Skip to contentSkip to conversation

fail io_uring zc with sockets not supporting it

PATCH Applied Pavel Begunkov 5 participants 16 messages Oct 21, 2022 – Oct 24, 2022

Subsystems (5)
Networking Network drivers TCP io_uring +1 more subsystems
lore
Share & tools
lore

Pavel Begunkov

author
#0 lore raw

cover fail io_uring zc with sockets not supporting it

Some sockets don't care about msghdr::ubuf_info and would execute the request by copying data. Such fallback behaviour was always a pain in my experience, so we'd rather want to fail such requests and have a more robust api in the future.

Mark struct socket that support it with a new SOCK_SUPPORT_ZC flag. I'm not entirely sure it's the best place for the flag but at least we don't have to do a bunch of extra dereferences in the hot path.

P.S. patches 2 and 3 are not combined for backporting purposes.

Pavel Begunkov (3):
net: flag sockets supporting msghdr originated zerocopy io_uring/net: fail zc send when unsupported by socket io_uring/net: fail zc sendmsg when unsupported by socket

 include/linux/net.h | 1 +
 io_uring/net.c      | 4 ++++
 net/ipv4/tcp.c      | 1 +
 net/ipv4/udp.c      | 1 +
 4 files changed, 7 insertions(+)
Signature · 1 line
2.38.0
author
#1 lore raw

PATCH 1/3 net: flag sockets supporting msghdr originated zerocopy

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>
---
 include/linux/net.h | 1 +
 net/ipv4/tcp.c      | 1 +
 net/ipv4/udp.c      | 1 +
 3 files changed, 3 insertions(+)
include/linux/net.h +1 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
 #define SOCK_NOSPACE		2
 #define SOCK_PASSCRED		3
 #define SOCK_PASSSEC		4
+#define SOCK_SUPPORT_ZC		5
 
 #ifndef ARCH_HAS_SOCKET_TYPES
 /**
net/ipv4/tcp.c +1 -0
index 0c51abeee172..aeb7b9eaddc7 100644
@@ -457,6 +457,7 @@ void tcp_init_sock(struct sock *sk)
 	WRITE_ONCE(sk->sk_sndbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[1]));
 	WRITE_ONCE(sk->sk_rcvbuf, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[1]));
 
+	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
 	sk_sockets_allocated_inc(sk);
 }
 EXPORT_SYMBOL(tcp_init_sock);
net/ipv4/udp.c +1 -0
index d63118ce5900..03616fc5162c 100644
@@ -1620,6 +1620,7 @@ int udp_init_sock(struct sock *sk)
 {
 	skb_queue_head_init(&udp_sk(sk)->reader_queue);
 	sk->sk_destruct = udp_destruct_sock;
+	set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
 	return 0;
 }
 EXPORT_SYMBOL_GPL(udp_init_sock);
Signature · 1 line
2.38.0
maintainer Acked-by on v1 1/3
#2 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Quoted 17 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>
---
 include/linux/net.h | 1 +
 net/ipv4/tcp.c      | 1 +
 net/ipv4/udp.c      | 1 +
 3 files changed, 3 insertions(+)
include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644

include/linux/net.h struct net; · line 44

Quoted 4 lines
include/linux/net.h +1 -0
@@ -41,6 +41,7 @@ struct net;
 #define SOCK_NOSPACE		2
 #define SOCK_PASSCRED		3
 #define SOCK_PASSSEC		4
+#define SOCK_SUPPORT_ZC		5
Acked-by Jakub Kicinski <[email protected]>

Any idea on when this will make it to Linus? If within a week we can probably delay:
https://lore.kernel.org/all/dc549a4d5c1d2031c64794c8e12bed55afb85c3e.1666287924.git.pabeni@redhat.com/
and avoid the conflict.

author
#3 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Quoted 25 lines · Jakub Kicinski
Quoted 17 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>
---
  include/linux/net.h | 1 +
  net/ipv4/tcp.c      | 1 +
  net/ipv4/udp.c      | 1 +
  3 files changed, 3 insertions(+)
include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644

include/linux/net.h struct net; · line 44

Quoted 4 lines
include/linux/net.h +1 -0
@@ -41,6 +41,7 @@ struct net;
  #define SOCK_NOSPACE		2
  #define SOCK_PASSCRED		3
  #define SOCK_PASSSEC		4
+#define SOCK_SUPPORT_ZC		5
Acked-by Jakub Kicinski <[email protected]>
Quoted 2 lines

Any idea on when this will make it to Linus? If within a week we can probably delay:

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Quoted 2 lines

https://lore.kernel.org/all/dc549a4d5c1d2031c64794c8e12bed55afb85c3e.1666287924.git.pabeni@redhat.com/
and avoid the conflict.

