1 /* 2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved. 4 * Copyright (c) 2006 Intel Corporation. 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 <linux/module.h> 36 #include <linux/init.h> 37 #include <linux/err.h> 38 #include <linux/random.h> 39 #include <linux/spinlock.h> 40 #include <linux/slab.h> 41 #include <linux/dma-mapping.h> 42 #include <linux/kref.h> 43 #include <linux/idr.h> 44 #include <linux/workqueue.h> 45 #include <uapi/linux/if_ether.h> 46 #include <rdma/ib_pack.h> 47 #include <rdma/ib_cache.h> 48 #include <rdma/rdma_netlink.h> 49 #include <net/netlink.h> 50 #include <uapi/rdma/ib_user_sa.h> 51 #include <rdma/ib_marshall.h> 52 #include <rdma/ib_addr.h> 53 #include <rdma/opa_addr.h> 54 #include "sa.h" 55 #include "core_priv.h" 56 57 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN 100 58 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT 2000 59 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX 200000 60 #define IB_SA_CPI_MAX_RETRY_CNT 3 61 #define IB_SA_CPI_RETRY_WAIT 1000 /*msecs */ 62 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT; 63 64 struct ib_sa_sm_ah { 65 struct ib_ah *ah; 66 struct kref ref; 67 u16 pkey_index; 68 u8 src_path_mask; 69 }; 70 71 enum rdma_class_port_info_type { 72 RDMA_CLASS_PORT_INFO_IB, 73 RDMA_CLASS_PORT_INFO_OPA 74 }; 75 76 struct rdma_class_port_info { 77 enum rdma_class_port_info_type type; 78 union { 79 struct ib_class_port_info ib; 80 struct opa_class_port_info opa; 81 }; 82 }; 83 84 struct ib_sa_classport_cache { 85 bool valid; 86 int retry_cnt; 87 struct rdma_class_port_info data; 88 }; 89 90 struct ib_sa_port { 91 struct ib_mad_agent *agent; 92 struct ib_sa_sm_ah *sm_ah; 93 struct work_struct update_task; 94 struct ib_sa_classport_cache classport_info; 95 struct delayed_work ib_cpi_work; 96 spinlock_t classport_lock; /* protects class port info set */ 97 spinlock_t ah_lock; 98 u8 port_num; 99 }; 100 101 struct ib_sa_device { 102 int start_port, end_port; 103 struct ib_event_handler event_handler; 104 struct ib_sa_port port[0]; 105 }; 106 107 struct ib_sa_query { 108 void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *); 109 void (*release)(struct ib_sa_query *); 110 struct ib_sa_client *client; 111 struct ib_sa_port *port; 112 struct ib_mad_send_buf *mad_buf; 113 struct ib_sa_sm_ah *sm_ah; 114 int id; 115 u32 flags; 116 struct list_head list; /* Local svc request list */ 117 u32 seq; /* Local svc request sequence number */ 118 unsigned long timeout; /* Local svc timeout */ 119 u8 path_use; /* How will the pathrecord be used */ 120 }; 121 122 #define IB_SA_ENABLE_LOCAL_SERVICE 0x00000001 123 #define IB_SA_CANCEL 0x00000002 124 #define IB_SA_QUERY_OPA 0x00000004 125 126 struct ib_sa_service_query { 127 void (*callback)(int, struct ib_sa_service_rec *, void *); 128 void *context; 129 struct ib_sa_query sa_query; 130 }; 131 132 struct ib_sa_path_query { 133 void (*callback)(int, struct sa_path_rec *, void *); 134 void *context; 135 struct ib_sa_query sa_query; 136 struct sa_path_rec *conv_pr; 137 }; 138 139 struct ib_sa_guidinfo_query { 140 void (*callback)(int, struct ib_sa_guidinfo_rec *, void *); 141 void *context; 142 struct ib_sa_query sa_query; 143 }; 144 145 struct ib_sa_classport_info_query { 146 void (*callback)(void *); 147 void *context; 148 struct ib_sa_query sa_query; 149 }; 150 151 struct ib_sa_mcmember_query { 152 void (*callback)(int, struct ib_sa_mcmember_rec *, void *); 153 void *context; 154 struct ib_sa_query sa_query; 155 }; 156 157 static LIST_HEAD(ib_nl_request_list); 158 static DEFINE_SPINLOCK(ib_nl_request_lock); 159 static atomic_t ib_nl_sa_request_seq; 160 static struct workqueue_struct *ib_nl_wq; 161 static struct delayed_work ib_nl_timed_work; 162 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = { 163 [LS_NLA_TYPE_PATH_RECORD] = {.type = NLA_BINARY, 164 .len = sizeof(struct ib_path_rec_data)}, 165 [LS_NLA_TYPE_TIMEOUT] = {.type = NLA_U32}, 166 [LS_NLA_TYPE_SERVICE_ID] = {.type = NLA_U64}, 167 [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY, 168 .len = sizeof(struct rdma_nla_ls_gid)}, 169 [LS_NLA_TYPE_SGID] = {.type = NLA_BINARY, 170 .len = sizeof(struct rdma_nla_ls_gid)}, 171 [LS_NLA_TYPE_TCLASS] = {.type = NLA_U8}, 172 [LS_NLA_TYPE_PKEY] = {.type = NLA_U16}, 173 [LS_NLA_TYPE_QOS_CLASS] = {.type = NLA_U16}, 174 }; 175 176 177 static void ib_sa_add_one(struct ib_device *device); 178 static void ib_sa_remove_one(struct ib_device *device, void *client_data); 179 180 static struct ib_client sa_client = { 181 .name = "sa", 182 .add = ib_sa_add_one, 183 .remove = ib_sa_remove_one 184 }; 185 186 static DEFINE_SPINLOCK(idr_lock); 187 static DEFINE_IDR(query_idr); 188 189 static DEFINE_SPINLOCK(tid_lock); 190 static u32 tid; 191 192 #define PATH_REC_FIELD(field) \ 193 .struct_offset_bytes = offsetof(struct sa_path_rec, field), \ 194 .struct_size_bytes = sizeof((struct sa_path_rec *)0)->field, \ 195 .field_name = "sa_path_rec:" #field 196 197 static const struct ib_field path_rec_table[] = { 198 { PATH_REC_FIELD(service_id), 199 .offset_words = 0, 200 .offset_bits = 0, 201 .size_bits = 64 }, 202 { PATH_REC_FIELD(dgid), 203 .offset_words = 2, 204 .offset_bits = 0, 205 .size_bits = 128 }, 206 { PATH_REC_FIELD(sgid), 207 .offset_words = 6, 208 .offset_bits = 0, 209 .size_bits = 128 }, 210 { PATH_REC_FIELD(ib.dlid), 211 .offset_words = 10, 212 .offset_bits = 0, 213 .size_bits = 16 }, 214 { PATH_REC_FIELD(ib.slid), 215 .offset_words = 10, 216 .offset_bits = 16, 217 .size_bits = 16 }, 218 { PATH_REC_FIELD(ib.raw_traffic), 219 .offset_words = 11, 220 .offset_bits = 0, 221 .size_bits = 1 }, 222 { RESERVED, 223 .offset_words = 11, 224 .offset_bits = 1, 225 .size_bits = 3 }, 226 { PATH_REC_FIELD(flow_label), 227 .offset_words = 11, 228 .offset_bits = 4, 229 .size_bits = 20 }, 230 { PATH_REC_FIELD(hop_limit), 231 .offset_words = 11, 232 .offset_bits = 24, 233 .size_bits = 8 }, 234 { PATH_REC_FIELD(traffic_class), 235 .offset_words = 12, 236 .offset_bits = 0, 237 .size_bits = 8 }, 238 { PATH_REC_FIELD(reversible), 239 .offset_words = 12, 240 .offset_bits = 8, 241 .size_bits = 1 }, 242 { PATH_REC_FIELD(numb_path), 243 .offset_words = 12, 244 .offset_bits = 9, 245 .size_bits = 7 }, 246 { PATH_REC_FIELD(pkey), 247 .offset_words = 12, 248 .offset_bits = 16, 249 .size_bits = 16 }, 250 { PATH_REC_FIELD(qos_class), 251 .offset_words = 13, 252 .offset_bits = 0, 253 .size_bits = 12 }, 254 { PATH_REC_FIELD(sl), 255 .offset_words = 13, 256 .offset_bits = 12, 257 .size_bits = 4 }, 258 { PATH_REC_FIELD(mtu_selector), 259 .offset_words = 13, 260 .offset_bits = 16, 261 .size_bits = 2 }, 262 { PATH_REC_FIELD(mtu), 263 .offset_words = 13, 264 .offset_bits = 18, 265 .size_bits = 6 }, 266 { PATH_REC_FIELD(rate_selector), 267 .offset_words = 13, 268 .offset_bits = 24, 269 .size_bits = 2 }, 270 { PATH_REC_FIELD(rate), 271 .offset_words = 13, 272 .offset_bits = 26, 273 .size_bits = 6 }, 274 { PATH_REC_FIELD(packet_life_time_selector), 275 .offset_words = 14, 276 .offset_bits = 0, 277 .size_bits = 2 }, 278 { PATH_REC_FIELD(packet_life_time), 279 .offset_words = 14, 280 .offset_bits = 2, 281 .size_bits = 6 }, 282 { PATH_REC_FIELD(preference), 283 .offset_words = 14, 284 .offset_bits = 8, 285 .size_bits = 8 }, 286 { RESERVED, 287 .offset_words = 14, 288 .offset_bits = 16, 289 .size_bits = 48 }, 290 }; 291 292 #define OPA_PATH_REC_FIELD(field) \ 293 .struct_offset_bytes = \ 294 offsetof(struct sa_path_rec, field), \ 295 .struct_size_bytes = \ 296 sizeof((struct sa_path_rec *)0)->field, \ 297 .field_name = "sa_path_rec:" #field 298 299 static const struct ib_field opa_path_rec_table[] = { 300 { OPA_PATH_REC_FIELD(service_id), 301 .offset_words = 0, 302 .offset_bits = 0, 303 .size_bits = 64 }, 304 { OPA_PATH_REC_FIELD(dgid), 305 .offset_words = 2, 306 .offset_bits = 0, 307 .size_bits = 128 }, 308 { OPA_PATH_REC_FIELD(sgid), 309 .offset_words = 6, 310 .offset_bits = 0, 311 .size_bits = 128 }, 312 { OPA_PATH_REC_FIELD(opa.dlid), 313 .offset_words = 10, 314 .offset_bits = 0, 315 .size_bits = 32 }, 316 { OPA_PATH_REC_FIELD(opa.slid), 317 .offset_words = 11, 318 .offset_bits = 0, 319 .size_bits = 32 }, 320 { OPA_PATH_REC_FIELD(opa.raw_traffic), 321 .offset_words = 12, 322 .offset_bits = 0, 323 .size_bits = 1 }, 324 { RESERVED, 325 .offset_words = 12, 326 .offset_bits = 1, 327 .size_bits = 3 }, 328 { OPA_PATH_REC_FIELD(flow_label), 329 .offset_words = 12, 330 .offset_bits = 4, 331 .size_bits = 20 }, 332 { OPA_PATH_REC_FIELD(hop_limit), 333 .offset_words = 12, 334 .offset_bits = 24, 335 .size_bits = 8 }, 336 { OPA_PATH_REC_FIELD(traffic_class), 337 .offset_words = 13, 338 .offset_bits = 0, 339 .size_bits = 8 }, 340 { OPA_PATH_REC_FIELD(reversible), 341 .offset_words = 13, 342 .offset_bits = 8, 343 .size_bits = 1 }, 344 { OPA_PATH_REC_FIELD(numb_path), 345 .offset_words = 13, 346 .offset_bits = 9, 347 .size_bits = 7 }, 348 { OPA_PATH_REC_FIELD(pkey), 349 .offset_words = 13, 350 .offset_bits = 16, 351 .size_bits = 16 }, 352 { OPA_PATH_REC_FIELD(opa.l2_8B), 353 .offset_words = 14, 354 .offset_bits = 0, 355 .size_bits = 1 }, 356 { OPA_PATH_REC_FIELD(opa.l2_10B), 357 .offset_words = 14, 358 .offset_bits = 1, 359 .size_bits = 1 }, 360 { OPA_PATH_REC_FIELD(opa.l2_9B), 361 .offset_words = 14, 362 .offset_bits = 2, 363 .size_bits = 1 }, 364 { OPA_PATH_REC_FIELD(opa.l2_16B), 365 .offset_words = 14, 366 .offset_bits = 3, 367 .size_bits = 1 }, 368 { RESERVED, 369 .offset_words = 14, 370 .offset_bits = 4, 371 .size_bits = 2 }, 372 { OPA_PATH_REC_FIELD(opa.qos_type), 373 .offset_words = 14, 374 .offset_bits = 6, 375 .size_bits = 2 }, 376 { OPA_PATH_REC_FIELD(opa.qos_priority), 377 .offset_words = 14, 378 .offset_bits = 8, 379 .size_bits = 8 }, 380 { RESERVED, 381 .offset_words = 14, 382 .offset_bits = 16, 383 .size_bits = 3 }, 384 { OPA_PATH_REC_FIELD(sl), 385 .offset_words = 14, 386 .offset_bits = 19, 387 .size_bits = 5 }, 388 { RESERVED, 389 .offset_words = 14, 390 .offset_bits = 24, 391 .size_bits = 8 }, 392 { OPA_PATH_REC_FIELD(mtu_selector), 393 .offset_words = 15, 394 .offset_bits = 0, 395 .size_bits = 2 }, 396 { OPA_PATH_REC_FIELD(mtu), 397 .offset_words = 15, 398 .offset_bits = 2, 399 .size_bits = 6 }, 400 { OPA_PATH_REC_FIELD(rate_selector), 401 .offset_words = 15, 402 .offset_bits = 8, 403 .size_bits = 2 }, 404 { OPA_PATH_REC_FIELD(rate), 405 .offset_words = 15, 406 .offset_bits = 10, 407 .size_bits = 6 }, 408 { OPA_PATH_REC_FIELD(packet_life_time_selector), 409 .offset_words = 15, 410 .offset_bits = 16, 411 .size_bits = 2 }, 412 { OPA_PATH_REC_FIELD(packet_life_time), 413 .offset_words = 15, 414 .offset_bits = 18, 415 .size_bits = 6 }, 416 { OPA_PATH_REC_FIELD(preference), 417 .offset_words = 15, 418 .offset_bits = 24, 419 .size_bits = 8 }, 420 }; 421 422 #define MCMEMBER_REC_FIELD(field) \ 423 .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field), \ 424 .struct_size_bytes = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \ 425 .field_name = "sa_mcmember_rec:" #field 426 427 static const struct ib_field mcmember_rec_table[] = { 428 { MCMEMBER_REC_FIELD(mgid), 429 .offset_words = 0, 430 .offset_bits = 0, 431 .size_bits = 128 }, 432 { MCMEMBER_REC_FIELD(port_gid), 433 .offset_words = 4, 434 .offset_bits = 0, 435 .size_bits = 128 }, 436 { MCMEMBER_REC_FIELD(qkey), 437 .offset_words = 8, 438 .offset_bits = 0, 439 .size_bits = 32 }, 440 { MCMEMBER_REC_FIELD(mlid), 441 .offset_words = 9, 442 .offset_bits = 0, 443 .size_bits = 16 }, 444 { MCMEMBER_REC_FIELD(mtu_selector), 445 .offset_words = 9, 446 .offset_bits = 16, 447 .size_bits = 2 }, 448 { MCMEMBER_REC_FIELD(mtu), 449 .offset_words = 9, 450 .offset_bits = 18, 451 .size_bits = 6 }, 452 { MCMEMBER_REC_FIELD(traffic_class), 453 .offset_words = 9, 454 .offset_bits = 24, 455 .size_bits = 8 }, 456 { MCMEMBER_REC_FIELD(pkey), 457 .offset_words = 10, 458 .offset_bits = 0, 459 .size_bits = 16 }, 460 { MCMEMBER_REC_FIELD(rate_selector), 461 .offset_words = 10, 462 .offset_bits = 16, 463 .size_bits = 2 }, 464 { MCMEMBER_REC_FIELD(rate), 465 .offset_words = 10, 466 .offset_bits = 18, 467 .size_bits = 6 }, 468 { MCMEMBER_REC_FIELD(packet_life_time_selector), 469 .offset_words = 10, 470 .offset_bits = 24, 471 .size_bits = 2 }, 472 { MCMEMBER_REC_FIELD(packet_life_time), 473 .offset_words = 10, 474 .offset_bits = 26, 475 .size_bits = 6 }, 476 { MCMEMBER_REC_FIELD(sl), 477 .offset_words = 11, 478 .offset_bits = 0, 479 .size_bits = 4 }, 480 { MCMEMBER_REC_FIELD(flow_label), 481 .offset_words = 11, 482 .offset_bits = 4, 483 .size_bits = 20 }, 484 { MCMEMBER_REC_FIELD(hop_limit), 485 .offset_words = 11, 486 .offset_bits = 24, 487 .size_bits = 8 }, 488 { MCMEMBER_REC_FIELD(scope), 489 .offset_words = 12, 490 .offset_bits = 0, 491 .size_bits = 4 }, 492 { MCMEMBER_REC_FIELD(join_state), 493 .offset_words = 12, 494 .offset_bits = 4, 495 .size_bits = 4 }, 496 { MCMEMBER_REC_FIELD(proxy_join), 497 .offset_words = 12, 498 .offset_bits = 8, 499 .size_bits = 1 }, 500 { RESERVED, 501 .offset_words = 12, 502 .offset_bits = 9, 503 .size_bits = 23 }, 504 }; 505 506 #define SERVICE_REC_FIELD(field) \ 507 .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field), \ 508 .struct_size_bytes = sizeof ((struct ib_sa_service_rec *) 0)->field, \ 509 .field_name = "sa_service_rec:" #field 510 511 static const struct ib_field service_rec_table[] = { 512 { SERVICE_REC_FIELD(id), 513 .offset_words = 0, 514 .offset_bits = 0, 515 .size_bits = 64 }, 516 { SERVICE_REC_FIELD(gid), 517 .offset_words = 2, 518 .offset_bits = 0, 519 .size_bits = 128 }, 520 { SERVICE_REC_FIELD(pkey), 521 .offset_words = 6, 522 .offset_bits = 0, 523 .size_bits = 16 }, 524 { SERVICE_REC_FIELD(lease), 525 .offset_words = 7, 526 .offset_bits = 0, 527 .size_bits = 32 }, 528 { SERVICE_REC_FIELD(key), 529 .offset_words = 8, 530 .offset_bits = 0, 531 .size_bits = 128 }, 532 { SERVICE_REC_FIELD(name), 533 .offset_words = 12, 534 .offset_bits = 0, 535 .size_bits = 64*8 }, 536 { SERVICE_REC_FIELD(data8), 537 .offset_words = 28, 538 .offset_bits = 0, 539 .size_bits = 16*8 }, 540 { SERVICE_REC_FIELD(data16), 541 .offset_words = 32, 542 .offset_bits = 0, 543 .size_bits = 8*16 }, 544 { SERVICE_REC_FIELD(data32), 545 .offset_words = 36, 546 .offset_bits = 0, 547 .size_bits = 4*32 }, 548 { SERVICE_REC_FIELD(data64), 549 .offset_words = 40, 550 .offset_bits = 0, 551 .size_bits = 2*64 }, 552 }; 553 554 #define CLASSPORTINFO_REC_FIELD(field) \ 555 .struct_offset_bytes = offsetof(struct ib_class_port_info, field), \ 556 .struct_size_bytes = sizeof((struct ib_class_port_info *)0)->field, \ 557 .field_name = "ib_class_port_info:" #field 558 559 static const struct ib_field ib_classport_info_rec_table[] = { 560 { CLASSPORTINFO_REC_FIELD(base_version), 561 .offset_words = 0, 562 .offset_bits = 0, 563 .size_bits = 8 }, 564 { CLASSPORTINFO_REC_FIELD(class_version), 565 .offset_words = 0, 566 .offset_bits = 8, 567 .size_bits = 8 }, 568 { CLASSPORTINFO_REC_FIELD(capability_mask), 569 .offset_words = 0, 570 .offset_bits = 16, 571 .size_bits = 16 }, 572 { CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time), 573 .offset_words = 1, 574 .offset_bits = 0, 575 .size_bits = 32 }, 576 { CLASSPORTINFO_REC_FIELD(redirect_gid), 577 .offset_words = 2, 578 .offset_bits = 0, 579 .size_bits = 128 }, 580 { CLASSPORTINFO_REC_FIELD(redirect_tcslfl), 581 .offset_words = 6, 582 .offset_bits = 0, 583 .size_bits = 32 }, 584 { CLASSPORTINFO_REC_FIELD(redirect_lid), 585 .offset_words = 7, 586 .offset_bits = 0, 587 .size_bits = 16 }, 588 { CLASSPORTINFO_REC_FIELD(redirect_pkey), 589 .offset_words = 7, 590 .offset_bits = 16, 591 .size_bits = 16 }, 592 593 { CLASSPORTINFO_REC_FIELD(redirect_qp), 594 .offset_words = 8, 595 .offset_bits = 0, 596 .size_bits = 32 }, 597 { CLASSPORTINFO_REC_FIELD(redirect_qkey), 598 .offset_words = 9, 599 .offset_bits = 0, 600 .size_bits = 32 }, 601 602 { CLASSPORTINFO_REC_FIELD(trap_gid), 603 .offset_words = 10, 604 .offset_bits = 0, 605 .size_bits = 128 }, 606 { CLASSPORTINFO_REC_FIELD(trap_tcslfl), 607 .offset_words = 14, 608 .offset_bits = 0, 609 .size_bits = 32 }, 610 611 { CLASSPORTINFO_REC_FIELD(trap_lid), 612 .offset_words = 15, 613 .offset_bits = 0, 614 .size_bits = 16 }, 615 { CLASSPORTINFO_REC_FIELD(trap_pkey), 616 .offset_words = 15, 617 .offset_bits = 16, 618 .size_bits = 16 }, 619 620 { CLASSPORTINFO_REC_FIELD(trap_hlqp), 621 .offset_words = 16, 622 .offset_bits = 0, 623 .size_bits = 32 }, 624 { CLASSPORTINFO_REC_FIELD(trap_qkey), 625 .offset_words = 17, 626 .offset_bits = 0, 627 .size_bits = 32 }, 628 }; 629 630 #define OPA_CLASSPORTINFO_REC_FIELD(field) \ 631 .struct_offset_bytes =\ 632 offsetof(struct opa_class_port_info, field), \ 633 .struct_size_bytes = \ 634 sizeof((struct opa_class_port_info *)0)->field, \ 635 .field_name = "opa_class_port_info:" #field 636 637 static const struct ib_field opa_classport_info_rec_table[] = { 638 { OPA_CLASSPORTINFO_REC_FIELD(base_version), 639 .offset_words = 0, 640 .offset_bits = 0, 641 .size_bits = 8 }, 642 { OPA_CLASSPORTINFO_REC_FIELD(class_version), 643 .offset_words = 0, 644 .offset_bits = 8, 645 .size_bits = 8 }, 646 { OPA_CLASSPORTINFO_REC_FIELD(cap_mask), 647 .offset_words = 0, 648 .offset_bits = 16, 649 .size_bits = 16 }, 650 { OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time), 651 .offset_words = 1, 652 .offset_bits = 0, 653 .size_bits = 32 }, 654 { OPA_CLASSPORTINFO_REC_FIELD(redirect_gid), 655 .offset_words = 2, 656 .offset_bits = 0, 657 .size_bits = 128 }, 658 { OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl), 659 .offset_words = 6, 660 .offset_bits = 0, 661 .size_bits = 32 }, 662 { OPA_CLASSPORTINFO_REC_FIELD(redirect_lid), 663 .offset_words = 7, 664 .offset_bits = 0, 665 .size_bits = 32 }, 666 { OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp), 667 .offset_words = 8, 668 .offset_bits = 0, 669 .size_bits = 32 }, 670 { OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey), 671 .offset_words = 9, 672 .offset_bits = 0, 673 .size_bits = 32 }, 674 { OPA_CLASSPORTINFO_REC_FIELD(trap_gid), 675 .offset_words = 10, 676 .offset_bits = 0, 677 .size_bits = 128 }, 678 { OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl), 679 .offset_words = 14, 680 .offset_bits = 0, 681 .size_bits = 32 }, 682 { OPA_CLASSPORTINFO_REC_FIELD(trap_lid), 683 .offset_words = 15, 684 .offset_bits = 0, 685 .size_bits = 32 }, 686 { OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp), 687 .offset_words = 16, 688 .offset_bits = 0, 689 .size_bits = 32 }, 690 { OPA_CLASSPORTINFO_REC_FIELD(trap_qkey), 691 .offset_words = 17, 692 .offset_bits = 0, 693 .size_bits = 32 }, 694 { OPA_CLASSPORTINFO_REC_FIELD(trap_pkey), 695 .offset_words = 18, 696 .offset_bits = 0, 697 .size_bits = 16 }, 698 { OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey), 699 .offset_words = 18, 700 .offset_bits = 16, 701 .size_bits = 16 }, 702 { OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd), 703 .offset_words = 19, 704 .offset_bits = 0, 705 .size_bits = 8 }, 706 { RESERVED, 707 .offset_words = 19, 708 .offset_bits = 8, 709 .size_bits = 24 }, 710 }; 711 712 #define GUIDINFO_REC_FIELD(field) \ 713 .