• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0
2  *
3  * Copyright (c) 2019 Facebook
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of version 2 of the GNU General Public
7  * License as published by the Free Software Foundation.
8  *
9  * Include file for sample Host Bandwidth Manager (HBM) BPF programs
10  */
11 #define KBUILD_MODNAME "foo"
12 #include <stddef.h>
13 #include <stdbool.h>
14 #include <uapi/linux/bpf.h>
15 #include <uapi/linux/if_ether.h>
16 #include <uapi/linux/if_packet.h>
17 #include <uapi/linux/ip.h>
18 #include <uapi/linux/ipv6.h>
19 #include <uapi/linux/in.h>
20 #include <uapi/linux/tcp.h>
21 #include <uapi/linux/filter.h>
22 #include <uapi/linux/pkt_cls.h>
23 #include <net/ipv6.h>
24 #include <net/inet_ecn.h>
25 #include "bpf_endian.h"
26 #include "bpf_helpers.h"
27 #include "hbm.h"
28 
29 #define DROP_PKT	0
30 #define ALLOW_PKT	1
31 #define TCP_ECN_OK	1
32 #define CWR		2
33 
34 #ifndef HBM_DEBUG  // Define HBM_DEBUG to enable debugging
35 #undef bpf_printk
36 #define bpf_printk(fmt, ...)
37 #endif
38 
39 #define INITIAL_CREDIT_PACKETS	100
40 #define MAX_BYTES_PER_PACKET	1500
41 #define MARK_THRESH		(40 * MAX_BYTES_PER_PACKET)
42 #define DROP_THRESH		(80 * 5 * MAX_BYTES_PER_PACKET)
43 #define LARGE_PKT_DROP_THRESH	(DROP_THRESH - (15 * MAX_BYTES_PER_PACKET))
44 #define MARK_REGION_SIZE	(LARGE_PKT_DROP_THRESH - MARK_THRESH)
45 #define LARGE_PKT_THRESH	120
46 #define MAX_CREDIT		(100 * MAX_BYTES_PER_PACKET)
47 #define INIT_CREDIT		(INITIAL_CREDIT_PACKETS * MAX_BYTES_PER_PACKET)
48 
49 // Time base accounting for fq's EDT
50 #define BURST_SIZE_NS		100000 // 100us
51 #define MARK_THRESH_NS		50000 // 50us
52 #define DROP_THRESH_NS		500000 // 500us
53 // Reserve 20us of queuing for small packets (less than 120 bytes)
54 #define LARGE_PKT_DROP_THRESH_NS (DROP_THRESH_NS - 20000)
55 #define MARK_REGION_SIZE_NS	(LARGE_PKT_DROP_THRESH_NS - MARK_THRESH_NS)
56 
57 // rate in bytes per ns << 20
58 #define CREDIT_PER_NS(delta, rate) ((((u64)(delta)) * (rate)) >> 20)
59 #define BYTES_PER_NS(delta, rate) ((((u64)(delta)) * (rate)) >> 20)
60 #define BYTES_TO_NS(bytes, rate) div64_u64(((u64)(bytes)) << 20, (u64)(rate))
61 
62 struct bpf_map_def SEC("maps") queue_state = {
63 	.type = BPF_MAP_TYPE_CGROUP_STORAGE,
64 	.key_size = sizeof(struct bpf_cgroup_storage_key),
65 	.value_size = sizeof(struct hbm_vqueue),
66 };
67 BPF_ANNOTATE_KV_PAIR(queue_state, struct bpf_cgroup_storage_key,
68 		     struct hbm_vqueue);
69 
70 struct bpf_map_def SEC("maps") queue_stats = {
71 	.type = BPF_MAP_TYPE_ARRAY,
72 	.key_size = sizeof(u32),
73 	.value_size = sizeof(struct hbm_queue_stats),
74 	.max_entries = 1,
75 };
76 BPF_ANNOTATE_KV_PAIR(queue_stats, int, struct hbm_queue_stats);
77 
78 struct hbm_pkt_info {
79 	int	cwnd;
80 	int	rtt;
81 	int	packets_out;
82 	bool	is_ip;
83 	bool	is_tcp;
84 	short	ecn;
85 };
86 
get_tcp_info(struct __sk_buff * skb,struct hbm_pkt_info * pkti)87 static int get_tcp_info(struct __sk_buff *skb, struct hbm_pkt_info *pkti)
88 {
89 	struct bpf_sock *sk;
90 	struct bpf_tcp_sock *tp;
91 
92 	sk = skb->sk;
93 	if (sk) {
94 		sk = bpf_sk_fullsock(sk);
95 		if (sk) {
96 			if (sk->protocol == IPPROTO_TCP) {
97 				tp = bpf_tcp_sock(sk);
98 				if (tp) {
99 					pkti->cwnd = tp->snd_cwnd;
100 					pkti->rtt = tp->srtt_us >> 3;
101 					pkti->packets_out = tp->packets_out;
102 					return 0;
103 				}
104 			}
105 		}
106 	}
107 	pkti->cwnd = 0;
108 	pkti->rtt = 0;
109 	pkti->packets_out = 0;
110 	return 1;
111 }
112 
hbm_get_pkt_info(struct __sk_buff * skb,struct hbm_pkt_info * pkti)113 static void hbm_get_pkt_info(struct __sk_buff *skb,
