1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * Generic netlink support functions to configure an SMC-R PNET table 6 * 7 * Copyright IBM Corp. 2016 8 * 9 * Author(s): Thomas Richter <tmricht@linux.vnet.ibm.com> 10 */ 11 12 #include <linux/module.h> 13 #include <linux/list.h> 14 #include <linux/ctype.h> 15 #include <linux/mutex.h> 16 #include <net/netlink.h> 17 #include <net/genetlink.h> 18 19 #include <uapi/linux/if.h> 20 #include <uapi/linux/smc.h> 21 22 #include <rdma/ib_verbs.h> 23 24 #include <net/netns/generic.h> 25 #include "smc_netns.h" 26 27 #include "smc_pnet.h" 28 #include "smc_ib.h" 29 #include "smc_ism.h" 30 #include "smc_core.h" 31 32 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev); 33 static struct net_device *pnet_find_base_ndev(struct net_device *ndev); 34 35 static const struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = { 36 [SMC_PNETID_NAME] = { 37 .type = NLA_NUL_STRING, 38 .len = SMC_MAX_PNETID_LEN 39 }, 40 [SMC_PNETID_ETHNAME] = { 41 .type = NLA_NUL_STRING, 42 .len = IFNAMSIZ - 1 43 }, 44 [SMC_PNETID_IBNAME] = { 45 .type = NLA_NUL_STRING, 46 .len = IB_DEVICE_NAME_MAX - 1 47 }, 48 [SMC_PNETID_IBPORT] = { .type = NLA_U8 } 49 }; 50 51 static struct genl_family smc_pnet_nl_family; 52 53 enum smc_pnet_nametype { 54 SMC_PNET_ETH = 1, 55 SMC_PNET_IB = 2, 56 }; 57 58 /* pnet entry stored in pnet table */ 59 struct smc_pnetentry { 60 struct list_head list; 61 char pnet_name[SMC_MAX_PNETID_LEN + 1]; 62 enum smc_pnet_nametype type; 63 union { 64 struct { 65 char eth_name[IFNAMSIZ + 1]; 66 struct net_device *ndev; 67 netdevice_tracker dev_tracker; 68 }; 69 struct { 70 char ib_name[IB_DEVICE_NAME_MAX + 1]; 71 u8 ib_port; 72 }; 73 }; 74 }; 75 76 /* Check if the pnetid is set */ 77 bool smc_pnet_is_pnetid_set(u8 *pnetid) 78 { 79 if (pnetid[0] == 0 || pnetid[0] == _S) 80 return false; 81 return true; 82 } 83 84 /* Check if two given pnetids match */ 85 static bool smc_pnet_match(u8 *pnetid1, u8 *pnetid2) 86 { 87 int i; 88 89 for (i = 0; i < SMC_MAX_PNETID_LEN; i++) { 90 if ((pnetid1[i] == 0 || pnetid1[i] == _S) && 91 (pnetid2[i] == 0 || pnetid2[i] == _S)) 92 break; 93 if (pnetid1[i] != pnetid2[i]) 94 return false; 95 } 96 return true; 97 } 98 99 /* Remove a pnetid from the pnet table. 100 */ 101 static int smc_pnet_remove_by_pnetid(struct net *net, char *pnet_name) 102 { 103 struct smc_pnetentry *pnetelem, *tmp_pe; 104 struct smc_pnettable *pnettable; 105 struct smc_ib_device *ibdev; 106 struct smcd_dev *smcd_dev; 107 struct smc_net *sn; 108 int rc = -ENOENT; 109 int ibport; 110 111 /* get pnettable for namespace */ 112 sn = net_generic(net, smc_net_id); 113 pnettable = &sn->pnettable; 114 115 /* remove table entry */ 116 write_lock(&pnettable->lock); 117 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, 118 list) { 119 if (!pnet_name || 120 smc_pnet_match(pnetelem->pnet_name, pnet_name)) { 121 list_del(&pnetelem->list); 122 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev) { 123 dev_put_track(pnetelem->ndev, &pnetelem->dev_tracker); 124 pr_warn_ratelimited("smc: net device %s " 125 "erased user defined " 126 "pnetid %.16s\n", 127 pnetelem->eth_name, 128 pnetelem->pnet_name); 129 } 130 kfree(pnetelem); 131 rc = 0; 132 } 133 } 134 write_unlock(&pnettable->lock); 135 136 /* if this is not the initial namespace, stop here */ 137 if (net != &init_net) 138 return rc; 139 140 /* remove ib devices */ 141 mutex_lock(&smc_ib_devices.mutex); 142 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 143 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) { 144 if (ibdev->pnetid_by_user[ibport] && 145 (!pnet_name || 146 smc_pnet_match(pnet_name, 147 ibdev->pnetid[ibport]))) { 148 pr_warn_ratelimited("smc: ib device %s ibport " 149 "%d erased user defined " 150 "pnetid %.16s\n", 151 ibdev->ibdev->name, 152 ibport + 1, 153 ibdev->pnetid[ibport]); 154 memset(ibdev->pnetid[ibport], 0, 155 SMC_MAX_PNETID_LEN); 156 ibdev->pnetid_by_user[ibport] = false; 157 rc = 0; 158 } 159 } 160 } 161 mutex_unlock(&smc_ib_devices.mutex); 162 /* remove smcd devices */ 163 mutex_lock(&smcd_dev_list.mutex); 164 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) { 165 if (smcd_dev->pnetid_by_user && 166 (!pnet_name || 167 smc_pnet_match(pnet_name, smcd_dev->pnetid))) { 168 pr_warn_ratelimited("smc: smcd device %s " 169 "erased user defined pnetid " 170 "%.16s\n", dev_name(&smcd_dev->dev), 171 smcd_dev->pnetid); 172 memset(smcd_dev->pnetid, 0, SMC_MAX_PNETID_LEN); 173 smcd_dev->pnetid_by_user = false; 174 rc = 0; 175 } 176 } 177 mutex_unlock(&smcd_dev_list.mutex); 178 return rc; 179 } 180 181 /* Add the reference to a given network device to the pnet table. 