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field), \ 714 .struct_size_bytes = sizeof((struct ib_sa_guidinfo_rec *) 0)->field, \ 715 .field_name = "sa_guidinfo_rec:" #field 716 717 static const struct ib_field guidinfo_rec_table[] = { 718 { GUIDINFO_REC_FIELD(lid), 719 .offset_words = 0, 720 .offset_bits = 0, 721 .size_bits = 16 }, 722 { GUIDINFO_REC_FIELD(block_num), 723 .offset_words = 0, 724 .offset_bits = 16, 725 .size_bits = 8 }, 726 { GUIDINFO_REC_FIELD(res1), 727 .offset_words = 0, 728 .offset_bits = 24, 729 .size_bits = 8 }, 730 { GUIDINFO_REC_FIELD(res2), 731 .offset_words = 1, 732 .offset_bits = 0, 733 .size_bits = 32 }, 734 { GUIDINFO_REC_FIELD(guid_info_list), 735 .offset_words = 2, 736 .offset_bits = 0, 737 .size_bits = 512 }, 738 }; 739 740 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query) 741 { 742 query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE; 743 } 744 745 static inline int ib_sa_query_cancelled(struct ib_sa_query *query) 746 { 747 return (query->flags & IB_SA_CANCEL); 748 } 749 750 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb, 751 struct ib_sa_query *query) 752 { 753 struct sa_path_rec *sa_rec = query->mad_buf->context[1]; 754 struct ib_sa_mad *mad = query->mad_buf->mad; 755 ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask; 756 u16 val16; 757 u64 val64; 758 struct rdma_ls_resolve_header *header; 759 760 query->mad_buf->context[1] = NULL; 761 762 /* Construct the family header first */ 763 header = skb_put(skb, NLMSG_ALIGN(sizeof(*header))); 764 memcpy(header->device_name, query->port->agent->device->name, 765 LS_DEVICE_NAME_MAX); 766 header->port_num = query->port->port_num; 767 768 if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) && 769 sa_rec->reversible != 0) 770 query->path_use = LS_RESOLVE_PATH_USE_GMP; 771 else 772 query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL; 773 header->path_use = query->path_use; 774 775 /* Now build the attributes */ 776 if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) { 777 val64 = be64_to_cpu(sa_rec->service_id); 778 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID, 779 sizeof(val64), &val64); 780 } 781 if (comp_mask & IB_SA_PATH_REC_DGID) 782 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID, 783 sizeof(sa_rec->dgid), &sa_rec->dgid); 784 if (comp_mask & IB_SA_PATH_REC_SGID) 785 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID, 786 sizeof(sa_rec->sgid), &sa_rec->sgid); 787 if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS) 788 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS, 789 sizeof(sa_rec->traffic_class), &sa_rec->traffic_class); 790 791 if (comp_mask & IB_SA_PATH_REC_PKEY) { 792 val16 = be16_to_cpu(sa_rec->pkey); 793 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY, 794 sizeof(val16), &val16); 795 } 796 if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) { 797 val16 = be16_to_cpu(sa_rec->qos_class); 798 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS, 799 sizeof(val16), &val16); 800 } 801 } 802 803 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask) 804 { 805 int len = 0; 806 807 if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) 808 len += nla_total_size(sizeof(u64)); 809 if (comp_mask & IB_SA_PATH_REC_DGID) 810 len += nla_total_size(sizeof(struct rdma_nla_ls_gid)); 811 if (comp_mask & IB_SA_PATH_REC_SGID) 812 len += nla_total_size(sizeof(struct rdma_nla_ls_gid)); 813 if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS) 814 len += nla_total_size(sizeof(u8)); 815 if (comp_mask & IB_SA_PATH_REC_PKEY) 816 len += nla_total_size(sizeof(u16)); 817 if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) 818 len += nla_total_size(sizeof(u16)); 819 820 /* 821 * Make sure that at least some of the required comp_mask bits are 822 * set. 823 */ 824 if (WARN_ON(len == 0)) 825 return len; 826 827 /* Add the family header */ 828 len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header)); 829 830 return len; 831 } 832 833 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask) 834 { 835 struct sk_buff *skb = NULL; 836 struct nlmsghdr *nlh; 837 void *data; 838 int ret = 0; 839 struct ib_sa_mad *mad; 840 int len; 841 842 mad = query->mad_buf->mad; 843 len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask); 844 if (len <= 0) 845 return -EMSGSIZE; 846 847 skb = nlmsg_new(len, gfp_mask); 848 if (!skb) 849 return -ENOMEM; 850 851 /* Put nlmsg header only for now */ 852 data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS, 853 RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST); 854 if (!data) { 855 nlmsg_free(skb); 856 return -EMSGSIZE; 857 } 858 859 /* Add attributes */ 860 ib_nl_set_path_rec_attrs(skb, query); 861 862 /* Repair the nlmsg header length */ 863 nlmsg_end(skb, nlh); 864 865 ret = rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, gfp_mask); 866 if (!ret) 867 ret = len; 868 else 869 ret = 0; 870 871 return ret; 872 } 873 874 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask) 875 { 876 unsigned long flags; 877 unsigned long delay; 878 int ret; 879 880 INIT_LIST_HEAD(&query->list); 881 query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq); 882 883 /* Put the request on the list first.*/ 884 spin_lock_irqsave(&ib_nl_request_lock, flags); 885 delay = msecs_to_jiffies(sa_local_svc_timeout_ms); 886 query->timeout = delay + jiffies; 887 list_add_tail(&query->list, &ib_nl_request_list); 888 /* Start the timeout if this is the only request */ 889 if (ib_nl_request_list.next == &query->list) 890 queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay); 891 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 892 893 ret = ib_nl_send_msg(query, gfp_mask); 894 if (ret <= 0) { 895 ret = -EIO; 896 /* Remove the request */ 897 spin_lock_irqsave(&ib_nl_request_lock, flags); 898 list_del(&query->list); 899 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 900 } else { 901 ret = 0; 902 } 903 904 return ret; 905 } 906 907 static int ib_nl_cancel_request(struct ib_sa_query *query) 908 { 909 unsigned long flags; 910 struct ib_sa_query *wait_query; 911 int found = 0; 912 913 spin_lock_irqsave(&ib_nl_request_lock, flags); 914 list_for_each_entry(wait_query, &ib_nl_request_list, list) { 915 /* Let the timeout to take care of the callback */ 916 if (query == wait_query) { 917 query->flags |= IB_SA_CANCEL; 918 query->timeout = jiffies; 919 list_move(&query->list, &ib_nl_request_list); 920 found = 1; 921 mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1); 922 break; 923 } 924 } 925 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 926 927 return found; 928 } 929 930 static void send_handler(struct ib_mad_agent *agent, 931 struct ib_mad_send_wc *mad_send_wc); 932 933 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query, 934 const struct nlmsghdr *nlh) 935 { 936 struct ib_mad_send_wc mad_send_wc; 937 struct ib_sa_mad *mad = NULL; 938 const struct nlattr *head, *curr; 939 struct ib_path_rec_data *rec; 940 int len, rem; 941 u32 mask = 0; 942 int status = -EIO; 943 944 if (query->callback) { 945 head = (const struct nlattr *) nlmsg_data(nlh); 946 len = nlmsg_len(nlh); 947 switch (query->path_use) { 948 case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL: 949 mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND; 950 break; 951 952 case LS_RESOLVE_PATH_USE_ALL: 953 case LS_RESOLVE_PATH_USE_GMP: 954 default: 955 mask = IB_PATH_PRIMARY | IB_PATH_GMP | 956 IB_PATH_BIDIRECTIONAL; 957 break; 958 } 959 nla_for_each_attr(curr, head, len, rem) { 960 if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) { 961 rec = nla_data(curr); 962 /* 963 * Get the first one. In the future, we may 964 * need to get up to 6 pathrecords. 