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->ops->get_dev(smcd)), 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->ops->get_dev(smcd_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 strncpy(new_pe->eth_name, eth_name, IFNAMSIZ); 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 struct device *dev; 417 bool new_ibdev; 418 419 /* try to apply the pnetid to active devices */ 420 ib_dev = smc_pnet_find_ib(ib_name); 421 if (ib_dev) { 422 ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name); 423 if (ibdev_applied) 424 pr_warn_ratelimited("smc: ib device %s ibport %d " 425 "applied user defined pnetid " 426 "%.16s\n", ib_dev->ibdev->name, 427 ib_port, 428 ib_dev->pnetid[ib_port - 1]); 429 } 430 smcd = smc_pnet_find_smcd(ib_name); 431 if (smcd) { 432 smcddev_applied = smc_pnet_apply_smcd(smcd, pnet_name); 433 if (smcddev_applied) { 434 dev = smcd->ops->get_dev(smcd); 435 pr_warn_ratelimited("smc: smcd device %s " 436 "applied user defined pnetid " 437 "%.16s\n", dev_name(dev), 438 smcd->pnetid); 439 } 440 } 441 /* Apply fails when a device has a hardware-defined pnetid set, do not 442 * add a pnet table entry in that case. 443 */ 444 if (!ibdev_applied || !smcddev_applied) 445 return -EEXIST; 446 447 /* add a new ib entry to the pnet table if there isn't one */ 448 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL); 449 if (!new_pe) 450 return -ENOMEM; 451 new_pe->type = SMC_PNET_IB; 452 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN); 453 strncpy(new_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX); 454 new_pe->ib_port = ib_port; 455 456 new_ibdev = true; 457 mutex_lock(&pnettable->lock); 458 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 459 if (tmp_pe->type == SMC_PNET_IB && 460 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 461 new_ibdev = false; 462 break; 463 } 464 } 465 if (new_ibdev) { 466 list_add_tail(&new_pe->list, &pnettable->pnetlist); 467 mutex_unlock(&pnettable->lock); 468 } else { 469 mutex_unlock(&pnettable->lock); 470 kfree(new_pe); 471 } 472 return (new_ibdev) ? 0 : -EEXIST; 473 } 474 475 /* Append a pnetid to the end of the pnet table if not already on this list. 476 */ 477 static int smc_pnet_enter(struct net *net, struct nlattr *tb[]) 478 { 479 char pnet_name[SMC_MAX_PNETID_LEN + 1]; 480 struct smc_pnettable *pnettable; 481 bool new_netdev = false; 482 bool new_ibdev = false; 483 struct smc_net *sn; 484 u8 ibport = 1; 485 char *string; 486 int rc; 487 488 /* get pnettable for namespace */ 489 sn = net_generic(net, smc_net_id); 490 pnettable = &sn->pnettable; 491 492 rc = -EINVAL; 493 if (!tb[SMC_PNETID_NAME]) 494 goto error; 495 string = (char *)nla_data(tb[SMC_PNETID_NAME]); 496 if (!smc_pnetid_valid(string, pnet_name)) 497 goto error; 498 499 if (tb[SMC_PNETID_ETHNAME]) { 500 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]); 501 rc = smc_pnet_add_eth(pnettable, net, string, pnet_name); 502 if (!rc) 503 new_netdev = true; 504 else if (rc != -EEXIST) 505 goto error; 506 } 507 508 /* if this is not the initial namespace, stop here */ 509 if (net != &init_net) 510 return new_netdev ? 0 : -EEXIST; 511 512 rc = -EINVAL; 513 if (tb[SMC_PNETID_IBNAME]) { 514 string = (char *)nla_data(tb[SMC_PNETID_IBNAME]); 515 string = strim(string); 516 if (tb[SMC_PNETID_IBPORT]) { 517 ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]); 518 if (ibport < 1 || ibport > SMC_MAX_PORTS) 519 goto error; 520 } 521 rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name); 522 if (!rc) 523 new_ibdev = true; 524 else if (rc != -EEXIST) 525 goto error; 526 } 527 return (new_netdev || new_ibdev) ? 