182 */ 183 static int smc_pnet_add_by_ndev(struct net_device *ndev) 184 { 185 struct smc_pnetentry *pnetelem, *tmp_pe; 186 struct smc_pnettable *pnettable; 187 struct net *net = dev_net(ndev); 188 struct smc_net *sn; 189 int rc = -ENOENT; 190 191 /* get pnettable for namespace */ 192 sn = net_generic(net, smc_net_id); 193 pnettable = &sn->pnettable; 194 195 write_lock(&pnettable->lock); 196 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) { 197 if (pnetelem->type == SMC_PNET_ETH && !pnetelem->ndev && 198 !strncmp(pnetelem->eth_name, ndev->name, IFNAMSIZ)) { 199 dev_hold_track(ndev, &pnetelem->dev_tracker, GFP_ATOMIC); 200 pnetelem->ndev = ndev; 201 rc = 0; 202 pr_warn_ratelimited("smc: adding net device %s with " 203 "user defined pnetid %.16s\n", 204 pnetelem->eth_name, 205 pnetelem->pnet_name); 206 break; 207 } 208 } 209 write_unlock(&pnettable->lock); 210 return rc; 211 } 212 213 /* Remove the reference to a given network device from the pnet table. 214 */ 215 static int smc_pnet_remove_by_ndev(struct net_device *ndev) 216 { 217 struct smc_pnetentry *pnetelem, *tmp_pe; 218 struct smc_pnettable *pnettable; 219 struct net *net = dev_net(ndev); 220 struct smc_net *sn; 221 int rc = -ENOENT; 222 223 /* get pnettable for namespace */ 224 sn = net_generic(net, smc_net_id); 225 pnettable = &sn->pnettable; 226 227 write_lock(&pnettable->lock); 228 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) { 229 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev == ndev) { 230 dev_put_track(pnetelem->ndev, &pnetelem->dev_tracker); 231 pnetelem->ndev = NULL; 232 rc = 0; 233 pr_warn_ratelimited("smc: removing net device %s with " 234 "user defined pnetid %.16s\n", 235 pnetelem->eth_name, 236 pnetelem->pnet_name); 237 break; 238 } 239 } 240 write_unlock(&pnettable->lock); 241 return rc; 242 } 243 244 /* Apply pnetid to ib device when no pnetid is set. 245 */ 246 static bool smc_pnet_apply_ib(struct smc_ib_device *ib_dev, u8 ib_port, 247 char *pnet_name) 248 { 249 bool applied = false; 250 251 mutex_lock(&smc_ib_devices.mutex); 252 if (!smc_pnet_is_pnetid_set(ib_dev->pnetid[ib_port - 1])) { 253 memcpy(ib_dev->pnetid[ib_port - 1], pnet_name, 254 SMC_MAX_PNETID_LEN); 255 ib_dev->pnetid_by_user[ib_port - 1] = true; 256 applied = true; 257 } 258 mutex_unlock(&smc_ib_devices.mutex); 259 return applied; 260 } 261 262 /* Apply pnetid to smcd device when no pnetid is set. 263 */ 264 static bool smc_pnet_apply_smcd(struct smcd_dev *smcd_dev, char *pnet_name) 265 { 266 bool applied = false; 267 268 mutex_lock(&smcd_dev_list.mutex); 269 if (!smc_pnet_is_pnetid_set(smcd_dev->pnetid)) { 270 memcpy(smcd_dev->pnetid, pnet_name, SMC_MAX_PNETID_LEN); 271 smcd_dev->pnetid_by_user = true; 272 applied = true; 273 } 274 mutex_unlock(&smcd_dev_list.mutex); 275 return applied; 276 } 277 278 /* The limit for pnetid is 16 characters. 279 * Valid characters should be (single-byte character set) a-z, A-Z, 0-9. 280 * Lower case letters are converted to upper case. 281 * Interior blanks should not be used. 282 */ 283 static bool smc_pnetid_valid(const char *pnet_name, char *pnetid) 284 { 285 char *bf = skip_spaces(pnet_name); 286 size_t len = strlen(bf); 287 char *end = bf + len; 288 289 if (!len) 290 return false; 291 while (--end >= bf && isspace(*end)) 292 ; 293 if (end - bf >= SMC_MAX_PNETID_LEN) 294 return false; 295 while (bf <= end) { 296 if (!isalnum(*bf)) 297 return false; 298 *pnetid++ = islower(*bf) ? toupper(*bf) : *bf; 299 bf++; 300 } 301 *pnetid = '\0'; 302 return true; 303 } 304 305 /* Find an infiniband device by a given name. The device might not exist. */ 306 static struct smc_ib_device *smc_pnet_find_ib(char *ib_name) 307 { 308 struct