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