xref: /freebsd/sys/dev/irdma/irdma_cm.h (revision 01fbb869)
1 /*-
2  * SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
3  *
4  * Copyright (c) 2015 - 2023 Intel Corporation
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenFabrics.org BSD license below:
11  *
12  *   Redistribution and use in source and binary forms, with or
13  *   without modification, are permitted provided that the following
14  *   conditions are met:
15  *
16  *    - Redistributions of source code must retain the above
17  *	copyright notice, this list of conditions and the following
18  *	disclaimer.
19  *
20  *    - Redistributions in binary form must reproduce the above
21  *	copyright notice, this list of conditions and the following
22  *	disclaimer in the documentation and/or other materials
23  *	provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #ifndef IRDMA_CM_H
36 #define IRDMA_CM_H
37 
38 #define IRDMA_MPA_REQUEST_ACCEPT	1
39 #define IRDMA_MPA_REQUEST_REJECT	2
40 
41 /* IETF MPA -- defines */
42 #define IEFT_MPA_KEY_REQ	"MPA ID Req Frame"
43 #define IEFT_MPA_KEY_REP	"MPA ID Rep Frame"
44 #define IETF_MPA_KEY_SIZE	16
45 #define IETF_MPA_VER		1
46 #define IETF_MAX_PRIV_DATA_LEN	512
47 #define IETF_MPA_FRAME_SIZE	20
48 #define IETF_RTR_MSG_SIZE	4
49 #define IETF_MPA_V2_FLAG	0x10
50 #define SNDMARKER_SEQNMASK	0x000001ff
51 #define IRDMA_MAX_IETF_SIZE	32
52 
53 /* IETF RTR MSG Fields */
54 #define IETF_PEER_TO_PEER	0x8000
55 #define IETF_FLPDU_ZERO_LEN	0x4000
56 #define IETF_RDMA0_WRITE	0x8000
57 #define IETF_RDMA0_READ		0x4000
58 #define IETF_NO_IRD_ORD		0x3fff
59 
60 #define MAX_PORTS	65536
61 
62 #define IRDMA_PASSIVE_STATE_INDICATED	0
63 #define IRDMA_DO_NOT_SEND_RESET_EVENT	1
64 #define IRDMA_SEND_RESET_EVENT		2
65 
66 #define MAX_IRDMA_IFS	4
67 
68 #define SET_ACK		1
69 #define SET_SYN		2
70 #define SET_FIN		4
71 #define SET_RST		8
72 
73 #define TCP_OPTIONS_PADDING	3
74 
75 #define IRDMA_DEFAULT_RETRYS	64
76 #define IRDMA_DEFAULT_RETRANS	32
77 #define IRDMA_DEFAULT_TTL		0x40
78 #define IRDMA_DEFAULT_RTT_VAR		6
79 #define IRDMA_DEFAULT_SS_THRESH		0x3fffffff
80 #define IRDMA_DEFAULT_REXMIT_THRESH	8
81 
82 #define IRDMA_RETRY_TIMEOUT	HZ
83 #define IRDMA_SHORT_TIME	10
84 #define IRDMA_LONG_TIME		(2 * HZ)
85 #define IRDMA_MAX_TIMEOUT	((unsigned long)(12 * HZ))
86 
87 #define IRDMA_CM_HASHTABLE_SIZE		1024
88 #define IRDMA_CM_TCP_TIMER_INTERVAL	3000
89 #define IRDMA_CM_DEFAULT_MTU		1540
90 #define IRDMA_CM_DEFAULT_FRAME_CNT	10
91 #define IRDMA_CM_THREAD_STACK_SIZE	256
92 #define IRDMA_CM_DEFAULT_RCV_WND	64240
93 #define IRDMA_CM_DEFAULT_RCV_WND_SCALED	0x3FFFC
94 #define IRDMA_CM_DEFAULT_RCV_WND_SCALE	2
95 #define IRDMA_CM_DEFAULT_FREE_PKTS	10
96 #define IRDMA_CM_FREE_PKT_LO_WATERMARK	2
97 #define IRDMA_CM_DEFAULT_MSS		536
98 #define IRDMA_CM_DEFAULT_MPA_VER	2
99 #define IRDMA_CM_DEFAULT_SEQ		0x159bf75f
100 #define IRDMA_CM_DEFAULT_LOCAL_ID	0x3b47
101 #define IRDMA_CM_DEFAULT_SEQ2		0x18ed5740
102 #define IRDMA_CM_DEFAULT_LOCAL_ID2	0xb807
103 #define IRDMA_MAX_CM_BUF		(IRDMA_MAX_IETF_SIZE + IETF_MAX_PRIV_DATA_LEN)
104 
105 enum ietf_mpa_flags {