smc_ib_device *ibdev; 309 310 mutex_lock(&smc_ib_devices.mutex); 311 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 312 if (!strncmp(ibdev->ibdev->name, ib_name, 313 sizeof(ibdev->ibdev->name)) || 314 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name, 315 IB_DEVICE_NAME_MAX - 1)) { 316 goto out; 317 } 318 } 319 ibdev = NULL; 320 out: 321 mutex_unlock(&smc_ib_devices.mutex); 322 return ibdev; 323 } 324 325 /* Find an smcd device by a given name. The device might not exist. */ 326 static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name) 327 { 328 struct smcd_dev *smcd_dev; 329 330 mutex_lock(&smcd_dev_list.mutex); 331 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) { 332 if (!strncmp(dev_name(&smcd_dev->dev), smcd_name, 333 IB_DEVICE_NAME_MAX - 1)) 334 goto out; 335 } 336 smcd_dev = NULL; 337 out: 338 mutex_unlock(&smcd_dev_list.mutex); 339 return smcd_dev; 340 } 341 342 static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net, 343 char *eth_name, char *pnet_name) 344 { 345 struct smc_pnetentry *tmp_pe, *new_pe; 346 struct net_device *ndev, *base_ndev; 347 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 348 bool new_netdev; 349 int rc; 350 351 /* check if (base) netdev already has a pnetid. If there is one, we do 352 * not want to add a pnet table entry 353 */ 354 rc = -EEXIST; 355 ndev = dev_get_by_name(net, eth_name); /* dev_hold() */ 356 if (ndev) { 357 base_ndev = pnet_find_base_ndev(ndev); 358 if (!smc_pnetid_by_dev_port(base_ndev->dev.parent, 359 base_ndev->dev_port, ndev_pnetid)) 360 goto out_put; 361 } 362 363 /* add a new netdev entry to the pnet table if there isn't one */ 364 rc = -ENOMEM; 365 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL); 366 if (!new_pe) 367 goto out_put; 368 new_pe->type = SMC_PNET_ETH; 369 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN); 370 strncpy(new_pe->eth_name, eth_name, IFNAMSIZ); 371 new_pe->ndev = ndev; 372 netdev_tracker_alloc(ndev, &new_pe->dev_tracker, GFP_KERNEL); 373 rc = -EEXIST; 374 new_netdev = true; 375 write_lock(&pnettable->lock); 376 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 377 if (tmp_pe->type == SMC_PNET_ETH && 378 !strncmp(tmp_pe->eth_name, eth_name, IFNAMSIZ)) { 379 new_netdev = false; 380 break; 381 } 382 } 383 if (new_netdev) { 384 list_add_tail(&new_pe->list, &pnettable->pnetlist); 385 write_unlock(&pnettable->lock); 386 } else { 387 write_unlock(&pnettable->lock); 388 kfree(new_pe); 389 goto out_put; 390 } 391 if (ndev) 392 pr_warn_ratelimited("smc: net device %s " 393 "applied user defined pnetid %.16s\n", 394 new_pe->eth_name, new_pe->pnet_name); 395 return 0; 396 397 out_put: 398 dev_put(ndev); 399 return rc; 400 } 401 402 static int smc_pnet_add_ib(struct smc_pnettable *pnettable, char *ib_name, 403 u8 ib_port, char *pnet_name) 404 { 405 struct smc_pnetentry *tmp_pe, *new_pe; 406 struct smc_ib_device *ib_dev; 407 bool smcddev_applied = true; 408 bool ibdev_applied = true; 409 struct smcd_dev *smcd_dev; 410 bool new_ibdev; 411 412 /* try to apply the pnetid to active devices */ 413 ib_dev = smc_pnet_find_ib(ib_name); 414 if (ib_dev) { 415 ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name); 416 if (ibdev_applied) 417 pr_warn_ratelimited("smc: ib device %s ibport %d " 418 "applied user defined pnetid " 419 "%.16s\n", ib_dev->ibdev->name, 420 ib_port, 421 ib_dev->pnetid[ib_port - 1]); 422 } 423 smcd_dev = smc_pnet_find_smcd(ib_name); 424 if (smcd_dev) { 425 smcddev_applied = smc_pnet_apply_smcd(smcd_dev, pnet_name); 426 if (smcddev_applied) 427 pr_warn_ratelimited("smc: smcd device %s " 428 "applied user defined pnetid " 429 "%.16s\n", dev_name(&smcd_dev->dev), 430 smcd_dev->pnetid); 431 } 432 /* Apply fails when a device has a hardware-defined pnetid set, do not 433 * add a pnet table entry in that case. 434 */ 435 if (!ibdev_applied || !smcddev_applied) 436 return -EEXIST; 437 438 /* add a new ib entry to the pnet table if there isn't one */ 439 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL); 440 if (!new_pe) 441 return -ENOMEM; 442 new_pe->type = SMC_PNET_IB; 443 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN); 444 strncpy(new_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX); 445 new_pe->ib_port = ib_port; 446 447 new_ibdev = true; 448 write_lock(&pnettable->lock); 449 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 450 if (tmp_pe->type == SMC_PNET_IB && 451 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 452 new_ibdev = false; 453 break; 454 } 455 } 456 if (new_ibdev) { 457 list_add_tail(&new_pe->list, &pnettable->pnetlist); 458 write_unlock(&pnettable->lock); 459 } else { 460 write_unlock(&pnettable->lock); 461 kfree(new_pe); 462 } 463 return (new_ibdev) ? 