114 			     struct hbm_pkt_info *pkti)
115 {
116 	struct iphdr iph;
117 	struct ipv6hdr *ip6h;
118 
119 	pkti->cwnd = 0;
120 	pkti->rtt = 0;
121 	bpf_skb_load_bytes(skb, 0, &iph, 12);
122 	if (iph.version == 6) {
123 		ip6h = (struct ipv6hdr *)&iph;
124 		pkti->is_ip = true;
125 		pkti->is_tcp = (ip6h->nexthdr == 6);
126 		pkti->ecn = (ip6h->flow_lbl[0] >> 4) & INET_ECN_MASK;
127 	} else if (iph.version == 4) {
128 		pkti->is_ip = true;
129 		pkti->is_tcp = (iph.protocol == 6);
130 		pkti->ecn = iph.tos & INET_ECN_MASK;
131 	} else {
132 		pkti->is_ip = false;
133 		pkti->is_tcp = false;
134 		pkti->ecn = 0;
135 	}
136 	if (pkti->is_tcp)
137 		get_tcp_info(skb, pkti);
138 }
139 
hbm_init_vqueue(struct hbm_vqueue * qdp,int rate)140 static __always_inline void hbm_init_vqueue(struct hbm_vqueue *qdp, int rate)
141 {
142 	bpf_printk("Initializing queue_state, rate:%d\n", rate * 128);
143 	qdp->lasttime = bpf_ktime_get_ns();
144 	qdp->credit = INIT_CREDIT;
145 	qdp->rate = rate * 128;
146 }
147 
hbm_init_edt_vqueue(struct hbm_vqueue * qdp,int rate)148 static __always_inline void hbm_init_edt_vqueue(struct hbm_vqueue *qdp,
149 						int rate)
150 {
151 	unsigned long long curtime;
152 
153 	curtime = bpf_ktime_get_ns();
154 	bpf_printk("Initializing queue_state, rate:%d\n", rate * 128);
155 	qdp->lasttime = curtime - BURST_SIZE_NS;	// support initial burst
156 	qdp->credit = 0;				// not used
157 	qdp->rate = rate * 128;
158 }
159 
hbm_update_stats(struct hbm_queue_stats * qsp,int len,unsigned long long curtime,bool congestion_flag,bool drop_flag,bool cwr_flag,bool ecn_ce_flag,struct hbm_pkt_info * pkti,int credit)160 static __always_inline void hbm_update_stats(struct hbm_queue_stats *qsp,
161 					     int len,
162 					     unsigned long long curtime,
163 					     bool congestion_flag,
164 					     bool drop_flag,
165 					     bool cwr_flag,
166 					     bool ecn_ce_flag,
167 					     struct hbm_pkt_info *pkti,
168 					     int credit)
169 {
170 	int rv = ALLOW_PKT;
171 
172 	if (qsp != NULL) {
173 		// Following is needed for work conserving
174 		__sync_add_and_fetch(&(qsp->bytes_total), len);
175 		if (qsp->stats) {
176 			// Optionally update statistics
177 			if (qsp->firstPacketTime == 0)
178 				qsp->firstPacketTime = curtime;
179 			qsp->lastPacketTime = curtime;
180 			__sync_add_and_fetch(&(qsp->pkts_total), 1);
181 			if (congestion_flag) {
182 				__sync_add_and_fetch(&(qsp->pkts_marked), 1);
183 				__sync_add_and_fetch(&(qsp->bytes_marked), len);
184 			}
185 			if (drop_flag) {
186 				__sync_add_and_fetch(&(qsp->pkts_dropped), 1);
187 				__sync_add_and_fetch(&(qsp->bytes_dropped),
188 						     len);
189 			}
190 			if (ecn_ce_flag)
191 				__sync_add_and_fetch(&(qsp->pkts_ecn_ce), 1);
192 			if (pkti->cwnd) {
193 				__sync_add_and_fetch(&(qsp->sum_cwnd),
194 						     pkti->cwnd);
195 				__sync_add_and_fetch(&(qsp->sum_cwnd_cnt), 1);
196 			}
197 			if (pkti->rtt)
198 				__sync_add_and_fetch(&(qsp->sum_rtt),
199 						     pkti->rtt);
200 			__sync_add_and_fetch(&(qsp->sum_credit), credit);
201 
202 			if (drop_flag)
203 				rv = DROP_PKT;
204 			if (cwr_flag)
205 				rv |= 2;
206 			if (rv == DROP_PKT)
207 				__sync_add_and_fetch(&(qsp->returnValCount[0]),
208 						     1);
209 			else if (rv == ALLOW_PKT)
210 				__sync_add_and_fetch(&(qsp->returnValCount[1]),
211 						     1);
212 			else if (rv == 2)
213 				__sync_add_and_fetch(&(qsp->returnValCount[2]),
214 						     1);
215 			else if (rv == 3)
216 				__sync_add_and_fetch(&(qsp->returnValCount[3]),
217 						     1);
218 		}
219 	}
220 }
221