public inbox for passt-dev@passt.top
 help / color / mirror / code / Atom feed
From: David Gibson <david@gibson.dropbear.id.au>
To: Stefano Brivio <sbrivio@redhat.com>
Cc: passt-dev@passt.top
Subject: Re: [PATCH 16/32] tcp: Use the same sockets to listen for spliced and non-spliced connections
Date: Thu, 17 Nov 2022 12:43:58 +1100	[thread overview]
Message-ID: <Y3WR3hiQy1azzhuV@yekko> (raw)
In-Reply-To: <20221117005405.70227aba@elisabeth>

[-- Attachment #1: Type: text/plain, Size: 11540 bytes --]

On Thu, Nov 17, 2022 at 12:54:05AM +0100, Stefano Brivio wrote:
> On Wed, 16 Nov 2022 15:41:56 +1100
> David Gibson <david@gibson.dropbear.id.au> wrote:
> 
> > In pasta mode, tcp_sock_init[46]() create separate sockets to listen for
> > spliced connections (these are bound to localhost) and non-spliced
> > connections (these are bound to the host address).  This introduces a
> > subtle behavioural difference between pasta and passt: by default, pasta
> > will listen only on a single host address, whereas passt will listen on
> > all addresses (0.0.0.0 or ::).  This also prevents us using some additional
> > optimizations that only work with the unspecified (0.0.0.0 or ::) address.
> > 
> > However, it turns out we don't need to do this.  We can splice a connection
> > if and only if it originates from the loopback address.  Currently we
> > ensure this by having the "spliced" listening sockets listening only on
> > loopback.  Instead, defer the decision about whether to splice a connection
> > until after accept(), by checking if the connection was made from the
> > loopback address.
> > 
> > Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
> > ---
> >  tcp.c        | 127 +++++++++++++--------------------------------------
> >  tcp_splice.c |  25 ++++++++--
> >  tcp_splice.h |   5 +-
> >  3 files changed, 55 insertions(+), 102 deletions(-)
> > 
> > diff --git a/tcp.c b/tcp.c
> > index e66a82a..4065da7 100644
> > --- a/tcp.c
> > +++ b/tcp.c
> > @@ -434,7 +434,6 @@ static const char *tcp_flag_str[] __attribute((__unused__)) = {
> >  };
> >  
> >  /* Listening sockets, used for automatic port forwarding in pasta mode only */
> > -static int tcp_sock_init_lo	[NUM_PORTS][IP_VERSIONS];
> >  static int tcp_sock_init_ext	[NUM_PORTS][IP_VERSIONS];
> >  static int tcp_sock_ns		[NUM_PORTS][IP_VERSIONS];
> >  
> > @@ -2851,21 +2850,31 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref,
> >  	socklen_t sl;
> >  	int s;
> >  
> > +	assert(ref.r.p.tcp.tcp.listen);
> > +	assert(!ref.r.p.tcp.tcp.splice);
> > +
> >  	if (c->tcp.conn_count >= TCP_MAX_CONNS)
> >  		return;
> >  
> >  	sl = sizeof(sa);
> > +	/* FIXME: Workaround clang-tidy not realizing that accept4()
> > +	 * writes the socket address.  See
> > +	 * https://github.com/llvm/llvm-project/issues/58992
> > +	 */
> > +	memset(&sa, 0, sizeof(struct sockaddr_in6));
> >  	s = accept4(ref.r.s, (struct sockaddr *)&sa, &sl, SOCK_NONBLOCK);
> 
> Ah, interesting. That looks new by the way -- not even valgrind
> complained about this.

Right, valgrind seems to have better modeling of the syscall here.
Note that this is harder for a static tool to get right, because the
amount that accept() writes is variable.

I got a reply on that bug report, saying apparently clang-tidy
*should* consider sa written here, and they weren't easily able to
reproduce the problem.  So, I'm not sure what's going on here :(.

