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/genetlink.h>
18#include <net/sock.h>
19
20#include "smc_ib.h"
21
22#define SMC_V1 1 /* SMC version V1 */
23#define SMC_V2 2 /* SMC version V2 */
24
25#define SMC_RELEASE_0 0
26#define SMC_RELEASE_1 1
27#define SMC_RELEASE SMC_RELEASE_1 /* the latest release version */
28
29#define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */
30#define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */
31
32#define SMC_AUTOCORKING_DEFAULT_SIZE 0x10000 /* 64K by default */
33
34extern struct proto smc_proto;
35extern struct proto smc_proto6;
36
37#ifdef ATOMIC64_INIT
38#define KERNEL_HAS_ATOMIC64
39#endif
40
41enum smc_state { /* possible states of an SMC socket */
42 SMC_ACTIVE = 1,
43 SMC_INIT = 2,
44 SMC_CLOSED = 7,
45 SMC_LISTEN = 10,
46 /* normal close */
47 SMC_PEERCLOSEWAIT1 = 20,
48 SMC_PEERCLOSEWAIT2 = 21,
49 SMC_APPFINCLOSEWAIT = 24,
50 SMC_APPCLOSEWAIT1 = 22,
51 SMC_APPCLOSEWAIT2 = 23,
52 SMC_PEERFINCLOSEWAIT = 25,
53 /* abnormal close */
54 SMC_PEERABORTWAIT = 26,
55 SMC_PROCESSABORT = 27,
56};
57
58enum smc_supplemental_features {
59 SMC_SPF_EMULATED_ISM_DEV = 0,
60};
61
62#define SMC_FEATURE_MASK \
63 (BIT(SMC_SPF_EMULATED_ISM_DEV))
64
65struct smc_link_group;
66
67struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
68 union {
69 u8 type;
70#if defined(__BIG_ENDIAN_BITFIELD)
71 struct {
72 u8 llc_version:4,
73 llc_type:4;
74 };
75#elif defined(__LITTLE_ENDIAN_BITFIELD)
76 struct {
77 u8 llc_type:4,
78 llc_version:4;
79 };
80#endif
81 };
82} __aligned(1);
83
84struct smc_cdc_conn_state_flags {
85#if defined(__BIG_ENDIAN_BITFIELD)
86 u8 peer_done_writing : 1; /* Sending done indicator */
87 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
88 u8 peer_conn_abort : 1; /* Abnormal close indicator */
89 u8 reserved : 5;
90#elif defined(__LITTLE_ENDIAN_BITFIELD)
91 u8 reserved : 5;
92 u8 peer_conn_abort : 1;
93 u8 peer_conn_closed : 1;
94 u8 peer_done_writing : 1;
95#endif
96};
97
98struct smc_cdc_producer_flags {
99#if defined(__BIG_ENDIAN_BITFIELD)
100 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
101 u8 urg_data_pending : 1; /* Urgent Data Pending */
102 u8 urg_data_present : 1; /* Urgent Data Present */
103 u8 cons_curs_upd_req : 1; /* cursor update requested */
104 u8 failover_validation : 1;/* message replay due to failover */
105 u8 reserved : 3;
106#elif defined(__LITTLE_ENDIAN_BITFIELD)
107 u8 reserved : 3;
108 u8 failover_validation : 1;
109 u8 cons_curs_upd_req : 1;
110 u8 urg_data_present : 1;
111 u8 urg_data_pending : 1;
112 u8 write_blocked : 1;
113#endif
114};
115
116/* in host byte order */
117union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
118 struct {
119 u16 reserved;
120 u16 wrap; /* window wrap sequence number */
121 u32 count; /* cursor (= offset) part */
122 };
123#ifdef KERNEL_HAS_ATOMIC64
124 atomic64_t acurs; /* for atomic processing */
125#else
126 u64 acurs; /* for atomic processing */
127#endif
128} __aligned(8);
129
130/* in host byte order, except for flag bitfields in network byte order */
131struct smc_host_cdc_msg { /* Connection Data Control message */
132 struct smc_wr_rx_hdr common; /* .type = 0xFE */
133 u8 len; /* length = 44 */
134 u16 seqno; /* connection seq # */
135 u32 token; /* alert_token */
136 union smc_host_cursor prod; /* producer cursor */
137 union smc_host_cursor cons; /* consumer cursor,
138 * piggy backed "ack"
139 */
140 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
141 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
142 u8 reserved[18];
143} __aligned(8);
144
145enum smc_urg_state {
146 SMC_URG_VALID = 1, /* data present */
147 SMC_URG_NOTYET = 2, /* data pending */
148 SMC_URG_READ = 3, /* data was already read */
149};
150
151struct smc_mark_woken {
152 bool woken;
153 void *key;
