xref: /linux/net/smc/smc_core.c (revision c4101e55974cc7d835fbd2d8e01553a3f61e9e75)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  *  Basic Transport Functions exploiting Infiniband API
6  *
7  *  Copyright IBM Corp. 2016
8  *
9  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
10  */
11 
12 #include <linux/socket.h>
13 #include <linux/if_vlan.h>
14 #include <linux/random.h>
15 #include <linux/workqueue.h>
16 #include <linux/wait.h>
17 #include <linux/reboot.h>
18 #include <linux/mutex.h>
19 #include <linux/list.h>
20 #include <linux/smc.h>
21 #include <net/tcp.h>
22 #include <net/sock.h>
23 #include <rdma/ib_verbs.h>
24 #include <rdma/ib_cache.h>
25 
26 #include "smc.h"
27 #include "smc_clc.h"
28 #include "smc_core.h"
29 #include "smc_ib.h"
30 #include "smc_wr.h"
31 #include "smc_llc.h"
32 #include "smc_cdc.h"
33 #include "smc_close.h"
34 #include "smc_ism.h"
35 #include "smc_netlink.h"
36 #include "smc_stats.h"
37 #include "smc_tracepoint.h"
38 
39 #define SMC_LGR_NUM_INCR		256
40 #define SMC_LGR_FREE_DELAY_SERV		(600 * HZ)
41 #define SMC_LGR_FREE_DELAY_CLNT		(SMC_LGR_FREE_DELAY_SERV + 10 * HZ)
42 
43 struct smc_lgr_list smc_lgr_list = {	/* established link groups */
44 	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
45 	.list = LIST_HEAD_INIT(smc_lgr_list.list),
46 	.num = 0,
47 };
48 
49 static atomic_t lgr_cnt = ATOMIC_INIT(0); /* number of existing link groups */
50 static DECLARE_WAIT_QUEUE_HEAD(lgrs_deleted);
51 
52 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
53 			 struct smc_buf_desc *buf_desc);
54 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft);
55 
56 static void smc_link_down_work(struct work_struct *work);
57 
58 /* return head of link group list and its lock for a given link group */
59 static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr,
60 						  spinlock_t **lgr_lock)
61 {
62 	if (lgr->is_smcd) {
63 		*lgr_lock = &lgr->smcd->lgr_lock;
64 		return &lgr->smcd->lgr_list;
65 	}
66 
67 	*lgr_lock = &smc_lgr_list.lock;
68 	return &smc_lgr_list.list;
69 }
70 
71 static void smc_ibdev_cnt_inc(struct smc_link *lnk)
72 {
73 	atomic_inc(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
74 }
75 
76 static void smc_ibdev_cnt_dec(struct smc_link *lnk)
77 {
78 	atomic_dec(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
79 }
80 
81 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr)
82 {
83 	/* client link group creation always follows the server link group
84 	 * creation. For client use a somewhat higher removal delay time,
85 	 * otherwise there is a risk of out-of-sync link groups.
86 	 */
87 	if (!lgr->freeing) {
88 		mod_delayed_work(system_wq, &lgr->free_work,
89 				 (!lgr->is_smcd && lgr->role == SMC_CLNT) ?
90 						SMC_LGR_FREE_DELAY_CLNT :
91 						SMC_LGR_FREE_DELAY_SERV);
92 	}
93 }
94 
95 /* Register connection's alert token in our lookup structure.
96  * To use rbtrees we have to implement our own insert core.
97  * Requires @conns_lock
98  * @smc		connection to register
99  * Returns 0 on success, != otherwise.
100  */
101 static void smc_lgr_add_alert_token(struct smc_connection *conn)
102 {
103 	struct rb_node **link, *parent = NULL;
104 	u32 token = conn->alert_token_local;
105 
106 	link = &conn->lgr->conns_all.rb_node;
107 	while (*link) {
108 		struct smc_connection *cur = rb_entry(*link,
109 					struct smc_connection, alert_node);
110 
111 		parent = *link;
112 		if (cur->alert_token_local > token)
113 			link = &parent->rb_left;
114 		else
115 			link = &parent->rb_right;
116 	}
117 	/* Put the new node there */
118 	rb_link_node(&conn->alert_node, parent, link);
119 	rb_insert_color(&conn->alert_node, &conn->lgr->conns_all);
120 }
121 
122 /* assign an SMC-R link to the connection */
123 static int smcr_lgr_conn_assign_link(struct smc_connection *conn, bool first)
124 {
125 	enum smc_link_state expected = first ? SMC_LNK_ACTIVATING :
126 				       SMC_LNK_ACTIVE;
127 	int i, j;
128 
129 	/* do link balancing */
130 	conn->lnk = NULL;	/* reset conn->lnk first */
131 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
132 		struct smc_link *lnk = &conn->lgr->lnk[i];
133 
134 		if (lnk->state != expected || lnk->link_is_asym)
135 			continue;
136 		if (conn->lgr->role == SMC_CLNT) {
137 			conn->lnk = lnk; /* temporary, SMC server assigns link*/
138 			break;
139 		}
140 		if (conn->lgr->conns_num % 2) {
141 			for (j = i + 1; j < SMC_LINKS_PER_LGR_MAX; j++) {
142 				struct smc_link *lnk2;
143 
144 				lnk2 = &conn->lgr->lnk[j];
145 				if (lnk2->state == expected &&
146 				    !lnk2->link_is_asym) {
147 					conn->lnk = lnk2;
148 					break;
149 				}
150 			}
151 		}
152 		if (!conn->lnk)
153 			conn->lnk = lnk;
154 		break;
155 	}
156 	if (!conn->lnk)
157 		return SMC_CLC_DECL_NOACTLINK;
158 	atomic_inc(&conn->lnk->conn_cnt);
159 	return 0;
160 }
161 
162 /* Register connection in link group by assigning an alert token
163  * registered in a search tree.
164  * Requires @conns_lock
165  * Note that '0' is a reserved value and not assigned.
166  */
167 static int smc_lgr_register_conn(struct smc_connection *conn, bool first)
168 {
169 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
170 	static atomic_t nexttoken = ATOMIC_INIT(0);
171 	int rc;
172 
173 	if (!conn->lgr->is_smcd) {
174 		rc = smcr_lgr_conn_assign_link(conn, first);
175 		if (rc) {
176 			conn->lgr = NULL;
177 			return rc;
178 		}
179 	}
180 	/* find a new alert_token_local value not yet used by some connection
181 	 * in this link group
182 	 */
183 	sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */
184 	while (!conn->alert_token_local) {
185 		conn->alert_token_local = atomic_inc_return(&nexttoken);
186 		if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr))
187 			conn->alert_token_local = 0;
188 	}
189 	smc_lgr_add_alert_token(conn);
190 	conn->lgr->conns_num++;
191 	return 0;
192 }
193 
194 /* Unregister connection and reset the alert token of the given connection<
195  */
196 static void __smc_lgr_unregister_conn(struct smc_connection *conn)
197 {
198 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
199 	struct smc_link_group *lgr = conn->lgr;
200 
201 	rb_erase(&conn->alert_node, &lgr->conns_all);
202 	if (conn->lnk)
203 		atomic_dec(&conn->lnk->conn_cnt);
204 	lgr->conns_num--;
205 	conn->alert_token_local = 0;
206 	sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */
207 }
208 
209 /* Unregister connection from lgr
210  */
211 static void smc_lgr_unregister_conn(struct smc_connection *conn)
212 {
213 	struct smc_link_group *lgr = conn->lgr;
214 
215 	if (!smc_conn_lgr_valid(conn))
216 		return;
217 	write_lock_bh(&lgr->conns_lock);
218 	if (conn->alert_token_local) {
219 		__smc_lgr_unregister_conn(conn);
220 	}
221 	write_unlock_bh(&lgr->conns_lock);
222 }
223 
224 int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb)
225 {
226 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
227 	char hostname[SMC_MAX_HOSTNAME_LEN + 1];
228 	char smc_seid[SMC_MAX_EID_LEN + 1];
229 	struct nlattr *attrs;
230 	u8 *seid = NULL;
231 	u8 *host = NULL;
232 	void *nlh;
233 
234 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
235 			  &smc_gen_nl_family, NLM_F_MULTI,
236 			  SMC_NETLINK_GET_SYS_INFO);
237 	if (!nlh)
238 		goto errmsg;
239 	if (cb_ctx->pos[0])
240 		goto errout;
241 	attrs = nla_nest_start(skb, SMC_GEN_SYS_INFO);
242 	if (!attrs)
243 		goto errout;
244 	if (nla_put_u8(skb, SMC_NLA_SYS_VER, SMC_V2))
245 		goto errattr;
246 	if (nla_put_u8(skb, SMC_NLA_SYS_REL, SMC_RELEASE))
247 		goto errattr;
248 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_ISM_V2, smc_ism_is_v2_capable()))
249 		goto errattr;
250 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_SMCR_V2, true))
251 		goto errattr;
252 	smc_clc_get_hostname(&host);
253 	if (host) {
254 		memcpy(hostname, host, SMC_MAX_HOSTNAME_LEN);
255 		hostname[SMC_MAX_HOSTNAME_LEN] = 0;
256 		if (nla_put_string(skb, SMC_NLA_SYS_LOCAL_HOST, hostname))
257 			goto errattr;
258 	}
259 	if (smc_ism_is_v2_capable()) {
260 		smc_ism_get_system_eid(&seid);
261 		memcpy(smc_seid, seid, SMC_MAX_EID_LEN);
262 		smc_seid[SMC_MAX_EID_LEN] = 0;
263 		if (nla_put_string(skb, SMC_NLA_SYS_SEID, smc_seid))
264 			goto errattr;
265 	}
266 	nla_nest_end(skb, attrs);
267 	genlmsg_end(skb, nlh);
268 	cb_ctx->pos[0] = 1;
269 	return skb->len;
270 
271 errattr:
272 	nla_nest_cancel(skb, attrs);
273 errout:
274 	genlmsg_cancel(skb, nlh);
275 errmsg:
276 	return skb->len;
277 }
278 
279 /* Fill SMC_NLA_LGR_D_V2_COMMON/SMC_NLA_LGR_R_V2_COMMON nested attributes */
280 static int smc_nl_fill_lgr_v2_common(struct smc_link_group *lgr,
281 				     struct sk_buff *skb,
282 				     struct netlink_callback *cb,
283 				     struct nlattr *v2_attrs)
284 {
285 	char smc_host[SMC_MAX_HOSTNAME_LEN + 1];
286 	char smc_eid[SMC_MAX_EID_LEN + 1];
287 
288 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_VER, lgr->smc_version))
289 		goto errv2attr;
290 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_REL, lgr->peer_smc_release))
291 		goto errv2attr;
292 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_OS, lgr->peer_os))
293 		goto errv2attr;
294 	memcpy(smc_host, lgr->peer_hostname, SMC_MAX_HOSTNAME_LEN);
295 	smc_host[SMC_MAX_HOSTNAME_LEN] = 0;
296 	if (nla_put_string(skb, SMC_NLA_LGR_V2_PEER_HOST, smc_host))
297 		goto errv2attr;
298 	memcpy(smc_eid, lgr->negotiated_eid, SMC_MAX_EID_LEN);
299 	smc_eid[SMC_MAX_EID_LEN] = 0;
300 	if (nla_put_string(skb, SMC_NLA_LGR_V2_NEG_EID, smc_eid))
301 		goto errv2attr;
302 
303 	nla_nest_end(skb, v2_attrs);
304 	return 0;
305 
306 errv2attr:
307 	nla_nest_cancel(skb, v2_attrs);
308 	return -EMSGSIZE;
309 }
310 
311 static int smc_nl_fill_smcr_lgr_v2(struct smc_link_group *lgr,
312 				   struct sk_buff *skb,
313 				   struct netlink_callback *cb)
314 {
315 	struct nlattr *v2_attrs;
316 
317 	v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2);
318 	if (!v2_attrs)
319 		goto errattr;
320 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_DIRECT, !lgr->uses_gateway))
321 		goto errv2attr;
322 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_MAX_CONNS, lgr->max_conns))
323 		goto errv2attr;
324 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_MAX_LINKS, lgr->max_links))
325 		goto errv2attr;
326 
327 	nla_nest_end(skb, v2_attrs);
328 	return 0;
329 
330 errv2attr:
331 	nla_nest_cancel(skb, v2_attrs);
332 errattr:
333 	return -EMSGSIZE;
334 }
335 
336 static int smc_nl_fill_lgr(struct smc_link_group *lgr,
337 			   struct sk_buff *skb,
338 			   struct netlink_callback *cb)
339 {
340 	char smc_target[SMC_MAX_PNETID_LEN + 1];
341 	struct nlattr *attrs, *v2_attrs;
342 
343 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCR);
344 	if (!attrs)
345 		goto errout;
346 
347 	if (nla_put_u32(skb, SMC_NLA_LGR_R_ID, *((u32 *)&lgr->id)))
348 		goto errattr;
349 	if (nla_put_u32(skb, SMC_NLA_LGR_R_CONNS_NUM, lgr->conns_num))
350 		goto errattr;
351 	if (nla_put_u8(skb, SMC_NLA_LGR_R_ROLE, lgr->role))
352 		goto errattr;
353 	if (nla_put_u8(skb, SMC_NLA_LGR_R_TYPE, lgr->type))
354 		goto errattr;
355 	if (nla_put_u8(skb, SMC_NLA_LGR_R_BUF_TYPE, lgr->buf_type))
356 		goto errattr;
357 	if (nla_put_u8(skb, SMC_NLA_LGR_R_VLAN_ID, lgr->vlan_id))
358 		goto errattr;
359 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_R_NET_COOKIE,
360 			      lgr->net->net_cookie, SMC_NLA_LGR_R_PAD))
361 		goto errattr;
362 	memcpy(smc_target, lgr->pnet_id, SMC_MAX_PNETID_LEN);
363 	smc_target[SMC_MAX_PNETID_LEN] = 0;
364 	if (nla_put_string(skb, SMC_NLA_LGR_R_PNETID, smc_target))
365 		goto errattr;
366 	if (lgr->smc_version > SMC_V1) {
367 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2_COMMON);
368 		if (!v2_attrs)
369 			goto errattr;
370 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
371 			goto errattr;
372 		if (smc_nl_fill_smcr_lgr_v2(lgr, skb, cb))
373 			goto errattr;
374 	}
375 
376 	nla_nest_end(skb, attrs);
377 	return 0;
378 errattr:
379 	nla_nest_cancel(skb, attrs);
380 errout:
381 	return -EMSGSIZE;
382 }
383 
384 static int smc_nl_fill_lgr_link(struct smc_link_group *lgr,