> >  	if (s < 0)
> >  		return;
> >  
> >  	conn = tc + c->tcp.conn_count++;
> >  
> > -	if (ref.r.p.tcp.tcp.splice)
> > -		tcp_splice_conn_from_sock(c, ref, &conn->splice, s);
> > -	else
> > -		tcp_tap_conn_from_sock(c, ref, &conn->tap, s,
> > -				       (struct sockaddr *)&sa, now);
> > +	if (c->mode == MODE_PASTA &&
> > +	    tcp_splice_conn_from_sock(c, ref, &conn->splice,
> > +				      s, (struct sockaddr *)&sa))
> > +		return;
> > +
> > +	tcp_tap_conn_from_sock(c, ref, &conn->tap, s,
> > +			       (struct sockaddr *)&sa, now);
> >  }
> >  
> >  /**
> > @@ -3018,47 +3027,16 @@ static void tcp_sock_init4(const struct ctx *c, const struct in_addr *addr,
> >  {
> >  	in_port_t idx = port + c->tcp.fwd_in.delta[port];
> >  	union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.index = idx };
> > -	bool spliced = false, tap = true;
> >  	int s;
> >  
> > -	if (c->mode == MODE_PASTA) {
> > -		spliced = !addr || IN4_IS_ADDR_UNSPECIFIED(addr) ||
> > -			IN4_IS_ADDR_LOOPBACK(addr);
> > -
> > -		if (!addr)
> > -			addr = &c->ip4.addr;
> > -
> > -		tap = !IN4_IS_ADDR_LOOPBACK(addr);
> > -	}
> > -
> > -	if (tap) {
> > -		s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port,
> > -			    tref.u32);
> > -		if (s >= 0)
> > -			tcp_sock_set_bufsize(c, s);
> > -		else
> > -			s = -1;
> > -
> > -		if (c->tcp.fwd_in.mode == FWD_AUTO)
> > -			tcp_sock_init_ext[port][V4] = s;
> > -	}
> > -
> > -	if (spliced) {
> > -		struct in_addr loopback = { htonl(INADDR_LOOPBACK) };
> > -		tref.tcp.splice = 1;
> > -
> > -		addr = &loopback;
> > -
> > -		s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port,
> > -			    tref.u32);
> > -		if (s >= 0)
> > -			tcp_sock_set_bufsize(c, s);
> > -		else
> > -			s = -1;
> > +	s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port, tref.u32);
> > +	if (s >= 0)
> > +		tcp_sock_set_bufsize(c, s);
> > +	else
> > +		s = -1;
> >  
> > -		if (c->tcp.fwd_out.mode == FWD_AUTO)
> > -			tcp_sock_init_lo[port][V4] = s;
> > -	}
> > +	if (c->tcp.fwd_in.mode == FWD_AUTO)
> > +		tcp_sock_init_ext[port][V4] = s;
> >  }
> >  
> >  /**
> > @@ -3075,47 +3053,16 @@ static void tcp_sock_init6(const struct ctx *c,
> >  	in_port_t idx = port + c->tcp.fwd_in.delta[port];
> >  	union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.v6 = 1,
> >  				     .tcp.index = idx	};
> > -	bool spliced = false, tap = true;
> >  	int s;
> >  
> > -	if (c->mode == MODE_PASTA) {
> > -		spliced = !addr ||
> > -			  IN6_IS_ADDR_UNSPECIFIED(addr) ||
> > -			  IN6_IS_ADDR_LOOPBACK(addr);
> > -
> > -		if (!addr)
> > -			addr = &c->ip6.addr;
> > -
> > -		tap = !IN6_IS_ADDR_LOOPBACK(addr);
> > -	}
> > -
> > -	if (tap) {
> > -		s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port,
> > -			    tref.u32);
> > -		if (s >= 0)
> > -			tcp_sock_set_bufsize(c, s);
> > -		else
> > -			s = -1;
> > -
> > -		if (c->tcp.fwd_in.mode == FWD_AUTO)
> > -			tcp_sock_init_ext[port][V6] = s;
> > -	}
> > -
> > -	if (spliced) {
> > -		tref.tcp.splice = 1;
> > -
> > -		addr = &in6addr_loopback;
> > -
> > -		s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port,
> > -			    tref.u32);
> > -		if (s >= 0)
> > -			tcp_sock_set_bufsize(c, s);
> > -		else
> > -			s = -1;
> > +	s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, tref.u32);
> > +	if (s >= 0)
> > +		tcp_sock_set_bufsize(c, s);
> > +	else
> > +		s = -1;
> >  
> > -		if (c->tcp.fwd_out.mode == FWD_AUTO)
> > -			tcp_sock_init_lo[port][V6] = s;
> > -	}
> > +	if (c->tcp.fwd_in.mode == FWD_AUTO)
> > +		tcp_sock_init_ext[port][V6] = s;
> >  }
> >  
> >  /**
> > @@ -3144,7 +3091,7 @@ static void tcp_ns_sock_init4(const struct ctx *c, in_port_t port)
> >  {
> >  	in_port_t idx = port + c->tcp.fwd_out.delta[port];
> >  	union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1,
> > -				     .tcp.splice = 1, .tcp.index = idx };
> > +				     .tcp.index = idx };
> >  	struct in_addr loopback = { htonl(INADDR_LOOPBACK) };
> >  	int s;
> >  
> > @@ -3169,8 +3116,7 @@ static void tcp_ns_sock_init6(const struct ctx *c, in_port_t port)
> >  {
> >  	in_port_t idx = port + c->tcp.fwd_out.delta[port];
> >  	union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1,
> > -				     .tcp.splice = 1, .tcp.v6 = 1,
> > -				     .tcp.index = idx};
> > +				     .tcp.v6 = 1, .tcp.index = idx};
> 
> Space missing here (from 14/32).

