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 <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 8 bits has the 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 5 flag bits are for processing
54 * by a stack.
55 *
56 */
57 #define TSTMP_LRO 0x0100
58 #define TSTMP_HDWR 0x0200
59 #define HAS_TSTMP 0x0400
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
66 struct inpcb;
67
68 union lro_address {
69 u_long raw[1];
70 struct {
71 uint8_t lro_type; /* internal */
72 #define LRO_TYPE_NONE 0
73 #define LRO_TYPE_IPV4_TCP 1
74 #define LRO_TYPE_IPV6_TCP 2
75 #define LRO_TYPE_IPV4_UDP 3
76 #define LRO_TYPE_IPV6_UDP 4
77 uint8_t lro_flags;
78 #define LRO_FLAG_DECRYPTED 1
79 uint16_t vlan_id; /* VLAN identifier */
80 uint16_t s_port; /* source TCP/UDP port */
81 uint16_t d_port; /* destination TCP/UDP port */
82 uint32_t vxlan_vni; /* VXLAN virtual network identifier */
83 union {
84 struct in_addr v4;
85 struct in6_addr v6;
86 } s_addr; /* source IPv4/IPv6 address */
87 union {
88 struct in_addr v4;
89 struct in6_addr v6;
90 } d_addr; /* destination IPv4/IPv6 address */
91 };
92 } __aligned(sizeof(u_long));
93
94 #define LRO_RAW_ADDRESS_MAX \
95 (sizeof(union lro_address) / sizeof(u_long))
96
97 /* Optimize address comparison by comparing one unsigned long at a time: */
98
99 static inline bool
lro_address_compare(const union lro_address * pa,const union lro_address * pb)100 lro_address_compare(const union lro_address *pa, const union lro_address *pb)
101 {
102 if (pa->lro_type == LRO_TYPE_NONE && pb->lro_type == LRO_TYPE_NONE) {
103 return (true);
104 } else for (unsigned i = 0; i < LRO_RAW_ADDRESS_MAX; i++) {
105 if (pa->raw[i] != pb->raw[i])
106 return (false);
107 }
108 return (true);
109 }
110
111 struct lro_parser {
112 union lro_address data;
113 union {
114 uint8_t *l3;
115 struct ip *ip4;
116 struct ip6_hdr *ip6;
117 };
118 union {
119 uint8_t *l4;
120 struct tcphdr *tcp;
121 struct udphdr *udp;
122 };
123 uint16_t total_hdr_len;
124 };
125
126 /* This structure is zeroed frequently, try to keep it small. */
127 struct lro_entry {
128 LIST_ENTRY(lro_entry) next;
129 LIST_ENTRY(lro_entry) hash_next;
130 struct mbuf *m_head;
131 struct mbuf *m_tail;
132 struct mbuf *m_last_mbuf;
133 struct lro_parser outer;
134 struct lro_parser inner;
135 uint32_t next_seq; /* tcp_seq */
136 uint32_t ack_seq; /* tcp_seq */
137 uint32_t tsval;
138 uint32_t tsecr;
139 uint16_t compressed;
140 uint16_t uncompressed;
141 uint16_t window;
142 uint8_t flags;
143 uint8_t timestamp : 1;
144 uint8_t needs_merge : 1;
145 struct bintime alloc_time; /* time when entry was allocated */
146 };
147
148 LIST_HEAD(lro_head, lro_entry);
149
150 struct lro_mbuf_sort {
151 uint64_t seq;
152 struct mbuf *mb;
153 };
154
155 /* NB: This is part of driver structs. */
156 struct lro_ctrl {
157 struct ifnet *ifp;
158 struct lro_mbuf_sort *lro_mbuf_data;
159 struct bintime lro_last_queue_time; /* last time data was queued */
160 uint64_t lro_queued;
161 uint64_t lro_flushed;
162 uint64_t lro_bad_csum;
163 unsigned lro_cnt;
164 unsigned lro_mbuf_count;
165 unsigned lro_mbuf_max;
166 unsigned short lro_ackcnt_lim; /* max # of aggregated ACKs */
167 unsigned short lro_cpu; /* Guess at the cpu we have affinity too */
168 unsigned lro_length_lim; /* max len of aggregated data */
169 u_long lro_hashsz;
170 uint32_t lro_last_cpu;
171 uint32_t lro_cnt_of_same_cpu;
172 struct lro_head *lro_hash;
173 struct lro_head lro_active;
174 struct lro_head lro_free;
175 uint8_t lro_cpu_is_set; /* Flag to say its ok to set the CPU on the inp */
176 };
177
178 struct tcp_ackent {
179 uint64_t timestamp; /* hardware or sofware timestamp, valid if TSTMP_LRO or TSTMP_HDRW set */
180 uint32_t seq; /* th_seq value */
181 uint32_t ack; /* th_ack value */
182 uint32_t ts_value; /* If ts option value, valid if HAS_TSTMP is set */
183 uint32_t ts_echo; /* If ts option echo, valid if HAS_TSTMP is set */
184 uint16_t win; /* TCP window */
185 uint16_t flags; /* Flags to say if TS is present and type of timestamp and th_flags */
186 uint8_t codepoint; /* IP level codepoint including ECN bits */
187 uint8_t ack_val_set; /* Classification of ack used by the stack */
188 uint8_t pad[2]; /* To 32 byte boundary */
189 };
190
191 /* We use two M_PROTO on the mbuf */
192 #define M_ACKCMP M_PROTO4 /* Indicates LRO is sending in a Ack-compression mbuf */
193 #define M_LRO_EHDRSTRP M_PROTO6 /* Indicates that LRO has stripped the etherenet header */
194
195 #define TCP_LRO_LENGTH_MAX (65535 - 255) /* safe value with room for outer headers */
196 #define TCP_LRO_ACKCNT_MAX 65535 /* unlimited */
197
198 int tcp_lro_init(struct lro_ctrl *);
199 int tcp_lro_init_args(struct lro_ctrl *, struct ifnet *, unsigned, unsigned);
200 void tcp_lro_free(struct lro_ctrl *);
201 void tcp_lro_flush_inactive(struct lro_ctrl *, const struct timeval *);
202 void tcp_lro_flush(struct lro_ctrl *, struct lro_entry *);
203 void tcp_lro_flush_all(struct lro_ctrl *);
204 int tcp_lro_rx(struct lro_ctrl *, struct mbuf *, uint32_t);
205 void tcp_lro_queue_mbuf(struct lro_ctrl *, struct mbuf *);
206 void tcp_lro_reg_mbufq(void);
207 void tcp_lro_dereg_mbufq(void);
208
209 #define TCP_LRO_NO_ENTRIES -2
210 #define TCP_LRO_CANNOT -1
211 #define TCP_LRO_NOT_SUPPORTED 1
212
213 #endif /* _TCP_LRO_H_ */
214