1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/accel_engine.h" 35 #include "spdk/stdinc.h" 36 #include "spdk/crc32.h" 37 #include "spdk/endian.h" 38 #include "spdk/assert.h" 39 #include "spdk/thread.h" 40 #include "spdk/nvmf_transport.h" 41 #include "spdk/string.h" 42 #include "spdk/trace.h" 43 #include "spdk/util.h" 44 #include "spdk/log.h" 45 46 #include "spdk_internal/assert.h" 47 #include "spdk_internal/nvme_tcp.h" 48 #include "spdk_internal/sock.h" 49 50 #include "nvmf_internal.h" 51 52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 54 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 55 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 56 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 57 58 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 59 60 /* spdk nvmf related structure */ 61 enum spdk_nvmf_tcp_req_state { 62 63 /* The request is not currently in use */ 64 TCP_REQUEST_STATE_FREE = 0, 65 66 /* Initial state when request first received */ 67 TCP_REQUEST_STATE_NEW, 68 69 /* The request is queued until a data buffer is available. */ 70 TCP_REQUEST_STATE_NEED_BUFFER, 71 72 /* The request is currently transferring data from the host to the controller. */ 73 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 74 75 /* The request is waiting for the R2T send acknowledgement. */ 76 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 77 78 /* The request is ready to execute at the block device */ 79 TCP_REQUEST_STATE_READY_TO_EXECUTE, 80 81 /* The request is currently executing at the block device */ 82 TCP_REQUEST_STATE_EXECUTING, 83 84 /* The request finished executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTED, 86 87 /* The request is ready to send a completion */ 88 TCP_REQUEST_STATE_READY_TO_COMPLETE, 89 90 /* The request is currently transferring final pdus from the controller to the host. */ 91 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 92 93 /* The request completed and can be marked free. */ 94 TCP_REQUEST_STATE_COMPLETED, 95 96 /* Terminator */ 97 TCP_REQUEST_NUM_STATES, 98 }; 99 100 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 101 "Invalid PDU Header Field", 102 "PDU Sequence Error", 103 "Header Digiest Error", 104 "Data Transfer Out of Range", 105 "R2T Limit Exceeded", 106 "Unsupported parameter", 107 }; 108 109 #define OBJECT_NVMF_TCP_IO 0x80 110 111 #define TRACE_GROUP_NVMF_TCP 0x5 112 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 113 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 115 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 117 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 118 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 119 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 120 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 121 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 122 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 123 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 124 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 132 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 133 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 136 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 139 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_EXECUTING", 142 TRACE_TCP_REQUEST_STATE_EXECUTING, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTED", 145 TRACE_TCP_REQUEST_STATE_EXECUTED, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 148 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 151 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_REQ_COMPLETED", 154 TRACE_TCP_REQUEST_STATE_COMPLETED, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 156 spdk_trace_register_description("TCP_WRITE_START", 157 TRACE_TCP_FLUSH_WRITEBUF_START, 158 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 159 spdk_trace_register_description("TCP_WRITE_DONE", 160 TRACE_TCP_FLUSH_WRITEBUF_DONE, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 spdk_trace_register_description("TCP_READ_DONE", 163 TRACE_TCP_READ_FROM_SOCKET_DONE, 164 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 165 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 166 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 167 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 168 } 169 170 struct spdk_nvmf_tcp_req { 171 struct spdk_nvmf_request req; 172 struct spdk_nvme_cpl rsp; 173 struct spdk_nvme_cmd cmd; 174 175 /* A PDU that can be used for sending responses. This is 176 * not the incoming PDU! */ 177 struct nvme_tcp_pdu *pdu; 178 179 /* In-capsule data buffer */ 180 uint8_t *buf; 181 /* 182 * The PDU for a request may be used multiple times in serial over 183 * the request's lifetime. For example, first to send an R2T, then 184 * to send a completion. To catch mistakes where the PDU is used 185 * twice at the same time, add a debug flag here for init/fini. 186 */ 187 bool pdu_in_use; 188 bool has_incapsule_data; 189 190 /* transfer_tag */ 191 uint16_t ttag; 192 193 enum spdk_nvmf_tcp_req_state state; 194 195 /* 196 * h2c_offset is used when we receive the h2c_data PDU. 197 */ 198 uint32_t h2c_offset; 199 200 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 201 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 202 }; 203 204 struct spdk_nvmf_tcp_qpair { 205 struct spdk_nvmf_qpair qpair; 206 struct spdk_nvmf_tcp_poll_group *group; 207 struct spdk_sock *sock; 208 209 enum nvme_tcp_pdu_recv_state recv_state; 210 enum nvme_tcp_qpair_state state; 211 212 /* PDU being actively received */ 213 struct nvme_tcp_pdu pdu_in_progress; 214 215 /* Queues to track the requests in all states */ 216 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 217 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 218 219 /* Number of requests in each state */ 220 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 221 222 uint8_t cpda; 223 224 bool host_hdgst_enable; 225 bool host_ddgst_enable; 226 227 /* This is a spare PDU used for sending special management 228 * operations. Primarily, this is used for the initial 229 * connection response and c2h termination request. */ 230 struct nvme_tcp_pdu *mgmt_pdu; 231 232 /* Arrays of in-capsule buffers, requests, and pdus. 233 * Each array is 'resource_count' number of elements */ 234 void *bufs; 235 struct spdk_nvmf_tcp_req *reqs; 236 struct nvme_tcp_pdu *pdus; 237 uint32_t resource_count; 238 uint32_t recv_buf_size; 239 240 struct spdk_nvmf_tcp_port *port; 241 242 /* IP address */ 243 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 244 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 245 246 /* IP port */ 247 uint16_t initiator_port; 248 uint16_t target_port; 249 250 /* Timer used to destroy qpair after detecting transport error issue if initiator does 251 * not close the connection. 252 */ 253 struct spdk_poller *timeout_poller; 254 255 256 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 257 }; 258 259 struct spdk_nvmf_tcp_control_msg { 260 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 261 }; 262 263 struct spdk_nvmf_tcp_control_msg_list { 264 void *msg_buf; 265 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 266 }; 267 268 struct spdk_nvmf_tcp_poll_group { 269 struct spdk_nvmf_transport_poll_group group; 270 struct spdk_sock_group *sock_group; 271 272 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 273 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 274 275 struct spdk_io_channel *accel_channel; 276 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 277 }; 278 279 struct spdk_nvmf_tcp_port { 280 const struct spdk_nvme_transport_id *trid; 281 struct spdk_sock *listen_sock; 282 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 283 }; 284 285 struct tcp_transport_opts { 286 bool c2h_success; 287 uint16_t control_msg_num; 288 uint32_t sock_priority; 289 }; 290 291 struct spdk_nvmf_tcp_transport { 292 struct spdk_nvmf_transport transport; 293 struct tcp_transport_opts tcp_opts; 294 295 pthread_mutex_t lock; 296 297 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 298 }; 299 300 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 301 { 302 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 303 spdk_json_decode_bool, true 304 }, 305 { 306 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 307 spdk_json_decode_uint16, true 308 }, 309 { 310 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 311 spdk_json_decode_uint32, true 312 }, 313 }; 314 315 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 316 struct spdk_nvmf_tcp_req *tcp_req); 317 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 318 319 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 320 struct spdk_nvmf_tcp_req *tcp_req); 321 322 static void 323 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 324 enum spdk_nvmf_tcp_req_state state) 325 { 326 struct spdk_nvmf_qpair *qpair; 327 struct spdk_nvmf_tcp_qpair *tqpair; 328 329 qpair = tcp_req->req.qpair; 330 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 331 332 assert(tqpair->state_cntr[tcp_req->state] > 0); 333 tqpair->state_cntr[tcp_req->state]--; 334 tqpair->state_cntr[state]++; 335 336 tcp_req->state = state; 337 } 338 339 static inline struct nvme_tcp_pdu * 340 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 341 { 342 assert(tcp_req->pdu_in_use == false); 343 tcp_req->pdu_in_use = true; 344 345 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 346 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 347 348 return tcp_req->pdu; 349 } 350 351 static inline void 352 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 353 { 354 tcp_req->pdu_in_use = false; 355 } 356 357 static struct spdk_nvmf_tcp_req * 358 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 359 { 360 struct spdk_nvmf_tcp_req *tcp_req; 361 362 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 363 if (!tcp_req) { 364 return NULL; 365 } 366 367 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 368 tcp_req->h2c_offset = 0; 369 tcp_req->has_incapsule_data = false; 370 tcp_req->req.dif.dif_insert_or_strip = false; 371 372 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 373 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 374 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 375 return tcp_req; 376 } 377 378 static inline void 379 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 380 { 381 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 382 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 384 } 385 386 static void 387 nvmf_tcp_request_free(void *cb_arg) 388 { 389 struct spdk_nvmf_tcp_transport *ttransport; 390 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 391 392 assert(tcp_req != NULL); 393 394 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 395 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 396 struct spdk_nvmf_tcp_transport, transport); 397 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 398 nvmf_tcp_req_process(ttransport, tcp_req); 399 } 400 401 static int 402 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 403 { 404 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 405 406 nvmf_tcp_request_free(tcp_req); 407 408 return 0; 409 } 410 411 static void 412 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 