0 : -EEXIST; 528 529 error: 530 return rc; 531 } 532 533 /* Convert an smc_pnetentry to a netlink attribute sequence */ 534 static int smc_pnet_set_nla(struct sk_buff *msg, 535 struct smc_pnetentry *pnetelem) 536 { 537 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name)) 538 return -1; 539 if (pnetelem->type == SMC_PNET_ETH) { 540 if (nla_put_string(msg, SMC_PNETID_ETHNAME, 541 pnetelem->eth_name)) 542 return -1; 543 } else { 544 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a")) 545 return -1; 546 } 547 if (pnetelem->type == SMC_PNET_IB) { 548 if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) || 549 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port)) 550 return -1; 551 } else { 552 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") || 553 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff)) 554 return -1; 555 } 556 557 return 0; 558 } 559 560 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info) 561 { 562 struct net *net = genl_info_net(info); 563 564 return smc_pnet_enter(net, info->attrs); 565 } 566 567 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info) 568 { 569 struct net *net = genl_info_net(info); 570 571 if (!info->attrs[SMC_PNETID_NAME]) 572 return -EINVAL; 573 return smc_pnet_remove_by_pnetid(net, 574 (char *)nla_data(info->attrs[SMC_PNETID_NAME])); 575 } 576 577 static int smc_pnet_dump_start(struct netlink_callback *cb) 578 { 579 cb->args[0] = 0; 580 return 0; 581 } 582 583 static int smc_pnet_dumpinfo(struct sk_buff *skb, 584 u32 portid, u32 seq, u32 flags, 585 struct smc_pnetentry *pnetelem) 586 { 587 void *hdr; 588 589 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family, 590 flags, SMC_PNETID_GET); 591 if (!hdr) 592 return -ENOMEM; 593 if (smc_pnet_set_nla(skb, pnetelem) < 0) { 594 genlmsg_cancel(skb, hdr); 595 return -EMSGSIZE; 596 } 597 genlmsg_end(skb, hdr); 598 return 0; 599 } 600 601 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid, 602 u32 seq, u8 *pnetid, int start_idx) 603 { 604 struct smc_pnettable *pnettable; 605 struct smc_pnetentry *pnetelem; 606 struct smc_net *sn; 607 int idx = 0; 608 609 /* get pnettable for namespace */ 610 sn = net_generic(net, smc_net_id); 611 pnettable = &sn->pnettable; 612 613 /* dump pnettable entries */ 614 mutex_lock(&pnettable->lock); 615 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 616 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid)) 617 continue; 618 if (idx++ < start_idx) 619 continue; 620 /* if this is not the initial namespace, dump only netdev */ 621 if (net != &init_net && pnetelem->type != SMC_PNET_ETH) 622 continue; 623 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI, 624 pnetelem)) { 625 --idx; 626 break; 627 } 628 } 629 mutex_unlock(&pnettable->lock); 630 return idx; 631 } 632 633 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb) 634 { 635 struct net *net = sock_net(skb->sk); 636 int idx; 637 638 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid, 639 cb->nlh->nlmsg_seq, NULL, cb->args[0]); 640 641 cb->args[0] = idx; 642 return skb->len; 643 } 644 645 /* Retrieve one PNETID entry */ 646 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info) 647 { 648 struct net *net = genl_info_net(info); 649 struct sk_buff *msg; 650 void *hdr; 651 652 if (!info->attrs[SMC_PNETID_NAME]) 653 return -EINVAL; 654 655 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 656 if (!msg) 657 return -ENOMEM; 658 659 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq, 660 nla_data(info->attrs[SMC_PNETID_NAME]), 0); 661 662 /* finish multi part message and send it */ 663 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0, 664 NLM_F_MULTI); 665 if (!hdr) { 666 nlmsg_free(msg); 667 return -EMSGSIZE; 668 } 669 return genlmsg_reply(msg, info); 670 } 671 672 /* Remove and delete all pnetids from pnet table. 