106 	IETF_MPA_FLAGS_REJECT  = 0x20,
107 	IETF_MPA_FLAGS_CRC     = 0x40,
108 	IETF_MPA_FLAGS_MARKERS = 0x80,
109 };
110 
111 enum irdma_timer_type {
112 	IRDMA_TIMER_TYPE_SEND,
113 	IRDMA_TIMER_TYPE_CLOSE,
114 };
115 
116 enum option_nums {
117 	OPTION_NUM_EOL,
118 	OPTION_NUM_NONE,
119 	OPTION_NUM_MSS,
120 	OPTION_NUM_WINDOW_SCALE,
121 	OPTION_NUM_SACK_PERM,
122 	OPTION_NUM_SACK,
123 	OPTION_NUM_WRITE0 = 0xbc,
124 };
125 
126 /* cm node transition states */
127 enum irdma_cm_node_state {
128 	IRDMA_CM_STATE_UNKNOWN,
129 	IRDMA_CM_STATE_INITED,
130 	IRDMA_CM_STATE_LISTENING,
131 	IRDMA_CM_STATE_SYN_RCVD,
132 	IRDMA_CM_STATE_SYN_SENT,
133 	IRDMA_CM_STATE_ONE_SIDE_ESTABLISHED,
134 	IRDMA_CM_STATE_ESTABLISHED,
135 	IRDMA_CM_STATE_ACCEPTING,
136 	IRDMA_CM_STATE_MPAREQ_SENT,
137 	IRDMA_CM_STATE_MPAREQ_RCVD,
138 	IRDMA_CM_STATE_MPAREJ_RCVD,
139 	IRDMA_CM_STATE_OFFLOADED,
140 	IRDMA_CM_STATE_FIN_WAIT1,
141 	IRDMA_CM_STATE_FIN_WAIT2,
142 	IRDMA_CM_STATE_CLOSE_WAIT,
143 	IRDMA_CM_STATE_TIME_WAIT,
144 	IRDMA_CM_STATE_LAST_ACK,
145 	IRDMA_CM_STATE_CLOSING,
146 	IRDMA_CM_STATE_LISTENER_DESTROYED,
147 	IRDMA_CM_STATE_CLOSED,
148 };
149 
150 enum mpa_frame_ver {
151 	IETF_MPA_V1 = 1,
152 	IETF_MPA_V2 = 2,
153 };
154 
155 enum mpa_frame_key {
156 	MPA_KEY_REQUEST,
157 	MPA_KEY_REPLY,
158 };
159 
160 enum send_rdma0 {
161 	SEND_RDMA_READ_ZERO  = 1,
162 	SEND_RDMA_WRITE_ZERO = 2,
163 };
164 
165 enum irdma_tcpip_pkt_type {
166 	IRDMA_PKT_TYPE_UNKNOWN,
167 	IRDMA_PKT_TYPE_SYN,
168 	IRDMA_PKT_TYPE_SYNACK,
169 	IRDMA_PKT_TYPE_ACK,
170 	IRDMA_PKT_TYPE_FIN,
171 	IRDMA_PKT_TYPE_RST,
172 };
173 
174 enum irdma_cm_listener_state {
175 	IRDMA_CM_LISTENER_PASSIVE_STATE = 1,
176 	IRDMA_CM_LISTENER_ACTIVE_STATE  = 2,
177 	IRDMA_CM_LISTENER_EITHER_STATE  = 3,
178 };
179 
180 /* CM event codes */
181 enum irdma_cm_event_type {
182 	IRDMA_CM_EVENT_UNKNOWN,
183 	IRDMA_CM_EVENT_ESTABLISHED,
184 	IRDMA_CM_EVENT_MPA_REQ,
185 	IRDMA_CM_EVENT_MPA_CONNECT,
186 	IRDMA_CM_EVENT_MPA_ACCEPT,
187 	IRDMA_CM_EVENT_MPA_REJECT,
188 	IRDMA_CM_EVENT_MPA_ESTABLISHED,
189 	IRDMA_CM_EVENT_CONNECTED,
190 	IRDMA_CM_EVENT_RESET,
191 	IRDMA_CM_EVENT_ABORTED,
192 };
193 
194 struct ietf_mpa_v1 {
195 	u8 key[IETF_MPA_KEY_SIZE];
196 	u8 flags;
197 	u8 rev;
198 	__be16 priv_data_len;
199 	u8 priv_data[];
200 };
201 
202 struct ietf_rtr_msg {
203 	__be16 ctrl_ird;
204 	__be16 ctrl_ord;
205 };
206 
207 struct ietf_mpa_v2 {
208 	u8 key[IETF_MPA_KEY_SIZE];
209 	u8 flags;
210 	u8 rev;
211 	__be16 priv_data_len;
212 	struct ietf_rtr_msg rtr_msg;
213 	u8 priv_data[];
214 };
215 
216 struct option_base {
217 	u8 optionnum;
218 	u8 len;
219 };
220 
221 struct option_mss {
222 	u8 optionnum;
223 	u8 len;
224 	__be16 mss;
225 };
226 
227 struct option_windowscale {
228 	u8 optionnum;
229 	u8 len;
230 	u8 shiftcount;
231 };
232 
233 union all_known_options {
234 	char eol;
235 	struct option_base base;
236 	struct option_mss mss;
237 	struct option_windowscale windowscale;
238 };
239 
240 struct irdma_timer_entry {
241 	struct list_head list;
242 	unsigned long timetosend; /* jiffies */