413 enum spdk_nvmf_tcp_req_state state) 414 { 415 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 416 417 assert(state != TCP_REQUEST_STATE_FREE); 418 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 419 if (state == tcp_req->state) { 420 nvmf_tcp_request_free(tcp_req); 421 } 422 } 423 } 424 425 static void 426 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 427 { 428 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 429 430 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 431 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 432 433 /* Wipe the requests waiting for buffer from the global list */ 434 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 435 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 436 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 437 spdk_nvmf_request, buf_link); 438 } 439 } 440 441 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 442 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 443 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 444 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 445 } 446 447 static void 448 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 449 { 450 int i; 451 struct spdk_nvmf_tcp_req *tcp_req; 452 453 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 454 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 455 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 456 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 457 if ((int)tcp_req->state == i) { 458 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 459 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 460 } 461 } 462 } 463 } 464 465 static void 466 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 467 { 468 int err = 0; 469 470 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 471 472 err = spdk_sock_close(&tqpair->sock); 473 assert(err == 0); 474 nvmf_tcp_cleanup_all_states(tqpair); 475 476 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 477 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 478 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 479 tqpair->resource_count); 480 err++; 481 } 482 483 if (err > 0) { 484 nvmf_tcp_dump_qpair_req_contents(tqpair); 485 } 486 487 spdk_dma_free(tqpair->pdus); 488 free(tqpair->reqs); 489 spdk_free(tqpair->bufs); 490 free(tqpair); 491 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 492 } 493 494 static void 495 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 496 { 497 struct spdk_nvmf_tcp_transport *ttransport; 498 assert(w != NULL); 499 500 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 501 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 502 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 503 } 504 505 static int 506 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 507 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 508 { 509 struct spdk_nvmf_tcp_transport *ttransport; 510 511 assert(transport != NULL); 512 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 513 514 pthread_mutex_destroy(&ttransport->lock); 515 free(ttransport); 516 517 if (cb_fn) { 518 cb_fn(cb_arg); 519 } 520 return 0; 521 } 522 523 static struct spdk_nvmf_transport * 524 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 525 { 526 struct spdk_nvmf_tcp_transport *ttransport; 527 uint32_t sge_count; 528 uint32_t min_shared_buffers; 529 530 ttransport = calloc(1, sizeof(*ttransport)); 531 if (!ttransport) { 532 return NULL; 533 } 534 535 TAILQ_INIT(&ttransport->ports); 536 537 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 538 539 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 540 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 541 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 542 if (opts->transport_specific != NULL && 543 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 544 SPDK_COUNTOF(tcp_transport_opts_decoder), 545 &ttransport->tcp_opts)) { 546 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 547 free(ttransport); 548 return NULL; 549 } 550 551 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 552 553 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 554 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 555 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 556 " in_capsule_data_size=%d, max_aq_depth=%d\n" 557 " num_shared_buffers=%d, c2h_success=%d,\n" 558 " dif_insert_or_strip=%d, sock_priority=%d\n" 559 " abort_timeout_sec=%d, control_msg_num=%hu\n", 560 opts->max_queue_depth, 561 opts->max_io_size, 562 opts->max_qpairs_per_ctrlr - 1, 563 opts->io_unit_size, 564 opts->in_capsule_data_size, 565 opts->max_aq_depth, 566 opts->num_shared_buffers, 567 ttransport->tcp_opts.c2h_success, 568 opts->dif_insert_or_strip, 569 ttransport->tcp_opts.sock_priority, 570 opts->abort_timeout_sec, 571 ttransport->tcp_opts.control_msg_num); 572 573 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 574 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 575 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 576 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 577 free(ttransport); 578 return NULL; 579 } 580 581 if (ttransport->tcp_opts.control_msg_num == 0 && 582 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 583 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 584 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 585 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 586 } 587 588 /* I/O unit size cannot be larger than max I/O size */ 589 if (opts->io_unit_size > opts->max_io_size) { 590 opts->io_unit_size = opts->max_io_size; 591 } 592 593 sge_count = opts->max_io_size / opts->io_unit_size; 594 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 595 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 596 free(ttransport); 597 return NULL; 598 } 599 600 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 601 if (min_shared_buffers > opts->num_shared_buffers) { 602 SPDK_ERRLOG("There are not enough buffers to satisfy" 603 "per-poll group caches for each thread. (%" PRIu32 ")" 604 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 605 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 606 free(ttransport); 607 return NULL; 608 } 609 610 pthread_mutex_init(&ttransport->lock, NULL); 611 612 return &ttransport->transport; 613 } 614 615 static int 616 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 617 { 618 unsigned long long ull; 619 char *end = NULL; 620 621 ull = strtoull(trsvcid, &end, 10); 622 if (end == NULL || end == trsvcid || *end != '\0') { 623 return -1; 624 } 625 626 /* Valid TCP/IP port numbers are in [0, 65535] */ 627 if (ull > 65535) { 628 return -1; 629 } 630 631 return (int)ull; 632 } 633 634 /** 635 * Canonicalize a listen address trid. 636 */ 637 static int 638 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 639 const struct spdk_nvme_transport_id *trid) 640 { 641 int trsvcid_int; 642 643 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 644 if (trsvcid_int < 0) { 645 return -EINVAL; 646 } 647 648 memset(canon_trid, 0, sizeof(*canon_trid)); 649 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 650 canon_trid->adrfam = trid->adrfam; 651 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 652 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 653 654 return 0; 655 } 656 657 /** 658 * Find an existing listening port. 659 * 660 * Caller must hold ttransport->lock. 661 */ 662 static struct spdk_nvmf_tcp_port * 663 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 664 const struct spdk_nvme_transport_id *trid) 665 { 666 struct spdk_nvme_transport_id canon_trid; 667 struct spdk_nvmf_tcp_port *port; 668 669 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 670 return NULL; 671 } 672 673 TAILQ_FOREACH(port, &ttransport->ports, link) { 674 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 675 return port; 676 } 677 } 678 679 return NULL; 680 } 681 682 static int 683 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 684 struct spdk_nvmf_listen_opts *listen_opts) 685 { 686 struct spdk_nvmf_tcp_transport *ttransport; 687 struct spdk_nvmf_tcp_port *port; 688 int trsvcid_int; 689 uint8_t adrfam; 690 struct spdk_sock_opts opts; 691 692 if (!strlen(trid->trsvcid)) { 693 SPDK_ERRLOG("Service id is required\n"); 694 return -EINVAL; 695 } 696 697 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 698 699 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 700 if (trsvcid_int < 0) { 701 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 702 return -EINVAL; 703 } 704 705 pthread_mutex_lock(&ttransport->lock); 706 port = calloc(1, sizeof(*port)); 707 if (!port) { 708 SPDK_ERRLOG("Port allocation failed\n"); 709 pthread_mutex_unlock(&ttransport->lock); 710 return -ENOMEM; 711 } 712 713 port->trid = trid; 714 opts.opts_size = sizeof(opts); 715 spdk_sock_get_default_opts(&opts); 716 opts.priority = ttransport->tcp_opts.sock_priority; 717 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 718 NULL, &opts); 719 if (port->listen_sock == NULL) { 720 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 721 trid->traddr, trsvcid_int, 722 spdk_strerror(errno), errno); 723 free(port); 724 pthread_mutex_unlock(&ttransport->lock); 725 return -errno; 726 } 727 728 if (spdk_sock_is_ipv4(port->listen_sock)) { 729 adrfam = SPDK_NVMF_ADRFAM_IPV4; 730 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 731 adrfam = SPDK_NVMF_ADRFAM_IPV6; 732 } else { 733 SPDK_ERRLOG("Unhandled socket type\n"); 734 adrfam = 0; 735 } 736 737 if (adrfam != trid->adrfam) { 738 SPDK_ERRLOG("Socket address family mismatch\n"); 739 spdk_sock_close(&port->listen_sock); 740 free(port); 741 pthread_mutex_unlock(&ttransport->lock); 742 return -EINVAL; 743 } 744 745 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 746 trid->traddr, trid->trsvcid); 747 748 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 749 pthread_mutex_unlock(&ttransport->lock); 750 return 0; 751 } 752 753 static void 754 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 755 const struct spdk_nvme_transport_id *trid) 756 { 757 struct spdk_nvmf_tcp_transport *ttransport; 758 struct spdk_nvmf_tcp_port *port; 759 760 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 761 762 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 763 trid->traddr, trid->trsvcid); 764 765 pthread_mutex_lock(&ttransport->lock); 766 port = nvmf_tcp_find_port(ttransport, trid); 767 if (port) { 768 TAILQ_REMOVE(&ttransport->ports, port, link); 769 spdk_sock_close(&port->listen_sock); 770 free(port); 771 } 772 773 pthread_mutex_unlock(&ttransport->lock); 774 } 775 776 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 777 enum nvme_tcp_pdu_recv_state state); 778 779 static void 780 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 781 { 782 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 783 784 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 785 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 786 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 787 spdk_poller_unregister(&tqpair->timeout_poller); 788 789 /* This will end up calling nvmf_tcp_close_qpair */ 790 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 791 } 792 } 793 794 static void 795 _pdu_write_done(void *_pdu, int