673 */ 674 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info) 675 { 676 struct net *net = genl_info_net(info); 677 678 smc_pnet_remove_by_pnetid(net, NULL); 679 return 0; 680 } 681 682 /* SMC_PNETID generic netlink operation definition */ 683 static const struct genl_ops smc_pnet_ops[] = { 684 { 685 .cmd = SMC_PNETID_GET, 686 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 687 /* can be retrieved by unprivileged users */ 688 .doit = smc_pnet_get, 689 .dumpit = smc_pnet_dump, 690 .start = smc_pnet_dump_start 691 }, 692 { 693 .cmd = SMC_PNETID_ADD, 694 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 695 .flags = GENL_ADMIN_PERM, 696 .doit = smc_pnet_add 697 }, 698 { 699 .cmd = SMC_PNETID_DEL, 700 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 701 .flags = GENL_ADMIN_PERM, 702 .doit = smc_pnet_del 703 }, 704 { 705 .cmd = SMC_PNETID_FLUSH, 706 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 707 .flags = GENL_ADMIN_PERM, 708 .doit = smc_pnet_flush 709 } 710 }; 711 712 /* SMC_PNETID family definition */ 713 static struct genl_family smc_pnet_nl_family __ro_after_init = { 714 .hdrsize = 0, 715 .name = SMCR_GENL_FAMILY_NAME, 716 .version = SMCR_GENL_FAMILY_VERSION, 717 .maxattr = SMC_PNETID_MAX, 718 .policy = smc_pnet_policy, 719 .netnsok = true, 720 .module = THIS_MODULE, 721 .ops = smc_pnet_ops, 722 .n_ops = ARRAY_SIZE(smc_pnet_ops), 723 .resv_start_op = SMC_PNETID_FLUSH + 1, 724 }; 725 726 bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid) 727 { 728 struct smc_net *sn = net_generic(net, smc_net_id); 729 struct smc_pnetids_ndev_entry *pe; 730 bool rc = false; 731 732 read_lock(&sn->pnetids_ndev.lock); 733 list_for_each_entry(pe, &sn->pnetids_ndev.list, list) { 734 if (smc_pnet_match(pnetid, pe->pnetid)) { 735 rc = true; 736 goto unlock; 737 } 738 } 739 740 unlock: 741 read_unlock(&sn->pnetids_ndev.lock); 742 return rc; 743 } 744 745 static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid) 746 { 747 struct smc_net *sn = net_generic(net, smc_net_id); 748 struct smc_pnetids_ndev_entry *pe, *pi; 749 750 pe = kzalloc(sizeof(*pe), GFP_KERNEL); 751 if (!pe) 752 return -ENOMEM; 753 754 write_lock(&sn->pnetids_ndev.lock); 755 list_for_each_entry(pi, &sn->pnetids_ndev.list, list) { 756 if (smc_pnet_match(pnetid, pe->pnetid)) { 757 refcount_inc(&pi->refcnt); 758 kfree(pe); 759 goto unlock; 760 } 761 } 762 refcount_set(&pe->refcnt, 1); 763 memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN); 764 list_add_tail(&pe->list, &sn->pnetids_ndev.list); 765 766 unlock: 767 write_unlock(&sn->pnetids_ndev.lock); 768 return 0; 769 } 770 771 static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid) 772 { 773 struct smc_net *sn = net_generic(net, smc_net_id); 774 struct smc_pnetids_ndev_entry *pe, *pe2; 775 776 write_lock(&sn->pnetids_ndev.lock); 777 list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) { 778 if (smc_pnet_match(pnetid, pe->pnetid)) { 779 if (refcount_dec_and_test(&pe->refcnt)) { 780 list_del(&pe->list); 781 kfree(pe); 782 } 783 break; 784 } 785 } 786 write_unlock(&sn->pnetids_ndev.lock); 787 } 788 789 static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev, 790 u8 *ndev_pnetid) 791 { 792 struct net_device *base_dev; 793 794 base_dev = __pnet_find_base_ndev(dev); 795 if (base_dev->flags & IFF_UP && 796 !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port, 797 ndev_pnetid)) { 798 /* add to PNETIDs list */ 799 smc_pnet_add_pnetid(net, ndev_pnetid); 800 } 801 } 802 803 /* create initial list of netdevice pnetids */ 804 static void smc_pnet_create_pnetids_list(struct net *net) 805 { 806 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 807 struct net_device *dev; 808 809 /* Newly created netns do not have devices. 810 * Do not even acquire rtnl. 811 */ 812 if (list_empty(&net->dev_base_head)) 813 return; 814 815 /* Note: This might not be needed, because smc_pnet_netdev_event() 816 * is also calling smc_pnet_add_base_pnetid() when handling 817 * NETDEV_UP event. 