1 /* 2 * This file is part of the Chelsio T4/T5/T6 Ethernet driver for Linux. 3 * 4 * Copyright (c) 2017 Chelsio Communications, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34 35 #include <net/tc_act/tc_mirred.h> 36 #include <net/tc_act/tc_pedit.h> 37 #include <net/tc_act/tc_gact.h> 38 #include <net/tc_act/tc_vlan.h> 39 40 #include "cxgb4.h" 41 #include "cxgb4_filter.h" 42 #include "cxgb4_tc_flower.h" 43 44 #define STATS_CHECK_PERIOD (HZ / 2) 45 46 static struct ch_tc_pedit_fields pedits[] = { 47 PEDIT_FIELDS(ETH_, DMAC_31_0, 4, dmac, 0), 48 PEDIT_FIELDS(ETH_, DMAC_47_32, 2, dmac, 4), 49 PEDIT_FIELDS(ETH_, SMAC_15_0, 2, smac, 0), 50 PEDIT_FIELDS(ETH_, SMAC_47_16, 4, smac, 2), 51 PEDIT_FIELDS(IP4_, SRC, 4, nat_fip, 0), 52 PEDIT_FIELDS(IP4_, DST, 4, nat_lip, 0), 53 PEDIT_FIELDS(IP6_, SRC_31_0, 4, nat_fip, 0), 54 PEDIT_FIELDS(IP6_, SRC_63_32, 4, nat_fip, 4), 55 PEDIT_FIELDS(IP6_, SRC_95_64, 4, nat_fip, 8), 56 PEDIT_FIELDS(IP6_, SRC_127_96, 4, nat_fip, 12), 57 PEDIT_FIELDS(IP6_, DST_31_0, 4, nat_lip, 0), 58 PEDIT_FIELDS(IP6_, DST_63_32, 4, nat_lip, 4), 59 PEDIT_FIELDS(IP6_, DST_95_64, 4, nat_lip, 8), 60 PEDIT_FIELDS(IP6_, DST_127_96, 4, nat_lip, 12), 61 }; 62 63 static const struct cxgb4_natmode_config cxgb4_natmode_config_array[] = { 64 /* Default supported NAT modes */ 65 { 66 .chip = CHELSIO_T5, 67 .flags = CXGB4_ACTION_NATMODE_NONE, 68 .natmode = NAT_MODE_NONE, 69 }, 70 { 71 .chip = CHELSIO_T5, 72 .flags = CXGB4_ACTION_NATMODE_DIP, 73 .natmode = NAT_MODE_DIP, 74 }, 75 { 76 .chip = CHELSIO_T5, 77 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT, 78 .natmode = NAT_MODE_DIP_DP, 79 }, 80 { 81 .chip = CHELSIO_T5, 82 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT | 83 CXGB4_ACTION_NATMODE_SIP, 84 .natmode = NAT_MODE_DIP_DP_SIP, 85 }, 86 { 87 .chip = CHELSIO_T5, 88 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_DPORT | 89 CXGB4_ACTION_NATMODE_SPORT, 90 .natmode = NAT_MODE_DIP_DP_SP, 91 }, 92 { 93 .chip = CHELSIO_T5, 94 .flags = CXGB4_ACTION_NATMODE_SIP | CXGB4_ACTION_NATMODE_SPORT, 95 .natmode = NAT_MODE_SIP_SP, 96 }, 97 { 98 .chip = CHELSIO_T5, 99 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP | 100 CXGB4_ACTION_NATMODE_SPORT, 101 .natmode = NAT_MODE_DIP_SIP_SP, 102 }, 103 { 104 .chip = CHELSIO_T5, 105 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP | 106 CXGB4_ACTION_NATMODE_DPORT | 107 CXGB4_ACTION_NATMODE_SPORT, 108 .natmode = NAT_MODE_ALL, 109 }, 110 /* T6+ can ignore L4 ports when they're disabled. */ 111 { 112 .chip = CHELSIO_T6, 113 .flags = CXGB4_ACTION_NATMODE_SIP, 114 .natmode = NAT_MODE_SIP_SP, 115 }, 116 { 117 .chip = CHELSIO_T6, 118 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SPORT, 119 .natmode = NAT_MODE_DIP_DP_SP, 120 }, 121 { 122 .chip = CHELSIO_T6, 123 .flags = CXGB4_ACTION_NATMODE_DIP | CXGB4_ACTION_NATMODE_SIP, 124 .natmode = NAT_MODE_ALL, 125 }, 126 }; 127 128 static void cxgb4_action_natmode_tweak(struct ch_filter_specification *fs, 129 u8 natmode_flags) 130 { 131 u8 i = 0; 132 133 /* Translate the enabled NAT 4-tuple fields to one of the 134 * hardware supported NAT mode configurations. This ensures 135 * that we pick a valid combination, where the disabled fields 136 * do not get overwritten to 0. 137 */ 138 for (i = 0; i < ARRAY_SIZE(cxgb4_natmode_config_array); i++) { 139 if (cxgb4_natmode_config_array[i].flags == natmode_flags) { 140 fs->nat_mode = cxgb4_natmode_config_array[i].natmode; 141 return; 142 } 143 } 144 } 145 146 static struct ch_tc_flower_entry *allocate_flower_entry(void) 147 { 148 struct ch_tc_flower_entry *new = kzalloc(sizeof(*new), GFP_KERNEL); 149 if (new) 150 spin_lock_init(&new->lock); 151 return new; 152 } 153 154 /* Must be called with either RTNL or rcu_read_lock */ 155 static struct ch_tc_flower_entry *ch_flower_lookup(struct adapter *adap, 156 unsigned long flower_cookie) 157 { 158 return rhashtable_lookup_fast(&adap->flower_tbl, &flower_cookie, 