xref: /openbmc/linux/net/smc/smc_core.c (revision 50dc9a85)
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 
38 #define SMC_LGR_NUM_INCR		256
39 #define SMC_LGR_FREE_DELAY_SERV		(600 * HZ)
40 #define SMC_LGR_FREE_DELAY_CLNT		(SMC_LGR_FREE_DELAY_SERV + 10 * HZ)
41 
42 struct smc_lgr_list smc_lgr_list = {	/* established link groups */
43 	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
44 	.list = LIST_HEAD_INIT(smc_lgr_list.list),
45 	.num = 0,
46 };
47 
48 static atomic_t lgr_cnt = ATOMIC_INIT(0); /* number of existing link groups */
49 static DECLARE_WAIT_QUEUE_HEAD(lgrs_deleted);
50 
51 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
52 			 struct smc_buf_desc *buf_desc);
53 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft);
54 
55 static void smc_link_down_work(struct work_struct *work);
56 
57 /* return head of link group list and its lock for a given link group */
58 static inline struct list_head *smc_lgr_list_head(struct smc_link_group *lgr,
59 						  spinlock_t **lgr_lock)
60 {
61 	if (lgr->is_smcd) {
62 		*lgr_lock = &lgr->smcd->lgr_lock;
63 		return &lgr->smcd->lgr_list;
64 	}
65 
66 	*lgr_lock = &smc_lgr_list.lock;
67 	return &smc_lgr_list.list;
68 }
69 
70 static void smc_ibdev_cnt_inc(struct smc_link *lnk)
71 {
72 	atomic_inc(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
73 }
74 
75 static void smc_ibdev_cnt_dec(struct smc_link *lnk)
76 {
77 	atomic_dec(&lnk->smcibdev->lnk_cnt_by_port[lnk->ibport - 1]);
78 }
79 
80 static void smc_lgr_schedule_free_work(struct smc_link_group *lgr)
81 {
82 	/* client link group creation always follows the server link group
83 	 * creation. For client use a somewhat higher removal delay time,
84 	 * otherwise there is a risk of out-of-sync link groups.
85 	 */
86 	if (!lgr->freeing) {
87 		mod_delayed_work(system_wq, &lgr->free_work,
88 				 (!lgr->is_smcd && lgr->role == SMC_CLNT) ?
89 						SMC_LGR_FREE_DELAY_CLNT :
90 						SMC_LGR_FREE_DELAY_SERV);
91 	}
92 }
93 
94 /* Register connection's alert token in our lookup structure.
95  * To use rbtrees we have to implement our own insert core.
96  * Requires @conns_lock
97  * @smc		connection to register
98  * Returns 0 on success, != otherwise.
99  */
100 static void smc_lgr_add_alert_token(struct smc_connection *conn)
101 {
102 	struct rb_node **link, *parent = NULL;
103 	u32 token = conn->alert_token_local;
104 
105 	link = &conn->lgr->conns_all.rb_node;
106 	while (*link) {
107 		struct smc_connection *cur = rb_entry(*link,
108 					struct smc_connection, alert_node);
109 
110 		parent = *link;
111 		if (cur->alert_token_local > token)
112 			link = &parent->rb_left;
113 		else
114 			link = &parent->rb_right;
115 	}
116 	/* Put the new node there */
117 	rb_link_node(&conn->alert_node, parent, link);
118 	rb_insert_color(&conn->alert_node, &conn->lgr->conns_all);
119 }
120 
121 /* assign an SMC-R link to the connection */
122 static int smcr_lgr_conn_assign_link(struct smc_connection *conn, bool first)
123 {
124 	enum smc_link_state expected = first ? SMC_LNK_ACTIVATING :
125 				       SMC_LNK_ACTIVE;
126 	int i, j;
127 
128 	/* do link balancing */
129 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
130 		struct smc_link *lnk = &conn->lgr->lnk[i];
131 
132 		if (lnk->state != expected || lnk->link_is_asym)
133 			continue;
134 		if (conn->lgr->role == SMC_CLNT) {
135 			conn->lnk = lnk; /* temporary, SMC server assigns link*/
136 			break;
137 		}
138 		if (conn->lgr->conns_num % 2) {
139 			for (j = i + 1; j < SMC_LINKS_PER_LGR_MAX; j++) {
140 				struct smc_link *lnk2;
141 
142 				lnk2 = &conn->lgr->lnk[j];
143 				if (lnk2->state == expected &&
144 				    !lnk2->link_is_asym) {
145 					conn->lnk = lnk2;
146 					break;
147 				}
148 			}
149 		}
150 		if (!conn->lnk)
151 			conn->lnk = lnk;
152 		break;
153 	}
154 	if (!conn->lnk)
155 		return SMC_CLC_DECL_NOACTLINK;
156 	atomic_inc(&conn->lnk->conn_cnt);
157 	return 0;
158 }
159 
160 /* Register connection in link group by assigning an alert token
161  * registered in a search tree.
162  * Requires @conns_lock
163  * Note that '0' is a reserved value and not assigned.
164  */
165 static int smc_lgr_register_conn(struct smc_connection *conn, bool first)
166 {
167 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
168 	static atomic_t nexttoken = ATOMIC_INIT(0);
169 	int rc;
170 
171 	if (!conn->lgr->is_smcd) {
172 		rc = smcr_lgr_conn_assign_link(conn, first);
173 		if (rc)
174 			return rc;
175 	}
176 	/* find a new alert_token_local value not yet used by some connection
177 	 * in this link group
178 	 */
179 	sock_hold(&smc->sk); /* sock_put in smc_lgr_unregister_conn() */
180 	while (!conn->alert_token_local) {
181 		conn->alert_token_local = atomic_inc_return(&nexttoken);
182 		if (smc_lgr_find_conn(conn->alert_token_local, conn->lgr))
183 			conn->alert_token_local = 0;
184 	}
185 	smc_lgr_add_alert_token(conn);
186 	conn->lgr->conns_num++;
187 	return 0;
188 }
189 
190 /* Unregister connection and reset the alert token of the given connection<
191  */
192 static void __smc_lgr_unregister_conn(struct smc_connection *conn)
193 {
194 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
195 	struct smc_link_group *lgr = conn->lgr;
196 
197 	rb_erase(&conn->alert_node, &lgr->conns_all);
198 	if (conn->lnk)
199 		atomic_dec(&conn->lnk->conn_cnt);
200 	lgr->conns_num--;
201 	conn->alert_token_local = 0;
202 	sock_put(&smc->sk); /* sock_hold in smc_lgr_register_conn() */
203 }
204 
205 /* Unregister connection from lgr
206  */
207 static void smc_lgr_unregister_conn(struct smc_connection *conn)
208 {
209 	struct smc_link_group *lgr = conn->lgr;
210 
211 	if (!lgr)
212 		return;
213 	write_lock_bh(&lgr->conns_lock);
214 	if (conn->alert_token_local) {
215 		__smc_lgr_unregister_conn(conn);
216 	}
217 	write_unlock_bh(&lgr->conns_lock);
218 	conn->lgr = NULL;
219 }
220 
221 int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb)
222 {
223 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
224 	char hostname[SMC_MAX_HOSTNAME_LEN + 1];
225 	char smc_seid[SMC_MAX_EID_LEN + 1];
226 	struct nlattr *attrs;
227 	u8 *seid = NULL;
228 	u8 *host = NULL;
229 	void *nlh;
230 
231 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
232 			  &smc_gen_nl_family, NLM_F_MULTI,
233 			  SMC_NETLINK_GET_SYS_INFO);
234 	if (!nlh)
235 		goto errmsg;
236 	if (cb_ctx->pos[0])
237 		goto errout;
238 	attrs = nla_nest_start(skb, SMC_GEN_SYS_INFO);
239 	if (!attrs)
240 		goto errout;
241 	if (nla_put_u8(skb, SMC_NLA_SYS_VER, SMC_V2))
242 		goto errattr;
243 	if (nla_put_u8(skb, SMC_NLA_SYS_REL, SMC_RELEASE))
244 		goto errattr;
245 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_ISM_V2, smc_ism_is_v2_capable()))
246 		goto errattr;
247 	if (nla_put_u8(skb, SMC_NLA_SYS_IS_SMCR_V2, true))
248 		goto errattr;
249 	smc_clc_get_hostname(&host);
250 	if (host) {
251 		memcpy(hostname, host, SMC_MAX_HOSTNAME_LEN);
252 		hostname[SMC_MAX_HOSTNAME_LEN] = 0;
253 		if (nla_put_string(skb, SMC_NLA_SYS_LOCAL_HOST, hostname))
254 			goto errattr;
255 	}
256 	if (smc_ism_is_v2_capable()) {
257 		smc_ism_get_system_eid(&seid);
258 		memcpy(smc_seid, seid, SMC_MAX_EID_LEN);
259 		smc_seid[SMC_MAX_EID_LEN] = 0;
260 		if (nla_put_string(skb, SMC_NLA_SYS_SEID, smc_seid))
261 			goto errattr;
262 	}
263 	nla_nest_end(skb, attrs);
264 	genlmsg_end(skb, nlh);
265 	cb_ctx->pos[0] = 1;
266 	return skb->len;
267 
268 errattr:
269 	nla_nest_cancel(skb, attrs);
270 errout:
271 	genlmsg_cancel(skb, nlh);
272 errmsg:
273 	return skb->len;
274 }
275 
276 /* Fill SMC_NLA_LGR_D_V2_COMMON/SMC_NLA_LGR_R_V2_COMMON nested attributes */
277 static int smc_nl_fill_lgr_v2_common(struct smc_link_group *lgr,
278 				     struct sk_buff *skb,
279 				     struct netlink_callback *cb,
280 				     struct nlattr *v2_attrs)
281 {
282 	char smc_host[SMC_MAX_HOSTNAME_LEN + 1];
283 	char smc_eid[SMC_MAX_EID_LEN + 1];
284 
285 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_VER, lgr->smc_version))
286 		goto errv2attr;
287 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_REL, lgr->peer_smc_release))
288 		goto errv2attr;
289 	if (nla_put_u8(skb, SMC_NLA_LGR_V2_OS, lgr->peer_os))
290 		goto errv2attr;
291 	memcpy(smc_host, lgr->peer_hostname, SMC_MAX_HOSTNAME_LEN);
292 	smc_host[SMC_MAX_HOSTNAME_LEN] = 0;
293 	if (nla_put_string(skb, SMC_NLA_LGR_V2_PEER_HOST, smc_host))
294 		goto errv2attr;
295 	memcpy(smc_eid, lgr->negotiated_eid, SMC_MAX_EID_LEN);
296 	smc_eid[SMC_MAX_EID_LEN] = 0;
297 	if (nla_put_string(skb, SMC_NLA_LGR_V2_NEG_EID, smc_eid))
298 		goto errv2attr;
299 
300 	nla_nest_end(skb, v2_attrs);
301 	return 0;
302 
303 errv2attr:
304 	nla_nest_cancel(skb, v2_attrs);
305 	return -EMSGSIZE;
306 }
307 
308 static int smc_nl_fill_smcr_lgr_v2(struct smc_link_group *lgr,
309 				   struct sk_buff *skb,
310 				   struct netlink_callback *cb)
311 {
312 	struct nlattr *v2_attrs;
313 
314 	v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2);
315 	if (!v2_attrs)
316 		goto errattr;
317 	if (nla_put_u8(skb, SMC_NLA_LGR_R_V2_DIRECT, !lgr->uses_gateway))
318 		goto errv2attr;
319 
320 	nla_nest_end(skb, v2_attrs);
321 	return 0;
322 
323 errv2attr:
324 	nla_nest_cancel(skb, v2_attrs);
325 errattr:
326 	return -EMSGSIZE;
327 }
328 
329 static int smc_nl_fill_lgr(struct smc_link_group *lgr,
330 			   struct sk_buff *skb,
331 			   struct netlink_callback *cb)