385 				struct smc_link *link,
386 				struct sk_buff *skb,
387 				struct netlink_callback *cb)
388 {
389 	char smc_ibname[IB_DEVICE_NAME_MAX];
390 	u8 smc_gid_target[41];
391 	struct nlattr *attrs;
392 	u32 link_uid = 0;
393 	void *nlh;
394 
395 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
396 			  &smc_gen_nl_family, NLM_F_MULTI,
397 			  SMC_NETLINK_GET_LINK_SMCR);
398 	if (!nlh)
399 		goto errmsg;
400 
401 	attrs = nla_nest_start(skb, SMC_GEN_LINK_SMCR);
402 	if (!attrs)
403 		goto errout;
404 
405 	if (nla_put_u8(skb, SMC_NLA_LINK_ID, link->link_id))
406 		goto errattr;
407 	if (nla_put_u32(skb, SMC_NLA_LINK_STATE, link->state))
408 		goto errattr;
409 	if (nla_put_u32(skb, SMC_NLA_LINK_CONN_CNT,
410 			atomic_read(&link->conn_cnt)))
411 		goto errattr;
412 	if (nla_put_u8(skb, SMC_NLA_LINK_IB_PORT, link->ibport))
413 		goto errattr;
414 	if (nla_put_u32(skb, SMC_NLA_LINK_NET_DEV, link->ndev_ifidx))
415 		goto errattr;
416 	snprintf(smc_ibname, sizeof(smc_ibname), "%s", link->ibname);
417 	if (nla_put_string(skb, SMC_NLA_LINK_IB_DEV, smc_ibname))
418 		goto errattr;
419 	memcpy(&link_uid, link->link_uid, sizeof(link_uid));
420 	if (nla_put_u32(skb, SMC_NLA_LINK_UID, link_uid))
421 		goto errattr;
422 	memcpy(&link_uid, link->peer_link_uid, sizeof(link_uid));
423 	if (nla_put_u32(skb, SMC_NLA_LINK_PEER_UID, link_uid))
424 		goto errattr;
425 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
426 	smc_gid_be16_convert(smc_gid_target, link->gid);
427 	if (nla_put_string(skb, SMC_NLA_LINK_GID, smc_gid_target))
428 		goto errattr;
429 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
430 	smc_gid_be16_convert(smc_gid_target, link->peer_gid);
431 	if (nla_put_string(skb, SMC_NLA_LINK_PEER_GID, smc_gid_target))
432 		goto errattr;
433 
434 	nla_nest_end(skb, attrs);
435 	genlmsg_end(skb, nlh);
436 	return 0;
437 errattr:
438 	nla_nest_cancel(skb, attrs);
439 errout:
440 	genlmsg_cancel(skb, nlh);
441 errmsg:
442 	return -EMSGSIZE;
443 }
444 
445 static int smc_nl_handle_lgr(struct smc_link_group *lgr,
446 			     struct sk_buff *skb,
447 			     struct netlink_callback *cb,
448 			     bool list_links)
449 {
450 	void *nlh;
451 	int i;
452 
453 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
454 			  &smc_gen_nl_family, NLM_F_MULTI,
455 			  SMC_NETLINK_GET_LGR_SMCR);
456 	if (!nlh)
457 		goto errmsg;
458 	if (smc_nl_fill_lgr(lgr, skb, cb))
459 		goto errout;
460 
461 	genlmsg_end(skb, nlh);
462 	if (!list_links)
463 		goto out;
464 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
465 		if (!smc_link_usable(&lgr->lnk[i]))
466 			continue;
467 		if (smc_nl_fill_lgr_link(lgr, &lgr->lnk[i], skb, cb))
468 			goto errout;
469 	}
470 out:
471 	return 0;
472 
473 errout:
474 	genlmsg_cancel(skb, nlh);
475 errmsg:
476 	return -EMSGSIZE;
477 }
478 
479 static void smc_nl_fill_lgr_list(struct smc_lgr_list *smc_lgr,
480 				 struct sk_buff *skb,
481 				 struct netlink_callback *cb,
482 				 bool list_links)
483 {
484 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
485 	struct smc_link_group *lgr;
486 	int snum = cb_ctx->pos[0];
487 	int num = 0;
488 
489 	spin_lock_bh(&smc_lgr->lock);
490 	list_for_each_entry(lgr, &smc_lgr->list, list) {
491 		if (num < snum)
492 			goto next;
493 		if (smc_nl_handle_lgr(lgr, skb, cb, list_links))
494 			goto errout;
495 next:
496 		num++;
497 	}
498 errout:
499 	spin_unlock_bh(&smc_lgr->lock);
500 	cb_ctx->pos[0] = num;
501 }
502 
503 static int smc_nl_fill_smcd_lgr(struct smc_link_group *lgr,
504 				struct sk_buff *skb,
505 				struct netlink_callback *cb)
506 {
507 	char smc_pnet[SMC_MAX_PNETID_LEN + 1];
508 	struct smcd_dev *smcd = lgr->smcd;
509 	struct smcd_gid smcd_gid;
510 	struct nlattr *attrs;
511 	void *nlh;
512 
513 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
514 			  &smc_gen_nl_family, NLM_F_MULTI,
515 			  SMC_NETLINK_GET_LGR_SMCD);
516 	if (!nlh)
517 		goto errmsg;
518 
519 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCD);
520 	if (!attrs)
521 		goto errout;
522 
523 	if (nla_put_u32(skb, SMC_NLA_LGR_D_ID, *((u32 *)&lgr->id)))
524 		goto errattr;
525 	smcd->ops->get_local_gid(smcd, &smcd_gid);
526 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_GID,
527 			      smcd_gid.gid, SMC_NLA_LGR_D_PAD))
528 		goto errattr;
529 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_EXT_GID,
530 			      smcd_gid.gid_ext, SMC_NLA_LGR_D_PAD))
531 		goto errattr;
532 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_GID, lgr->peer_gid.gid,
533 			      SMC_NLA_LGR_D_PAD))
534 		goto errattr;
535 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_EXT_GID,
536 			      lgr->peer_gid.gid_ext, SMC_NLA_LGR_D_PAD))
537 		goto errattr;
538 	if (nla_put_u8(skb, SMC_NLA_LGR_D_VLAN_ID, lgr->vlan_id))
539 		goto errattr;
540 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CONNS_NUM, lgr->conns_num))
541 		goto errattr;
542 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CHID, smc_ism_get_chid(lgr->smcd)))
543 		goto errattr;
544 	memcpy(smc_pnet, lgr->smcd->pnetid, SMC_MAX_PNETID_LEN);
545 	smc_pnet[SMC_MAX_PNETID_LEN] = 0;
546 	if (nla_put_string(skb, SMC_NLA_LGR_D_PNETID, smc_pnet))
547 		goto errattr;
548 	if (lgr->smc_version > SMC_V1) {
549 		struct nlattr *v2_attrs;
550 
551 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_D_V2_COMMON);
552 		if (!v2_attrs)
553 			goto errattr;
554 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
555 			goto errattr;
556 	}
557 	nla_nest_end(skb, attrs);
558 	genlmsg_end(skb, nlh);
559 	return 0;
560 
561 errattr:
562 	nla_nest_cancel(skb, attrs);
563 errout:
564 	genlmsg_cancel(skb, nlh);
565 errmsg:
566 	return -EMSGSIZE;
567 }
568 
569 static int smc_nl_handle_smcd_lgr(struct smcd_dev *dev,
570 				  struct sk_buff *skb,
571 				  struct netlink_callback *cb)
572 {
573 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
574 	struct smc_link_group *lgr;
575 	int snum = cb_ctx->pos[1];
576 	int rc = 0, num = 0;
577 
578 	spin_lock_bh(&dev->lgr_lock);
579 	list_for_each_entry(lgr, &dev->lgr_list, list) {
580 		if (!lgr->is_smcd)
581 			continue;
582 		if (num < snum)
583 			goto next;
584 		rc = smc_nl_fill_smcd_lgr(lgr, skb, cb);
585 		if (rc)
586 			goto errout;
587 next:
588 		num++;
589 	}
590 errout:
591 	spin_unlock_bh(&dev->lgr_lock);
592 	cb_ctx->pos[1] = num;
593 	return rc;
594 }
595 
596 static int smc_nl_fill_smcd_dev(struct smcd_dev_list *dev_list,
597 				struct sk_buff *skb,
598 				struct netlink_callback *cb)
599 {
600 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
601 	struct smcd_dev *smcd_dev;
602 	int snum = cb_ctx->pos[0];
603 	int rc = 0, num = 0;
604 
605 	mutex_lock(&dev_list->mutex);
606 	list_for_each_entry(smcd_dev, &dev_list->list, list) {
607 		if (list_empty(&smcd_dev->lgr_list))
608 			continue;
609 		if (num < snum)
610 			goto next;
611 		rc = smc_nl_handle_smcd_lgr(smcd_dev, skb, cb);
612 		if (rc)
613 			goto errout;
614 next:
615 		num++;
616 	}
617 errout:
618 	mutex_unlock(&dev_list->mutex);
619 	cb_ctx->pos[0] = num;
620 	return rc;
621 }
622 
623 int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
624 {
625 	bool list_links = false;
626 
627 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
628 	return skb->len;
629 }
630 
631 int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb)
632 {
633 	bool list_links = true;
634 
635 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
636 	return skb->len;
637 }
638 
639 int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
640 {
641 	smc_nl_fill_smcd_dev(&smcd_dev_list, skb, cb);
642 	return skb->len;
643 }
644 
645 void smc_lgr_cleanup_early(struct smc_link_group *lgr)
646 {
647 	spinlock_t *lgr_lock;
648 
649 	if (!lgr)
650 		return;
651 
652 	smc_lgr_list_head(lgr, &lgr_lock);
653 	spin_lock_bh(lgr_lock);
654 	/* do not use this link group for new connections */
655 	if (!list_empty(&lgr->list))
656 		list_del_init(&lgr->list);
657 	spin_unlock_bh(lgr_lock);
658 	__smc_lgr_terminate(lgr, true);
659 }
660 
661 static void smcr_lgr_link_deactivate_all(struct smc_link_group *lgr)
662 {
663 	int i;
664 
665 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
666 		struct smc_link *lnk = &lgr->lnk[i];
667 
668 		if (smc_link_sendable(lnk))
669 			lnk->state = SMC_LNK_INACTIVE;
670 	}
671 	wake_up_all(&lgr->llc_msg_waiter);
672 	wake_up_all(&lgr->llc_flow_waiter);
673 }
674 
675 static void smc_lgr_free(struct smc_link_group *lgr);
676 
677 static void smc_lgr_free_work(struct work_struct *work)
678 {
679 	struct smc_link_group *lgr = container_of(to_delayed_work(work),
680 						  struct smc_link_group,
681 						  free_work);
682 	spinlock_t *lgr_lock;
683 	bool conns;
684 
685 	smc_lgr_list_head(lgr, &lgr_lock);
686 	spin_lock_bh(lgr_lock);
687 	if (lgr->freeing) {
688 		spin_unlock_bh(lgr_lock);
689 		return;
690 	}
691 	read_lock_bh(&lgr->conns_lock);
692 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
693 	read_unlock_bh(&lgr->conns_lock);
694 	if (!conns) { /* number of lgr connections is no longer zero */
695 		spin_unlock_bh(lgr_lock);
696 		return;
697 	}
698 	list_del_init(&lgr->list); /* remove from smc_lgr_list */
699 	lgr->freeing = 1; /* this instance does the freeing, no new schedule */
700 	spin_unlock_bh(lgr_lock);
701 	cancel_delayed_work(&lgr->free_work);
702 
703 	if (!lgr->is_smcd && !lgr->terminating)
704 		smc_llc_send_link_delete_all(lgr, true,
705 					     SMC_LLC_DEL_PROG_INIT_TERM);
706 	if (lgr->is_smcd && !lgr->terminating)
707 		smc_ism_signal_shutdown(lgr);
708 	if (!lgr->is_smcd)
709 		smcr_lgr_link_deactivate_all(lgr);
710 	smc_lgr_free(lgr);
711 }
712 
713 static void smc_lgr_terminate_work(struct work_struct *work)
714 {
715 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
716 						  terminate_work);
717 
718 	__smc_lgr_terminate(lgr, true);
719 }
720 
721 /* return next unique link id for the lgr */
722 static u8 smcr_next_link_id(struct smc_link_group *lgr)
723 {
724 	u8 link_id;
725 	int i;
726 
727 	while (1) {
728 again:
729 		link_id = ++lgr->next_link_id;
730 		if (!link_id)	/* skip zero as link_id */
731 			link_id = ++lgr->next_link_id;
732 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
733 			if (smc_link_usable(&lgr->lnk[i]) &&
734 			    lgr->lnk[i].link_id == link_id)
735 				goto again;
736 		}
737 		break;
738 	}
739 	return link_id;
740 }
741 
742 static void smcr_copy_dev_info_to_link(struct smc_link *link)
743 {
744 	struct smc_ib_device *smcibdev = link->smcibdev;
745 
746 	snprintf(link->ibname, sizeof(link->ibname), "%s",
747 		 smcibdev->ibdev->name);
748 	link->ndev_ifidx = smcibdev->ndev_ifidx[link->ibport - 1];
749 }
750 
751 int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk,
752 		   u8 link_idx, struct smc_init_info *ini)
753 {
754 	struct smc_ib_device *smcibdev;
755 	u8 rndvec[3];
756 	int rc;
757 
758 	if (lgr->smc_version == SMC_V2) {
759 		lnk->smcibdev = ini->smcrv2.ib_dev_v2;
760 		lnk->ibport = ini->smcrv2.ib_port_v2;
761 	} else {
762 		lnk->smcibdev = ini->ib_dev;
763 		lnk->ibport = ini->ib_port;
764 	}
765 	get_device(&lnk->smcibdev->ibdev->dev);
766 	atomic_inc(&lnk->smcibdev->lnk_cnt);
767 	refcount_set(&lnk->refcnt, 1); /* link refcnt is set to 1 */
768 	lnk->clearing = 0;
769 	lnk->path_mtu = lnk->smcibdev->pattr[lnk->ibport - 1].active_mtu;
770 	lnk->link_id = smcr_next_link_id(lgr);
771 	lnk->lgr = lgr;
772 	smc_lgr_hold(lgr); /* lgr_put in smcr_link_clear() */
773 	lnk->link_idx = link_idx;
774 	lnk->wr_rx_id_compl = 0;
775 	smc_ibdev_cnt_inc(lnk);
776 	smcr_copy_dev_info_to_link(lnk);
777 	atomic_set(&lnk->conn_cnt, 0);
778 	smc_llc_link_set_uid(lnk);
779 	INIT_WORK(&lnk->link_down_wrk, smc_link_down_work);
780 	if (!lnk->smcibdev->initialized) {
781 		rc = (int)smc_ib_setup_per_ibdev(lnk->smcibdev);
782 		if (rc)
783 			goto out;
784 	}
785 	get_random_bytes(rndvec, sizeof(rndvec));
786 	lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) +
787 		(rndvec[2] << 16);
788 	rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport,
789 				  ini->vlan_id, lnk->gid, &lnk->sgid_index,
790 				  lgr->smc_version == SMC_V2 ?