0 : -EEXIST; 464 } 465 466 /* Append a pnetid to the end of the pnet table if not already on this list. 467 */ 468 static int smc_pnet_enter(struct net *net, struct nlattr *tb[]) 469 { 470 char pnet_name[SMC_MAX_PNETID_LEN + 1]; 471 struct smc_pnettable *pnettable; 472 bool new_netdev = false; 473 bool new_ibdev = false; 474 struct smc_net *sn; 475 u8 ibport = 1; 476 char *string; 477 int rc; 478 479 /* get pnettable for namespace */ 480 sn = net_generic(net, smc_net_id); 481 pnettable = &sn->pnettable; 482 483 rc = -EINVAL; 484 if (!tb[SMC_PNETID_NAME]) 485 goto error; 486 string = (char *)nla_data(tb[SMC_PNETID_NAME]); 487 if (!smc_pnetid_valid(string, pnet_name)) 488 goto error; 489 490 if (tb[SMC_PNETID_ETHNAME]) { 491 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]); 492 rc = smc_pnet_add_eth(pnettable, net, string, pnet_name); 493 if (!rc) 494 new_netdev = true; 495 else if (rc != -EEXIST) 496 goto error; 497 } 498 499 /* if this is not the initial namespace, stop here */ 500 if (net != &init_net) 501 return new_netdev ? 0 : -EEXIST; 502 503 rc = -EINVAL; 504 if (tb[SMC_PNETID_IBNAME]) { 505 string = (char *)nla_data(tb[SMC_PNETID_IBNAME]); 506 string = strim(string); 507 if (tb[SMC_PNETID_IBPORT]) { 508 ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]); 509 if (ibport < 1 || ibport > SMC_MAX_PORTS) 510 goto error; 511 } 512 rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name); 513 if (!rc) 514 new_ibdev = true; 515 else if (rc != -EEXIST) 516 goto error; 517 } 518 return (new_netdev || new_ibdev) ? 0 : -EEXIST; 519 520 error: 521 return rc; 522 } 523 524 /* Convert an smc_pnetentry to a netlink attribute sequence */ 525 static int smc_pnet_set_nla(struct sk_buff *msg, 526 struct smc_pnetentry *pnetelem) 527 { 528 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name)) 529 return -1; 530 if (pnetelem->type == SMC_PNET_ETH) { 531 if (nla_put_string(msg, SMC_PNETID_ETHNAME, 532 pnetelem->eth_name)) 533 return -1; 534 } else { 535 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a")) 536 return -1; 537 } 538 if (pnetelem->type == SMC_PNET_IB) { 539 if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) || 540 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port)) 541 return -1; 542 } else { 543 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") || 544 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff)) 545 return -1; 546 } 547 548 return 0; 549 } 550 551 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info) 552 { 553 struct net *net = genl_info_net(info); 554 555 return smc_pnet_enter(net, info->attrs); 556 } 557 558 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info) 559 { 560 struct net *net = genl_info_net(info); 561 562 if (!info->attrs[SMC_PNETID_NAME]) 563 return -EINVAL; 564 return smc_pnet_remove_by_pnetid(net, 565 (char *)nla_data(info->attrs[SMC_PNETID_NAME])); 566 } 567 568 static int smc_pnet_dump_start(struct netlink_callback *cb) 569 { 570 cb->args[0] = 0; 571 return 0; 572 } 573 574 static int smc_pnet_dumpinfo(struct sk_buff *skb, 575 u32 portid, u32 seq, u32 flags, 576 struct smc_pnetentry *pnetelem) 577 { 578 void *hdr; 579 580 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family, 581 flags, SMC_PNETID_GET); 582 if (!hdr) 583 return -ENOMEM; 584 if (smc_pnet_set_nla(skb, pnetelem) < 0) { 585 genlmsg_cancel(skb, hdr); 586 return -EMSGSIZE; 587 } 588 genlmsg_end(skb, hdr); 589 return 0; 590 } 591 592 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid, 593 u32 seq, u8 *pnetid, int start_idx) 594 { 595 struct smc_pnettable *pnettable; 596 struct smc_pnetentry *pnetelem; 597 struct