Signature · 1 line
Pavel Begunkov
maintainer
#4 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Quoted 26 lines · Pavel Begunkov
Quoted 25 lines · Jakub Kicinski
Quoted 17 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>

---

? include/linux/net.h | 1 +
? net/ipv4/tcp.c????? | 1 +
? net/ipv4/udp.c????? | 1 +

? 3 files changed, 3 insertions(+)

include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
Quoted 4 lines

? #define SOCK_NOSPACE??????? 2
? #define SOCK_PASSCRED??????? 3
? #define SOCK_PASSSEC??????? 4
+#define SOCK_SUPPORT_ZC??????? 5

Acked-by Jakub Kicinski <[email protected]>
Quoted 5 lines
Quoted 2 lines

Any idea on when this will make it to Linus? If within a week we can probably delay:

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Yes, and let's retain the name for now, can always be changed if we need to make it more granular. I'll ship this off before -rc2.

Signature · 1 line
Jens Axboe
#5 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Am 22.10.22 um 18:07 schrieb Jens Axboe:

Quoted 27 lines · Pavel Begunkov
Quoted 26 lines · Jakub Kicinski
Quoted 25 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:
Quoted 17 lines

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>

---

? include/linux/net.h | 1 +
? net/ipv4/tcp.c????? | 1 +
? net/ipv4/udp.c????? | 1 +

? 3 files changed, 3 insertions(+)

include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
Quoted 4 lines

? #define SOCK_NOSPACE??????? 2
? #define SOCK_PASSCRED??????? 3
? #define SOCK_PASSSEC??????? 4
+#define SOCK_SUPPORT_ZC??????? 5

Acked-by Jakub Kicinski <[email protected]>
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Any idea on when this will make it to Linus? If within a week we can probably delay:

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Yes, and let's retain the name for now, can always be changed if we need to make it more granular. I'll ship this off before -rc2.

I'm now always getting -EOPNOTSUPP from SENDMSG_ZC for tcp connections...

metze

#6 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Am 24.10.22 um 14:49 schrieb Stefan Metzmacher:

Quoted 31 lines

Am 22.10.22 um 18:07 schrieb Jens Axboe:

Quoted 30 lines · Pavel Begunkov
Quoted 29 lines · Jakub Kicinski
Quoted 27 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:
Quoted 17 lines

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>

---

? include/linux/net.h | 1 +
? net/ipv4/tcp.c????? | 1 +
? net/ipv4/udp.c????? | 1 +

? 3 files changed, 3 insertions(+)

include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
Quoted 4 lines

? #define SOCK_NOSPACE??????? 2
? #define SOCK_PASSCRED??????? 3
? #define SOCK_PASSSEC??????? 4
+#define SOCK_SUPPORT_ZC??????? 5

Acked-by Jakub Kicinski <[email protected]>

Any idea on when this will make it to Linus? If within a week we can probably delay:

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Yes, and let's retain the name for now, can always be changed if we need to make it more granular. I'll ship this off before -rc2.

I'm now always getting -EOPNOTSUPP from SENDMSG_ZC for tcp connections...

The problem is that inet_accept() doesn't set SOCK_SUPPORT_ZC...

This hack below fixes it for me, but I'm sure there's a nicer way...

The natural way would be to overload inet_csk_accept for tcp, but sk1->sk_prot->accept() is called with sk2->sk_socket == NULL, because sock_graft() is called later...

metze

net/ipv4/af_inet.c +3 -0
index 3dd02396517d..0f35f2a32756 100644
@@ -736,6 +736,7 @@ EXPORT_SYMBOL(inet_stream_connect);
  *	Accept a pending connection. The TCP layer now gives BSD semantics.
  */
 
+#include <net/transp_v6.h>
 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 		bool kern)
 {

@@ -754,6 +755,8 @@ int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 
+	if (sk2->sk_prot == &tcp_prot || sk2->sk_prot == &tcpv6_prot)
+		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 	sock_graft(sk2, newsock);
 
 	newsock->state = SS_CONNECTED;
author
#7 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Quoted 35 lines · Stefan Metzmacher

Am 24.10.22 um 14:49 schrieb Stefan Metzmacher:

Quoted 34 lines

Am 22.10.22 um 18:07 schrieb Jens Axboe:

