1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2006, Myricom Inc. 5 * Copyright (c) 2008, Intel Corporation. 6 * Copyright (c) 2016-2021 Mellanox Technologies. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $FreeBSD$ 31 */ 32 33 #ifndef _TCP_LRO_H_ 34 #define _TCP_LRO_H_ 35 36 #include <sys/time.h> 37 38 #ifndef TCP_LRO_ENTRIES 39 /* Define default number of LRO entries per RX queue */ 40 #define TCP_LRO_ENTRIES 8 41 #endif 42 43 /* 44 * Flags for ACK entry for compression 45 * the bottom 8 bits has the th_flags. 46 * LRO itself adds only the TSTMP flags 47 * to indicate if either of the types 48 * of timestamps are filled and the 49 * HAS_TSTMP option to indicate if the 50 * TCP timestamp option is valid. 51 * 52 * The other 5 flag bits are for processing 53 * by a stack. 54 * 55 */ 56 #define TSTMP_LRO 0x0100 57 #define TSTMP_HDWR 0x0200 58 #define HAS_TSTMP 0x0400 59 /* 60 * Default number of interrupts on the same cpu in a row 61 * that will cause us to declare a "affinity cpu". 62 */ 63 #define TCP_LRO_CPU_DECLARATION_THRESH 50 64 65 struct inpcb; 66 67 union lro_address { 68 u_long raw[1]; 69 struct { 70 uint16_t lro_type; /* internal */ 71 #define LRO_TYPE_NONE 0 72 #define LRO_TYPE_IPV4_TCP 1 73 #define LRO_TYPE_IPV6_TCP 2 74 #define LRO_TYPE_IPV4_UDP 3 75 #define LRO_TYPE_IPV6_UDP 4 76 uint16_t vlan_id; /* VLAN identifier */ 77 uint16_t s_port; /* source TCP/UDP port */ 78 uint16_t d_port; /* destination TCP/UDP port */ 79 uint32_t vxlan_vni; /* VXLAN virtual network identifier */ 80 union { 81 #ifdef INET 82 struct in_addr v4; 83 #endif 84 #ifdef INET6 85 struct in6_addr v6; 86 #endif 87 } s_addr; /* source IPv4/IPv6 address */ 88 union { 89 #ifdef INET 90 struct in_addr v4; 91 #endif 92 #ifdef INET6 93 struct in6_addr v6; 94 #endif 95 } d_addr; /* destination IPv4/IPv6 address */ 96 }; 97 } __aligned(sizeof(u_long)); 98 99 #define LRO_RAW_ADDRESS_MAX \ 100 (sizeof(union lro_address) / sizeof(u_long)) 101 102 /* Optimize address comparison by comparing one unsigned long at a time: */ 103 104 static inline bool 105 lro_address_compare(const union lro_address *pa, const union lro_address *pb) 106 { 107 if (pa->lro_type == LRO_TYPE_NONE && pb->lro_type == LRO_TYPE_NONE) { 108 return (true); 109 } else for (unsigned i = 0; i < LRO_RAW_ADDRESS_MAX; i++) { 110 if (pa->raw[i] != pb->raw[i]) 111 return (false); 112 } 113 return (true); 114 } 115 116 struct lro_parser { 117 union lro_address data; 118 union { 119 uint8_t *l3; 120 struct ip *ip4; 121 struct ip6_hdr *ip6; 122 }; 123 union { 124 uint8_t *l4; 125 struct tcphdr *tcp; 126 struct udphdr *udp; 127 }; 128 uint16_t total_hdr_len; 129 }; 130 131 /* This structure is zeroed frequently, try to keep it small. */ 132 struct lro_entry { 133 LIST_ENTRY(lro_entry) next; 134 LIST_ENTRY(lro_entry) hash_next; 135 struct mbuf *m_head; 136 struct mbuf *m_tail; 137 struct mbuf *m_last_mbuf; 138 struct lro_parser outer; 139 struct lro_parser inner; 140 uint32_t