Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Definitions for the SMC module (socket related)
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
10 */
11#ifndef __SMC_H
12#define __SMC_H
13
14#include <linux/socket.h>
15#include <linux/types.h>
16#include <linux/compiler.h> /* __aligned */
17#include <net/sock.h>
18
19#include "smc_ib.h"
20
21#define SMC_V1 1 /* SMC version V1 */
22#define SMC_V2 2 /* SMC version V2 */
23#define SMC_RELEASE 0
24
25#define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */
26#define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */
27
28#define SMC_MAX_ISM_DEVS 8 /* max # of proposed non-native ISM
29 * devices
30 */
31
32#define SMC_MAX_HOSTNAME_LEN 32
33#define SMC_MAX_EID_LEN 32
34
35extern struct proto smc_proto;
36extern struct proto smc_proto6;
37
38#ifdef ATOMIC64_INIT
39#define KERNEL_HAS_ATOMIC64
40#endif
41
42enum smc_state { /* possible states of an SMC socket */
43 SMC_ACTIVE = 1,
44 SMC_INIT = 2,
45 SMC_CLOSED = 7,
46 SMC_LISTEN = 10,
47 /* normal close */
48 SMC_PEERCLOSEWAIT1 = 20,
49 SMC_PEERCLOSEWAIT2 = 21,
50 SMC_APPFINCLOSEWAIT = 24,
51 SMC_APPCLOSEWAIT1 = 22,
52 SMC_APPCLOSEWAIT2 = 23,
53 SMC_PEERFINCLOSEWAIT = 25,
54 /* abnormal close */
55 SMC_PEERABORTWAIT = 26,
56 SMC_PROCESSABORT = 27,
57};
58
59struct smc_link_group;
60
61struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
62 u8 type;
63} __aligned(1);
64
65struct smc_cdc_conn_state_flags {
66#if defined(__BIG_ENDIAN_BITFIELD)
67 u8 peer_done_writing : 1; /* Sending done indicator */
68 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
69 u8 peer_conn_abort : 1; /* Abnormal close indicator */
70 u8 reserved : 5;
71#elif defined(__LITTLE_ENDIAN_BITFIELD)
72 u8 reserved : 5;
73 u8 peer_conn_abort : 1;
74 u8 peer_conn_closed : 1;
75 u8 peer_done_writing : 1;
76#endif
77};
78
79struct smc_cdc_producer_flags {
80#if defined(__BIG_ENDIAN_BITFIELD)
81 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
82 u8 urg_data_pending : 1; /* Urgent Data Pending */
83 u8 urg_data_present : 1; /* Urgent Data Present */
84 u8 cons_curs_upd_req : 1; /* cursor update requested */
85 u8 failover_validation : 1;/* message replay due to failover */
86 u8 reserved : 3;
87#elif defined(__LITTLE_ENDIAN_BITFIELD)
88 u8 reserved : 3;
89 u8 failover_validation : 1;
90 u8 cons_curs_upd_req : 1;
91 u8 urg_data_present : 1;
92 u8 urg_data_pending : 1;
93 u8 write_blocked : 1;
94#endif
95};
96
97/* in host byte order */
98union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
99 struct {
100 u16 reserved;
101 u16 wrap; /* window wrap sequence number */
102 u32 count; /* cursor (= offset) part */
103 };
104#ifdef KERNEL_HAS_ATOMIC64
105 atomic64_t acurs; /* for atomic processing */
106#else
107 u64 acurs; /* for atomic processing */
108#endif
109} __aligned(8);
110
111/* in host byte order, except for flag bitfields in network byte order */
112struct smc_host_cdc_msg { /* Connection Data Control message */
113 struct smc_wr_rx_hdr common; /* .type = 0xFE */
114 u8 len; /* length = 44 */
115 u16 seqno; /* connection seq # */
116 u32 token; /* alert_token */
117 union smc_host_cursor prod; /* producer cursor */
118 union smc_host_cursor cons; /* consumer cursor,
119 * piggy backed "ack"
120 */
121 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
122 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
123 u8 reserved[18];
124} __aligned(8);
125
126enum smc_urg_state {
127 SMC_URG_VALID = 1, /* data present */
128 SMC_URG_NOTYET = 2, /* data pending */
129 SMC_URG_READ = 3, /* data was already read */
130};
131
132struct smc_connection {
133 struct rb_node alert_node;
134 struct smc_link_group *lgr; /* link group of connection */
135 struct smc_link *lnk; /* assigned SMC-R link */
136 u32 alert_token_local; /* unique conn. id */
137 u8 peer_rmbe_idx; /* from tcp handshake */
138 int peer_rmbe_size; /* size of peer rx buffer */
139 atomic_t peer_rmbe_space;/* remaining free bytes in peer
140 * rmbe
141 */
142 int rtoken_idx; /* idx to peer RMB rkey/addr */
143
144 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
145 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
146 int rmbe_size_short;/* compressed notation */
147 int rmbe_update_limit;
148 /* lower limit for consumer
149 * cursor update
150 */
151
152 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
153 * buffer for CDC msg send
154 * .prod cf. TCP snd_nxt
155 * .cons cf. TCP sends ack
156 */
157 union smc_host_cursor local_tx_ctrl_fin;
158 /* prod crsr - confirmed by peer
159 */
160 union smc_host_cursor tx_curs_prep; /* tx - prepared data
161 * snd_max..wmem_alloc
162 */
163 union smc_host_cursor tx_curs_sent; /* tx - sent data
164 * snd_nxt ?