159 adap->flower_ht_params); 160 } 161 162 static void cxgb4_process_flow_match(struct net_device *dev, 163 struct flow_rule *rule, 164 struct ch_filter_specification *fs) 165 { 166 u16 addr_type = 0; 167 168 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) { 169 struct flow_match_control match; 170 171 flow_rule_match_control(rule, &match); 172 addr_type = match.key->addr_type; 173 } else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) { 174 addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 175 } else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS)) { 176 addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 177 } 178 179 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) { 180 struct flow_match_basic match; 181 u16 ethtype_key, ethtype_mask; 182 183 flow_rule_match_basic(rule, &match); 184 ethtype_key = ntohs(match.key->n_proto); 185 ethtype_mask = ntohs(match.mask->n_proto); 186 187 if (ethtype_key == ETH_P_ALL) { 188 ethtype_key = 0; 189 ethtype_mask = 0; 190 } 191 192 if (ethtype_key == ETH_P_IPV6) 193 fs->type = 1; 194 195 fs->val.ethtype = ethtype_key; 196 fs->mask.ethtype = ethtype_mask; 197 fs->val.proto = match.key->ip_proto; 198 fs->mask.proto = match.mask->ip_proto; 199 } 200 201 if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 202 struct flow_match_ipv4_addrs match; 203 204 flow_rule_match_ipv4_addrs(rule, &match); 205 fs->type = 0; 206 memcpy(&fs->val.lip[0], &match.key->dst, sizeof(match.key->dst)); 207 memcpy(&fs->val.fip[0], &match.key->src, sizeof(match.key->src)); 208 memcpy(&fs->mask.lip[0], &match.mask->dst, sizeof(match.mask->dst)); 209 memcpy(&fs->mask.fip[0], &match.mask->src, sizeof(match.mask->src)); 210 211 /* also initialize nat_lip/fip to same values */ 212 memcpy(&fs->nat_lip[0], &match.key->dst, sizeof(match.key->dst)); 213 memcpy(&fs->nat_fip[0], &match.key->src, sizeof(match.key->src)); 214 } 215 216 if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 217 struct flow_match_ipv6_addrs match; 218 219 flow_rule_match_ipv6_addrs(rule, &match); 220 fs->type = 1; 221 memcpy(&fs->val.lip[0], match.key->dst.s6_addr, 222 sizeof(match.key->dst)); 223 memcpy(&fs->val.fip[0], match.key->src.s6_addr, 224 sizeof(match.key->src)); 225 memcpy(&fs->mask.lip[0], match.mask->dst.s6_addr, 226 sizeof(match.mask->dst)); 227 memcpy(&fs->mask.fip[0], match.mask->src.s6_addr, 228 sizeof(match.mask->src)); 229 230 /* also initialize nat_lip/fip to same values */ 231 memcpy(&fs->nat_lip[0], match.key->dst.s6_addr, 232 sizeof(match.key->dst)); 233 memcpy(&fs->nat_fip[0], match.key->src.s6_addr, 234 sizeof(match.key->src)); 235 } 236 237 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) { 238 struct flow_match_ports match; 239 240 flow_rule_match_ports(rule, &match); 241 fs->val.lport = be16_to_cpu(match.key->dst); 242 fs->mask.lport = be16_to_cpu(match.mask->dst); 243 fs->val.fport = be16_to_cpu(match.key->src); 244 fs->mask.fport = be16_to_cpu(match.mask->src); 245 246 /* also initialize nat_lport/fport to same values */ 247 fs->nat_lport = fs->val.lport; 248 fs->nat_fport = fs->val.fport; 249 } 250 251 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) { 252 struct flow_match_ip match; 253 254 flow_rule_match_ip(rule, &match); 255 fs->val.tos = match.key->tos; 256 fs->mask.tos = match.mask->tos; 257 } 258 259 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_KEYID)) { 260 struct flow_match_enc_keyid match; 261 262 flow_rule_match_enc_keyid(rule, &match); 263 fs->val.vni = be32_to_cpu(match.key->keyid); 264 fs->mask.vni = be32_to_cpu(match.mask->keyid); 265 if (fs->mask.vni) { 266 fs->val.encap_vld = 1; 267 fs->mask.encap_vld = 1; 268 } 269 } 270 271 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) { 272 struct flow_match_vlan match; 273 u16 vlan_tci, vlan_tci_mask; 274 275 flow_rule_match_vlan(rule, &match); 276 vlan_tci = match.key->vlan_id | (match.key->vlan_priority << 277 VLAN_PRIO_SHIFT); 278 vlan_tci_mask = match.mask->vlan_id | (match.mask->vlan_priority << 279 VLAN_PRIO_SHIFT); 280 fs->val.ivlan = vlan_tci; 281 fs->mask.ivlan = vlan_tci_mask; 282 283 fs->val.ivlan_vld = 1; 284 fs->mask.ivlan_vld = 1; 285 286 /* Chelsio adapters use ivlan_vld bit to match vlan packets 287 * as 802.1Q. Also, when vlan tag is present in packets, 288 * ethtype match is used then to match on ethtype of inner 289 * header ie. the header following the vlan header. 