332 {
333 	char smc_target[SMC_MAX_PNETID_LEN + 1];
334 	struct nlattr *attrs, *v2_attrs;
335 
336 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCR);
337 	if (!attrs)
338 		goto errout;
339 
340 	if (nla_put_u32(skb, SMC_NLA_LGR_R_ID, *((u32 *)&lgr->id)))
341 		goto errattr;
342 	if (nla_put_u32(skb, SMC_NLA_LGR_R_CONNS_NUM, lgr->conns_num))
343 		goto errattr;
344 	if (nla_put_u8(skb, SMC_NLA_LGR_R_ROLE, lgr->role))
345 		goto errattr;
346 	if (nla_put_u8(skb, SMC_NLA_LGR_R_TYPE, lgr->type))
347 		goto errattr;
348 	if (nla_put_u8(skb, SMC_NLA_LGR_R_VLAN_ID, lgr->vlan_id))
349 		goto errattr;
350 	memcpy(smc_target, lgr->pnet_id, SMC_MAX_PNETID_LEN);
351 	smc_target[SMC_MAX_PNETID_LEN] = 0;
352 	if (nla_put_string(skb, SMC_NLA_LGR_R_PNETID, smc_target))
353 		goto errattr;
354 	if (lgr->smc_version > SMC_V1) {
355 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_R_V2_COMMON);
356 		if (!v2_attrs)
357 			goto errattr;
358 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
359 			goto errattr;
360 		if (smc_nl_fill_smcr_lgr_v2(lgr, skb, cb))
361 			goto errattr;
362 	}
363 
364 	nla_nest_end(skb, attrs);
365 	return 0;
366 errattr:
367 	nla_nest_cancel(skb, attrs);
368 errout:
369 	return -EMSGSIZE;
370 }
371 
372 static int smc_nl_fill_lgr_link(struct smc_link_group *lgr,
373 				struct smc_link *link,
374 				struct sk_buff *skb,
375 				struct netlink_callback *cb)
376 {
377 	char smc_ibname[IB_DEVICE_NAME_MAX];
378 	u8 smc_gid_target[41];
379 	struct nlattr *attrs;
380 	u32 link_uid = 0;
381 	void *nlh;
382 
383 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
384 			  &smc_gen_nl_family, NLM_F_MULTI,
385 			  SMC_NETLINK_GET_LINK_SMCR);
386 	if (!nlh)
387 		goto errmsg;
388 
389 	attrs = nla_nest_start(skb, SMC_GEN_LINK_SMCR);
390 	if (!attrs)
391 		goto errout;
392 
393 	if (nla_put_u8(skb, SMC_NLA_LINK_ID, link->link_id))
394 		goto errattr;
395 	if (nla_put_u32(skb, SMC_NLA_LINK_STATE, link->state))
396 		goto errattr;
397 	if (nla_put_u32(skb, SMC_NLA_LINK_CONN_CNT,
398 			atomic_read(&link->conn_cnt)))
399 		goto errattr;
400 	if (nla_put_u8(skb, SMC_NLA_LINK_IB_PORT, link->ibport))
401 		goto errattr;
402 	if (nla_put_u32(skb, SMC_NLA_LINK_NET_DEV, link->ndev_ifidx))
403 		goto errattr;
404 	snprintf(smc_ibname, sizeof(smc_ibname), "%s", link->ibname);
405 	if (nla_put_string(skb, SMC_NLA_LINK_IB_DEV, smc_ibname))
406 		goto errattr;
407 	memcpy(&link_uid, link->link_uid, sizeof(link_uid));
408 	if (nla_put_u32(skb, SMC_NLA_LINK_UID, link_uid))
409 		goto errattr;
410 	memcpy(&link_uid, link->peer_link_uid, sizeof(link_uid));
411 	if (nla_put_u32(skb, SMC_NLA_LINK_PEER_UID, link_uid))
412 		goto errattr;
413 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
414 	smc_gid_be16_convert(smc_gid_target, link->gid);
415 	if (nla_put_string(skb, SMC_NLA_LINK_GID, smc_gid_target))
416 		goto errattr;
417 	memset(smc_gid_target, 0, sizeof(smc_gid_target));
418 	smc_gid_be16_convert(smc_gid_target, link->peer_gid);
419 	if (nla_put_string(skb, SMC_NLA_LINK_PEER_GID, smc_gid_target))
420 		goto errattr;
421 
422 	nla_nest_end(skb, attrs);
423 	genlmsg_end(skb, nlh);
424 	return 0;
425 errattr:
426 	nla_nest_cancel(skb, attrs);
427 errout:
428 	genlmsg_cancel(skb, nlh);
429 errmsg:
430 	return -EMSGSIZE;
431 }
432 
433 static int smc_nl_handle_lgr(struct smc_link_group *lgr,
434 			     struct sk_buff *skb,
435 			     struct netlink_callback *cb,
436 			     bool list_links)
437 {
438 	void *nlh;
439 	int i;
440 
441 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
442 			  &smc_gen_nl_family, NLM_F_MULTI,
443 			  SMC_NETLINK_GET_LGR_SMCR);
444 	if (!nlh)
445 		goto errmsg;
446 	if (smc_nl_fill_lgr(lgr, skb, cb))
447 		goto errout;
448 
449 	genlmsg_end(skb, nlh);
450 	if (!list_links)
451 		goto out;
452 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
453 		if (!smc_link_usable(&lgr->lnk[i]))
454 			continue;
455 		if (smc_nl_fill_lgr_link(lgr, &lgr->lnk[i], skb, cb))
456 			goto errout;
457 	}
458 out:
459 	return 0;
460 
461 errout:
462 	genlmsg_cancel(skb, nlh);
463 errmsg:
464 	return -EMSGSIZE;
465 }
466 
467 static void smc_nl_fill_lgr_list(struct smc_lgr_list *smc_lgr,
468 				 struct sk_buff *skb,
469 				 struct netlink_callback *cb,
470 				 bool list_links)
471 {
472 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
473 	struct smc_link_group *lgr;
474 	int snum = cb_ctx->pos[0];
475 	int num = 0;
476 
477 	spin_lock_bh(&smc_lgr->lock);
478 	list_for_each_entry(lgr, &smc_lgr->list, list) {
479 		if (num < snum)
480 			goto next;
481 		if (smc_nl_handle_lgr(lgr, skb, cb, list_links))
482 			goto errout;
483 next:
484 		num++;
485 	}
486 errout:
487 	spin_unlock_bh(&smc_lgr->lock);
488 	cb_ctx->pos[0] = num;
489 }
490 
491 static int smc_nl_fill_smcd_lgr(struct smc_link_group *lgr,
492 				struct sk_buff *skb,
493 				struct netlink_callback *cb)
494 {
495 	char smc_pnet[SMC_MAX_PNETID_LEN + 1];
496 	struct nlattr *attrs;
497 	void *nlh;
498 
499 	nlh = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
500 			  &smc_gen_nl_family, NLM_F_MULTI,
501 			  SMC_NETLINK_GET_LGR_SMCD);
502 	if (!nlh)
503 		goto errmsg;
504 
505 	attrs = nla_nest_start(skb, SMC_GEN_LGR_SMCD);
506 	if (!attrs)
507 		goto errout;
508 
509 	if (nla_put_u32(skb, SMC_NLA_LGR_D_ID, *((u32 *)&lgr->id)))
510 		goto errattr;
511 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_GID, lgr->smcd->local_gid,
512 			      SMC_NLA_LGR_D_PAD))
513 		goto errattr;
514 	if (nla_put_u64_64bit(skb, SMC_NLA_LGR_D_PEER_GID, lgr->peer_gid,
515 			      SMC_NLA_LGR_D_PAD))
516 		goto errattr;
517 	if (nla_put_u8(skb, SMC_NLA_LGR_D_VLAN_ID, lgr->vlan_id))
518 		goto errattr;
519 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CONNS_NUM, lgr->conns_num))
520 		goto errattr;
521 	if (nla_put_u32(skb, SMC_NLA_LGR_D_CHID, smc_ism_get_chid(lgr->smcd)))
522 		goto errattr;
523 	memcpy(smc_pnet, lgr->smcd->pnetid, SMC_MAX_PNETID_LEN);
524 	smc_pnet[SMC_MAX_PNETID_LEN] = 0;
525 	if (nla_put_string(skb, SMC_NLA_LGR_D_PNETID, smc_pnet))
526 		goto errattr;
527 	if (lgr->smc_version > SMC_V1) {
528 		struct nlattr *v2_attrs;
529 
530 		v2_attrs = nla_nest_start(skb, SMC_NLA_LGR_D_V2_COMMON);
531 		if (!v2_attrs)
532 			goto errattr;
533 		if (smc_nl_fill_lgr_v2_common(lgr, skb, cb, v2_attrs))
534 			goto errattr;
535 	}
536 	nla_nest_end(skb, attrs);
537 	genlmsg_end(skb, nlh);
538 	return 0;
539 
540 errattr:
541 	nla_nest_cancel(skb, attrs);
542 errout:
543 	genlmsg_cancel(skb, nlh);
544 errmsg:
545 	return -EMSGSIZE;
546 }
547 
548 static int smc_nl_handle_smcd_lgr(struct smcd_dev *dev,
549 				  struct sk_buff *skb,
550 				  struct netlink_callback *cb)
551 {
552 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
553 	struct smc_link_group *lgr;
554 	int snum = cb_ctx->pos[1];
555 	int rc = 0, num = 0;
556 
557 	spin_lock_bh(&dev->lgr_lock);
558 	list_for_each_entry(lgr, &dev->lgr_list, list) {
559 		if (!lgr->is_smcd)
560 			continue;
561 		if (num < snum)
562 			goto next;
563 		rc = smc_nl_fill_smcd_lgr(lgr, skb, cb);
564 		if (rc)
565 			goto errout;
566 next:
567 		num++;
568 	}
569 errout:
570 	spin_unlock_bh(&dev->lgr_lock);
571 	cb_ctx->pos[1] = num;
572 	return rc;
573 }
574 
575 static int smc_nl_fill_smcd_dev(struct smcd_dev_list *dev_list,
576 				struct sk_buff *skb,
577 				struct netlink_callback *cb)
578 {
579 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
580 	struct smcd_dev *smcd_dev;
581 	int snum = cb_ctx->pos[0];
582 	int rc = 0, num = 0;
583 
584 	mutex_lock(&dev_list->mutex);
585 	list_for_each_entry(smcd_dev, &dev_list->list, list) {
586 		if (list_empty(&smcd_dev->lgr_list))
587 			continue;
588 		if (num < snum)
589 			goto next;
590 		rc = smc_nl_handle_smcd_lgr(smcd_dev, skb, cb);
591 		if (rc)
592 			goto errout;
593 next:
594 		num++;
595 	}
596 errout:
597 	mutex_unlock(&dev_list->mutex);
598 	cb_ctx->pos[0] = num;
599 	return rc;
600 }
601 
602 int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
603 {
604 	bool list_links = false;
605 
606 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
607 	return skb->len;
608 }
609 
610 int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb)
611 {
612 	bool list_links = true;
613 
614 	smc_nl_fill_lgr_list(&smc_lgr_list, skb, cb, list_links);
615 	return skb->len;
616 }
617 
618 int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb)
619 {
620 	smc_nl_fill_smcd_dev(&smcd_dev_list, skb, cb);
621 	return skb->len;
622 }
623 
624 void smc_lgr_cleanup_early(struct smc_connection *conn)
625 {
626 	struct smc_link_group *lgr = conn->lgr;
627 	struct list_head *lgr_list;
628 	spinlock_t *lgr_lock;
629 
630 	if (!lgr)
631 		return;
632 
633 	smc_conn_free(conn);
634 	lgr_list = smc_lgr_list_head(lgr, &lgr_lock);
635 	spin_lock_bh(lgr_lock);
636 	/* do not use this link group for new connections */
637 	if (!list_empty(lgr_list))
638 		list_del_init(lgr_list);
639 	spin_unlock_bh(lgr_lock);
640 	__smc_lgr_terminate(lgr, true);
641 }
642 
643 static void smcr_lgr_link_deactivate_all(struct smc_link_group *lgr)
644 {
645 	int i;
646 
647 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
648 		struct smc_link *lnk = &lgr->lnk[i];
649 
650 		if (smc_link_usable(lnk))
651 			lnk->state = SMC_LNK_INACTIVE;
652 	}
653 	wake_up_all(&lgr->llc_msg_waiter);
654 	wake_up_all(&lgr->llc_flow_waiter);
655 }
656 