smc_net *sn; 598 int idx = 0; 599 600 /* get pnettable for namespace */ 601 sn = net_generic(net, smc_net_id); 602 pnettable = &sn->pnettable; 603 604 /* dump pnettable entries */ 605 read_lock(&pnettable->lock); 606 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 607 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid)) 608 continue; 609 if (idx++ < start_idx) 610 continue; 611 /* if this is not the initial namespace, dump only netdev */ 612 if (net != &init_net && pnetelem->type != SMC_PNET_ETH) 613 continue; 614 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI, 615 pnetelem)) { 616 --idx; 617 break; 618 } 619 } 620 read_unlock(&pnettable->lock); 621 return idx; 622 } 623 624 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb) 625 { 626 struct net *net = sock_net(skb->sk); 627 int idx; 628 629 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid, 630 cb->nlh->nlmsg_seq, NULL, cb->args[0]); 631 632 cb->args[0] = idx; 633 return skb->len; 634 } 635 636 /* Retrieve one PNETID entry */ 637 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info) 638 { 639 struct net *net = genl_info_net(info); 640 struct sk_buff *msg; 641 void *hdr; 642 643 if (!info->attrs[SMC_PNETID_NAME]) 644 return -EINVAL; 645 646 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 647 if (!msg) 648 return -ENOMEM; 649 650 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq, 651 nla_data(info->attrs[SMC_PNETID_NAME]), 0); 652 653 /* finish multi part message and send it */ 654 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0, 655 NLM_F_MULTI); 656 if (!hdr) { 657 nlmsg_free(msg); 658 return -EMSGSIZE; 659 } 660 return genlmsg_reply(msg, info); 661 } 662 663 /* Remove and delete all pnetids from pnet table. 664 */ 665 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info) 666 { 667 struct net *net = genl_info_net(info); 668 669 smc_pnet_remove_by_pnetid(net, NULL); 670 return 0; 671 } 672 673 /* SMC_PNETID generic netlink operation definition */ 674 static const struct genl_ops smc_pnet_ops[] = { 675 { 676 .cmd = SMC_PNETID_GET, 677 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 678 /* can be retrieved by unprivileged users */ 679 .doit = smc_pnet_get, 680 .dumpit = smc_pnet_dump, 681 .start = smc_pnet_dump_start 682 }, 683 { 684 .cmd = SMC_PNETID_ADD, 685 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 686 .flags = GENL_ADMIN_PERM, 687 .doit = smc_pnet_add 688 }, 689 { 690 .cmd = SMC_PNETID_DEL, 691 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 692 .flags = GENL_ADMIN_PERM, 693 .doit = smc_pnet_del 694 }, 695 { 696 .cmd = SMC_PNETID_FLUSH, 697 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 698 .flags = GENL_ADMIN_PERM, 699 .doit = smc_pnet_flush 700 } 701 }; 702 703 /* SMC_PNETID family definition */ 704 static struct genl_family smc_pnet_nl_family __ro_after_init = { 705 .hdrsize = 0, 706 .name = SMCR_GENL_FAMILY_NAME, 707 .version = SMCR_GENL_FAMILY_VERSION, 708 .maxattr = SMC_PNETID_MAX, 709 .policy = smc_pnet_policy, 710 .netnsok = true, 711 .module = THIS_MODULE, 712 .ops = smc_pnet_ops, 713 .n_ops = ARRAY_SIZE(smc_pnet_ops) 714 }; 715 716 bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid) 717 { 718 struct smc_net *sn = net_generic(net, smc_net_id); 719 struct smc_pnetids_ndev_entry *pe; 720 bool rc = false; 721 722 read_lock(&sn->pnetids_ndev.lock); 723 list_for_each_entry(pe, &sn->pnetids_ndev.list, list) { 724 if (smc_pnet_match(pnetid, pe->pnetid)) { 725 rc = true; 726 goto unlock; 727 } 728 } 729 730 unlock: 731 read_unlock(&sn->pnetids_ndev.lock); 732 return rc; 733 } 734 735 static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid) 736 { 737 struct smc_net *sn = net_generic(net, smc_net_id); 738 struct smc_pnetids_ndev_entry *pe, *pi; 739 740 pe = kzalloc(sizeof(*pe), GFP_KERNEL); 741 if (!pe) 742 return -ENOMEM; 743 744 write_lock(&sn->pnetids_ndev.lock); 745 list_for_each_entry(pi, &sn->pnetids_ndev.list, list) { 746 if (smc_pnet_match(pnetid, pe->pnetid)) { 747 refcount_inc(&pi->refcnt); 748 kfree(pe); 749 goto unlock; 750 } 751 } 752 refcount_set(&pe->refcnt, 1); 753 memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN); 754 