projects/linux·blob master · refs

434 lines · 12.9 KiB

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 <[email protected]>
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
37extern struct smc_hashinfo smc_v4_hashinfo;
38extern struct smc_hashinfo smc_v6_hashinfo;
39
40int smc_hash_sk(struct sock *sk);
41void smc_unhash_sk(struct sock *sk);
42void smc_release_cb(struct sock *sk);
43
44int smc_release(struct socket *sock);
45int smc_bind(struct socket *sock, struct sockaddr_unsized *uaddr,
46 int addr_len);
47int smc_connect(struct socket *sock, struct sockaddr_unsized *addr,
48 int alen, int flags);
49int smc_accept(struct socket *sock, struct socket *new_sock,
50 struct proto_accept_arg *arg);
51int smc_getname(struct socket *sock, struct sockaddr *addr,
52 int peer);
53__poll_t smc_poll(struct file *file, struct socket *sock,
54 poll_table *wait);
55int smc_ioctl(struct socket *sock, unsigned int cmd,
56 unsigned long arg);
57int smc_listen(struct socket *sock, int backlog);
58int smc_shutdown(struct socket *sock, int how);
59int smc_setsockopt(struct socket *sock, int level, int optname,
60 sockptr_t optval, unsigned int optlen);
61int smc_getsockopt(struct socket *sock, int level, int optname,
62 char __user *optval, int __user *optlen);
63int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len);
64int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
65 int flags);
66ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
67 struct pipe_inode_info *pipe, size_t len,
68 unsigned int flags);
69
70/* smc sock initialization */
71void smc_sk_init(struct net *net, struct sock *sk, int protocol);
72/* clcsock initialization */
73int smc_create_clcsk(struct net *net, struct sock *sk, int family);
74
75#ifdef ATOMIC64_INIT
76#define KERNEL_HAS_ATOMIC64
77#endif
78
79enum smc_state { /* possible states of an SMC socket */
80 SMC_ACTIVE = 1,
81 SMC_INIT = 2,
82 SMC_CLOSED = 7,
83 SMC_LISTEN = 10,
84 /* normal close */
85 SMC_PEERCLOSEWAIT1 = 20,
86 SMC_PEERCLOSEWAIT2 = 21,
87 SMC_APPFINCLOSEWAIT = 24,
88 SMC_APPCLOSEWAIT1 = 22,
89 SMC_APPCLOSEWAIT2 = 23,
90 SMC_PEERFINCLOSEWAIT = 25,
91 /* abnormal close */
92 SMC_PEERABORTWAIT = 26,
93 SMC_PROCESSABORT = 27,
94};
95
96enum smc_supplemental_features {
97 SMC_SPF_EMULATED_ISM_DEV = 0,
98};
99
100#define SMC_FEATURE_MASK \
101 (BIT(SMC_SPF_EMULATED_ISM_DEV))
102
103struct smc_link_group;
104
105struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */
106 union {
107 u8 type;
108#if defined(__BIG_ENDIAN_BITFIELD)
109 struct {
110 u8 llc_version:4,
111 llc_type:4;
112 };
113#elif defined(__LITTLE_ENDIAN_BITFIELD)
114 struct {
115 u8 llc_type:4,
116 llc_version:4;
117 };
118#endif
119 };
120} __aligned(1);
121
122struct smc_cdc_conn_state_flags {
123#if defined(__BIG_ENDIAN_BITFIELD)
124 u8 peer_done_writing : 1; /* Sending done indicator */
125 u8 peer_conn_closed : 1; /* Peer connection closed indicator */
126 u8 peer_conn_abort : 1; /* Abnormal close indicator */
127 u8 reserved : 5;
128#elif defined(__LITTLE_ENDIAN_BITFIELD)
129 u8 reserved : 5;
130 u8 peer_conn_abort : 1;
131 u8 peer_conn_closed : 1;
132 u8 peer_done_writing : 1;
133#endif
134};
135
136struct smc_cdc_producer_flags {
137#if defined(__BIG_ENDIAN_BITFIELD)
138 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */
139 u8 urg_data_pending : 1; /* Urgent Data Pending */
140 u8 urg_data_present : 1; /* Urgent Data Present */
141 u8 cons_curs_upd_req : 1; /* cursor update requested */
142 u8 failover_validation : 1;/* message replay due to failover */
143 u8 reserved : 3;
144#elif defined(__LITTLE_ENDIAN_BITFIELD)
145 u8 reserved : 3;
146 u8 failover_validation : 1;
147 u8 cons_curs_upd_req : 1;
148 u8 urg_data_present : 1;
149 u8 urg_data_pending : 1;
150 u8 write_blocked : 1;
151#endif
152};
153
154/* in host byte order */
155union smc_host_cursor { /* SMC cursor - an offset in an RMBE */
156 struct {
157 u16 reserved;
158 u16 wrap; /* window wrap sequence number */
159 u32 count; /* cursor (= offset) part */
160 };
161#ifdef KERNEL_HAS_ATOMIC64
162 atomic64_t acurs; /* for atomic processing */
163#else
164 u64 acurs; /* for atomic processing */
165#endif
166} __aligned(8);
167
168/* in host byte order, except for flag bitfields in network byte order */
