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; 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 mutex_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 netdev_put(pnetelem->ndev, 124 &pnetelem->dev_tracker); 125 pr_warn_ratelimited("smc: net device %s " 126 "erased user defined " 127 "pnetid %.16s\n", 128 pnetelem->eth_name, 129 pnetelem->pnet_name); 130 } 131 kfree(pnetelem); 132 rc = 0; 133 } 134 } 135 mutex_unlock(&pnettable->lock); 136 137 /* if this is not the initial namespace, stop here */ 138 if (net != &init_net) 139 return rc; 140 141 /* remove ib devices */ 142 mutex_lock(&smc_ib_devices.mutex); 143 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 144 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) { 145 if (ibdev->pnetid_by_user[ibport] && 146 (!pnet_name || 147 smc_pnet_match(pnet_name, 148 ibdev->pnetid[ibport]))) { 149 pr_warn_ratelimited("smc: ib device %s ibport " 150 "%d erased user defined " 151 "pnetid %.16s\n", 152 ibdev->ibdev->name, 153 ibport + 1, 154 ibdev->pnetid[ibport]); 155 memset(ibdev->pnetid[ibport], 0, 156 SMC_MAX_PNETID_LEN); 157 ibdev->pnetid_by_user[ibport] = false; 158 rc = 0; 159 } 160 } 161 } 162 mutex_unlock(&smc_ib_devices.mutex); 163 /* remove smcd devices */ 164 mutex_lock(&smcd_dev_list.mutex); 165 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 166 if (smcd->pnetid_by_user && 167 (!pnet_name || 168 smc_pnet_match(pnet_name, smcd->pnetid))) { 169 pr_warn_ratelimited("smc: smcd device %s " 170 "erased user defined pnetid " 171 "%.16s\n", 172 dev_name(&smcd->dibs->dev), 173 smcd->pnetid); 174 memset(smcd->pnetid, 0, SMC_MAX_PNETID_LEN); 175 smcd->pnetid_by_user = false; 176 rc = 0; 177 } 178 } 179 mutex_unlock(&smcd_dev_list.mutex); 180 return rc; 181 } 182 183 /* Add the reference to a given network device to the pnet table. 184 */ 185 static int smc_pnet_add_by_ndev(struct net_device *ndev) 186 { 187 struct smc_pnetentry *pnetelem, *tmp_pe; 188 struct smc_pnettable *pnettable; 189 struct net *net = dev_net(ndev); 190 struct smc_net *sn; 191 int rc = -ENOENT; 192 193 /* get pnettable for namespace */ 194 sn = net_generic(net, smc_net_id); 195 pnettable = &sn->pnettable; 196 197 mutex_lock(&pnettable->lock); 198 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) { 199 if (pnetelem->type == SMC_PNET_ETH && !pnetelem->ndev && 200 !strncmp(pnetelem->eth_name, ndev->name, IFNAMSIZ)) { 201 netdev_hold(ndev, &pnetelem->dev_tracker, GFP_ATOMIC); 202 pnetelem->ndev = ndev; 203 rc = 0; 204 pr_warn_ratelimited("smc: adding net device %s with " 205 "user defined pnetid %.16s\n", 206 pnetelem->eth_name, 207 pnetelem->pnet_name); 208 break; 209 } 210 } 211 mutex_unlock(&pnettable->lock); 212 return rc; 213 } 214 215 /* Remove the reference to a given network device from the pnet table. 216 */ 217 static int smc_pnet_remove_by_ndev(struct net_device *ndev) 218 { 219 struct smc_pnetentry *pnetelem, *tmp_pe; 220 struct smc_pnettable *pnettable; 221 struct net *net = dev_net(ndev); 222 struct smc_net *sn; 223 int rc = -ENOENT; 224 225 /* get pnettable for namespace */ 226 sn = net_generic(net, smc_net_id); 227 pnettable = &sn->pnettable; 228 229 mutex_lock(&pnettable->lock); 230 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) { 231 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev == ndev) { 232 netdev_put(pnetelem->ndev, &pnetelem->dev_tracker); 233 pnetelem->ndev = NULL; 234 rc = 0; 235 pr_warn_ratelimited("smc: removing net device %s with " 236 "user defined pnetid %.16s\n", 237 pnetelem->eth_name, 238 pnetelem->pnet_name); 239 break; 240 } 241 } 242 mutex_unlock(&pnettable->lock); 243 return rc; 244 } 245 246 /* Apply pnetid to ib device when no pnetid is set. 247 */ 248 static bool smc_pnet_apply_ib(struct smc_ib_device *ib_dev, u8 ib_port, 249 char *pnet_name) 250 { 251 bool applied = false; 252 253 mutex_lock(&smc_ib_devices.mutex); 254 