list_add_tail(&pe->list, &sn->pnetids_ndev.list); 755 756 unlock: 757 write_unlock(&sn->pnetids_ndev.lock); 758 return 0; 759 } 760 761 static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid) 762 { 763 struct smc_net *sn = net_generic(net, smc_net_id); 764 struct smc_pnetids_ndev_entry *pe, *pe2; 765 766 write_lock(&sn->pnetids_ndev.lock); 767 list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) { 768 if (smc_pnet_match(pnetid, pe->pnetid)) { 769 if (refcount_dec_and_test(&pe->refcnt)) { 770 list_del(&pe->list); 771 kfree(pe); 772 } 773 break; 774 } 775 } 776 write_unlock(&sn->pnetids_ndev.lock); 777 } 778 779 static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev, 780 u8 *ndev_pnetid) 781 { 782 struct net_device *base_dev; 783 784 base_dev = __pnet_find_base_ndev(dev); 785 if (base_dev->flags & IFF_UP && 786 !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port, 787 ndev_pnetid)) { 788 /* add to PNETIDs list */ 789 smc_pnet_add_pnetid(net, ndev_pnetid); 790 } 791 } 792 793 /* create initial list of netdevice pnetids */ 794 static void smc_pnet_create_pnetids_list(struct net *net) 795 { 796 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 797 struct net_device *dev; 798 799 rtnl_lock(); 800 for_each_netdev(net, dev) 801 smc_pnet_add_base_pnetid(net, dev, ndev_pnetid); 802 rtnl_unlock(); 803 } 804 805 /* clean up list of netdevice pnetids */ 806 static void smc_pnet_destroy_pnetids_list(struct net *net) 807 { 808 struct smc_net *sn = net_generic(net, smc_net_id); 809 struct smc_pnetids_ndev_entry *pe, *temp_pe; 810 811 write_lock(&sn->pnetids_ndev.lock); 812 list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) { 813 list_del(&pe->list); 814 kfree(pe); 815 } 816 write_unlock(&sn->pnetids_ndev.lock); 817 } 818 819 static int smc_pnet_netdev_event(struct notifier_block *this, 820 unsigned long event, void *ptr) 821 { 822 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 823 struct net *net = dev_net(event_dev); 824 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 825 826 switch (event) { 827 case NETDEV_REBOOT: 828 case NETDEV_UNREGISTER: 829 smc_pnet_remove_by_ndev(event_dev); 830 smc_ib_ndev_change(event_dev, event); 831 return NOTIFY_OK; 832 case NETDEV_REGISTER: 833 smc_pnet_add_by_ndev(event_dev); 834 smc_ib_ndev_change(event_dev, event); 835 return NOTIFY_OK; 836 case NETDEV_UP: 837 smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid); 838 return NOTIFY_OK; 839 case NETDEV_DOWN: 840 event_dev = __pnet_find_base_ndev(event_dev); 841 if (!smc_pnetid_by_dev_port(event_dev->dev.parent, 842 event_dev->dev_port, ndev_pnetid)) { 843 /* remove from PNETIDs list */ 844 smc_pnet_remove_pnetid(net, ndev_pnetid); 845 } 846 return NOTIFY_OK; 847 default: 848 return NOTIFY_DONE; 849 } 850 } 851 852 static struct notifier_block smc_netdev_notifier = { 853 .notifier_call = smc_pnet_netdev_event 854 }; 855 856 /* init network namespace */ 857 int smc_pnet_net_init(struct net *net) 858 { 859 struct smc_net *sn = net_generic(net, smc_net_id); 860 struct smc_pnettable *pnettable = &sn->pnettable; 861 struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev; 862 863 INIT_LIST_HEAD(&pnettable->pnetlist); 864 rwlock_init(&pnettable->lock); 865 INIT_LIST_HEAD(&pnetids_ndev->list); 866 rwlock_init(&pnetids_ndev->lock); 867 868 smc_pnet_create_pnetids_list(net); 869 870 return 0; 871 } 872 873 int __init smc_pnet_init(void) 874 { 875 int rc; 876 877 rc = genl_register_family(&smc_pnet_nl_family); 878 if (rc) 879 return rc; 880 rc = register_netdevice_notifier(&smc_netdev_notifier); 881 if (rc) 882 genl_unregister_family(&smc_pnet_nl_family); 883 884 return rc; 885 } 886 887 /* exit network namespace */ 888 void smc_pnet_net_exit(struct net *net) 889 { 890 /* flush pnet table */ 891 smc_pnet_remove_by_pnetid(net, NULL); 892 smc_pnet_destroy_pnetids_list(net); 893 } 894 895 void smc_pnet_exit(void) 896 { 897 unregister_netdevice_notifier(&smc_netdev_notifier); 898 genl_unregister_family(&smc_pnet_nl_family); 899 } 900 901 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev) 902 { 903 int i, nest_lvl; 904 905 ASSERT_RTNL(); 906 nest_lvl = ndev->lower_level; 907 for (i = 0; i < nest_lvl; i++) { 908 struct list_head *lower = &ndev->adj_list.lower; 909 910 if (list_empty(lower)) 911 break; 912 lower = lower->next; 913 ndev = netdev_lower_get_next(ndev, &lower); 914 } 915 return ndev; 916 } 917 918 /* Determine one base device for stacked net devices. 