165 */
166 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
167 * snd-wnd-begin ?
168 */
169 atomic_t sndbuf_space; /* remaining space in sndbuf */
170 u16 tx_cdc_seq; /* sequence # for CDC send */
171 u16 tx_cdc_seq_fin; /* sequence # - tx completed */
172 spinlock_t send_lock; /* protect wr_sends */
173 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
174 u32 tx_off; /* base offset in peer rmb */
175
176 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
177 * .prod cf. TCP rcv_nxt
178 * .cons cf. TCP snd_una
179 */
180 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
181 * source of snd_una ?
182 */
183 union smc_host_cursor urg_curs; /* points at urgent byte */
184 enum smc_urg_state urg_state;
185 bool urg_tx_pend; /* urgent data staged */
186 bool urg_rx_skip_pend;
187 /* indicate urgent oob data
188 * read, but previous regular
189 * data still pending
190 */
191 char urg_rx_byte; /* urgent byte */
192 atomic_t bytes_to_rcv; /* arrived data,
193 * not yet received
194 */
195 atomic_t splice_pending; /* number of spliced bytes
196 * pending processing
197 */
198#ifndef KERNEL_HAS_ATOMIC64
199 spinlock_t acurs_lock; /* protect cursors */
200#endif
201 struct work_struct close_work; /* peer sent some closing */
202 struct work_struct abort_work; /* abort the connection */
203 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */
204 u8 rx_off; /* receive offset:
205 * 0 for SMC-R, 32 for SMC-D
206 */
207 u64 peer_token; /* SMC-D token of peer */
208 u8 killed : 1; /* abnormal termination */
209 u8 out_of_sync : 1; /* out of sync with peer */
210};
211
212struct smc_sock { /* smc sock container */
213 struct sock sk;
214 struct socket *clcsock; /* internal tcp socket */
215 void (*clcsk_data_ready)(struct sock *sk);
216 /* original data_ready fct. **/
217 struct smc_connection conn; /* smc connection */
218 struct smc_sock *listen_smc; /* listen parent */
219 struct work_struct connect_work; /* handle non-blocking connect*/
220 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
221 struct work_struct smc_listen_work;/* prepare new accept socket */
222 struct list_head accept_q; /* sockets to be accepted */
223 spinlock_t accept_q_lock; /* protects accept_q */
224 bool use_fallback; /* fallback to tcp */
225 int fallback_rsn; /* reason for fallback */
226 u32 peer_diagnosis; /* decline reason from peer */
227 int sockopt_defer_accept;
228 /* sockopt TCP_DEFER_ACCEPT
229 * value
230 */
231 u8 wait_close_tx_prepared : 1;
232 /* shutdown wr or close
233 * started, waiting for unsent
234 * data to be sent
235 */
236 u8 connect_nonblock : 1;
237 /* non-blocking connect in
238 * flight
239 */
240 struct mutex clcsock_release_lock;
241 /* protects clcsock of a listen
242 * socket
243 * */
244};
245
246static inline struct smc_sock *smc_sk(const struct sock *sk)
247{
248 return (struct smc_sock *)sk;
249}
250
251extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
252extern struct workqueue_struct *smc_close_wq; /* wq for close work */
253
254#define SMC_SYSTEMID_LEN 8
255
256extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
257
258#define ntohll(x) be64_to_cpu(x)
259#define htonll(x) cpu_to_be64(x)
260
261/* convert an u32 value into network byte order, store it into a 3 byte field */
262static inline void hton24(u8 *net, u32 host)
263{
264 __be32 t;
265
266 t = cpu_to_be32(host);
267 memcpy(net, ((u8 *)&t) + 1, 3);
268}
269
270/* convert a received 3 byte field into host byte order*/
271static inline u32 ntoh24(u8 *net)
272{
273 __be32 t = 0;
274
275 memcpy(((u8 *)&t) + 1, net, 3);
276 return be32_to_cpu(t);
277}
278
279#ifdef CONFIG_XFRM
280static inline bool using_ipsec(struct smc_sock *smc)
281{
282 return (smc->clcsock->sk->sk_policy[0] ||
283 smc->clcsock->sk->sk_policy[1]) ? true : false;
284}
285#else
286static inline bool using_ipsec(struct smc_sock *smc)
287{
288 return false;
289}
290#endif
291
292struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
293void smc_close_non_accepted(struct sock *sk);
294
295#endif /* __SMC_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Definitions for the SMC module (socket related)