965 */ 966 if ((rec->flags & mask) == mask) { 967 mad = query->mad_buf->mad; 968 mad->mad_hdr.method |= 969 IB_MGMT_METHOD_RESP; 970 memcpy(mad->data, rec->path_rec, 971 sizeof(rec->path_rec)); 972 status = 0; 973 break; 974 } 975 } 976 } 977 query->callback(query, status, mad); 978 } 979 980 mad_send_wc.send_buf = query->mad_buf; 981 mad_send_wc.status = IB_WC_SUCCESS; 982 send_handler(query->mad_buf->mad_agent, &mad_send_wc); 983 } 984 985 static void ib_nl_request_timeout(struct work_struct *work) 986 { 987 unsigned long flags; 988 struct ib_sa_query *query; 989 unsigned long delay; 990 struct ib_mad_send_wc mad_send_wc; 991 int ret; 992 993 spin_lock_irqsave(&ib_nl_request_lock, flags); 994 while (!list_empty(&ib_nl_request_list)) { 995 query = list_entry(ib_nl_request_list.next, 996 struct ib_sa_query, list); 997 998 if (time_after(query->timeout, jiffies)) { 999 delay = query->timeout - jiffies; 1000 if ((long)delay <= 0) 1001 delay = 1; 1002 queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay); 1003 break; 1004 } 1005 1006 list_del(&query->list); 1007 ib_sa_disable_local_svc(query); 1008 /* Hold the lock to protect against query cancellation */ 1009 if (ib_sa_query_cancelled(query)) 1010 ret = -1; 1011 else 1012 ret = ib_post_send_mad(query->mad_buf, NULL); 1013 if (ret) { 1014 mad_send_wc.send_buf = query->mad_buf; 1015 mad_send_wc.status = IB_WC_WR_FLUSH_ERR; 1016 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1017 send_handler(query->port->agent, &mad_send_wc); 1018 spin_lock_irqsave(&ib_nl_request_lock, flags); 1019 } 1020 } 1021 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1022 } 1023 1024 int ib_nl_handle_set_timeout(struct sk_buff *skb, 1025 struct nlmsghdr *nlh, 1026 struct netlink_ext_ack *extack) 1027 { 1028 int timeout, delta, abs_delta; 1029 const struct nlattr *attr; 1030 unsigned long flags; 1031 struct ib_sa_query *query; 1032 long delay = 0; 1033 struct nlattr *tb[LS_NLA_TYPE_MAX]; 1034 int ret; 1035 1036 if (!(nlh->nlmsg_flags & NLM_F_REQUEST) || 1037 !(NETLINK_CB(skb).sk)) 1038 return -EPERM; 1039 1040 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh), 1041 nlmsg_len(nlh), ib_nl_policy, NULL); 1042 attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT]; 1043 if (ret || !attr) 1044 goto settimeout_out; 1045 1046 timeout = *(int *) nla_data(attr); 1047 if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN) 1048 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN; 1049 if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX) 1050 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX; 1051 1052 delta = timeout - sa_local_svc_timeout_ms; 1053 if (delta < 0) 1054 abs_delta = -delta; 1055 else 1056 abs_delta = delta; 1057 1058 if (delta != 0) { 1059 spin_lock_irqsave(&ib_nl_request_lock, flags); 1060 sa_local_svc_timeout_ms = timeout; 1061 list_for_each_entry(query, &ib_nl_request_list, list) { 1062 if (delta < 0 && abs_delta > query->timeout) 1063 query->timeout = 0; 1064 else 1065 query->timeout += delta; 1066 1067 /* Get the new delay from the first entry */ 1068 if (!delay) { 1069 delay = query->timeout - jiffies; 1070 if (delay <= 0) 1071 delay = 1; 1072 } 1073 } 1074 if (delay) 1075 mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1076 (unsigned long)delay); 1077 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1078 } 1079 1080 settimeout_out: 1081 return skb->len; 1082 } 1083 1084 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh) 1085 { 1086 struct nlattr *tb[LS_NLA_TYPE_MAX]; 1087 int ret; 1088 1089 if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR) 1090 return 0; 1091 1092 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh), 1093 nlmsg_len(nlh), ib_nl_policy, NULL); 1094 if (ret) 1095 return 0; 1096 1097 return 1; 1098 } 1099 1100 int ib_nl_handle_resolve_resp(struct sk_buff *skb, 1101 struct nlmsghdr *nlh, 1102 struct netlink_ext_ack *extack) 1103 { 1104 unsigned long flags; 1105 struct ib_sa_query *query; 1106 struct ib_mad_send_buf *send_buf; 1107 struct ib_mad_send_wc mad_send_wc; 1108 int found = 0; 1109 int ret; 1110 1111 if ((nlh->nlmsg_flags & NLM_F_REQUEST) || 1112 !(NETLINK_CB(skb).sk)) 1113 return -EPERM; 1114 1115 spin_lock_irqsave(&ib_nl_request_lock, flags); 1116 list_for_each_entry(query, &ib_nl_request_list, list) { 1117 /* 1118 * If the query is cancelled, let the timeout routine 1119 * take care of it. 1120 */ 1121 if (nlh->nlmsg_seq == query->seq) { 1122 found = !ib_sa_query_cancelled(query); 1123 if (found) 1124 list_del(&query->list); 1125 break; 1126 } 1127 } 1128 1129 if (!found) { 1130 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1131 goto resp_out; 1132 } 1133 1134 send_buf = query->mad_buf; 1135 1136 if (!ib_nl_is_good_resolve_resp(nlh)) { 1137 /* if the result is a failure, send out the packet via IB */ 1138 ib_sa_disable_local_svc(query); 1139 ret = ib_post_send_mad(query->mad_buf, NULL); 1140 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1141 if (ret) { 1142 mad_send_wc.send_buf = send_buf; 1143 mad_send_wc.status = IB_WC_GENERAL_ERR; 1144 send_handler(query->port->agent, &mad_send_wc); 1145 } 1146 } else { 1147 spin_unlock_irqrestore(&ib_nl_request_lock, flags); 1148 ib_nl_process_good_resolve_rsp(query, nlh); 1149 } 1150 1151 resp_out: 1152 return skb->len; 1153 } 1154 1155 static void free_sm_ah(struct kref *kref) 1156 { 1157 struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref); 1158 1159 rdma_destroy_ah(sm_ah->ah); 1160 kfree(sm_ah); 1161 } 1162 1163 void ib_sa_register_client(struct ib_sa_client *client) 1164 { 1165 atomic_set(&client->users, 1); 1166 init_completion(&client->comp); 1167 } 1168 EXPORT_SYMBOL(ib_sa_register_client); 1169 1170 void ib_sa_unregister_client(struct ib_sa_client *client) 1171 { 1172 ib_sa_client_put(client); 1173 wait_for_completion(&client->comp); 1174 } 1175 EXPORT_SYMBOL(ib_sa_unregister_client); 1176 1177 /** 1178 * ib_sa_cancel_query - try to cancel an SA query 1179 * @id:ID of query to cancel 1180 * @query:query pointer to cancel 1181 * 1182 * Try to cancel an SA query. If the id and query don't match up or 1183 * the query has already completed, nothing is done. Otherwise the 1184 * query is canceled and will complete with a status of -EINTR. 1185 */ 1186 void ib_sa_cancel_query(int id, struct ib_sa_query *query) 1187 { 1188 unsigned long flags; 1189 struct ib_mad_agent *agent; 1190 struct ib_mad_send_buf *mad_buf; 1191 1192 spin_lock_irqsave(&idr_lock, flags); 1193 if (idr_find(&query_idr, id) != query) { 1194 spin_unlock_irqrestore(&idr_lock, flags); 1195 return; 1196 } 1197 agent = query->port->agent; 1198 mad_buf = query->mad_buf; 1199 spin_unlock_irqrestore(&idr_lock, flags); 1200 1201 /* 1202 * If the query is still on the netlink request list, schedule 1203 * it to be cancelled by the timeout routine. Otherwise, it has been 1204 * sent to the MAD layer and has to be cancelled from there. 1205 */ 1206 if (!ib_nl_cancel_request(query)) 1207 ib_cancel_mad(agent, mad_buf); 1208 } 1209 EXPORT_SYMBOL(ib_sa_cancel_query); 1210 1211 static u8 get_src_path_mask(struct ib_device *device, u8 port_num) 1212 { 1213 struct ib_sa_device *sa_dev; 1214 struct ib_sa_port *port; 1215 unsigned long flags; 1216 u8 src_path_mask; 1217 1218 sa_dev = ib_get_client_data(device, &sa_client); 1219 if (!sa_dev) 1220 return 0x7f; 1221 1222 port = &sa_dev->port[port_num - sa_dev->start_port]; 1223 spin_lock_irqsave(&port->ah_lock, flags); 1224 src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f; 1225 spin_unlock_irqrestore(&port->ah_lock, flags); 1226 1227 return src_path_mask; 1228 } 1229 1230 int ib_init_ah_from_path(struct ib_device *device, u8 port_num, 1231 struct sa_path_rec *rec, 1232 struct rdma_ah_attr *ah_attr) 1233 { 1234 int ret; 1235 u16 gid_index; 1236 int use_roce; 1237 struct net_device *ndev = NULL; 1238 1239 memset(ah_attr, 0, sizeof *ah_attr); 1240 ah_attr->type = rdma_ah_find_type(device, port_num); 1241 1242 rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec))); 1243 1244 if ((ah_attr->type == RDMA_AH_ATTR_TYPE_OPA) && 1245 (rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))) 1246 rdma_ah_set_make_grd(ah_attr, true); 1247 1248 rdma_ah_set_sl(ah_attr, rec->sl); 1249 rdma_ah_set_path_bits(ah_attr, be32_to_cpu(sa_path_get_slid(rec)) & 1250 get_src_path_mask(device, port_num)); 1251 rdma_ah_set_port_num(ah_attr, port_num); 1252 rdma_ah_set_static_rate(ah_attr, rec->rate); 1253 use_roce = rdma_cap_eth_ah(device, port_num); 1254 1255 if (use_roce) { 1256 struct net_device *idev; 1257 struct net_device *resolved_dev; 1258 struct rdma_dev_addr dev_addr = { 1259 .bound_dev_if = ((sa_path_get_ifindex(rec) >= 0) ? 1260 sa_path_get_ifindex(rec) : 0), 1261 .net = sa_path_get_ndev(rec) ? 