| 1 | /*- |
| 2 | * |
| 3 | * SPDX-License-Identifier: BSD-3-Clause |
| 4 | * |
| 5 | * Copyright (c) 2018-2020 |
| 6 | *	Netflix Inc. |
| 7 | * |
| 8 | * Redistribution and use in source and binary forms, with or without |
| 9 | * modification, are permitted provided that the following conditions |
| 10 | * are met: |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in the |
| 15 | * documentation and/or other materials provided with the distribution. |
| 16 | * |
| 17 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 18 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 19 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 20 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 21 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 22 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 23 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 24 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 25 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 26 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 27 | * SUCH DAMAGE. |
| 28 | * |
| 29 | */ |
| 30 | /** |
| 31 | * Author: Randall Stewart <rrs@netflix.com> |
| 32 | */ |
| 33 | #ifndef __tcp_ratelimit_h__ |
| 34 | #define __tcp_ratelimit_h__ |
| 35 | |
| 36 | struct m_snd_tag; |
| 37 | |
| 38 | #define RL_MIN_DIVISOR 50 |
| 39 | #define RL_DEFAULT_DIVISOR 1000 |
| 40 | |
| 41 | /* Flags on an individual rate */ |
| 42 | #define HDWRPACE_INITED 	0x0001 |
| 43 | #define HDWRPACE_TAGPRESENT	0x0002 |
| 44 | #define HDWRPACE_IFPDEPARTED	0x0004 |
| 45 | struct tcp_hwrate_limit_table { |
| 46 | 	const struct tcp_rate_set *ptbl;	/* Pointer to parent table */ |
| 47 | 	struct m_snd_tag *tag;	/* Send tag if needed (chelsio) */ |
| 48 | 	long	 rate;		/* Rate we get in Bytes per second (Bps) */ |
| 49 | 	long	 using;		/* How many flows are using this hdwr rate. */ |
| 50 | 	long	 rs_num_enobufs; |
| 51 | 	uint32_t time_between;	/* Time-Gap between packets at this rate */ |
| 52 | 	uint32_t flags; |
| 53 | }; |
| 54 | |
| 55 | /* Rateset flags */ |
| 56 | #define RS_IS_DEFF 0x0001	/* Its a lagg, do a double lookup */ |
| 57 | #define RS_IS_INTF 0x0002	/* Its a plain interface */ |
| 58 | #define RS_NO_PRE 0x0004	/* The interfacd has set rates */ |
| 59 | #define RS_INT_TBL 0x0010	/* |
| 60 | 				 * The table is the internal version |
| 61 | 				 * which has special setup requirements. |
| 62 | 				 */ |
| 63 | #define RS_IS_DEAD 0x0020	/* The RS is dead list */ |
| 64 | #define RS_FUNERAL_SCHD 0x0040 /* Is a epoch call scheduled to bury this guy?*/ |
| 65 | #define RS_INTF_NO_SUP 0x0100 	/* The interface does not support the ratelimiting */ |
| 66 | |
| 67 | struct tcp_rate_set { |
| 68 | 	struct sysctl_ctx_list sysctl_ctx; |
| 69 | 	CK_LIST_ENTRY(tcp_rate_set) next; |
| 70 | 	struct ifnet *rs_ifp; |
| 71 | 	struct tcp_hwrate_limit_table *rs_rlt; |
| 72 | 	uint64_t rs_flows_using; |
| 73 | 	uint64_t rs_flow_limit; |
| 74 | 	uint32_t rs_if_dunit; |
| 75 | 	int rs_rate_cnt; |
| 76 | 	int rs_min_seg; |
| 77 | 	int rs_highest_valid; |