Fixed.

> >  	int s;
> >  
> >  	assert(c->mode == MODE_PASTA);
> > @@ -3337,7 +3283,6 @@ int tcp_init(struct ctx *c)
> >  	memset(init_sock_pool6,		0xff,	sizeof(init_sock_pool6));
> >  	memset(ns_sock_pool4,		0xff,	sizeof(ns_sock_pool4));
> >  	memset(ns_sock_pool6,		0xff,	sizeof(ns_sock_pool6));
> > -	memset(tcp_sock_init_lo,	0xff,	sizeof(tcp_sock_init_lo));
> >  	memset(tcp_sock_init_ext,	0xff,	sizeof(tcp_sock_init_ext));
> >  	memset(tcp_sock_ns,		0xff,	sizeof(tcp_sock_ns));
> >  
> > @@ -3445,16 +3390,6 @@ static int tcp_port_rebind(void *arg)
> >  					close(tcp_sock_init_ext[port][V6]);
> >  					tcp_sock_init_ext[port][V6] = -1;
> >  				}
> > -
> > -				if (tcp_sock_init_lo[port][V4] >= 0) {
> > -					close(tcp_sock_init_lo[port][V4]);
> > -					tcp_sock_init_lo[port][V4] = -1;
> > -				}
> > -
> > -				if (tcp_sock_init_lo[port][V6] >= 0) {
> > -					close(tcp_sock_init_lo[port][V6]);
> > -					tcp_sock_init_lo[port][V6] = -1;
> > -				}
> >  				continue;
> >  			}
> >  
> > diff --git a/tcp_splice.c b/tcp_splice.c
> > index 7007501..30d49d4 100644
> > --- a/tcp_splice.c
> > +++ b/tcp_splice.c
> > @@ -502,19 +502,35 @@ static void tcp_splice_dir(struct tcp_splice_conn *conn, int ref_sock,
> >  }
> >  
> >  /**
> > - * tcp_splice_conn_from_sock() - Initialize state for spliced connection
> > + * tcp_splice_conn_from_sock() - Attempt to init state for a spliced connection
> >   * @c:		Execution context
> >   * @ref:	epoll reference of listening socket
> >   * @conn:	connection structure to initialize
> >   * @s:		Accepted socket
> > + * @sa:		Peer address of connection
> >   *
> > + * Return: true if able to create a spliced connection, false otherwise
> >   * #syscalls:pasta setsockopt
> >   */
> > -void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref,
> > -			       struct tcp_splice_conn *conn, int s)
> > +bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref,
> > +			       struct tcp_splice_conn *conn, int s,
> > +			       const struct sockaddr *sa)
> >  {
> >  	assert(c->mode == MODE_PASTA);
> >  
> > +	if (ref.r.p.tcp.tcp.v6) {
> > +		const struct sockaddr_in6 *sa6
> > +			= (const struct sockaddr_in6 *)sa;
> 
> Maybe you could split declaration and assignment here.