if (!smc_pnet_is_pnetid_set(ib_dev->pnetid[ib_port - 1])) { 255 memcpy(ib_dev->pnetid[ib_port - 1], pnet_name, 256 SMC_MAX_PNETID_LEN); 257 ib_dev->pnetid_by_user[ib_port - 1] = true; 258 applied = true; 259 } 260 mutex_unlock(&smc_ib_devices.mutex); 261 return applied; 262 } 263 264 /* Apply pnetid to smcd device when no pnetid is set. 265 */ 266 static bool smc_pnet_apply_smcd(struct smcd_dev *smcd_dev, char *pnet_name) 267 { 268 bool applied = false; 269 270 mutex_lock(&smcd_dev_list.mutex); 271 if (!smc_pnet_is_pnetid_set(smcd_dev->pnetid)) { 272 memcpy(smcd_dev->pnetid, pnet_name, SMC_MAX_PNETID_LEN); 273 smcd_dev->pnetid_by_user = true; 274 applied = true; 275 } 276 mutex_unlock(&smcd_dev_list.mutex); 277 return applied; 278 } 279 280 /* The limit for pnetid is 16 characters. 281 * Valid characters should be (single-byte character set) a-z, A-Z, 0-9. 282 * Lower case letters are converted to upper case. 283 * Interior blanks should not be used. 284 */ 285 static bool smc_pnetid_valid(const char *pnet_name, char *pnetid) 286 { 287 char *bf = skip_spaces(pnet_name); 288 size_t len = strlen(bf); 289 char *end = bf + len; 290 291 if (!len) 292 return false; 293 while (--end >= bf && isspace(*end)) 294 ; 295 if (end - bf >= SMC_MAX_PNETID_LEN) 296 return false; 297 while (bf <= end) { 298 if (!isalnum(*bf)) 299 return false; 300 *pnetid++ = islower(*bf) ? toupper(*bf) : *bf; 301 bf++; 302 } 303 *pnetid = '\0'; 304 return true; 305 } 306 307 /* Find an infiniband device by a given name. The device might not exist. */ 308 static struct smc_ib_device *smc_pnet_find_ib(char *ib_name) 309 { 310 struct smc_ib_device *ibdev; 311 312 mutex_lock(&smc_ib_devices.mutex); 313 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 314 if (!strncmp(ibdev->ibdev->name, ib_name, 315 sizeof(ibdev->ibdev->name)) || 316 (ibdev->ibdev->dev.parent && 317 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name, 318 IB_DEVICE_NAME_MAX - 1))) { 319 goto out; 320 } 321 } 322 ibdev = NULL; 323 out: 324 mutex_unlock(&smc_ib_devices.mutex); 325 return ibdev; 326 } 327 328 /* Find an smcd device by a given name. The device might not exist. */ 329 static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name) 330 { 331 struct smcd_dev *smcd_dev; 332 333 mutex_lock(&smcd_dev_list.mutex); 334 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) { 335 if (!strncmp(dev_name(&smcd_dev->dibs->dev), 336 smcd_name, IB_DEVICE_NAME_MAX - 1)) 337 goto out; 338 } 339 smcd_dev = NULL; 340 out: 341 mutex_unlock(&smcd_dev_list.mutex); 342 return smcd_dev; 343 } 344 345 static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net, 346 char *eth_name, char *pnet_name) 347 { 348 struct smc_pnetentry *tmp_pe, *new_pe; 349 struct net_device *ndev, *base_ndev; 350 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 351 bool new_netdev; 352 int rc; 353 354 /* check if (base) netdev already has a pnetid. If there is one, we do 355 * not want to add a pnet table entry 356 */ 357 rc = -EEXIST; 358 ndev = dev_get_by_name(net, eth_name); /* dev_hold() */ 359 if (ndev) { 360 base_ndev = pnet_find_base_ndev(ndev); 361 if (!smc_pnetid_by_dev_port(base_ndev->dev.parent, 362 base_ndev->dev_port, ndev_pnetid)) 363 goto out_put; 364 } 365 366 /* add a new netdev entry to the pnet table if there isn't one */ 367 rc = -ENOMEM; 368 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL); 369 if (!new_pe) 370 goto out_put; 371 new_pe->type = SMC_PNET_ETH; 372 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN); 373 strscpy(new_pe->eth_name, eth_name); 374 rc = -EEXIST; 375 new_netdev = true; 376 mutex_lock(&pnettable->lock); 377 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 378 if (tmp_pe->type == SMC_PNET_ETH && 379 !strncmp(tmp_pe->eth_name, eth_name, IFNAMSIZ)) { 380 new_netdev = false; 381 break; 382 } 383 } 384 if (new_netdev) { 385 if (ndev) { 386 new_pe->ndev = ndev; 387 netdev_tracker_alloc(ndev, &new_pe->dev_tracker, 388 GFP_ATOMIC); 389 } 390 list_add_tail(&new_pe->list, &pnettable->pnetlist); 391 mutex_unlock(&pnettable->lock); 392 } else { 393 mutex_unlock(&pnettable->lock); 394 kfree(new_pe); 395 goto out_put; 396 } 397 if (ndev) 398 pr_warn_ratelimited("smc: net device %s " 399 "applied user defined pnetid %.16s\n", 400 new_pe->eth_name, new_pe->pnet_name); 401 return 0; 402 403 out_put: 404 dev_put(ndev); 405 return rc; 406 } 407 408 static int smc_pnet_add_ib(struct smc_pnettable *pnettable, char *ib_name, 409 u8 ib_port, char *pnet_name) 410 { 411 struct smc_pnetentry *tmp_pe, *new_pe; 412 struct smc_ib_device *ib_dev; 413 bool smcddev_applied = true; 414 bool ibdev_applied = true; 415 struct smcd_dev *smcd; 416 bool new_ibdev; 417 418 /* try to apply the pnetid to active devices */ 419 ib_dev = smc_pnet_find_ib(ib_name); 420 if (ib_dev) { 421 ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name); 422 if (ibdev_applied) 423 pr_warn_ratelimited("smc: ib device %s ibport %d " 424 "applied user defined pnetid " 425 "%.16s\n", ib_dev->ibdev->name, 426 ib_port, 427 ib_dev->pnetid[ib_port - 1]); 428 } 429 smcd = smc_pnet_find_smcd(ib_name); 430 if (smcd) { 431 smcddev_applied = smc_pnet_apply_smcd(smcd, pnet_name); 432 if (smcddev_applied) { 433 pr_warn_ratelimited("smc: smcd device %s applied user defined pnetid %.16s\n", 434 dev_name(&smcd->dibs->dev), 435 smcd->pnetid); 436 } 437 } 438 /* Apply fails when a device has a hardware-defined pnetid set, do not 439 * add a pnet table entry in that case. 440 */ 441 if (!ibdev_applied || !smcddev_applied) 442 return -EEXIST; 443 444 /* add a new ib entry to the pnet table if there isn't one */ 445 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL); 446 if (!new_pe) 447 return -ENOMEM; 448 new_pe->type = SMC_PNET_IB; 449 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN); 450 strscpy(new_pe->ib_name, ib_name); 451 new_pe->ib_port = ib_port; 452 453 new_ibdev = true; 454 mutex_lock(&pnettable->lock); 455 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 456 if (tmp_pe->type == SMC_PNET_IB && 457 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 458 new_ibdev = false; 459 break; 460 } 461 } 462 if (new_ibdev) { 463 list_add_tail(&new_pe->list, &pnettable->pnetlist); 464 mutex_unlock(&pnettable->lock); 465 } else { 466 mutex_unlock(&pnettable->lock); 467 kfree(new_pe); 468 } 469 return (new_ibdev) ? 0 : -EEXIST; 470 } 471 472 /* Append a pnetid to the end of the pnet table if not already on this list. 473 */ 474 static int smc_pnet_enter(struct net *net, struct nlattr *tb[]) 475 { 476 char pnet_name[SMC_MAX_PNETID_LEN + 1]; 477 struct smc_pnettable *pnettable; 478 bool new_netdev = false; 479 bool new_ibdev = false; 480 struct smc_net *sn; 481 u8 ibport = 1; 482 char *string; 483 int rc; 484 485 /* get pnettable for namespace */ 486 sn = net_generic(net, smc_net_id); 487 pnettable = &sn->pnettable; 488 489 rc = -EINVAL; 490 if (!tb[SMC_PNETID_NAME]) 491 goto error; 492 string = (char *)nla_data(tb[SMC_PNETID_NAME]); 493 if (!smc_pnetid_valid(string, pnet_name)) 494 goto error; 495 496 if (tb[SMC_PNETID_ETHNAME]) { 497 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]); 498 rc = smc_pnet_add_eth(pnettable, net, string, pnet_name); 499 if (!rc) 500 new_netdev = true; 501 else if (rc != -EEXIST) 502 goto error; 503 } 504 505 /* if this is not the initial namespace, stop here */ 506 if (net != &init_net) 507 return new_netdev ? 0 : -EEXIST; 508 509 rc = -EINVAL; 510 if (tb[SMC_PNETID_IBNAME]) { 511 string = (char *)nla_data(tb[SMC_PNETID_IBNAME]); 512 string = strim(string); 513 if (tb[SMC_PNETID_IBPORT]) { 514 ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]); 515 if (ibport < 1 || ibport > SMC_MAX_PORTS) 516 goto error; 517 } 518 rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name); 519 if (!rc) 520 new_ibdev = true; 521 else if (rc != -EEXIST) 522 goto error; 523 } 524 return (new_netdev || new_ibdev) ? 