657 static void smc_lgr_free(struct smc_link_group *lgr);
658 
659 static void smc_lgr_free_work(struct work_struct *work)
660 {
661 	struct smc_link_group *lgr = container_of(to_delayed_work(work),
662 						  struct smc_link_group,
663 						  free_work);
664 	spinlock_t *lgr_lock;
665 	bool conns;
666 
667 	smc_lgr_list_head(lgr, &lgr_lock);
668 	spin_lock_bh(lgr_lock);
669 	if (lgr->freeing) {
670 		spin_unlock_bh(lgr_lock);
671 		return;
672 	}
673 	read_lock_bh(&lgr->conns_lock);
674 	conns = RB_EMPTY_ROOT(&lgr->conns_all);
675 	read_unlock_bh(&lgr->conns_lock);
676 	if (!conns) { /* number of lgr connections is no longer zero */
677 		spin_unlock_bh(lgr_lock);
678 		return;
679 	}
680 	list_del_init(&lgr->list); /* remove from smc_lgr_list */
681 	lgr->freeing = 1; /* this instance does the freeing, no new schedule */
682 	spin_unlock_bh(lgr_lock);
683 	cancel_delayed_work(&lgr->free_work);
684 
685 	if (!lgr->is_smcd && !lgr->terminating)
686 		smc_llc_send_link_delete_all(lgr, true,
687 					     SMC_LLC_DEL_PROG_INIT_TERM);
688 	if (lgr->is_smcd && !lgr->terminating)
689 		smc_ism_signal_shutdown(lgr);
690 	if (!lgr->is_smcd)
691 		smcr_lgr_link_deactivate_all(lgr);
692 	smc_lgr_free(lgr);
693 }
694 
695 static void smc_lgr_terminate_work(struct work_struct *work)
696 {
697 	struct smc_link_group *lgr = container_of(work, struct smc_link_group,
698 						  terminate_work);
699 
700 	__smc_lgr_terminate(lgr, true);
701 }
702 
703 /* return next unique link id for the lgr */
704 static u8 smcr_next_link_id(struct smc_link_group *lgr)
705 {
706 	u8 link_id;
707 	int i;
708 
709 	while (1) {
710 		link_id = ++lgr->next_link_id;
711 		if (!link_id)	/* skip zero as link_id */
712 			link_id = ++lgr->next_link_id;
713 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
714 			if (smc_link_usable(&lgr->lnk[i]) &&
715 			    lgr->lnk[i].link_id == link_id)
716 				continue;
717 		}
718 		break;
719 	}
720 	return link_id;
721 }
722 
723 static void smcr_copy_dev_info_to_link(struct smc_link *link)
724 {
725 	struct smc_ib_device *smcibdev = link->smcibdev;
726 
727 	snprintf(link->ibname, sizeof(link->ibname), "%s",
728 		 smcibdev->ibdev->name);
729 	link->ndev_ifidx = smcibdev->ndev_ifidx[link->ibport - 1];
730 }
731 
732 int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk,
733 		   u8 link_idx, struct smc_init_info *ini)
734 {
735 	struct smc_ib_device *smcibdev;
736 	u8 rndvec[3];
737 	int rc;
738 
739 	if (lgr->smc_version == SMC_V2) {
740 		lnk->smcibdev = ini->smcrv2.ib_dev_v2;
741 		lnk->ibport = ini->smcrv2.ib_port_v2;
742 	} else {
743 		lnk->smcibdev = ini->ib_dev;
744 		lnk->ibport = ini->ib_port;
745 	}
746 	get_device(&lnk->smcibdev->ibdev->dev);
747 	atomic_inc(&lnk->smcibdev->lnk_cnt);
748 	lnk->path_mtu = lnk->smcibdev->pattr[lnk->ibport - 1].active_mtu;
749 	lnk->link_id = smcr_next_link_id(lgr);
750 	lnk->lgr = lgr;
751 	lnk->link_idx = link_idx;
752 	smc_ibdev_cnt_inc(lnk);
753 	smcr_copy_dev_info_to_link(lnk);
754 	atomic_set(&lnk->conn_cnt, 0);
755 	smc_llc_link_set_uid(lnk);
756 	INIT_WORK(&lnk->link_down_wrk, smc_link_down_work);
757 	if (!lnk->smcibdev->initialized) {
758 		rc = (int)smc_ib_setup_per_ibdev(lnk->smcibdev);
759 		if (rc)
760 			goto out;
761 	}
762 	get_random_bytes(rndvec, sizeof(rndvec));
763 	lnk->psn_initial = rndvec[0] + (rndvec[1] << 8) +
764 		(rndvec[2] << 16);
765 	rc = smc_ib_determine_gid(lnk->smcibdev, lnk->ibport,
766 				  ini->vlan_id, lnk->gid, &lnk->sgid_index,
767 				  lgr->smc_version == SMC_V2 ?
768 						  &ini->smcrv2 : NULL);
769 	if (rc)
770 		goto out;
771 	rc = smc_llc_link_init(lnk);
772 	if (rc)
773 		goto out;
774 	rc = smc_wr_alloc_link_mem(lnk);
775 	if (rc)
776 		goto clear_llc_lnk;
777 	rc = smc_ib_create_protection_domain(lnk);
778 	if (rc)
779 		goto free_link_mem;
780 	rc = smc_ib_create_queue_pair(lnk);
781 	if (rc)
782 		goto dealloc_pd;
783 	rc = smc_wr_create_link(lnk);
784 	if (rc)
785 		goto destroy_qp;
786 	lnk->state = SMC_LNK_ACTIVATING;
787 	return 0;
788 
789 destroy_qp:
790 	smc_ib_destroy_queue_pair(lnk);
791 dealloc_pd:
792 	smc_ib_dealloc_protection_domain(lnk);
793 free_link_mem:
794 	smc_wr_free_link_mem(lnk);
795 clear_llc_lnk:
796 	smc_llc_link_clear(lnk, false);
797 out:
798 	smc_ibdev_cnt_dec(lnk);
799 	put_device(&lnk->smcibdev->ibdev->dev);
800 	smcibdev = lnk->smcibdev;
801 	memset(lnk, 0, sizeof(struct smc_link));
802 	lnk->state = SMC_LNK_UNUSED;
803 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
804 		wake_up(&smcibdev->lnks_deleted);
805 	return rc;
806 }
807 
808 /* create a new SMC link group */
809 static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)
810 {
811 	struct smc_link_group *lgr;
812 	struct list_head *lgr_list;
813 	struct smc_link *lnk;
814 	spinlock_t *lgr_lock;
815 	u8 link_idx;
816 	int rc = 0;
817 	int i;
818 
819 	if (ini->is_smcd && ini->vlan_id) {
820 		if (smc_ism_get_vlan(ini->ism_dev[ini->ism_selected],
821 				     ini->vlan_id)) {
822 			rc = SMC_CLC_DECL_ISMVLANERR;
823 			goto out;
824 		}
825 	}
826 
827 	lgr = kzalloc(sizeof(*lgr), GFP_KERNEL);
828 	if (!lgr) {
829 		rc = SMC_CLC_DECL_MEM;
830 		goto ism_put_vlan;
831 	}
832 	lgr->tx_wq = alloc_workqueue("smc_tx_wq-%*phN", 0, 0,
833 				     SMC_LGR_ID_SIZE, &lgr->id);
834 	if (!lgr->tx_wq) {
835 		rc = -ENOMEM;
836 		goto free_lgr;
837 	}
838 	lgr->is_smcd = ini->is_smcd;
839 	lgr->sync_err = 0;
840 	lgr->terminating = 0;
841 	lgr->freeing = 0;
842 	lgr->vlan_id = ini->vlan_id;
843 	mutex_init(&lgr->sndbufs_lock);
844 	mutex_init(&lgr->rmbs_lock);
845 	rwlock_init(&lgr->conns_lock);
846 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
847 		INIT_LIST_HEAD(&lgr->sndbufs[i]);
848 		INIT_LIST_HEAD(&lgr->rmbs[i]);
849 	}
850 	lgr->next_link_id = 0;
851 	smc_lgr_list.num += SMC_LGR_NUM_INCR;
852 	memcpy(&lgr->id, (u8 *)&smc_lgr_list.num, SMC_LGR_ID_SIZE);
853 	INIT_DELAYED_WORK(&lgr->free_work, smc_lgr_free_work);
854 	INIT_WORK(&lgr->terminate_work, smc_lgr_terminate_work);
855 	lgr->conns_all = RB_ROOT;
856 	if (ini->is_smcd) {
857 		/* SMC-D specific settings */
858 		get_device(&ini->ism_dev[ini->ism_selected]->dev);
859 		lgr->peer_gid = ini->ism_peer_gid[ini->ism_selected];
860 		lgr->smcd = ini->ism_dev[ini->ism_selected];
861 		lgr_list = &ini->ism_dev[ini->ism_selected]->lgr_list;
862 		lgr_lock = &lgr->smcd->lgr_lock;
863 		lgr->smc_version = ini->smcd_version;
864 		lgr->peer_shutdown = 0;
865 		atomic_inc(&ini->ism_dev[ini->ism_selected]->lgr_cnt);
866 	} else {
867 		/* SMC-R specific settings */
868 		struct smc_ib_device *ibdev;
869 		int ibport;
870 
871 		lgr->role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
872 		lgr->smc_version = ini->smcr_version;
873 		memcpy(lgr->peer_systemid, ini->peer_systemid,
874 		       SMC_SYSTEMID_LEN);
875 		if (lgr->smc_version == SMC_V2) {
876 			ibdev = ini->smcrv2.ib_dev_v2;
877 			ibport = ini->smcrv2.ib_port_v2;
878 			lgr->saddr = ini->smcrv2.saddr;
879 			lgr->uses_gateway = ini->smcrv2.uses_gateway;
880 			memcpy(lgr->nexthop_mac, ini->smcrv2.nexthop_mac,
881 			       ETH_ALEN);
882 		} else {
883 			ibdev = ini->ib_dev;
884 			ibport = ini->ib_port;
885 		}
886 		memcpy(lgr->pnet_id, ibdev->pnetid[ibport - 1],
887 		       SMC_MAX_PNETID_LEN);
888 		if (smc_wr_alloc_lgr_mem(lgr))
889 			goto free_wq;
890 		smc_llc_lgr_init(lgr, smc);
891 
892 		link_idx = SMC_SINGLE_LINK;
893 		lnk = &lgr->lnk[link_idx];
894 		rc = smcr_link_init(lgr, lnk, link_idx, ini);
895 		if (rc) {
896 			smc_wr_free_lgr_mem(lgr);
897 			goto free_wq;
898 		}
899 		lgr_list = &smc_lgr_list.list;
900 		lgr_lock = &smc_lgr_list.lock;
901 		atomic_inc(&lgr_cnt);
902 	}
903 	smc->conn.lgr = lgr;
904 	spin_lock_bh(lgr_lock);
905 	list_add_tail(&lgr->list, lgr_list);
906 	spin_unlock_bh(lgr_lock);
907 	return 0;
908 
909 free_wq:
910 	destroy_workqueue(lgr->tx_wq);
911 free_lgr:
912 	kfree(lgr);
913 ism_put_vlan:
914 	if (ini->is_smcd && ini->vlan_id)
915 		smc_ism_put_vlan(ini->ism_dev[ini->ism_selected], ini->vlan_id);
916 out:
917 	if (rc < 0) {
918 		if (rc == -ENOMEM)
919 			rc = SMC_CLC_DECL_MEM;
920 		else
921 			rc = SMC_CLC_DECL_INTERR;
922 	}
923 	return rc;
924 }
925 
926 static int smc_write_space(struct smc_connection *conn)
927 {
928 	int buffer_len = conn->peer_rmbe_size;
929 	union smc_host_cursor prod;
930 	union smc_host_cursor cons;
931 	int space;
932 
933 	smc_curs_copy(&prod, &conn->local_tx_ctrl.prod, conn);
934 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
935 	/* determine rx_buf space */
936 	space = buffer_len - smc_curs_diff(buffer_len, &cons, &prod);
937 	return space;
938 }
939 
940 static int smc_switch_cursor(struct smc_sock *smc, struct smc_cdc_tx_pend *pend,
941 			     struct smc_wr_buf *wr_buf)
942 {
943 	struct smc_connection *conn = &smc->conn;
944 	union smc_host_cursor cons, fin;
945 	int rc = 0;
946 	int diff;
947 
948 	smc_curs_copy(&conn->tx_curs_sent, &conn->tx_curs_fin, conn);
949 	smc_curs_copy(&fin, &conn->local_tx_ctrl_fin, conn);
950 	/* set prod cursor to old state, enforce tx_rdma_writes() */
951 	smc_curs_copy(&conn->local_tx_ctrl.prod, &fin, conn);
952 	smc_curs_copy(&cons, &conn->local_rx_ctrl.cons, conn);
953 
954 	if (smc_curs_comp(conn->peer_rmbe_size, &cons, &fin) < 0) {
955 		/* cons cursor advanced more than fin, and prod was set
956 		 * fin above, so now prod is smaller than cons. Fix that.