Good idea, done.

> > +		if (!IN6_IS_ADDR_LOOPBACK(&sa6->sin6_addr))
> > +			return false;
> > +		conn->flags = SPLICE_V6;
> > +	} else {
> > +		const struct sockaddr_in *sa4 = (const struct sockaddr_in *)sa;
> > +		if (!IN4_IS_ADDR_LOOPBACK(&sa4->sin_addr))
> > +			return false;
> > +		conn->flags = 0;
> > +	}
> > +
> >  	if (setsockopt(s, SOL_TCP, TCP_QUICKACK, &((int){ 1 }),
> >  		       sizeof(int))) {
> >  		trace("TCP (spliced): failed to set TCP_QUICKACK on %i",
> > @@ -524,11 +540,12 @@ void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref,
> >  	conn->c.spliced = true;
> >  	c->tcp.splice_conn_count++;
> >  	conn->a = s;
> > -	conn->flags = ref.r.p.tcp.tcp.v6 ? SPLICE_V6 : 0;
> >  
> >  	if (tcp_splice_new(c, conn, ref.r.p.tcp.tcp.index,
> >  			   ref.r.p.tcp.tcp.outbound))
> >  		conn_flag(c, conn, CLOSING);
> > +
> > +	return true;
> >  }
> >  
> >  /**
> > diff --git a/tcp_splice.h b/tcp_splice.h
> > index f9462ae..1a915dd 100644
> > --- a/tcp_splice.h
> > +++ b/tcp_splice.h
> > @@ -10,8 +10,9 @@ struct tcp_splice_conn;
> >  
> >  void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref,
> >  			     uint32_t events);
> > -void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref,
> > -			       struct tcp_splice_conn *conn, int s);
> > +bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref,
> > +			       struct tcp_splice_conn *conn, int s,
> > +			       const struct sockaddr *sa);
> >  void tcp_splice_init(struct ctx *c);
> >  
> >  #endif /* TCP_SPLICE_H */
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

  reply	other threads:[~2022-11-17  2:08 UTC|newest]

