| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAL+tcoCe-Ee92r5B7LwV8GCxEBWDzq3X_g_oOWWzo7-u4wYZzw@mail.gmail.com>
Date: Thu, 16 Jan 2025 08:41:52 +0800
From: Jason Xing <kerneljasonxing@...il.com>
To: Martin KaFai Lau <martin.lau@...ux.dev>
Cc: davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
pabeni@...hat.com, dsahern@...nel.org, willemdebruijn.kernel@...il.com,
willemb@...gle.com, ast@...nel.org, daniel@...earbox.net, andrii@...nel.org,
eddyz87@...il.com, song@...nel.org, yonghong.song@...ux.dev,
john.fastabend@...il.com, kpsingh@...nel.org, sdf@...ichev.me,
haoluo@...gle.com, jolsa@...nel.org, horms@...nel.org, bpf@...r.kernel.org,
netdev@...r.kernel.org
Subject: Re: [PATCH net-next v5 13/15] net-timestamp: support tcp_sendmsg for
bpf extension
On Thu, Jan 16, 2025 at 8:04 AM Martin KaFai Lau <martin.lau@...ux.dev> wrote:
>
> On 1/12/25 3:37 AM, Jason Xing wrote:
> > Introduce tskey_bpf to correlate tcp_sendmsg timestamp with other
> > three points (SND/SW/ACK). More details can be found in the
> > selftest.
> >
> > For TCP, tskey_bpf is used to store the initial write_seq value
> > the moment tcp_sendmsg is called, so that the last skb of this
> > call will have the same tskey_bpf with tcp_sendmsg bpf callback.
> >
> > UDP works similarly because tskey_bpf can increase by one everytime
> > udp_sendmsg gets called. It will be implemented soon.
> >
> > Signed-off-by: Jason Xing <kerneljasonxing@...il.com>
> > ---
> > include/linux/skbuff.h | 2 ++
> > include/uapi/linux/bpf.h | 3 +++
> > net/core/sock.c | 3 ++-
> > net/ipv4/tcp.c | 10 ++++++++--
> > tools/include/uapi/linux/bpf.h | 3 +++
> > 5 files changed, 18 insertions(+), 3 deletions(-)
> >
> > diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> > index d3ef8db94a94..3b7b470d5d89 100644
> > --- a/include/linux/skbuff.h
> > +++ b/include/linux/skbuff.h
> > @@ -609,6 +609,8 @@ struct skb_shared_info {
> > };
> > unsigned int gso_type;
> > u32 tskey;
> > + /* For TCP, it records the initial write_seq when sendmsg is called */
> > + u32 tskey_bpf;
>
> I would suggest to remove this tskey_bpf addition to skb_shared_info. My
> understanding is the intention is to get the delay spent in the
> tcp_sendmsg_locked(). I think this can be done in bpf_sk_storage. More below.
>
> >
> > /*
> > * Warning : all fields before dataref are cleared in __alloc_skb()
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index a0aff1b4eb61..87420c0f2235 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -7037,6 +7037,9 @@ enum {
> > * feature is on. It indicates the
> > * recorded timestamp.
> > */
> > + BPF_SOCK_OPS_TS_TCP_SND_CB, /* Called when every tcp_sendmsg
> > + * syscall is triggered
> > + */
>
> UDP will need this also?
Yep.