169struct smc_host_cdc_msg { /* Connection Data Control message */
170 struct smc_wr_rx_hdr common; /* .type = 0xFE */
171 u8 len; /* length = 44 */
172 u16 seqno; /* connection seq # */
173 u32 token; /* alert_token */
174 union smc_host_cursor prod; /* producer cursor */
175 union smc_host_cursor cons; /* consumer cursor,
176 * piggy backed "ack"
177 */
178 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */
179 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/
180 u8 reserved[18];
181} __aligned(8);
182
183enum smc_urg_state {
184 SMC_URG_VALID = 1, /* data present */
185 SMC_URG_NOTYET = 2, /* data pending */
186 SMC_URG_READ = 3, /* data was already read */
187};
188
189struct smc_mark_woken {
190 bool woken;
191 void *key;
192 wait_queue_entry_t wait_entry;
193};
194
195struct smc_connection {
196 struct rb_node alert_node;
197 struct smc_link_group *lgr; /* link group of connection */
198 struct smc_link *lnk; /* assigned SMC-R link */
199 u32 alert_token_local; /* unique conn. id */
200 u8 peer_rmbe_idx; /* from tcp handshake */
201 int peer_rmbe_size; /* size of peer rx buffer */
202 atomic_t peer_rmbe_space;/* remaining free bytes in peer
203 * rmbe
204 */
205 int rtoken_idx; /* idx to peer RMB rkey/addr */
206
207 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */
208 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */
209 int rmbe_size_comp; /* compressed notation */
210 int rmbe_update_limit;
211 /* lower limit for consumer
212 * cursor update
213 */
214
215 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging
216 * buffer for CDC msg send
217 * .prod cf. TCP snd_nxt
218 * .cons cf. TCP sends ack
219 */
220 union smc_host_cursor local_tx_ctrl_fin;
221 /* prod crsr - confirmed by peer
222 */
223 union smc_host_cursor tx_curs_prep; /* tx - prepared data
224 * snd_max..wmem_alloc
225 */
226 union smc_host_cursor tx_curs_sent; /* tx - sent data
227 * snd_nxt ?
228 */
229 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer
230 * snd-wnd-begin ?
231 */
232 atomic_t sndbuf_space; /* remaining space in sndbuf */
233 u16 tx_cdc_seq; /* sequence # for CDC send */
234 u16 tx_cdc_seq_fin; /* sequence # - tx completed */
235 spinlock_t send_lock; /* protect wr_sends */
236 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe
237 * - inc when post wqe,
238 * - dec on polled tx cqe
239 */
240 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/
241 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */
242 u32 tx_off; /* base offset in peer rmb */
243
244 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl.
245 * .prod cf. TCP rcv_nxt
246 * .cons cf. TCP snd_una
247 */
248 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer
249 * source of snd_una ?
250 */
251 union smc_host_cursor urg_curs; /* points at urgent byte */
252 enum smc_urg_state urg_state;
253 bool urg_tx_pend; /* urgent data staged */
254 bool urg_rx_skip_pend;
255 /* indicate urgent oob data
256 * read, but previous regular
257 * data still pending
258 */
259 char urg_rx_byte; /* urgent byte */
260 bool tx_in_release_sock;
261 /* flush pending tx data in
262 * sock release_cb()
263 */
264 atomic_t bytes_to_rcv; /* arrived data,
265 * not yet received
266 */
267 atomic_t splice_pending; /* number of spliced bytes
268 * pending processing
269 */
270#ifndef KERNEL_HAS_ATOMIC64
271 spinlock_t acurs_lock; /* protect cursors */
272#endif
273 struct work_struct close_work; /* peer sent some closing */
274 struct work_struct abort_work; /* abort the connection */
275 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */
276 u8 rx_off; /* receive offset:
277 * 0 for SMC-R, 32 for SMC-D
278 */
279 u64 peer_token; /* SMC-D token of peer */
280 u8 killed; /* abnormal termination */
281 u8 freed; /* normal termination */
282 u8 out_of_sync; /* out of sync with peer */
283};
284
285struct smc_sock { /* smc sock container */
286 union {
287 struct sock sk;
288 struct inet_sock icsk_inet;
289 };
290 struct socket *clcsock; /* internal tcp socket */
291 void (*clcsk_state_change)(struct sock *sk);
292 /* original stat_change fct. */
293 void (*clcsk_data_ready)(struct sock *sk);
294 /* original data_ready fct. */
295 void (*clcsk_write_space)(struct sock *sk);
296 /* original write_space fct. */