154 wait_queue_entry_t wait_entry;
155};
156
157struct smc_connection {
158 struct rb_node alert_node;
159 struct smc_link_group *lgr; /* link group of connection */
160 struct smc_link *lnk; /* assigned SMC-R link */
161 u32 alert_token_local; /* unique conn. id */
162 u8 peer_rmbe_idx; /* from tcp handshake */
163 int peer_rmbe_size; /* size of peer rx buffer */
164 atomic_t peer_rmbe_space;/* remaining free bytes in peer
165 * rmbe
166 */
167 int rtoken_idx; /* idx to peer RMB rkey/addr */
168
169 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
170 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
171 int rmbe_size_comp; /* compressed notation */
172 int rmbe_update_limit;
173 /* lower limit for consumer
174 * cursor update
175 */
176
177 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
178 * buffer for CDC msg send
179 * .prod cf. TCP snd_nxt
180 * .cons cf. TCP sends ack
181 */
182 union smc_host_cursor local_tx_ctrl_fin;
183 /* prod crsr - confirmed by peer
184 */
185 union smc_host_cursor tx_curs_prep; /* tx - prepared data
186 * snd_max..wmem_alloc
187 */
188 union smc_host_cursor tx_curs_sent; /* tx - sent data
189 * snd_nxt ?
190 */
191 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
192 * snd-wnd-begin ?
193 */
194 atomic_t sndbuf_space; /* remaining space in sndbuf */
195 u16 tx_cdc_seq; /* sequence # for CDC send */
196 u16 tx_cdc_seq_fin; /* sequence # - tx completed */
197 spinlock_t send_lock; /* protect wr_sends */
198 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe
199 * - inc when post wqe,
200 * - dec on polled tx cqe
201 */
202 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/
203 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
204 u32 tx_off; /* base offset in peer rmb */
205
206 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
207 * .prod cf. TCP rcv_nxt
208 * .cons cf. TCP snd_una
209 */
210 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
211 * source of snd_una ?
212 */
213 union smc_host_cursor urg_curs; /* points at urgent byte */
214 enum smc_urg_state urg_state;
215 bool urg_tx_pend; /* urgent data staged */
216 bool urg_rx_skip_pend;
217 /* indicate urgent oob data
218 * read, but previous regular
219 * data still pending
220 */
221 char urg_rx_byte; /* urgent byte */
222 bool tx_in_release_sock;
223 /* flush pending tx data in
224 * sock release_cb()
225 */
226 atomic_t bytes_to_rcv; /* arrived data,
227 * not yet received
228 */
229 atomic_t splice_pending; /* number of spliced bytes
230 * pending processing
231 */
232#ifndef KERNEL_HAS_ATOMIC64
233 spinlock_t acurs_lock; /* protect cursors */
234#endif
235 struct work_struct close_work; /* peer sent some closing */
236 struct work_struct abort_work; /* abort the connection */
237 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */
238 u8 rx_off; /* receive offset:
239 * 0 for SMC-R, 32 for SMC-D
240 */
241 u64 peer_token; /* SMC-D token of peer */
242 u8 killed : 1; /* abnormal termination */
243 u8 freed : 1; /* normal termiation */
244 u8 out_of_sync : 1; /* out of sync with peer */
245};
246
247struct smc_sock { /* smc sock container */
248 struct sock sk;
249 struct socket *clcsock; /* internal tcp socket */
250 void (*clcsk_state_change)(struct sock *sk);
251 /* original stat_change fct. */
252 void (*clcsk_data_ready)(struct sock *sk);
253 /* original data_ready fct. */
254 void (*clcsk_write_space)(struct sock *sk);
255 /* original write_space fct. */
256 void (*clcsk_error_report)(struct sock *sk);
257 /* original error_report fct. */
258 struct smc_connection conn; /* smc connection */
259 struct smc_sock *listen_smc; /* listen parent */
260 struct work_struct connect_work; /* handle non-blocking connect*/
261 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