Quoted 33 lines · Pavel Begunkov
Quoted 31 lines · Jakub Kicinski
Quoted 25 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:
Quoted 17 lines

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>

---

? include/linux/net.h | 1 +
? net/ipv4/tcp.c????? | 1 +
? net/ipv4/udp.c????? | 1 +

? 3 files changed, 3 insertions(+)

include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
Quoted 4 lines

? #define SOCK_NOSPACE??????? 2
? #define SOCK_PASSCRED??????? 3
? #define SOCK_PASSSEC??????? 4
+#define SOCK_SUPPORT_ZC??????? 5

Acked-by Jakub Kicinski <[email protected]>
Quoted 2 lines

Any idea on when this will make it to Linus? If within a week we can probably delay:

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Yes, and let's retain the name for now, can always be changed if we need to make it more granular. I'll ship this off before -rc2.

I'm now always getting -EOPNOTSUPP from SENDMSG_ZC for tcp connections...

The problem is that inet_accept() doesn't set SOCK_SUPPORT_ZC...

Yeah, you're right, accept is not handled, not great. Thanks for tracking it down. And I'll add a test for it

Quoted 5 lines

This hack below fixes it for me, but I'm sure there's a nicer way...

The natural way would be to overload inet_csk_accept for tcp, but sk1->sk_prot->accept() is called with sk2->sk_socket == NULL, because sock_graft() is called later...

I think it's acceptable for a fix. sk_is_tcp() sounds a bit better assuming that sk_type/protocol are set there.

Quoted 21 lines
net/ipv4/af_inet.c +3 -0
index 3dd02396517d..0f35f2a32756 100644
@@ -736,6 +736,7 @@ EXPORT_SYMBOL(inet_stream_connect);
   *    Accept a pending connection. The TCP layer now gives BSD semantics.
   */
 
+#include <net/transp_v6.h>
  int inet_accept(struct socket *sock, struct socket *newsock, int flags,
          bool kern)
  {

@@ -754,6 +755,8 @@ int inet_accept(struct socket *sock, struct socket *newsock, int flags,
            (TCPF_ESTABLISHED | TCPF_SYN_RECV |
            TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 
+    if (sk2->sk_prot == &tcp_prot || sk2->sk_prot == &tcpv6_prot)
+        set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
      sock_graft(sk2, newsock);
 
      newsock->state = SS_CONNECTED;
Signature · 1 line
Pavel Begunkov
#8 lore raw

Re: [PATCH for-6.1 1/3] net: flag sockets supporting msghdr originated zerocopy

Am 24.10.22 um 19:00 schrieb Pavel Begunkov:

Quoted 73 lines · Stefan Metzmacher
Quoted 35 lines

Am 24.10.22 um 14:49 schrieb Stefan Metzmacher:

Quoted 34 lines

Am 22.10.22 um 18:07 schrieb Jens Axboe:

Quoted 33 lines · Pavel Begunkov
Quoted 31 lines · Jakub Kicinski
Quoted 25 lines · On Fri, 21 Oct 2022 11:16:39 +0100 Pavel Begunkov wrote:
Quoted 17 lines

We need an efficient way in io_uring to check whether a socket supports zerocopy with msghdr provided ubuf_info. Add a new flag into the struct socket flags fields.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>

---

? include/linux/net.h | 1 +
? net/ipv4/tcp.c????? | 1 +
? net/ipv4/udp.c????? | 1 +

? 3 files changed, 3 insertions(+)

include/linux/net.h +0 -0
index 711c3593c3b8..18d942bbdf6e 100644
@@ -41,6 +41,7 @@ struct net;
Quoted 4 lines

? #define SOCK_NOSPACE??????? 2
? #define SOCK_PASSCRED??????? 3
? #define SOCK_PASSSEC??????? 4
+#define SOCK_SUPPORT_ZC??????? 5

Acked-by Jakub Kicinski <[email protected]>
Quoted 2 lines

Any idea on when this will make it to Linus? If within a week we can probably delay:

After a chat with Jens, IIUC he can take it and send out to Linus early. So, sounds like a good plan

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Yes, and let's retain the name for now, can always be changed if we need to make it more granular. I'll ship this off before -rc2.

I'm now always getting -EOPNOTSUPP from SENDMSG_ZC for tcp connections...

The problem is that inet_accept() doesn't set SOCK_SUPPORT_ZC...

Yeah, you're right, accept is not handled, not great. Thanks for tracking it down. And I'll add a test for it

Quoted 5 lines

This hack below fixes it for me, but I'm sure there's a nicer way...

