From b5947e5d1e710c35ea281247bd27e6975250285c Mon Sep 17 00:00:00 2001 From: Willem de Bruijn Date: Fri, 30 Nov 2018 15:32:39 -0500 Subject: [PATCH 1/3] udp: msg_zerocopy Extend zerocopy to udp sockets. Allow setting sockopt SO_ZEROCOPY and interpret flag MSG_ZEROCOPY. This patch was previously part of the zerocopy RFC patchsets. Zerocopy is not effective at small MTU. With segmentation offload building larger datagrams, the benefit of page flipping outweights the cost of generating a completion notification. tools/testing/selftests/net/msg_zerocopy.sh after applying follow-on test patch and making skb_orphan_frags_rx same as skb_orphan_frags: ipv4 udp -t 1 tx=191312 (11938 MB) txc=0 zc=n rx=191312 (11938 MB) ipv4 udp -z -t 1 tx=304507 (19002 MB) txc=304507 zc=y rx=304507 (19002 MB) ok ipv6 udp -t 1 tx=174485 (10888 MB) txc=0 zc=n rx=174485 (10888 MB) ipv6 udp -z -t 1 tx=294801 (18396 MB) txc=294801 zc=y rx=294801 (18396 MB) ok Changes v1 -> v2 - Fixup reverse christmas tree violation v2 -> v3 - Split refcount avoidance optimization into separate patch - Fix refcount leak on error in fragmented case (thanks to Paolo Abeni for pointing this one out!) - Fix refcount inc on zero - Test sock_flag SOCK_ZEROCOPY directly in __ip_append_data. This is needed since commit 5cf4a8532c99 ("tcp: really ignore MSG_ZEROCOPY if no SO_ZEROCOPY") did the same for tcp. Signed-off-by: Willem de Bruijn Acked-by: Paolo Abeni Signed-off-by: David S. Miller --- include/linux/skbuff.h | 1 + net/core/skbuff.c | 6 ++++++ net/core/sock.c | 5 ++++- net/ipv4/ip_output.c | 23 ++++++++++++++++++++++- net/ipv6/ip6_output.c | 23 ++++++++++++++++++++++- 5 files changed, 55 insertions(+), 3 deletions(-) diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index 73902acf2b71..04f52e719571 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -485,6 +485,7 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg); void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); +int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len); int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, struct msghdr *msg, int len, struct ubuf_info *uarg); diff --git a/net/core/skbuff.c b/net/core/skbuff.c index 3c814565ed7c..1350901c5cb8 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -1105,6 +1105,12 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, struct iov_iter *from, size_t length); +int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len) +{ + return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); +} +EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram); + int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, struct msghdr *msg, int len, struct ubuf_info *uarg) diff --git a/net/core/sock.c b/net/core/sock.c index 6d7e189e3cd9..f5bb89785e47 100644 --- a/net/core/sock.c +++ b/net/core/sock.c @@ -1018,7 +1018,10 @@ set_rcvbuf: case SO_ZEROCOPY: if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { - if (sk->sk_protocol != IPPROTO_TCP) + if (!((sk->sk_type == SOCK_STREAM && + sk->sk_protocol == IPPROTO_TCP) || + (sk->sk_type == SOCK_DGRAM && + sk->sk_protocol == IPPROTO_UDP))) ret = -ENOTSUPP; } else if (sk->sk_family != PF_RDS) { ret = -ENOTSUPP; diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c index 5dbec21856f4..6f843aff628c 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -867,6 +867,7 @@ static int __ip_append_data(struct sock *sk, unsigned int flags) { struct inet_sock *inet = inet_sk(sk); + struct ubuf_info *uarg = NULL; struct sk_buff *skb; struct ip_options *opt = cork->opt; @@ -916,6 +917,19 @@ static int __ip_append_data(struct sock *sk, (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM))) csummode = CHECKSUM_PARTIAL; + if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { + uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); + if (!uarg) + return -ENOBUFS; + if (rt->dst.dev->features & NETIF_F_SG && + csummode == CHECKSUM_PARTIAL) { + paged = true; + } else { + uarg->zerocopy = 0; + skb_zcopy_set(skb, uarg); + } + } + cork->length += length; /* So, what's going on in the loop below? @@ -1006,6 +1020,7 @@ alloc_new_skb: cork->tx_flags = 0; skb_shinfo(skb)->tskey = tskey; tskey = 