From: Joe Stringer <joe@wand.net.nz>
To: bpf@vger.kernel.org
Cc: netdev@vger.kernel.org, daniel@iogearbox.net, ast@kernel.org,
eric.dumazet@gmail.com, lmb@cloudflare.com
Subject: [PATCH bpf-next 2/7] dst: Add socket prefetch metadata destinations
Date: Thu, 12 Mar 2020 16:36:43 -0700 [thread overview]
Message-ID: <20200312233648.1767-3-joe@wand.net.nz> (raw)
In-Reply-To: <20200312233648.1767-1-joe@wand.net.nz>
Metadata destinations were introduced in commit f38a9eb1f77b
("dst: Metadata destinations") to "carry per packet metadata
between forwarding and processing elements via the skb->dst pointer".
The aim of this new METADATA_SK_PREFETCH destination type is to allow
early forwarding elements to store a socket destination for the duration
of receiving into the IP stack, which can be later be identified to
avoid orphaning the skb and losing the prefetched socket in ip_rcv_core().
The destination is stored temporarily in a per-CPU buffer to ensure that
if applications attempt to reach out from loopback address to loopback
address, they may restore the original destination and avoid martian
packet drops.
Signed-off-by: Joe Stringer <joe@wand.net.nz>
---
include/net/dst_metadata.h | 31 +++++++++++++++++++++++++++++++
net/core/dst.c | 30 ++++++++++++++++++++++++++++++
2 files changed, 61 insertions(+)
diff --git a/include/net/dst_metadata.h b/include/net/dst_metadata.h
index 56cb3c38569a..31574c553a07 100644
--- a/include/net/dst_metadata.h
+++ b/include/net/dst_metadata.h
@@ -9,6 +9,7 @@
enum metadata_type {
METADATA_IP_TUNNEL,
METADATA_HW_PORT_MUX,
+ METADATA_SK_PREFETCH,
};
struct hw_port_info {
@@ -80,6 +81,8 @@ static inline int skb_metadata_dst_cmp(const struct sk_buff *skb_a,
return memcmp(&a->u.tun_info, &b->u.tun_info,
sizeof(a->u.tun_info) +
a->u.tun_info.options_len);
+ case METADATA_SK_PREFETCH:
+ return 0;
default:
return 1;
}
@@ -214,4 +217,32 @@ static inline struct metadata_dst *ipv6_tun_rx_dst(struct sk_buff *skb,
0, ip6_flowlabel(ip6h), flags, tunnel_id,
md_size);
}
+
+extern const struct metadata_dst dst_sk_prefetch;
+
+static inline bool dst_is_sk_prefetch(const struct dst_entry *dst)
+{
+ return dst == &dst_sk_prefetch.dst;
+}
+
+static inline bool skb_dst_is_sk_prefetch(const struct sk_buff *skb)
+{
+ return dst_is_sk_prefetch(skb_dst(skb));
+}
+
+void dst_sk_prefetch_store(struct sk_buff *skb);
+void dst_sk_prefetch_fetch(struct sk_buff *skb);
+
+/**
+ * dst_sk_prefetch_reset - reset prefetched socket dst
+ * @skb: buffer
+ *
+ * Reverts the dst back to the originally stored dst if present.
+ */
+static inline void dst_sk_prefetch_reset(struct sk_buff *skb)
+{
+ if (unlikely(skb_dst_is_sk_prefetch(skb)))
+ dst_sk_prefetch_fetch(skb);
+}
+
#endif /* __NET_DST_METADATA_H */
diff --git a/net/core/dst.c b/net/core/dst.c
index 193af526e908..cf1a1d5b6b0a 100644
--- a/net/core/dst.c
+++ b/net/core/dst.c
@@ -330,3 +330,33 @@ void metadata_dst_free_percpu(struct metadata_dst __percpu *md_dst)
free_percpu(md_dst);
}
EXPORT_SYMBOL_GPL(metadata_dst_free_percpu);
+
+const struct metadata_dst dst_sk_prefetch = {
+ .dst = {
+ .ops = &md_dst_ops,
+ .input = dst_md_discard,
+ .output = dst_md_discard_out,
+ .flags = DST_NOCOUNT | DST_METADATA,
+ .obsolete = DST_OBSOLETE_NONE,
+ .__refcnt = ATOMIC_INIT(1),
+ },
+ .type = METADATA_SK_PREFETCH,
+};
+EXPORT_SYMBOL(dst_sk_prefetch);
+
+DEFINE_PER_CPU(unsigned long, dst_sk_prefetch_dst);
+
+void dst_sk_prefetch_store(struct sk_buff *skb)
+{
+ unsigned long refdst;
+
+ refdst = skb->_skb_refdst;
+ __this_cpu_write(dst_sk_prefetch_dst, refdst);
+ skb_dst_set_noref(skb, (struct dst_entry *)&dst_sk_prefetch.dst);
+}
+
+void dst_sk_prefetch_fetch(struct sk_buff *skb)
+{
+ skb->_skb_refdst = __this_cpu_read(dst_sk_prefetch_dst);
+}
+EXPORT_SYMBOL(dst_sk_prefetch_fetch);
--
2.20.1
next prev parent reply other threads:[~2020-03-12 23:36 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-03-12 23:36 [PATCH bpf-next 0/7] Add bpf_sk_assign eBPF helper Joe Stringer
2020-03-12 23:36 ` [PATCH bpf-next 1/7] dst: Move skb_dst_drop to skbuff.c Joe Stringer
2020-03-12 23:36 ` Joe Stringer [this message]
2020-03-12 23:36 ` [PATCH bpf-next 3/7] bpf: Add socket assign support Joe Stringer
2020-03-16 10:08 ` Jakub Sitnicki
2020-03-16 21:23 ` Joe Stringer
2020-03-16 22:57 ` Martin KaFai Lau
2020-03-17 3:06 ` Joe Stringer
2020-03-17 6:26 ` Martin KaFai Lau
2020-03-18 0:46 ` Joe Stringer
2020-03-18 10:03 ` Jakub Sitnicki
2020-03-19 5:49 ` Joe Stringer
2020-03-18 18:48 ` Martin KaFai Lau
2020-03-19 6:24 ` Joe Stringer
2020-03-20 1:54 ` Martin KaFai Lau
2020-03-20 4:28 ` Joe Stringer
2020-03-17 10:09 ` Lorenz Bauer
2020-03-18 1:10 ` Joe Stringer
2020-03-18 2:03 ` Joe Stringer
2020-03-12 23:36 ` [PATCH bpf-next 4/7] dst: Prefetch established socket destinations Joe Stringer
2020-03-16 23:03 ` Martin KaFai Lau
2020-03-17 3:17 ` Joe Stringer
2020-03-12 23:36 ` [PATCH bpf-next 5/7] selftests: bpf: add test for sk_assign Joe Stringer
2020-03-17 6:30 ` Martin KaFai Lau
2020-03-17 20:56 ` Joe Stringer
2020-03-18 17:27 ` Martin KaFai Lau
2020-03-19 5:45 ` Joe Stringer
2020-03-19 17:36 ` Andrii Nakryiko
2020-03-12 23:36 ` [PATCH bpf-next 6/7] selftests: bpf: Extend sk_assign for address proxy Joe Stringer
2020-03-12 23:36 ` [PATCH bpf-next 7/7] selftests: bpf: Improve debuggability of sk_assign Joe Stringer
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200312233648.1767-3-joe@wand.net.nz \
--to=joe@wand.net.nz \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=eric.dumazet@gmail.com \
--cc=lmb@cloudflare.com \
--cc=netdev@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).