818 */ 819 rtnl_lock(); 820 for_each_netdev(net, dev) 821 smc_pnet_add_base_pnetid(net, dev, ndev_pnetid); 822 rtnl_unlock(); 823 } 824 825 /* clean up list of netdevice pnetids */ 826 static void smc_pnet_destroy_pnetids_list(struct net *net) 827 { 828 struct smc_net *sn = net_generic(net, smc_net_id); 829 struct smc_pnetids_ndev_entry *pe, *temp_pe; 830 831 write_lock(&sn->pnetids_ndev.lock); 832 list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) { 833 list_del(&pe->list); 834 kfree(pe); 835 } 836 write_unlock(&sn->pnetids_ndev.lock); 837 } 838 839 static int smc_pnet_netdev_event(struct notifier_block *this, 840 unsigned long event, void *ptr) 841 { 842 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 843 struct net *net = dev_net(event_dev); 844 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 845 846 switch (event) { 847 case NETDEV_REBOOT: 848 case NETDEV_UNREGISTER: 849 smc_pnet_remove_by_ndev(event_dev); 850 smc_ib_ndev_change(event_dev, event); 851 return NOTIFY_OK; 852 case NETDEV_REGISTER: 853 smc_pnet_add_by_ndev(event_dev); 854 smc_ib_ndev_change(event_dev, event); 855 return NOTIFY_OK; 856 case NETDEV_UP: 857 smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid); 858 return NOTIFY_OK; 859 case NETDEV_DOWN: 860 event_dev = __pnet_find_base_ndev(event_dev); 861 if (!smc_pnetid_by_dev_port(event_dev->dev.parent, 862 event_dev->dev_port, ndev_pnetid)) { 863 /* remove from PNETIDs list */ 864 smc_pnet_remove_pnetid(net, ndev_pnetid); 865 } 866 return NOTIFY_OK; 867 default: 868 return NOTIFY_DONE; 869 } 870 } 871 872 static struct notifier_block smc_netdev_notifier = { 873 .notifier_call = smc_pnet_netdev_event 874 }; 875 876 /* init network namespace */ 877 int smc_pnet_net_init(struct net *net) 878 { 879 struct smc_net *sn = net_generic(net, smc_net_id); 880 struct smc_pnettable *pnettable = &sn->pnettable; 881 struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev; 882 883 INIT_LIST_HEAD(&pnettable->pnetlist); 884 mutex_init(&pnettable->lock); 885 INIT_LIST_HEAD(&pnetids_ndev->list); 886 rwlock_init(&pnetids_ndev->lock); 887 888 smc_pnet_create_pnetids_list(net); 889 890 return 0; 891 } 892 893 int __init smc_pnet_init(void) 894 { 895 int rc; 896 897 rc = genl_register_family(&smc_pnet_nl_family); 898 if (rc) 899 return rc; 900 rc = register_netdevice_notifier(&smc_netdev_notifier); 901 if (rc) 902 genl_unregister_family(&smc_pnet_nl_family); 903 904 return rc; 905 } 906 907 /* exit network namespace */ 908 void smc_pnet_net_exit(struct net *net) 909 { 910 /* flush pnet table */ 911 smc_pnet_remove_by_pnetid(net, NULL); 912 smc_pnet_destroy_pnetids_list(net); 913 } 914 915 void smc_pnet_exit(void) 916 { 917 unregister_netdevice_notifier(&smc_netdev_notifier); 918 genl_unregister_family(&smc_pnet_nl_family); 919 } 920 921 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev) 922 { 923 int i, nest_lvl; 924 925 ASSERT_RTNL(); 926 nest_lvl = ndev->lower_level; 927 for (i = 0; i < nest_lvl; i++) { 928 struct list_head *lower = &ndev->adj_list.lower; 929 930 if (list_empty(lower)) 931 break; 932 lower = lower->next; 933 ndev = netdev_lower_get_next(ndev, &lower); 934 } 935 return ndev; 936 } 937 938 /* Determine one base device for stacked net devices. 939 * If the lower device level contains more than one devices 940 * (for instance with bonding slaves), just the first device 941 * is used to reach a base device. 