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