919 * If the lower device level contains more than one devices 920 * (for instance with bonding slaves), just the first device 921 * is used to reach a base device. 922 */ 923 static struct net_device *pnet_find_base_ndev(struct net_device *ndev) 924 { 925 rtnl_lock(); 926 ndev = __pnet_find_base_ndev(ndev); 927 rtnl_unlock(); 928 return ndev; 929 } 930 931 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev, 932 u8 *pnetid) 933 { 934 struct smc_pnettable *pnettable; 935 struct net *net = dev_net(ndev); 936 struct smc_pnetentry *pnetelem; 937 struct smc_net *sn; 938 int rc = -ENOENT; 939 940 /* get pnettable for namespace */ 941 sn = net_generic(net, smc_net_id); 942 pnettable = &sn->pnettable; 943 944 read_lock(&pnettable->lock); 945 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 946 if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) { 947 /* get pnetid of netdev device */ 948 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN); 949 rc = 0; 950 break; 951 } 952 } 953 read_unlock(&pnettable->lock); 954 return rc; 955 } 956 957 static int smc_pnet_determine_gid(struct smc_ib_device *ibdev, int i, 958 struct smc_init_info *ini) 959 { 960 if (!ini->check_smcrv2 && 961 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->ib_gid, NULL, 962 NULL)) { 963 ini->ib_dev = ibdev; 964 ini->ib_port = i; 965 return 0; 966 } 967 if (ini->check_smcrv2 && 968 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->smcrv2.ib_gid_v2, 969 NULL, &ini->smcrv2)) { 970 ini->smcrv2.ib_dev_v2 = ibdev; 971 ini->smcrv2.ib_port_v2 = i; 972 return 0; 973 } 974 return -ENODEV; 975 } 976 977 /* find a roce device for the given pnetid */ 978 static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id, 979 struct smc_init_info *ini, 980 struct smc_ib_device *known_dev, 981 struct net *net) 982 { 983 struct smc_ib_device *ibdev; 984 int i; 985 986 mutex_lock(&smc_ib_devices.mutex); 987 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 988 if (ibdev == known_dev || 989 !rdma_dev_access_netns(ibdev->ibdev, net)) 990 continue; 991 for (i = 1; i <= SMC_MAX_PORTS; i++) { 992 if (!rdma_is_port_valid(ibdev->ibdev, i)) 993 continue; 994 if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) && 995 smc_ib_port_active(ibdev, i) && 996 !test_bit(i - 1, ibdev->ports_going_away)) { 997 if (!smc_pnet_determine_gid(ibdev, i, ini)) 998 goto out; 999 } 1000 } 1001 } 1002 out: 1003 mutex_unlock(&smc_ib_devices.mutex); 1004 } 1005 1006 /* find alternate roce device with same pnet_id, vlan_id and net namespace */ 1007 void smc_pnet_find_alt_roce(struct smc_link_group *lgr, 1008 struct smc_init_info *ini, 1009 struct smc_ib_device *known_dev) 1010 { 1011 struct net *net = lgr->net; 1012 1013 _smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev, net); 1014 } 1015 1016 /* if handshake network device belongs to a roce device, return its 1017 * IB device and port 1018 */ 1019 static void smc_pnet_find_rdma_dev(struct net_device *netdev, 1020 struct smc_init_info *ini) 1021 { 1022 struct net *net = dev_net(netdev); 1023 struct smc_ib_device *ibdev; 1024 1025 mutex_lock(&smc_ib_devices.mutex); 1026 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 1027 struct net_device *ndev; 1028 int i; 1029 1030 /* check rdma net namespace */ 1031 if (!rdma_dev_access_netns(ibdev->ibdev, net)) 1032 continue; 1033 1034 for (i = 1; i <= SMC_MAX_PORTS; i++) { 1035 if (!rdma_is_port_valid(ibdev->ibdev, i)) 1036 continue; 1037 if (!ibdev->ibdev->ops.get_netdev) 1038 continue; 1039 ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i); 1040 if (!ndev) 1041 continue; 1042 dev_put(ndev); 1043 if (netdev == ndev && 1044 smc_ib_port_active(ibdev, i) && 1045 !test_bit(i - 1, ibdev->ports_going_away)) { 1046 if (!smc_pnet_determine_gid(ibdev, i, ini)) 1047 break; 1048 } 1049 } 1050 } 1051 mutex_unlock(&smc_ib_devices.mutex); 1052 } 1053 1054 /* Determine the corresponding IB device port based on the hardware PNETID. 