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 #include <sys/param.h> 38 39 #include <netinet/in.h> 40 41 #ifndef TCP_LRO_ENTRIES 42 /* Define default number of LRO entries per RX queue */ 43 #define TCP_LRO_ENTRIES 8 44 #endif 45 46 /* 47 * Flags for ACK entry for compression 48 * the bottom 12 bits has the th_x2|th_flags. 49 * LRO itself adds only the TSTMP flags 50 * to indicate if either of the types 51 * of timestamps are filled and the 52 * HAS_TSTMP option to indicate if the 53 * TCP timestamp option is valid. 54 * 55 * The other 1 flag bits are for processing 56 * by a stack. 57 * 58 */ 59 #define TSTMP_LRO 0x1000 60 #define TSTMP_HDWR 0x2000 61 #define HAS_TSTMP 0x4000 62 /* 63 * Default number of interrupts on the same cpu in a row 64 * that will cause us to declare a "affinity cpu". 65 */ 66 #define TCP_LRO_CPU_DECLARATION_THRESH 50 67 68 struct inpcb; 69 70 /* Precompute the LRO_RAW_ADDRESS_MAX value: */ 71 #define LRO_RAW_ADDRESS_MAX \ 72 howmany(12 + 2 * sizeof(struct in6_addr), sizeof(u_long)) 73 74 union lro_address { 75 u_long raw[LRO_RAW_ADDRESS_MAX]; 76 struct { 77 uint8_t lro_type; /* internal */ 78 #define LRO_TYPE_NONE 0 79 #define LRO_TYPE_IPV4_TCP 1 80 #define LRO_TYPE_IPV6_TCP 2 81 #define LRO_TYPE_IPV4_UDP 3 82 #define LRO_TYPE_IPV6_UDP 4 83 uint8_t lro_flags; 84 #define LRO_FLAG_DECRYPTED 1 85 uint16_t vlan_id; /* VLAN identifier */ 86 uint16_t s_port; /* source TCP/UDP port */ 87 uint16_t d_port; /* destination TCP/UDP port */ 88 uint32_t vxlan_vni; /* VXLAN virtual network identifier */ 89 union { 90 struct in_addr v4; 91 struct in6_addr v6; 92 } s_addr; /* source IPv4/IPv6 address */ 93 union { 94 struct in_addr v4; 95 struct in6_addr v6; 96 } d_addr; /* destination IPv4/IPv6 address */ 97 }; 98 }; 99 100 _Static_assert(sizeof(union lro_address) == sizeof(u_long) * LRO_RAW_ADDRESS_MAX, 101 "The raw field in the lro_address union does not cover the whole structure."); 102 103 /* Optimize address comparison by comparing one unsigned long at a time: */ 104 105 static inline bool 106 lro_address_compare(const union lro_address *pa, const union lro_address *pb) 107 { 108 if (pa->lro_type == LRO_TYPE_NONE && pb->lro_type == LRO_TYPE_NONE) { 109 return (true); 110 } else for (unsigned i = 0; i < LRO_RAW_ADDRESS_MAX; i++) { 111 if (pa->raw[i] != pb->raw[i]) 112 return (false); 113 } 114 return (true); 115 } 116 117 struct lro_parser { 118 union lro_address data; 119 union { 120 uint8_t *l3; 121 struct ip *ip4; 122 struct ip6_hdr *ip6; 123 }; 124 union { 125 uint8_t *l4; 126 struct tcphdr *tcp; 127 struct udphdr *udp; 128 }; 129 uint16_t total_hdr_len; 130 }; 131 132 /* This structure is zeroed frequently, try to keep it small. */ 133 struct lro_entry { 134 LIST_ENTRY(lro_entry) next; 135 LIST_ENTRY(lro_entry) hash_next; 136 struct mbuf *m_head; 137 struct mbuf *m_tail; 138 struct mbuf *m_last_mbuf; 139 struct lro_parser outer; 140 struct lro_parser inner; 141 uint32_t next_seq; /* tcp_seq */ 142 uint32_t