| Message ID | 0eb1cb5746e9ac938a7ba7848b33ccf680d30030.1657643355.git.asml.silence@gmail.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | io_uring zerocopy send | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Clearly marked for net-next, async |
| netdev/apply | fail | Patch does not apply to net-next |
On Tue, 12 Jul 2022 21:52:25 +0100 Pavel Begunkov wrote: > Even when zerocopy transmission is requested and possible, > __ip_append_data() will still copy a small chunk of data just because it > allocated some extra linear space (e.g. 148 bytes). It wastes CPU cycles > on copy and iter manipulations and also misalignes potentially aligned > data. Avoid such coies. And as a bonus we can allocate smaller skb. s/coies/copies/ can fix when applying > > Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> > --- > net/ipv4/ip_output.c | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c > index 00b4bf26fd93..581d1e233260 100644 > --- a/net/ipv4/ip_output.c > +++ b/net/ipv4/ip_output.c > @@ -969,7 +969,6 @@ static int __ip_append_data(struct sock *sk, > struct inet_sock *inet = inet_sk(sk); > struct ubuf_info *uarg = NULL; > struct sk_buff *skb; > - > struct ip_options *opt = cork->opt; > int hh_len; > int exthdrlen; > @@ -977,6 +976,7 @@ static int __ip_append_data(struct sock *sk, > int copy; > int err; > int offset = 0; > + bool zc = false; > unsigned int maxfraglen, fragheaderlen, maxnonfragsize; > int csummode = CHECKSUM_NONE; > struct rtable *rt = (struct rtable *)cork->dst; > @@ -1025,6 +1025,7 @@ static int __ip_append_data(struct sock *sk, > if (rt->dst.dev->features & NETIF_F_SG && > csummode == CHECKSUM_PARTIAL) { > paged = true; > + zc = true; > } else { > uarg->zerocopy = 0; > skb_zcopy_set(skb, uarg, &extra_uref); > @@ -1091,9 +1092,12 @@ static int __ip_append_data(struct sock *sk, > (fraglen + alloc_extra < SKB_MAX_ALLOC || > !(rt->dst.dev->features & NETIF_F_SG))) > alloclen = fraglen; > - else { > + else if (!zc) { > alloclen = min_t(int, fraglen, MAX_HEADER); Willem, I think this came in with your GSO work, is there a reason we use MAX_HEADER here? I thought MAX_HEADER is for headers (i.e. more or less to be reserved) not for the min amount of data to be included. I wanna make sure we're not missing something about GSO here. Otherwise I don't think we need the extra branch but that can be a follow up. > pagedlen = fraglen - alloclen; > + } else { > + alloclen = fragheaderlen + transhdrlen; > + pagedlen = datalen - transhdrlen; > } > > alloclen += alloc_extra;
On Tue, Jul 19, 2022 at 3:54 AM Jakub Kicinski <kuba@kernel.org> wrote: > > On Tue, 12 Jul 2022 21:52:25 +0100 Pavel Begunkov wrote: > > Even when zerocopy transmission is requested and possible, > > __ip_append_data() will still copy a small chunk of data just because it > > allocated some extra linear space (e.g. 148 bytes). It wastes CPU cycles > > on copy and iter manipulations and also misalignes potentially aligned > > data. Avoid such coies. And as a bonus we can allocate smaller skb. > > s/coies/copies/ can fix when applying > > > > > Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> > > --- > > net/ipv4/ip_output.c | 8 ++++++-- > > 1 file changed, 6 insertions(+), 2 deletions(-) > > > > diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c > > index 00b4bf26fd93..581d1e233260 100644 > > --- a/net/ipv4/ip_output.c > > +++ b/net/ipv4/ip_output.c > > @@ -969,7 +969,6 @@ static int __ip_append_data(struct sock *sk, > > struct inet_sock *inet = inet_sk(sk); > > struct ubuf_info *uarg = NULL; > > struct sk_buff *skb; > > - > > struct ip_options *opt = cork->opt; > > int hh_len; > > int exthdrlen; > > @@ -977,6 +976,7 @@ static int __ip_append_data(struct sock *sk, > > int copy; > > int err; > > int