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 #include "spdk_internal/trace_defs.h" 53 54 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 55 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 56 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 57 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 58 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 59 60 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 61 62 /* spdk nvmf related structure */ 63 enum spdk_nvmf_tcp_req_state { 64 65 /* The request is not currently in use */ 66 TCP_REQUEST_STATE_FREE = 0, 67 68 /* Initial state when request first received */ 69 TCP_REQUEST_STATE_NEW, 70 71 /* The request is queued until a data buffer is available. */ 72 TCP_REQUEST_STATE_NEED_BUFFER, 73 74 /* The request is currently transferring data from the host to the controller. */ 75 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 76 77 /* The request is waiting for the R2T send acknowledgement. */ 78 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 79 80 /* The request is ready to execute at the block device */ 81 TCP_REQUEST_STATE_READY_TO_EXECUTE, 82 83 /* The request is currently executing at the block device */ 84 TCP_REQUEST_STATE_EXECUTING, 85 86 /* The request finished executing at the block device */ 87 TCP_REQUEST_STATE_EXECUTED, 88 89 /* The request is ready to send a completion */ 90 TCP_REQUEST_STATE_READY_TO_COMPLETE, 91 92 /* The request is currently transferring final pdus from the controller to the host. */ 93 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 94 95 /* The request completed and can be marked free. */ 96 TCP_REQUEST_STATE_COMPLETED, 97 98 /* Terminator */ 99 TCP_REQUEST_NUM_STATES, 100 }; 101 102 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 103 "Invalid PDU Header Field", 104 "PDU Sequence Error", 105 "Header Digiest Error", 106 "Data Transfer Out of Range", 107 "R2T Limit Exceeded", 108 "Unsupported parameter", 109 }; 110 111 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 112 { 113 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 114 spdk_trace_register_description("TCP_REQ_NEW", 115 TRACE_TCP_REQUEST_STATE_NEW, 116 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 117 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 118 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 119 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 120 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 121 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 122 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 123 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 124 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 125 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 126 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 127 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 128 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 129 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 130 spdk_trace_register_description("TCP_REQ_EXECUTING", 131 TRACE_TCP_REQUEST_STATE_EXECUTING, 132 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 133 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 134 spdk_trace_register_description("TCP_REQ_EXECUTED", 135 TRACE_TCP_REQUEST_STATE_EXECUTED, 136 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 137 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 138 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 139 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 141 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 142 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 143 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 144 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 145 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 146 spdk_trace_register_description("TCP_REQ_COMPLETED", 147 TRACE_TCP_REQUEST_STATE_COMPLETED, 148 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 149 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 150 spdk_trace_register_description("TCP_WRITE_START", 151 TRACE_TCP_FLUSH_WRITEBUF_START, 152 OWNER_NONE, OBJECT_NONE, 0, 153 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 154 spdk_trace_register_description("TCP_WRITE_DONE", 155 TRACE_TCP_FLUSH_WRITEBUF_DONE, 156 OWNER_NONE, OBJECT_NONE, 0, 157 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 158 spdk_trace_register_description("TCP_READ_DONE", 159 TRACE_TCP_READ_FROM_SOCKET_DONE, 160 OWNER_NONE, OBJECT_NONE, 0, 161 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 162 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 163 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 164 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 165 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 166 167 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 168 OWNER_NONE, OBJECT_NONE, 0, 169 SPDK_TRACE_ARG_TYPE_INT, ""); 170 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 171 OWNER_NONE, OBJECT_NONE, 0, 172 SPDK_TRACE_ARG_TYPE_INT, ""); 173 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 174 OWNER_NONE, OBJECT_NONE, 0, 175 SPDK_TRACE_ARG_TYPE_INT, "state"); 176 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 177 OWNER_NONE, OBJECT_NONE, 0, 178 SPDK_TRACE_ARG_TYPE_INT, ""); 179 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 180 OWNER_NONE, OBJECT_NONE, 0, 181 SPDK_TRACE_ARG_TYPE_INT, ""); 182 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 183 OWNER_NONE, OBJECT_NONE, 0, 184 SPDK_TRACE_ARG_TYPE_INT, ""); 185 spdk_trace_register_description("TCP_QP_RECV_STATE_CHANGE", TRACE_TCP_QP_RECV_STATE_CHANGE, 186 OWNER_NONE, OBJECT_NONE, 0, 187 SPDK_TRACE_ARG_TYPE_INT, "state"); 188 } 189 190 struct spdk_nvmf_tcp_req { 191 struct spdk_nvmf_request req; 192 struct spdk_nvme_cpl rsp; 193 struct spdk_nvme_cmd cmd; 194 195 /* A PDU that can be used for sending responses. This is 196 * not the incoming PDU! */ 197 struct nvme_tcp_pdu *pdu; 198 199 /* In-capsule data buffer */ 200 uint8_t *buf; 201 /* 202 * The PDU for a request may be used multiple times in serial over 203 * the request's lifetime. For example, first to send an R2T, then 204 * to send a completion. To catch mistakes where the PDU is used 205 * twice at the same time, add a debug flag here for init/fini. 206 */ 207 bool pdu_in_use; 208 bool has_incapsule_data; 209 210 /* transfer_tag */ 211 uint16_t ttag; 212 213 enum spdk_nvmf_tcp_req_state state; 214 215 /* 216 * h2c_offset is used when we receive the h2c_data PDU. 217 */ 218 uint32_t h2c_offset; 219 220 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 221 