The natural way would be to overload inet_csk_accept for tcp, but sk1->sk_prot->accept() is called with sk2->sk_socket == NULL, because sock_graft() is called later...

I think it's acceptable for a fix. sk_is_tcp() sounds a bit better assuming that sk_type/protocol are set there.

Quoted 16 lines
net/ipv4/af_inet.c +1 -0
index 3dd02396517d..0f35f2a32756 100644
@@ -736,6 +736,7 @@ EXPORT_SYMBOL(inet_stream_connect);
    *    Accept a pending connection. The TCP layer now gives BSD semantics.
    */
 
+#include <net/transp_v6.h>
   int inet_accept(struct socket *sock, struct socket *newsock, int flags,
           bool kern)
   {

@@ -754,6 +755,8 @@ int inet_accept(struct socket *sock, struct socket *newsock, int flags,

            (TCPF_ESTABLISHED | TCPF_SYN_RECV |
            TCPF_CLOSE_WAIT | TCPF_CLOSE)));

Quoted 3 lines
Quoted 3 lines

+    if (sk2->sk_prot == &tcp_prot || sk2->sk_prot == &tcpv6_prot) +        set_bit(SOCK_SUPPORT_ZC, &newsock->flags); sock_graft(sk2, newsock);

Hmm, I think we just need to check if the flag was set on the listening socket, otherwise we have a hard coded value again...

This also works for me, and seems like a much nicer patch:

net/ipv4/af_inet.c +2 -0
index 3dd02396517d..4728087c42a5 100644
@@ -754,6 +754,8 @@ int inet_accept(struct socket *sock, struct socket *newsock, int flags,
 		  (TCPF_ESTABLISHED | TCPF_SYN_RECV |
 		  TCPF_CLOSE_WAIT | TCPF_CLOSE)));
 
+	if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
+		set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
 	sock_graft(sk2, newsock);
 
 	newsock->state = SS_CONNECTED;
author
#9 lore raw

PATCH 2/3 io_uring/net: fail zc send when unsupported by socket

In reply to Pavel Begunkov

If a protocol doesn't support zerocopy it will silently fall back to copying. This type of behaviour has always been a source of troubles so it's better to fail such requests instead.

Cc <[email protected]> # 6.0
Signed-off-by Pavel Begunkov <[email protected]>
---
 io_uring/net.c | 2 ++
 1 file changed, 2 insertions(+)
io_uring/net.c +2 -0
index 8c7226b5bf41..26ff3675214d 100644
@@ -1056,6 +1056,8 @@ int io_send_zc(struct io_kiocb *req, unsigned int issue_flags)
 	sock = sock_from_file(req->file);
 	if (unlikely(!sock))
 		return -ENOTSOCK;
+	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
+		return -EOPNOTSUPP;
 
 	msg.msg_name = NULL;
 	msg.msg_control = NULL;
Signature · 1 line
2.38.0
author
#10 lore raw

PATCH 3/3 io_uring/net: fail zc sendmsg when unsupported by socket

In reply to Pavel Begunkov

The previous patch fails zerocopy send requests for protocols that don't support it, do the same for zerocopy sendmsg.

Signed-off-by Pavel Begunkov <[email protected]>
---
 io_uring/net.c | 2 ++
 1 file changed, 2 insertions(+)
io_uring/net.c +2 -0
index 26ff3675214d..15dea91625e2 100644
@@ -1153,6 +1153,8 @@ int io_sendmsg_zc(struct io_kiocb *req, unsigned int issue_flags)
 	sock = sock_from_file(req->file);
 	if (unlikely(!sock))
 		return -ENOTSOCK;
+	if (!test_bit(SOCK_SUPPORT_ZC, &sock->flags))
+		return -EOPNOTSUPP;
 
 	if (req_has_async_data(req)) {
 		kmsg = req->async_data;
Signature · 1 line
2.38.0
#11 lore raw
In reply to Pavel Begunkov

Hi Pavel,

Quoted 8 lines

Some sockets don't care about msghdr::ubuf_info and would execute the request by copying data. Such fallback behaviour was always a pain in my experience, so we'd rather want to fail such requests and have a more robust api in the future.

Mark struct socket that support it with a new SOCK_SUPPORT_ZC flag. I'm not entirely sure it's the best place for the flag but at least we don't have to do a bunch of extra dereferences in the hot path.