290 * So, set the ivlan_vld based on ethtype info supplied by 291 * TC for vlan packets if its 802.1Q. And then reset the 292 * ethtype value else, hw will try to match the supplied 293 * ethtype value with ethtype of inner header. 294 */ 295 if (fs->val.ethtype == ETH_P_8021Q) { 296 fs->val.ethtype = 0; 297 fs->mask.ethtype = 0; 298 } 299 } 300 301 /* Match only packets coming from the ingress port where this 302 * filter will be created. 303 */ 304 fs->val.iport = netdev2pinfo(dev)->port_id; 305 fs->mask.iport = ~0; 306 } 307 308 static int cxgb4_validate_flow_match(struct netlink_ext_ack *extack, 309 struct flow_rule *rule) 310 { 311 struct flow_dissector *dissector = rule->match.dissector; 312 u16 ethtype_mask = 0; 313 u16 ethtype_key = 0; 314 315 if (dissector->used_keys & 316 ~(BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) | 317 BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) | 318 BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) | 319 BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) | 320 BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) | 321 BIT_ULL(FLOW_DISSECTOR_KEY_ENC_KEYID) | 322 BIT_ULL(FLOW_DISSECTOR_KEY_VLAN) | 323 BIT_ULL(FLOW_DISSECTOR_KEY_IP))) { 324 NL_SET_ERR_MSG_FMT_MOD(extack, 325 "Unsupported key used: 0x%llx", 326 dissector->used_keys); 327 return -EOPNOTSUPP; 328 } 329 330 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) { 331 struct flow_match_basic match; 332 333 flow_rule_match_basic(rule, &match); 334 ethtype_key = ntohs(match.key->n_proto); 335 ethtype_mask = ntohs(match.mask->n_proto); 336 } 337 338 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) { 339 u16 eth_ip_type = ethtype_key & ethtype_mask; 340 struct flow_match_ip match; 341 342 if (eth_ip_type != ETH_P_IP && eth_ip_type != ETH_P_IPV6) { 343 NL_SET_ERR_MSG_MOD(extack, 344 "IP Key supported only with IPv4/v6"); 345 return -EINVAL; 346 } 347 348 flow_rule_match_ip(rule, &match); 349 if (match.mask->ttl) { 350 NL_SET_ERR_MSG_MOD(extack, 351 "ttl match unsupported for offload"); 352 return -EOPNOTSUPP; 353 } 354 } 355 356 return 0; 357 } 358 359 static void offload_pedit(struct ch_filter_specification *fs, u32 val, u32 mask, 360 u8 field) 361 { 362 u32 set_val = val & ~mask; 363 u32 offset = 0; 364 u8 size = 1; 365 int i; 366 367 for (i = 0; i < ARRAY_SIZE(pedits); i++) { 368 if (pedits[i].field == field) { 369 offset = pedits[i].offset; 370 size = pedits[i].size; 371 break; 372 } 373 } 374 memcpy((u8 *)fs + offset, &set_val, size); 375 } 376 377 static void process_pedit_field(struct ch_filter_specification *fs, u32 val, 378 u32 mask, u32 offset, u8 htype, 379 u8 *natmode_flags) 380 { 381 switch (htype) { 382 case FLOW_ACT_MANGLE_HDR_TYPE_ETH: 383 switch (offset) { 384 case PEDIT_ETH_DMAC_31_0: 385 fs->newdmac = 1; 386 offload_pedit(fs, val, mask, ETH_DMAC_31_0); 387 break; 388 case PEDIT_ETH_DMAC_47_32_SMAC_15_0: 389 if (~mask & PEDIT_ETH_DMAC_MASK) 390 offload_pedit(fs, val, mask, ETH_DMAC_47_32); 391 else 392 offload_pedit(fs, val >> 16, mask >> 16, 393 ETH_SMAC_15_0); 394 break; 395 case PEDIT_ETH_SMAC_47_16: 396 fs->newsmac = 1; 397 offload_pedit(fs, val, mask, ETH_SMAC_47_16); 398 } 399 break; 400 case FLOW_ACT_MANGLE_HDR_TYPE_IP4: 401 switch (offset) { 402 case PEDIT_IP4_SRC: 403 offload_pedit(fs, val, mask, IP4_SRC); 404 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 405 break; 406 