942 */ 943 static struct net_device *pnet_find_base_ndev(struct net_device *ndev) 944 { 945 rtnl_lock(); 946 ndev = __pnet_find_base_ndev(ndev); 947 rtnl_unlock(); 948 return ndev; 949 } 950 951 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev, 952 u8 *pnetid) 953 { 954 struct smc_pnettable *pnettable; 955 struct net *net = dev_net(ndev); 956 struct smc_pnetentry *pnetelem; 957 struct smc_net *sn; 958 int rc = -ENOENT; 959 960 /* get pnettable for namespace */ 961 sn = net_generic(net, smc_net_id); 962 pnettable = &sn->pnettable; 963 964 mutex_lock(&pnettable->lock); 965 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) { 966 if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) { 967 /* get pnetid of netdev device */ 968 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN); 969 rc = 0; 970 break; 971 } 972 } 973 mutex_unlock(&pnettable->lock); 974 return rc; 975 } 976 977 static int smc_pnet_determine_gid(struct smc_ib_device *ibdev, int i, 978 struct smc_init_info *ini) 979 { 980 if (!ini->check_smcrv2 && 981 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->ib_gid, NULL, 982 NULL)) { 983 ini->ib_dev = ibdev; 984 ini->ib_port = i; 985 return 0; 986 } 987 if (ini->check_smcrv2 && 988 !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->smcrv2.ib_gid_v2, 989 NULL, &ini->smcrv2)) { 990 ini->smcrv2.ib_dev_v2 = ibdev; 991 ini->smcrv2.ib_port_v2 = i; 992 return 0; 993 } 994 return -ENODEV; 995 } 996 997 /* find a roce device for the given pnetid */ 998 static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id, 999 struct smc_init_info *ini, 1000 struct smc_ib_device *known_dev, 1001 struct net *net) 1002 { 1003 struct smc_ib_device *ibdev; 1004 int i; 1005 1006 mutex_lock(&smc_ib_devices.mutex); 1007 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 1008 if (ibdev == known_dev || 1009 !rdma_dev_access_netns(ibdev->ibdev, net)) 1010 continue; 1011 for (i = 1; i <= SMC_MAX_PORTS; i++) { 1012 if (!rdma_is_port_valid(ibdev->ibdev, i)) 1013 continue; 1014 if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) && 1015 smc_ib_port_active(ibdev, i) && 1016 !test_bit(i - 1, ibdev->ports_going_away)) { 1017 if (!smc_pnet_determine_gid(ibdev, i, ini)) 1018 goto out; 1019 } 1020 } 1021 } 1022 out: 1023 mutex_unlock(&smc_ib_devices.mutex); 1024 } 1025 1026 /* find alternate roce device with same pnet_id, vlan_id and net namespace */ 1027 void smc_pnet_find_alt_roce(struct smc_link_group *lgr, 1028 struct smc_init_info *ini, 1029 struct smc_ib_device *known_dev) 1030 { 1031 struct net *net = lgr->net; 1032 1033 _smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev, net); 1034 } 1035 1036 /* if handshake network device belongs to a roce device, return its 1037 * IB device and port 1038 */ 1039 static void smc_pnet_find_rdma_dev(struct net_device *netdev, 1040 struct smc_init_info *ini) 1041 { 1042 struct net *net = dev_net(netdev); 1043 struct smc_ib_device *ibdev; 1044 1045 mutex_lock(&smc_ib_devices.mutex); 1046 list_for_each_entry(ibdev, &smc_ib_devices.list, list) { 1047 struct net_device *ndev; 1048 int i; 1049 1050 /* check rdma net namespace */ 1051 if (!rdma_dev_access_netns(ibdev->ibdev, net)) 1052 continue; 1053 1054 for (i = 1; i <= SMC_MAX_PORTS; i++) { 1055 if (!rdma_is_port_valid(ibdev->ibdev, i)) 1056 continue; 1057 if (!ibdev->ibdev->ops.get_netdev) 1058 continue; 1059 ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i); 1060 if (!ndev) 1061 continue; 1062 dev_put(ndev); 1063 if (netdev == ndev && 1064 smc_ib_port_active(ibdev, i) && 1065 !test_bit(i - 1, ibdev->ports_going_away)) { 1066 if (!smc_pnet_determine_gid(ibdev, i, ini)) 1067 break; 1068 } 1069 } 1070 } 1071 mutex_unlock(&smc_ib_devices.mutex); 1072 } 1073 1074 /* Determine the corresponding IB device port based on the hardware PNETID. 1075 * Searching stops at the first matching active IB device port with vlan_id 1076 * configured. 1077 * If nothing found, check pnetid table. 