| 78 | 	int rs_lowest_valid; |
| 79 | 	int rs_disable; |
| 80 | 	int rs_flags; |
| 81 | 	struct epoch_context rs_epoch_ctx; |
| 82 | }; |
| 83 | |
| 84 | CK_LIST_HEAD(head_tcp_rate_set, tcp_rate_set); |
| 85 | |
| 86 | /* Request flags */ |
| 87 | #define RS_PACING_EXACT_MATCH	0x0001	/* Need an exact match for rate */ |
| 88 | #define RS_PACING_GT		0x0002	/* Greater than requested */ |
| 89 | #define RS_PACING_GEQ		0x0004	/* Greater than or equal too */ |
| 90 | #define RS_PACING_LT		0x0008	/* Less than requested rate */ |
| 91 | #define RS_PACING_SUB_OK	0x0010	/* If a rate can't be found get the |
| 92 | 					 * next best rate (highest or lowest). */ |
| 93 | #ifdef _KERNEL |
| 94 | #ifndef ETHERNET_SEGMENT_SIZE |
| 95 | #define ETHERNET_SEGMENT_SIZE 1514 |
| 96 | #endif |
| 97 | struct tcpcb; |
| 98 | |
| 99 | #ifdef RATELIMIT |
| 100 | #define DETAILED_RATELIMIT_SYSCTL 1	/* |
| 101 | 					 * Undefine this if you don't want |
| 102 | 					 * detailed rates to appear in |
| 103 | 					 * net.inet.tcp.rl. |
| 104 | 					 * With the defintion each rate |
| 105 | 					 * shows up in your sysctl tree |
| 106 | 					 * this can be big. |
| 107 | 					 */ |
| 108 | uint64_t inline |
| 109 | tcp_hw_highest_rate(const struct tcp_hwrate_limit_table *rle) |
| 110 | { |
| 111 | 	return (rle->ptbl->rs_rlt[rle->ptbl->rs_highest_valid].rate); |
| 112 | } |
| 113 | |
| 114 | uint64_t |
| 115 | tcp_hw_highest_rate_ifp(struct ifnet *ifp, struct inpcb *inp); |
| 116 | |
| 117 | const struct tcp_hwrate_limit_table * |
| 118 | tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, |
| 119 | uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_rate); |
| 120 | |
| 121 | const struct tcp_hwrate_limit_table * |
| 122 | tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, |
| 123 | struct tcpcb *tp, struct ifnet *ifp, |
| 124 | uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_rate); |
| 125 | void |
| 126 | tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte, |
| 127 | struct tcpcb *tp); |
| 128 | |
| 129 | uint32_t |
| 130 | tcp_get_pacing_burst_size_w_divisor(struct tcpcb *tp, uint64_t bw, uint32_t segsiz, int can_use_1mss, |
| 131 | const struct tcp_hwrate_limit_table *te, int *err, int divisor); |
| 132 | |
| 133 | void |
| 134 | tcp_rl_log_enobuf(const struct tcp_hwrate_limit_table *rte); |
| 135 | |
| 136 | void |
| 137 | tcp_rl_release_ifnet(struct ifnet *ifp); |
| 138 | |
| 139 | #else |
| 140 | static inline const struct tcp_hwrate_limit_table * |
| 141 | tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, |
| 142 | uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_rate) |
| 143 | { |
| 144 | 	if (error) |
| 145 | 		*error = EOPNOTSUPP; |
| 146 | 	return (NULL); |
| 147 | } |
| 148 | |
| 149 | static inline const struct tcp_hwrate_limit_table * |
| 150 | tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, |
| 151 | struct tcpcb *tp, struct ifnet *ifp, |
| 152 | uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_rate) |
| 153 | { |
| 154 | 	if (error) |
| 155 | 		*error = EOPNOTSUPP; |
| 156 | 	return (NULL); |