0; + skb_zcopy_set(skb, uarg); /* * Find where to start putting bytes. @@ -1068,7 +1083,7 @@ alloc_new_skb: err = -EFAULT; goto error; } - } else { + } else if (!uarg || !uarg->zerocopy) { int i = skb_shinfo(skb)->nr_frags; err = -ENOMEM; @@ -1098,6 +1113,10 @@ alloc_new_skb: skb->data_len += copy; skb->truesize += copy; wmem_alloc_delta += copy; + } else { + err = skb_zerocopy_iter_dgram(skb, from, copy); + if (err < 0) + goto error; } offset += copy; length -= copy; @@ -1105,11 +1124,13 @@ alloc_new_skb: if (wmem_alloc_delta) refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); + sock_zerocopy_put(uarg); return 0; error_efault: err = -EFAULT; error: + sock_zerocopy_put_abort(uarg); cork->length -= length; IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c index 827a3f5ff3bb..7df04d20a91f 100644 --- a/net/ipv6/ip6_output.c +++ b/net/ipv6/ip6_output.c @@ -1245,6 +1245,7 @@ static int __ip6_append_data(struct sock *sk, { struct sk_buff *skb, *skb_prev = NULL; unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu; + struct ubuf_info *uarg = NULL; int exthdrlen = 0; int dst_exthdrlen = 0; int hh_len; @@ -1322,6 +1323,19 @@ emsgsize: rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM)) csummode = CHECKSUM_PARTIAL; + if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) { + uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); + if (!uarg) + return -ENOBUFS; + if (rt->dst.dev->features & NETIF_F_SG && + csummode == CHECKSUM_PARTIAL) { + paged = true; + } else { + uarg->zerocopy = 0; + skb_zcopy_set(skb, uarg); + } + } + /* * Let's try using as much space as possible. * Use MTU if total length of the message fits into the MTU. @@ -1445,6 +1459,7 @@ alloc_new_skb: cork->tx_flags = 0; skb_shinfo(skb)->tskey = tskey; tskey = 0; + skb_zcopy_set(skb, uarg); /* * Find where to start putting bytes @@ -1506,7 +1521,7 @@ alloc_new_skb: err = -EFAULT; goto error; } - } else { + } else if (!uarg || !uarg->zerocopy) { int i = skb_shinfo(skb)->nr_frags; err = -ENOMEM; @@ -1536,6 +1551,10 @@ alloc_new_skb: skb->data_len += copy; skb->truesize += copy; wmem_alloc_delta += copy; + } else { + err = skb_zerocopy_iter_dgram(skb, from, copy); + if (err < 0) + goto error; } offset += copy; length -= copy; @@ -1543,11 +1562,13 @@ alloc_new_skb: if (wmem_alloc_delta) refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); + sock_zerocopy_put(uarg); return 0; error_efault: err = -EFAULT; error: + sock_zerocopy_put_abort(uarg); cork->length -= length; IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); From 52900d22288e7d45846037e1db277c665bbc40db Mon Sep 17 00:00:00 2001 From: Willem de Bruijn Date: Fri, 30 Nov 2018 15:32:40 -0500 Subject: [PATCH 2/3] udp: elide zerocopy operation in hot path With MSG_ZEROCOPY, each skb holds a reference to a struct ubuf_info. Release of its last reference triggers a completion notification. The TCP stack in tcp_sendmsg_locked holds an extra ref independent of the skbs, because it can build, send and free skbs within its loop, possibly reaching refcount zero and freeing the ubuf_info too soon. The UDP stack currently also takes this extra ref, but does not need it as all skbs are sent after return from __ip(6)_append_data. Avoid the extra refcount_inc and refcount_dec_and_test, and generally the sock_zerocopy_put in the common path, by passing the initial reference to the first skb. This approach is taken instead of initializing the refcount to 0, as that would generate error "refcount_t: increment on 0" on the next skb_zcopy_set. Changes v3 -> v4 - Move skb_zcopy_set below the only kfree_skb that might cause a premature uarg destroy before skb_zerocopy_put_abort - Move the entire skb_shinfo assignment block, to keep that cacheline access in one place Signed-off-by: Willem de Bruijn Acked-by: Paolo Abeni Signed-off-by: David S. Miller --- include/linux/skbuff.h | 12 ++++++++---- net/core/skbuff.c | 9 +++++---- net/ipv4/ip_output.c | 22 +++++++++++----------- net/ipv4/tcp.c | 2 +- net/ipv6/ip6_output.c | 22 +++++++++++----------- 5 files changed, 36 insertions(+), 31 deletions(-) diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index 04f52e719571..75d50ab7997c 