297 void (*clcsk_error_report)(struct sock *sk);
298 /* original error_report fct. */
299 struct smc_connection conn; /* smc connection */
300 struct smc_sock *listen_smc; /* listen parent */
301 struct work_struct connect_work; /* handle non-blocking connect*/
302 struct work_struct tcp_listen_work;/* handle tcp socket accepts */
303 struct work_struct smc_listen_work;/* prepare new accept socket */
304 struct list_head accept_q; /* sockets to be accepted */
305 spinlock_t accept_q_lock; /* protects accept_q */
306 bool limit_smc_hs; /* put constraint on handshake */
307 bool use_fallback; /* fallback to tcp */
308 int fallback_rsn; /* reason for fallback */
309 u32 peer_diagnosis; /* decline reason from peer */
310 atomic_t queued_smc_hs; /* queued smc handshakes */
311 struct inet_connection_sock_af_ops af_ops;
312 const struct inet_connection_sock_af_ops *ori_af_ops;
313 /* original af ops */
314 int sockopt_defer_accept;
315 /* sockopt TCP_DEFER_ACCEPT
316 * value
317 */
318 u8 wait_close_tx_prepared : 1;
319 /* shutdown wr or close
320 * started, waiting for unsent
321 * data to be sent
322 */
323 u8 connect_nonblock : 1;
324 /* non-blocking connect in
325 * flight
326 */
327 struct mutex clcsock_release_lock;
328 /* protects clcsock of a listen
329 * socket
330 * */
331};
332
333#define smc_sk(ptr) container_of_const(ptr, struct smc_sock, sk)
334
335static inline void smc_init_saved_callbacks(struct smc_sock *smc)
336{
337 smc->clcsk_state_change = NULL;
338 smc->clcsk_data_ready = NULL;
339 smc->clcsk_write_space = NULL;
340 smc->clcsk_error_report = NULL;
341}
342
343static inline struct smc_sock *smc_clcsock_user_data(const struct sock *clcsk)
344{
345 return (struct smc_sock *)
346 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY);
347}
348
349static inline struct smc_sock *smc_clcsock_user_data_rcu(const struct sock *clcsk)
350{
351 return (struct smc_sock *)rcu_dereference_sk_user_data(clcsk);
352}
353
354/* save target_cb in saved_cb, and replace target_cb with new_cb */
355static inline void smc_clcsock_replace_cb(void (**target_cb)(struct sock *),
356 void (*new_cb)(struct sock *),
357 void (**saved_cb)(struct sock *))
358{
359 /* only save once */
360 if (!*saved_cb)
361 *saved_cb = *target_cb;
362 *target_cb = new_cb;
363}
364
365/* restore target_cb to saved_cb, and reset saved_cb to NULL */
366static inline void smc_clcsock_restore_cb(void (**target_cb)(struct sock *),
367 void (**saved_cb)(struct sock *))
368{
369 if (!*saved_cb)
370 return;
371 *target_cb = *saved_cb;
372 *saved_cb = NULL;
373}
374
375extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
376extern struct workqueue_struct *smc_close_wq; /* wq for close work */
377
378#define SMC_SYSTEMID_LEN 8
379
380extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */
381
382#define ntohll(x) be64_to_cpu(x)
383#define htonll(x) cpu_to_be64(x)
384
385/* convert an u32 value into network byte order, store it into a 3 byte field */
386static inline void hton24(u8 *net, u32 host)
387{
388 __be32 t;
389
390 t = cpu_to_be32(host);
391 memcpy(net, ((u8 *)&t) + 1, 3);
392}
393
394/* convert a received 3 byte field into host byte order*/
395static inline u32 ntoh24(u8 *net)
396{
397 __be32 t = 0;
398
399 memcpy(((u8 *)&t) + 1, net, 3);
400 return be32_to_cpu(t);
401}
402
403#ifdef CONFIG_XFRM
404static inline bool using_ipsec(struct smc_sock *smc)
405{
406 return (smc->clcsock->sk->sk_policy[0] ||
407 smc->clcsock->sk->sk_policy[1]) ? true : false;
408}
409#else
410static inline bool using_ipsec(struct smc_sock *smc)
411{
412 return false;
413}
414#endif
415
416struct smc_gidlist;
417
418struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock);
419void smc_close_non_accepted(struct sock *sk);
420void smc_fill_gid_list(struct smc_link_group *lgr,
421 struct smc_gidlist *gidlist,
422 struct smc_ib_device *known_dev, u8 *known_gid);
423
424/* smc handshake limitation interface for netlink */
425int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb);
426int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info);
427int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info);
428
429static inline void smc_sock_set_flag(struct sock *sk, enum sock_flags flag)
430{
431 set_bit(flag, &sk->sk_flags);
432}
433
434#endif /* __SMC_H */