aljaz/lightning
1
0
Fork 0
mirror of https://github.com/aljazceru/lightning.git synced 2025-12-24 01:24:26 +01:00
Code Issues Packages Projects Releases Wiki Activity

connectd: do nagle by packet type.

Browse Source 
channeld can't do it any more: it's using local sockets.  Connectd
can do it, and simply does it by type.

Amazingly, on my machine the timing change *always* caused
test_channel_receivable() to fail, due to a latent race.

Includes feedback from @cdecker.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
This commit is contained in:
Rusty Russell
2022-01-08 23:56:29 +10:30
parent 7a514112ec
commit e37a638c0c
7 changed files with 118 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
common/peer_io.c
View File

@@ -22,50 +22,6 @@ void peer_write(struct per_peer_state *pps, const void *msg TAKES)
peer_failed_connection_lost();
}
/* We're happy for the kernel to batch update and gossip messages, but a
* commitment message, for example, should be instantly sent. There's no
* great way of doing this, unfortunately.
*
* Setting TCP_NODELAY on Linux flushes the socket, which really means
* we'd want to toggle on then off it *after* sending. But Linux has
* TCP_CORK. On FreeBSD, it seems (looking at source) not to, so
* there we'd want to set it before the send, and reenable it
* afterwards. Even if this is wrong on other non-Linux platforms, it
* only means one extra packet.
*/
void peer_write_no_delay(struct per_peer_state *pps, const void *msg TAKES)
{
int val;
int opt;
const char *optname;
static bool complained = false;
#ifdef TCP_CORK
opt = TCP_CORK;
optname = "TCP_CORK";
#elif defined(TCP_NODELAY)
opt = TCP_NODELAY;
optname = "TCP_NODELAY";
#else
#error "Please report platform with neither TCP_CORK nor TCP_NODELAY?"
#endif
val = 1;
if (setsockopt(pps->peer_fd, IPPROTO_TCP, opt, &val, sizeof(val)) != 0) {
/* This actually happens in testing, where we blackhole the fd */
if (!complained) {
status_unusual("setsockopt %s=1: %s",
optname,
strerror(errno));
complained = true;
}
}
peer_write(pps, msg);
val = 0;
setsockopt(pps->peer_fd, IPPROTO_TCP, opt, &val, sizeof(val));
}
u8 *peer_read(const tal_t *ctx, struct per_peer_state *pps)
{
u8 *dec = wire_sync_read(ctx, pps->peer_fd);