>
> > };
> >
> > /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect
> > diff --git a/net/core/sock.c b/net/core/sock.c
> > index 2f54e60a50d4..e74ab0e2979d 100644
> > --- a/net/core/sock.c
> > +++ b/net/core/sock.c
> > @@ -958,7 +958,8 @@ void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
> > if (sk_is_tcp(sk) && sk_fullsock(sk))
> > sock_ops.is_fullsock = 1;
> > sock_ops.sk = sk;
> > - bpf_skops_init_skb(&sock_ops, skb, 0);
> > + if (skb)
> > + bpf_skops_init_skb(&sock_ops, skb, 0);
> > sock_ops.timestamp_used = 1;
> > __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS);
> > }
> > diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> > index 0a41006b10d1..b6e0db5e4ead 100644
> > --- a/net/ipv4/tcp.c
> > +++ b/net/ipv4/tcp.c
> > @@ -477,7 +477,7 @@ void tcp_init_sock(struct sock *sk)
> > }
> > EXPORT_SYMBOL(tcp_init_sock);
> >
> > -static void tcp_tx_timestamp(struct sock *sk, struct sockcm_cookie *sockc)
> > +static void tcp_tx_timestamp(struct sock *sk, struct sockcm_cookie *sockc, u32 first_write_seq)
> > {
> > struct sk_buff *skb = tcp_write_queue_tail(sk);
> > u32 tsflags = sockc->tsflags;
> > @@ -500,6 +500,7 @@ static void tcp_tx_timestamp(struct sock *sk, struct sockcm_cookie *sockc)
> > tcb->txstamp_ack_bpf = 1;
> > shinfo->tx_flags |= SKBTX_BPF;
> > shinfo->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1;
>
> Add the bpf prog callout here instead:
>
> bpf_skops_tx_timestamping(sk, skb, BPF_SOCK_OPS_TS_TCP_SND_CB);
If we trigger the first time callback here by using
BPF_SOCK_OPS_TS_TCP_SND_CB, it's a little bit late. Do you mean that
we can use the same new callback at the beginning of
tcp_sendmsg_locked() and tcp_tx_timestamp()?
>
> If the bpf prog wants to figure out the delay from the very beginning of the
> tcp_sendmsg_locked(), a bpf prog (either by tracing the tcp_sendmsg_locked or by
> adding a new callout at the beginning of tcp_sendmsg_locked like this patch) can
> store a bpf_ktime_get_ns() in the bpf_sk_storage. The bpf prog running here (at
Thanks for a new lesson about the usage of bpf_sk_storage here. I'll
dig into it.
> tcp_tx_timestamp) can get that timestamp from the bpf_sk_storage since it has a
> hold on the same sk pointer. There is no need to add a new shinfo->tskey_bpf to
> measure this part of the delay.
>
> > + shinfo->tskey_bpf = first_write_seq;
> > }
> > }
> >
> > @@ -1067,10 +1068,15 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
> > int flags, err, copied = 0;
> > int mss_now = 0, size_goal, copied_syn = 0;
> > int process_backlog = 0;
> > + u32 first_write_seq = 0;
> > int zc = 0;
> > long timeo;
> >
> > flags = msg->msg_flags;
> > + if (SK_BPF_CB_FLAG_TEST(sk, SK_BPF_CB_TX_TIMESTAMPING)) {
> > + first_write_seq = tp->write_seq;
> > + bpf_skops_tx_timestamping(sk, NULL, BPF_SOCK_OPS_TS_TCP_SND_CB);
>
> My preference is to skip this bpf callout for now and depends on a bpf trace
> program if it is really needed.
I have no idea if the bpf program wants to record the timestamp here
without the above three lines? Please enlighten me more. Thanks in
advance.
I guess there is one way which I don't know yet to monitor at the
beginning of tcp_sendmsg_locked(). Then let the bpf program call
BPF_SOCK_OPS_TS_TCP_SND_CB in the tcp_tx_timestamp() like your
previous comment with the help of bpf storage feature to correlate
them. Am I understanding right?
Thanks,
Jason
>
> > + }
> >
> > if ((flags & MSG_ZEROCOPY) && size) {
> > if (msg->msg_ubuf) {
> > @@ -1331,7 +1337,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
> >
> > out:
> > if (copied) {
> > - tcp_tx_timestamp(sk, &sockc);
> > + tcp_tx_timestamp(sk, &sockc, first_write_seq);
> > tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
> > }
> > out_nopush:
> > diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
> > index 0fe7d663a244..3769e38e052d 100644
> > --- a/tools/include/uapi/linux/bpf.h
> > +++ b/tools/include/uapi/linux/bpf.h
> > @@ -7030,6 +7030,9 @@ enum {
> > * feature is on. It indicates the
> > * recorded timestamp.
> > */
> > + BPF_SOCK_OPS_TS_TCP_SND_CB, /* Called when every tcp_sendmsg
> > + * syscall is triggered
> > + */
> > };
> >
> > /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect
>
Powered by blists - more mailing lists