case PEDIT_IP4_DST: 407 offload_pedit(fs, val, mask, IP4_DST); 408 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 409 } 410 break; 411 case FLOW_ACT_MANGLE_HDR_TYPE_IP6: 412 switch (offset) { 413 case PEDIT_IP6_SRC_31_0: 414 offload_pedit(fs, val, mask, IP6_SRC_31_0); 415 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 416 break; 417 case PEDIT_IP6_SRC_63_32: 418 offload_pedit(fs, val, mask, IP6_SRC_63_32); 419 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 420 break; 421 case PEDIT_IP6_SRC_95_64: 422 offload_pedit(fs, val, mask, IP6_SRC_95_64); 423 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 424 break; 425 case PEDIT_IP6_SRC_127_96: 426 offload_pedit(fs, val, mask, IP6_SRC_127_96); 427 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 428 break; 429 case PEDIT_IP6_DST_31_0: 430 offload_pedit(fs, val, mask, IP6_DST_31_0); 431 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 432 break; 433 case PEDIT_IP6_DST_63_32: 434 offload_pedit(fs, val, mask, IP6_DST_63_32); 435 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 436 break; 437 case PEDIT_IP6_DST_95_64: 438 offload_pedit(fs, val, mask, IP6_DST_95_64); 439 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 440 break; 441 case PEDIT_IP6_DST_127_96: 442 offload_pedit(fs, val, mask, IP6_DST_127_96); 443 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 444 } 445 break; 446 case FLOW_ACT_MANGLE_HDR_TYPE_TCP: 447 switch (offset) { 448 case PEDIT_TCP_SPORT_DPORT: 449 if (~mask & PEDIT_TCP_UDP_SPORT_MASK) { 450 fs->nat_fport = val; 451 *natmode_flags |= CXGB4_ACTION_NATMODE_SPORT; 452 } else { 453 fs->nat_lport = val >> 16; 454 *natmode_flags |= CXGB4_ACTION_NATMODE_DPORT; 455 } 456 } 457 break; 458 case FLOW_ACT_MANGLE_HDR_TYPE_UDP: 459 switch (offset) { 460 case PEDIT_UDP_SPORT_DPORT: 461 if (~mask & PEDIT_TCP_UDP_SPORT_MASK) { 462 fs->nat_fport = val; 463 *natmode_flags |= CXGB4_ACTION_NATMODE_SPORT; 464 } else { 465 fs->nat_lport = val >> 16; 466 *natmode_flags |= CXGB4_ACTION_NATMODE_DPORT; 467 } 468 } 469 break; 470 } 471 } 472 473 static int cxgb4_action_natmode_validate(struct adapter *adap, u8 natmode_flags, 474 struct netlink_ext_ack *extack) 475 { 476 u8 i = 0; 477 478 /* Extract the NAT mode to enable based on what 4-tuple fields 479 * are enabled to be overwritten. This ensures that the 480 * disabled fields don't get overwritten to 0. 481 */ 482 for (i = 0; i < ARRAY_SIZE(cxgb4_natmode_config_array); i++) { 483 const struct cxgb4_natmode_config *c; 484 485 c = &cxgb4_natmode_config_array[i]; 486 if (CHELSIO_CHIP_VERSION(adap->params.chip) >= c->chip && 487 natmode_flags == c->flags) 488 return 0; 489 } 490 NL_SET_ERR_MSG_MOD(extack, "Unsupported NAT mode 4-tuple combination"); 491 return -EOPNOTSUPP; 492 } 493 494 void cxgb4_process_flow_actions(struct net_device *in, 495 struct flow_action *actions, 496 struct ch_filter_specification *fs) 497 { 498 struct flow_action_entry *act; 499 u8 natmode_flags = 0; 500 int i; 501 502 flow_action_for_each(i, act, actions) { 503 switch (act->id) { 504 case FLOW_ACTION_ACCEPT: 505 fs->action = FILTER_PASS; 506 break; 507 case FLOW_ACTION_DROP: 508 fs->action = FILTER_DROP; 509 break; 510 case FLOW_ACTION_MIRRED: 511 case FLOW_ACTION_REDIRECT: { 512 struct net_device *out = act->dev; 513 struct port_info *pi = netdev_priv(out); 514 515 fs->action = FILTER_SWITCH; 516 fs->eport = pi->port_id; 517 } 518 break; 519 case FLOW_ACTION_VLAN_POP: 520 case FLOW_ACTION_VLAN_PUSH: 521 case FLOW_ACTION_VLAN_MANGLE: { 522 u8 prio = act->vlan.prio; 523 u16 vid = act->vlan.vid; 524 u16 vlan_tci = (prio << VLAN_PRIO_SHIFT) | vid; 525 switch (act->id) { 526 case FLOW_ACTION_VLAN_POP: 527 fs->newvlan |= VLAN_REMOVE; 528 break; 529 case FLOW_ACTION_VLAN_PUSH: 530 fs->newvlan |= VLAN_INSERT; 531 fs->vlan = vlan_tci; 532 break; 533 case FLOW_ACTION_VLAN_MANGLE: 534 fs->newvlan |= VLAN_REWRITE; 535 fs->vlan = vlan_tci; 536 break; 537 default: 538 break; 539 } 540 } 541 break; 542 case FLOW_ACTION_MANGLE: { 543 u32 mask, val, offset; 544 u8 htype; 545 546 htype = act->mangle.htype; 547 mask = act->mangle.mask; 548 val = act->mangle.val; 549 offset = act->mangle.offset; 550 551 process_pedit_field(fs, val, mask, offset, htype, 552 &natmode_flags); 553 } 554 break; 555 case FLOW_ACTION_QUEUE: 556 fs->action = FILTER_PASS; 557 fs->dirsteer = 1; 558 fs->iq = act->queue.index; 559 break; 560 default: 561 break; 562 } 563 } 564 if (natmode_flags) 565 cxgb4_action_natmode_tweak(fs, natmode_flags); 566 567 } 568 569 static bool valid_l4_mask(u32 mask) 570 { 571 u16 hi, lo; 572 573 /* Either the upper 16-bits (SPORT) OR the lower 574 * 16-bits (DPORT) can be set, but NOT BOTH. 575 */ 576 hi = (mask >> 16) & 0xFFFF; 577 lo = mask & 0xFFFF; 578 579 return hi && lo ? false : true; 580 } 581 582 static bool valid_pedit_action(struct netlink_ext_ack *extack, 583 const struct flow_action_entry *act, 584 u8 *natmode_flags) 585 { 586 u32 mask, offset; 587 u8 htype; 588 589 htype = act->mangle.htype; 590 mask = act->mangle.mask; 591 offset = act->mangle.offset; 592 593 switch (htype) { 594 case FLOW_ACT_MANGLE_HDR_TYPE_ETH: 595 switch (offset) { 596 case PEDIT_ETH_DMAC_31_0: 597 case PEDIT_ETH_DMAC_47_32_SMAC_15_0: 598 case PEDIT_ETH_SMAC_47_16: 599 break; 600 default: 601 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field"); 602 return false; 603 } 604 break; 605 case FLOW_ACT_MANGLE_HDR_TYPE_IP4: 606 switch (offset) { 607 case PEDIT_IP4_SRC: 608 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 609 break; 610 case PEDIT_IP4_DST: 611 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 612 break; 613 default: 614 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field"); 615 return false; 616 } 617 break; 618 case FLOW_ACT_MANGLE_HDR_TYPE_IP6: 619 switch (offset) { 620 case PEDIT_IP6_SRC_31_0: 621 case PEDIT_IP6_SRC_63_32: 622 case PEDIT_IP6_SRC_95_64: 623 case PEDIT_IP6_SRC_127_96: 624 *natmode_flags |= CXGB4_ACTION_NATMODE_SIP; 625 break; 626 case PEDIT_IP6_DST_31_0: 627 case PEDIT_IP6_DST_63_32: 628 case PEDIT_IP6_DST_95_64: 629 case PEDIT_IP6_DST_127_96: 630 *natmode_flags |= CXGB4_ACTION_NATMODE_DIP; 631 break; 632 default: 633 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field"); 634 return false; 635 } 636 break; 637 case FLOW_ACT_MANGLE_HDR_TYPE_TCP: 638 switch (offset) { 639 case PEDIT_TCP_SPORT_DPORT: 640 if (!valid_l4_mask(~mask)) { 641 NL_SET_ERR_MSG_MOD(extack, 642 "Unsupported mask for TCP L4 ports"); 643 return false; 644 } 645 if (~mask & PEDIT_TCP_UDP_SPORT_MASK) 646 *natmode_flags |= CXGB4_ACTION_NATMODE_SPORT; 647 else 648 *natmode_flags |= CXGB4_ACTION_NATMODE_DPORT; 649 break; 650 default: 651 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field"); 652 return false; 653 } 654 break; 655 case FLOW_ACT_MANGLE_HDR_TYPE_UDP: 656 switch (offset) { 657 case PEDIT_UDP_SPORT_DPORT: 658 if (!valid_l4_mask(~mask)) { 659 NL_SET_ERR_MSG_MOD(extack, 660 "Unsupported mask for UDP L4 ports"); 661 return false; 662 } 663 if (~mask & PEDIT_TCP_UDP_SPORT_MASK) 664 *natmode_flags |= CXGB4_ACTION_NATMODE_SPORT; 665 else 666 *natmode_flags |= CXGB4_ACTION_NATMODE_DPORT; 667 break; 668 default: 669 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit field"); 670 return false; 671 } 672 break; 673 default: 674 NL_SET_ERR_MSG_MOD(extack, "Unsupported pedit type"); 675 return false; 676 } 677 return true; 678 } 679 680 int cxgb4_validate_flow_actions(struct net_device *dev, 681 struct flow_action *actions, 682 struct netlink_ext_ack *extack, 683 u8 matchall_filter) 684 { 685 struct adapter *adap = netdev2adap(dev); 686 struct flow_action_entry *act; 687 bool act_redir = false; 688 bool act_pedit = false; 689 bool act_vlan = false; 690 u8 natmode_flags = 0; 691 int i; 692 693 if (!flow_action_basic_hw_stats_check(actions, extack)) 