791 						  &ini->smcrv2 : NULL);
792 	if (rc)
793 		goto out;
794 	rc = smc_llc_link_init(lnk);
795 	if (rc)
796 		goto out;
797 	rc = smc_wr_alloc_link_mem(lnk);
798 	if (rc)
799 		goto clear_llc_lnk;
800 	rc = smc_ib_create_protection_domain(lnk);
801 	if (rc)
802 		goto free_link_mem;
803 	rc = smc_ib_create_queue_pair(lnk);
804 	if (rc)
805 		goto dealloc_pd;
806 	rc = smc_wr_create_link(lnk);
807 	if (rc)
808 		goto destroy_qp;
809 	lnk->state = SMC_LNK_ACTIVATING;
810 	return 0;
811 
812 destroy_qp:
813 	smc_ib_destroy_queue_pair(lnk);
814 dealloc_pd:
815 	smc_ib_dealloc_protection_domain(lnk);
816 free_link_mem:
817 	smc_wr_free_link_mem(lnk);
818 clear_llc_lnk:
819 	smc_llc_link_clear(lnk, false);
820 out:
821 	smc_ibdev_cnt_dec(lnk);
822 	put_device(&lnk->smcibdev->ibdev->dev);
823 	smcibdev = lnk->smcibdev;
824 	memset(lnk, 0, sizeof(struct smc_link));
825 	lnk->state = SMC_LNK_UNUSED;
826 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
827 		wake_up(&smcibdev->lnks_deleted);
828 	smc_lgr_put(lgr); /* lgr_hold above */
829 	return rc;
830 }
831 
832 /* create a new SMC link group */
833 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
834 {
835 	struct smc_link_group *lgr;
836 	struct list_head *lgr_list;
837 	struct smcd_dev *smcd;
838 	struct smc_link *lnk;
839 	spinlock_t *lgr_lock;
840 	u8 link_idx;
841 	int rc = 0;
842 	int i;
843 
844 	if (ini->is_smcd && ini->vlan_id) {
845 		if (smc_ism_get_vlan(ini->ism_dev[ini->ism_selected],
846 				     ini->vlan_id)) {
847 			rc = SMC_CLC_DECL_ISMVLANERR;
848 			goto out;
849 		}
850 	}
851 
852 	lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
853 	if (!lgr) {
854 		rc = SMC_CLC_DECL_MEM;
855 		goto ism_put_vlan;
856 	}
857 	lgr->tx_wq = alloc_workqueue("smc_tx_wq-%*phN", 0, 0,
858 				     SMC_LGR_ID_SIZE, &lgr->id);
859 	if (!lgr->tx_wq) {
860 		rc = -ENOMEM;
861 		goto free_lgr;
862 	}
863 	lgr->is_smcd = ini->is_smcd;
864 	lgr->sync_err = 0;
865 	lgr->terminating = 0;
866 	lgr->freeing = 0;
867 	lgr->vlan_id = ini->vlan_id;
868 	refcount_set(&lgr->refcnt, 1); /* set lgr refcnt to 1 */
869 	init_rwsem(&lgr->sndbufs_lock);
870 	init_rwsem(&lgr->rmbs_lock);
871 	rwlock_init(&lgr->conns_lock);
872 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
873 		INIT_LIST_HEAD(&lgr->sndbufs[i]);
874 		INIT_LIST_HEAD(&lgr->rmbs[i]);
875 	}
876 	lgr->next_link_id = 0;
877 	smc_lgr_list.num += SMC_LGR_NUM_INCR;
878 	memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE);
879 	INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
880 	INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work);
881 	lgr->conns_all = RB_ROOT;
882 	if (ini->is_smcd) {
883 		/* SMC-D specific settings */
884 		smcd = ini->ism_dev[ini->ism_selected];
885 		get_device(smcd->ops->get_dev(smcd));
886 		lgr->peer_gid.gid =
887 			ini->ism_peer_gid[ini->ism_selected].gid;
888 		lgr->peer_gid.gid_ext =
889 			ini->ism_peer_gid[ini->ism_selected].gid_ext;
890 		lgr->smcd = ini->ism_dev[ini->ism_selected];
891 		lgr_list = &ini->ism_dev[ini->ism_selected]->lgr_list;
892 		lgr_lock = &lgr->smcd->lgr_lock;
893 		lgr->smc_version = ini->smcd_version;
894 		lgr->peer_shutdown = 0;
895 		atomic_inc(&ini->ism_dev[ini->ism_selected]->lgr_cnt);
896 	} else {
897 		/* SMC-R specific settings */
898 		struct smc_ib_device *ibdev;
899 		int ibport;
900 
901 		lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
902 		lgr->smc_version = ini->smcr_version;
903 		memcpy(lgr->peer_systemid, ini->peer_systemid,
904 		       SMC_SYSTEMID_LEN);
905 		if (lgr->smc_version == SMC_V2) {
906 			ibdev = ini->smcrv2.ib_dev_v2;
907 			ibport = ini->smcrv2.ib_port_v2;
908 			lgr->saddr = ini->smcrv2.saddr;
909 			lgr->uses_gateway = ini->smcrv2.uses_gateway;
910 			memcpy(lgr->nexthop_mac, ini->smcrv2.nexthop_mac,
911 			       ETH_ALEN);
912 			lgr->max_conns = ini->max_conns;
913 			lgr->max_links = ini->max_links;
914 		} else {
915 			ibdev = ini->ib_dev;
916 			ibport = ini->ib_port;
917 			lgr->max_conns = SMC_CONN_PER_LGR_MAX;
918 			lgr->max_links = SMC_LINKS_ADD_LNK_MAX;
919 		}
920 		memcpy(lgr->pnet_id, ibdev->pnetid[ibport - 1],
921 		       SMC_MAX_PNETID_LEN);
922 		rc = smc_wr_alloc_lgr_mem(lgr);
923 		if (rc)
924 			goto free_wq;
925 		smc_llc_lgr_init(lgr, smc);
926 
927 		link_idx = SMC_SINGLE_LINK;
928 		lnk = &lgr->lnk[link_idx];
929 		rc = smcr_link_init(lgr, lnk, link_idx, ini);
930 		if (rc) {
931 			smc_wr_free_lgr_mem(lgr);
932 			goto free_wq;
933 		}
934 		lgr->net = smc_ib_net(lnk->smcibdev);
935 		lgr_list = &smc_lgr_list.list;
936 		lgr_lock = &smc_lgr_list.lock;
937 		lgr->buf_type = lgr->net->smc.sysctl_smcr_buf_type;
938 		atomic_inc(&lgr_cnt);
939 	}
940 	smc->conn.lgr = lgr;
941 	spin_lock_bh(lgr_lock);
942 	list_add_tail(&lgr->list, lgr_list);
943 	spin_unlock_bh(lgr_lock);
944 	return 0;
945 
946 free_wq:
947 	destroy_workqueue(lgr->tx_wq);
948 free_lgr:
949 	kfree(lgr);
950 ism_put_vlan:
951 	if (ini->is_smcd && ini->vlan_id)
952 		smc_ism_put_vlan(ini->ism_dev[ini->ism_selected], ini->vlan_id);
953 out:
954 	if (rc < 0) {
955 		if (rc == -ENOMEM)
956 			rc = SMC_CLC_DECL_MEM;
957 		else
958 			rc = SMC_CLC_DECL_INTERR;
959 	}
960 	return rc;
961 }
962 
963 static int smc_write_space(struct smc_connection *conn)
964 {
965 	int buffer_len = conn->peer_rmbe_size;
966 	union smc_host_cursor prod;
967 	union smc_host_cursor cons;
968 	int space;
969 
970 	smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
971 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
972 	/* determine rx_buf space */
973 	space = buffer_len - smc_curs_diff(buffer_len, &cons, &prod);
974 	return space;
975 }
976 
977 static int smc_switch_cursor(struct smc_sock *smc, struct smc_cdc_tx_pend *pend,
978 			     struct smc_wr_buf *wr_buf)
979 {
980 	struct smc_connection *conn = &smc->conn;
981 	union smc_host_cursor cons, fin;
982 	int rc = 0;
983 	int diff;
984 
985 	smc_curs_copy(&conn->tx_curs_sent, &conn->tx_curs_fin, conn);
986 	smc_curs_copy(&fin, &conn->local_tx_ctrl_fin, conn);
987 	/* set prod cursor to old state, enforce tx_rdma_writes() */
988 	smc_curs_copy(&conn->local_tx_ctrl.prod, &fin, conn);
989 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
990 
991 	if (smc_curs_comp(conn->peer_rmbe_size, &cons, &fin) < 0) {
992 		/* cons cursor advanced more than fin, and prod was set
993 		 * fin above, so now prod is smaller than cons. Fix that.