1055 * Searching stops at the first matching active IB device port with vlan_id 1056 * configured. 1057 * If nothing found, check pnetid table. 1058 * If nothing found, try to use handshake device 1059 */ 1060 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev, 1061 struct smc_init_info *ini) 1062 { 1063 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1064 struct net *net; 1065 1066 ndev = pnet_find_base_ndev(ndev); 1067 net = dev_net(ndev); 1068 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port, 1069 ndev_pnetid) && 1070 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) { 1071 smc_pnet_find_rdma_dev(ndev, ini); 1072 return; /* pnetid could not be determined */ 1073 } 1074 _smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL, net); 1075 } 1076 1077 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev, 1078 struct smc_init_info *ini) 1079 { 1080 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1081 struct smcd_dev *ismdev; 1082 1083 ndev = pnet_find_base_ndev(ndev); 1084 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port, 1085 ndev_pnetid) && 1086 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) 1087 return; /* pnetid could not be determined */ 1088 1089 mutex_lock(&smcd_dev_list.mutex); 1090 list_for_each_entry(ismdev, &smcd_dev_list.list, list) { 1091 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) && 1092 !ismdev->going_away && 1093 (!ini->ism_peer_gid[0] || 1094 !smc_ism_cantalk(ini->ism_peer_gid[0], ini->vlan_id, 1095 ismdev))) { 1096 ini->ism_dev[0] = ismdev; 1097 break; 1098 } 1099 } 1100 mutex_unlock(&smcd_dev_list.mutex); 1101 } 1102 1103 /* PNET table analysis for a given sock: 1104 * determine ib_device and port belonging to used internal TCP socket 1105 * ethernet interface. 1106 */ 1107 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini) 1108 { 1109 struct dst_entry *dst = sk_dst_get(sk); 1110 1111 if (!dst) 1112 goto out; 1113 if (!dst->dev) 1114 goto out_rel; 1115 1116 smc_pnet_find_roce_by_pnetid(dst->dev, ini); 1117 1118 out_rel: 1119 dst_release(dst); 1120 out: 1121 return; 1122 } 1123 1124 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini) 1125 { 1126 struct dst_entry *dst = sk_dst_get(sk); 1127 1128 ini->ism_dev[0] = NULL; 1129 if (!dst) 1130 goto out; 1131 if (!dst->dev) 1132 goto out_rel; 1133 1134 smc_pnet_find_ism_by_pnetid(dst->dev, ini); 1135 1136 out_rel: 1137 dst_release(dst); 1138 out: 1139 return; 1140 } 1141 1142 /* Lookup and apply a pnet table entry to the given ib device. 1143 */ 1144 int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port) 1145 { 1146 char *ib_name = smcibdev->ibdev->name; 1147 struct smc_pnettable *pnettable; 1148 struct smc_pnetentry *tmp_pe; 1149 struct smc_net *sn; 1150 int rc = -ENOENT; 1151 1152 /* get pnettable for init namespace */ 1153 sn = net_generic(&init_net, smc_net_id); 1154 pnettable = &sn->pnettable; 1155 1156 read_lock(&pnettable->lock); 1157 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1158 if (tmp_pe->type == SMC_PNET_IB && 1159 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) && 1160 tmp_pe->ib_port == ib_port) { 1161 smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name); 1162 rc = 0; 1163 break; 1164 } 1165 } 1166 read_unlock(&pnettable->lock); 1167 1168 return rc; 1169 } 1170 1171 /* Lookup and apply a pnet table entry to the given smcd device. 1172 */ 1173 int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev) 1174 { 1175 const char *ib_name = dev_name(&smcddev->dev); 1176 struct smc_pnettable *pnettable; 1177 struct smc_pnetentry *tmp_pe; 1178 struct smc_net *sn; 1179 int rc = -ENOENT; 1180 1181 /* get pnettable for init namespace */ 1182 sn = net_generic(&init_net, smc_net_id); 1183 pnettable = &sn->pnettable; 1184 1185 read_lock(&pnettable->lock); 1186 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1187 if (tmp_pe->type == SMC_PNET_IB && 1188 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 1189 smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name); 1190 rc = 0; 1191 break; 1192 } 1193 } 1194 read_unlock(&pnettable->lock); 1195 1196 return rc; 1197 } 1198