694 return -EOPNOTSUPP; 695 696 flow_action_for_each(i, act, actions) { 697 switch (act->id) { 698 case FLOW_ACTION_ACCEPT: 699 case FLOW_ACTION_DROP: 700 /* Do nothing */ 701 break; 702 case FLOW_ACTION_MIRRED: 703 case FLOW_ACTION_REDIRECT: { 704 struct net_device *n_dev, *target_dev; 705 bool found = false; 706 unsigned int i; 707 708 if (act->id == FLOW_ACTION_MIRRED && 709 !matchall_filter) { 710 NL_SET_ERR_MSG_MOD(extack, 711 "Egress mirror action is only supported for tc-matchall"); 712 return -EOPNOTSUPP; 713 } 714 715 target_dev = act->dev; 716 for_each_port(adap, i) { 717 n_dev = adap->port[i]; 718 if (target_dev == n_dev) { 719 found = true; 720 break; 721 } 722 } 723 724 /* If interface doesn't belong to our hw, then 725 * the provided output port is not valid 726 */ 727 if (!found) { 728 NL_SET_ERR_MSG_MOD(extack, "Out port invalid"); 729 return -EINVAL; 730 } 731 act_redir = true; 732 } 733 break; 734 case FLOW_ACTION_VLAN_POP: 735 case FLOW_ACTION_VLAN_PUSH: 736 case FLOW_ACTION_VLAN_MANGLE: { 737 u16 proto = be16_to_cpu(act->vlan.proto); 738 739 switch (act->id) { 740 case FLOW_ACTION_VLAN_POP: 741 break; 742 case FLOW_ACTION_VLAN_PUSH: 743 case FLOW_ACTION_VLAN_MANGLE: 744 if (proto != ETH_P_8021Q) { 745 NL_SET_ERR_MSG_MOD(extack, 746 "Unsupported vlan proto"); 747 return -EOPNOTSUPP; 748 } 749 break; 750 default: 751 NL_SET_ERR_MSG_MOD(extack, 752 "Unsupported vlan action"); 753 return -EOPNOTSUPP; 754 } 755 act_vlan = true; 756 } 757 break; 758 case FLOW_ACTION_MANGLE: { 759 bool pedit_valid = valid_pedit_action(extack, act, 760 &natmode_flags); 761 762 if (!pedit_valid) 763 return -EOPNOTSUPP; 764 act_pedit = true; 765 } 766 break; 767 case FLOW_ACTION_QUEUE: 768 /* Do nothing. cxgb4_set_filter will validate */ 769 break; 770 default: 771 NL_SET_ERR_MSG_MOD(extack, "Unsupported action"); 772 return -EOPNOTSUPP; 773 } 774 } 775 776 if ((act_pedit || act_vlan) && !act_redir) { 777 NL_SET_ERR_MSG_MOD(extack, 778 "pedit/vlan rewrite invalid without egress redirect"); 779 return -EINVAL; 780 } 781 782 if (act_pedit) { 783 int ret; 784 785 ret = cxgb4_action_natmode_validate(adap, natmode_flags, 786 extack); 787 if (ret) 788 return ret; 789 } 790 791 return 0; 792 } 793 794 static void cxgb4_tc_flower_hash_prio_add(struct adapter *adap, u32 tc_prio) 795 { 796 spin_lock_bh(&adap->tids.ftid_lock); 797 if (adap->tids.tc_hash_tids_max_prio < tc_prio) 798 adap->tids.tc_hash_tids_max_prio = tc_prio; 799 spin_unlock_bh(&adap->tids.ftid_lock); 800 } 801 802 static void cxgb4_tc_flower_hash_prio_del(struct adapter *adap, u32 tc_prio) 803 { 804 struct tid_info *t = &adap->tids; 805 struct ch_tc_flower_entry *fe; 806 struct rhashtable_iter iter; 807 u32 found = 0; 808 809 spin_lock_bh(&t->ftid_lock); 810 /* Bail if the current rule is not the one with the max 811 * prio. 812 */ 813 if (t->tc_hash_tids_max_prio != tc_prio) 814 goto out_unlock; 815 816 /* Search for the next rule having the same or next lower 817 * max prio. 818 */ 819 rhashtable_walk_enter(&adap->flower_tbl, &iter); 820 do { 821 rhashtable_walk_start(&iter); 822 823 fe = rhashtable_walk_next(&iter); 824 while (!IS_ERR_OR_NULL(fe)) { 825 if (fe->fs.hash && 826 fe->fs.tc_prio <= t->tc_hash_tids_max_prio) { 827 t->tc_hash_tids_max_prio = fe->fs.tc_prio; 828 found++; 829 830 /* Bail if we found another rule 831 * having the same prio as the 832 * current max one. 833 */ 834 if (fe->fs.tc_prio == tc_prio) 835 break; 836 } 837 838 fe = rhashtable_walk_next(&iter); 839 } 840 841 rhashtable_walk_stop(&iter); 842 } while (fe == ERR_PTR(-EAGAIN)); 843 rhashtable_walk_exit(&iter); 844 845 if (!found) 846 t->tc_hash_tids_max_prio = 0; 847 848 out_unlock: 849 spin_unlock_bh(&t->ftid_lock); 850 } 851 852 int cxgb4_flow_rule_replace(struct net_device *dev, struct flow_rule *rule, 853 u32 tc_prio, struct netlink_ext_ack *extack, 854 struct ch_filter_specification *fs, u32 *tid) 