1262 sa_path_get_ndev(rec) : 1263 &init_net 1264 }; 1265 union { 1266 struct sockaddr _sockaddr; 1267 struct sockaddr_in _sockaddr_in; 1268 struct sockaddr_in6 _sockaddr_in6; 1269 } sgid_addr, dgid_addr; 1270 1271 if (!device->get_netdev) 1272 return -EOPNOTSUPP; 1273 1274 rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid); 1275 rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid); 1276 1277 /* validate the route */ 1278 ret = rdma_resolve_ip_route(&sgid_addr._sockaddr, 1279 &dgid_addr._sockaddr, &dev_addr); 1280 if (ret) 1281 return ret; 1282 1283 if ((dev_addr.network == RDMA_NETWORK_IPV4 || 1284 dev_addr.network == RDMA_NETWORK_IPV6) && 1285 rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2) 1286 return -EINVAL; 1287 1288 idev = device->get_netdev(device, port_num); 1289 if (!idev) 1290 return -ENODEV; 1291 1292 resolved_dev = dev_get_by_index(dev_addr.net, 1293 dev_addr.bound_dev_if); 1294 if (resolved_dev->flags & IFF_LOOPBACK) { 1295 dev_put(resolved_dev); 1296 resolved_dev = idev; 1297 dev_hold(resolved_dev); 1298 } 1299 ndev = ib_get_ndev_from_path(rec); 1300 rcu_read_lock(); 1301 if ((ndev && ndev != resolved_dev) || 1302 (resolved_dev != idev && 1303 !rdma_is_upper_dev_rcu(idev, resolved_dev))) 1304 ret = -EHOSTUNREACH; 1305 rcu_read_unlock(); 1306 dev_put(idev); 1307 dev_put(resolved_dev); 1308 if (ret) { 1309 if (ndev) 1310 dev_put(ndev); 1311 return ret; 1312 } 1313 } 1314 1315 if (rec->hop_limit > 0 || use_roce) { 1316 enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec); 1317 1318 ret = ib_find_cached_gid_by_port(device, &rec->sgid, type, 1319 port_num, ndev, &gid_index); 1320 if (ret) { 1321 if (ndev) 1322 dev_put(ndev); 1323 return ret; 1324 } 1325 1326 rdma_ah_set_grh(ah_attr, &rec->dgid, 1327 be32_to_cpu(rec->flow_label), 1328 gid_index, rec->hop_limit, 1329 rec->traffic_class); 1330 if (ndev) 1331 dev_put(ndev); 1332 } 1333 1334 if (use_roce) { 1335 u8 *dmac = sa_path_get_dmac(rec); 1336 1337 if (!dmac) 1338 return -EINVAL; 1339 memcpy(ah_attr->roce.dmac, dmac, ETH_ALEN); 1340 } 1341 1342 return 0; 1343 } 1344 EXPORT_SYMBOL(ib_init_ah_from_path); 1345 1346 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask) 1347 { 1348 unsigned long flags; 1349 1350 spin_lock_irqsave(&query->port->ah_lock, flags); 1351 if (!query->port->sm_ah) { 1352 spin_unlock_irqrestore(&query->port->ah_lock, flags); 1353 return -EAGAIN; 1354 } 1355 kref_get(&query->port->sm_ah->ref); 1356 query->sm_ah = query->port->sm_ah; 1357 spin_unlock_irqrestore(&query->port->ah_lock, flags); 1358 1359 query->mad_buf = ib_create_send_mad(query->port->agent, 1, 1360 query->sm_ah->pkey_index, 1361 0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA, 1362 gfp_mask, 1363 ((query->flags & IB_SA_QUERY_OPA) ? 1364 OPA_MGMT_BASE_VERSION : 1365 IB_MGMT_BASE_VERSION)); 1366 if (IS_ERR(query->mad_buf)) { 1367 kref_put(&query->sm_ah->ref, free_sm_ah); 1368 return -ENOMEM; 1369 } 1370 1371 query->mad_buf->ah = query->sm_ah->ah; 1372 1373 return 0; 1374 } 1375 1376 static void free_mad(struct ib_sa_query *query) 1377 { 1378 ib_free_send_mad(query->mad_buf); 1379 kref_put(&query->sm_ah->ref, free_sm_ah); 1380 } 1381 1382 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent) 1383 { 1384 struct ib_sa_mad *mad = query->mad_buf->mad; 1385 unsigned long flags; 1386 1387 memset(mad, 0, sizeof *mad); 1388 1389 if (query->flags & IB_SA_QUERY_OPA) { 1390 mad->mad_hdr.base_version = OPA_MGMT_BASE_VERSION; 1391 mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION; 1392 } else { 1393 mad->mad_hdr.base_version = IB_MGMT_BASE_VERSION; 1394 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION; 1395 } 1396 mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM; 1397 spin_lock_irqsave(&tid_lock, flags); 1398 mad->mad_hdr.tid = 1399 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++); 1400 spin_unlock_irqrestore(&tid_lock, flags); 1401 } 1402 1403 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask) 1404 { 1405 bool preload = gfpflags_allow_blocking(gfp_mask); 1406 unsigned long flags; 1407 int ret, id; 1408 1409 if (preload) 1410 idr_preload(gfp_mask); 1411 spin_lock_irqsave(&idr_lock, flags); 1412 1413 id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT); 1414 1415 spin_unlock_irqrestore(&idr_lock, flags); 1416 if (preload) 1417 idr_preload_end(); 1418 if (id < 0) 1419 return id; 1420 1421 query->mad_buf->timeout_ms = timeout_ms; 1422 query->mad_buf->context[0] = query; 1423 query->id = id; 1424 1425 if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) && 1426 (!(query->flags & IB_SA_QUERY_OPA))) { 1427 if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) { 1428 if (!ib_nl_make_request(query, gfp_mask)) 1429 return id; 1430 } 1431 ib_sa_disable_local_svc(query); 1432 } 1433 1434 ret = ib_post_send_mad(query->mad_buf, NULL); 1435 if (ret) { 1436 spin_lock_irqsave(&idr_lock, flags); 1437 idr_remove(&query_idr, id); 1438 spin_unlock_irqrestore(&idr_lock, flags); 1439 } 1440 1441 /* 1442 * It's not safe to dereference query any more, because the 1443 * send may already have completed and freed the query in 1444 * another context. 1445 */ 1446 return ret ? ret : id; 1447 } 1448 1449 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec) 1450 { 1451 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec); 1452 } 1453 EXPORT_SYMBOL(ib_sa_unpack_path); 1454 1455 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute) 1456 { 1457 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute); 1458 } 1459 EXPORT_SYMBOL(ib_sa_pack_path); 1460 1461 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client, 1462 struct ib_device *device, 1463 u8 port_num) 1464 { 1465 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1466 struct ib_sa_port *port; 1467 unsigned long flags; 1468 bool ret = false; 1469 1470 if (!sa_dev) 1471 return ret; 1472 1473 port = &sa_dev->port[port_num - sa_dev->start_port]; 1474 spin_lock_irqsave(&port->classport_lock, flags); 1475 if (!port->classport_info.valid) 1476 goto ret; 1477 1478 if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA) 1479 ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) & 1480 OPA_CLASS_PORT_INFO_PR_SUPPORT; 1481 ret: 1482 spin_unlock_irqrestore(&port->classport_lock, flags); 1483 return ret; 1484 } 1485 1486 enum opa_pr_supported { 1487 PR_NOT_SUPPORTED, 1488 PR_OPA_SUPPORTED, 1489 PR_IB_SUPPORTED 1490 }; 1491 1492 /** 1493 * Check if current PR query can be an OPA query. 1494 * Retuns PR_NOT_SUPPORTED if a path record query is not 1495 * possible, PR_OPA_SUPPORTED if an OPA path record query 1496 * is possible and PR_IB_SUPPORTED if an IB path record 1497 * query is possible. 1498 */ 1499 static int opa_pr_query_possible(struct ib_sa_client *client, 1500 struct ib_device *device, 1501 u8 port_num, 1502 struct sa_path_rec *rec) 1503 { 1504 struct ib_port_attr port_attr; 1505 1506 if (ib_query_port(device, port_num, &port_attr)) 1507 return PR_NOT_SUPPORTED; 1508 1509 if (ib_sa_opa_pathrecord_support(client, device, port_num)) 1510 return PR_OPA_SUPPORTED; 1511 1512 if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) 1513 return PR_NOT_SUPPORTED; 1514 else 1515 return PR_IB_SUPPORTED; 1516 } 1517 1518 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query, 1519 int status, 1520 struct ib_sa_mad *mad) 1521 { 1522 struct ib_sa_path_query *query = 1523 container_of(sa_query, struct ib_sa_path_query, sa_query); 1524 1525 if (mad) { 1526 struct sa_path_rec rec; 1527 1528 if (sa_query->flags & IB_SA_QUERY_OPA) { 1529 ib_unpack(opa_path_rec_table, 1530 ARRAY_SIZE(opa_path_rec_table), 1531 mad->data, &rec); 1532 rec.rec_type = SA_PATH_REC_TYPE_OPA; 1533 query->callback(status, &rec, query->context); 1534 } else { 1535 ib_unpack(path_rec_table, 1536 ARRAY_SIZE(path_rec_table), 1537 mad->data, &rec); 1538 rec.rec_type = SA_PATH_REC_TYPE_IB; 1539 sa_path_set_ndev(&rec, NULL); 1540 sa_path_set_ifindex(&rec, 0); 1541 sa_path_set_dmac_zero(&rec); 1542 1543 if (query->conv_pr) { 1544 struct sa_path_rec opa; 1545 1546 memset(&opa, 0, sizeof(struct sa_path_rec)); 1547 sa_convert_path_ib_to_opa(&opa, &rec); 1548 query->callback(status, &opa, query->context); 1549 } else { 1550 query->callback(status, &rec, query->context); 1551 } 1552 } 1553 } else 1554 query->callback(status, NULL, query->context); 1555 } 1556 1557 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query) 1558 { 1559 struct ib_sa_path_query *query = 1560 container_of(sa_query, struct ib_sa_path_query, sa_query); 1561 1562 kfree(query->conv_pr); 1563 kfree(query); 1564 } 1565 1566 /** 1567 * ib_sa_path_rec_get - Start a Path get query 1568 * @client:SA client 1569 * @device:device to send query on 1570 * @port_num: port number to send query on 1571 * @rec:Path Record to send in query 1572 * @comp_mask:component mask to send in query 1573 * @timeout_ms:time to wait for response 1574 * @gfp_mask:GFP mask to use for internal allocations 1575 * @callback:function called when query completes, times out or is 1576 * canceled 1577 * @context:opaque user context passed to callback 1578 * @sa_query:query context, used to cancel query 1579 * 1580 * Send a Path Record Get query to the SA to look up a path. The 1581 * callback function will be called when the query completes (or 1582 * fails); status is 0 for a successful response, -EINTR if the query 1583 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 1584 * occurred sending the query. The resp parameter of the callback is 1585 * only valid if status is 0. 