0 : -EEXIST; 525 526 error: 527 return rc; 528 } 529 530 /* Convert an smc_pnetentry to a netlink attribute sequence */ 531 static int smc_pnet_set_nla(struct sk_buff *msg, 532 struct smc_pnetentry *pnetelem) 533 { 534 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name)) 535 return -1; 536 if (pnetelem->type == SMC_PNET_ETH) { 537 if (nla_put_string(msg, SMC_PNETID_ETHNAME, 538 pnetelem->eth_name)) 539 return -1; 540 } else { 541 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a")) 542 return -1; 543 } 544 if (pnetelem->type == SMC_PNET_IB) { 545 if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) || 546 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port)) 547 return -1; 548 } else { 549 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") || 550 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff)) 551 return -1; 552 } 553 554 return 0; 555 } 556 557 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info) 558 { 559 struct net *net = genl_info_net(info); 560 561 return smc_pnet_enter(net, info->attrs); 562 } 563 564 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info) 565 { 566 struct net *net = genl_info_net(info); 567 568 if (!info->attrs[SMC_PNETID_NAME]) 569 return -EINVAL; 570 return smc_pnet_remove_by_pnetid(net, 571 (char *)nla_data(info->attrs[SMC_PNETID_NAME])); 572 } 573 574 static int smc_pnet_dump_start(struct netlink_callback *cb) 575 { 576 cb->args[0] = 0; 577 return 0; 578 } 579 580 static int smc_pnet_dumpinfo(struct sk_buff *skb, 581 u32 portid, u32 seq, u32 flags, 582 struct smc_pnetentry *pnetelem) 583 { 584 void *hdr; 585 586 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family, 587 flags, SMC_PNETID_GET); 588 if (!hdr) 589 return -ENOMEM; 590 if (smc_pnet_set_nla(skb, pnetelem) < 0) { 591 genlmsg_cancel(skb, hdr); 592 return -EMSGSIZE; 593 } 594 genlmsg_end(skb, hdr); 595 return 0; 596 } 597 598 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid, 599 u32 seq, u8 *pnetid, int start_idx) 600 { 601 struct smc_pnettable *pnettable; 602 struct smc_pnetentry *pnetelem; 603 struct smc_net *sn; 604 int idx = 0; 605 606 /* get pnettable for namespace */ 607 sn = net_generic(net, smc_net_id); 608 pnettable = &sn->pnettable; 609 610 /* dump pnettable entries */ 611 mutex_lock(&pnettable->lock); 612 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 613 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid)) 614 continue; 615 if (idx++ < start_idx) 616 continue; 617 /* if this is not the initial namespace, dump only netdev */ 618 if (net != &init_net && pnetelem->type != SMC_PNET_ETH) 619 continue; 620 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI, 621 pnetelem)) { 622 --idx; 623 break; 624 } 625 } 626 mutex_unlock(&pnettable->lock); 627 return idx; 628 } 629 630 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb) 631 { 632 struct net *net = sock_net(skb->sk); 633 int idx; 634 635 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid, 636 cb->nlh->nlmsg_seq, NULL, cb->args[0]); 637 638 cb->args[0] = idx; 639 return skb->len; 640 } 641 642 /* Retrieve one PNETID entry */ 643 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info) 644 { 645 struct net *net = genl_info_net(info); 646 struct sk_buff *msg; 647 void *hdr; 648 649 if (!info->attrs[SMC_PNETID_NAME]) 650 return -EINVAL; 651 652 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 653 if (!msg) 654 return -ENOMEM; 655 656 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq, 657 nla_data(info->attrs[SMC_PNETID_NAME]), 0); 658 659 /* finish multi part message and send it */ 660 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0, 661 NLM_F_MULTI); 662 if (!hdr) { 663 nlmsg_free(msg); 664 return -EMSGSIZE; 665 } 666 return genlmsg_reply(msg, info); 667 } 668 669 /* Remove and delete all pnetids from pnet table. 