243 	struct irdma_puda_buf *sqbuf;
244 	u32 type;
245 	u32 retrycount;
246 	u32 retranscount;
247 	u32 context;
248 	u32 send_retrans;
249 	int close_when_complete;
250 };
251 
252 /* CM context params */
253 struct irdma_cm_tcp_context {
254 	u8 client;
255 	u32 loc_seq_num;
256 	u32 loc_ack_num;
257 	u32 rem_ack_num;
258 	u32 rcv_nxt;
259 	u32 loc_id;
260 	u32 rem_id;
261 	u32 snd_wnd;
262 	u32 max_snd_wnd;
263 	u32 rcv_wnd;
264 	u32 mss;
265 	u8 snd_wscale;
266 	u8 rcv_wscale;
267 };
268 
269 struct irdma_apbvt_entry {
270 	struct hlist_node hlist;
271 	u32 use_cnt;
272 	u16 port;
273 };
274 
275 struct irdma_cm_listener {
276 	struct list_head list;
277 	struct iw_cm_id *cm_id;
278 	struct irdma_cm_core *cm_core;
279 	struct irdma_device *iwdev;
280 	struct list_head child_listen_list;
281 	struct irdma_apbvt_entry *apbvt_entry;
282 	enum irdma_cm_listener_state listener_state;
283 	atomic_t refcnt;
284 	atomic_t pend_accepts_cnt;
285 	u32 loc_addr[4];
286 	u32 reused_node;
287 	int backlog;
288 	u16 loc_port;
289 	u16 vlan_id;
290 	u8 loc_mac[ETHER_ADDR_LEN];
291 	u8 user_pri;
292 	u8 tos;
293 	bool qhash_set:1;
294 	bool ipv4:1;
295 };
296 
297 struct irdma_kmem_info {
298 	void *addr;
299 	u32 size;
300 };
301 
302 struct irdma_mpa_priv_info {
303 	const void *addr;
304 	u32 size;
305 };
306 
307 struct irdma_cm_node {
308 	struct irdma_qp *iwqp;
309 	struct irdma_device *iwdev;
310 	struct irdma_sc_dev *dev;
311 	struct irdma_cm_tcp_context tcp_cntxt;
312 	struct irdma_cm_core *cm_core;
313 	struct irdma_timer_entry *send_entry;
314 	struct irdma_timer_entry *close_entry;
315 	struct irdma_cm_listener *listener;
316 	struct list_head timer_entry;
317 	struct list_head reset_entry;
318 	struct list_head teardown_entry;
319 	struct irdma_apbvt_entry *apbvt_entry;
320 	struct rcu_head rcu_head;
321 	struct irdma_mpa_priv_info pdata;
322 	struct irdma_sc_ah *ah;
323 	struct irdma_kmem_info mpa_hdr;
324 	struct iw_cm_id *cm_id;
325 	struct hlist_node list;
326 	struct completion establish_comp;
327 	spinlock_t retrans_list_lock; /* protect CM node rexmit updates*/
328 	atomic_t passive_state;
329 	atomic_t refcnt;
330 	enum irdma_cm_node_state state;
331 	enum send_rdma0 send_rdma0_op;
332 	enum mpa_frame_ver mpa_frame_rev;
333 	u32 loc_addr[4], rem_addr[4];
334 	u16 loc_port, rem_port;
335 	int apbvt_set;
336 	int accept_pend;
337 	u16 vlan_id;
338 	u16 ird_size;
339 	u16 ord_size;
340 	u16 mpav2_ird_ord;
341 	u16 lsmm_size;
342 	u8 pdata_buf[IETF_MAX_PRIV_DATA_LEN];
343 	u8 loc_mac[ETHER_ADDR_LEN];
344 	u8 rem_mac[ETHER_ADDR_LEN];
345 	u8 user_pri;
346 	u8 tos;
347 	bool ack_rcvd:1;
348 	bool qhash_set:1;
349 	bool ipv4:1;
350 	bool snd_mark_en:1;
351 	bool rcv_mark_en:1;
352 	bool do_lpb:1;
353 	bool accelerated:1;
354 	struct ietf_mpa_v2 mpa_v2_frame;
355 };
356 
357 /* Used by internal CM APIs to pass CM information*/
358 struct irdma_cm_info {
359 	struct iw_cm_id *cm_id;
360 	struct irdma_cqp_request *cqp_request;
361 	u16 loc_port;
362 	u16 rem_port;
363 	u32 loc_addr[4];
364 	u32 rem_addr[4];
365 	u32 qh_qpid;
366 	u16 vlan_id;