1586 * 1587 * If the return value of ib_sa_path_rec_get() is negative, it is an 1588 * error code. Otherwise it is a query ID that can be used to cancel 1589 * the query. 1590 */ 1591 int ib_sa_path_rec_get(struct ib_sa_client *client, 1592 struct ib_device *device, u8 port_num, 1593 struct sa_path_rec *rec, 1594 ib_sa_comp_mask comp_mask, 1595 int timeout_ms, gfp_t gfp_mask, 1596 void (*callback)(int status, 1597 struct sa_path_rec *resp, 1598 void *context), 1599 void *context, 1600 struct ib_sa_query **sa_query) 1601 { 1602 struct ib_sa_path_query *query; 1603 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1604 struct ib_sa_port *port; 1605 struct ib_mad_agent *agent; 1606 struct ib_sa_mad *mad; 1607 enum opa_pr_supported status; 1608 int ret; 1609 1610 if (!sa_dev) 1611 return -ENODEV; 1612 1613 if ((rec->rec_type != SA_PATH_REC_TYPE_IB) && 1614 (rec->rec_type != SA_PATH_REC_TYPE_OPA)) 1615 return -EINVAL; 1616 1617 port = &sa_dev->port[port_num - sa_dev->start_port]; 1618 agent = port->agent; 1619 1620 query = kzalloc(sizeof(*query), gfp_mask); 1621 if (!query) 1622 return -ENOMEM; 1623 1624 query->sa_query.port = port; 1625 if (rec->rec_type == SA_PATH_REC_TYPE_OPA) { 1626 status = opa_pr_query_possible(client, device, port_num, rec); 1627 if (status == PR_NOT_SUPPORTED) { 1628 ret = -EINVAL; 1629 goto err1; 1630 } else if (status == PR_OPA_SUPPORTED) { 1631 query->sa_query.flags |= IB_SA_QUERY_OPA; 1632 } else { 1633 query->conv_pr = 1634 kmalloc(sizeof(*query->conv_pr), gfp_mask); 1635 if (!query->conv_pr) { 1636 ret = -ENOMEM; 1637 goto err1; 1638 } 1639 } 1640 } 1641 1642 ret = alloc_mad(&query->sa_query, gfp_mask); 1643 if (ret) 1644 goto err2; 1645 1646 ib_sa_client_get(client); 1647 query->sa_query.client = client; 1648 query->callback = callback; 1649 query->context = context; 1650 1651 mad = query->sa_query.mad_buf->mad; 1652 init_mad(&query->sa_query, agent); 1653 1654 query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL; 1655 query->sa_query.release = ib_sa_path_rec_release; 1656 mad->mad_hdr.method = IB_MGMT_METHOD_GET; 1657 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_PATH_REC); 1658 mad->sa_hdr.comp_mask = comp_mask; 1659 1660 if (query->sa_query.flags & IB_SA_QUERY_OPA) { 1661 ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table), 1662 rec, mad->data); 1663 } else if (query->conv_pr) { 1664 sa_convert_path_opa_to_ib(query->conv_pr, rec); 1665 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), 1666 query->conv_pr, mad->data); 1667 } else { 1668 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), 1669 rec, mad->data); 1670 } 1671 1672 *sa_query = &query->sa_query; 1673 1674 query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE; 1675 query->sa_query.mad_buf->context[1] = (query->conv_pr) ? 1676 query->conv_pr : rec; 1677 1678 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 1679 if (ret < 0) 1680 goto err3; 1681 1682 return ret; 1683 1684 err3: 1685 *sa_query = NULL; 1686 ib_sa_client_put(query->sa_query.client); 1687 free_mad(&query->sa_query); 1688 err2: 1689 kfree(query->conv_pr); 1690 err1: 1691 kfree(query); 1692 return ret; 1693 } 1694 EXPORT_SYMBOL(ib_sa_path_rec_get); 1695 1696 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query, 1697 int status, 1698 struct ib_sa_mad *mad) 1699 { 1700 struct ib_sa_service_query *query = 1701 container_of(sa_query, struct ib_sa_service_query, sa_query); 1702 1703 if (mad) { 1704 struct ib_sa_service_rec rec; 1705 1706 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table), 1707 mad->data, &rec); 1708 query->callback(status, &rec, query->context); 1709 } else 1710 query->callback(status, NULL, query->context); 1711 } 1712 1713 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query) 1714 { 1715 kfree(container_of(sa_query, struct ib_sa_service_query, sa_query)); 1716 } 1717 1718 /** 1719 * ib_sa_service_rec_query - Start Service Record operation 1720 * @client:SA client 1721 * @device:device to send request on 1722 * @port_num: port number to send request on 1723 * @method:SA method - should be get, set, or delete 1724 * @rec:Service Record to send in request 1725 * @comp_mask:component mask to send in request 1726 * @timeout_ms:time to wait for response 1727 * @gfp_mask:GFP mask to use for internal allocations 1728 * @callback:function called when request completes, times out or is 1729 * canceled 1730 * @context:opaque user context passed to callback 1731 * @sa_query:request context, used to cancel request 1732 * 1733 * Send a Service Record set/get/delete to the SA to register, 1734 * unregister or query a service record. 1735 * The callback function will be called when the request completes (or 1736 * fails); status is 0 for a successful response, -EINTR if the query 1737 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 1738 * occurred sending the query. The resp parameter of the callback is 1739 * only valid if status is 0. 1740 * 1741 * If the return value of ib_sa_service_rec_query() is negative, it is an 1742 * error code. Otherwise it is a request ID that can be used to cancel 1743 * the query. 1744 */ 1745 int ib_sa_service_rec_query(struct ib_sa_client *client, 1746 struct ib_device *device, u8 port_num, u8 method, 1747 struct ib_sa_service_rec *rec, 1748 ib_sa_comp_mask comp_mask, 1749 int timeout_ms, gfp_t gfp_mask, 1750 void (*callback)(int status, 1751 struct ib_sa_service_rec *resp, 1752 void *context), 1753 void *context, 1754 struct ib_sa_query **sa_query) 1755 { 1756 struct ib_sa_service_query *query; 1757 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1758 struct ib_sa_port *port; 1759 struct ib_mad_agent *agent; 1760 struct ib_sa_mad *mad; 1761 int ret; 1762 1763 if (!sa_dev) 1764 return -ENODEV; 1765 1766 port = &sa_dev->port[port_num - sa_dev->start_port]; 1767 agent = port->agent; 1768 1769 if (method != IB_MGMT_METHOD_GET && 1770 method != IB_MGMT_METHOD_SET && 1771 method != IB_SA_METHOD_DELETE) 1772 return -EINVAL; 1773 1774 query = kzalloc(sizeof(*query), gfp_mask); 1775 if (!query) 1776 return -ENOMEM; 1777 1778 query->sa_query.port = port; 1779 ret = alloc_mad(&query->sa_query, gfp_mask); 1780 if (ret) 1781 goto err1; 1782 1783 ib_sa_client_get(client); 1784 query->sa_query.client = client; 1785 query->callback = callback; 1786 query->context = context; 1787 1788 mad = query->sa_query.mad_buf->mad; 1789 init_mad(&query->sa_query, agent); 1790 1791 query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL; 1792 query->sa_query.release = ib_sa_service_rec_release; 1793 mad->mad_hdr.method = method; 1794 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_SERVICE_REC); 1795 mad->sa_hdr.comp_mask = comp_mask; 1796 1797 ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table), 1798 rec, mad->data); 1799 1800 *sa_query = &query->sa_query; 1801 1802 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 1803 if (ret < 0) 1804 goto err2; 1805 1806 return ret; 1807 1808 err2: 1809 *sa_query = NULL; 1810 ib_sa_client_put(query->sa_query.client); 1811 free_mad(&query->sa_query); 1812 1813 err1: 1814 kfree(query); 1815 return ret; 1816 } 1817 EXPORT_SYMBOL(ib_sa_service_rec_query); 1818 1819 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query, 1820 int status, 1821 struct ib_sa_mad *mad) 1822 { 1823 struct ib_sa_mcmember_query *query = 1824 container_of(sa_query, struct ib_sa_mcmember_query, sa_query); 1825 1826 if (mad) { 1827 struct ib_sa_mcmember_rec rec; 1828 1829 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 1830 mad->data, &rec); 1831 query->callback(status, &rec, query->context); 1832 } else 1833 query->callback(status, NULL, query->context); 1834 } 1835 1836 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query) 1837 { 1838 kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query)); 1839 } 1840 1841 int ib_sa_mcmember_rec_query(struct ib_sa_client *client, 1842 struct ib_device *device, u8 port_num, 1843 u8 method, 1844 struct ib_sa_mcmember_rec *rec, 1845 ib_sa_comp_mask comp_mask, 1846 int timeout_ms, gfp_t gfp_mask, 1847 void (*callback)(int status, 1848 struct ib_sa_mcmember_rec *resp, 1849 void *context), 1850 void *context, 1851 struct ib_sa_query **sa_query) 1852 { 1853 struct ib_sa_mcmember_query *query; 1854 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1855 struct ib_sa_port *port; 1856 struct ib_mad_agent *agent; 1857 struct ib_sa_mad *mad; 1858 int ret; 1859 1860 if (!sa_dev) 1861 return -ENODEV; 1862 1863 port = &sa_dev->port[port_num - sa_dev->start_port]; 1864 agent = port->agent; 1865 1866 query = kzalloc(sizeof(*query), gfp_mask); 1867 if (!query) 1868 return -ENOMEM; 1869 1870 query->sa_query.port = port; 1871 ret = alloc_mad(&query->sa_query, gfp_mask); 