670 */ 671 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info) 672 { 673 struct net *net = genl_info_net(info); 674 675 smc_pnet_remove_by_pnetid(net, NULL); 676 return 0; 677 } 678 679 /* SMC_PNETID generic netlink operation definition */ 680 static const struct genl_ops smc_pnet_ops[] = { 681 { 682 .cmd = SMC_PNETID_GET, 683 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 684 /* can be retrieved by unprivileged users */ 685 .doit = smc_pnet_get, 686 .dumpit = smc_pnet_dump, 687 .start = smc_pnet_dump_start 688 }, 689 { 690 .cmd = SMC_PNETID_ADD, 691 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 692 .flags = GENL_ADMIN_PERM, 693 .doit = smc_pnet_add 694 }, 695 { 696 .cmd = SMC_PNETID_DEL, 697 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 698 .flags = GENL_ADMIN_PERM, 699 .doit = smc_pnet_del 700 }, 701 { 702 .cmd = SMC_PNETID_FLUSH, 703 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 704 .flags = GENL_ADMIN_PERM, 705 .doit = smc_pnet_flush 706 } 707 }; 708 709 /* SMC_PNETID family definition */ 710 static struct genl_family smc_pnet_nl_family __ro_after_init = { 711 .hdrsize = 0, 712 .name = SMCR_GENL_FAMILY_NAME, 713 .version = SMCR_GENL_FAMILY_VERSION, 714 .maxattr = SMC_PNETID_MAX, 715 .policy = smc_pnet_policy, 716 .netnsok = true, 717 .module = THIS_MODULE, 718 .ops = smc_pnet_ops, 719 .n_ops = ARRAY_SIZE(smc_pnet_ops), 720 .resv_start_op = SMC_PNETID_FLUSH + 1, 721 }; 722 723 bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid) 724 { 725 struct smc_net *sn = net_generic(net, smc_net_id); 726 struct smc_pnetids_ndev_entry *pe; 727 bool rc = false; 728 729 read_lock(&sn->pnetids_ndev.lock); 730 list_for_each_entry(pe, &sn->pnetids_ndev.list, list) { 731 if (smc_pnet_match(pnetid, pe->pnetid)) { 732 rc = true; 733 goto unlock; 734 } 735 } 736 737 unlock: 738 read_unlock(&sn->pnetids_ndev.lock); 739 return rc; 740 } 741 742 static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid) 743 { 744 struct smc_net *sn = net_generic(net, smc_net_id); 745 struct smc_pnetids_ndev_entry *pe, *pi; 746 747 pe = kzalloc(sizeof(*pe), GFP_KERNEL); 748 if (!pe) 749 return -ENOMEM; 750 751 write_lock(&sn->pnetids_ndev.lock); 752 list_for_each_entry(pi, &sn->pnetids_ndev.list, list) { 753 if (smc_pnet_match(pnetid, pi->pnetid)) { 754 refcount_inc(&pi->refcnt); 755 kfree(pe); 756 goto unlock; 757 } 758 } 759 refcount_set(&pe->refcnt, 1); 760 memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN); 761 list_add_tail(&pe->list, &sn->pnetids_ndev.list); 762 763 unlock: 764 write_unlock(&sn->pnetids_ndev.lock); 765 return 0; 766 } 767 768 static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid) 769 { 770 struct smc_net *sn = net_generic(net, smc_net_id); 771 struct smc_pnetids_ndev_entry *pe, *pe2; 772 773 write_lock(&sn->pnetids_ndev.lock); 774 list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) { 775 if (smc_pnet_match(pnetid, pe->pnetid)) { 776 if (refcount_dec_and_test(&pe->refcnt)) { 777 list_del(&pe->list); 778 kfree(pe); 779 } 780 break; 781 } 782 } 783 write_unlock(&sn->pnetids_ndev.lock); 784 } 785 786 static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev, 787 u8 *ndev_pnetid) 788 { 789 struct net_device *base_dev; 790 791 base_dev = __pnet_find_base_ndev(dev); 792 if (base_dev->flags & IFF_UP && 793 !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port, 794 ndev_pnetid)) { 795 /* add to PNETIDs list */ 796 smc_pnet_add_pnetid(net, ndev_pnetid); 797 } 798 } 799 800 /* create initial list of netdevice pnetids */ 801 static void smc_pnet_create_pnetids_list(struct net *net) 802 { 803 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 804 struct net_device *dev; 805 806 /* Newly created netns do not have devices. 