| 157 | } |
| 158 | |
| 159 | static inline void |
| 160 | tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte, |
| 161 | struct tcpcb *tp) |
| 162 | { |
| 163 | 	return; |
| 164 | } |
| 165 | |
| 166 | static uint64_t inline |
| 167 | tcp_hw_highest_rate(const struct tcp_hwrate_limit_table *rle) |
| 168 | { |
| 169 | 	return (0); |
| 170 | } |
| 171 | |
| 172 | static uint64_t inline |
| 173 | tcp_hw_highest_rate_ifp(struct ifnet *ifp, struct inpcb *inp) |
| 174 | { |
| 175 | 	return (0); |
| 176 | } |
| 177 | |
| 178 | static inline uint32_t |
| 179 | tcp_get_pacing_burst_size_w_divisor(struct tcpcb *tp, uint64_t bw, uint32_t segsiz, int can_use_1mss, |
| 180 | const struct tcp_hwrate_limit_table *te, int *err, int divisor) |
| 181 | { |
| 182 | 	/* |
| 183 | 	 * We use the google formula to calculate the |
| 184 | 	 * TSO size. I.E. |
| 185 | 	 * bw < 24Meg |
| 186 | 	 * tso = 2mss |
| 187 | 	 * else |
| 188 | 	 * tso = min(bw/(div=1000), 64k) |
| 189 | 	 * |
| 190 | 	 * Note for these calculations we ignore the |
| 191 | 	 * packet overhead (enet hdr, ip hdr and tcp hdr). |
| 192 | 	 * We only get the google formula when we have |
| 193 | 	 * divisor = 1000, which is the default for now. |
| 194 | 	 */ |
| 195 | 	uint64_t bytes; |
| 196 | 	uint32_t new_tso, min_tso_segs; |
| 197 | |
| 198 | 	/* It can't be zero */ |
| 199 | 	if ((divisor == 0) || |
| 200 | 	 (divisor < RL_MIN_DIVISOR)) { |
| 201 | 		bytes = bw / RL_DEFAULT_DIVISOR; |
| 202 | 	} else |
| 203 | 		bytes = bw / divisor; |
| 204 | 	/* We can't ever send more than 65k in a TSO */ |
| 205 | 	if (bytes > 0xffff) { |
| 206 | 		bytes = 0xffff; |
| 207 | 	} |
| 208 | 	/* Round up */ |
| 209 | 	new_tso = (bytes + segsiz - 1) / segsiz; |
| 210 | 	if (can_use_1mss) |
| 211 | 		min_tso_segs = 1; |
| 212 | 	else |
| 213 | 		min_tso_segs = 2; |
| 214 | 	if (new_tso < min_tso_segs) |
| 215 | 		new_tso = min_tso_segs; |
| 216 | 	new_tso *= segsiz; |
| 217 | 	return (new_tso); |
| 218 | } |
| 219 | |
| 220 | /* Do nothing if RATELIMIT is not defined */ |
| 221 | static inline void |
| 222 | tcp_rl_log_enobuf(const struct tcp_hwrate_limit_table *rte) |
| 223 | { |
| 224 | } |
| 225 | |
| 226 | static inline void |
| 227 | tcp_rl_release_ifnet(struct ifnet *ifp) |
| 228 | { |
| 229 | } |
| 230 | #endif |
| 231 | |
| 232 | /* |
| 233 | * Given a b/w and a segsiz, and optional hardware |
| 234 | * rate limit, return the ideal size to burst |
| 235 | * out at once. Note the parameter can_use_1mss |
| 236 | * dictates if the transport will tolerate a 1mss |
| 237 | * limit, if not it will bottom out at 2mss (think |
| 238 | * delayed ack). |
| 239 | */ |
| 240 | static inline uint32_t |
| 241 | tcp_get_pacing_burst_size(struct tcpcb *tp, uint64_t bw, uint32_t segsiz, int can_use_1mss, |
| 242 | 			 const struct tcp_hwrate_limit_table *te, int *err) |
| 243 | { |
| 244 | |
| 245 | 	return (tcp_get_pacing_burst_size_w_divisor(tp, bw, segsiz, |
| 246 | 						 can_use_1mss, |
| 247 | 						 te, err, 0)); |
| 248 | } |
| 249 | |
| 250 | #endif |
| 251 | #endif |