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