1872 if (ret) 1873 goto err1; 1874 1875 ib_sa_client_get(client); 1876 query->sa_query.client = client; 1877 query->callback = callback; 1878 query->context = context; 1879 1880 mad = query->sa_query.mad_buf->mad; 1881 init_mad(&query->sa_query, agent); 1882 1883 query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL; 1884 query->sa_query.release = ib_sa_mcmember_rec_release; 1885 mad->mad_hdr.method = method; 1886 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC); 1887 mad->sa_hdr.comp_mask = comp_mask; 1888 1889 ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 1890 rec, mad->data); 1891 1892 *sa_query = &query->sa_query; 1893 1894 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 1895 if (ret < 0) 1896 goto err2; 1897 1898 return ret; 1899 1900 err2: 1901 *sa_query = NULL; 1902 ib_sa_client_put(query->sa_query.client); 1903 free_mad(&query->sa_query); 1904 1905 err1: 1906 kfree(query); 1907 return ret; 1908 } 1909 1910 /* Support GuidInfoRecord */ 1911 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query, 1912 int status, 1913 struct ib_sa_mad *mad) 1914 { 1915 struct ib_sa_guidinfo_query *query = 1916 container_of(sa_query, struct ib_sa_guidinfo_query, sa_query); 1917 1918 if (mad) { 1919 struct ib_sa_guidinfo_rec rec; 1920 1921 ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), 1922 mad->data, &rec); 1923 query->callback(status, &rec, query->context); 1924 } else 1925 query->callback(status, NULL, query->context); 1926 } 1927 1928 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query) 1929 { 1930 kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query)); 1931 } 1932 1933 int ib_sa_guid_info_rec_query(struct ib_sa_client *client, 1934 struct ib_device *device, u8 port_num, 1935 struct ib_sa_guidinfo_rec *rec, 1936 ib_sa_comp_mask comp_mask, u8 method, 1937 int timeout_ms, gfp_t gfp_mask, 1938 void (*callback)(int status, 1939 struct ib_sa_guidinfo_rec *resp, 1940 void *context), 1941 void *context, 1942 struct ib_sa_query **sa_query) 1943 { 1944 struct ib_sa_guidinfo_query *query; 1945 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1946 struct ib_sa_port *port; 1947 struct ib_mad_agent *agent; 1948 struct ib_sa_mad *mad; 1949 int ret; 1950 1951 if (!sa_dev) 1952 return -ENODEV; 1953 1954 if (method != IB_MGMT_METHOD_GET && 1955 method != IB_MGMT_METHOD_SET && 1956 method != IB_SA_METHOD_DELETE) { 1957 return -EINVAL; 1958 } 1959 1960 port = &sa_dev->port[port_num - sa_dev->start_port]; 1961 agent = port->agent; 1962 1963 query = kzalloc(sizeof(*query), gfp_mask); 1964 if (!query) 1965 return -ENOMEM; 1966 1967 query->sa_query.port = port; 1968 ret = alloc_mad(&query->sa_query, gfp_mask); 1969 if (ret) 1970 goto err1; 1971 1972 ib_sa_client_get(client); 1973 query->sa_query.client = client; 1974 query->callback = callback; 1975 query->context = context; 1976 1977 mad = query->sa_query.mad_buf->mad; 1978 init_mad(&query->sa_query, agent); 1979 1980 query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL; 1981 query->sa_query.release = ib_sa_guidinfo_rec_release; 1982 1983 mad->mad_hdr.method = method; 1984 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC); 1985 mad->sa_hdr.comp_mask = comp_mask; 1986 1987 ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec, 1988 mad->data); 1989 1990 *sa_query = &query->sa_query; 1991 1992 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 1993 if (ret < 0) 1994 goto err2; 1995 1996 return ret; 1997 1998 err2: 1999 *sa_query = NULL; 2000 ib_sa_client_put(query->sa_query.client); 2001 free_mad(&query->sa_query); 2002 2003 err1: 2004 kfree(query); 2005 return ret; 2006 } 2007 EXPORT_SYMBOL(ib_sa_guid_info_rec_query); 2008 2009 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client, 2010 struct ib_device *device, 2011 u8 port_num) 2012 { 2013 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 2014 struct ib_sa_port *port; 2015 bool ret = false; 2016 unsigned long flags; 2017 2018 if (!sa_dev) 2019 return ret; 2020 2021 port = &sa_dev->port[port_num - sa_dev->start_port]; 2022 2023 spin_lock_irqsave(&port->classport_lock, flags); 2024 if ((port->classport_info.valid) && 2025 (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB)) 2026 ret = ib_get_cpi_capmask2(&port->classport_info.data.ib) 2027 & IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT; 2028 spin_unlock_irqrestore(&port->classport_lock, flags); 2029 return ret; 2030 } 2031 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support); 2032 2033 struct ib_classport_info_context { 2034 struct completion done; 2035 struct ib_sa_query *sa_query; 2036 }; 2037 2038 static void ib_classportinfo_cb(void *context) 2039 { 2040 struct ib_classport_info_context *cb_ctx = context; 2041 2042 complete(&cb_ctx->done); 2043 } 2044 2045 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query, 2046 int status, 2047 struct ib_sa_mad *mad) 2048 { 2049 unsigned long flags; 2050 struct ib_sa_classport_info_query *query = 2051 container_of(sa_query, struct ib_sa_classport_info_query, sa_query); 2052 struct ib_sa_classport_cache *info = &sa_query->port->classport_info; 2053 2054 if (mad) { 2055 if (sa_query->flags & IB_SA_QUERY_OPA) { 2056 struct opa_class_port_info rec; 2057 2058 ib_unpack(opa_classport_info_rec_table, 2059 ARRAY_SIZE(opa_classport_info_rec_table), 2060 mad->data, &rec); 2061 2062 spin_lock_irqsave(&sa_query->port->classport_lock, 2063 flags); 2064 if (!status && !info->valid) { 2065 memcpy(&info->data.opa, &rec, 2066 sizeof(info->data.opa)); 2067 2068 info->valid = true; 2069 info->data.type = RDMA_CLASS_PORT_INFO_OPA; 2070 } 2071 spin_unlock_irqrestore(&sa_query->port->classport_lock, 2072 flags); 2073 2074 } else { 2075 struct ib_class_port_info rec; 2076 2077 ib_unpack(ib_classport_info_rec_table, 2078 ARRAY_SIZE(ib_classport_info_rec_table), 2079 mad->data, &rec); 2080 2081 spin_lock_irqsave(&sa_query->port->classport_lock, 2082 flags); 2083 if (!status && !info->valid) { 2084 memcpy(&info->data.ib, &rec, 2085 sizeof(info->data.ib)); 2086 2087 info->valid = true; 2088 info->data.type = RDMA_CLASS_PORT_INFO_IB; 2089 } 2090 spin_unlock_irqrestore(&sa_query->port->classport_lock, 2091 flags); 2092 } 2093 } 2094 query->callback(query->context); 2095 } 2096 2097 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query) 2098 { 2099 kfree(container_of(sa_query, struct ib_sa_classport_info_query, 2100 sa_query)); 2101 } 2102 2103 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port, 2104 int timeout_ms, 2105 void (*callback)(void *context), 2106 void *context, 2107 struct ib_sa_query **sa_query) 2108 { 2109 struct ib_mad_agent *agent; 2110 struct ib_sa_classport_info_query *query; 2111 struct ib_sa_mad *mad; 2112 gfp_t gfp_mask = GFP_KERNEL; 2113 int ret; 2114 2115 agent = port->agent; 2116 2117 query = kzalloc(sizeof(*query), gfp_mask); 2118 if (!query) 2119 return -ENOMEM; 2120 2121 query->sa_query.port = port; 2122 query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device, 2123 port->port_num) ? 2124 IB_SA_QUERY_OPA : 0; 2125 ret = alloc_mad(&query->sa_query, gfp_mask); 2126 if (ret) 2127 goto err_free; 2128 2129 query->callback = callback; 2130 query->context = context; 2131 2132 mad = query->sa_query.mad_buf->mad; 2133 init_mad(&query->sa_query, agent); 2134 2135 query->sa_query.callback = ib_sa_classport_info_rec_callback; 2136 query->sa_query.release = ib_sa_classport_info_rec_release; 2137 mad->mad_hdr.method = IB_MGMT_METHOD_GET; 2138 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO); 2139 mad->sa_hdr.comp_mask = 0; 2140 *sa_query = &query->sa_query; 2141 2142 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 2143 if (ret < 0) 2144 goto err_free_mad; 2145 2146 return ret; 2147 2148 err_free_mad: 2149 *sa_query = NULL; 2150 free_mad(&query->sa_query); 2151 2152 err_free: 2153 kfree(query); 2154 return ret; 2155 } 2156 2157 static void update_ib_cpi(struct work_struct *work) 2158 { 2159 struct ib_sa_port *port = 2160 container_of(work, struct ib_sa_port, ib_cpi_work.work); 2161 struct ib_classport_info_context *cb_context; 2162 unsigned long flags; 2163 int ret; 2164 2165 /* If the classport info is valid, nothing 2166 * to do here. 2167 */ 2168 spin_lock_irqsave(&port->classport_lock, flags); 2169 if (port->classport_info.valid) { 2170 spin_unlock_irqrestore(&port->classport_lock, flags); 2171 return; 2172 } 2173 spin_unlock_irqrestore(&port->classport_lock, flags); 2174 2175 cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL); 2176 if (!cb_context) 2177 goto err_nomem; 2178 2179 init_completion(&cb_context->done); 2180 2181 ret = ib_sa_classport_info_rec_query(port, 3000, 2182 ib_classportinfo_cb, cb_context, 2183 &cb_context->sa_query); 2184 if (ret < 0) 2185 goto free_cb_err; 2186 wait_for_completion(&cb_context->done); 2187 free_cb_err: 2188 kfree(cb_context); 2189 spin_lock_irqsave(&port->classport_lock, flags); 2190 2191 /* If the classport info is still not valid, the query should