957 		 */
958 		diff = smc_curs_diff(conn->peer_rmbe_size, &fin, &cons);
959 		smc_curs_add(conn->sndbuf_desc->len,
960 			     &conn->tx_curs_sent, diff);
961 		smc_curs_add(conn->sndbuf_desc->len,
962 			     &conn->tx_curs_fin, diff);
963 
964 		smp_mb__before_atomic();
965 		atomic_add(diff, &conn->sndbuf_space);
966 		smp_mb__after_atomic();
967 
968 		smc_curs_add(conn->peer_rmbe_size,
969 			     &conn->local_tx_ctrl.prod, diff);
970 		smc_curs_add(conn->peer_rmbe_size,
971 			     &conn->local_tx_ctrl_fin, diff);
972 	}
973 	/* recalculate, value is used by tx_rdma_writes() */
974 	atomic_set(&smc->conn.peer_rmbe_space, smc_write_space(conn));
975 
976 	if (smc->sk.sk_state != SMC_INIT &&
977 	    smc->sk.sk_state != SMC_CLOSED) {
978 		rc = smcr_cdc_msg_send_validation(conn, pend, wr_buf);
979 		if (!rc) {
980 			queue_delayed_work(conn->lgr->tx_wq, &conn->tx_work, 0);
981 			smc->sk.sk_data_ready(&smc->sk);
982 		}
983 	} else {
984 		smc_wr_tx_put_slot(conn->lnk,
985 				   (struct smc_wr_tx_pend_priv *)pend);
986 	}
987 	return rc;
988 }
989 
990 void smc_switch_link_and_count(struct smc_connection *conn,
991 			       struct smc_link *to_lnk)
992 {
993 	atomic_dec(&conn->lnk->conn_cnt);
994 	conn->lnk = to_lnk;
995 	atomic_inc(&conn->lnk->conn_cnt);
996 }
997 
998 struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
999 				  struct smc_link *from_lnk, bool is_dev_err)
1000 {
1001 	struct smc_link *to_lnk = NULL;
1002 	struct smc_cdc_tx_pend *pend;
1003 	struct smc_connection *conn;
1004 	struct smc_wr_buf *wr_buf;
1005 	struct smc_sock *smc;
1006 	struct rb_node *node;
1007 	int i, rc = 0;
1008 
1009 	/* link is inactive, wake up tx waiters */
1010 	smc_wr_wakeup_tx_wait(from_lnk);
1011 
1012 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1013 		if (!smc_link_active(&lgr->lnk[i]) || i == from_lnk->link_idx)
1014 			continue;
1015 		if (is_dev_err && from_lnk->smcibdev == lgr->lnk[i].smcibdev &&
1016 		    from_lnk->ibport == lgr->lnk[i].ibport) {
1017 			continue;
1018 		}
1019 		to_lnk = &lgr->lnk[i];
1020 		break;
1021 	}
1022 	if (!to_lnk || !smc_wr_tx_link_hold(to_lnk)) {
1023 		smc_lgr_terminate_sched(lgr);
1024 		return NULL;
1025 	}
1026 again:
1027 	read_lock_bh(&lgr->conns_lock);
1028 	for (node = rb_first(&lgr->conns_all); node; node = rb_next(node)) {
1029 		conn = rb_entry(node, struct smc_connection, alert_node);
1030 		if (conn->lnk != from_lnk)
1031 			continue;
1032 		smc = container_of(conn, struct smc_sock, conn);
1033 		/* conn->lnk not yet set in SMC_INIT state */
1034 		if (smc->sk.sk_state == SMC_INIT)
1035 			continue;
1036 		if (smc->sk.sk_state == SMC_CLOSED ||
1037 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT1 ||
1038 		    smc->sk.sk_state == SMC_PEERCLOSEWAIT2 ||
1039 		    smc->sk.sk_state == SMC_APPFINCLOSEWAIT ||
1040 		    smc->sk.sk_state == SMC_APPCLOSEWAIT1 ||
1041 		    smc->sk.sk_state == SMC_APPCLOSEWAIT2 ||
1042 		    smc->sk.sk_state == SMC_PEERFINCLOSEWAIT ||
1043 		    smc->sk.sk_state == SMC_PEERABORTWAIT ||
1044 		    smc->sk.sk_state == SMC_PROCESSABORT) {
1045 			spin_lock_bh(&conn->send_lock);
1046 			smc_switch_link_and_count(conn, to_lnk);
1047 			spin_unlock_bh(&conn->send_lock);
1048 			continue;
1049 		}
1050 		sock_hold(&smc->sk);
1051 		read_unlock_bh(&lgr->conns_lock);
1052 		/* pre-fetch buffer outside of send_lock, might sleep */
1053 		rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
1054 		if (rc)
1055 			goto err_out;
1056 		/* avoid race with smcr_tx_sndbuf_nonempty() */
1057 		spin_lock_bh(&conn->send_lock);
1058 		smc_switch_link_and_count(conn, to_lnk);
1059 		rc = smc_switch_cursor(smc, pend, wr_buf);
1060 		spin_unlock_bh(&conn->send_lock);
1061 		sock_put(&smc->sk);
1062 		if (rc)
1063 			goto err_out;
1064 		goto again;
1065 	}
1066 	read_unlock_bh(&lgr->conns_lock);
1067 	smc_wr_tx_link_put(to_lnk);
1068 	return to_lnk;
1069 
1070 err_out:
1071 	smcr_link_down_cond_sched(to_lnk);
1072 	smc_wr_tx_link_put(to_lnk);
1073 	return NULL;
1074 }
1075 
1076 static void smcr_buf_unuse(struct smc_buf_desc *rmb_desc,
1077 			   struct smc_link_group *lgr)
1078 {
1079 	int rc;
1080 
1081 	if (rmb_desc->is_conf_rkey && !list_empty(&lgr->list)) {
1082 		/* unregister rmb with peer */
1083 		rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
1084 		if (!rc) {
1085 			/* protect against smc_llc_cli_rkey_exchange() */
1086 			mutex_lock(&lgr->llc_conf_mutex);
1087 			smc_llc_do_delete_rkey(lgr, rmb_desc);
1088 			rmb_desc->is_conf_rkey = false;
1089 			mutex_unlock(&lgr->llc_conf_mutex);
1090 			smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
1091 		}
1092 	}
1093 
1094 	if (rmb_desc->is_reg_err) {
1095 		/* buf registration failed, reuse not possible */
1096 		mutex_lock(&lgr->rmbs_lock);
1097 		list_del(&rmb_desc->list);
1098 		mutex_unlock(&lgr->rmbs_lock);
1099 
1100 		smc_buf_free(lgr, true, rmb_desc);
1101 	} else {
1102 		rmb_desc->used = 0;
1103 	}
1104 }
1105 
1106 static void smc_buf_unuse(struct smc_connection *conn,
1107 			  struct smc_link_group *lgr)
1108 {
1109 	if (conn->sndbuf_desc)
1110 		conn->sndbuf_desc->used = 0;
1111 	if (conn->rmb_desc && lgr->is_smcd)
1112 		conn->rmb_desc->used = 0;
1113 	else if (conn->rmb_desc)
1114 		smcr_buf_unuse(conn->rmb_desc, lgr);
1115 }
1116 
1117 /* remove a finished connection from its link group */
1118 void smc_conn_free(struct smc_connection *conn)
1119 {
1120 	struct smc_link_group *lgr = conn->lgr;
1121 
1122 	if (!lgr)
1123 		return;
1124 	if (lgr->is_smcd) {
1125 		if (!list_empty(&lgr->list))
1126 			smc_ism_unset_conn(conn);
1127 		tasklet_kill(&conn->rx_tsklet);
1128 	} else {
1129 		smc_cdc_tx_dismiss_slots(conn);
1130 		if (current_work() != &conn->abort_work)
1131 			cancel_work_sync(&conn->abort_work);
1132 	}
1133 	if (!list_empty(&lgr->list)) {
1134 		smc_lgr_unregister_conn(conn);
1135 		smc_buf_unuse(conn, lgr); /* allow buffer reuse */
1136 	}
1137 
1138 	if (!lgr->conns_num)
1139 		smc_lgr_schedule_free_work(lgr);
1140 }
1141 
1142 /* unregister a link from a buf_desc */
1143 static void smcr_buf_unmap_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1144 				struct smc_link *lnk)
1145 {
1146 	if (is_rmb)
1147 		buf_desc->is_reg_mr[lnk->link_idx] = false;
1148 	if (!buf_desc->is_map_ib[lnk->link_idx])
1149 		return;
1150 	if (is_rmb) {
1151 		if (buf_desc->mr_rx[lnk->link_idx]) {
1152 			smc_ib_put_memory_region(
1153 					buf_desc->mr_rx[lnk->link_idx]);
1154 			buf_desc->mr_rx[lnk->link_idx] = NULL;
1155 		}
1156 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_FROM_DEVICE);
1157 	} else {
1158 		smc_ib_buf_unmap_sg(lnk, buf_desc, DMA_TO_DEVICE);
1159 	}
1160 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1161 	buf_desc->is_map_ib[lnk->link_idx] = false;
1162 }
1163 
1164 /* unmap all buffers of lgr for a deleted link */
1165 static void smcr_buf_unmap_lgr(struct smc_link *lnk)
1166 {
1167 	struct smc_link_group *lgr = lnk->lgr;
1168 	struct smc_buf_desc *buf_desc, *bf;
1169 	int i;
1170 
1171 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1172 		mutex_lock(&lgr->rmbs_lock);
1173 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list)
1174 			smcr_buf_unmap_link(buf_desc, true, lnk);
1175 		mutex_unlock(&lgr->rmbs_lock);
1176 		mutex_lock(&lgr->sndbufs_lock);
1177 		list_for_each_entry_safe(buf_desc, bf, &lgr->sndbufs[i],
1178 					 list)
1179 			smcr_buf_unmap_link(buf_desc, false, lnk);
1180 		mutex_unlock(&lgr->sndbufs_lock);
1181 	}
1182 }
1183 
1184 static void smcr_rtoken_clear_link(struct smc_link *lnk)
1185 {
1186 	struct smc_link_group *lgr = lnk->lgr;
1187 	int i;
1188 
1189 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
1190 		lgr->rtokens[i][lnk->link_idx].rkey = 0;
1191 		lgr->rtokens[i][lnk->link_idx].dma_addr = 0;
1192 	}
1193 }
1194 
1195 /* must be called under lgr->llc_conf_mutex lock */
1196 void smcr_link_clear(struct smc_link *lnk, bool log)
1197 {
1198 	struct smc_ib_device *smcibdev;
1199 
1200 	if (!lnk->lgr || lnk->state == SMC_LNK_UNUSED)
1201 		return;
1202 	lnk->peer_qpn = 0;
1203 	smc_llc_link_clear(lnk, log);
1204 	smcr_buf_unmap_lgr(lnk);