err) 796 { 797 struct nvme_tcp_pdu *pdu = _pdu; 798 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 799 800 if (err != 0) { 801 nvmf_tcp_qpair_disconnect(tqpair); 802 return; 803 } 804 805 assert(pdu->cb_fn != NULL); 806 pdu->cb_fn(pdu->cb_arg); 807 } 808 809 static void 810 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 811 { 812 uint32_t mapped_length = 0; 813 ssize_t rc; 814 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 815 816 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 817 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 818 &mapped_length); 819 pdu->sock_req.cb_fn = _pdu_write_done; 820 pdu->sock_req.cb_arg = pdu; 821 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 822 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 823 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 824 if (rc == mapped_length) { 825 _pdu_write_done(pdu, 0); 826 } else { 827 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 828 _pdu_write_done(pdu, -1); 829 } 830 } else { 831 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 832 } 833 } 834 835 static void 836 data_crc32_accel_done(void *cb_arg, int status) 837 { 838 struct nvme_tcp_pdu *pdu = cb_arg; 839 840 if (spdk_unlikely(status)) { 841 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 842 _pdu_write_done(pdu, status); 843 return; 844 } 845 846 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 847 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 848 849 _tcp_write_pdu(pdu); 850 } 851 852 static void 853 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 854 { 855 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 856 uint32_t crc32c; 857 858 /* Data Digest */ 859 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 860 /* Only suport this limitated case for the first step */ 861 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 862 && tqpair->group)) { 863 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, 864 pdu->data_iov, pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 865 return; 866 } 867 868 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 869 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 870 } 871 872 _tcp_write_pdu(pdu); 873 } 874 875 static void 876 header_crc32_accel_done(void *cb_arg, int status) 877 { 878 struct nvme_tcp_pdu *pdu = cb_arg; 879 880 pdu->header_digest_crc32 ^= SPDK_CRC32C_XOR; 881 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, pdu->header_digest_crc32); 882 if (spdk_unlikely(status)) { 883 SPDK_ERRLOG("Failed to compute header digest on pdu=%p\n", pdu); 884 _pdu_write_done(pdu, status); 885 return; 886 } 887 888 pdu_data_crc32_compute(pdu); 889 } 890 891 static void 892 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 893 struct nvme_tcp_pdu *pdu, 894 nvme_tcp_qpair_xfer_complete_cb cb_fn, 895 void *cb_arg) 896 { 897 int hlen; 898 899 assert(&tqpair->pdu_in_progress != pdu); 900 901 hlen = pdu->hdr.common.hlen; 902 pdu->cb_fn = cb_fn; 903 pdu->cb_arg = cb_arg; 904 pdu->qpair = tqpair; 905 906 pdu->iov[0].iov_base = &pdu->hdr.raw; 907 pdu->iov[0].iov_len = hlen; 908 909 /* Header Digest */ 910 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable && tqpair->group) { 911 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->header_digest_crc32, 912 pdu->iov, 1, 0, header_crc32_accel_done, pdu); 913 return; 914 } 915 916 pdu_data_crc32_compute(pdu); 917 } 918 919 static int 920 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 921 { 922 uint32_t i; 923 struct spdk_nvmf_transport_opts *opts; 924 uint32_t in_capsule_data_size; 925 926 opts = &tqpair->qpair.transport->opts; 927 928 in_capsule_data_size = opts->in_capsule_data_size; 929 if (opts->dif_insert_or_strip) { 930 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 931 } 932 933 tqpair->resource_count = opts->max_queue_depth; 934 935 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 936 if (!tqpair->reqs) { 937 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 938 return -1; 939 } 940 941 if (in_capsule_data_size) { 942 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 943 NULL, SPDK_ENV_LCORE_ID_ANY, 944 SPDK_MALLOC_DMA); 945 if (!tqpair->bufs) { 946 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 947 return -1; 948 } 949 } 950 951 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 952 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 953 if (!tqpair->pdus) { 954 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 955 return -1; 956 } 957 958 for (i = 0; i < tqpair->resource_count; i++) { 959 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 960 961 tcp_req->ttag = i + 1; 962 tcp_req->req.qpair = &tqpair->qpair; 963 964 tcp_req->pdu = &tqpair->pdus[i]; 965 tcp_req->pdu->qpair = tqpair; 966 967 /* Set up memory to receive commands */ 968 if (tqpair->bufs) { 969 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 970 } 971 972 /* Set the cmdn and rsp */ 973 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 974 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 975 976 /* Initialize request state to FREE */ 977 tcp_req->state = TCP_REQUEST_STATE_FREE; 978 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 979 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 980 } 981 982 tqpair->mgmt_pdu = &tqpair->pdus[i]; 983 tqpair->mgmt_pdu->qpair = tqpair; 984 985 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 986 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 987 988 return 0; 989 } 990 991 static int 992 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 993 { 994 struct spdk_nvmf_tcp_qpair *tqpair; 995 996 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 997 998 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 999 1000 /* Initialise request state queues of the qpair */ 1001 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1002 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1003 1004 tqpair->host_hdgst_enable = true; 1005 tqpair->host_ddgst_enable = true; 1006 1007 return 0; 1008 } 1009 1010 static int 1011 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1012 { 1013 int rc; 1014 1015 /* set low water mark */ 1016 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1017 if (rc != 0) { 1018 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1019 return rc; 1020 } 1021 1022 return 0; 1023 } 1024 1025 static void 1026 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1027 struct spdk_nvmf_tcp_port *port, 1028 struct spdk_sock *sock) 1029 { 1030 struct spdk_nvmf_tcp_qpair *tqpair; 1031 int rc; 1032 1033 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1034 port->trid->traddr, port->trid->trsvcid); 1035 1036 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1037 if (tqpair == NULL) { 1038 SPDK_ERRLOG("Could not allocate new connection.\n"); 1039 spdk_sock_close(&sock); 1040 return; 1041 } 1042 1043 tqpair->sock = sock; 1044 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1045 tqpair->port = port; 1046 tqpair->qpair.transport = transport; 1047 1048 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1049 sizeof(tqpair->target_addr), &tqpair->target_port, 1050 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1051 &tqpair->initiator_port); 1052 if (rc < 0) { 1053 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1054 nvmf_tcp_qpair_destroy(tqpair); 1055 return; 1056 } 1057 1058 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1059 } 1060 1061 static uint32_t 1062 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1063 { 1064 struct spdk_sock *sock; 1065 uint32_t count = 0; 1066 int i; 1067 1068 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1069 sock = spdk_sock_accept(port->listen_sock); 1070 if (sock == NULL) { 1071 break; 1072 } 1073 count++; 1074 nvmf_tcp_handle_connect(transport, port, sock); 1075 } 1076 1077 return count; 1078 } 1079 1080 static uint32_t 1081 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1082 { 1083 struct spdk_nvmf_tcp_transport *ttransport; 1084 struct spdk_nvmf_tcp_port *port; 1085 uint32_t count = 0; 1086 1087 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1088 1089 TAILQ_FOREACH(port, &ttransport->ports, link) { 1090 count += nvmf_tcp_port_accept(transport, port); 1091 } 1092 1093 return count; 1094 } 1095 1096 static void 1097 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1098 struct spdk_nvme_transport_id *trid, 1099 struct spdk_nvmf_discovery_log_page_entry *entry) 1100 { 1101 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1102 entry->adrfam = trid->adrfam; 1103 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1104 1105 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1106 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1107 1108 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1109 } 1110 1111 static struct spdk_nvmf_tcp_control_msg_list * 1112 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1113 { 1114 struct spdk_nvmf_tcp_control_msg_list *list; 1115 struct spdk_nvmf_tcp_control_msg *msg; 1116 uint16_t i; 1117 1118 list = calloc(1, sizeof(*list)); 1119 if (!list) { 1120 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1121 return NULL; 1122 } 1123 1124 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1125 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1126 if (!list->msg_buf) { 1127 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1128 free(list); 1129 return NULL; 1130 } 1131 1132 STAILQ_INIT(&list->free_msgs); 1133 1134 for (i = 0; i < num_messages; i++) { 1135 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1136 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1137 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1138 } 1139 1140 return list; 1141 } 1142 1143 static void 1144 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1145 { 1146 if (!list) { 1147 return; 1148 } 1149 1150 spdk_free(list->msg_buf); 1151 free(list); 1152 } 1153 1154 static struct spdk_nvmf_transport_poll_group * 1155 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1156 { 1157 struct spdk_nvmf_tcp_transport *ttransport; 1158 struct spdk_nvmf_tcp_poll_group *tgroup; 1159 1160 tgroup = calloc(1, sizeof(*tgroup)); 1161 if (!tgroup) { 1162 return NULL; 1163 } 1164 1165 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1166 if (!tgroup->sock_group) { 1167 goto cleanup; 1168 } 1169 1170 TAILQ_INIT(&tgroup->qpairs); 1171 TAILQ_INIT(&tgroup->await_req); 1172 1173 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1174 1175 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1176 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1177 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1178 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1179 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1180 if (!tgroup->control_msg_list) { 1181 goto cleanup; 1182 } 1183 } 1184 1185 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1186 if (spdk_unlikely(!tgroup->accel_channel)) { 1187 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1188 goto cleanup; 1189 } 1190 1191 return &tgroup->group; 1192 1193 cleanup: 1194 nvmf_tcp_poll_group_destroy(&tgroup->group); 1195 return NULL; 1196 } 1197 1198 static struct spdk_nvmf_transport_poll_group * 1199 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1200 { 1201 struct spdk_nvmf_tcp_qpair *tqpair; 1202 struct spdk_sock_group *group = NULL; 1203 int rc; 1204 1205 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1206 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1207 if (!rc && group != NULL) { 1208 return spdk_sock_group_get_ctx(group); 1209 } 1210 1211 return NULL; 1212 } 1213 1214 static void 1215 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1216 { 1217 struct spdk_nvmf_tcp_poll_group *tgroup; 1218 1219 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1220 spdk_sock_group_close(&tgroup->sock_group); 1221 if (tgroup->control_msg_list) { 1222 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1223 } 1224 1225 if (tgroup->accel_channel) { 1226 spdk_put_io_channel(tgroup->accel_channel); 1227 } 1228 1229 free(tgroup); 1230 } 1231 1232 static void 1233 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1234 enum nvme_tcp_pdu_recv_state state) 1235 { 1236 if (tqpair->recv_state == state) { 1237 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1238 tqpair, state); 1239 return; 1240 } 1241 1242 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1243 /* When leaving the await req state, move the qpair to the main list */ 1244 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1245 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1246 } 1247 1248 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1249 tqpair->recv_state = state; 1250 1251 switch (state) { 1252 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1253 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1254 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1255 break; 1256 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1257 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1258 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1259 break; 1260 case NVME_TCP_PDU_RECV_STATE_ERROR: 1261 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1262 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1263 break; 1264 default: 1265 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1266 abort(); 1267 break; 1268 } 1269 } 1270 1271 static int 1272 nvmf_tcp_qpair_handle_timeout(void *ctx) 1273 { 1274 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1275 1276 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1277 1278 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1279 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1280 1281 nvmf_tcp_qpair_disconnect(tqpair); 1282 return SPDK_POLLER_BUSY; 1283 } 1284 1285 static void 1286 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1287 { 1288 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1289 1290 if (!tqpair->timeout_poller) { 1291 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1292 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1293 } 1294 } 1295 1296 static void 1297 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1298 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1299 { 1300 struct nvme_tcp_pdu *rsp_pdu; 1301 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1302 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1303 uint32_t copy_len; 1304 1305 rsp_pdu = tqpair->mgmt_pdu; 1306 1307 c2h_term_req = &rsp_pdu->hdr.term_req; 1308 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1309 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1310 1311 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1312 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1313 DSET32(&c2h_term_req->fei, error_offset); 1314 } 1315 1316 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1317 1318 /* Copy the error info into the buffer */ 1319 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1320 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1321 1322 /* Contain the header of the wrong received pdu */ 1323 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1324 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1325 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1326 } 1327 1328 static void 1329 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1330 struct spdk_nvmf_tcp_qpair *tqpair, 1331 struct nvme_tcp_pdu *pdu) 1332 { 1333 struct spdk_nvmf_tcp_req *tcp_req; 1334 1335 assert(pdu->psh_valid_bytes == pdu->psh_len); 1336 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1337 1338 tcp_req = nvmf_tcp_req_get(tqpair); 1339 if (!tcp_req) { 1340 /* Directly return and make the allocation retry again */ 1341 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1342 return; 1343 } 1344 1345 /* The host sent more commands than the maximum queue depth. */ 1346 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1347 nvmf_tcp_qpair_disconnect(tqpair); 1348 return; 1349 } 1350 1351 pdu->req = tcp_req; 1352 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1353 nvmf_tcp_req_process(ttransport, tcp_req); 1354 } 1355 1356 static void 1357 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1358 struct spdk_nvmf_tcp_qpair *tqpair, 1359 struct nvme_tcp_pdu *pdu) 1360 { 1361 struct spdk_nvmf_tcp_req *tcp_req; 1362 struct spdk_nvme_tcp_cmd *capsule_cmd; 1363 uint32_t error_offset = 0; 1364 enum spdk_nvme_tcp_term_req_fes fes; 1365 1366 capsule_cmd = &pdu->hdr.capsule_cmd; 1367 tcp_req = pdu->req; 1368 assert(tcp_req != NULL); 1369 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1370 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1371 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1372 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1373 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1374 goto err; 1375 } 1376 1377 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1378 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1379 nvmf_tcp_req_process(ttransport, tcp_req); 1380 1381 return; 1382 err: 1383 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1384 } 1385 1386 static int 1387 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1388 enum spdk_nvmf_tcp_req_state state, 1389 uint16_t cid, uint16_t tag, 1390 struct spdk_nvmf_tcp_req **req) 1391 { 1392 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1393 1394 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1395 if (tcp_req->state != state) { 1396 continue; 1397 } 1398 1399 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1400 continue; 1401 } 1402 1403 if (tcp_req->ttag == tag) { 1404 *req = tcp_req; 1405 return 0; 1406 } 1407 1408 *req = NULL; 1409 return -1; 1410 } 1411 1412 /* Didn't find it, but not an error */ 1413 *req = NULL; 1414 return 0; 1415 } 1416 1417 static void 1418 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1419 struct spdk_nvmf_tcp_qpair *tqpair, 1420 struct nvme_tcp_pdu *pdu) 1421 { 1422 struct spdk_nvmf_tcp_req *tcp_req; 1423 uint32_t error_offset = 0; 1424 enum spdk_nvme_tcp_term_req_fes fes = 0; 1425 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1426 int rc; 1427 1428 h2c_data = &pdu->hdr.h2c_data; 1429 1430 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1431 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1432 1433 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1434 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1435 if (rc == 0 && tcp_req == NULL) { 1436 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1437 h2c_data->ttag, &tcp_req); 1438 } 1439 1440 if (!tcp_req) { 1441 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1442 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1443 if (rc == 0) { 1444 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1445 } else { 1446 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1447 } 1448 goto err; 1449 } 1450 1451 if (tcp_req->h2c_offset != h2c_data->datao) { 1452 SPDK_DEBUGLOG(nvmf_tcp, 1453 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1454 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1455 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1456 goto err; 1457 } 1458 1459 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1460 SPDK_DEBUGLOG(nvmf_tcp, 1461 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1462 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1463 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1464 goto err; 1465 } 1466 1467 pdu->req = tcp_req; 1468 1469 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1470 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1471 } 1472 1473 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1474 h2c_data->datao, h2c_data->datal); 1475 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1476 return; 1477 1478 err: 1479 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1480 } 1481 1482 static void 1483 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1484 struct spdk_nvmf_tcp_qpair *tqpair) 1485 { 1486 struct nvme_tcp_pdu *rsp_pdu; 1487 struct spdk_nvme_tcp_rsp *capsule_resp; 1488 1489 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1490 1491 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1492 assert(rsp_pdu != NULL); 1493 1494 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1495 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1496 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1497 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1498 if (tqpair->host_hdgst_enable) { 1499 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1500 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1501 } 1502 1503 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1504 } 1505 1506 static void 1507 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1508 { 1509 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1510 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1511 struct spdk_nvmf_tcp_qpair, qpair); 1512 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 1513 tcp_req->req.qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1514 1515 assert(tqpair != NULL); 1516 1517 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1518 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1519 tcp_req->req.length); 1520 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1521 return; 1522 } 1523 1524 if (ttransport->tcp_opts.c2h_success) { 1525 nvmf_tcp_request_free(tcp_req); 1526 } else { 1527 nvmf_tcp_req_pdu_fini(tcp_req); 1528 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1529 } 1530 } 1531 1532 static void 1533 nvmf_tcp_r2t_complete(void *cb_arg) 1534 { 1535 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1536 struct spdk_nvmf_tcp_transport *ttransport; 1537 1538 nvmf_tcp_req_pdu_fini(tcp_req); 1539 1540 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1541 struct spdk_nvmf_tcp_transport, transport); 1542 1543 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1544 1545 if (tcp_req->h2c_offset == tcp_req->req.length) { 1546 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1547 nvmf_tcp_req_process(ttransport, tcp_req); 1548 } 1549 } 1550 1551 static void 1552 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1553 struct spdk_nvmf_tcp_req *tcp_req) 1554 { 1555 struct nvme_tcp_pdu *rsp_pdu; 1556 struct spdk_nvme_tcp_r2t_hdr *r2t; 1557 1558 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1559 assert(rsp_pdu != NULL); 1560 1561 r2t = &rsp_pdu->hdr.r2t; 1562 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1563 