Thread overview: 57+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-11-16  4:41 [PATCH 00/32] Use dual stack sockets to listen for inbound TCP connections David Gibson
2022-11-16  4:41 ` [PATCH 01/32] clang-tidy: Suppress warning about assignments in if statements David Gibson
2022-11-16 23:10   ` Stefano Brivio
2022-11-17  1:20     ` David Gibson
2022-11-16  4:41 ` [PATCH 02/32] style: Minor corrections to function comments David Gibson
2022-11-16 23:11   ` Stefano Brivio
2022-11-17  1:21     ` David Gibson
2022-11-16  4:41 ` [PATCH 03/32] tcp_splice: #include tcp_splice.h in tcp_splice.c David Gibson
2022-11-16  4:41 ` [PATCH 04/32] tcp: Remove unused TCP_MAX_SOCKS constant David Gibson
2022-11-16  4:41 ` [PATCH 05/32] tcp: Better helpers for converting between connection pointer and index David Gibson
2022-11-16 23:11   ` Stefano Brivio
2022-11-17  1:24     ` David Gibson
2022-11-16  4:41 ` [PATCH 06/32] tcp_splice: Helpers for converting from index to/from tcp_splice_conn David Gibson
2022-11-16  4:41 ` [PATCH 07/32] tcp: Move connection state structures into a shared header David Gibson
2022-11-16  4:41 ` [PATCH 08/32] tcp: Add connection union type David Gibson
2022-11-16  4:41 ` [PATCH 09/32] tcp: Improved helpers to update connections after moving David Gibson
2022-11-16  4:41 ` [PATCH 10/32] tcp: Unify spliced and non-spliced connection tables David Gibson
2022-11-16  4:41 ` [PATCH 11/32] tcp: Unify tcp_defer_handler and tcp_splice_defer_handler() David Gibson
2022-11-16  4:41 ` [PATCH 12/32] tcp: Partially unify tcp_timer() and tcp_splice_timer() David Gibson
2022-11-16  4:41 ` [PATCH 13/32] tcp: Unify the IN_EPOLL flag David Gibson
2022-11-16  4:41 ` [PATCH 14/32] tcp: Separate helpers to create ns listening sockets David Gibson
2022-11-16 23:51   ` Stefano Brivio
2022-11-17  1:32     ` David Gibson
2022-11-16  4:41 ` [PATCH 15/32] tcp: Unify part of spliced and non-spliced conn_from_sock path David Gibson
2022-11-16 23:53   ` Stefano Brivio
2022-11-17  1:37     ` David Gibson
2022-11-17  7:30       ` Stefano Brivio
2022-11-17  8:58         ` David Gibson
2022-11-16  4:41 ` [PATCH 16/32] tcp: Use the same sockets to listen for spliced and non-spliced connections David Gibson
2022-11-16 23:54   ` Stefano Brivio
2022-11-17  1:43     ` David Gibson [this message]
2022-11-16  4:41 ` [PATCH 17/32] tcp: Remove splice from tcp_epoll_ref David Gibson
2022-11-16  4:41 ` [PATCH 18/32] tcp: Don't store hash bucket in connection structures David Gibson
2022-11-16  4:41 ` [PATCH 19/32] inany: Helper functions for handling addresses which could be IPv4 or IPv6 David Gibson
2022-11-16 23:54   ` Stefano Brivio
2022-11-17  1:48     ` David Gibson
2022-11-16  4:42 ` [PATCH 20/32] tcp: Hash IPv4 and IPv4-mapped-IPv6 addresses the same David Gibson
2022-11-16  4:42 ` [PATCH 21/32] tcp: Take tcp_hash_insert() address from struct tcp_conn David Gibson
2022-11-16  4:42 ` [PATCH 22/32] tcp: Simplify tcp_hash_match() to take an inany_addr David Gibson
2022-11-16  4:42 ` [PATCH 23/32] tcp: Unify initial sequence number calculation for IPv4 and IPv6 David Gibson
2022-11-16  4:42 ` [PATCH 24/32] tcp: Have tcp_seq_init() take its parameters from struct tcp_conn David Gibson
2022-11-16  4:42 ` [PATCH 25/32] tcp: Fix small errors in tcp_seq_init() time handling David Gibson
2022-11-16  4:42 ` [PATCH 26/32] tcp: Remove v6 flag from tcp_epoll_ref David Gibson
2022-11-17  0:15   ` Stefano Brivio
2022-11-17  1:50     ` David Gibson
2022-11-16  4:42 ` [PATCH 27/32] tcp: NAT IPv4-mapped IPv6 addresses like IPv4 addresses David Gibson
2022-11-17  0:15   ` Stefano Brivio
2022-11-17  2:00     ` David Gibson
2022-11-16  4:42 ` [PATCH 28/32] tcp_splice: Allow splicing of connections from IPv4-mapped loopback David Gibson
2022-11-17  0:15   ` Stefano Brivio
2022-11-17  2:05     ` David Gibson
2022-11-16  4:42 ` [PATCH 29/32] tcp: Consolidate tcp_sock_init[46] David Gibson
2022-11-16  4:42 ` [PATCH 30/32] util: Allow sock_l4() to open dual stack sockets David Gibson
2022-11-16  4:42 ` [PATCH 31/32] util: Always return -1 on error in sock_l4() David Gibson
2022-11-16  4:42 ` [PATCH 32/32] tcp: Use dual stack sockets for port forwarding when possible David Gibson
2022-11-17  0:15   ` Stefano Brivio
2022-11-17  2:08     ` David Gibson

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=Y3WR3hiQy1azzhuV@yekko \
    --to=david@gibson.dropbear.id.au \
    --cc=passt-dev@passt.top \
    --cc=sbrivio@redhat.com \
    /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.
Code repositories for project(s) associated with this public inbox

	https://passt.top/passt

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 IMAP folder(s).