807 * Do not even acquire rtnl. 808 */ 809 if (list_empty(&net->dev_base_head)) 810 return; 811 812 /* Note: This might not be needed, because smc_pnet_netdev_event() 813 * is also calling smc_pnet_add_base_pnetid() when handling 814 * NETDEV_UP event. 815 */ 816 rtnl_lock(); 817 for_each_netdev(net, dev) 818 smc_pnet_add_base_pnetid(net, dev, ndev_pnetid); 819 rtnl_unlock(); 820 } 821 822 /* clean up list of netdevice pnetids */ 823 static void smc_pnet_destroy_pnetids_list(struct net *net) 824 { 825 struct smc_net *sn = net_generic(net, smc_net_id); 826 struct smc_pnetids_ndev_entry *pe, *temp_pe; 827 828 write_lock(&sn->pnetids_ndev.lock); 829 list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) { 830 list_del(&pe->list); 831 kfree(pe); 832 } 833 write_unlock(&sn->pnetids_ndev.lock); 834 } 835 836 static int smc_pnet_netdev_event(struct notifier_block *this, 837 unsigned long event, void *ptr) 838 { 839 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 840 struct net *net = dev_net(event_dev); 841 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 842 843 switch (event) { 844 case NETDEV_REBOOT: 845 case NETDEV_UNREGISTER: 846 smc_pnet_remove_by_ndev(event_dev); 847 smc_ib_ndev_change(event_dev, event); 848 return NOTIFY_OK; 849 case NETDEV_REGISTER: 850 smc_pnet_add_by_ndev(event_dev); 851 smc_ib_ndev_change(event_dev, event); 852 return NOTIFY_OK; 853 case NETDEV_UP: 854 smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid); 855 return NOTIFY_OK; 856 case NETDEV_DOWN: 857 event_dev = __pnet_find_base_ndev(event_dev); 858 if (!smc_pnetid_by_dev_port(event_dev->dev.parent, 859 event_dev->dev_port, ndev_pnetid)) { 860 /* remove from PNETIDs list */ 861 smc_pnet_remove_pnetid(net, ndev_pnetid); 862 } 863 return NOTIFY_OK; 864 default: 865 return NOTIFY_DONE; 866 } 867 } 868 869 static struct notifier_block smc_netdev_notifier = { 870 .notifier_call = smc_pnet_netdev_event 871 }; 872 873 /* init network namespace */ 874 int smc_pnet_net_init(struct net *net) 875 { 876 struct smc_net *sn = net_generic(net, smc_net_id); 877 struct smc_pnettable *pnettable = &sn->pnettable; 878 struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev; 879 880 INIT_LIST_HEAD(&pnettable->pnetlist); 881 mutex_init(&pnettable->lock); 882 INIT_LIST_HEAD(&pnetids_ndev->list); 883 rwlock_init(&pnetids_ndev->lock); 884 885 smc_pnet_create_pnetids_list(net); 886 887 return 0; 888 } 889 890 int __init smc_pnet_init(void) 891 { 892 int rc; 893 894 rc = genl_register_family(&smc_pnet_nl_family); 895 if (rc) 896 return rc; 897 rc = register_netdevice_notifier(&smc_netdev_notifier); 898 if (rc) 899 genl_unregister_family(&smc_pnet_nl_family); 900 901 return rc; 902 } 903 904 /* exit network namespace */ 905 void smc_pnet_net_exit(struct net *net) 906 { 907 /* flush pnet table */ 908 smc_pnet_remove_by_pnetid(net, NULL); 909 smc_pnet_destroy_pnetids_list(net); 910 } 911 912 void smc_pnet_exit(void) 913 { 914 unregister_netdevice_notifier(&smc_netdev_notifier); 915 genl_unregister_family(&smc_pnet_nl_family); 916 } 917 918 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev) 919 { 920 int i, nest_lvl; 921 922 ASSERT_RTNL(); 923 nest_lvl = ndev->lower_level; 924 for (i = 0; i < nest_lvl; i++) { 925 struct list_head *lower = &ndev->adj_list.lower; 926 927 if (list_empty(lower)) 928 break; 929 lower = lower->next; 930 ndev = netdev_lower_get_next(ndev, &lower); 931 } 932 return ndev; 933 } 934 935 /* Determine one base device for stacked net devices. 936 * If the lower device level contains more than one devices 937 * (for instance with bonding slaves), just the first device 938 * is used to reach a base device. 