next_seq; /* tcp_seq */ 141 uint32_t ack_seq; /* tcp_seq */ 142 uint32_t tsval; 143 uint32_t tsecr; 144 uint16_t compressed; 145 uint16_t uncompressed; 146 uint16_t window; 147 uint16_t timestamp; /* flag, not a TCP hdr field. */ 148 struct bintime alloc_time; /* time when entry was allocated */ 149 }; 150 151 LIST_HEAD(lro_head, lro_entry); 152 153 struct lro_mbuf_sort { 154 uint64_t seq; 155 struct mbuf *mb; 156 }; 157 158 /* NB: This is part of driver structs. */ 159 struct lro_ctrl { 160 struct ifnet *ifp; 161 struct lro_mbuf_sort *lro_mbuf_data; 162 struct bintime lro_last_queue_time; /* last time data was queued */ 163 uint64_t lro_queued; 164 uint64_t lro_flushed; 165 uint64_t lro_bad_csum; 166 unsigned lro_cnt; 167 unsigned lro_mbuf_count; 168 unsigned lro_mbuf_max; 169 unsigned short lro_ackcnt_lim; /* max # of aggregated ACKs */ 170 unsigned short lro_cpu; /* Guess at the cpu we have affinity too */ 171 unsigned lro_length_lim; /* max len of aggregated data */ 172 u_long lro_hashsz; 173 uint32_t lro_last_cpu; 174 uint32_t lro_cnt_of_same_cpu; 175 struct lro_head *lro_hash; 176 struct lro_head lro_active; 177 struct lro_head lro_free; 178 uint8_t lro_cpu_is_set; /* Flag to say its ok to set the CPU on the inp */ 179 }; 180 181 struct tcp_ackent { 182 uint64_t timestamp; /* hardware or sofware timestamp, valid if TSTMP_LRO or TSTMP_HDRW set */ 183 uint32_t seq; /* th_seq value */ 184 uint32_t ack; /* th_ack value */ 185 uint32_t ts_value; /* If ts option value, valid if HAS_TSTMP is set */ 186 uint32_t ts_echo; /* If ts option echo, valid if HAS_TSTMP is set */ 187 uint16_t win; /* TCP window */ 188 uint16_t flags; /* Flags to say if TS is present and type of timestamp and th_flags */ 189 uint8_t codepoint; /* IP level codepoint including ECN bits */ 190 uint8_t ack_val_set; /* Classification of ack used by the stack */ 191 uint8_t pad[2]; /* To 32 byte boundary */ 192 }; 193 194 /* We use two M_PROTO on the mbuf */ 195 #define M_ACKCMP M_PROTO4 /* Indicates LRO is sending in a Ack-compression mbuf */ 196 #define M_LRO_EHDRSTRP M_PROTO6 /* Indicates that LRO has stripped the etherenet header */ 197 198 #define TCP_LRO_LENGTH_MAX (65535 - 255) /* safe value with room for outer headers */ 199 #define TCP_LRO_ACKCNT_MAX 65535 /* unlimited */ 200 201 int tcp_lro_init(struct lro_ctrl *); 202 int tcp_lro_init_args(struct lro_ctrl *, struct ifnet *, unsigned, unsigned); 203 void tcp_lro_free(struct lro_ctrl *); 204 void tcp_lro_flush_inactive(struct lro_ctrl *, const struct timeval *); 205 void tcp_lro_flush(struct lro_ctrl *, struct lro_entry *); 206 void tcp_lro_flush_all(struct lro_ctrl *); 207 int tcp_lro_rx(struct lro_ctrl *, struct mbuf *, uint32_t); 208 void tcp_lro_queue_mbuf(struct lro_ctrl *, struct mbuf *); 209 void tcp_lro_reg_mbufq(void); 210 void tcp_lro_dereg_mbufq(void); 211 212 #define TCP_LRO_NO_ENTRIES -2 213 #define TCP_LRO_CANNOT -1 214 #define TCP_LRO_NOT_SUPPORTED 1 215 216 #endif /* _TCP_LRO_H_ */ 217