offset = 0; > > + bool zc = false; > > unsigned int maxfraglen, fragheaderlen, maxnonfragsize; > > int csummode = CHECKSUM_NONE; > > struct rtable *rt = (struct rtable *)cork->dst; > > @@ -1025,6 +1025,7 @@ static int __ip_append_data(struct sock *sk, > > if (rt->dst.dev->features & NETIF_F_SG && > > csummode == CHECKSUM_PARTIAL) { > > paged = true; > > + zc = true; > > } else { > > uarg->zerocopy = 0; > > skb_zcopy_set(skb, uarg, &extra_uref); > > @@ -1091,9 +1092,12 @@ static int __ip_append_data(struct sock *sk, > > (fraglen + alloc_extra < SKB_MAX_ALLOC || > > !(rt->dst.dev->features & NETIF_F_SG))) > > alloclen = fraglen; > > - else { > > + else if (!zc) { > > alloclen = min_t(int, fraglen, MAX_HEADER); > > Willem, I think this came in with your GSO work, is there a reason we > use MAX_HEADER here? I thought MAX_HEADER is for headers (i.e. more or > less to be reserved) not for the min amount of data to be included. > > I wanna make sure we're not missing something about GSO here. > > Otherwise I don't think we need the extra branch but that can > be a follow up. The change was introduced for UDP GSO, to avoid copying most payload on software segmentation: " commit 15e36f5b8e982debe43e425d2e12d34e022d51e9 Author: Willem de Bruijn <willemb@google.com> Date: Thu Apr 26 13:42:19 2018 -0400 udp: paged allocation with gso When sending large datagrams that are later segmented, store data in page frags to avoid copying from linear in skb_segment. " and in code - else - alloclen = datalen + fragheaderlen; + else if (!paged) + alloclen = fraglen; + else { + alloclen = min_t(int, fraglen, MAX_HEADER); + pagedlen = fraglen - alloclen; + } MAX_HEADER was a short-hand for the exact header length. "alloclen = fragheaderlen + transhdrlen;" is probably a better choice indeed. Whether with branch or without, the same change needs to be made to __ip6_append_data, just as in the referenced commit. Let's keep the stacks in sync. This is tricky code. If in doubt, run the msg_zerocopy and udp_gso tests from tools/testing/selftests/net, ideally with KASAN.
On 7/19/22 10:35, Willem de Bruijn wrote: > On Tue, Jul 19, 2022 at 3:54 AM Jakub Kicinski <kuba@kernel.org> wrote: >> >> On Tue, 12 Jul 2022 21:52:25 +0100 Pavel Begunkov wrote: >>> Even when zerocopy transmission is requested and possible, >>> __ip_append_data() will still copy a small chunk of data just because it >>> allocated some extra linear space (e.g. 148 bytes). It wastes CPU cycles >>> on copy and iter manipulations and also misalignes potentially aligned >>> data. Avoid such coies. And as a bonus we can allocate smaller skb. >> >> s/coies/copies/ can fix when applying >> >>> >>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> >>> --- >>> net/ipv4/ip_output.c | 8 ++++++-- >>> 1 file changed, 6 insertions(+), 2 deletions(-) >>> >>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c >>> index 00b4bf26fd93..581d1e233260 100644 >>> --- a/net/ipv4/ip_output.c >>> +++ b/net/ipv4/ip_output.c >>> @@ -969,7 +969,6 @@ static int __ip_append_data(struct sock *sk, >>> struct inet_sock *inet = inet_sk(sk); >>> struct ubuf_info *uarg = NULL; >>> struct sk_buff *skb; >>> - >>> struct ip_options *opt = cork->opt; >>> int hh_len; >>> int exthdrlen; >>> @@ -977,6 +976,7 @@ static int __ip_append_data(struct sock *sk, >>> int copy; >>> int err; >>> int offset = 0; >>> + bool zc = false; >>> unsigned int maxfraglen, fragheaderlen, maxnonfragsize; >>> int csummode = CHECKSUM_NONE; >>> struct rtable *rt = (struct rtable *)cork->dst; >>> @@ -1025,6 +1025,7 @@ static int __ip_append_data(struct sock *sk, >>> if (rt->dst.dev->features & NETIF_F_SG && >>> csummode == CHECKSUM_PARTIAL) { >>> paged = true; >>> + zc = true; >>> } else { >>> uarg->zerocopy = 0; >>> skb_zcopy_set(skb, uarg, &extra_uref); >>> @@ -1091,9 +1092,12 @@ static int __ip_append_data(struct sock *sk, >>> (fraglen + alloc_extra < SKB_MAX_ALLOC || >>> !