994 		 */
995 		diff = smc_curs_diff(conn->peer_rmbe_size, &fin, &cons);
996 		smc_curs_add(conn->sndbuf_desc->len,
997 			     &conn->tx_curs_sent, diff);
998 		smc_curs_add(conn->sndbuf_desc->len,
999 			     &conn->tx_curs_fin, diff);
1000 
1001 		smp_mb__before_atomic();
1002 		atomic_add(diff, &conn->sndbuf_space);
1003 		smp_mb__after_atomic();
1004 
1005 		smc_curs_add(conn->peer_rmbe_size,
1006 			     &conn->local_tx_ctrl.prod, diff);
1007 		smc_curs_add(conn->peer_rmbe_size,
1008 			     &conn->local_tx_ctrl_fin, diff);
1009 	}
1010 	/* recalculate, value is used by tx_rdma_writes() */
1011 	atomic_set(&smc->conn.peer_rmbe_space, smc_write_space(conn));
1012 
1013 	if (smc->sk.sk_state != SMC_INIT &&
1014 	    smc->sk.sk_state != SMC_CLOSED) {
1015 		rc = smcr_cdc_msg_send_validation(conn, pend, wr_buf);
1016 		if (!rc) {
1017 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work, 0);
1018 			smc->sk.sk_data_ready(&smc->sk);
1019 		}
1020 	} else {
1021 		smc_wr_tx_put_slot(conn->lnk,
1022 				   (struct smc_wr_tx_pend_priv *)pend);
1023 	}
1024 	return rc;
1025 }
1026 
1027 void smc_switch_link_and_count(struct smc_connection *conn,
1028 			       struct smc_link *to_lnk)
1029 {
1030 	atomic_dec(&conn->lnk->conn_cnt);
1031 	/* link_hold in smc_conn_create() */
1032 	smcr_link_put(conn->lnk);
1033 	conn->lnk = to_lnk;
1034 	atomic_inc(&conn->lnk->conn_cnt);
1035 	/* link_put in smc_conn_free() */
1036 	smcr_link_hold(conn->lnk);
1037 }
1038 
1039 struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
1040 				  struct smc_link *from_lnk, bool is_dev_err)
1041 {
1042 	struct smc_link *to_lnk = NULL;
1043 	struct smc_cdc_tx_pend *pend;
1044 	struct smc_connection *conn;
1045 	struct smc_wr_buf *wr_buf;
1046 	struct smc_sock *smc;
1047 	struct rb_node *node;
1048 	int i, rc = 0;
1049 
1050 	/* link is inactive, wake up tx waiters */
1051 	smc_wr_wakeup_tx_wait(from_lnk);
1052 
1053 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1054 		if (!smc_link_active(&lgr->lnk[i]) || i == from_lnk->link_idx)
1055 			continue;
1056 		if (is_dev_err && from_lnk->smcibdev == lgr->lnk[i].smcibdev &&
1057 		    from_lnk->ibport == lgr->lnk[i].ibport) {
1058 			continue;
1059 		}
1060 		to_lnk = &lgr->lnk[i];
1061 		break;
1062 	}
1063 	if (!to_lnk || !smc_wr_tx_link_hold(to_lnk)) {
1064 		smc_lgr_terminate_sched(lgr);
1065 		return NULL;
1066 	}
1067 again:
1068 	read_lock_bh(&lgr->conns_lock);
1069 	for (node = rb_first(&lgr->conns_all); node; node = rb_next(node)) {
1070 		conn = rb_entry(node, struct smc_connection, alert_node);
1071 		if (conn->lnk != from_lnk)
1072 			continue;
1073 		smc = container_of(conn, struct smc_sock, conn);
1074 		/* conn->lnk not yet set in SMC_INIT state */
1075 		if (smc->sk.sk_state == SMC_INIT)
1076 			continue;
1077 		if (smc->sk.sk_state == SMC_CLOSED ||
1078 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT1 ||
1079 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT2 ||
1080 		    smc->sk.sk_state == SMC_APPFINCLOSEWAIT ||
1081 		    smc->sk.sk_state == SMC_APPCLOSEWAIT1 ||
1082 		    smc->sk.sk_state == SMC_APPCLOSEWAIT2 ||
1083 		    smc->sk.sk_state == SMC_PEERFINCLOSEWAIT ||
1084 		    smc->sk.sk_state == SMC_PEERABORTWAIT ||
1085 		    smc->sk.sk_state == SMC_PROCESSABORT) {
1086 			spin_lock_bh(&conn->send_lock);
1087 			smc_switch_link_and_count(conn, to_lnk);
1088 			spin_unlock_bh(&conn->send_lock);
1089 			continue;
1090 		}
1091 		sock_hold(&smc->sk);
1092 		read_unlock_bh(&lgr->conns_lock);
1093 		/* pre-fetch buffer outside of send_lock, might sleep */
1094 		rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
1095 		if (rc)
1096 			goto err_out;
1097 		/* avoid race with smcr_tx_sndbuf_nonempty() */
1098 		spin_lock_bh(&conn->send_lock);
1099 		smc_switch_link_and_count(conn, to_lnk);
1100 		rc = smc_switch_cursor(smc, pend, wr_buf);
1101 		spin_unlock_bh(&conn->send_lock);
1102 		sock_put(&smc->sk);
1103 		if (rc)
1104 			goto err_out;
1105 		goto again;
1106 	}
1107 	read_unlock_bh(&lgr->conns_lock);
1108 	smc_wr_tx_link_put(to_lnk);
1109 	return to_lnk;
1110 
1111 err_out:
1112 	smcr_link_down_cond_sched(to_lnk);
1113 	smc_wr_tx_link_put(to_lnk);
1114 	return NULL;
1115 }
1116 
1117 static void smcr_buf_unuse(struct smc_buf_desc *buf_desc, bool is_rmb,
1118 			   struct smc_link_group *lgr)
1119 {
1120 	struct rw_semaphore *lock;	/* lock buffer list */
1121 	int rc;
1122 
1123 	if (is_rmb && buf_desc->is_conf_rkey && !list_empty(&lgr->list)) {
1124 		/* unregister rmb with peer */
1125 		rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
1126 		if (!rc) {
1127 			/* protect against smc_llc_cli_rkey_exchange() */
1128 			down_read(&lgr->llc_conf_mutex);
1129 			smc_llc_do_delete_rkey(lgr, buf_desc);
1130 			buf_desc->is_conf_rkey = false;
1131 			up_read(&lgr->llc_conf_mutex);
1132 			smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1133 		}
1134 	}
1135 
1136 	if (buf_desc->is_reg_err) {
1137 		/* buf registration failed, reuse not possible */
1138 		lock = is_rmb ? &lgr->rmbs_lock :
1139 				&lgr->sndbufs_lock;
1140 		down_write(lock);
1141 		list_del(&buf_desc->list);
1142 		up_write(lock);
1143 
1144 		smc_buf_free(lgr, is_rmb, buf_desc);
1145 	} else {
1146 		/* memzero_explicit provides potential memory barrier semantics */
1147 		memzero_explicit(buf_desc->cpu_addr, buf_desc->len);
1148 		WRITE_ONCE(buf_desc->used, 0);
1149 	}
1150 }
1151 
1152 static void smc_buf_unuse(struct smc_connection *conn,
1153 			  struct smc_link_group *lgr)
1154 {
1155 	if (conn->sndbuf_desc) {
1156 		if (!lgr->is_smcd && conn->sndbuf_desc->is_vm) {
1157 			smcr_buf_unuse(conn->sndbuf_desc, false, lgr);
1158 		} else {
1159 			memzero_explicit(conn->sndbuf_desc->cpu_addr, conn->sndbuf_desc->len);
1160 			WRITE_ONCE(conn->sndbuf_desc->used, 0);
1161 		}
1162 	}
1163 	if (conn->rmb_desc) {
1164 		if (!lgr->is_smcd) {
1165 			smcr_buf_unuse(conn->rmb_desc, true, lgr);
1166 		} else {
1167 			memzero_explicit(conn->rmb_desc->cpu_addr,
1168 					 conn->rmb_desc->len + sizeof(struct smcd_cdc_msg));
1169 			WRITE_ONCE(conn->rmb_desc->used, 0);
1170 		}
1171 	}
1172 }
1173 
1174 /* remove a finished connection from its link group */
1175 void smc_conn_free(struct smc_connection *conn)
1176 {
1177 	struct smc_link_group *lgr = conn->lgr;
1178 
1179 	if (!lgr || conn->freed)
1180 		/* Connection has never been registered in a
1181 		 * link group, or has already been freed.
1182 		 */
1183 		return;
1184 
1185 	conn->freed = 1;
1186 	if (!smc_conn_lgr_valid(conn))
1187 		/* Connection has already unregistered from
1188 		 * link group.
1189 		 */
1190 		goto lgr_put;
1191 
1192 	if (lgr->is_smcd) {
1193 		if (!list_empty(&lgr->list))
1194 			smc_ism_unset_conn(conn);
1195 		tasklet_kill(&conn->rx_tsklet);
1196 	} else {
1197 		smc_cdc_wait_pend_tx_wr(conn);
1198 		if (current_work() != &conn->abort_work)
1199 			cancel_work_sync(&conn->abort_work);
1200 	}
1201 	if (!list_empty(&lgr->list)) {
1202 		smc_buf_unuse(conn, lgr); /* allow buffer reuse */
1203 		smc_lgr_unregister_conn(conn);
1204 	}
1205 
1206 	if (!lgr->conns_num)
1207 		smc_lgr_schedule_free_work(lgr);
1208 lgr_put:
1209 	if (!lgr->is_smcd)
1210 		smcr_link_put(conn->lnk); /* link_hold in smc_conn_create() */
1211 	smc_lgr_put(lgr); /* lgr_hold in smc_conn_create() */
1212 }
1213 
1214 /* unregister a link from a buf_desc */
1215 static void smcr_buf_unmap_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1216 				struct smc_link *lnk)
1217 {
1218 	if (is_rmb || buf_desc->is_vm)
1219 		buf_desc->is_reg_mr[lnk->link_idx] = false;
1220 	if (!buf_desc->is_map_ib[lnk->link_idx])
1221 		return;
1222 
1223 	if ((is_rmb || buf_desc->is_vm) &&
1224 	    buf_desc->mr[lnk->link_idx]) {
1225 		smc_ib_put_memory_region(buf_desc->mr[lnk->link_idx]);
1226 		buf_desc->mr[lnk->link_idx] = NULL;
1227 	}
1228 	if (is_rmb)
1229 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_FROM_DEVICE);
1230 	else
1231 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_TO_DEVICE);
1232 
1233 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1234 	buf_desc->is_map_ib[lnk->link_idx] = false;
1235 }
1236 
1237 /* unmap all buffers of lgr for a deleted link */
1238 static void smcr_buf_unmap_lgr(struct smc_link *lnk)
1239 {
1240 	struct smc_link_group *lgr = lnk->lgr;
1241 	struct smc_buf_desc *buf_desc, *bf;
1242 	int i;
1243 
1244 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1245 		down_write(&lgr->rmbs_lock);
1246 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list)
1247 			smcr_buf_unmap_link(buf_desc, true, lnk);
1248 		up_write(&lgr->rmbs_lock);
1249 
1250 		down_write(&lgr->sndbufs_lock);
1251 		list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i],
1252 					 list)
1253 			smcr_buf_unmap_link(buf_desc, false, lnk);
1254 		up_write(&lgr->sndbufs_lock);
1255 	}
1256 }
1257 
1258 static void smcr_rtoken_clear_link(struct smc_link *lnk)
1259 {
1260 	struct smc_link_group *lgr = lnk->lgr;
1261 	int i;
1262 
1263 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
1264 		lgr->rtokens[i][lnk->link_idx].rkey = 0;
1265 		lgr->rtokens[i][lnk->link_idx].dma_addr = 0;
1266 	}
1267 }
1268 
1269 static void __smcr_link_clear(struct smc_link *lnk)
1270 {
1271 	struct smc_link_group *lgr = lnk->lgr;
1272 	struct smc_ib_device *smcibdev;
1273 
1274 	smc_wr_free_link_mem(lnk);
1275 	smc_ibdev_cnt_dec(lnk);
1276 	put_device(&lnk->smcibdev->ibdev->dev);