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1564 1565 if (tqpair->host_hdgst_enable) { 1566 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1567 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1568 } 1569 1570 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1571 r2t->ttag = tcp_req->ttag; 1572 r2t->r2to = tcp_req->h2c_offset; 1573 r2t->r2tl = tcp_req->req.length; 1574 1575 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1576 1577 SPDK_DEBUGLOG(nvmf_tcp, 1578 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1579 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1580 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1581 } 1582 1583 static void 1584 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1585 struct spdk_nvmf_tcp_qpair *tqpair, 1586 struct nvme_tcp_pdu *pdu) 1587 { 1588 struct spdk_nvmf_tcp_req *tcp_req; 1589 1590 tcp_req = pdu->req; 1591 assert(tcp_req != NULL); 1592 1593 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1594 1595 tcp_req->h2c_offset += pdu->data_len; 1596 1597 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1598 1599 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1600 * acknowledged before moving on. */ 1601 if (tcp_req->h2c_offset == tcp_req->req.length && 1602 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1603 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1604 nvmf_tcp_req_process(ttransport, tcp_req); 1605 } 1606 } 1607 1608 static void 1609 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1610 { 1611 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1612 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1613 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1614 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1615 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1616 DGET32(h2c_term_req->fei)); 1617 } 1618 } 1619 1620 static void 1621 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1622 struct nvme_tcp_pdu *pdu) 1623 { 1624 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1625 uint32_t error_offset = 0; 1626 enum spdk_nvme_tcp_term_req_fes fes; 1627 1628 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1629 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1630 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1631 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1632 goto end; 1633 } 1634 1635 /* set the data buffer */ 1636 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1637 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1638 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1639 return; 1640 end: 1641 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1642 } 1643 1644 static void 1645 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1646 struct nvme_tcp_pdu *pdu) 1647 { 1648 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1649 1650 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1651 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1652 } 1653 1654 static void 1655 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1656 struct spdk_nvmf_tcp_transport *ttransport) 1657 { 1658 int rc = 0; 1659 struct nvme_tcp_pdu *pdu; 1660 uint32_t crc32c, error_offset = 0; 1661 enum spdk_nvme_tcp_term_req_fes fes; 1662 1663 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1664 pdu = &tqpair->pdu_in_progress; 1665 1666 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1667 /* check data digest if need */ 1668 if (pdu->ddgst_enable) { 1669 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1670 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1671 if (rc == 0) { 1672 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1673 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1674 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1675 return; 1676 1677 } 1678 } 1679 1680 switch (pdu->hdr.common.pdu_type) { 1681 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1682 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1683 break; 1684 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1685 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1686 break; 1687 1688 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1689 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1690 break; 1691 1692 default: 1693 /* The code should not go to here */ 1694 SPDK_ERRLOG("The code should not go to here\n"); 1695 break; 1696 } 1697 } 1698 1699 static void 1700 nvmf_tcp_send_icresp_complete(void *cb_arg) 1701 { 1702 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1703 1704 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1705 } 1706 1707 static void 1708 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1709 struct spdk_nvmf_tcp_qpair *tqpair, 1710 struct nvme_tcp_pdu *pdu) 1711 { 1712 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1713 struct nvme_tcp_pdu *rsp_pdu; 1714 struct spdk_nvme_tcp_ic_resp *ic_resp; 1715 uint32_t error_offset = 0; 1716 enum spdk_nvme_tcp_term_req_fes fes; 1717 1718 /* Only PFV 0 is defined currently */ 1719 if (ic_req->pfv != 0) { 1720 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1721 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1722 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1723 goto end; 1724 } 1725 1726 /* MAXR2T is 0's based */ 1727 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1728 1729 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1730 if (!tqpair->host_hdgst_enable) { 1731 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1732 } 1733 1734 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1735 if (!tqpair->host_ddgst_enable) { 1736 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1737 } 1738 1739 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1740 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1741 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1742 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1743 tqpair, 1744 tqpair->recv_buf_size); 1745 /* Not fatal. */ 1746 } 1747 1748 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1749 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1750 1751 rsp_pdu = tqpair->mgmt_pdu; 1752 1753 ic_resp = &rsp_pdu->hdr.ic_resp; 1754 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1755 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1756 ic_resp->pfv = 0; 1757 ic_resp->cpda = tqpair->cpda; 1758 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1759 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1760 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1761 1762 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1763 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1764 1765 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1766 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1767 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1768 return; 1769 end: 1770 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1771 } 1772 1773 static void 1774 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1775 struct spdk_nvmf_tcp_transport *ttransport) 1776 { 1777 struct nvme_tcp_pdu *pdu; 1778 int rc; 1779 uint32_t crc32c, error_offset = 0; 1780 enum spdk_nvme_tcp_term_req_fes fes; 1781 1782 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1783 pdu = &tqpair->pdu_in_progress; 1784 1785 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1786 pdu->hdr.common.pdu_type); 1787 /* check header digest if needed */ 1788 if (pdu->has_hdgst) { 1789 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1790 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1791 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1792 if (rc == 0) { 1793 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1794 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1795 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1796 return; 1797 1798 } 1799 } 1800 1801 switch (pdu->hdr.common.pdu_type) { 1802 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1803 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1804 break; 1805 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1806 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1807 break; 1808 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1809 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1810 break; 1811 1812 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1813 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1814 break; 1815 1816 default: 1817 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1818 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1819 error_offset = 1; 1820 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1821 break; 1822 } 1823 } 1824 1825 static void 1826 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1827 { 1828 struct nvme_tcp_pdu *pdu; 1829 uint32_t error_offset = 0; 1830 enum spdk_nvme_tcp_term_req_fes fes; 1831 uint8_t expected_hlen, pdo; 1832 bool plen_error = false, pdo_error = false; 1833 1834 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1835 pdu = &tqpair->pdu_in_progress; 1836 1837 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1838 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1839 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1840 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1841 goto err; 1842 } 1843 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1844 if (pdu->hdr.common.plen != expected_hlen) { 1845 plen_error = true; 1846 } 1847 } else { 1848 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1849 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1850 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1851 goto err; 1852 } 1853 1854 switch (pdu->hdr.common.pdu_type) { 1855 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1856 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1857 pdo = pdu->hdr.common.pdo; 1858 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1859 pdo_error = true; 1860 break; 1861 } 1862 1863 if (pdu->hdr.common.plen < expected_hlen) { 1864 plen_error = true; 1865 } 1866 break; 1867 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1868 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1869 pdo = pdu->hdr.common.pdo; 1870 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1871 pdo_error = true; 1872 break; 1873 } 1874 if (pdu->hdr.common.plen < expected_hlen) { 1875 plen_error = true; 1876 } 1877 break; 1878 1879 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1880 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1881 if ((pdu->hdr.common.plen <= expected_hlen) || 1882 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1883 plen_error = true; 1884 } 1885 break; 1886 1887 default: 1888 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1889 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1890 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1891 goto err; 1892 } 1893 } 1894 1895 if (pdu->hdr.common.hlen != expected_hlen) { 1896 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1897 pdu->hdr.common.pdu_type, 1898 expected_hlen, pdu->hdr.common.hlen, tqpair); 1899 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1900 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1901 goto err; 1902 } else if (pdo_error) { 1903 