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -481,7 +481,7 @@ static inline void sock_zerocopy_get(struct ubuf_info *uarg) } void sock_zerocopy_put(struct ubuf_info *uarg); -void sock_zerocopy_put_abort(struct ubuf_info *uarg); +void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); @@ -1326,10 +1326,14 @@ static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) return is_zcopy ? skb_uarg(skb) : NULL; } -static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) +static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, + bool *have_ref) { if (skb && uarg && !skb_zcopy(skb)) { - sock_zerocopy_get(uarg); + if (unlikely(have_ref && *have_ref)) + *have_ref = false; + else + sock_zerocopy_get(uarg); skb_shinfo(skb)->destructor_arg = uarg; skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; } @@ -1374,7 +1378,7 @@ static inline void skb_zcopy_abort(struct sk_buff *skb) struct ubuf_info *uarg = skb_zcopy(skb); if (uarg) { - sock_zerocopy_put_abort(uarg); + sock_zerocopy_put_abort(uarg, false); skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; } } diff --git a/net/core/skbuff.c b/net/core/skbuff.c index 1350901c5cb8..c78ce114537e 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -1089,7 +1089,7 @@ void sock_zerocopy_put(struct ubuf_info *uarg) } EXPORT_SYMBOL_GPL(sock_zerocopy_put); -void sock_zerocopy_put_abort(struct ubuf_info *uarg) +void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) { if (uarg) { struct sock *sk = skb_from_uarg(uarg)->sk; @@ -1097,7 +1097,8 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg) atomic_dec(&sk->sk_zckey); uarg->len--; - sock_zerocopy_put(uarg); + if (have_uref) + sock_zerocopy_put(uarg); } } EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort); @@ -1137,7 +1138,7 @@ int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, return err; } - skb_zcopy_set(skb, uarg); + skb_zcopy_set(skb, uarg, NULL); return skb->len - orig_len; } EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream); @@ -1157,7 +1158,7 @@ static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig, if (skb_copy_ubufs(nskb, GFP_ATOMIC)) return -EIO; } - skb_zcopy_set(nskb, skb_uarg(orig)); + skb_zcopy_set(nskb, skb_uarg(orig), NULL); } return 0; } diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c index 6f843aff628c..78f028bdad30 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -881,8 +881,8 @@ static int __ip_append_data(struct sock *sk, int csummode = CHECKSUM_NONE; struct rtable *rt = (struct rtable *)cork->dst; unsigned int wmem_alloc_delta = 0; + bool paged, extra_uref; u32 tskey = 0; - bool paged; skb = skb_peek_tail(queue); @@ -921,12 +921,13 @@ static int __ip_append_data(struct sock *sk, uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); if (!uarg) return -ENOBUFS; + extra_uref = true; if (rt->dst.dev->features & NETIF_F_SG && csummode == CHECKSUM_PARTIAL) { paged = true; } else { uarg->zerocopy = 0; - skb_zcopy_set(skb, uarg); + skb_zcopy_set(skb, uarg, &extra_uref); } } @@ -1015,13 +1016,6 @@ alloc_new_skb: skb->csum = 0; skb_reserve(skb, hh_len); - /* only the initial fragment is time stamped */ - skb_shinfo(skb)->tx_flags = cork->tx_flags; - cork->tx_flags = 0; - skb_shinfo(skb)->tskey = tskey; - tskey = 0; - skb_zcopy_set(skb, uarg); - /* * Find where to start putting bytes. */ @@ -1054,6 +1048,13 @@ alloc_new_skb: exthdrlen = 0; csummode = CHECKSUM_NONE; + /* only the initial fragment is time stamped */ + skb_shinfo(skb)->tx_flags = cork->tx_flags; + cork->tx_flags = 0; + skb_shinfo(skb)->tskey = tskey; + tskey = 0; + skb_zcopy_set(skb, uarg, &extra_uref); + if ((flags & MSG_CONFIRM) && !skb_prev) skb_set_dst_pending_confirm(skb, 1); @@ -1124,13 +1125,12 @@ alloc_new_skb: if (wmem_alloc_delta) refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); - sock_zerocopy_put(uarg); return 0; error_efault: err = -EFAULT; error: - sock_zerocopy_put_abort(uarg); + sock_zerocopy_put_abort(uarg, extra_uref); cork->length -= length; IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 215e4d3b3616..dc68c408bba0 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -1423,7 +1423,7 @@ do_error: if (copied + copied_syn) goto out; out_err: - sock_zerocopy_put_abort(uarg); + sock_zerocopy_put_abort(uarg, true); err = sk_stream_error(sk, flags, err); /* make sure we wake any epoll edge trigger waiter */ if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c index 7df04d20a91f..ec8c235ea891 100644 --- a/net/ipv6/ip6_output.c +++ b/net/ipv6/ip6_output.c @@ -1258,7 +1258,7 @@ static int __ip6_append_data(struct sock *sk, int csummode = CHECKSUM_NONE; unsigned int maxnonfragsize, headersize; unsigned int wmem_alloc_delta = 0; - bool paged; + bool paged, extra_uref; skb = skb_peek_tail(queue); if (!skb) { @@ -1327,12 +1327,13 @@ emsgsize: uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb)); if (!uarg) return -ENOBUFS; + extra_uref = true; if (rt->dst.dev->features & NETIF_F_SG && csummode == CHECKSUM_PARTIAL) { paged = true; } else { uarg->zerocopy = 0; - skb_zcopy_set(skb, uarg); + skb_zcopy_set(skb, uarg, &extra_uref); } } @@ -1454,13 +1455,6 @@ alloc_new_skb: skb_reserve(skb, hh_len + sizeof(struct frag_hdr) + dst_exthdrlen); - /* Only the initial fragment is time stamped */ - skb_shinfo(skb)->tx_flags = cork->tx_flags; - cork->tx_flags = 0; - skb_shinfo(skb)->tskey = tskey; - tskey = 0; - skb_zcopy_set(skb, uarg); - /* * Find where to start putting bytes */ @@ -1492,6 +1486,13 @@ alloc_new_skb: exthdrlen = 0; dst_exthdrlen = 0; + /* Only the initial fragment is time stamped */ + skb_shinfo(skb)->tx_flags = cork->tx_flags; + cork->tx_flags = 0; + skb_shinfo(skb)->tskey = tskey; + tskey = 0; + skb_zcopy_set(skb, uarg, &extra_uref); + if ((flags & MSG_CONFIRM) && !skb_prev) skb_set_dst_pending_confirm(skb, 1); @@ -1562,13 +1563,12 @@ alloc_new_skb: if (wmem_alloc_delta) refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); - sock_zerocopy_put(uarg); return 0; error_efault: err = -EFAULT; error: - sock_zerocopy_put_abort(uarg); + sock_zerocopy_put_abort(uarg, extra_uref); cork->length -= length; IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS); refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc); From db63e489c7aa03bab86960d6bcb6bf1fad8eea4e Mon Sep 17 00:00:00 2001 From: Willem de Bruijn Date: Fri, 30 Nov 2018 15:32:41 -0500 Subject: [PATCH 3/3] selftests: extend zerocopy tests to udp Both msg_zerocopy and udpgso_bench have udp zerocopy variants. Exercise these as part of the standard kselftest run. With udp, msg_zerocopy has no control channel. Ensure that the receiver exits after the sender by accounting for the initial delay in starting them (in msg_zerocopy.sh). Signed-off-by: Willem de Bruijn Signed-off-by: David S. Miller --- tools/testing/selftests/net/msg_zerocopy.c | 3 ++- tools/testing/selftests/net/msg_zerocopy.sh | 2 ++ tools/testing/selftests/net/udpgso_bench.sh | 3 +++ 3 files changed, 7 insertions(+), 1 deletion(-) diff --git a/tools/testing/selftests/net/msg_zerocopy.c b/tools/testing/selftests/net/msg_zerocopy.c index 406cc70c571d..4b02933cab8a 100644 --- a/tools/testing/selftests/net/msg_zerocopy.c +++ b/tools/testing/selftests/net/msg_zerocopy.c @@ -651,12 +651,13 @@ static void do_flush_datagram(int fd, int type) static void do_rx(int domain, int type, int protocol) { + const int cfg_receiver_wait_ms = 400; uint64_t tstop; int fd; fd = do_setup_rx(domain, type, protocol); - tstop = gettimeofday_ms() + cfg_runtime_ms; + tstop = gettimeofday_ms() + cfg_runtime_ms + cfg_receiver_wait_ms; do { if (type == SOCK_STREAM) do_flush_tcp(fd); diff --git a/tools/testing/selftests/net/msg_zerocopy.sh b/tools/testing/selftests/net/msg_zerocopy.sh index c43c6debda06..825ffec85cea 100755 --- a/tools/testing/selftests/net/msg_zerocopy.sh +++ b/tools/testing/selftests/net/msg_zerocopy.sh @@ -25,6 +25,8 @@ readonly path_sysctl_mem="net.core.optmem_max" if [[ "$#" -eq "0" ]]; then $0 4 tcp -t 1 $0 6 tcp -t 1 + $0 4 udp -t 1 + $0 6 udp -t 1 echo "OK. All tests passed" exit 0 fi diff --git a/tools/testing/selftests/net/udpgso_bench.sh b/tools/testing/selftests/net/udpgso_bench.sh index 0f0628613f81..5670a9ffd8eb 100755 --- a/tools/testing/selftests/net/udpgso_bench.sh +++ b/tools/testing/selftests/net/udpgso_bench.sh @@ -35,6 +35,9 @@ run_udp() { echo "udp gso" run_in_netns ${args} -S 0 + + echo "udp gso zerocopy" + run_in_netns ${args} -S 0 -z } run_tcp() {