855 { 856 struct adapter *adap = netdev2adap(dev); 857 struct filter_ctx ctx; 858 u8 inet_family; 859 int fidx, ret; 860 861 if (cxgb4_validate_flow_actions(dev, &rule->action, extack, 0)) 862 return -EOPNOTSUPP; 863 864 if (cxgb4_validate_flow_match(extack, rule)) 865 return -EOPNOTSUPP; 866 867 cxgb4_process_flow_match(dev, rule, fs); 868 cxgb4_process_flow_actions(dev, &rule->action, fs); 869 870 fs->hash = is_filter_exact_match(adap, fs); 871 inet_family = fs->type ? PF_INET6 : PF_INET; 872 873 /* Get a free filter entry TID, where we can insert this new 874 * rule. Only insert rule if its prio doesn't conflict with 875 * existing rules. 876 */ 877 fidx = cxgb4_get_free_ftid(dev, inet_family, fs->hash, 878 tc_prio); 879 if (fidx < 0) { 880 NL_SET_ERR_MSG_MOD(extack, 881 "No free LETCAM index available"); 882 return -ENOMEM; 883 } 884 885 if (fidx < adap->tids.nhpftids) { 886 fs->prio = 1; 887 fs->hash = 0; 888 } 889 890 /* If the rule can be inserted into HASH region, then ignore 891 * the index to normal FILTER region. 892 */ 893 if (fs->hash) 894 fidx = 0; 895 896 fs->tc_prio = tc_prio; 897 898 init_completion(&ctx.completion); 899 ret = __cxgb4_set_filter(dev, fidx, fs, &ctx); 900 if (ret) { 901 NL_SET_ERR_MSG_FMT_MOD(extack, "filter creation err %d", ret); 902 return ret; 903 } 904 905 /* Wait for reply */ 906 ret = wait_for_completion_timeout(&ctx.completion, 10 * HZ); 907 if (!ret) 908 return -ETIMEDOUT; 909 910 /* Check if hw returned error for filter creation */ 911 if (ctx.result) 912 return ctx.result; 913 914 *tid = ctx.tid; 915 916 if (fs->hash) 917 cxgb4_tc_flower_hash_prio_add(adap, tc_prio); 918 919 return 0; 920 } 921 922 int cxgb4_tc_flower_replace(struct net_device *dev, 923 struct flow_cls_offload *cls) 924 { 925 struct flow_rule *rule = flow_cls_offload_flow_rule(cls); 926 struct netlink_ext_ack *extack = cls->common.extack; 927 struct adapter *adap = netdev2adap(dev); 928 struct ch_tc_flower_entry *ch_flower; 929 struct ch_filter_specification *fs; 930 int ret; 931 932 ch_flower = allocate_flower_entry(); 933 if (!ch_flower) { 934 netdev_err(dev, "%s: ch_flower alloc failed.\n", __func__); 935 return -ENOMEM; 936 } 937 938 fs = &ch_flower->fs; 939 fs->hitcnts = 1; 940 fs->tc_cookie = cls->cookie; 941 942 ret = cxgb4_flow_rule_replace(dev, rule, cls->common.prio, extack, fs, 943 &ch_flower->filter_id); 944 if (ret) 945 goto free_entry; 946 947 ch_flower->tc_flower_cookie = cls->cookie; 948 ret = rhashtable_insert_fast(&adap->flower_tbl, &ch_flower->node, 949 adap->flower_ht_params); 950 if (ret) 951 goto del_filter; 952 953 return 0; 954 955 del_filter: 956 if (fs->hash) 957 cxgb4_tc_flower_hash_prio_del(adap, cls->common.prio); 958 959 cxgb4_del_filter(dev, ch_flower->filter_id, &ch_flower->fs); 960 961 free_entry: 962 kfree(ch_flower); 963 return ret; 964 } 965 966 int cxgb4_flow_rule_destroy(struct net_device *dev, u32 tc_prio, 967 struct ch_filter_specification *fs, int tid) 968 { 969 struct adapter *adap = netdev2adap(dev); 970 u8 hash; 971 int ret; 972 973 hash = fs->hash; 974 975 ret = cxgb4_del_filter(dev, tid, fs); 976 if (ret) 977 return ret; 978 979 if (hash) 980 cxgb4_tc_flower_hash_prio_del(adap, tc_prio); 981 982 return ret; 983 } 984 985 int cxgb4_tc_flower_destroy(struct net_device *dev, 986 struct flow_cls_offload *cls) 987 { 988 struct adapter *adap = netdev2adap(dev); 989 struct ch_tc_flower_entry *ch_flower; 990 int ret; 991 992 ch_flower = ch_flower_lookup(adap, cls->cookie); 993 if (!ch_flower) 994 return -ENOENT; 995 996 rhashtable_remove_fast(&adap->flower_tbl, &ch_flower->node, 997 adap->flower_ht_params); 998 999 ret = cxgb4_flow_rule_destroy(dev, ch_flower->fs.tc_prio, 1000 &ch_flower->fs, ch_flower->filter_id); 1001 if (ret) 1002 netdev_err(dev, "Flow rule destroy failed for tid: %u, ret: %d", 1003 ch_flower->filter_id, ret); 1004 1005 kfree_rcu(ch_flower, rcu); 1006 return ret; 