have 2192 * failed for some reason. Retry issuing the query 2193 */ 2194 if (!port->classport_info.valid) { 2195 port->classport_info.retry_cnt++; 2196 if (port->classport_info.retry_cnt <= 2197 IB_SA_CPI_MAX_RETRY_CNT) { 2198 unsigned long delay = 2199 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT); 2200 2201 queue_delayed_work(ib_wq, &port->ib_cpi_work, delay); 2202 } 2203 } 2204 spin_unlock_irqrestore(&port->classport_lock, flags); 2205 2206 err_nomem: 2207 return; 2208 } 2209 2210 static void send_handler(struct ib_mad_agent *agent, 2211 struct ib_mad_send_wc *mad_send_wc) 2212 { 2213 struct ib_sa_query *query = mad_send_wc->send_buf->context[0]; 2214 unsigned long flags; 2215 2216 if (query->callback) 2217 switch (mad_send_wc->status) { 2218 case IB_WC_SUCCESS: 2219 /* No callback -- already got recv */ 2220 break; 2221 case IB_WC_RESP_TIMEOUT_ERR: 2222 query->callback(query, -ETIMEDOUT, NULL); 2223 break; 2224 case IB_WC_WR_FLUSH_ERR: 2225 query->callback(query, -EINTR, NULL); 2226 break; 2227 default: 2228 query->callback(query, -EIO, NULL); 2229 break; 2230 } 2231 2232 spin_lock_irqsave(&idr_lock, flags); 2233 idr_remove(&query_idr, query->id); 2234 spin_unlock_irqrestore(&idr_lock, flags); 2235 2236 free_mad(query); 2237 if (query->client) 2238 ib_sa_client_put(query->client); 2239 query->release(query); 2240 } 2241 2242 static void recv_handler(struct ib_mad_agent *mad_agent, 2243 struct ib_mad_send_buf *send_buf, 2244 struct ib_mad_recv_wc *mad_recv_wc) 2245 { 2246 struct ib_sa_query *query; 2247 2248 if (!send_buf) 2249 return; 2250 2251 query = send_buf->context[0]; 2252 if (query->callback) { 2253 if (mad_recv_wc->wc->status == IB_WC_SUCCESS) 2254 query->callback(query, 2255 mad_recv_wc->recv_buf.mad->mad_hdr.status ? 2256 -EINVAL : 0, 2257 (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad); 2258 else 2259 query->callback(query, -EIO, NULL); 2260 } 2261 2262 ib_free_recv_mad(mad_recv_wc); 2263 } 2264 2265 static void update_sm_ah(struct work_struct *work) 2266 { 2267 struct ib_sa_port *port = 2268 container_of(work, struct ib_sa_port, update_task); 2269 struct ib_sa_sm_ah *new_ah; 2270 struct ib_port_attr port_attr; 2271 struct rdma_ah_attr ah_attr; 2272 2273 if (ib_query_port(port->agent->device, port->port_num, &port_attr)) { 2274 pr_warn("Couldn't query port\n"); 2275 return; 2276 } 2277 2278 new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL); 2279 if (!new_ah) 2280 return; 2281 2282 kref_init(&new_ah->ref); 2283 new_ah->src_path_mask = (1 << port_attr.lmc) - 1; 2284 2285 new_ah->pkey_index = 0; 2286 if (ib_find_pkey(port->agent->device, port->port_num, 2287 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index)) 2288 pr_err("Couldn't find index for default PKey\n"); 2289 2290 memset(&ah_attr, 0, sizeof(ah_attr)); 2291 ah_attr.type = rdma_ah_find_type(port->agent->device, 2292 port->port_num); 2293 rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid); 2294 rdma_ah_set_sl(&ah_attr, port_attr.sm_sl); 2295 rdma_ah_set_port_num(&ah_attr, port->port_num); 2296 if (port_attr.grh_required) { 2297 if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA) { 2298 rdma_ah_set_make_grd(&ah_attr, true); 2299 } else { 2300 rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH); 2301 rdma_ah_set_subnet_prefix(&ah_attr, 2302 cpu_to_be64(port_attr.subnet_prefix)); 2303 rdma_ah_set_interface_id(&ah_attr, 2304 cpu_to_be64(IB_SA_WELL_KNOWN_GUID)); 2305 } 2306 } 2307 2308 new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr); 2309 if (IS_ERR(new_ah->ah)) { 2310 pr_warn("Couldn't create new SM AH\n"); 2311 kfree(new_ah); 2312 return; 2313 } 2314 2315 spin_lock_irq(&port->ah_lock); 2316 if (port->sm_ah) 2317 kref_put(&port->sm_ah->ref, free_sm_ah); 2318 port->sm_ah = new_ah; 2319 spin_unlock_irq(&port->ah_lock); 2320 } 2321 2322 static void ib_sa_event(struct ib_event_handler *handler, 2323 struct ib_event *event) 2324 { 2325 if (event->event == IB_EVENT_PORT_ERR || 2326 event->event == IB_EVENT_PORT_ACTIVE || 2327 event->event == IB_EVENT_LID_CHANGE || 2328 event->event == IB_EVENT_PKEY_CHANGE || 2329 event->event == IB_EVENT_SM_CHANGE || 2330 event->event == IB_EVENT_CLIENT_REREGISTER) { 2331 unsigned long flags; 2332 struct ib_sa_device *sa_dev = 2333 container_of(handler, typeof(*sa_dev), event_handler); 2334 u8 port_num = event->element.port_num - sa_dev->start_port; 2335 struct ib_sa_port *port = &sa_dev->port[port_num]; 2336 2337 if (!rdma_cap_ib_sa(handler->device, port->port_num)) 2338 return; 2339 2340 spin_lock_irqsave(&port->ah_lock, flags); 2341 if (port->sm_ah) 2342 kref_put(&port->sm_ah->ref, free_sm_ah); 2343 port->sm_ah = NULL; 2344 spin_unlock_irqrestore(&port->ah_lock, flags); 2345 2346 if (event->event == IB_EVENT_SM_CHANGE || 2347 event->event == IB_EVENT_CLIENT_REREGISTER || 2348 event->event == IB_EVENT_LID_CHANGE || 2349 event->event == IB_EVENT_PORT_ACTIVE) { 2350 unsigned long delay = 2351 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT); 2352 2353 spin_lock_irqsave(&port->classport_lock, flags); 2354 port->classport_info.valid = false; 2355 port->classport_info.retry_cnt = 0; 2356 spin_unlock_irqrestore(&port->classport_lock, flags); 2357 queue_delayed_work(ib_wq, 2358 &port->ib_cpi_work, delay); 2359 } 2360 queue_work(ib_wq, &sa_dev->port[port_num].update_task); 2361 } 2362 } 2363 2364 static void ib_sa_add_one(struct ib_device *device) 2365 { 2366 struct ib_sa_device *sa_dev; 2367 int s, e, i; 2368 int count = 0; 2369 2370 s = rdma_start_port(device); 2371 e = rdma_end_port(device); 2372 2373 sa_dev = kzalloc(sizeof *sa_dev + 2374 (e - s + 1) * sizeof (struct ib_sa_port), 2375 GFP_KERNEL); 2376 if (!sa_dev) 2377 return; 2378 2379 sa_dev->start_port = s; 2380 sa_dev->end_port = e; 2381 2382 for (i = 0; i <= e - s; ++i) { 2383 spin_lock_init(&sa_dev->port[i].ah_lock); 2384 if (!rdma_cap_ib_sa(device, i + 1)) 2385 continue; 2386 2387 sa_dev->port[i].sm_ah = NULL; 2388 sa_dev->port[i].port_num = i + s; 2389 2390 spin_lock_init(&sa_dev->port[i].classport_lock); 2391 sa_dev->port[i].classport_info.valid = false; 2392 2393 sa_dev->port[i].agent = 2394 ib_register_mad_agent(device, i + s, IB_QPT_GSI, 2395 NULL, 0, send_handler, 2396 recv_handler, sa_dev, 0); 2397 if (IS_ERR(sa_dev->port[i].agent)) 2398 goto err; 2399 2400 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah); 2401 INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work, 2402 update_ib_cpi); 2403 2404 count++; 2405 } 2406 2407 if (!count) 2408 goto free; 2409 2410 ib_set_client_data(device, &sa_client, sa_dev); 2411 2412 /* 2413 * We register our event handler after everything is set up, 2414 * and then update our cached info after the event handler is 2415 * registered to avoid any problems if a port changes state 2416 * during our initialization. 2417 */ 2418 2419 INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event); 2420 ib_register_event_handler(&sa_dev->event_handler); 2421 2422 for (i = 0; i <= e - s; ++i) { 2423 if (rdma_cap_ib_sa(device, i + 1)) 2424 update_sm_ah(&sa_dev->port[i].update_task); 2425 } 2426 2427 return; 2428 2429 err: 2430 while (--i >= 0) { 2431 if (rdma_cap_ib_sa(device, i + 1)) 2432 ib_unregister_mad_agent(sa_dev->port[i].agent); 2433 } 2434 free: 2435 kfree(sa_dev); 2436 return; 2437 } 2438 2439 static void ib_sa_remove_one(struct ib_device *device, void *client_data) 2440 { 2441 struct ib_sa_device *sa_dev = client_data; 2442 int i; 2443 2444 if (!sa_dev) 2445 return; 2446 2447 ib_unregister_event_handler(&sa_dev->event_handler); 2448 flush_workqueue(ib_wq); 2449 2450 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) { 2451 if (rdma_cap_ib_sa(device, i + 1)) { 2452 cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work); 2453 ib_unregister_mad_agent(sa_dev->port[i].agent); 2454 if (sa_dev->port[i].sm_ah) 2455 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah); 2456 } 2457 2458 } 2459 2460 kfree(sa_dev); 2461 } 2462 2463 int ib_sa_init(void) 2464 { 2465 int ret; 2466 2467 get_random_bytes(&tid, sizeof tid); 2468 2469 atomic_set(&ib_nl_sa_request_seq, 0); 2470 2471 ret = ib_register_client(&sa_client); 2472 if (ret) { 2473 pr_err("Couldn't register ib_sa client\n"); 2474 goto err1; 2475 } 2476 2477 ret = mcast_init(); 2478 if (ret) { 2479 pr_err("Couldn't initialize multicast handling\n"); 2480 goto err2; 2481 } 2482 2483 ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM); 2484 if (!ib_nl_wq) { 2485 ret = -ENOMEM; 2486 goto err3; 2487 } 2488 2489 INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout); 2490 2491 return 0; 2492 2493 err3: 2494 mcast_cleanup(); 2495 err2: 2496 ib_unregister_client(&sa_client); 2497 err1: 2498 return ret; 2499 } 2500 2501 void ib_sa_cleanup(void) 2502 { 2503 cancel_delayed_work(&ib_nl_timed_work); 2504 flush_workqueue(ib_nl_wq); 2505 destroy_workqueue(ib_nl_wq); 2506 mcast_cleanup(); 2507 ib_unregister_client(&sa_client); 2508 idr_destroy(&query_idr); 2509 } 2510