1277 	smcibdev = lnk->smcibdev;
1278 	memset(lnk, 0, sizeof(struct smc_link));
1279 	lnk->state = SMC_LNK_UNUSED;
1280 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
1281 		wake_up(&smcibdev->lnks_deleted);
1282 	smc_lgr_put(lgr); /* lgr_hold in smcr_link_init() */
1283 }
1284 
1285 /* must be called under lgr->llc_conf_mutex lock */
1286 void smcr_link_clear(struct smc_link *lnk, bool log)
1287 {
1288 	if (!lnk->lgr || lnk->clearing ||
1289 	    lnk->state == SMC_LNK_UNUSED)
1290 		return;
1291 	lnk->clearing = 1;
1292 	lnk->peer_qpn = 0;
1293 	smc_llc_link_clear(lnk, log);
1294 	smcr_buf_unmap_lgr(lnk);
1295 	smcr_rtoken_clear_link(lnk);
1296 	smc_ib_modify_qp_error(lnk);
1297 	smc_wr_free_link(lnk);
1298 	smc_ib_destroy_queue_pair(lnk);
1299 	smc_ib_dealloc_protection_domain(lnk);
1300 	smcr_link_put(lnk); /* theoretically last link_put */
1301 }
1302 
1303 void smcr_link_hold(struct smc_link *lnk)
1304 {
1305 	refcount_inc(&lnk->refcnt);
1306 }
1307 
1308 void smcr_link_put(struct smc_link *lnk)
1309 {
1310 	if (refcount_dec_and_test(&lnk->refcnt))
1311 		__smcr_link_clear(lnk);
1312 }
1313 
1314 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb,
1315 			  struct smc_buf_desc *buf_desc)
1316 {
1317 	int i;
1318 
1319 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1320 		smcr_buf_unmap_link(buf_desc, is_rmb, &lgr->lnk[i]);
1321 
1322 	if (!buf_desc->is_vm && buf_desc->pages)
1323 		__free_pages(buf_desc->pages, buf_desc->order);
1324 	else if (buf_desc->is_vm && buf_desc->cpu_addr)
1325 		vfree(buf_desc->cpu_addr);
1326 	kfree(buf_desc);
1327 }
1328 
1329 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb,
1330 			  struct smc_buf_desc *buf_desc)
1331 {
1332 	if (is_dmb) {
1333 		/* restore original buf len */
1334 		buf_desc->len += sizeof(struct smcd_cdc_msg);
1335 		smc_ism_unregister_dmb(lgr->smcd, buf_desc);
1336 	} else {
1337 		kfree(buf_desc->cpu_addr);
1338 	}
1339 	kfree(buf_desc);
1340 }
1341 
1342 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
1343 			 struct smc_buf_desc *buf_desc)
1344 {
1345 	if (lgr->is_smcd)
1346 		smcd_buf_free(lgr, is_rmb, buf_desc);
1347 	else
1348 		smcr_buf_free(lgr, is_rmb, buf_desc);
1349 }
1350 
1351 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb)
1352 {
1353 	struct smc_buf_desc *buf_desc, *bf_desc;
1354 	struct list_head *buf_list;
1355 	int i;
1356 
1357 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1358 		if (is_rmb)
1359 			buf_list = &lgr->rmbs[i];
1360 		else
1361 			buf_list = &lgr->sndbufs[i];
1362 		list_for_each_entry_safe(buf_desc, bf_desc, buf_list,
1363 					 list) {
1364 			list_del(&buf_desc->list);
1365 			smc_buf_free(lgr, is_rmb, buf_desc);
1366 		}
1367 	}
1368 }
1369 
1370 static void smc_lgr_free_bufs(struct smc_link_group *lgr)
1371 {
1372 	/* free send buffers */
1373 	__smc_lgr_free_bufs(lgr, false);
1374 	/* free rmbs */
1375 	__smc_lgr_free_bufs(lgr, true);
1376 }
1377 
1378 /* won't be freed until no one accesses to lgr anymore */
1379 static void __smc_lgr_free(struct smc_link_group *lgr)
1380 {
1381 	smc_lgr_free_bufs(lgr);
1382 	if (lgr->is_smcd) {
1383 		if (!atomic_dec_return(&lgr->smcd->lgr_cnt))
1384 			wake_up(&lgr->smcd->lgrs_deleted);
1385 	} else {
1386 		smc_wr_free_lgr_mem(lgr);
1387 		if (!atomic_dec_return(&lgr_cnt))
1388 			wake_up(&lgrs_deleted);
1389 	}
1390 	kfree(lgr);
1391 }
1392 
1393 /* remove a link group */
1394 static void smc_lgr_free(struct smc_link_group *lgr)
1395 {
1396 	int i;
1397 
1398 	if (!lgr->is_smcd) {
1399 		down_write(&lgr->llc_conf_mutex);
1400 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1401 			if (lgr->lnk[i].state != SMC_LNK_UNUSED)
1402 				smcr_link_clear(&lgr->lnk[i], false);
1403 		}
1404 		up_write(&lgr->llc_conf_mutex);
1405 		smc_llc_lgr_clear(lgr);
1406 	}
1407 
1408 	destroy_workqueue(lgr->tx_wq);
1409 	if (lgr->is_smcd) {
1410 		smc_ism_put_vlan(lgr->smcd, lgr->vlan_id);
1411 		put_device(lgr->smcd->ops->get_dev(lgr->smcd));
1412 	}
1413 	smc_lgr_put(lgr); /* theoretically last lgr_put */
1414 }
1415 
1416 void smc_lgr_hold(struct smc_link_group *lgr)
1417 {
1418 	refcount_inc(&lgr->refcnt);
1419 }
1420 
1421 void smc_lgr_put(struct smc_link_group *lgr)
1422 {
1423 	if (refcount_dec_and_test(&lgr->refcnt))
1424 		__smc_lgr_free(lgr);
1425 }
1426 
1427 static void smc_sk_wake_ups(struct smc_sock *smc)
1428 {
1429 	smc->sk.sk_write_space(&smc->sk);
1430 	smc->sk.sk_data_ready(&smc->sk);
1431 	smc->sk.sk_state_change(&smc->sk);
1432 }
1433 
1434 /* kill a connection */
1435 static void smc_conn_kill(struct smc_connection *conn, bool soft)
1436 {
1437 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1438 
1439 	if (conn->lgr->is_smcd && conn->lgr->peer_shutdown)
1440 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
1441 	else
1442 		smc_close_abort(conn);
1443 	conn->killed = 1;
1444 	smc->sk.sk_err = ECONNABORTED;
1445 	smc_sk_wake_ups(smc);
1446 	if (conn->lgr->is_smcd) {
1447 		smc_ism_unset_conn(conn);
1448 		if (soft)
1449 			tasklet_kill(&conn->rx_tsklet);
1450 		else
1451 			tasklet_unlock_wait(&conn->rx_tsklet);
1452 	} else {
1453 		smc_cdc_wait_pend_tx_wr(conn);
1454 	}
1455 	smc_lgr_unregister_conn(conn);
1456 	smc_close_active_abort(smc);
1457 }
1458 
1459 static void smc_lgr_cleanup(struct smc_link_group *lgr)
1460 {
1461 	if (lgr->is_smcd) {
1462 		smc_ism_signal_shutdown(lgr);
1463 	} else {
1464 		u32 rsn = lgr->llc_termination_rsn;
1465 
1466 		if (!rsn)
1467 			rsn = SMC_LLC_DEL_PROG_INIT_TERM;
1468 		smc_llc_send_link_delete_all(lgr, false, rsn);
1469 		smcr_lgr_link_deactivate_all(lgr);
1470 	}
1471 }
1472 
1473 /* terminate link group
1474  * @soft: true if link group shutdown can take its time
1475  *	  false if immediate link group shutdown is required
1476  */
1477 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft)
1478 {
1479 	struct smc_connection *conn;
1480 	struct smc_sock *smc;
1481 	struct rb_node *node;
1482 
1483 	if (lgr->terminating)
1484 		return;	/* lgr already terminating */
1485 	/* cancel free_work sync, will terminate when lgr->freeing is set */
1486 	cancel_delayed_work(&lgr->free_work);
1487 	lgr->terminating = 1;
1488 
1489 	/* kill remaining link group connections */
1490 	read_lock_bh(&lgr->conns_lock);
1491 	node = rb_first(&lgr->conns_all);
1492 	while (node) {
1493 		read_unlock_bh(&lgr->conns_lock);
1494 		conn = rb_entry(node, struct smc_connection, alert_node);
1495 		smc = container_of(conn, struct smc_sock, conn);
1496 		sock_hold(&smc->sk); /* sock_put below */
1497 		lock_sock(&smc->sk);
1498 		smc_conn_kill(conn, soft);
1499 		release_sock(&smc->sk);
1500 		sock_put(&smc->sk); /* sock_hold above */
1501 		read_lock_bh(&lgr->conns_lock);
1502 		node = rb_first(&lgr->conns_all);
1503 	}
1504 	read_unlock_bh(&lgr->conns_lock);
1505 	smc_lgr_cleanup(lgr);
1506 	smc_lgr_free(lgr);
1507 }
1508 
1509 /* unlink link group and schedule termination */
1510 void smc_lgr_terminate_sched(struct smc_link_group *lgr)
1511 {
1512 	spinlock_t *lgr_lock;
1513 
1514 	smc_lgr_list_head(lgr, &lgr_lock);
1515 	spin_lock_bh(lgr_lock);
1516 	if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) {
1517 		spin_unlock_bh(lgr_lock);
1518 		return;	/* lgr already terminating */
1519 	}
1520 	list_del_init(&lgr->list);
1521 	lgr->freeing = 1;
1522 	spin_unlock_bh(lgr_lock);
1523 	schedule_work(&lgr->terminate_work);
1524 }
1525 
1526 /* Called when peer lgr shutdown (regularly or abnormally) is received */
1527 void smc_smcd_terminate(struct smcd_dev *dev, struct smcd_gid *peer_gid,
1528 			unsigned short vlan)
1529 {
1530 	struct smc_link_group *lgr, *l;
1531 	LIST_HEAD(lgr_free_list);
1532 
1533 	/* run common cleanup function and build free list */
1534 	spin_lock_bh(&dev->lgr_lock);
1535 	list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) {
1536 		if ((!peer_gid->gid ||
1537 		     (lgr->peer_gid.gid == peer_gid->gid &&
1538 		      !smc_ism_is_virtual(dev) ? 1 :
1539 		      lgr->peer_gid.gid_ext == peer_gid->gid_ext)) &&
1540 		    (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) {
1541 			if (peer_gid->gid) /* peer triggered termination */
1542 				lgr->peer_shutdown = 1;
1543 			list_move(&lgr->list, &lgr_free_list);
1544 			lgr->freeing = 1;
1545 		}
1546 	}
1547 	spin_unlock_bh(&dev->lgr_lock);
1548 
1549 	/* cancel the regular free workers and actually free lgrs */
1550 	list_for_each_entry_safe(lgr, l, &lgr_free_list, list) {
1551 		list_del_init(&lgr->list);
1552 		schedule_work(&lgr->terminate_work);
1553 	}
1554 }
1555 
1556 /* Called when an SMCD device is removed or the smc module is unloaded */
1557 void smc_smcd_terminate_all(struct smcd_dev *smcd)
1558 {
1559 	struct smc_link_group *lgr, *lg;
1560 	LIST_HEAD(lgr_free_list);
1561 
1562 	spin_lock_bh(&smcd->lgr_lock);
1563 	list_splice_init(&smcd->lgr_list, &lgr_free_list);
1564 	list_for_each_entry(lgr, &lgr_free_list, list)
1565 		lgr->freeing = 1;
1566 	spin_unlock_bh(&smcd->lgr_lock);
1567 
1568 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1569 		list_del_init(&lgr->list);
1570 		__smc_lgr_terminate(lgr, false);
1571 	}
1572 
1573 	if (atomic_read(&smcd->lgr_cnt))
1574 		wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt));
1575 }
1576 
1577 /* Called when an SMCR device is removed or the smc module is unloaded.
1578  * If smcibdev is given, all SMCR link groups using this device are terminated.
1579  * If smcibdev is NULL, all SMCR link groups are terminated.