1205 	smcr_rtoken_clear_link(lnk);
1206 	smc_ib_modify_qp_reset(lnk);
1207 	smc_wr_free_link(lnk);
1208 	smc_ib_destroy_queue_pair(lnk);
1209 	smc_ib_dealloc_protection_domain(lnk);
1210 	smc_wr_free_link_mem(lnk);
1211 	smc_ibdev_cnt_dec(lnk);
1212 	put_device(&lnk->smcibdev->ibdev->dev);
1213 	smcibdev = lnk->smcibdev;
1214 	memset(lnk, 0, sizeof(struct smc_link));
1215 	lnk->state = SMC_LNK_UNUSED;
1216 	if (!atomic_dec_return(&smcibdev->lnk_cnt))
1217 		wake_up(&smcibdev->lnks_deleted);
1218 }
1219 
1220 static void smcr_buf_free(struct smc_link_group *lgr, bool is_rmb,
1221 			  struct smc_buf_desc *buf_desc)
1222 {
1223 	int i;
1224 
1225 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1226 		smcr_buf_unmap_link(buf_desc, is_rmb, &lgr->lnk[i]);
1227 
1228 	if (buf_desc->pages)
1229 		__free_pages(buf_desc->pages, buf_desc->order);
1230 	kfree(buf_desc);
1231 }
1232 
1233 static void smcd_buf_free(struct smc_link_group *lgr, bool is_dmb,
1234 			  struct smc_buf_desc *buf_desc)
1235 {
1236 	if (is_dmb) {
1237 		/* restore original buf len */
1238 		buf_desc->len += sizeof(struct smcd_cdc_msg);
1239 		smc_ism_unregister_dmb(lgr->smcd, buf_desc);
1240 	} else {
1241 		kfree(buf_desc->cpu_addr);
1242 	}
1243 	kfree(buf_desc);
1244 }
1245 
1246 static void smc_buf_free(struct smc_link_group *lgr, bool is_rmb,
1247 			 struct smc_buf_desc *buf_desc)
1248 {
1249 	if (lgr->is_smcd)
1250 		smcd_buf_free(lgr, is_rmb, buf_desc);
1251 	else
1252 		smcr_buf_free(lgr, is_rmb, buf_desc);
1253 }
1254 
1255 static void __smc_lgr_free_bufs(struct smc_link_group *lgr, bool is_rmb)
1256 {
1257 	struct smc_buf_desc *buf_desc, *bf_desc;
1258 	struct list_head *buf_list;
1259 	int i;
1260 
1261 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1262 		if (is_rmb)
1263 			buf_list = &lgr->rmbs[i];
1264 		else
1265 			buf_list = &lgr->sndbufs[i];
1266 		list_for_each_entry_safe(buf_desc, bf_desc, buf_list,
1267 					 list) {
1268 			list_del(&buf_desc->list);
1269 			smc_buf_free(lgr, is_rmb, buf_desc);
1270 		}
1271 	}
1272 }
1273 
1274 static void smc_lgr_free_bufs(struct smc_link_group *lgr)
1275 {
1276 	/* free send buffers */
1277 	__smc_lgr_free_bufs(lgr, false);
1278 	/* free rmbs */
1279 	__smc_lgr_free_bufs(lgr, true);
1280 }
1281 
1282 /* remove a link group */
1283 static void smc_lgr_free(struct smc_link_group *lgr)
1284 {
1285 	int i;
1286 
1287 	if (!lgr->is_smcd) {
1288 		mutex_lock(&lgr->llc_conf_mutex);
1289 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1290 			if (lgr->lnk[i].state != SMC_LNK_UNUSED)
1291 				smcr_link_clear(&lgr->lnk[i], false);
1292 		}
1293 		mutex_unlock(&lgr->llc_conf_mutex);
1294 		smc_llc_lgr_clear(lgr);
1295 	}
1296 
1297 	smc_lgr_free_bufs(lgr);
1298 	destroy_workqueue(lgr->tx_wq);
1299 	if (lgr->is_smcd) {
1300 		smc_ism_put_vlan(lgr->smcd, lgr->vlan_id);
1301 		put_device(&lgr->smcd->dev);
1302 		if (!atomic_dec_return(&lgr->smcd->lgr_cnt))
1303 			wake_up(&lgr->smcd->lgrs_deleted);
1304 	} else {
1305 		smc_wr_free_lgr_mem(lgr);
1306 		if (!atomic_dec_return(&lgr_cnt))
1307 			wake_up(&lgrs_deleted);
1308 	}
1309 	kfree(lgr);
1310 }
1311 
1312 static void smc_sk_wake_ups(struct smc_sock *smc)
1313 {
1314 	smc->sk.sk_write_space(&smc->sk);
1315 	smc->sk.sk_data_ready(&smc->sk);
1316 	smc->sk.sk_state_change(&smc->sk);
1317 }
1318 
1319 /* kill a connection */
1320 static void smc_conn_kill(struct smc_connection *conn, bool soft)
1321 {
1322 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1323 
1324 	if (conn->lgr->is_smcd && conn->lgr->peer_shutdown)
1325 		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
1326 	else
1327 		smc_close_abort(conn);
1328 	conn->killed = 1;
1329 	smc->sk.sk_err = ECONNABORTED;
1330 	smc_sk_wake_ups(smc);
1331 	if (conn->lgr->is_smcd) {
1332 		smc_ism_unset_conn(conn);
1333 		if (soft)
1334 			tasklet_kill(&conn->rx_tsklet);
1335 		else
1336 			tasklet_unlock_wait(&conn->rx_tsklet);
1337 	} else {
1338 		smc_cdc_tx_dismiss_slots(conn);
1339 	}
1340 	smc_lgr_unregister_conn(conn);
1341 	smc_close_active_abort(smc);
1342 }
1343 
1344 static void smc_lgr_cleanup(struct smc_link_group *lgr)
1345 {
1346 	if (lgr->is_smcd) {
1347 		smc_ism_signal_shutdown(lgr);
1348 	} else {
1349 		u32 rsn = lgr->llc_termination_rsn;
1350 
1351 		if (!rsn)
1352 			rsn = SMC_LLC_DEL_PROG_INIT_TERM;
1353 		smc_llc_send_link_delete_all(lgr, false, rsn);
1354 		smcr_lgr_link_deactivate_all(lgr);
1355 	}
1356 }
1357 
1358 /* terminate link group
1359  * @soft: true if link group shutdown can take its time
1360  *	  false if immediate link group shutdown is required
1361  */
1362 static void __smc_lgr_terminate(struct smc_link_group *lgr, bool soft)
1363 {
1364 	struct smc_connection *conn;
1365 	struct smc_sock *smc;
1366 	struct rb_node *node;
1367 
1368 	if (lgr->terminating)
1369 		return;	/* lgr already terminating */
1370 	/* cancel free_work sync, will terminate when lgr->freeing is set */
1371 	cancel_delayed_work_sync(&lgr->free_work);
1372 	lgr->terminating = 1;
1373 
1374 	/* kill remaining link group connections */
1375 	read_lock_bh(&lgr->conns_lock);
1376 	node = rb_first(&lgr->conns_all);
1377 	while (node) {
1378 		read_unlock_bh(&lgr->conns_lock);
1379 		conn = rb_entry(node, struct smc_connection, alert_node);
1380 		smc = container_of(conn, struct smc_sock, conn);
1381 		sock_hold(&smc->sk); /* sock_put below */
1382 		lock_sock(&smc->sk);
1383 		smc_conn_kill(conn, soft);
1384 		release_sock(&smc->sk);
1385 		sock_put(&smc->sk); /* sock_hold above */
1386 		read_lock_bh(&lgr->conns_lock);
1387 		node = rb_first(&lgr->conns_all);
1388 	}
1389 	read_unlock_bh(&lgr->conns_lock);
1390 	smc_lgr_cleanup(lgr);
1391 	smc_lgr_free(lgr);
1392 }
1393 
1394 /* unlink link group and schedule termination */
1395 void smc_lgr_terminate_sched(struct smc_link_group *lgr)
1396 {
1397 	spinlock_t *lgr_lock;
1398 
1399 	smc_lgr_list_head(lgr, &lgr_lock);
1400 	spin_lock_bh(lgr_lock);
1401 	if (list_empty(&lgr->list) || lgr->terminating || lgr->freeing) {
1402 		spin_unlock_bh(lgr_lock);
1403 		return;	/* lgr already terminating */
1404 	}
1405 	list_del_init(&lgr->list);
1406 	lgr->freeing = 1;
1407 	spin_unlock_bh(lgr_lock);
1408 	schedule_work(&lgr->terminate_work);
1409 }
1410 
1411 /* Called when peer lgr shutdown (regularly or abnormally) is received */
1412 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid, unsigned short vlan)
1413 {
1414 	struct smc_link_group *lgr, *l;
1415 	LIST_HEAD(lgr_free_list);
1416 
1417 	/* run common cleanup function and build free list */
1418 	spin_lock_bh(&dev->lgr_lock);
1419 	list_for_each_entry_safe(lgr, l, &dev->lgr_list, list) {
1420 		if ((!peer_gid || lgr->peer_gid == peer_gid) &&
1421 		    (vlan == VLAN_VID_MASK || lgr->vlan_id == vlan)) {
1422 			if (peer_gid) /* peer triggered termination */
1423 				lgr->peer_shutdown = 1;
1424 			list_move(&lgr->list, &lgr_free_list);
1425 			lgr->freeing = 1;
1426 		}
1427 	}
1428 	spin_unlock_bh(&dev->lgr_lock);
1429 
1430 	/* cancel the regular free workers and actually free lgrs */
1431 	list_for_each_entry_safe(lgr, l, &lgr_free_list, list) {
1432 		list_del_init(&lgr->list);
1433 		schedule_work(&lgr->terminate_work);
1434 	}
1435 }
1436 
1437 /* Called when an SMCD device is removed or the smc module is unloaded */
1438 void smc_smcd_terminate_all(struct smcd_dev *smcd)
1439 {
1440 	struct smc_link_group *lgr, *lg;
1441 	LIST_HEAD(lgr_free_list);
1442 
1443 	spin_lock_bh(&smcd->lgr_lock);
1444 	list_splice_init(&smcd->lgr_list, &lgr_free_list);
1445 	list_for_each_entry(lgr, &lgr_free_list, list)
1446 		lgr->freeing = 1;
1447 	spin_unlock_bh(&smcd->lgr_lock);
1448 
1449 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1450 		list_del_init(&lgr->list);
1451 		__smc_lgr_terminate(lgr, false);
1452 	}
1453 
1454 	if (atomic_read(&smcd->lgr_cnt))
1455 		wait_event(smcd->lgrs_deleted, !atomic_read(&smcd->lgr_cnt));
1456 }
1457 
1458 /* Called when an SMCR device is removed or the smc module is unloaded.
1459  * If smcibdev is given, all SMCR link groups using this device are terminated.
1460  * If smcibdev is NULL, all SMCR link groups are terminated.