(rt->dst.dev->features & NETIF_F_SG))) >>> alloclen = fraglen; >>> - else { >>> + else if (!zc) { >>> alloclen = min_t(int, fraglen, MAX_HEADER); >> >> Willem, I think this came in with your GSO work, is there a reason we >> use MAX_HEADER here? I thought MAX_HEADER is for headers (i.e. more or >> less to be reserved) not for the min amount of data to be included. >> >> I wanna make sure we're not missing something about GSO here. >> >> Otherwise I don't think we need the extra branch but that can >> be a follow up. I brought it up before but left it for later as I don't know workloads and there might be perf implications. I'll send a follow up. > The change was introduced for UDP GSO, to avoid copying most payload > on software segmentation: > > " > commit 15e36f5b8e982debe43e425d2e12d34e022d51e9 > Author: Willem de Bruijn <willemb@google.com> > Date: Thu Apr 26 13:42:19 2018 -0400 > > udp: paged allocation with gso > > When sending large datagrams that are later segmented, store data in > page frags to avoid copying from linear in skb_segment. > " > > and in code > > - else > - alloclen = datalen + fragheaderlen; > + else if (!paged) > + alloclen = fraglen; > + else { > + alloclen = min_t(int, fraglen, MAX_HEADER); > + pagedlen = fraglen - alloclen; > + } > > > MAX_HEADER was a short-hand for the exact header length. "alloclen = > fragheaderlen + transhdrlen;" is probably a better choice indeed. Great, thanks for taking a look! > > Whether with branch or without, the same change needs to be made to > __ip6_append_data, just as in the referenced commit. Let's keep the > stacks in sync. __ip6_append_data() is changed as well but in the following patch. I had doubts whether it's preferable to keep ipv4 and ipv6 changes separately. > This is tricky code. If in doubt, run the msg_zerocopy and udp_gso > tests from tools/testing/selftests/net, ideally with KASAN.
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c index 00b4bf26fd93..581d1e233260 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -969,7 +969,6 @@ static int __ip_append_data(struct sock *sk, struct inet_sock *inet = inet_sk(sk); struct ubuf_info *uarg = NULL; struct sk_buff *skb; - struct ip_options *opt = cork->opt; int hh_len; int exthdrlen; @@ -977,6 +976,7 @@ static int __ip_append_data(struct sock *sk, int copy; int err; int offset = 0; + bool zc = false; unsigned int maxfraglen, fragheaderlen, maxnonfragsize; int csummode = CHECKSUM_NONE; struct rtable *rt = (struct rtable *)cork->dst; @@ -1025,6 +1025,7 @@ static int __ip_append_data(struct sock *sk, if (rt->dst.dev->features & NETIF_F_SG && csummode == CHECKSUM_PARTIAL) { paged = true; + zc = true; } else { uarg->zerocopy = 0; skb_zcopy_set(skb, uarg, &extra_uref); @@ -1091,9 +1092,12 @@ static int __ip_append_data(struct sock *sk, (fraglen + alloc_extra < SKB_MAX_ALLOC || !(rt->dst.dev->features & NETIF_F_SG))) alloclen = fraglen; - else { + else if (!zc) { alloclen = min_t(int, fraglen, MAX_HEADER); pagedlen = fraglen - alloclen; + } else { + alloclen = fragheaderlen + transhdrlen; + pagedlen = datalen - transhdrlen; } alloclen += alloc_extra;
Even when zerocopy transmission is requested and possible, __ip_append_data() will still copy a small chunk of data just because it allocated some extra linear space (e.g. 148 bytes). It wastes CPU cycles on copy and iter manipulations and also misalignes potentially aligned data. Avoid such coies. And as a bonus we can allocate smaller skb. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> --- net/ipv4/ip_output.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)