1 /* Copyright (c) 2017 Facebook 1 /* Copyright (c) 2017 Facebook
2 * 2 *
3 * This program is free software; you can redi 3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of t 4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software F 5 * License as published by the Free Software Foundation.
6 * 6 *
7 * BPF program to set congestion control to dc 7 * BPF program to set congestion control to dctcp when both hosts are
8 * in the same datacenter (as deteremined by I 8 * in the same datacenter (as deteremined by IPv6 prefix).
9 * 9 *
10 * Use "bpftool cgroup attach $cg sock_ops $pr !! 10 * Use load_sock_ops to load this BPF program.
11 */ 11 */
12 12
13 #include <uapi/linux/bpf.h> 13 #include <uapi/linux/bpf.h>
14 #include <uapi/linux/tcp.h> 14 #include <uapi/linux/tcp.h>
15 #include <uapi/linux/if_ether.h> 15 #include <uapi/linux/if_ether.h>
16 #include <uapi/linux/if_packet.h> 16 #include <uapi/linux/if_packet.h>
17 #include <uapi/linux/ip.h> 17 #include <uapi/linux/ip.h>
18 #include <linux/socket.h> 18 #include <linux/socket.h>
19 #include <bpf/bpf_helpers.h> !! 19 #include "bpf_helpers.h"
20 #include <bpf/bpf_endian.h> !! 20 #include "bpf_endian.h"
21 21
22 #define DEBUG 1 22 #define DEBUG 1
>> 23
>> 24 #define bpf_printk(fmt, ...) \
>> 25 ({ \
>> 26 char ____fmt[] = fmt; \
>> 27 bpf_trace_printk(____fmt, sizeof(____fmt), \
>> 28 ##__VA_ARGS__); \
>> 29 })
23 30
24 SEC("sockops") 31 SEC("sockops")
25 int bpf_cong(struct bpf_sock_ops *skops) 32 int bpf_cong(struct bpf_sock_ops *skops)
26 { 33 {
27 char cong[] = "dctcp"; 34 char cong[] = "dctcp";
28 int rv = 0; 35 int rv = 0;
29 int op; 36 int op;
30 37
31 /* For testing purposes, only execute 38 /* For testing purposes, only execute rest of BPF program
32 * if neither port numberis 55601 39 * if neither port numberis 55601
33 */ 40 */
34 if (bpf_ntohl(skops->remote_port) != 5 41 if (bpf_ntohl(skops->remote_port) != 55601 &&
35 skops->local_port != 55601) { 42 skops->local_port != 55601) {
36 skops->reply = -1; 43 skops->reply = -1;
37 return 1; 44 return 1;
38 } 45 }
39 46
40 op = (int) skops->op; 47 op = (int) skops->op;
41 48
42 #ifdef DEBUG 49 #ifdef DEBUG
43 bpf_printk("BPF command: %d\n", op); 50 bpf_printk("BPF command: %d\n", op);
44 #endif 51 #endif
45 52
46 /* Check if both hosts are in the same 53 /* Check if both hosts are in the same datacenter. For this
47 * example they are if the 1st 5.5 byt 54 * example they are if the 1st 5.5 bytes in the IPv6 address
48 * are the same. 55 * are the same.
49 */ 56 */
50 if (skops->family == AF_INET6 && 57 if (skops->family == AF_INET6 &&
51 skops->local_ip6[0] == skops->remo 58 skops->local_ip6[0] == skops->remote_ip6[0] &&
52 (bpf_ntohl(skops->local_ip6[1]) & 59 (bpf_ntohl(skops->local_ip6[1]) & 0xfff00000) ==
53 (bpf_ntohl(skops->remote_ip6[1]) & 60 (bpf_ntohl(skops->remote_ip6[1]) & 0xfff00000)) {
54 switch (op) { 61 switch (op) {
55 case BPF_SOCK_OPS_NEEDS_ECN: 62 case BPF_SOCK_OPS_NEEDS_ECN:
56 rv = 1; 63 rv = 1;
57 break; 64 break;
58 case BPF_SOCK_OPS_ACTIVE_ESTAB 65 case BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB:
59 rv = bpf_setsockopt(sk 66 rv = bpf_setsockopt(skops, SOL_TCP, TCP_CONGESTION,
60 co 67 cong, sizeof(cong));
61 break; 68 break;
62 case BPF_SOCK_OPS_PASSIVE_ESTA 69 case BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB:
63 rv = bpf_setsockopt(sk 70 rv = bpf_setsockopt(skops, SOL_TCP, TCP_CONGESTION,
64 co 71 cong, sizeof(cong));
65 break; 72 break;
66 default: 73 default:
67 rv = -1; 74 rv = -1;
68 } 75 }
69 } else { 76 } else {
70 rv = -1; 77 rv = -1;
71 } 78 }
72 #ifdef DEBUG 79 #ifdef DEBUG
73 bpf_printk("Returning %d\n", rv); 80 bpf_printk("Returning %d\n", rv);
74 #endif 81 #endif
75 skops->reply = rv; 82 skops->reply = rv;
76 return 1; 83 return 1;
77 } 84 }
78 char _license[] SEC("license") = "GPL"; 85 char _license[] SEC("license") = "GPL";
79 86
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.