1461  */
1462 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev)
1463 {
1464 	struct smc_link_group *lgr, *lg;
1465 	LIST_HEAD(lgr_free_list);
1466 	int i;
1467 
1468 	spin_lock_bh(&smc_lgr_list.lock);
1469 	if (!smcibdev) {
1470 		list_splice_init(&smc_lgr_list.list, &lgr_free_list);
1471 		list_for_each_entry(lgr, &lgr_free_list, list)
1472 			lgr->freeing = 1;
1473 	} else {
1474 		list_for_each_entry_safe(lgr, lg, &smc_lgr_list.list, list) {
1475 			for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1476 				if (lgr->lnk[i].smcibdev == smcibdev)
1477 					smcr_link_down_cond_sched(&lgr->lnk[i]);
1478 			}
1479 		}
1480 	}
1481 	spin_unlock_bh(&smc_lgr_list.lock);
1482 
1483 	list_for_each_entry_safe(lgr, lg, &lgr_free_list, list) {
1484 		list_del_init(&lgr->list);
1485 		smc_llc_set_termination_rsn(lgr, SMC_LLC_DEL_OP_INIT_TERM);
1486 		__smc_lgr_terminate(lgr, false);
1487 	}
1488 
1489 	if (smcibdev) {
1490 		if (atomic_read(&smcibdev->lnk_cnt))
1491 			wait_event(smcibdev->lnks_deleted,
1492 				   !atomic_read(&smcibdev->lnk_cnt));
1493 	} else {
1494 		if (atomic_read(&lgr_cnt))
1495 			wait_event(lgrs_deleted, !atomic_read(&lgr_cnt));
1496 	}
1497 }
1498 
1499 /* set new lgr type and clear all asymmetric link tagging */
1500 void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type)
1501 {
1502 	char *lgr_type = "";
1503 	int i;
1504 
1505 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++)
1506 		if (smc_link_usable(&lgr->lnk[i]))
1507 			lgr->lnk[i].link_is_asym = false;
1508 	if (lgr->type == new_type)
1509 		return;
1510 	lgr->type = new_type;
1511 
1512 	switch (lgr->type) {
1513 	case SMC_LGR_NONE:
1514 		lgr_type = "NONE";
1515 		break;
1516 	case SMC_LGR_SINGLE:
1517 		lgr_type = "SINGLE";
1518 		break;
1519 	case SMC_LGR_SYMMETRIC:
1520 		lgr_type = "SYMMETRIC";
1521 		break;
1522 	case SMC_LGR_ASYMMETRIC_PEER:
1523 		lgr_type = "ASYMMETRIC_PEER";
1524 		break;
1525 	case SMC_LGR_ASYMMETRIC_LOCAL:
1526 		lgr_type = "ASYMMETRIC_LOCAL";
1527 		break;
1528 	}
1529 	pr_warn_ratelimited("smc: SMC-R lg %*phN state changed: "
1530 			    "%s, pnetid %.16s\n", SMC_LGR_ID_SIZE, &lgr->id,
1531 			    lgr_type, lgr->pnet_id);
1532 }
1533 
1534 /* set new lgr type and tag a link as asymmetric */
1535 void smcr_lgr_set_type_asym(struct smc_link_group *lgr,
1536 			    enum smc_lgr_type new_type, int asym_lnk_idx)
1537 {
1538 	smcr_lgr_set_type(lgr, new_type);
1539 	lgr->lnk[asym_lnk_idx].link_is_asym = true;
1540 }
1541 
1542 /* abort connection, abort_work scheduled from tasklet context */
1543 static void smc_conn_abort_work(struct work_struct *work)
1544 {
1545 	struct smc_connection *conn = container_of(work,
1546 						   struct smc_connection,
1547 						   abort_work);
1548 	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
1549 
1550 	lock_sock(&smc->sk);
1551 	smc_conn_kill(conn, true);
1552 	release_sock(&smc->sk);
1553 	sock_put(&smc->sk); /* sock_hold done by schedulers of abort_work */
1554 }
1555 
1556 void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport)
1557 {
1558 	struct smc_link_group *lgr, *n;
1559 
1560 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1561 		struct smc_link *link;
1562 
1563 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1564 			    SMC_MAX_PNETID_LEN) ||
1565 		    lgr->type == SMC_LGR_SYMMETRIC ||
1566 		    lgr->type == SMC_LGR_ASYMMETRIC_PEER)
1567 			continue;
1568 
1569 		/* trigger local add link processing */
1570 		link = smc_llc_usable_link(lgr);
1571 		if (link)
1572 			smc_llc_add_link_local(link);
1573 	}
1574 }
1575 
1576 /* link is down - switch connections to alternate link,
1577  * must be called under lgr->llc_conf_mutex lock
1578  */
1579 static void smcr_link_down(struct smc_link *lnk)
1580 {
1581 	struct smc_link_group *lgr = lnk->lgr;
1582 	struct smc_link *to_lnk;
1583 	int del_link_id;
1584 
1585 	if (!lgr || lnk->state == SMC_LNK_UNUSED || list_empty(&lgr->list))
1586 		return;
1587 
1588 	smc_ib_modify_qp_reset(lnk);
1589 	to_lnk = smc_switch_conns(lgr, lnk, true);
1590 	if (!to_lnk) { /* no backup link available */
1591 		smcr_link_clear(lnk, true);
1592 		return;
1593 	}
1594 	smcr_lgr_set_type(lgr, SMC_LGR_SINGLE);
1595 	del_link_id = lnk->link_id;
1596 
1597 	if (lgr->role == SMC_SERV) {
1598 		/* trigger local delete link processing */
1599 		smc_llc_srv_delete_link_local(to_lnk, del_link_id);
1600 	} else {
1601 		if (lgr->llc_flow_lcl.type != SMC_LLC_FLOW_NONE) {
1602 			/* another llc task is ongoing */
1603 			mutex_unlock(&lgr->llc_conf_mutex);
1604 			wait_event_timeout(lgr->llc_flow_waiter,
1605 				(list_empty(&lgr->list) ||
1606 				 lgr->llc_flow_lcl.type == SMC_LLC_FLOW_NONE),
1607 				SMC_LLC_WAIT_TIME);
1608 			mutex_lock(&lgr->llc_conf_mutex);
1609 		}
1610 		if (!list_empty(&lgr->list)) {
1611 			smc_llc_send_delete_link(to_lnk, del_link_id,
1612 						 SMC_LLC_REQ, true,
1613 						 SMC_LLC_DEL_LOST_PATH);
1614 			smcr_link_clear(lnk, true);
1615 		}
1616 		wake_up(&lgr->llc_flow_waiter);	/* wake up next waiter */
1617 	}
1618 }
1619 
1620 /* must be called under lgr->llc_conf_mutex lock */
1621 void smcr_link_down_cond(struct smc_link *lnk)
1622 {
1623 	if (smc_link_downing(&lnk->state))
1624 		smcr_link_down(lnk);
1625 }
1626 
1627 /* will get the lgr->llc_conf_mutex lock */
1628 void smcr_link_down_cond_sched(struct smc_link *lnk)
1629 {
1630 	if (smc_link_downing(&lnk->state))
1631 		schedule_work(&lnk->link_down_wrk);
1632 }
1633 
1634 void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport)
1635 {
1636 	struct smc_link_group *lgr, *n;
1637 	int i;
1638 
1639 	list_for_each_entry_safe(lgr, n, &smc_lgr_list.list, list) {
1640 		if (strncmp(smcibdev->pnetid[ibport - 1], lgr->pnet_id,
1641 			    SMC_MAX_PNETID_LEN))
1642 			continue; /* lgr is not affected */
1643 		if (list_empty(&lgr->list))
1644 			continue;
1645 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1646 			struct smc_link *lnk = &lgr->lnk[i];
1647 
1648 			if (smc_link_usable(lnk) &&
1649 			    lnk->smcibdev == smcibdev && lnk->ibport == ibport)
1650 				smcr_link_down_cond_sched(lnk);
1651 		}
1652 	}
1653 }
1654 
1655 static void smc_link_down_work(struct work_struct *work)
1656 {
1657 	struct smc_link *link = container_of(work, struct smc_link,
1658 					     link_down_wrk);
1659 	struct smc_link_group *lgr = link->lgr;
1660 
1661 	if (list_empty(&lgr->list))
1662 		return;
1663 	wake_up_all(&lgr->llc_msg_waiter);
1664 	mutex_lock(&lgr->llc_conf_mutex);
1665 	smcr_link_down(link);
1666 	mutex_unlock(&lgr->llc_conf_mutex);
1667 }
1668 
1669 /* Determine vlan of internal TCP socket.
1670  * @vlan_id: address to store the determined vlan id into
1671  */
1672 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini)
1673 {
1674 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
1675 	struct net_device *ndev;
1676 	int i, nest_lvl, rc = 0;
1677 
1678 	ini->vlan_id = 0;
1679 	if (!dst) {
1680 		rc = -ENOTCONN;
1681 		goto out;
1682 	}
1683 	if (!dst->dev) {
1684 		rc = -ENODEV;
1685 		goto out_rel;
1686 	}
1687 
1688 	ndev = dst->dev;
1689 	if (is_vlan_dev(ndev)) {
1690 		ini->vlan_id = vlan_dev_vlan_id(ndev);
1691 		goto out_rel;
1692 	}
1693 
1694 	rtnl_lock();
1695 	nest_lvl = ndev->lower_level;
1696 	for (i = 0; i < nest_lvl; i++) {
1697 		struct list_head *lower = &ndev->adj_list.lower;
1698 
1699 		if (list_empty(lower))
1700 			break;
1701 		lower = lower->next;
1702 		ndev = (struct net_device *)netdev_lower_get_next(ndev, &lower);
1703 		if (is_vlan_dev(ndev)) {
1704 			ini->vlan_id = vlan_dev_vlan_id(ndev);
1705 			break;
1706 		}
1707 	}
1708 	rtnl_unlock();
1709 
1710 out_rel:
1711 	dst_release(dst);
1712 out:
1713 	return rc;
1714 }
1715 
1716 static bool smcr_lgr_match(struct smc_link_group *lgr, u8 smcr_version,
1717 			   u8 peer_systemid[],
1718 			   u8 peer_gid[],
1719 			   u8 peer_mac_v1[],
1720 			   enum smc_lgr_role role, u32 clcqpn)
1721 {
1722 	int i;
1723 
1724 	if (memcmp(lgr->peer_systemid, peer_systemid, SMC_SYSTEMID_LEN) ||
1725 	    lgr->role != role)
1726 		return false;
1727 
1728 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1729 		if (!smc_link_active(&lgr->lnk[i]))
1730 			continue;
1731 		if ((lgr->role == SMC_SERV || lgr->lnk[i].peer_qpn == clcqpn) &&
1732 		    !memcmp(lgr->lnk[i].peer_gid, peer_gid, SMC_GID_SIZE) &&
1733 		    (smcr_version == SMC_V2 ||
1734 		     !memcmp(lgr->lnk[i].peer_mac, peer_mac_v1, ETH_ALEN)))
1735 			return true;
1736 	}
1737 	return false;
1738 }
1739 
1740 static bool smcd_lgr_match(struct smc_link_group *lgr,
1741 			   struct smcd_dev *smcismdev, u64 peer_gid)
1742 {
1743 	return lgr->peer_gid == peer_gid && lgr->smcd == smcismdev;
1744 }
1745 
1746 /* create a new SMC connection (and a new link group if necessary) */
1747 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini)
1748 {
1749 	struct smc_connection *conn = &smc->conn;
1750 	struct list_head *lgr_list;
1751 	struct smc_link_group *lgr;
1752 	enum smc_lgr_role role;
1753 	spinlock_t *lgr_lock;
1754 	int rc = 0;
1755 
1756 	lgr_list = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_list :
1757 				  &smc_lgr_list.list;
1758 	lgr_lock = ini->is_smcd ? &ini->ism_dev[ini->ism_selected]->lgr_lock :
1759 				  &smc_lgr_list.lock;
1760 	ini->first_contact_local = 1;
1761 	role = smc->listen_smc ? SMC_SERV : SMC_CLNT;
1762 	if (role == SMC_CLNT && ini->first_contact_peer)
1763 		/* create new link group as well */
1764 		goto create;
1765 
1766 	/* determine if an existing link group can be reused */
1767 	spin_lock_bh(lgr_lock);
1768 	list_for_each_entry(lgr, lgr_list, list) {
1769 		write_lock_bh(&lgr->conns_lock);
1770 		if ((ini->is_smcd ?
1771 		     smcd_lgr_match(lgr, ini->ism_dev[ini->ism_selected],
1772 				    ini->ism_peer_gid[ini->ism_selected]) :
1773 		     smcr_lgr_match(lgr, ini->smcr_version,
1774 				    ini->peer_systemid,
1775 				    ini->peer_gid, ini->peer_mac, role,
1776 				    ini->ib_clcqpn)) &&
1777 		    !lgr->sync_err &&
1778 		    (ini->smcd_version == SMC_V2 ||
1779 		     lgr->vlan_id == ini->vlan_id) &&
1780 		    (role == SMC_CLNT || ini->is_smcd ||
1781 		     lgr->conns_num < SMC_RMBS_PER_LGR_MAX)) {
1782 			/* link group found */
1783 			ini->first_contact_local = 0;
1784 			conn->lgr = lgr;
1785 			rc = smc_lgr_register_conn(conn, false);
1786 			write_unlock_bh(&lgr->conns_lock);
1787 			if (!rc && delayed_work_pending(&lgr->free_work))
1788 				cancel_delayed_work(&lgr->free_work);
1789 			break;
1790 		}
1791 		write_unlock_bh(&lgr->conns_lock);
1792 	}
1793 	spin_unlock_bh(lgr_lock);
1794 	if (rc)
1795 		return rc;
1796 
1797 	if (role == SMC_CLNT && !ini->first_contact_peer &&
1798 	    ini->first_contact_local) {
1799 		/* Server reuses a link group, but Client wants to start
1800 		 * a new one
1801 		 * send out_of_sync decline, reason synchr. error
1802 		 */
1803 		return SMC_CLC_DECL_SYNCERR;
1804 	}
1805 
1806 create:
1807 	if (ini->first_contact_local) {
1808 		rc = smc_lgr_create(smc, ini);
1809 		if (rc)
1810 			goto out;
1811 		lgr = conn->lgr;
1812 		write_lock_bh(&lgr->conns_lock);
1813 		rc = smc_lgr_register_conn(conn, true);
1814 		write_unlock_bh(&lgr->conns_lock);
1815 		if (rc)
1816 			goto out;
1817 	}
1818 	conn->local_tx_ctrl.common.type = SMC_CDC_MSG_TYPE;
1819 	conn->local_tx_ctrl.len = SMC_WR_TX_SIZE;
1820 	conn->urg_state = SMC_URG_READ;
1821 	INIT_WORK(&smc->conn.abort_work, smc_conn_abort_work);
1822 	if (ini->is_smcd) {
1823 		conn->rx_off = sizeof(struct smcd_cdc_msg);
1824 		smcd_cdc_rx_init(conn); /* init tasklet for this conn */
1825 	} else {
1826 		conn->rx_off = 0;
1827 	}
1828 #ifndef KERNEL_HAS_ATOMIC64
1829 	spin_lock_init(&conn->acurs_lock);
1830 #endif
1831 
1832 out:
1833 	return rc;
1834 }
1835 
1836 #define SMCD_DMBE_SIZES		6 /* 0 -> 16KB, 1 -> 32KB, .. 6 -> 1MB */
1837 #define SMCR_RMBE_SIZES		5 /* 0 -> 16KB, 1 -> 32KB, .. 5 -> 512KB */
1838 
1839 /* convert the RMB size into the compressed notation (minimum 16K, see
1840  * SMCD/R_DMBE_SIZES.
