master
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
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
31#ifndef _TCP_LRO_H_
32#define _TCP_LRO_H_
33
34#include <sys/time.h>
35#include <sys/param.h>
36#include <sys/mbuf.h>
37#include <netinet/in.h>
38
39#ifndef TCP_LRO_ENTRIES
40/* Define default number of LRO entries per RX queue */
41#define TCP_LRO_ENTRIES 8
42#endif
43
44/*
45 * Flags for ACK entry for compression
46 * the bottom 12 bits has the th_x2|th_flags.
47 * LRO itself adds only the TSTMP flags
48 * to indicate if either of the types
49 * of timestamps are filled and the
50 * HAS_TSTMP option to indicate if the
51 * TCP timestamp option is valid.
52 *
53 * The other 1 flag bits are for processing
54 * by a stack.
55 *
56 */
57#define TSTMP_LRO 0x1000
58#define TSTMP_HDWR 0x2000
59#define HAS_TSTMP 0x4000
60/*
61 * Default number of interrupts on the same cpu in a row
62 * that will cause us to declare a "affinity cpu".
63 */
64#define TCP_LRO_CPU_DECLARATION_THRESH 50
65
66struct inpcb;
67
68/* Precompute the LRO_RAW_ADDRESS_MAX value: */
69#define LRO_RAW_ADDRESS_MAX \
70 howmany(12 + 2 * sizeof(struct in6_addr), sizeof(u_long))
71
72union lro_address {
73 u_long raw[LRO_RAW_ADDRESS_MAX];
74 struct {
75 uint8_t lro_type; /* internal */
76#define LRO_TYPE_NONE 0
77#define LRO_TYPE_IPV4_TCP 1
78#define LRO_TYPE_IPV6_TCP 2
79#define LRO_TYPE_IPV4_UDP 3
80#define LRO_TYPE_IPV6_UDP 4
81 uint8_t lro_flags;
82#define LRO_FLAG_DECRYPTED 1
83 uint16_t vlan_id; /* VLAN identifier */
84 uint16_t s_port; /* source TCP/UDP port */
85 uint16_t d_port; /* destination TCP/UDP port */
86 uint32_t vxlan_vni; /* VXLAN virtual network identifier */
87 union {
88 struct in_addr v4;
89 struct in6_addr v6;
90 } s_addr; /* source IPv4/IPv6 address */
91 union {
92 struct in_addr v4;
93 struct in6_addr v6;
94 } d_addr; /* destination IPv4/IPv6 address */
95 };
96};
97
98_Static_assert(sizeof(union lro_address) == sizeof(u_long) * LRO_RAW_ADDRESS_MAX,
99 "The raw field in the lro_address union does not cover the whole structure.");
100
101/* Optimize address comparison by comparing one unsigned long at a time: */
102
103static inline bool
104lro_address_compare(const union lro_address *pa, const union lro_address *pb)
105{
106 if (pa->lro_type == LRO_TYPE_NONE && pb->lro_type == LRO_TYPE_NONE) {
107 return (true);
108 } else for (unsigned i = 0; i < LRO_RAW_ADDRESS_MAX; i++) {
109 if (pa->raw[i] != pb->raw[i])
110 return (false);
111 }
112 return (true);
113}
114
115struct lro_parser {
116 union lro_address data;
117 union {
118 uint8_t *l3;
119 struct ip *ip4;
120 struct ip6_hdr *ip6;
121 };
122 union {
123 uint8_t *l4;
124 struct tcphdr *tcp;
125 struct udphdr *udp;
126 };
127 uint16_t total_hdr_len;
128};
129
130/* This structure is zeroed frequently, try to keep it small. */
131struct lro_entry {
132 LIST_ENTRY(lro_entry) next;
133 LIST_ENTRY(lro_entry) hash_next;
134 struct mbuf *m_head;
135 struct mbuf *m_tail;
136 struct mbuf *m_last_mbuf;
137 struct lro_parser outer;
138 struct lro_parser inner;
139 uint32_t next_seq; /* tcp_seq */
140 uint32_t ack_seq; /* tcp_seq */
141 uint32_t tsval;
142 uint32_t tsecr;
143 uint16_t compressed;
144 uint16_t uncompressed;