262 struct work_struct smc_listen_work;/* prepare new accept socket */
263 struct list_head accept_q; /* sockets to be accepted */
264 spinlock_t accept_q_lock; /* protects accept_q */
265 bool limit_smc_hs; /* put constraint on handshake */
266 bool use_fallback; /* fallback to tcp */
267 int fallback_rsn; /* reason for fallback */
268 u32 peer_diagnosis; /* decline reason from peer */
269 atomic_t queued_smc_hs; /* queued smc handshakes */
270 struct inet_connection_sock_af_ops af_ops;
271 const struct inet_connection_sock_af_ops *ori_af_ops;
272 /* original af ops */
273 int sockopt_defer_accept;
274 /* sockopt TCP_DEFER_ACCEPT
275 * value
276 */
277 u8 wait_close_tx_prepared : 1;
278 /* shutdown wr or close
279 * started, waiting for unsent
280 * data to be sent
281 */
282 u8 connect_nonblock : 1;
283 /* non-blocking connect in
284 * flight
285 */
286 struct mutex clcsock_release_lock;
287 /* protects clcsock of a listen
288 * socket
289 * */
290};
291
292#define smc_sk(ptr) container_of_const(ptr, struct smc_sock, sk)
293
294static inline void smc_init_saved_callbacks(struct smc_sock *smc)
295{
296 smc->clcsk_state_change = NULL;
297 smc->clcsk_data_ready = NULL;
298 smc->clcsk_write_space = NULL;
299 smc->clcsk_error_report = NULL;
300}
301
302static inline struct smc_sock *smc_clcsock_user_data(const struct sock *clcsk)
303{
304 return (struct smc_sock *)
305 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY);
306}
307
308/* save target_cb in saved_cb, and replace target_cb with new_cb */
309static inline void smc_clcsock_replace_cb(void (**target_cb)(struct sock *),
310 void (*new_cb)(struct sock *),
311 void (**saved_cb)(struct sock *))
312{
313 /* only save once */
314 if (!*saved_cb)
315 *saved_cb = *target_cb;
316 *target_cb = new_cb;
317}
318
319/* restore target_cb to saved_cb, and reset saved_cb to NULL */
320static inline void smc_clcsock_restore_cb(void (**target_cb)(struct sock *),
321 void (**saved_cb)(struct sock *))
322{
323 if (!*saved_cb)
324 return;
325 *target_cb = *saved_cb;
326 *saved_cb = NULL;
327}
328
329extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
330extern struct workqueue_struct *smc_close_wq; /* wq for close work */
331
332#define SMC_SYSTEMID_LEN 8
333
334extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
335
336#define ntohll(x) be64_to_cpu(x)
337#define htonll(x) cpu_to_be64(x)
338
339/* convert an u32 value into network byte order, store it into a 3 byte field */
340static inline void hton24(u8 *net, u32 host)
341{
342 __be32 t;
343
344 t = cpu_to_be32(host);
345 memcpy(net, ((u8 *)&t) + 1, 3);
346}
347
348/* convert a received 3 byte field into host byte order*/
349static inline u32 ntoh24(u8 *net)
350{
351 __be32 t = 0;
352
353 memcpy(((u8 *)&t) + 1, net, 3);
354 return be32_to_cpu(t);
355}
356
357#ifdef CONFIG_XFRM
358static inline bool using_ipsec(struct smc_sock *smc)
359{
360 return (smc->clcsock->sk->sk_policy[0] ||
361 smc->clcsock->sk->sk_policy[1]) ? true : false;
362}
363#else
364static inline bool using_ipsec(struct smc_sock *smc)
365{
366 return false;
367}
368#endif
369
370struct smc_gidlist;
371
372struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
373void smc_close_non_accepted(struct sock *sk);
374void smc_fill_gid_list(struct smc_link_group *lgr,
375 struct smc_gidlist *gidlist,
376 struct smc_ib_device *known_dev, u8 *known_gid);
377
378/* smc handshake limitation interface for netlink */
379int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb);
380int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info);
381int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info);
382
383static inline void smc_sock_set_flag(struct sock *sk, enum sock_flags flag)
384{
385 set_bit(nr: flag, addr: &sk->sk_flags);
386}
387
388#endif /* __SMC_H */
389

source code of linux/net/smc/smc.h