1841  * In contrast to plain ilog2, this rounds towards the next power of 2,
1842  * so the socket application gets at least its desired sndbuf / rcvbuf size.
1843  */
1844 static u8 smc_compress_bufsize(int size, bool is_smcd, bool is_rmb)
1845 {
1846 	const unsigned int max_scat = SG_MAX_SINGLE_ALLOC * PAGE_SIZE;
1847 	u8 compressed;
1848 
1849 	if (size <= SMC_BUF_MIN_SIZE)
1850 		return 0;
1851 
1852 	size = (size - 1) >> 14;  /* convert to 16K multiple */
1853 	compressed = min_t(u8, ilog2(size) + 1,
1854 			   is_smcd ? SMCD_DMBE_SIZES : SMCR_RMBE_SIZES);
1855 
1856 	if (!is_smcd && is_rmb)
1857 		/* RMBs are backed by & limited to max size of scatterlists */
1858 		compressed = min_t(u8, compressed, ilog2(max_scat >> 14));
1859 
1860 	return compressed;
1861 }
1862 
1863 /* convert the RMB size from compressed notation into integer */
1864 int smc_uncompress_bufsize(u8 compressed)
1865 {
1866 	u32 size;
1867 
1868 	size = 0x00000001 << (((int)compressed) + 14);
1869 	return (int)size;
1870 }
1871 
1872 /* try to reuse a sndbuf or rmb description slot for a certain
1873  * buffer size; if not available, return NULL
1874  */
1875 static struct smc_buf_desc *smc_buf_get_slot(int compressed_bufsize,
1876 					     struct mutex *lock,
1877 					     struct list_head *buf_list)
1878 {
1879 	struct smc_buf_desc *buf_slot;
1880 
1881 	mutex_lock(lock);
1882 	list_for_each_entry(buf_slot, buf_list, list) {
1883 		if (cmpxchg(&buf_slot->used, 0, 1) == 0) {
1884 			mutex_unlock(lock);
1885 			return buf_slot;
1886 		}
1887 	}
1888 	mutex_unlock(lock);
1889 	return NULL;
1890 }
1891 
1892 /* one of the conditions for announcing a receiver's current window size is
1893  * that it "results in a minimum increase in the window size of 10% of the
1894  * receive buffer space" [RFC7609]
1895  */
1896 static inline int smc_rmb_wnd_update_limit(int rmbe_size)
1897 {
1898 	return min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2);
1899 }
1900 
1901 /* map an rmb buf to a link */
1902 static int smcr_buf_map_link(struct smc_buf_desc *buf_desc, bool is_rmb,
1903 			     struct smc_link *lnk)
1904 {
1905 	int rc;
1906 
1907 	if (buf_desc->is_map_ib[lnk->link_idx])
1908 		return 0;
1909 
1910 	rc = sg_alloc_table(&buf_desc->sgt[lnk->link_idx], 1, GFP_KERNEL);
1911 	if (rc)
1912 		return rc;
1913 	sg_set_buf(buf_desc->sgt[lnk->link_idx].sgl,
1914 		   buf_desc->cpu_addr, buf_desc->len);
1915 
1916 	/* map sg table to DMA address */
1917 	rc = smc_ib_buf_map_sg(lnk, buf_desc,
1918 			       is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
1919 	/* SMC protocol depends on mapping to one DMA address only */
1920 	if (rc != 1) {
1921 		rc = -EAGAIN;
1922 		goto free_table;
1923 	}
1924 
1925 	/* create a new memory region for the RMB */
1926 	if (is_rmb) {
1927 		rc = smc_ib_get_memory_region(lnk->roce_pd,
1928 					      IB_ACCESS_REMOTE_WRITE |
1929 					      IB_ACCESS_LOCAL_WRITE,
1930 					      buf_desc, lnk->link_idx);
1931 		if (rc)
1932 			goto buf_unmap;
1933 		smc_ib_sync_sg_for_device(lnk, buf_desc, DMA_FROM_DEVICE);
1934 	}
1935 	buf_desc->is_map_ib[lnk->link_idx] = true;
1936 	return 0;
1937 
1938 buf_unmap:
1939 	smc_ib_buf_unmap_sg(lnk, buf_desc,
1940 			    is_rmb ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
1941 free_table:
1942 	sg_free_table(&buf_desc->sgt[lnk->link_idx]);
1943 	return rc;
1944 }
1945 
1946 /* register a new rmb on IB device,
1947  * must be called under lgr->llc_conf_mutex lock
1948  */
1949 int smcr_link_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc)
1950 {
1951 	if (list_empty(&link->lgr->list))
1952 		return -ENOLINK;
1953 	if (!rmb_desc->is_reg_mr[link->link_idx]) {
1954 		/* register memory region for new rmb */
1955 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[link->link_idx])) {
1956 			rmb_desc->is_reg_err = true;
1957 			return -EFAULT;
1958 		}
1959 		rmb_desc->is_reg_mr[link->link_idx] = true;
1960 	}
1961 	return 0;
1962 }
1963 
1964 static int _smcr_buf_map_lgr(struct smc_link *lnk, struct mutex *lock,
1965 			     struct list_head *lst, bool is_rmb)
1966 {
1967 	struct smc_buf_desc *buf_desc, *bf;
1968 	int rc = 0;
1969 
1970 	mutex_lock(lock);
1971 	list_for_each_entry_safe(buf_desc, bf, lst, list) {
1972 		if (!buf_desc->used)
1973 			continue;
1974 		rc = smcr_buf_map_link(buf_desc, is_rmb, lnk);
1975 		if (rc)
1976 			goto out;
1977 	}
1978 out:
1979 	mutex_unlock(lock);
1980 	return rc;
1981 }
1982 
1983 /* map all used buffers of lgr for a new link */
1984 int smcr_buf_map_lgr(struct smc_link *lnk)
1985 {
1986 	struct smc_link_group *lgr = lnk->lgr;
1987 	int i, rc = 0;
1988 
1989 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
1990 		rc = _smcr_buf_map_lgr(lnk, &lgr->rmbs_lock,
1991 				       &lgr->rmbs[i], true);
1992 		if (rc)
1993 			return rc;
1994 		rc = _smcr_buf_map_lgr(lnk, &lgr->sndbufs_lock,
1995 				       &lgr->sndbufs[i], false);
1996 		if (rc)
1997 			return rc;
1998 	}
1999 	return 0;
2000 }
2001 
2002 /* register all used buffers of lgr for a new link,
2003  * must be called under lgr->llc_conf_mutex lock
2004  */
2005 int smcr_buf_reg_lgr(struct smc_link *lnk)
2006 {
2007 	struct smc_link_group *lgr = lnk->lgr;
2008 	struct smc_buf_desc *buf_desc, *bf;
2009 	int i, rc = 0;
2010 
2011 	mutex_lock(&lgr->rmbs_lock);
2012 	for (i = 0; i < SMC_RMBE_SIZES; i++) {
2013 		list_for_each_entry_safe(buf_desc, bf, &lgr->rmbs[i], list) {
2014 			if (!buf_desc->used)
2015 				continue;
2016 			rc = smcr_link_reg_rmb(lnk, buf_desc);
2017 			if (rc)
2018 				goto out;
2019 		}
2020 	}
2021 out:
2022 	mutex_unlock(&lgr->rmbs_lock);
2023 	return rc;
2024 }
2025 
2026 static struct smc_buf_desc *smcr_new_buf_create(struct smc_link_group *lgr,
2027 						bool is_rmb, int bufsize)
2028 {
2029 	struct smc_buf_desc *buf_desc;
2030 
2031 	/* try to alloc a new buffer */
2032 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2033 	if (!buf_desc)
2034 		return ERR_PTR(-ENOMEM);
2035 
2036 	buf_desc->order = get_order(bufsize);
2037 	buf_desc->pages = alloc_pages(GFP_KERNEL | __GFP_NOWARN |
2038 				      __GFP_NOMEMALLOC | __GFP_COMP |
2039 				      __GFP_NORETRY | __GFP_ZERO,
2040 				      buf_desc->order);
2041 	if (!buf_desc->pages) {
2042 		kfree(buf_desc);
2043 		return ERR_PTR(-EAGAIN);
2044 	}
2045 	buf_desc->cpu_addr = (void *)page_address(buf_desc->pages);
2046 	buf_desc->len = bufsize;
2047 	return buf_desc;
2048 }
2049 
2050 /* map buf_desc on all usable links,
2051  * unused buffers stay mapped as long as the link is up
2052  */
2053 static int smcr_buf_map_usable_links(struct smc_link_group *lgr,
2054 				     struct smc_buf_desc *buf_desc, bool is_rmb)
2055 {
2056 	int i, rc = 0;
2057 
2058 	/* protect against parallel link reconfiguration */
2059 	mutex_lock(&lgr->llc_conf_mutex);
2060 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2061 		struct smc_link *lnk = &lgr->lnk[i];
2062 
2063 		if (!smc_link_usable(lnk))
2064 			continue;
2065 		if (smcr_buf_map_link(buf_desc, is_rmb, lnk)) {
2066 			rc = -ENOMEM;
2067 			goto out;
2068 		}
2069 	}
2070 out:
2071 	mutex_unlock(&lgr->llc_conf_mutex);
2072 	return rc;
2073 }
2074 
2075 static struct smc_buf_desc *smcd_new_buf_create(struct smc_link_group *lgr,
2076 						bool is_dmb, int bufsize)
2077 {
2078 	struct smc_buf_desc *buf_desc;
2079 	int rc;
2080 
2081 	/* try to alloc a new DMB */
2082 	buf_desc = kzalloc(sizeof(*buf_desc), GFP_KERNEL);
2083 	if (!buf_desc)
2084 		return ERR_PTR(-ENOMEM);
2085 	if (is_dmb) {
2086 		rc = smc_ism_register_dmb(lgr, bufsize, buf_desc);
2087 		if (rc) {
2088 			kfree(buf_desc);
2089 			if (rc == -ENOMEM)
2090 				return ERR_PTR(-EAGAIN);
2091 			if (rc == -ENOSPC)
2092 				return ERR_PTR(-ENOSPC);
2093 			return ERR_PTR(-EIO);
2094 		}
2095 		buf_desc->pages = virt_to_page(buf_desc->cpu_addr);
2096 		/* CDC header stored in buf. So, pretend it was smaller */
2097 		buf_desc->len = bufsize - sizeof(struct smcd_cdc_msg);
2098 	} else {
2099 		buf_desc->cpu_addr = kzalloc(bufsize, GFP_KERNEL |
2100 					     __GFP_NOWARN | __GFP_NORETRY |
2101 					     __GFP_NOMEMALLOC);
2102 		if (!buf_desc->cpu_addr) {
2103 			kfree(buf_desc);
2104 			return ERR_PTR(-EAGAIN);
2105 		}
2106 		buf_desc->len = bufsize;
2107 	}
2108 	return buf_desc;
2109 }
2110 
2111 static int __smc_buf_create(struct smc_sock *smc, bool is_smcd, bool is_rmb)
2112 {
2113 	struct smc_buf_desc *buf_desc = ERR_PTR(-ENOMEM);
2114 	struct smc_connection *conn = &smc->conn;
2115 	struct smc_link_group *lgr = conn->lgr;
2116 	struct list_head *buf_list;
2117 	int bufsize, bufsize_short;
2118 	bool is_dgraded = false;
2119 	struct mutex *lock;	/* lock buffer list */
2120 	int sk_buf_size;
2121 
2122 	if (is_rmb)
2123 		/* use socket recv buffer size (w/o overhead) as start value */
2124 		sk_buf_size = smc->sk.sk_rcvbuf / 2;
2125 	else
2126 		/* use socket send buffer size (w/o overhead) as start value */
2127 		sk_buf_size = smc->sk.sk_sndbuf / 2;
2128 
2129 	for (bufsize_short = smc_compress_bufsize(sk_buf_size, is_smcd, is_rmb);
2130 	     bufsize_short >= 0; bufsize_short--) {
2131 		if (is_rmb) {
2132 			lock = &lgr->rmbs_lock;
2133 			buf_list = &lgr->rmbs[bufsize_short];
2134 		} else {
2135 			lock = &lgr->sndbufs_lock;
2136 			buf_list = &lgr->sndbufs[bufsize_short];
2137 		}