1580  */
1581 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev)
1582 {
1583 	struct smc_link_group *lgr, *lg;
1584 	LIST_HEAD(lgr_free_list);
1585 	int i;
1586 
1587 	spin_lock_bh(&smc_lgr_list.lock);
1588 	if (!smcibdev) {
1589 		list_splice_init(&smc_lgr_list.list, &lgr_free_list);
1590 		list_for_each_entry(lgr, &lgr_free_list, list)
1591 			lgr->freeing = 1;
1592 	} else {
1593 		list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) {
1594 			for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1595 				if (lgr->lnk[i].smcibdev == smcibdev)
1596 					smcr_link_down_cond_sched(&lgr->lnk[i]);
1597 			}
1598 		}
1599 	}
1600 	spin_unlock_bh(&smc_lgr_list.lock);
1601 
1602 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1603 		list_del_init(&lgr->list);
1604 		smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_OP_INIT_TERM);
1605 		__smc_lgr_terminate(lgr, false);
1606 	}
1607 
1608 	if (smcibdev) {
1609 		if (atomic_read(&smcibdev->lnk_cnt))
1610 			wait_event(smcibdev->lnks_deleted,
1611 				   !atomic_read(&smcibdev->lnk_cnt));
1612 	} else {
1613 		if (atomic_read(&lgr_cnt))
1614 			wait_event(lgrs_deleted, !atomic_read(&lgr_cnt));
1615 	}
1616 }
1617 
1618 /* set new lgr type and clear all asymmetric link tagging */
1619 void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type)
1620 {
1621 	char *lgr_type = "";
1622 	int i;
1623 
1624 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1625 		if (smc_link_usable(&lgr->lnk[i]))
1626 			lgr->lnk[i].link_is_asym = false;
1627 	if (lgr->type == new_type)
1628 		return;
1629 	lgr->type = new_type;
1630 
1631 	switch (lgr->type) {
1632 	case SMC_LGR_NONE:
1633 		lgr_type = "NONE";
1634 		break;
1635 	case SMC_LGR_SINGLE:
1636 		lgr_type = "SINGLE";
1637 		break;
1638 	case SMC_LGR_SYMMETRIC:
1639 		lgr_type = "SYMMETRIC";
1640 		break;
1641 	case SMC_LGR_ASYMMETRIC_PEER:
1642 		lgr_type = "ASYMMETRIC_PEER";
1643 		break;
1644 	case SMC_LGR_ASYMMETRIC_LOCAL:
1645 		lgr_type = "ASYMMETRIC_LOCAL";
1646 		break;
1647 	}
1648 	pr_warn_ratelimited("smc: SMC-R lg %*phN net %llu state changed: "
1649 			    "%s, pnetid %.16s\n", SMC_LGR_ID_SIZE, &lgr->id,
1650 			    lgr->net->net_cookie, lgr_type, lgr->pnet_id);
1651 }
1652 
1653 /* set new lgr type and tag a link as asymmetric */
1654 void smcr_lgr_set_type_asym(struct smc_link_group *lgr,
1655 			    enum smc_lgr_type new_type, int asym_lnk_idx)
1656 {
1657 	smcr_lgr_set_type(lgr, new_type);
1658 	lgr->lnk[asym_lnk_idx].link_is_asym = true;
1659 }
1660 
1661 /* abort connection, abort_work scheduled from tasklet context */
1662 static void smc_conn_abort_work(struct work_struct *work)
1663 {
1664 	struct smc_connection *conn = container_of(work,
1665 						   struct smc_connection,
1666 						   abort_work);
1667 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1668 
1669 	lock_sock(&smc->sk);
1670 	smc_conn_kill(conn, true);
1671 	release_sock(&smc->sk);
1672 	sock_put(&smc->sk); /* sock_hold done by schedulers of abort_work */
1673 }
1674 
1675 void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport)
1676 {
1677 	struct smc_link_group *lgr, *n;
1678 
1679 	spin_lock_bh(&smc_lgr_list.lock);
1680 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1681 		struct smc_link *link;
1682 
1683 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1684 			    SMC_MAX_PNETID_LEN) ||
1685 		    lgr->type == SMC_LGR_SYMMETRIC ||
1686 		    lgr->type == SMC_LGR_ASYMMETRIC_PEER ||
1687 		    !rdma_dev_access_netns(smcibdev->ibdev, lgr->net))
1688 			continue;
1689 
1690 		if (lgr->type == SMC_LGR_SINGLE && lgr->max_links <= 1)
1691 			continue;
1692 
1693 		/* trigger local add link processing */
1694 		link = smc_llc_usable_link(lgr);
1695 		if (link)
1696 			smc_llc_add_link_local(link);
1697 	}
1698 	spin_unlock_bh(&smc_lgr_list.lock);
1699 }
1700 
1701 /* link is down - switch connections to alternate link,
1702  * must be called under lgr->llc_conf_mutex lock
1703  */
1704 static void smcr_link_down(struct smc_link *lnk)
1705 {
1706 	struct smc_link_group *lgr = lnk->lgr;
1707 	struct smc_link *to_lnk;
1708 	int del_link_id;
1709 
1710 	if (!lgr || lnk->state == SMC_LNK_UNUSED || list_empty(&lgr->list))
1711 		return;
1712 
1713 	to_lnk = smc_switch_conns(lgr, lnk, true);
1714 	if (!to_lnk) { /* no backup link available */
1715 		smcr_link_clear(lnk, true);
1716 		return;
1717 	}
1718 	smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1719 	del_link_id = lnk->link_id;
1720 
1721 	if (lgr->role == SMC_SERV) {
1722 		/* trigger local delete link processing */
1723 		smc_llc_srv_delete_link_local(to_lnk, del_link_id);
1724 	} else {
1725 		if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) {
1726 			/* another llc task is ongoing */
1727 			up_write(&lgr->llc_conf_mutex);
1728 			wait_event_timeout(lgr->llc_flow_waiter,
1729 				(list_empty(&lgr->list) ||
1730 				 lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE),
1731 				SMC_LLC_WAIT_TIME);
1732 			down_write(&lgr->llc_conf_mutex);
1733 		}
1734 		if (!list_empty(&lgr->list)) {
1735 			smc_llc_send_delete_link(to_lnk, del_link_id,
1736 						 SMC_LLC_REQ, true,
1737 						 SMC_LLC_DEL_LOST_PATH);
1738 			smcr_link_clear(lnk, true);
1739 		}
1740 		wake_up(&lgr->llc_flow_waiter);	/* wake up next waiter */
1741 	}
1742 }
1743 
1744 /* must be called under lgr->llc_conf_mutex lock */
1745 void smcr_link_down_cond(struct smc_link *lnk)
1746 {
1747 	if (smc_link_downing(&lnk->state)) {
1748 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1749 		smcr_link_down(lnk);
1750 	}
1751 }
1752 
1753 /* will get the lgr->llc_conf_mutex lock */
1754 void smcr_link_down_cond_sched(struct smc_link *lnk)
1755 {
1756 	if (smc_link_downing(&lnk->state)) {
1757 		trace_smcr_link_down(lnk, __builtin_return_address(0));
1758 		schedule_work(&lnk->link_down_wrk);
1759 	}
1760 }
1761 
1762 void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport)
1763 {
1764 	struct smc_link_group *lgr, *n;
1765 	int i;
1766 
1767 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1768 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1769 			    SMC_MAX_PNETID_LEN))
1770 			continue; /* lgr is not affected */
1771 		if (list_empty(&lgr->list))
1772 			continue;
1773 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1774 			struct smc_link *lnk = &lgr->lnk[i];
1775 
1776 			if (smc_link_usable(lnk) &&
1777 			    lnk->smcibdev == smcibdev && lnk->ibport == ibport)
1778 				smcr_link_down_cond_sched(lnk);
1779 		}
1780 	}
1781 }
1782 
1783 static void smc_link_down_work(struct work_struct *work)
1784 {
1785 	struct smc_link *link = container_of(work, struct smc_link,
1786 					     link_down_wrk);
1787 	struct smc_link_group *lgr = link->lgr;
1788 
1789 	if (list_empty(&lgr->list))
1790 		return;
1791 	wake_up_all(&lgr->llc_msg_waiter);
1792 	down_write(&lgr->llc_conf_mutex);
1793 	smcr_link_down(link);
1794 	up_write(&lgr->llc_conf_mutex);
1795 }
1796 
1797 static int smc_vlan_by_tcpsk_walk(struct net_device *lower_dev,
1798 				  struct netdev_nested_priv *priv)
1799 {
1800 	unsigned short *vlan_id = (unsigned short *)priv->data;
1801 
1802 	if (is_vlan_dev(lower_dev)) {
1803 		*vlan_id = vlan_dev_vlan_id(lower_dev);
1804 		return 1;
1805 	}
1806 
1807 	return 0;
1808 }
1809 
1810 /* Determine vlan of internal TCP socket. */
1811 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini)
1812 {
1813 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
1814 	struct netdev_nested_priv priv;
1815 	struct net_device *ndev;
1816 	int rc = 0;
1817 
1818 	ini->vlan_id = 0;
1819 	if (!dst) {
1820 		rc = -ENOTCONN;
1821 		goto out;
1822 	}
1823 	if (!dst->dev) {
1824 		rc = -ENODEV;
1825 		goto out_rel;
1826 	}
1827 
1828 	ndev = dst->dev;
1829 	if (is_vlan_dev(ndev)) {
1830 		ini->vlan_id = vlan_dev_vlan_id(ndev);
1831 		goto out_rel;
1832 	}
1833 
1834 	priv.data = (void *)&ini->vlan_id;
1835 	rtnl_lock();
1836 	netdev_walk_all_lower_dev(ndev, smc_vlan_by_tcpsk_walk, &priv);
1837 	rtnl_unlock();
1838 
1839 out_rel:
1840 	dst_release(dst);
1841 out:
1842 	return rc;
1843 }
1844 
1845 static bool smcr_lgr_match(struct smc_link_group *lgr, u8 smcr_version,
1846 			   u8 peer_systemid[],
1847 			   u8 peer_gid[],
1848 			   u8 peer_mac_v1[],
1849 			   enum smc_lgr_role role, u32 clcqpn,
1850 			   struct net *net)
1851 {
1852 	struct smc_link *lnk;
1853 	int i;
1854 
1855 	if (memcmp(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN) ||
1856 	    lgr->role != role)
1857 		return false;
1858 
1859 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1860 		lnk = &lgr->lnk[i];
1861 
1862 		if (!smc_link_active(lnk))
1863 			continue;
1864 		/* use verbs API to check netns, instead of lgr->net */
1865 		if (!rdma_dev_access_netns(lnk->smcibdev->ibdev, net))
1866 			return false;
1867 		if ((lgr->role == SMC_SERV || lnk->peer_qpn == clcqpn) &&
1868 		    !memcmp(lnk->peer_gid, peer_gid, SMC_GID_SIZE) &&
1869 		    (smcr_version == SMC_V2 ||
1870 		     !memcmp(lnk->peer_mac, peer_mac_v1, ETH_ALEN)))
1871 			return true;
1872 	}
1873 	return false;
1874 }
1875 
1876 static bool smcd_lgr_match(struct smc_link_group *lgr,
1877 			   struct smcd_dev *smcismdev,
1878 			   struct smcd_gid *peer_gid)
1879 {
1880 	return lgr->peer_gid.gid == peer_gid->gid && lgr->smcd == smcismdev &&
1881 		smc_ism_is_virtual(smcismdev) ?
1882 		(lgr->peer_gid.gid_ext == peer_gid->gid_ext) : 1;
1883 }
1884 
1885 /* create a new SMC connection (and a new link group if necessary) */
1886 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
1887 {
1888 	struct smc_connection *conn = &smc->conn;
1889 	struct net *net = sock_net(&smc->sk);
1890 	struct list_head *lgr_list;
1891 	struct smc_link_group *lgr;
1892 	enum smc_lgr_role role;
1893 	spinlock_t *lgr_lock;
1894 	int rc = 0;
1895 
1896 	lgr_list = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_list :
1897 				  &smc_lgr_list.list;
1898 	lgr_lock = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_lock :
1899 				  &smc_lgr_list.lock;
1900 	ini->first_contact_local = 1;
1901 	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
1902 	if (role == SMC_CLNT && ini->first_contact_peer)
1903 		/* create new link group as well */
1904 		goto create;
1905 
1906 	/* determine if an existing link group can be reused */
1907 	spin_lock_bh(lgr_lock);
1908 	list_for_each_entry(lgr, lgr_list, list) {
1909 		write_lock_bh(&lgr->conns_lock);
1910 		if ((ini->is_smcd ?
1911 		     smcd_lgr_match(lgr, ini->ism_dev[ini->ism_selected],
1912 				    &ini->ism_peer_gid[ini->ism_selected]) :
1913 		     smcr_lgr_match(lgr, ini->smcr_version,
1914 				    ini->peer_systemid,
1915 				    ini->peer_gid, ini->peer_mac, role,
1916 				    ini->ib_clcqpn, net)) &&
1917 		    !lgr->sync_err &&
1918 		    (ini->smcd_version == SMC_V2 ||
1919 		     lgr->vlan_id == ini->vlan_id) &&
1920 		    (role == SMC_CLNT || ini->is_smcd ||
1921 		    (lgr->conns_num < lgr->max_conns &&
1922 		      !bitmap_full(lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX)))) {
1923 			/* link group found */
1924 			ini->first_contact_local = 0;
1925 			conn->lgr = lgr;
1926 			rc = smc_lgr_register_conn(conn, false);
1927 			write_unlock_bh(&lgr->conns_lock);
1928 			if (!rc && delayed_work_pending(&lgr->free_work))
1929 				cancel_delayed_work(&lgr->free_work);
1930 			break;
1931 		}
1932 		write_unlock_bh(&lgr->conns_lock);
1933 	}
1934 	spin_unlock_bh(lgr_lock);
1935 	if (rc)
1936 		return rc;
1937 
1938 	if (role == SMC_CLNT && !ini->first_contact_peer &&
1939 	    ini->first_contact_local) {
1940 		/* Server reuses a link group, but Client wants to start
1941 		 * a new one
1942 		 * send out_of_sync decline, reason synchr. error
1943 		 */
1944 		return SMC_CLC_DECL_SYNCERR;
1945 	}
1946 
1947 create:
1948 	if (ini->first_contact_local) {
1949 		rc = smc_lgr_create(smc, ini);
1950 		if (rc)
1951 			goto out;
1952 		lgr = conn->lgr;
1953 		write_lock_bh(&lgr->conns_lock);
1954 		rc = smc_lgr_register_conn(conn, true);
1955 		write_unlock_bh(&lgr->conns_lock);
1956 		if (rc) {
1957 			smc_lgr_cleanup_early(lgr);
1958 			goto out;
1959 		}
1960 	}
1961 	smc_lgr_hold(conn->lgr); /* lgr_put in smc_conn_free() */
1962 	if (!conn->lgr->is_smcd)
1963 		smcr_link_hold(conn->lnk); /* link_put in smc_conn_free() */
1964 	conn->freed = 0;
1965 	conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
1966 	conn->local_tx_ctrl.len = SMC_WR_TX_SIZE;
1967 	conn->urg_state = SMC_URG_READ;
1968 	init_waitqueue_head(&conn->cdc_pend_tx_wq);
1969 	INIT_WORK(&smc->conn.abort_work, smc_conn_abort_work);
1970 	if (ini->is_smcd) {
1971 		conn->rx_off = sizeof(struct smcd_cdc_msg);
1972 		smcd_cdc_rx_init(conn); /* init tasklet for this conn */
1973 	} else {
1974 		conn->rx_off = 0;
1975 	}
1976 #ifndef KERNEL_HAS_ATOMIC64
1977 	spin_lock_init(&conn->acurs_lock);
1978 #endif
1979 
1980 out:
1981 	return rc;
1982 }
1983 
1984 #define SMCD_DMBE_SIZES		6 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */
1985 #define SMCR_RMBE_SIZES		5 /* 0 -> 16KB, 1 -> 32KB, .. 5 -> 512KB */
1986 
1987 /* convert the RMB size into the compressed notation (minimum 16K, see
1988  * SMCD/R_DMBE_SIZES.