1007 } 1008 1009 static void ch_flower_stats_handler(struct work_struct *work) 1010 { 1011 struct adapter *adap = container_of(work, struct adapter, 1012 flower_stats_work); 1013 struct ch_tc_flower_entry *flower_entry; 1014 struct ch_tc_flower_stats *ofld_stats; 1015 struct rhashtable_iter iter; 1016 u64 packets; 1017 u64 bytes; 1018 int ret; 1019 1020 rhashtable_walk_enter(&adap->flower_tbl, &iter); 1021 do { 1022 rhashtable_walk_start(&iter); 1023 1024 while ((flower_entry = rhashtable_walk_next(&iter)) && 1025 !IS_ERR(flower_entry)) { 1026 ret = cxgb4_get_filter_counters(adap->port[0], 1027 flower_entry->filter_id, 1028 &packets, &bytes, 1029 flower_entry->fs.hash); 1030 if (!ret) { 1031 spin_lock(&flower_entry->lock); 1032 ofld_stats = &flower_entry->stats; 1033 1034 if (ofld_stats->prev_packet_count != packets) { 1035 ofld_stats->prev_packet_count = packets; 1036 ofld_stats->last_used = jiffies; 1037 } 1038 spin_unlock(&flower_entry->lock); 1039 } 1040 } 1041 1042 rhashtable_walk_stop(&iter); 1043 1044 } while (flower_entry == ERR_PTR(-EAGAIN)); 1045 rhashtable_walk_exit(&iter); 1046 mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD); 1047 } 1048 1049 static void ch_flower_stats_cb(struct timer_list *t) 1050 { 1051 struct adapter *adap = from_timer(adap, t, flower_stats_timer); 1052 1053 schedule_work(&adap->flower_stats_work); 1054 } 1055 1056 int cxgb4_tc_flower_stats(struct net_device *dev, 1057 struct flow_cls_offload *cls) 1058 { 1059 struct adapter *adap = netdev2adap(dev); 1060 struct ch_tc_flower_stats *ofld_stats; 1061 struct ch_tc_flower_entry *ch_flower; 1062 u64 packets; 1063 u64 bytes; 1064 int ret; 1065 1066 ch_flower = ch_flower_lookup(adap, cls->cookie); 1067 if (!ch_flower) { 1068 ret = -ENOENT; 1069 goto err; 1070 } 1071 1072 ret = cxgb4_get_filter_counters(dev, ch_flower->filter_id, 1073 &packets, &bytes, 1074 ch_flower->fs.hash); 1075 if (ret < 0) 1076 goto err; 1077 1078 spin_lock_bh(&ch_flower->lock); 1079 ofld_stats = &ch_flower->stats; 1080 if (ofld_stats->packet_count != packets) { 1081 if (ofld_stats->prev_packet_count != packets) 1082 ofld_stats->last_used = jiffies; 1083 flow_stats_update(&cls->stats, bytes - ofld_stats->byte_count, 1084 packets - ofld_stats->packet_count, 0, 1085 ofld_stats->last_used, 1086 FLOW_ACTION_HW_STATS_IMMEDIATE); 1087 1088 ofld_stats->packet_count = packets; 1089 ofld_stats->byte_count = bytes; 1090 ofld_stats->prev_packet_count = packets; 1091 } 1092 spin_unlock_bh(&ch_flower->lock); 1093 return 0; 1094 1095 err: 1096 return ret; 1097 } 1098 1099 static const struct rhashtable_params cxgb4_tc_flower_ht_params = { 1100 .nelem_hint = 384, 1101 .head_offset = offsetof(struct ch_tc_flower_entry, node), 1102 .key_offset = offsetof(struct ch_tc_flower_entry, tc_flower_cookie), 1103 .key_len = sizeof(((struct ch_tc_flower_entry *)0)->tc_flower_cookie), 1104 .max_size = 524288, 1105 .min_size = 512, 1106 .automatic_shrinking = true 1107 }; 1108 1109 int cxgb4_init_tc_flower(struct adapter *adap) 1110 { 1111 int ret; 1112 1113 if (adap->tc_flower_initialized) 1114 return -EEXIST; 1115 1116 adap->flower_ht_params = cxgb4_tc_flower_ht_params; 1117 ret = rhashtable_init(&adap->flower_tbl, &adap->flower_ht_params); 1118 if (ret) 1119 return ret; 1120 1121 INIT_WORK(&adap->flower_stats_work, ch_flower_stats_handler); 1122 timer_setup(&adap->flower_stats_timer, ch_flower_stats_cb, 0); 1123 mod_timer(&adap->flower_stats_timer, jiffies + STATS_CHECK_PERIOD); 1124 adap->tc_flower_initialized = true; 1125 return 0; 1126 } 1127 1128 void cxgb4_cleanup_tc_flower(struct adapter *adap) 1129 { 1130 if (!adap->tc_flower_initialized) 1131 return; 1132 1133 if (adap->flower_stats_timer.function) 1134 timer_shutdown_sync(&adap->flower_stats_timer); 1135 cancel_work_sync(&adap->flower_stats_work); 1136 rhashtable_destroy(&adap->flower_tbl); 1137 adap->tc_flower_initialized = false; 1138 } 1139