939 */ 940 static struct net_device *pnet_find_base_ndev(struct net_device *ndev) 941 { 942 rtnl_lock(); 943 ndev = __pnet_find_base_ndev(ndev); 944 rtnl_unlock(); 945 return ndev; 946 } 947 948 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev, 949 u8 *pnetid) 950 { 951 struct smc_pnettable *pnettable; 952 struct net *net = dev_net(ndev); 953 struct smc_pnetentry *pnetelem; 954 struct smc_net *sn; 955 int rc = -ENOENT; 956 957 /* get pnettable for namespace */ 958 sn = net_generic(net, smc_net_id); 959 pnettable = &sn->pnettable; 960 961 mutex_lock(&pnettable->lock); 962 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 963 if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) { 964 /* get pnetid of netdev device */ 965 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN); 966 rc = 0; 967 break; 968 } 969 } 970 mutex_unlock(&pnettable->lock); 971 return rc; 972 } 973 974 static int smc_pnet_determine_gid(struct smc_ib_device *ibdev, int i, 975 struct smc_init_info *ini) 976 { 977 if (!ini->check_smcrv2 && 978 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->ib_gid, NULL, 979 NULL)) { 980 ini->ib_dev = ibdev; 981 ini->ib_port = i; 982 return 0; 983 } 984 if (ini->check_smcrv2 && 985 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->smcrv2.ib_gid_v2, 986 NULL, &ini->smcrv2)) { 987 ini->smcrv2.ib_dev_v2 = ibdev; 988 ini->smcrv2.ib_port_v2 = i; 989 return 0; 990 } 991 return -ENODEV; 992 } 993 994 /* find a roce device for the given pnetid */ 995 static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id, 996 struct smc_init_info *ini, 997 struct smc_ib_device *known_dev, 998 struct net *net) 999 { 1000 struct smc_ib_device *ibdev; 1001 int i; 1002 1003 mutex_lock(&smc_ib_devices.mutex); 1004 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 1005 if (ibdev == known_dev || 1006 !rdma_dev_access_netns(ibdev->ibdev, net)) 1007 continue; 1008 for (i = 1; i <= SMC_MAX_PORTS; i++) { 1009 if (!rdma_is_port_valid(ibdev->ibdev, i)) 1010 continue; 1011 if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) && 1012 smc_ib_port_active(ibdev, i) && 1013 !test_bit(i - 1, ibdev->ports_going_away)) { 1014 if (!smc_pnet_determine_gid(ibdev, i, ini)) 1015 goto out; 1016 } 1017 } 1018 } 1019 out: 1020 mutex_unlock(&smc_ib_devices.mutex); 1021 } 1022 1023 /* find alternate roce device with same pnet_id, vlan_id and net namespace */ 1024 void smc_pnet_find_alt_roce(struct smc_link_group *lgr, 1025 struct smc_init_info *ini, 1026 struct smc_ib_device *known_dev) 1027 { 1028 struct net *net = lgr->net; 1029 1030 _smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev, net); 1031 } 1032 1033 /* if handshake network device belongs to a roce device, return its 1034 * IB device and port 1035 */ 1036 static void smc_pnet_find_rdma_dev(struct net_device *netdev, 1037 struct smc_init_info *ini) 1038 { 1039 struct net *net = dev_net(netdev); 1040 struct smc_ib_device *ibdev; 1041 1042 mutex_lock(&smc_ib_devices.mutex); 1043 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 1044 struct net_device *ndev; 1045 int i; 1046 1047 /* check rdma net namespace */ 1048 if (!rdma_dev_access_netns(ibdev->ibdev, net)) 1049 continue; 1050 1051 for (i = 1; i <= SMC_MAX_PORTS; i++) { 1052 if (!rdma_is_port_valid(ibdev->ibdev, i)) 1053 continue; 1054 ndev = ib_device_get_netdev(ibdev->ibdev, i); 1055 if (!ndev) 1056 continue; 1057 dev_put(ndev); 1058 if (netdev == ndev && 1059 smc_ib_port_active(ibdev, i) && 1060 !test_bit(i - 1, ibdev->ports_going_away)) { 1061 if (!smc_pnet_determine_gid(ibdev, i, ini)) 1062 break; 1063 } 1064 } 1065 } 1066 mutex_unlock(&smc_ib_devices.mutex); 1067 } 1068 1069 /* Determine the corresponding IB device port based on the hardware PNETID. 1070 * Searching stops at the first matching active IB device port with vlan_id 1071 * configured. 1072 * If nothing found, check pnetid table. 