1989  * In contrast to plain ilog2, this rounds towards the next power of 2,
1990  * so the socket application gets at least its desired sndbuf / rcvbuf size.
1991  */
1992 static u8 smc_compress_bufsize(int size, bool is_smcd, bool is_rmb)
1993 {
1994 	const unsigned int max_scat = SG_MAX_SINGLE_ALLOC * PAGE_SIZE;
1995 	u8 compressed;
1996 
1997 	if (size <= SMC_BUF_MIN_SIZE)
1998 		return 0;
1999 
2000 	size = (size - 1) >> 14;  /* convert to 16K multiple */
2001 	compressed = min_t(u8, ilog2(size) + 1,
2002 			   is_smcd ? SMCD_DMBE_SIZES : SMCR_RMBE_SIZES);
2003 
2004 	if (!is_smcd && is_rmb)
2005 		/* RMBs are backed by & limited to max size of scatterlists */
2006 		compressed = min_t(u8, compressed, ilog2(max_scat >> 14));
2007 
2008 	return compressed;
2009 }
2010 
2011 /* convert the RMB size from compressed notation into integer */
2012 int smc_uncompress_bufsize(u8 compressed)
2013 {
2014 	u32 size;
2015 
2016 	size = 0x00000001 << (((int)compressed) + 14);
2017 	return (int)size;
2018 }
2019 
2020 /* try to reuse a sndbuf or rmb description slot for a certain
2021  * buffer size; if not available, return NULL
2022  */
2023 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize,
2024 					     struct rw_semaphore *lock,
2025 					     struct list_head *buf_list)
2026 {
2027 	struct smc_buf_desc *buf_slot;
2028 
2029 	down_read(lock);
2030 	list_for_each_entry(buf_slot, buf_list, list) {
2031 		if (cmpxchg(&buf_slot->used, 0, 1) == 0) {
2032 			up_read(lock);
2033 			return buf_slot;
2034 		}
2035 	}
2036 	up_read(lock);
2037 	return NULL;
2038 }
2039 
2040 /* one of the conditions for announcing a receiver's current window size is
2041  * that it "results in a minimum increase in the window size of 10% of the
2042  * receive buffer space" [RFC7609]
2043  */
2044 static inline int smc_rmb_wnd_update_limit(int rmbe_size)
2045 {
2046 	return max_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
2047 }
2048 
2049 /* map an buf to a link */
2050 static int smcr_buf_map_link(struct smc_buf_desc *buf_desc, bool is_rmb,
2051 			     struct smc_link *lnk)
2052 {
2053 	int rc, i, nents, offset, buf_size, size, access_flags;
2054 	struct scatterlist *sg;
2055 	void *buf;
2056 
2057 	if (buf_desc->is_map_ib[lnk->link_idx])
2058 		return 0;
2059 
2060 	if (buf_desc->is_vm) {
2061 		buf = buf_desc->cpu_addr;
2062 		buf_size = buf_desc->len;
2063 		offset = offset_in_page(buf_desc->cpu_addr);
2064 		nents = PAGE_ALIGN(buf_size + offset) / PAGE_SIZE;
2065 	} else {
2066 		nents = 1;
2067 	}
2068 
2069 	rc = sg_alloc_table(&buf_desc->sgt[lnk->link_idx], nents, GFP_KERNEL);
2070 	if (rc)
2071 		return rc;
2072 
2073 	if (buf_desc->is_vm) {
2074 		/* virtually contiguous buffer */
2075 		for_each_sg(buf_desc->sgt[lnk->link_idx].sgl, sg, nents, i) {
2076 			size = min_t(int, PAGE_SIZE - offset, buf_size);
2077 			sg_set_page(sg, vmalloc_to_page(buf), size, offset);
2078 			buf += size / sizeof(*buf);
2079 			buf_size -= size;
2080 			offset = 0;
2081 		}
2082 	} else {
2083 		/* physically contiguous buffer */
2084 		sg_set_buf(buf_desc->sgt[lnk->link_idx].sgl,
2085 			   buf_desc->cpu_addr, buf_desc->len);
2086 	}
2087 
2088 	/* map sg table to DMA address */
2089 	rc = smc_ib_buf_map_sg(lnk, buf_desc,
2090 			       is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2091 	/* SMC protocol depends on mapping to one DMA address only */
2092 	if (rc != nents) {
2093 		rc = -EAGAIN;
2094 		goto free_table;
2095 	}
2096 
2097 	buf_desc->is_dma_need_sync |=
2098 		smc_ib_is_sg_need_sync(lnk, buf_desc) << lnk->link_idx;
2099 
2100 	if (is_rmb || buf_desc->is_vm) {
2101 		/* create a new memory region for the RMB or vzalloced sndbuf */
2102 		access_flags = is_rmb ?
2103 			       IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
2104 			       IB_ACCESS_LOCAL_WRITE;
2105 
2106 		rc = smc_ib_get_memory_region(lnk->roce_pd, access_flags,
2107 					      buf_desc, lnk->link_idx);
2108 		if (rc)
2109 			goto buf_unmap;
2110 		smc_ib_sync_sg_for_device(lnk, buf_desc,
2111 					  is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2112 	}
2113 	buf_desc->is_map_ib[lnk->link_idx] = true;
2114 	return 0;
2115 
2116 buf_unmap:
2117 	smc_ib_buf_unmap_sg(lnk, buf_desc,
2118 			    is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
2119 free_table:
2120 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
2121 	return rc;
2122 }
2123 
2124 /* register a new buf on IB device, rmb or vzalloced sndbuf
2125  * must be called under lgr->llc_conf_mutex lock
2126  */
2127 int smcr_link_reg_buf(struct smc_link *link, struct smc_buf_desc *buf_desc)
2128 {
2129 	if (list_empty(&link->lgr->list))
2130 		return -ENOLINK;
2131 	if (!buf_desc->is_reg_mr[link->link_idx]) {
2132 		/* register memory region for new buf */
2133 		if (buf_desc->is_vm)
2134 			buf_desc->mr[link->link_idx]->iova =
2135 				(uintptr_t)buf_desc->cpu_addr;
2136 		if (smc_wr_reg_send(link, buf_desc->mr[link->link_idx])) {
2137 			buf_desc->is_reg_err = true;
2138 			return -EFAULT;
2139 		}
2140 		buf_desc->is_reg_mr[link->link_idx] = true;
2141 	}
2142 	return 0;
2143 }
2144 
2145 static int _smcr_buf_map_lgr(struct smc_link *lnk, struct rw_semaphore *lock,
2146 			     struct list_head *lst, bool is_rmb)
2147 {
2148 	struct smc_buf_desc *buf_desc, *bf;
2149 	int rc = 0;
2150 
2151 	down_write(lock);
2152 	list_for_each_entry_safe(buf_desc, bf, lst, list) {
2153 		if (!buf_desc->used)
2154 			continue;
2155 		rc = smcr_buf_map_link(buf_desc, is_rmb, lnk);
2156 		if (rc)
2157 			goto out;
2158 	}
2159 out:
2160 	up_write(lock);
2161 	return rc;
2162 }
2163 
2164 /* map all used buffers of lgr for a new link */
2165 int smcr_buf_map_lgr(struct smc_link *lnk)
2166 {
2167 	struct smc_link_group *lgr = lnk->lgr;
2168 	int i, rc = 0;
2169 
2170 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2171 		rc = _smcr_buf_map_lgr(lnk, &lgr->rmbs_lock,
2172 				       &lgr->rmbs[i], true);
2173 		if (rc)
2174 			return rc;
2175 		rc = _smcr_buf_map_lgr(lnk, &lgr->sndbufs_lock,
2176 				       &lgr->sndbufs[i], false);
2177 		if (rc)
2178 			return rc;
2179 	}
2180 	return 0;
2181 }
2182 
2183 /* register all used buffers of lgr for a new link,
2184  * must be called under lgr->llc_conf_mutex lock
2185  */
2186 int smcr_buf_reg_lgr(struct smc_link *lnk)
2187 {
2188 	struct smc_link_group *lgr = lnk->lgr;
2189 	struct smc_buf_desc *buf_desc, *bf;
2190 	int i, rc = 0;
2191 
2192 	/* reg all RMBs for a new link */
2193 	down_write(&lgr->rmbs_lock);
2194 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2195 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) {
2196 			if (!buf_desc->used)
2197 				continue;
2198 			rc = smcr_link_reg_buf(lnk, buf_desc);
2199 			if (rc) {
2200 				up_write(&lgr->rmbs_lock);
2201 				return rc;
2202 			}
2203 		}
2204 	}
2205 	up_write(&lgr->rmbs_lock);
2206 
2207 	if (lgr->buf_type == SMCR_PHYS_CONT_BUFS)
2208 		return rc;
2209 
2210 	/* reg all vzalloced sndbufs for a new link */
2211 	down_write(&lgr->sndbufs_lock);
2212 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2213 		list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i], list) {
2214 			if (!buf_desc->used || !buf_desc->is_vm)
2215 				continue;
2216 			rc = smcr_link_reg_buf(lnk, buf_desc);
2217 			if (rc) {
2218 				up_write(&lgr->sndbufs_lock);
2219 				return rc;
2220 			}
2221 		}
2222 	}
2223 	up_write(&lgr->sndbufs_lock);
2224 	return rc;
2225 }
2226 
2227 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr,
2228 						bool is_rmb, int bufsize)
2229 {
2230 	struct smc_buf_desc *buf_desc;
2231 
2232 	/* try to alloc a new buffer */
2233 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2234 	if (!buf_desc)
2235 		return ERR_PTR(-ENOMEM);
2236 
2237 	switch (lgr->buf_type) {
2238 	case SMCR_PHYS_CONT_BUFS:
2239 	case SMCR_MIXED_BUFS:
2240 		buf_desc->order = get_order(bufsize);
2241 		buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN |
2242 					      __GFP_NOMEMALLOC | __GFP_COMP |
2243 					      __GFP_NORETRY | __GFP_ZERO,
2244 					      buf_desc->order);
2245 		if (buf_desc->pages) {
2246 			buf_desc->cpu_addr =
2247 				(void *)page_address(buf_desc->pages);
2248 			buf_desc->len = bufsize;
2249 			buf_desc->is_vm = false;
2250 			break;
2251 		}
2252 		if (lgr->buf_type == SMCR_PHYS_CONT_BUFS)
2253 			goto out;
2254 		fallthrough;	// try virtually continguous buf
2255 	case SMCR_VIRT_CONT_BUFS:
2256 		buf_desc->order = get_order(bufsize);
2257 		buf_desc->cpu_addr = vzalloc(PAGE_SIZE << buf_desc->order);
2258 		if (!buf_desc->cpu_addr)
2259 			goto out;
2260 		buf_desc->pages = NULL;
2261 		buf_desc->len = bufsize;
2262 		buf_desc->is_vm = true;
2263 		break;
2264 	}
2265 	return buf_desc;
2266 
2267 out:
2268 	kfree(buf_desc);
2269 	return ERR_PTR(-EAGAIN);
2270 }
2271 
2272 /* map buf_desc on all usable links,
2273  * unused buffers stay mapped as long as the link is up
2274  */
2275 static int smcr_buf_map_usable_links(struct smc_link_group *lgr,
2276 				     struct smc_buf_desc *buf_desc, bool is_rmb)
2277 {
2278 	int i, rc = 0, cnt = 0;
2279 
2280 	/* protect against parallel link reconfiguration */
2281 	down_read(&lgr->llc_conf_mutex);
2282 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2283 		struct smc_link *lnk = &lgr->lnk[i];
2284 
2285 		if (!smc_link_usable(lnk))
2286 			continue;
2287 		if (smcr_buf_map_link(buf_desc, is_rmb, lnk)) {
2288 			rc = -ENOMEM;
2289 			goto out;
2290 		}
2291 		cnt++;
2292 	}
2293 out:
2294 	up_read(&lgr->llc_conf_mutex);
2295 	if (!rc && !cnt)
2296 		rc = -EINVAL;
2297 	return rc;
2298 }
2299 
2300 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr,
2301 						bool is_dmb, int bufsize)
2302 {
2303 	struct smc_buf_desc *buf_desc;
2304 	int rc;
2305 
2306 	/* try to alloc a new DMB */
2307 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2308 	if (!buf_desc)
2309 		return ERR_PTR(-ENOMEM);
2310 	if (is_dmb) {
2311 		rc = smc_ism_register_dmb(lgr, bufsize, buf_desc);
2312 		if (rc) {
2313 			kfree(buf_desc);
2314 			if (rc == -ENOMEM)
2315 				return ERR_PTR(-EAGAIN);
2316 			if (rc == -ENOSPC)
2317 				return ERR_PTR(-ENOSPC);
2318 			return ERR_PTR(-EIO);
2319 		}
2320 		buf_desc->pages = virt_to_page(buf_desc->cpu_addr);