2138 		bufsize = smc_uncompress_bufsize(bufsize_short);
2139 
2140 		/* check for reusable slot in the link group */
2141 		buf_desc = smc_buf_get_slot(bufsize_short, lock, buf_list);
2142 		if (buf_desc) {
2143 			SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2144 			SMC_STAT_BUF_REUSE(smc, is_smcd, is_rmb);
2145 			memset(buf_desc->cpu_addr, 0, bufsize);
2146 			break; /* found reusable slot */
2147 		}
2148 
2149 		if (is_smcd)
2150 			buf_desc = smcd_new_buf_create(lgr, is_rmb, bufsize);
2151 		else
2152 			buf_desc = smcr_new_buf_create(lgr, is_rmb, bufsize);
2153 
2154 		if (PTR_ERR(buf_desc) == -ENOMEM)
2155 			break;
2156 		if (IS_ERR(buf_desc)) {
2157 			if (!is_dgraded) {
2158 				is_dgraded = true;
2159 				SMC_STAT_RMB_DOWNGRADED(smc, is_smcd, is_rmb);
2160 			}
2161 			continue;
2162 		}
2163 
2164 		SMC_STAT_RMB_ALLOC(smc, is_smcd, is_rmb);
2165 		SMC_STAT_RMB_SIZE(smc, is_smcd, is_rmb, bufsize);
2166 		buf_desc->used = 1;
2167 		mutex_lock(lock);
2168 		list_add(&buf_desc->list, buf_list);
2169 		mutex_unlock(lock);
2170 		break; /* found */
2171 	}
2172 
2173 	if (IS_ERR(buf_desc))
2174 		return PTR_ERR(buf_desc);
2175 
2176 	if (!is_smcd) {
2177 		if (smcr_buf_map_usable_links(lgr, buf_desc, is_rmb)) {
2178 			smcr_buf_unuse(buf_desc, lgr);
2179 			return -ENOMEM;
2180 		}
2181 	}
2182 
2183 	if (is_rmb) {
2184 		conn->rmb_desc = buf_desc;
2185 		conn->rmbe_size_short = bufsize_short;
2186 		smc->sk.sk_rcvbuf = bufsize * 2;
2187 		atomic_set(&conn->bytes_to_rcv, 0);
2188 		conn->rmbe_update_limit =
2189 			smc_rmb_wnd_update_limit(buf_desc->len);
2190 		if (is_smcd)
2191 			smc_ism_set_conn(conn); /* map RMB/smcd_dev to conn */
2192 	} else {
2193 		conn->sndbuf_desc = buf_desc;
2194 		smc->sk.sk_sndbuf = bufsize * 2;
2195 		atomic_set(&conn->sndbuf_space, bufsize);
2196 	}
2197 	return 0;
2198 }
2199 
2200 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn)
2201 {
2202 	if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk))
2203 		return;
2204 	smc_ib_sync_sg_for_cpu(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2205 }
2206 
2207 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn)
2208 {
2209 	if (!conn->lgr || conn->lgr->is_smcd || !smc_link_active(conn->lnk))
2210 		return;
2211 	smc_ib_sync_sg_for_device(conn->lnk, conn->sndbuf_desc, DMA_TO_DEVICE);
2212 }
2213 
2214 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn)
2215 {
2216 	int i;
2217 
2218 	if (!conn->lgr || conn->lgr->is_smcd)
2219 		return;
2220 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2221 		if (!smc_link_active(&conn->lgr->lnk[i]))
2222 			continue;
2223 		smc_ib_sync_sg_for_cpu(&conn->lgr->lnk[i], conn->rmb_desc,
2224 				       DMA_FROM_DEVICE);
2225 	}
2226 }
2227 
2228 void smc_rmb_sync_sg_for_device(struct smc_connection *conn)
2229 {
2230 	int i;
2231 
2232 	if (!conn->lgr || conn->lgr->is_smcd)
2233 		return;
2234 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
2235 		if (!smc_link_active(&conn->lgr->lnk[i]))
2236 			continue;
2237 		smc_ib_sync_sg_for_device(&conn->lgr->lnk[i], conn->rmb_desc,
2238 					  DMA_FROM_DEVICE);
2239 	}
2240 }
2241 
2242 /* create the send and receive buffer for an SMC socket;
2243  * receive buffers are called RMBs;
2244  * (even though the SMC protocol allows more than one RMB-element per RMB,
2245  * the Linux implementation uses just one RMB-element per RMB, i.e. uses an
2246  * extra RMB for every connection in a link group
2247  */
2248 int smc_buf_create(struct smc_sock *smc, bool is_smcd)
2249 {
2250 	int rc;
2251 
2252 	/* create send buffer */
2253 	rc = __smc_buf_create(smc, is_smcd, false);
2254 	if (rc)
2255 		return rc;
2256 	/* create rmb */
2257 	rc = __smc_buf_create(smc, is_smcd, true);
2258 	if (rc) {
2259 		mutex_lock(&smc->conn.lgr->sndbufs_lock);
2260 		list_del(&smc->conn.sndbuf_desc->list);
2261 		mutex_unlock(&smc->conn.lgr->sndbufs_lock);
2262 		smc_buf_free(smc->conn.lgr, false, smc->conn.sndbuf_desc);
2263 		smc->conn.sndbuf_desc = NULL;
2264 	}
2265 	return rc;
2266 }
2267 
2268 static inline int smc_rmb_reserve_rtoken_idx(struct smc_link_group *lgr)
2269 {
2270 	int i;
2271 
2272 	for_each_clear_bit(i, lgr->rtokens_used_mask, SMC_RMBS_PER_LGR_MAX) {
2273 		if (!test_and_set_bit(i, lgr->rtokens_used_mask))
2274 			return i;
2275 	}
2276 	return -ENOSPC;
2277 }
2278 
2279 static int smc_rtoken_find_by_link(struct smc_link_group *lgr, int lnk_idx,
2280 				   u32 rkey)
2281 {
2282 	int i;
2283 
2284 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2285 		if (test_bit(i, lgr->rtokens_used_mask) &&
2286 		    lgr->rtokens[i][lnk_idx].rkey == rkey)
2287 			return i;
2288 	}
2289 	return -ENOENT;
2290 }
2291 
2292 /* set rtoken for a new link to an existing rmb */
2293 void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new,
2294 		    __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey)
2295 {
2296 	int rtok_idx;
2297 
2298 	rtok_idx = smc_rtoken_find_by_link(lgr, link_idx, ntohl(nw_rkey_known));
2299 	if (rtok_idx == -ENOENT)
2300 		return;
2301 	lgr->rtokens[rtok_idx][link_idx_new].rkey = ntohl(nw_rkey);
2302 	lgr->rtokens[rtok_idx][link_idx_new].dma_addr = be64_to_cpu(nw_vaddr);
2303 }
2304 
2305 /* set rtoken for a new link whose link_id is given */
2306 void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id,
2307 		     __be64 nw_vaddr, __be32 nw_rkey)
2308 {
2309 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2310 	u32 rkey = ntohl(nw_rkey);
2311 	bool found = false;
2312 	int link_idx;
2313 
2314 	for (link_idx = 0; link_idx < SMC_LINKS_PER_LGR_MAX; link_idx++) {
2315 		if (lgr->lnk[link_idx].link_id == link_id) {
2316 			found = true;
2317 			break;
2318 		}
2319 	}
2320 	if (!found)
2321 		return;
2322 	lgr->rtokens[rtok_idx][link_idx].rkey = rkey;
2323 	lgr->rtokens[rtok_idx][link_idx].dma_addr = dma_addr;
2324 }
2325 
2326 /* add a new rtoken from peer */
2327 int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey)
2328 {
2329 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2330 	u64 dma_addr = be64_to_cpu(nw_vaddr);
2331 	u32 rkey = ntohl(nw_rkey);
2332 	int i;
2333 
2334 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2335 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2336 		    lgr->rtokens[i][lnk->link_idx].dma_addr == dma_addr &&
2337 		    test_bit(i, lgr->rtokens_used_mask)) {
2338 			/* already in list */
2339 			return i;
2340 		}
2341 	}
2342 	i = smc_rmb_reserve_rtoken_idx(lgr);
2343 	if (i < 0)
2344 		return i;
2345 	lgr->rtokens[i][lnk->link_idx].rkey = rkey;
2346 	lgr->rtokens[i][lnk->link_idx].dma_addr = dma_addr;
2347 	return i;
2348 }
2349 
2350 /* delete an rtoken from all links */
2351 int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey)
2352 {
2353 	struct smc_link_group *lgr = smc_get_lgr(lnk);
2354 	u32 rkey = ntohl(nw_rkey);
2355 	int i, j;
2356 
2357 	for (i = 0; i < SMC_RMBS_PER_LGR_MAX; i++) {
2358 		if (lgr->rtokens[i][lnk->link_idx].rkey == rkey &&
2359 		    test_bit(i, lgr->rtokens_used_mask)) {
2360 			for (j = 0; j < SMC_LINKS_PER_LGR_MAX; j++) {
2361 				lgr->rtokens[i][j].rkey = 0;
2362 				lgr->rtokens[i][j].dma_addr = 0;
2363 			}
2364 			clear_bit(i, lgr->rtokens_used_mask);
2365 			return 0;
2366 		}
2367 	}
2368 	return -ENOENT;
2369 }
2370 
2371 /* save rkey and dma_addr received from peer during clc handshake */
2372 int smc_rmb_rtoken_handling(struct smc_connection *conn,
2373 			    struct smc_link *lnk,
2374 			    struct smc_clc_msg_accept_confirm *clc)
2375 {
2376 	conn->rtoken_idx = smc_rtoken_add(lnk, clc->r0.rmb_dma_addr,
2377 					  clc->r0.rmb_rkey);
2378 	if (conn->rtoken_idx < 0)
2379 		return conn->rtoken_idx;
2380 	return 0;
2381 }
2382 
2383 static void smc_core_going_away(void)
2384 {
2385 	struct smc_ib_device *smcibdev;
2386 	struct smcd_dev *smcd;
2387 
2388 	mutex_lock(&smc_ib_devices.mutex);
2389 	list_for_each_entry(smcibdev, &smc_ib_devices.list, list) {
2390 		int i;
2391 
2392 		for (i = 0; i < SMC_MAX_PORTS; i++)
2393 			set_bit(i, smcibdev->ports_going_away);
2394 	}
2395 	mutex_unlock(&smc_ib_devices.mutex);
2396 
2397 	mutex_lock(&smcd_dev_list.mutex);
2398 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2399 		smcd->going_away = 1;
2400 	}
2401 	mutex_unlock(&smcd_dev_list.mutex);
2402 }
2403 
2404 /* Clean up all SMC link groups */
2405 static void smc_lgrs_shutdown(void)
2406 {
2407 	struct smcd_dev *smcd;
2408 
2409 	smc_core_going_away();
2410 
2411 	smc_smcr_terminate_all(NULL);
2412 
2413 	mutex_lock(&smcd_dev_list.mutex);
2414 	list_for_each_entry(smcd, &smcd_dev_list.list, list)
2415 		smc_smcd_terminate_all(smcd);
2416 	mutex_unlock(&smcd_dev_list.mutex);
2417 }
2418 
2419 static int smc_core_reboot_event(struct notifier_block *this,
2420 				 unsigned long event, void *ptr)
2421 {
2422 	smc_lgrs_shutdown();
2423 	smc_ib_unregister_client();
2424 	return 0;
2425 }
2426 
2427 static struct notifier_block smc_reboot_notifier = {
2428 	.notifier_call = smc_core_reboot_event,
2429 };
2430 
2431 int __init smc_core_init(void)
2432 {
2433 	return register_reboot_notifier(&smc_reboot_notifier);
2434 }
2435 
2436 /* Called (from smc_exit) when module is removed */
2437 void smc_core_exit(void)
2438 {
2439 	unregister_reboot_notifier(&smc_reboot_notifier);
2440 	smc_lgrs_shutdown();
2441 }
2442