfo? Thus only zero-copy will use it, and no need for the reference count. >Which means --> good bye performance especially on SMP. > >You're going to be adding new serialization points and at >least two new atomics per packet. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
>-----Original Message----- >From: kvm-owner@vger.kernel.org [mailto:kvm-owner@vger.kernel.org] On Behalf Of Xin, >Xiaohui >Sent: Thursday, November 11, 2010 4:28 PM >To: David Miller >Cc: netdev@vger.kernel.org; kvm@vger.kernel.org; linux-kernel@vger.kernel.org; >mst@redhat.com; mingo@elte.hu; herbert@gondor.apana.org.au; jdike@linux.intel.com >Subject: RE: [PATCH v15 00/17] Provide a zero-copy method on KVM virtio-net. > >>-----Original Message----- >>From: David Miller [mailto:davem@davemloft.net] >>Sent: Thursday, November 11, 2010 1:47 AM >>To: Xin, Xiaohui >>Cc: netdev@vger.kernel.org; kvm@vger.kernel.org; linux-kernel@vger.kernel.org; >>mst@redhat.com; mingo@elte.hu; herbert@gondor.apana.org.au; jdike@linux.intel.com >>Subject: Re: [PATCH v15 00/17] Provide a zero-copy method on KVM virtio-net. >> >>From: xiaohui.xin@intel.com >>Date: Wed, 10 Nov 2010 17:23:28 +0800 >> >>> From: Xin Xiaohui <xiaohui.xin@intel.com> >>> >>>>2) The idea to key off of skb->dev in skb_release_data() is >>>> fundamentally flawed since many actions can change skb->dev on you, >>>> which will end up causing a leak of your external data areas. >>> >>> How about this one? If the destructor_arg is not a good candidate, >>> then I have to add an apparent field in shinfo. >> >>If destructor_arg is actually a net_device pointer or similar, >>you will need to take a reference count on it or similar. >> >Do you mean destructor_arg will be consumed by other user? >If that case, may I add a new structure member in shinfo? >Thus only zero-copy will use it, and no need for the reference count. > How about this? It really needs somewhere to track the external data area, and if something wrong with it, we can also release the data area. We think skb_release_data() is the right place to deal with it. If I understood right, that destructor_arg will be used by other else that why reference count is needed, then how about add a new structure member in shinfo? Thanks Xiaohui >>Which means --> good bye performance especially on SMP. >> >>You're going to be adding new serialization points and at >>least two new atomics per packet. >-- >To unsubscribe from this list: send the line "unsubscribe kvm" in >the body of a message to majordomo@vger.kernel.org >More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index 10ba67d..ad4636e 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -199,14 +199,15 @@ struct skb_shared_info { struct sk_buff *frag_list; struct skb_shared_hwtstamps hwtstamps; + /* Intermediate layers must ensure that destructor_arg + * remains valid until skb destructor */ + void * destructor_arg; + /* * Warning : all fields before dataref are cleared in __alloc_skb() */ atomic_t dataref; - /* Intermediate layers must ensure that destructor_arg - * remains valid until skb destructor */ - void * destructor_arg; /* must be last field, see pskb_expand_head() */ skb_frag_t frags[MAX_SKB_FRAGS]; }; diff --git a/net/core/skbuff.c b/net/core/skbuff.c index c83b421..eb040f4 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -343,6 +343,13 @@ static void skb_release_data(struct sk_buff *skb) if (skb_has_frags(skb)) skb_drop_fraglist(skb); + if (skb_shinfo(skb)->destructor_arg) { + struct skb_ext_page *ext_page = + skb_shinfo(skb)->destructor_arg; + if (ext_page->dtor) + ext_page->dtor(ext_page); + } + kfree(skb->head); } }