2321 		/* CDC header stored in buf. So, pretend it was smaller */
2322 		buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg);
2323 	} else {
2324 		buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL |
2325 					     __GFP_NOWARN | __GFP_NORETRY |
2326 					     __GFP_NOMEMALLOC);
2327 		if (!buf_desc->cpu_addr) {
2328 			kfree(buf_desc);
2329 			return ERR_PTR(-EAGAIN);
2330 		}
2331 		buf_desc->len = bufsize;
2332 	}
2333 	return buf_desc;
2334 }
2335 
2336 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb)
2337 {
2338 	struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM);
2339 	struct smc_connection *conn = &smc->conn;
2340 	struct smc_link_group *lgr = conn->lgr;
2341 	struct list_head *buf_list;
2342 	int bufsize, bufsize_comp;
2343 	struct rw_semaphore *lock;	/* lock buffer list */
2344 	bool is_dgraded = false;
2345 
2346 	if (is_rmb)
2347 		/* use socket recv buffer size (w/o overhead) as start value */
2348 		bufsize = smc->sk.sk_rcvbuf / 2;
2349 	else
2350 		/* use socket send buffer size (w/o overhead) as start value */
2351 		bufsize = smc->sk.sk_sndbuf / 2;
2352 
2353 	for (bufsize_comp = smc_compress_bufsize(bufsize, is_smcd, is_rmb);
2354 	     bufsize_comp >= 0; bufsize_comp--) {
2355 		if (is_rmb) {
2356 			lock = &lgr->rmbs_lock;
2357 			buf_list = &lgr->rmbs[bufsize_comp];
2358 		} else {
2359 			lock = &lgr->sndbufs_lock;
2360 			buf_list = &lgr->sndbufs[bufsize_comp];
2361 		}
2362 		bufsize = smc_uncompress_bufsize(bufsize_comp);
2363 
2364 		/* check for reusable slot in the link group */
2365 		buf_desc = smc_buf_get_slot(bufsize_comp, lock, buf_list);
2366 		if (buf_desc) {
2367 			buf_desc->is_dma_need_sync = 0;
2368 			SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2369 			SMC_STAT_BUF_REUSE(smc, is_smcd, is_rmb);
2370 			break; /* found reusable slot */
2371 		}
2372 
2373 		if (is_smcd)
2374 			buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize);
2375 		else
2376 			buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize);
2377 
2378 		if (PTR_ERR(buf_desc) == -ENOMEM)
2379 			break;
2380 		if (IS_ERR(buf_desc)) {
2381 			if (!is_dgraded) {
2382 				is_dgraded = true;
2383 				SMC_STAT_RMB_DOWNGRADED(smc, is_smcd, is_rmb);
2384 			}
2385 			continue;
2386 		}
2387 
2388 		SMC_STAT_RMB_ALLOC(smc, is_smcd, is_rmb);
2389 		SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2390 		buf_desc->used = 1;
2391 		down_write(lock);
2392 		list_add(&buf_desc->list, buf_list);
2393 		up_write(lock);
2394 		break; /* found */
2395 	}
2396 
2397 	if (IS_ERR(buf_desc))
2398 		return PTR_ERR(buf_desc);
2399 
2400 	if (!is_smcd) {
2401 		if (smcr_buf_map_usable_links(lgr, buf_desc, is_rmb)) {
2402 			smcr_buf_unuse(buf_desc, is_rmb, lgr);
2403 			return -ENOMEM;
2404 		}
2405 	}
2406 
2407 	if (is_rmb) {
2408 		conn->rmb_desc = buf_desc;
2409 		conn->rmbe_size_comp = bufsize_comp;
2410 		smc->sk.sk_rcvbuf = bufsize * 2;
2411 		atomic_set(&conn->bytes_to_rcv, 0);
2412 		conn->rmbe_update_limit =
2413 			smc_rmb_wnd_update_limit(buf_desc->len);
2414 		if (is_smcd)
2415 			smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */
2416 	} else {
2417 		conn->sndbuf_desc = buf_desc;
2418 		smc->sk.sk_sndbuf = bufsize * 2;
2419 		atomic_set(&conn->sndbuf_space, bufsize);
2420 	}
2421 	return 0;
2422 }
2423 
2424 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn)
2425 {
2426 	if (!conn->sndbuf_desc->is_dma_need_sync)
2427 		return;
2428 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd ||
2429 	    !smc_link_active(conn->lnk))
2430 		return;
2431 	smc_ib_sync_sg_for_device(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2432 }
2433 
2434 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn)
2435 {
2436 	int i;
2437 
2438 	if (!conn->rmb_desc->is_dma_need_sync)
2439 		return;
2440 	if (!smc_conn_lgr_valid(conn) || conn->lgr->is_smcd)
2441 		return;
2442 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2443 		if (!smc_link_active(&conn->lgr->lnk[i]))
2444 			continue;
2445 		smc_ib_sync_sg_for_cpu(&conn->lgr->lnk[i], conn->rmb_desc,
2446 				       DMA_FROM_DEVICE);
2447 	}
2448 }
2449 
2450 /* create the send and receive buffer for an SMC socket;
2451  * receive buffers are called RMBs;
2452  * (even though the SMC protocol allows more than one RMB-element per RMB,
2453  * the Linux implementation uses just one RMB-element per RMB, i.e. uses an
2454  * extra RMB for every connection in a link group
2455  */
2456 int smc_buf_create(struct smc_sock *smc, bool is_smcd)
2457 {
2458 	int rc;
2459 
2460 	/* create send buffer */
2461 	rc = __smc_buf_create(smc, is_smcd, false);
2462 	if (rc)
2463 		return rc;
2464 	/* create rmb */
2465 	rc = __smc_buf_create(smc, is_smcd, true);
2466 	if (rc) {
2467 		down_write(&smc->conn.lgr->sndbufs_lock);
2468 		list_del(&smc->conn.sndbuf_desc->list);
2469 		up_write(&smc->conn.lgr->sndbufs_lock);
2470 		smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc);
2471 		smc->conn.sndbuf_desc = NULL;
2472 	}
2473 	return rc;
2474 }
2475 
2476 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
2477 {
2478 	int i;
2479 
2480 	for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
2481 		if (!test_and_set_bit(i, lgr->rtokens_used_mask))
2482 			return i;
2483 	}
2484 	return -ENOSPC;
2485 }
2486 
2487 static int smc_rtoken_find_by_link(struct smc_link_group *lgr, int lnk_idx,
2488 				   u32 rkey)
2489 {
2490 	int i;
2491 
2492 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2493 		if (test_bit(i, lgr->rtokens_used_mask) &&
2494 		    lgr->rtokens[i][lnk_idx].rkey == rkey)
2495 			return i;
2496 	}
2497 	return -ENOENT;
2498 }
2499 
2500 /* set rtoken for a new link to an existing rmb */
2501 void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new,
2502 		    __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey)
2503 {
2504 	int rtok_idx;
2505 
2506 	rtok_idx = smc_rtoken_find_by_link(lgr, link_idx, ntohl(nw_rkey_known));
2507 	if (rtok_idx == -ENOENT)
2508 		return;
2509 	lgr->rtokens[rtok_idx][link_idx_new].rkey = ntohl(nw_rkey);
2510 	lgr->rtokens[rtok_idx][link_idx_new].dma_addr = be64_to_cpu(nw_vaddr);
2511 }
2512 
2513 /* set rtoken for a new link whose link_id is given */
2514 void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id,
2515 		     __be64 nw_vaddr, __be32 nw_rkey)
2516 {
2517 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2518 	u32 rkey = ntohl(nw_rkey);
2519 	bool found = false;
2520 	int link_idx;
2521 
2522 	for (link_idx = 0; link_idx < SMC_LINKS_PER_LGR_MAX; link_idx++) {
2523 		if (lgr->lnk[link_idx].link_id == link_id) {
2524 			found = true;
2525 			break;
2526 		}
2527 	}
2528 	if (!found)
2529 		return;
2530 	lgr->rtokens[rtok_idx][link_idx].rkey = rkey;
2531 	lgr->rtokens[rtok_idx][link_idx].dma_addr = dma_addr;
2532 }
2533 
2534 /* add a new rtoken from peer */
2535 int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey)
2536 {
2537 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2538 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2539 	u32 rkey = ntohl(nw_rkey);
2540 	int i;
2541 
2542 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2543 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2544 		    lgr->rtokens[i][lnk->link_idx].dma_addr == dma_addr &&
2545 		    test_bit(i, lgr->rtokens_used_mask)) {
2546 			/* already in list */
2547 			return i;
2548 		}
2549 	}
2550 	i = smc_rmb_reserve_rtoken_idx(lgr);
2551 	if (i < 0)
2552 		return i;
2553 	lgr->rtokens[i][lnk->link_idx].rkey = rkey;
2554 	lgr->rtokens[i][lnk->link_idx].dma_addr = dma_addr;
2555 	return i;
2556 }
2557 
2558 /* delete an rtoken from all links */
2559 int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey)
2560 {
2561 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2562 	u32 rkey = ntohl(nw_rkey);
2563 	int i, j;
2564 
2565 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2566 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2567 		    test_bit(i, lgr->rtokens_used_mask)) {
2568 			for (j = 0; j < SMC_LINKS_PER_LGR_MAX; j++) {
2569 				lgr->rtokens[i][j].rkey = 0;
2570 				lgr->rtokens[i][j].dma_addr = 0;
2571 			}
2572 			clear_bit(i, lgr->rtokens_used_mask);
2573 			return 0;
2574 		}
2575 	}
2576 	return -ENOENT;
2577 }
2578 
2579 /* save rkey and dma_addr received from peer during clc handshake */
2580 int smc_rmb_rtoken_handling(struct smc_connection *conn,
2581 			    struct smc_link *lnk,
2582 			    struct smc_clc_msg_accept_confirm *clc)
2583 {
2584 	conn->rtoken_idx = smc_rtoken_add(lnk, clc->r0.rmb_dma_addr,
2585 					  clc->r0.rmb_rkey);
2586 	if (conn->rtoken_idx < 0)
2587 		return conn->rtoken_idx;
2588 	return 0;
2589 }
2590 
2591 static void smc_core_going_away(void)
2592 {
2593 	struct smc_ib_device *smcibdev;
2594 	struct smcd_dev *smcd;
2595 
2596 	mutex_lock(&smc_ib_devices.mutex);
2597 	list_for_each_entry(smcibdev, &smc_ib_devices.list, list) {
2598 		int i;
2599 
2600 		for (i = 0; i < SMC_MAX_PORTS; i++)
2601 			set_bit(i, smcibdev->ports_going_away);
2602 	}
2603 	mutex_unlock(&smc_ib_devices.mutex);
2604 
2605 	mutex_lock(&smcd_dev_list.mutex);
2606 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2607 		smcd->going_away = 1;
2608 	}
2609 	mutex_unlock(&smcd_dev_list.mutex);
2610 }
2611 
2612 /* Clean up all SMC link groups */
2613 static void smc_lgrs_shutdown(void)
2614 {
2615 	struct smcd_dev *smcd;
2616 
2617 	smc_core_going_away();
2618 
2619 	smc_smcr_terminate_all(NULL);
2620 
2621 	mutex_lock(&smcd_dev_list.mutex);
2622 	list_for_each_entry(smcd, &smcd_dev_list.list, list)
2623 		smc_smcd_terminate_all(smcd);
2624 	mutex_unlock(&smcd_dev_list.mutex);
2625 }
2626 
2627 static int smc_core_reboot_event(struct notifier_block *this,
2628 				 unsigned long event, void *ptr)
2629 {
2630 	smc_lgrs_shutdown();
2631 	smc_ib_unregister_client();
2632 	smc_ism_exit();
2633 	return 0;
2634 }
2635 
2636 static struct notifier_block smc_reboot_notifier = {
2637 	.notifier_call = smc_core_reboot_event,
2638 };
2639 
2640 int __init smc_core_init(void)
2641 {
2642 	return register_reboot_notifier(&smc_reboot_notifier);
2643 }
2644 
2645 /* Called (from smc_exit) when module is removed */
2646 void smc_core_exit(void)
2647 {
2648 	unregister_reboot_notifier(&smc_reboot_notifier);
2649 	smc_lgrs_shutdown();
2650 }
2651