colin/passt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

tcp: Rework timers to use timerfd instead of periodic bitmap scan

Browse Source 
With a lot of concurrent connections, the bitmap scan approach is
not really sustainable.

Switch to per-connection timerfd timers, set based on events and on
two new flags, ACK_FROM_TAP_DUE and ACK_TO_TAP_DUE. Timers are added
to the common epoll list, and implement the existing timeouts.

While at it, drop the CONN_ prefix from flag names, otherwise they
get quite long, and fix the logic to decide if a connection has a
local, possibly unreachable endpoint: we shouldn't go through the
rest of tcp_conn_from_tap() if we reset the connection due to a
successful bind(2), and we'll get EACCES if the port number is low.

Suggested by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This commit is contained in:
Stefano Brivio
2022-03-18 12:18:19 +01:00
parent 3eb19cfd8a
commit be5bbb9b06
5 changed files with 288 additions and 241 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tcp.h
View File

@@ -6,7 +6,9 @@
#ifndef TCP_H
#define TCP_H
#define TCP_TIMER_INTERVAL 20 /* ms */
#define REFILL_INTERVAL 1000 /* ms */
#define PORT_DETECT_INTERVAL 1000
#define TCP_TIMER_INTERVAL MIN(REFILL_INTERVAL, PORT_DETECT_INTERVAL)
#define TCP_MAX_CONNS (128 * 1024)
#define TCP_MAX_SOCKS (TCP_MAX_CONNS + USHRT_MAX * 2)
@@ -21,7 +23,7 @@ int tcp_tap_handler(struct ctx *c, int af, void *addr,
struct tap_l4_msg *msg, int count, struct timespec *now);
int tcp_sock_init(struct ctx *c, struct timespec *now);
void tcp_timer(struct ctx *c, struct timespec *now);
void tcp_defer_handler(struct ctx *c, struct timespec *now);
void tcp_defer_handler(struct ctx *c);
void tcp_sock_set_bufsize(struct ctx *c, int s);
void tcp_update_l2_buf(unsigned char *eth_d, unsigned char *eth_s,
@@ -34,6 +36,7 @@ void tcp_remap_to_init(in_port_t port, in_port_t delta);
* @listen: Set if this file descriptor is a listening socket
* @splice: Set if descriptor is associated to a spliced connection
* @v6: Set for IPv6 sockets or connections
* @timer: Reference is a timerfd descriptor for connection
* @index: Index of connection in table, or port for bound sockets
* @u32: Opaque u32 value of reference
*/
@@ -42,6 +45,7 @@ union tcp_epoll_ref {
uint32_t listen:1,
splice:1,
v6:1,
timer:1,
index:20;
} tcp;
uint32_t u32;