ack_seq; /* tcp_seq */ 143 uint32_t tsval; 144 uint32_t tsecr; 145 uint16_t compressed; 146 uint16_t uncompressed; 147 uint16_t window; 148 uint16_t flags : 12, /* 12 TCP header bits */ 149 timestamp : 1, 150 needs_merge : 1, 151 reserved : 2; /* unused */ 152 struct bintime alloc_time; /* time when entry was allocated */ 153 }; 154 155 LIST_HEAD(lro_head, lro_entry); 156 157 struct lro_mbuf_sort { 158 uint64_t seq; 159 struct mbuf *mb; 160 }; 161 162 /* NB: This is part of driver structs. */ 163 struct lro_ctrl { 164 struct ifnet *ifp; 165 struct lro_mbuf_sort *lro_mbuf_data; 166 struct bintime lro_last_queue_time; /* last time data was queued */ 167 uint64_t lro_queued; 168 uint64_t lro_flushed; 169 uint64_t lro_bad_csum; 170 unsigned lro_cnt; 171 unsigned lro_mbuf_count; 172 unsigned lro_mbuf_max; 173 unsigned short lro_ackcnt_lim; /* max # of aggregated ACKs */ 174 unsigned short lro_cpu; /* Guess at the cpu we have affinity too */ 175 unsigned lro_length_lim; /* max len of aggregated data */ 176 u_long lro_hashsz; 177 uint32_t lro_last_cpu; 178 uint32_t lro_cnt_of_same_cpu; 179 struct lro_head *lro_hash; 180 struct lro_head lro_active; 181 struct lro_head lro_free; 182 uint8_t lro_cpu_is_set; /* Flag to say its ok to set the CPU on the inp */ 183 }; 184 185 struct tcp_ackent { 186 uint64_t timestamp; /* hardware or sofware timestamp, valid if TSTMP_LRO or TSTMP_HDRW set */ 187 uint32_t seq; /* th_seq value */ 188 uint32_t ack; /* th_ack value */ 189 uint32_t ts_value; /* If ts option value, valid if HAS_TSTMP is set */ 190 uint32_t ts_echo; /* If ts option echo, valid if HAS_TSTMP is set */ 191 uint16_t win; /* TCP window */ 192 uint16_t flags; /* Flags to say if TS is present and type of timestamp and th_flags */ 193 uint8_t codepoint; /* IP level codepoint including ECN bits */ 194 uint8_t ack_val_set; /* Classification of ack used by the stack */ 195 uint8_t pad[2]; /* To 32 byte boundary */ 196 }; 197 198 /* We use two M_PROTO on the mbuf */ 199 #define M_ACKCMP M_PROTO4 /* Indicates LRO is sending in a Ack-compression mbuf */ 200 #define M_LRO_EHDRSTRP M_PROTO6 /* Indicates that LRO has stripped the etherenet header */ 201 202 #define TCP_LRO_LENGTH_MAX (65535 - 255) /* safe value with room for outer headers */ 203 #define TCP_LRO_ACKCNT_MAX 65535 /* unlimited */ 204 205 int tcp_lro_init(struct lro_ctrl *); 206 int tcp_lro_init_args(struct lro_ctrl *, struct ifnet *, unsigned, unsigned); 207 void tcp_lro_free(struct lro_ctrl *); 208 void tcp_lro_flush_inactive(struct lro_ctrl *, const struct timeval *); 209 void tcp_lro_flush(struct lro_ctrl *, struct lro_entry *); 210 void tcp_lro_flush_all(struct lro_ctrl *); 211 int tcp_lro_rx(struct lro_ctrl *, struct mbuf *, uint32_t); 212 void tcp_lro_queue_mbuf(struct lro_ctrl *, struct mbuf *); 213 void tcp_lro_reg_mbufq(void); 214 void tcp_lro_dereg_mbufq(void); 215 216 #define TCP_LRO_NO_ENTRIES -2 217 #define TCP_LRO_CANNOT -1 218 #define TCP_LRO_NOT_SUPPORTED 1 219 220 #endif /* _TCP_LRO_H_ */ 221