367 	int backlog;
368 	u8 user_pri;
369 	u8 tos;
370 	bool ipv4;
371 };
372 
373 struct irdma_cm_event {
374 	enum irdma_cm_event_type type;
375 	struct irdma_cm_info cm_info;
376 	struct work_struct event_work;
377 	struct irdma_cm_node *cm_node;
378 };
379 
380 struct irdma_cm_core {
381 	struct irdma_device *iwdev;
382 	struct irdma_sc_dev *dev;
383 	struct list_head listen_list;
384 	DECLARE_HASHTABLE(cm_hash_tbl, 8);
385 	DECLARE_HASHTABLE(apbvt_hash_tbl, 8);
386 	struct timer_list tcp_timer;
387 	struct workqueue_struct *event_wq;
388 	spinlock_t ht_lock; /* protect CM node (active side) list */
389 	spinlock_t listen_list_lock; /* protect listener list */
390 	spinlock_t apbvt_lock; /*serialize apbvt add/del entries*/
391 	u64 stats_nodes_created;
392 	u64 stats_nodes_destroyed;
393 	u64 stats_listen_created;
394 	u64 stats_listen_destroyed;
395 	u64 stats_listen_nodes_created;
396 	u64 stats_listen_nodes_destroyed;
397 	u64 stats_lpbs;
398 	u64 stats_accepts;
399 	u64 stats_rejects;
400 	u64 stats_connect_errs;
401 	u64 stats_passive_errs;
402 	u64 stats_pkt_retrans;
403 	u64 stats_backlog_drops;
404 	struct irdma_puda_buf *(*form_cm_frame)(struct irdma_cm_node *cm_node,
405 						struct irdma_kmem_info *options,
406 						struct irdma_kmem_info *hdr,
407 						struct irdma_mpa_priv_info *pdata,
408 						u8 flags);
409 	int (*cm_create_ah)(struct irdma_cm_node *cm_node, bool wait);
410 	void (*cm_free_ah)(struct irdma_cm_node *cm_node);
411 };
412 
413 struct irdma_add_mqh_cbs {
414 	struct irdma_device *iwdev;
415 	struct irdma_cm_info *cm_info;
416 	struct irdma_cm_listener *cm_listen_node;
417 };
418 
419 int irdma_schedule_cm_timer(struct irdma_cm_node *cm_node,
420 			    struct irdma_puda_buf *sqbuf,
421 			    enum irdma_timer_type type, int send_retrans,
422 			    int close_when_complete);
423 
irdma_tos2dscp(u8 tos)424 static inline u8 irdma_tos2dscp(u8 tos)
425 {
426 #define IRDMA_DSCP_S 2
427 #define IRDMA_DSCP GENMASK(7, 2)
428 	return FIELD_GET(IRDMA_DSCP, tos);
429 }
430 
431 int irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param);
432 int irdma_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len);
433 int irdma_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param);
434 int irdma_create_listen(struct iw_cm_id *cm_id, int backlog);
435 int irdma_destroy_listen(struct iw_cm_id *cm_id);
436 int irdma_add_arp(struct irdma_pci_f *rf, u32 *ip, const u8 *mac);
437 int irdma_cm_start(struct irdma_device *dev);
438 int irdma_cm_stop(struct irdma_device *dev);
439 bool irdma_ipv4_is_lpb(u32 loc_addr, u32 rem_addr);
440 bool irdma_ipv6_is_lpb(u32 *loc_addr, u32 *rem_addr);
441 int irdma_arp_table(struct irdma_pci_f *rf, u32 *ip_addr,
442 		    const u8 *mac_addr, u32 action);
443 bool irdma_port_in_use(struct irdma_cm_core *cm_core, u16 port);
444 void irdma_send_ack(struct irdma_cm_node *cm_node);
445 void irdma_lpb_nop(struct irdma_sc_qp *qp);
446 void irdma_rem_ref_cm_node(struct irdma_cm_node *cm_node);
447 void irdma_add_conn_est_qh(struct irdma_cm_node *cm_node);
448 #endif /* IRDMA_CM_H */
449