1078 * If nothing found, try to use handshake device 1079 */ 1080 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev, 1081 struct smc_init_info *ini) 1082 { 1083 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1084 struct net *net; 1085 1086 ndev = pnet_find_base_ndev(ndev); 1087 net = dev_net(ndev); 1088 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port, 1089 ndev_pnetid) && 1090 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) { 1091 smc_pnet_find_rdma_dev(ndev, ini); 1092 return; /* pnetid could not be determined */ 1093 } 1094 _smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL, net); 1095 } 1096 1097 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev, 1098 struct smc_init_info *ini) 1099 { 1100 u8 ndev_pnetid[SMC_MAX_PNETID_LEN]; 1101 struct smcd_dev *ismdev; 1102 1103 ndev = pnet_find_base_ndev(ndev); 1104 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port, 1105 ndev_pnetid) && 1106 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) 1107 return; /* pnetid could not be determined */ 1108 1109 mutex_lock(&smcd_dev_list.mutex); 1110 list_for_each_entry(ismdev, &smcd_dev_list.list, list) { 1111 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) && 1112 !ismdev->going_away && 1113 (!ini->ism_peer_gid[0].gid || 1114 !smc_ism_cantalk(&ini->ism_peer_gid[0], ini->vlan_id, 1115 ismdev))) { 1116 ini->ism_dev[0] = ismdev; 1117 break; 1118 } 1119 } 1120 mutex_unlock(&smcd_dev_list.mutex); 1121 } 1122 1123 /* PNET table analysis for a given sock: 1124 * determine ib_device and port belonging to used internal TCP socket 1125 * ethernet interface. 1126 */ 1127 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini) 1128 { 1129 struct dst_entry *dst = sk_dst_get(sk); 1130 1131 if (!dst) 1132 goto out; 1133 if (!dst->dev) 1134 goto out_rel; 1135 1136 smc_pnet_find_roce_by_pnetid(dst->dev, ini); 1137 1138 out_rel: 1139 dst_release(dst); 1140 out: 1141 return; 1142 } 1143 1144 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini) 1145 { 1146 struct dst_entry *dst = sk_dst_get(sk); 1147 1148 ini->ism_dev[0] = NULL; 1149 if (!dst) 1150 goto out; 1151 if (!dst->dev) 1152 goto out_rel; 1153 1154 smc_pnet_find_ism_by_pnetid(dst->dev, ini); 1155 1156 out_rel: 1157 dst_release(dst); 1158 out: 1159 return; 1160 } 1161 1162 /* Lookup and apply a pnet table entry to the given ib device. 1163 */ 1164 int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port) 1165 { 1166 char *ib_name = smcibdev->ibdev->name; 1167 struct smc_pnettable *pnettable; 1168 struct smc_pnetentry *tmp_pe; 1169 struct smc_net *sn; 1170 int rc = -ENOENT; 1171 1172 /* get pnettable for init namespace */ 1173 sn = net_generic(&init_net, smc_net_id); 1174 pnettable = &sn->pnettable; 1175 1176 mutex_lock(&pnettable->lock); 1177 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1178 if (tmp_pe->type == SMC_PNET_IB && 1179 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) && 1180 tmp_pe->ib_port == ib_port) { 1181 smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name); 1182 rc = 0; 1183 break; 1184 } 1185 } 1186 mutex_unlock(&pnettable->lock); 1187 1188 return rc; 1189 } 1190 1191 /* Lookup and apply a pnet table entry to the given smcd device. 1192 */ 1193 int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev) 1194 { 1195 const char *ib_name = dev_name(smcddev->ops->get_dev(smcddev)); 1196 struct smc_pnettable *pnettable; 1197 struct smc_pnetentry *tmp_pe; 1198 struct smc_net *sn; 1199 int rc = -ENOENT; 1200 1201 /* get pnettable for init namespace */ 1202 sn = net_generic(&init_net, smc_net_id); 1203 pnettable = &sn->pnettable; 1204 1205 mutex_lock(&pnettable->lock); 1206 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) { 1207 if (tmp_pe->type == SMC_PNET_IB && 1208 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) { 1209 smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name); 1210 rc = 0; 1211 break; 1212 } 1213 } 1214 mutex_unlock(&pnettable->lock); 1215 1216 return rc; 1217 } 1218