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
10 */
11#ifndef __SMC_H
12#define __SMC_H
13
14#include <linux/socket.h>
15#include <linux/types.h>
16#include <linux/compiler.h> /* __aligned */
17#include <net/sock.h>
18
19#include "smc_ib.h"
20
21#define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */
22#define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */
23
24#define SMC_MAX_PORTS 2 /* Max # of ports */
25
26extern struct proto smc_proto;
27extern struct proto smc_proto6;
28
29#ifdef ATOMIC64_INIT
30#define KERNEL_HAS_ATOMIC64
31#endif
32
33enum smc_state { /* possible states of an SMC socket */
34 SMC_ACTIVE = 1,
35 SMC_INIT = 2,
36 SMC_CLOSED = 7,
37 SMC_LISTEN = 10,
38 /* normal close */
39 SMC_PEERCLOSEWAIT1 = 20,
40 SMC_PEERCLOSEWAIT2 = 21,
41 SMC_APPFINCLOSEWAIT = 24,
42 SMC_APPCLOSEWAIT1 = 22,
43 SMC_APPCLOSEWAIT2 = 23,
44 SMC_PEERFINCLOSEWAIT = 25,
45 /* abnormal close */
46 SMC_PEERABORTWAIT = 26,
47 SMC_PROCESSABORT = 27,
48};
49
50struct smc_link_group;
51
52struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
53 u8 type;
54} __aligned(1);
55
56struct smc_cdc_conn_state_flags {
57#if defined(__BIG_ENDIAN_BITFIELD)
58 u8 peer_done_writing : 1; /* Sending done indicator */
59 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
60 u8 peer_conn_abort : 1; /* Abnormal close indicator */
61 u8 reserved : 5;
62#elif defined(__LITTLE_ENDIAN_BITFIELD)
63 u8 reserved : 5;
64 u8 peer_conn_abort : 1;
65 u8 peer_conn_closed : 1;
66 u8 peer_done_writing : 1;
67#endif
68};
69
70struct smc_cdc_producer_flags {
71#if defined(__BIG_ENDIAN_BITFIELD)
72 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
73 u8 urg_data_pending : 1; /* Urgent Data Pending */
74 u8 urg_data_present : 1; /* Urgent Data Present */
75 u8 cons_curs_upd_req : 1; /* cursor update requested */
76 u8 failover_validation : 1;/* message replay due to failover */
77 u8 reserved : 3;
78#elif defined(__LITTLE_ENDIAN_BITFIELD)
79 u8 reserved : 3;
80 u8 failover_validation : 1;
81 u8 cons_curs_upd_req : 1;
82 u8 urg_data_present : 1;
83 u8 urg_data_pending : 1;
84 u8 write_blocked : 1;
85#endif
86};
87
88/* in host byte order */
89union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
90 struct {
91 u16 reserved;
92 u16 wrap; /* window wrap sequence number */
93 u32 count; /* cursor (= offset) part */
94 };
95#ifdef KERNEL_HAS_ATOMIC64
96 atomic64_t acurs; /* for atomic processing */
97#else
98 u64 acurs; /* for atomic processing */
99#endif
100} __aligned(8);
101
102/* in host byte order, except for flag bitfields in network byte order */
103struct smc_host_cdc_msg { /* Connection Data Control message */
104 struct smc_wr_rx_hdr common; /* .type = 0xFE */
105 u8 len; /* length = 44 */
106 u16 seqno; /* connection seq # */
107 u32 token; /* alert_token */
108 union smc_host_cursor prod; /* producer cursor */
109 union smc_host_cursor cons; /* consumer cursor,
110 * piggy backed "ack"
111 */
112 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
113 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
114 u8 reserved[18];
115} __aligned(8);
116
117struct smc_connection {
118 struct rb_node alert_node;
119 struct smc_link_group *lgr; /* link group of connection */
120 u32 alert_token_local; /* unique conn. id */
121 u8 peer_conn_idx; /* from tcp handshake */
122 int peer_rmbe_size; /* size of peer rx buffer */
123 atomic_t peer_rmbe_space;/* remaining free bytes in peer
124 * rmbe
125 */
126 int rtoken_idx; /* idx to peer RMB rkey/addr */
127
128 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
129 int sndbuf_size; /* sndbuf size <== sock wmem */
130 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
131 int rmbe_size; /* RMBE size <== sock rmem */
132 int rmbe_size_short;/* compressed notation */
133 int rmbe_update_limit;
134 /* lower limit for consumer
135 * cursor update
136 */
137
138 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
139 * buffer for CDC msg send
140 * .prod cf. TCP snd_nxt
141 * .cons cf. TCP sends ack
142 */
143 union smc_host_cursor tx_curs_prep; /* tx - prepared data
144 * snd_max..wmem_alloc
145 */
146 union smc_host_cursor tx_curs_sent; /* tx - sent data
147 * snd_nxt ?