I'd give the flag another name that indicates msg_ubuf and
have a 2nd flag that can indicate support for SO_ZEROCOPY in sk_setsockopt()

The SO_ZEROCOPY version is also provided by AF_RDS.

metze

author
#12 lore raw
Quoted 7 lines · Stefan Metzmacher

Hi Pavel,

Quoted 5 lines

Some sockets don't care about msghdr::ubuf_info and would execute the request by copying data. Such fallback behaviour was always a pain in my experience, so we'd rather want to fail such requests and have a more robust api in the future.

Quoted 5 lines
Quoted 3 lines

Mark struct socket that support it with a new SOCK_SUPPORT_ZC flag. I'm not entirely sure it's the best place for the flag but at least we don't have to do a bunch of extra dereferences in the hot path.

I'd give the flag another name that indicates msg_ubuf and

Could be renamed, e.g. SOCK_SUPPORT_MSGHDR_UBUF or maybe SOCK_SUPPORT_EXTERNAL_UBUF

Quoted 1 line

have a 2nd flag that can indicate support for SO_ZEROCOPY in sk_setsockopt()

There is absolutely no reason to introduce a second flag here, it has nothing to do with SO_ZEROCOPY.

Quoted 1 line

The SO_ZEROCOPY version is also provided by AF_RDS.

Signature · 1 line
Pavel Begunkov
#13 lore raw

Am 21.10.22 um 12:42 schrieb Pavel Begunkov:

Quoted 8 lines · Stefan Metzmacher
Quoted 7 lines

Hi Pavel,

Quoted 5 lines

Some sockets don't care about msghdr::ubuf_info and would execute the request by copying data. Such fallback behaviour was always a pain in my experience, so we'd rather want to fail such requests and have a more robust api in the future.

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Mark struct socket that support it with a new SOCK_SUPPORT_ZC flag. I'm not entirely sure it's the best place for the flag but at least we don't have to do a bunch of extra dereferences in the hot path.

I'd give the flag another name that indicates msg_ubuf and

Could be renamed, e.g. SOCK_SUPPORT_MSGHDR_UBUF

That's good or SOCK_SUPPORT_ZC_MSGHDR_UBUF.

Quoted 4 lines
Quoted 1 line

have a 2nd flag that can indicate support for SO_ZEROCOPY in sk_setsockopt()

There is absolutely no reason to introduce a second flag here, it has nothing to do with SO_ZEROCOPY.

I meant as a separate change to replace the hard coded logic in sk_setsockopt()... But I don't care much about it, it's unlikely that I ever want to use SO_ZEROCOPY...

metze

CI Applied
#14 lore raw
In reply to Pavel Begunkov
This series was applied to netdev/net.git (master) by Jens Axboe <[email protected]>:

Hello:

This series was applied to netdev/net.git (master) by Jens Axboe <[email protected]>:

Quoted 10 lines · On Fri, 21 Oct 2022 11:16:38 +0100 you wrote:

Some sockets don't care about msghdr::ubuf_info and would execute the request by copying data. Such fallback behaviour was always a pain in my experience, so we'd rather want to fail such requests and have a more robust api in the future.

Mark struct socket that support it with a new SOCK_SUPPORT_ZC flag. I'm not entirely sure it's the best place for the flag but at least we don't have to do a bunch of extra dereferences in the hot path.

[...]

Here is the summary with links:
- [for-6.1,1/3] net: flag sockets supporting msghdr originated zerocopy

    https://git.kernel.org/netdev/net/c/e993ffe3da4b
  - [for-6.1,2/3] io_uring/net: fail zc send when unsupported by socket
    https://git.kernel.org/netdev/net/c/edf81438799c
  - [for-6.1,3/3] io_uring/net: fail zc sendmsg when unsupported by socket
    https://git.kernel.org/netdev/net/c/cc767e7c6913

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
maintainer
#15 lore raw
Quoted 7 lines · On Mon, 24 Oct 2022 20:40:38 +0000 [email protected] wrote:

Here is the summary with links:
- [for-6.1,1/3] net: flag sockets supporting msghdr originated zerocopy

    https://git.kernel.org/netdev/net/c/e993ffe3da4b
  - [for-6.1,2/3] io_uring/net: fail zc send when unsupported by socket
    https://git.kernel.org/netdev/net/c/edf81438799c
  - [for-6.1,3/3] io_uring/net: fail zc sendmsg when unsupported by socket
    https://git.kernel.org/netdev/net/c/cc767e7c6913

Ugh, ignore, looks like we forgot to drop this series from our PW and bot thought we've applied it when we forwarded the tree.