1073 * If nothing found, try to use handshake device 1074 */ 1075 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev, 1076 struct smc_init_info *ini) 1077 { 1078 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1079 struct net_device *base_ndev; 1080 struct net *net; 1081 1082 base_ndev = pnet_find_base_ndev(ndev); 1083 net = dev_net(ndev); 1084 if (smc_pnetid_by_dev_port(base_ndev->dev.parent, base_ndev->dev_port, 1085 ndev_pnetid) && 1086 smc_pnet_find_ndev_pnetid_by_table(base_ndev, ndev_pnetid) && 1087 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) { 1088 smc_pnet_find_rdma_dev(base_ndev, ini); 1089 return; /* pnetid could not be determined */ 1090 } 1091 _smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL, net); 1092 } 1093 1094 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev, 1095 struct smc_init_info *ini) 1096 { 1097 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1098 struct smcd_dev *ismdev; 1099 1100 ndev = pnet_find_base_ndev(ndev); 1101 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port, 1102 ndev_pnetid) && 1103 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) 1104 return; /* pnetid could not be determined */ 1105 1106 mutex_lock(&smcd_dev_list.mutex); 1107 list_for_each_entry(ismdev, &smcd_dev_list.list, list) { 1108 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) && 1109 !ismdev->going_away && 1110 (!ini->ism_peer_gid[0].gid || 1111 !smc_ism_cantalk(&ini->ism_peer_gid[0], ini->vlan_id, 1112 ismdev))) { 1113 ini->ism_dev[0] = ismdev; 1114 break; 1115 } 1116 } 1117 mutex_unlock(&smcd_dev_list.mutex); 1118 } 1119 1120 /* PNET table analysis for a given sock: 1121 * determine ib_device and port belonging to used internal TCP socket 1122 * ethernet interface. 1123 */ 1124 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini) 1125 { 1126 struct net_device *dev; 1127 struct dst_entry *dst; 1128 1129 rcu_read_lock(); 1130 dst = __sk_dst_get(sk); 1131 dev = dst ? dst_dev_rcu(dst) : NULL; 1132 dev_hold(dev); 1133 rcu_read_unlock(); 1134 1135 if (dev) { 1136 smc_pnet_find_roce_by_pnetid(dev, ini); 1137 dev_put(dev); 1138 } 1139 } 1140 1141 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini) 1142 { 1143 struct net_device *dev; 1144 struct dst_entry *dst; 1145 1146 ini->ism_dev[0] = NULL; 1147 1148 rcu_read_lock(); 1149 dst = __sk_dst_get(sk); 1150 dev = dst ? dst_dev_rcu(dst) : NULL; 1151 dev_hold(dev); 1152 rcu_read_unlock(); 1153 1154 if (dev) { 1155 smc_pnet_find_ism_by_pnetid(dev, ini); 1156 dev_put(dev); 1157 } 1158 } 1159 1160 /* Lookup and apply a pnet table entry to the given ib device. 1161 */ 1162 int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port) 1163 { 1164 char *ib_name = smcibdev->ibdev->name; 1165 struct smc_pnettable *pnettable; 1166 struct smc_pnetentry *tmp_pe; 1167 struct smc_net *sn; 1168 int rc = -ENOENT; 1169 1170 /* get pnettable for init namespace */ 1171 sn = net_generic(&init_net, smc_net_id); 1172 pnettable = &sn->pnettable; 1173 1174 mutex_lock(&pnettable->lock); 1175 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1176 if (tmp_pe->type == SMC_PNET_IB && 1177 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) && 1178 tmp_pe->ib_port == ib_port) { 1179 smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name); 1180 rc = 0; 1181 break; 1182 } 1183 } 1184 mutex_unlock(&pnettable->lock); 1185 1186 return rc; 1187 } 1188 1189 /* Lookup and apply a pnet table entry to the given smcd device. 1190 */ 1191 int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev) 1192 { 1193 const char *ib_name = dev_name(&smcddev->dibs->dev); 1194 struct smc_pnettable *pnettable; 1195 struct smc_pnetentry *tmp_pe; 1196 struct smc_net *sn; 1197 int rc = -ENOENT; 1198 1199 /* get pnettable for init namespace */ 1200 sn = net_generic(&init_net, smc_net_id); 1201 pnettable = &sn->pnettable; 1202 1203 mutex_lock(&pnettable->lock); 1204 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1205 if (tmp_pe->type == SMC_PNET_IB && 1206 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 1207 smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name); 1208 rc = 0; 1209 break; 1210 } 1211 } 1212 mutex_unlock(&pnettable->lock); 1213 1214 return rc; 1215 } 1216