145 uint16_t window;
146 uint16_t flags : 12, /* 12 TCP header bits */
147 timestamp : 1,
148 needs_merge : 1,
149 reserved : 2; /* unused */
150 struct bintime alloc_time; /* time when entry was allocated */
151};
152
153LIST_HEAD(lro_head, lro_entry);
154
155struct lro_mbuf_sort {
156 uint64_t seq;
157 struct mbuf *mb;
158};
159
160/* NB: This is part of driver structs. */
161struct lro_ctrl {
162 struct ifnet *ifp;
163 struct lro_mbuf_sort *lro_mbuf_data;
164 struct bintime lro_last_queue_time; /* last time data was queued */
165 uint64_t lro_queued;
166 uint64_t lro_flushed;
167 uint64_t lro_bad_csum;
168 unsigned lro_cnt;
169 unsigned lro_mbuf_count;
170 unsigned lro_mbuf_max;
171 unsigned short lro_ackcnt_lim; /* max # of aggregated ACKs */
172 unsigned short lro_cpu; /* Guess at the cpu we have affinity too */
173 unsigned lro_length_lim; /* max len of aggregated data */
174 u_long lro_hashsz;
175 uint32_t lro_last_cpu;
176 uint32_t lro_cnt_of_same_cpu;
177 struct lro_head *lro_hash;
178 struct lro_head lro_active;
179 struct lro_head lro_free;
180 uint8_t lro_cpu_is_set; /* Flag to say its ok to set the CPU on the inp */
181};
182
183struct tcp_ackent {
184 uint64_t timestamp; /* hardware or sofware timestamp, valid if TSTMP_LRO or TSTMP_HDRW set */
185 uint32_t seq; /* th_seq value */
186 uint32_t ack; /* th_ack value */
187 uint32_t ts_value; /* If ts option value, valid if HAS_TSTMP is set */
188 uint32_t ts_echo; /* If ts option echo, valid if HAS_TSTMP is set */
189 uint16_t win; /* TCP window */
190 uint16_t flags; /* Flags to say if TS is present and type of timestamp and th_flags */
191 uint8_t codepoint; /* IP level codepoint including ECN bits */
192 uint8_t ack_val_set; /* Classification of ack used by the stack */
193 uint8_t pad[2]; /* To 32 byte boundary */
194};
195
196/* We use two M_PROTO on the mbuf */
197#define M_ACKCMP M_PROTO4 /* Indicates LRO is sending in a Ack-compression mbuf */
198#define M_LRO_EHDRSTRP M_PROTO6 /* Indicates that LRO has stripped the etherenet header */
199
200#define TCP_LRO_LENGTH_MAX (65535 - 255) /* safe value with room for outer headers */
201#define TCP_LRO_ACKCNT_MAX 65535 /* unlimited */
202
203#define TCP_LRO_TS_OPTION ntohl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |\
204 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)
205
206static inline struct tcphdr *
207tcp_lro_get_th(struct mbuf *m)
208{
209 return ((struct tcphdr *)((char *)m->m_data +
210 m->m_pkthdr.lro_tcp_h_off));
211}
212
213extern long tcplro_stacks_wanting_mbufq;
214
215int tcp_lro_init(struct lro_ctrl *);
216int tcp_lro_init_args(struct lro_ctrl *, struct ifnet *, unsigned, unsigned);
217void tcp_lro_free(struct lro_ctrl *);
218void tcp_lro_flush_inactive(struct lro_ctrl *, const struct timeval *);
219void tcp_lro_flush_all(struct lro_ctrl *);
220extern int (*tcp_lro_flush_tcphpts)(struct lro_ctrl *, struct lro_entry *);
221int tcp_lro_rx(struct lro_ctrl *, struct mbuf *, uint32_t);
222void tcp_lro_queue_mbuf(struct lro_ctrl *, struct mbuf *);
223void tcp_lro_reg_mbufq(void);
224void tcp_lro_dereg_mbufq(void);
225
226#define TCP_LRO_NO_ENTRIES -2
227#define TCP_LRO_CANNOT -1
228#define TCP_LRO_NOT_SUPPORTED 1
229
230#endif /* _TCP_LRO_H_ */