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1904 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1905 } else if (plen_error) { 1906 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1907 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1908 goto err; 1909 } else { 1910 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1911 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1912 return; 1913 } 1914 err: 1915 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1916 } 1917 1918 static int 1919 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1920 int read_len) 1921 { 1922 int rc; 1923 1924 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1925 read_offset, read_len, pdu->dif_ctx); 1926 if (rc != 0) { 1927 SPDK_ERRLOG("DIF generate failed\n"); 1928 } 1929 1930 return rc; 1931 } 1932 1933 static int 1934 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1935 { 1936 int rc = 0; 1937 struct nvme_tcp_pdu *pdu; 1938 enum nvme_tcp_pdu_recv_state prev_state; 1939 uint32_t data_len; 1940 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1941 struct spdk_nvmf_tcp_transport, transport); 1942 1943 /* The loop here is to allow for several back-to-back state changes. */ 1944 do { 1945 prev_state = tqpair->recv_state; 1946 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1947 1948 pdu = &tqpair->pdu_in_progress; 1949 switch (tqpair->recv_state) { 1950 /* Wait for the common header */ 1951 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1952 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1953 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1954 return rc; 1955 } 1956 1957 rc = nvme_tcp_read_data(tqpair->sock, 1958 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1959 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1960 if (rc < 0) { 1961 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1962 return NVME_TCP_PDU_FATAL; 1963 } else if (rc > 0) { 1964 pdu->ch_valid_bytes += rc; 1965 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1966 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1967 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1968 } 1969 } 1970 1971 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1972 return NVME_TCP_PDU_IN_PROGRESS; 1973 } 1974 1975 /* The command header of this PDU has now been read from the socket. */ 1976 nvmf_tcp_pdu_ch_handle(tqpair); 1977 break; 1978 /* Wait for the pdu specific header */ 1979 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1980 rc = nvme_tcp_read_data(tqpair->sock, 1981 pdu->psh_len - pdu->psh_valid_bytes, 1982 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1983 if (rc < 0) { 1984 return NVME_TCP_PDU_FATAL; 1985 } else if (rc > 0) { 1986 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1987 0, rc, 0, 0); 1988 pdu->psh_valid_bytes += rc; 1989 } 1990 1991 if (pdu->psh_valid_bytes < pdu->psh_len) { 1992 return NVME_TCP_PDU_IN_PROGRESS; 1993 } 1994 1995 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1996 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1997 break; 1998 /* Wait for the req slot */ 1999 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2000 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2001 break; 2002 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2003 /* check whether the data is valid, if not we just return */ 2004 if (!pdu->data_len) { 2005 return NVME_TCP_PDU_IN_PROGRESS; 2006 } 2007 2008 data_len = pdu->data_len; 2009 /* data digest */ 2010 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2011 tqpair->host_ddgst_enable)) { 2012 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2013 pdu->ddgst_enable = true; 2014 } 2015 2016 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2017 if (rc < 0) { 2018 return NVME_TCP_PDU_FATAL; 2019 } 2020 pdu->rw_offset += rc; 2021 2022 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2023 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2024 if (rc != 0) { 2025 return NVME_TCP_PDU_FATAL; 2026 } 2027 } 2028 2029 if (pdu->rw_offset < data_len) { 2030 return NVME_TCP_PDU_IN_PROGRESS; 2031 } 2032 2033 /* All of this PDU has now been read from the socket. */ 2034 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2035 break; 2036 case NVME_TCP_PDU_RECV_STATE_ERROR: 2037 if (!spdk_sock_is_connected(tqpair->sock)) { 2038 return NVME_TCP_PDU_FATAL; 2039 } 2040 break; 2041 default: 2042 assert(0); 2043 SPDK_ERRLOG("code should not come to here"); 2044 break; 2045 } 2046 } while (tqpair->recv_state != prev_state); 2047 2048 return rc; 2049 } 2050 2051 static inline void * 2052 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2053 { 2054 struct spdk_nvmf_tcp_control_msg *msg; 2055 2056 assert(list); 2057 2058 msg = STAILQ_FIRST(&list->free_msgs); 2059 if (!msg) { 2060 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2061 return NULL; 2062 } 2063 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2064 return msg; 2065 } 2066 2067 static inline void 2068 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2069 { 2070 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2071 2072 assert(list); 2073 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2074 } 2075 2076 static int 2077 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2078 struct spdk_nvmf_transport *transport, 2079 struct spdk_nvmf_transport_poll_group *group) 2080 { 2081 struct spdk_nvmf_request *req = &tcp_req->req; 2082 struct spdk_nvme_cmd *cmd; 2083 struct spdk_nvme_cpl *rsp; 2084 struct spdk_nvme_sgl_descriptor *sgl; 2085 struct spdk_nvmf_tcp_poll_group *tgroup; 2086 uint32_t length; 2087 2088 cmd = &req->cmd->nvme_cmd; 2089 rsp = &req->rsp->nvme_cpl; 2090 sgl = &cmd->dptr.sgl1; 2091 2092 length = sgl->unkeyed.length; 2093 2094 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2095 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2096 if (length > transport->opts.max_io_size) { 2097 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2098 length, transport->opts.max_io_size); 2099 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2100 return -1; 2101 } 2102 2103 /* fill request length and populate iovs */ 2104 req->length = length; 2105 2106 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2107 2108 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2109 req->dif.orig_length = length; 2110 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2111 req->dif.elba_length = length; 2112 } 2113 2114 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2115 /* No available buffers. Queue this request up. */ 2116 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2117 tcp_req); 2118 return 0; 2119 } 2120 2121 /* backward compatible */ 2122 req->data = req->iov[0].iov_base; 2123 2124 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2125 tcp_req, req->iovcnt, req->data); 2126 2127 return 0; 2128 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2129 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2130 uint64_t offset = sgl->address; 2131 uint32_t max_len = transport->opts.in_capsule_data_size; 2132 assert(tcp_req->has_incapsule_data); 2133 2134 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2135 offset, length); 2136 2137 if (offset > max_len) { 2138 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2139 offset, max_len); 2140 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2141 return -1; 2142 } 2143 max_len -= (uint32_t)offset; 2144 2145 if (spdk_unlikely(length > max_len)) { 2146 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2147 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2148 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2149 2150 /* Get a buffer from dedicated list */ 2151 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2152 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2153 assert(tgroup->control_msg_list); 2154 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2155 if (!req->data) { 2156 /* No available buffers. Queue this request up. */ 2157 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2158 return 0; 2159 } 2160 } else { 2161 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2162 length, max_len); 2163 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2164 return -1; 2165 } 2166 } else { 2167 req->data = tcp_req->buf; 2168 } 2169 2170 req->length = length; 2171 req->data_from_pool = false; 2172 2173 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2174 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2175 req->dif.elba_length = length; 2176 } 2177 2178 req->iov[0].iov_base = req->data; 2179 req->iov[0].iov_len = length; 2180 req->iovcnt = 1; 2181 2182 return 0; 2183 } 2184 2185 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2186 sgl->generic.type, sgl->generic.subtype); 2187 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2188 return -1; 2189 } 2190 2191 static inline enum spdk_nvme_media_error_status_code 2192 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2193 enum spdk_nvme_media_error_status_code result; 2194 2195 switch (err_type) 2196 { 2197 case SPDK_DIF_REFTAG_ERROR: 2198 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2199 break; 2200 case SPDK_DIF_APPTAG_ERROR: 2201 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2202 break; 2203 case SPDK_DIF_GUARD_ERROR: 2204 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2205 break; 2206 default: 2207 SPDK_UNREACHABLE(); 2208 break; 2209 } 2210 2211 return result; 2212 } 2213 2214 static void 2215 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2216 struct spdk_nvmf_tcp_req *tcp_req) 2217 { 2218 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2219 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2220 struct nvme_tcp_pdu *rsp_pdu; 2221 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2222 uint32_t plen, pdo, alignment; 2223 int rc; 2224 2225 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2226 2227 rsp_pdu = tcp_req->pdu; 2228 assert(rsp_pdu != NULL); 2229 assert(tcp_req->pdu_in_use); 2230 2231 c2h_data = &rsp_pdu->hdr.c2h_data; 2232 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2233 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2234 2235 if (tqpair->host_hdgst_enable) { 2236 plen += SPDK_NVME_TCP_DIGEST_LEN; 2237 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2238 } 2239 2240 /* set the psh */ 2241 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2242 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2243 c2h_data->datao = tcp_req->pdu->rw_offset; 2244 2245 /* set the padding */ 2246 rsp_pdu->padding_len = 0; 2247 pdo = plen; 2248 if (tqpair->cpda) { 2249 alignment = (tqpair->cpda + 1) << 2; 2250 if (plen % alignment != 0) { 2251 pdo = (plen + alignment) / alignment * alignment; 2252 rsp_pdu->padding_len = pdo - plen; 2253 plen = pdo; 2254 } 2255 } 2256 2257 c2h_data->common.pdo = pdo; 2258 plen += c2h_data->datal; 2259 if (tqpair->host_ddgst_enable) { 2260 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2261 plen += SPDK_NVME_TCP_DIGEST_LEN; 2262 } 2263 2264 c2h_data->common.plen = plen; 2265 2266 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2267 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2268 } 2269 2270 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2271 c2h_data->datao, c2h_data->datal); 