148 */
149 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
150 * snd-wnd-begin ?
151 */
152 atomic_t sndbuf_space; /* remaining space in sndbuf */
153 u16 tx_cdc_seq; /* sequence # for CDC send */
154 spinlock_t send_lock; /* protect wr_sends */
155 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
156
157 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
158 * .prod cf. TCP rcv_nxt
159 * .cons cf. TCP snd_una
160 */
161 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
162 * source of snd_una ?
163 */
164 atomic_t bytes_to_rcv; /* arrived data,
165 * not yet received
166 */
167#ifndef KERNEL_HAS_ATOMIC64
168 spinlock_t acurs_lock; /* protect cursors */
169#endif
170 struct work_struct close_work; /* peer sent some closing */
171};
172
173struct smc_sock { /* smc sock container */
174 struct sock sk;
175 struct socket *clcsock; /* internal tcp socket */
176 struct smc_connection conn; /* smc connection */
177 struct smc_sock *listen_smc; /* listen parent */
178 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
179 struct work_struct smc_listen_work;/* prepare new accept socket */
180 struct list_head accept_q; /* sockets to be accepted */
181 spinlock_t accept_q_lock; /* protects accept_q */
182 bool use_fallback; /* fallback to tcp */
183 u8 wait_close_tx_prepared : 1;
184 /* shutdown wr or close
185 * started, waiting for unsent
186 * data to be sent
187 */
188};
189
190static inline struct smc_sock *smc_sk(const struct sock *sk)
191{
192 return (struct smc_sock *)sk;
193}
194
195#define SMC_SYSTEMID_LEN 8
196
197extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
198
199/* convert an u32 value into network byte order, store it into a 3 byte field */
200static inline void hton24(u8 *net, u32 host)
201{
202 __be32 t;
203
204 t = cpu_to_be32(host);
205 memcpy(net, ((u8 *)&t) + 1, 3);
206}
207
208/* convert a received 3 byte field into host byte order*/
209static inline u32 ntoh24(u8 *net)
210{
211 __be32 t = 0;
212
213 memcpy(((u8 *)&t) + 1, net, 3);
214 return be32_to_cpu(t);
215}
216
217#define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */
218
219#define SMC_RMBE_SIZES 16 /* number of distinct sizes for an RMBE */
220/* theoretically, the RFC states that largest size would be 512K,
221 * i.e. compressed 5 and thus 6 sizes (0..5), despite
222 * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
223 */
224
225/* convert the RMB size into the compressed notation - minimum 16K.
226 * In contrast to plain ilog2, this rounds towards the next power of 2,
227 * so the socket application gets at least its desired sndbuf / rcvbuf size.
228 */
229static inline u8 smc_compress_bufsize(int size)
230{
231 u8 compressed;
232
233 if (size <= SMC_BUF_MIN_SIZE)
234 return 0;
235
236 size = (size - 1) >> 14;
237 compressed = ilog2(size) + 1;
238 if (compressed >= SMC_RMBE_SIZES)
239 compressed = SMC_RMBE_SIZES - 1;
240 return compressed;
241}
242
243/* convert the RMB size from compressed notation into integer */
244static inline int smc_uncompress_bufsize(u8 compressed)
245{
246 u32 size;
247
248 size = 0x00000001 << (((int)compressed) + 14);
249 return (int)size;
250}
251
252#ifdef CONFIG_XFRM
253static inline bool using_ipsec(struct smc_sock *smc)
254{
255 return (smc->clcsock->sk->sk_policy[0] ||
256 smc->clcsock->sk->sk_policy[1]) ? true : false;
257}
258#else
259static inline bool using_ipsec(struct smc_sock *smc)
260{
261 return false;
262}
263#endif
264
265struct smc_clc_msg_local;
266
267void smc_conn_free(struct smc_connection *conn);
268int smc_conn_create(struct smc_sock *smc,
269 struct smc_ib_device *smcibdev, u8 ibport,
270 struct smc_clc_msg_local *lcl, int srv_first_contact);
271struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
272void smc_close_non_accepted(struct sock *sk);
273
274#endif /* __SMC_H */