2272 2273 2274 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2275 if (ttransport->tcp_opts.c2h_success) { 2276 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2277 } 2278 2279 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2280 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2281 struct spdk_dif_error err_blk = {}; 2282 uint32_t mapped_length = 0; 2283 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2284 uint32_t ddgst_len = 0; 2285 2286 if (tqpair->host_ddgst_enable) { 2287 /* Data digest consumes additional iov entry */ 2288 available_iovs--; 2289 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2290 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2291 c2h_data->common.plen -= ddgst_len; 2292 } 2293 /* Temp call to estimate if data can be described by limited number of iovs. 2294 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2295 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2296 false, &mapped_length); 2297 2298 if (mapped_length != c2h_data->common.plen) { 2299 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2300 SPDK_DEBUGLOG(nvmf_tcp, 2301 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2302 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2303 c2h_data->common.plen = mapped_length; 2304 2305 /* Rebuild pdu->data_iov since data length is changed */ 2306 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2307 c2h_data->datal); 2308 2309 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2310 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2311 } 2312 2313 c2h_data->common.plen += ddgst_len; 2314 2315 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2316 2317 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2318 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2319 if (rc != 0) { 2320 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2321 err_blk.err_type, err_blk.err_offset); 2322 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2323 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2324 nvmf_tcp_req_pdu_fini(tcp_req); 2325 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2326 return; 2327 } 2328 } 2329 2330 rsp_pdu->rw_offset += c2h_data->datal; 2331 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2332 } 2333 2334 static void 2335 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2336 struct spdk_nvmf_tcp_req *tcp_req) 2337 { 2338 nvmf_tcp_req_pdu_init(tcp_req); 2339 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2340 } 2341 2342 static int 2343 request_transfer_out(struct spdk_nvmf_request *req) 2344 { 2345 struct spdk_nvmf_tcp_req *tcp_req; 2346 struct spdk_nvmf_qpair *qpair; 2347 struct spdk_nvmf_tcp_qpair *tqpair; 2348 struct spdk_nvme_cpl *rsp; 2349 2350 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2351 2352 qpair = req->qpair; 2353 rsp = &req->rsp->nvme_cpl; 2354 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2355 2356 /* Advance our sq_head pointer */ 2357 if (qpair->sq_head == qpair->sq_head_max) { 2358 qpair->sq_head = 0; 2359 } else { 2360 qpair->sq_head++; 2361 } 2362 rsp->sqhd = qpair->sq_head; 2363 2364 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2365 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2366 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2367 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2368 } else { 2369 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2370 } 2371 2372 return 0; 2373 } 2374 2375 static void 2376 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2377 struct spdk_nvmf_tcp_req *tcp_req) 2378 { 2379 struct nvme_tcp_pdu *pdu; 2380 uint32_t plen = 0; 2381 2382 pdu = &tqpair->pdu_in_progress; 2383 plen = pdu->hdr.common.hlen; 2384 2385 if (tqpair->host_hdgst_enable) { 2386 plen += SPDK_NVME_TCP_DIGEST_LEN; 2387 } 2388 2389 if (pdu->hdr.common.plen != plen) { 2390 tcp_req->has_incapsule_data = true; 2391 } 2392 } 2393 2394 static bool 2395 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2396 struct spdk_nvmf_tcp_req *tcp_req) 2397 { 2398 struct spdk_nvmf_tcp_qpair *tqpair; 2399 int rc; 2400 enum spdk_nvmf_tcp_req_state prev_state; 2401 bool progress = false; 2402 struct spdk_nvmf_transport *transport = &ttransport->transport; 2403 struct spdk_nvmf_transport_poll_group *group; 2404 struct spdk_nvmf_tcp_poll_group *tgroup; 2405 2406 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2407 group = &tqpair->group->group; 2408 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2409 2410 /* If the qpair is not active, we need to abort the outstanding requests. */ 2411 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2412 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2413 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2414 } 2415 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2416 } 2417 2418 /* The loop here is to allow for several back-to-back state changes. */ 2419 do { 2420 prev_state = tcp_req->state; 2421 2422 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2423 tqpair); 2424 2425 switch (tcp_req->state) { 2426 case TCP_REQUEST_STATE_FREE: 2427 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2428 * to escape this state. */ 2429 break; 2430 case TCP_REQUEST_STATE_NEW: 2431 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2432 2433 /* copy the cmd from the receive pdu */ 2434 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2435 2436 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2437 tcp_req->req.dif.dif_insert_or_strip = true; 2438 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2439 } 2440 2441 /* The next state transition depends on the data transfer needs of this request. */ 2442 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2443 2444 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2445 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2446 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2447 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2448 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2449 break; 2450 } 2451 2452 /* If no data to transfer, ready to execute. */ 2453 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2454 /* Reset the tqpair receving pdu state */ 2455 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2456 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2457 break; 2458 } 2459 2460 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2461 2462 if (!tcp_req->has_incapsule_data) { 2463 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2464 } 2465 2466 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2467 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2468 break; 2469 case TCP_REQUEST_STATE_NEED_BUFFER: 2470 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2471 2472 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2473 2474 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2475 SPDK_DEBUGLOG(nvmf_tcp, 2476 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2477 tcp_req, tqpair); 2478 /* This request needs to wait in line to obtain a buffer */ 2479 break; 2480 } 2481 2482 /* Try to get a data buffer */ 2483 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2484 if (rc < 0) { 2485 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2486 /* Reset the tqpair receving pdu state */ 2487 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2488 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2489 break; 2490 } 2491 2492 if (!tcp_req->req.data) { 2493 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2494 tcp_req, tqpair); 2495 /* No buffers available. */ 2496 break; 2497 } 2498 2499 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2500 2501 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2502 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2503 if (tcp_req->req.data_from_pool) { 2504 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2505 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2506 } else { 2507 struct nvme_tcp_pdu *pdu; 2508 2509 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2510 2511 pdu = &tqpair->pdu_in_progress; 2512 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2513 tqpair); 2514 /* No need to send r2t, contained in the capsuled data */ 2515 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2516 0, tcp_req->req.length); 2517 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2518 } 2519 break; 2520 } 2521 2522 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2523 break; 2524 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2525 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2526 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2527 break; 2528 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2529 2530 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2531 (uintptr_t)tcp_req, 0); 2532 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2533 * to escape this state. */ 2534 break; 2535 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2536 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2537 2538 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2539 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2540 tcp_req->req.length = tcp_req->req.dif.elba_length; 2541 } 2542 2543 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2544 spdk_nvmf_request_exec(&tcp_req->req); 2545 break; 2546 case TCP_REQUEST_STATE_EXECUTING: 2547 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2548 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2549 * to escape this state. */ 2550 break; 2551 case TCP_REQUEST_STATE_EXECUTED: 2552 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2553 2554 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2555 tcp_req->req.length = tcp_req->req.dif.orig_length; 2556 } 2557 2558 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2559 break; 2560 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2561 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2562 rc = request_transfer_out(&tcp_req->req); 2563 assert(rc == 0); /* No good way to handle this currently */ 2564 break; 2565 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2566 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2567 (uintptr_t)tcp_req, 2568 0); 2569 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2570 * to escape this state. */ 2571 break; 2572 case TCP_REQUEST_STATE_COMPLETED: 2573 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2574 if (tcp_req->req.data_from_pool) { 2575 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2576 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2577 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2578 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2579 assert(tgroup->control_msg_list); 2580 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2581 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2582 } 2583 tcp_req->req.length = 0; 2584 tcp_req->req.iovcnt = 0; 2585 tcp_req->req.data = NULL; 2586 2587 nvmf_tcp_req_pdu_fini(tcp_req); 2588 2589 nvmf_tcp_req_put(tqpair, tcp_req); 2590 break; 2591 case TCP_REQUEST_NUM_STATES: 2592 default: 2593 assert(0); 2594 break; 2595 } 2596 2597 if (tcp_req->state != prev_state) { 2598 progress = true; 2599 } 2600 } while (tcp_req->state != prev_state); 2601 2602 return progress; 2603 } 2604 2605 static void 2606 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2607 { 2608 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2609 int rc; 2610 2611 assert(tqpair != NULL); 2612 rc = nvmf_tcp_sock_process(tqpair); 2613 2614 /* If there was a new socket error, disconnect */ 2615 if (rc < 0) { 2616 nvmf_tcp_qpair_disconnect(tqpair); 2617 } 2618 } 2619 2620 static int 2621 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2622 struct spdk_nvmf_qpair *qpair) 2623 { 2624 struct spdk_nvmf_tcp_poll_group *tgroup; 2625 struct spdk_nvmf_tcp_qpair *tqpair; 2626 int rc; 2627 2628 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2629 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2630 2631 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2632 nvmf_tcp_sock_cb, tqpair); 2633 if (rc != 0) { 2634 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2635 spdk_strerror(errno), errno); 2636 return -1; 2637 } 2638 2639 rc = nvmf_tcp_qpair_sock_init(tqpair); 2640 if (rc != 0) { 2641 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2642 return -1; 2643 } 2644 2645 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2646 if (rc < 0) { 2647 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2648 return -1; 2649 } 2650 2651 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2652 if (rc < 0) { 2653 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2654 return -1; 2655 } 2656 2657 tqpair->group = tgroup; 2658 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2659 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2660 2661 return 0; 2662 } 2663 2664 static int 2665 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2666 struct spdk_nvmf_qpair *qpair) 2667 { 2668 struct spdk_nvmf_tcp_poll_group *tgroup; 2669 struct spdk_nvmf_tcp_qpair *tqpair; 2670 int rc; 2671 2672 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2673 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2674 2675 assert(tqpair->group == tgroup); 2676 2677 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2678 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2679 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2680 } else { 2681 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2682 } 2683 2684 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2685 if (rc != 0) { 2686 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2687 spdk_strerror(errno), errno); 2688 } 2689 2690 return rc; 2691 } 2692 2693 static int 2694 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2695 { 2696 struct spdk_nvmf_tcp_transport *ttransport; 2697 struct spdk_nvmf_tcp_req *tcp_req; 2698 2699 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2700 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2701 2702 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2703 nvmf_tcp_req_process(ttransport, tcp_req); 2704 2705 return 0; 2706 } 2707 2708 static void 2709 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2710 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2711 { 2712 struct spdk_nvmf_tcp_qpair *tqpair; 2713 2714 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2715 2716 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2717 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2718 nvmf_tcp_qpair_destroy(tqpair); 2719 2720 if (cb_fn) { 2721 cb_fn(cb_arg); 2722 } 2723 } 2724 2725 static int 2726 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2727 { 2728 struct spdk_nvmf_tcp_poll_group *tgroup; 2729 int rc; 2730 struct spdk_nvmf_request *req, *req_tmp; 2731 struct spdk_nvmf_tcp_req *tcp_req; 2732 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2733 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2734 struct spdk_nvmf_tcp_transport, transport); 2735 2736 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2737 2738 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2739 return 0; 2740 } 2741 2742 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2743 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2744 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2745 break; 2746 } 2747 } 2748 2749 rc = spdk_sock_group_poll(tgroup->sock_group); 2750 if (rc < 0) { 2751 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2752 } 2753 2754 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2755 nvmf_tcp_sock_process(tqpair); 2756 } 2757 2758 return rc; 2759 } 2760 2761 static int 2762 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2763 struct spdk_nvme_transport_id *trid, bool peer) 2764 { 2765 struct spdk_nvmf_tcp_qpair *tqpair; 2766 uint16_t port; 2767 2768 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2769 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2770 2771 if (peer) { 2772 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2773 port = tqpair->initiator_port; 2774 } else { 2775 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2776 port = tqpair->target_port; 2777 } 2778 2779 if (spdk_sock_is_ipv4(tqpair->sock)) { 2780 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2781 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2782 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2783 } else { 2784 return -1; 2785 } 2786 2787 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2788 return 0; 2789 } 2790 2791 static int 2792 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2793 struct spdk_nvme_transport_id *trid) 2794 { 2795 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2796 } 2797 2798 static int 2799 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2800 struct spdk_nvme_transport_id *trid) 2801 { 2802 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2803 } 2804 2805 static int 2806 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2807 struct spdk_nvme_transport_id *trid) 2808 { 2809 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2810 } 2811 2812 static void 2813 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2814 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2815 { 2816 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2817 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2818 2819 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2820 2821 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2822 } 2823 2824 static int 2825 _nvmf_tcp_qpair_abort_request(void *ctx) 2826 { 2827 struct spdk_nvmf_request *req = ctx; 2828 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2829 struct spdk_nvmf_tcp_req, req); 2830 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2831 struct spdk_nvmf_tcp_qpair, qpair); 2832 int rc; 2833 2834 spdk_poller_unregister(&req->poller); 2835 2836 switch (tcp_req_to_abort->state) { 2837 case TCP_REQUEST_STATE_EXECUTING: 2838 rc = nvmf_ctrlr_abort_request(req); 2839 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2840 return SPDK_POLLER_BUSY; 2841 } 2842 break; 2843 2844 case TCP_REQUEST_STATE_NEED_BUFFER: 2845 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2846 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2847 2848 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2849 break; 2850 2851 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2852 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2853 break; 2854 2855 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2856 if (spdk_get_ticks() < req->timeout_tsc) { 2857 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2858 return SPDK_POLLER_BUSY; 2859 } 2860 break; 2861 2862 default: 2863 break; 2864 } 2865 2866 spdk_nvmf_request_complete(req); 2867 return SPDK_POLLER_BUSY; 2868 } 2869 2870 static void 2871 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2872 struct spdk_nvmf_request *req) 2873 { 2874 struct spdk_nvmf_tcp_qpair *tqpair; 2875 struct spdk_nvmf_tcp_transport *ttransport; 2876 struct spdk_nvmf_transport *transport; 2877 uint16_t cid; 2878 uint32_t i; 2879 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2880 2881 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2882 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2883 transport = &ttransport->transport; 2884 2885 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2886 2887 for (i = 0; i < tqpair->resource_count; i++) { 2888 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2889 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2890 tcp_req_to_abort = &tqpair->reqs[i]; 2891 break; 2892 } 2893 } 2894 2895 if (tcp_req_to_abort == NULL) { 2896 spdk_nvmf_request_complete(req); 2897 return; 2898 } 2899 2900 req->req_to_abort = &tcp_req_to_abort->req; 2901 req->timeout_tsc = spdk_get_ticks() + 2902 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2903 req->poller = NULL; 2904 2905 _nvmf_tcp_qpair_abort_request(req); 2906 } 2907 2908 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2909 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2910 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2911 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2912 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2913 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2914 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2915 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2916 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2917 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2918 2919 static void 2920 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2921 { 2922 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2923 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2924 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2925 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2926 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2927 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2928 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2929 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2930 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2931 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2932 opts->transport_specific = NULL; 2933 } 2934 2935 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2936 .name = "TCP", 2937 .type = SPDK_NVME_TRANSPORT_TCP, 2938 .opts_init = nvmf_tcp_opts_init, 2939 .create = nvmf_tcp_create, 2940 .dump_opts = nvmf_tcp_dump_opts, 2941 .destroy = nvmf_tcp_destroy, 2942 2943 .listen = nvmf_tcp_listen, 2944 .stop_listen = nvmf_tcp_stop_listen, 2945 .accept = nvmf_tcp_accept, 2946 2947 .listener_discover = nvmf_tcp_discover, 2948 2949 .poll_group_create = nvmf_tcp_poll_group_create, 2950 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2951 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2952 .poll_group_add = nvmf_tcp_poll_group_add, 2953 .poll_group_remove = nvmf_tcp_poll_group_remove, 2954 .poll_group_poll = nvmf_tcp_poll_group_poll, 2955 2956 .req_free = nvmf_tcp_req_free, 2957 .req_complete = nvmf_tcp_req_complete, 2958 2959 .qpair_fini = nvmf_tcp_close_qpair, 2960 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2961 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2962 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2963 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2964 }; 2965 2966 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2967 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2968