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 222 }; 223 224 struct spdk_nvmf_tcp_qpair { 225 struct spdk_nvmf_qpair qpair; 226 struct spdk_nvmf_tcp_poll_group *group; 227 struct spdk_sock *sock; 228 229 enum nvme_tcp_pdu_recv_state recv_state; 230 enum nvme_tcp_qpair_state state; 231 232 /* PDU being actively received */ 233 struct nvme_tcp_pdu *pdu_in_progress; 234 235 /* Queues to track the requests in all states */ 236 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 237 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 238 239 /* Number of requests in each state */ 240 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 241 242 uint8_t cpda; 243 244 bool host_hdgst_enable; 245 bool host_ddgst_enable; 246 247 /* This is a spare PDU used for sending special management 248 * operations. Primarily, this is used for the initial 249 * connection response and c2h termination request. */ 250 struct nvme_tcp_pdu *mgmt_pdu; 251 252 /* Arrays of in-capsule buffers, requests, and pdus. 253 * Each array is 'resource_count' number of elements */ 254 void *bufs; 255 struct spdk_nvmf_tcp_req *reqs; 256 struct nvme_tcp_pdu *pdus; 257 uint32_t resource_count; 258 uint32_t recv_buf_size; 259 260 struct spdk_nvmf_tcp_port *port; 261 262 /* IP address */ 263 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 264 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 265 266 /* IP port */ 267 uint16_t initiator_port; 268 uint16_t target_port; 269 270 /* Timer used to destroy qpair after detecting transport error issue if initiator does 271 * not close the connection. 272 */ 273 struct spdk_poller *timeout_poller; 274 275 276 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 277 }; 278 279 struct spdk_nvmf_tcp_control_msg { 280 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 281 }; 282 283 struct spdk_nvmf_tcp_control_msg_list { 284 void *msg_buf; 285 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 286 }; 287 288 struct spdk_nvmf_tcp_poll_group { 289 struct spdk_nvmf_transport_poll_group group; 290 struct spdk_sock_group *sock_group; 291 292 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 293 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 294 295 struct spdk_io_channel *accel_channel; 296 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 297 }; 298 299 struct spdk_nvmf_tcp_port { 300 const struct spdk_nvme_transport_id *trid; 301 struct spdk_sock *listen_sock; 302 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 303 }; 304 305 struct tcp_transport_opts { 306 bool c2h_success; 307 uint16_t control_msg_num; 308 uint32_t sock_priority; 309 }; 310 311 struct spdk_nvmf_tcp_transport { 312 struct spdk_nvmf_transport transport; 313 struct tcp_transport_opts tcp_opts; 314 315 pthread_mutex_t lock; 316 317 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 318 }; 319 320 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 321 { 322 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 323 spdk_json_decode_bool, true 324 }, 325 { 326 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 327 spdk_json_decode_uint16, true 328 }, 329 { 330 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 331 spdk_json_decode_uint32, true 332 }, 333 }; 334 335 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 336 struct spdk_nvmf_tcp_req *tcp_req); 337 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 338 339 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 340 struct spdk_nvmf_tcp_req *tcp_req); 341 342 static void 343 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 344 enum spdk_nvmf_tcp_req_state state) 345 { 346 struct spdk_nvmf_qpair *qpair; 347 struct spdk_nvmf_tcp_qpair *tqpair; 348 349 qpair = tcp_req->req.qpair; 350 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 351 352 assert(tqpair->state_cntr[tcp_req->state] > 0); 353 tqpair->state_cntr[tcp_req->state]--; 354 tqpair->state_cntr[state]++; 355 356 tcp_req->state = state; 357 } 358 359 static inline struct nvme_tcp_pdu * 360 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 361 { 362 assert(tcp_req->pdu_in_use == false); 363 tcp_req->pdu_in_use = true; 364 365 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 366 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 367 368 return tcp_req->pdu; 369 } 370 371 static inline void 372 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 373 { 374 tcp_req->pdu_in_use = false; 375 } 376 377 static struct spdk_nvmf_tcp_req * 378 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 379 { 380 struct spdk_nvmf_tcp_req *tcp_req; 381 382 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 383 if (!tcp_req) { 384 return NULL; 385 } 386 387 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 388 tcp_req->h2c_offset = 0; 389 tcp_req->has_incapsule_data = false; 390 tcp_req->req.dif_enabled = false; 391 392 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 393 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 394 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 395 return tcp_req; 396 } 397 398 static inline void 399 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 400 { 401 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 402 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 403 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 404 } 405 406 static void 407 nvmf_tcp_request_free(void *cb_arg) 408 { 409 struct spdk_nvmf_tcp_transport *ttransport; 410 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 411 412 assert(tcp_req != NULL); 413 414 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 415 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 416 struct spdk_nvmf_tcp_transport, transport); 417 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 418 nvmf_tcp_req_process(ttransport, tcp_req); 419 } 420 421 static int 422 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 423 { 424 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 425 426 nvmf_tcp_request_free(tcp_req); 427 428 return 0; 429 } 430 431 static void 432 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 433 enum spdk_nvmf_tcp_req_state state) 434 { 435 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 436 437 assert(state != TCP_REQUEST_STATE_FREE); 438 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 439 if (state == tcp_req->state) { 440 nvmf_tcp_request_free(tcp_req); 441 } 442 } 443 } 444 445 static void 446 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 447 { 448 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 449 450 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 451 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 452 453 /* Wipe the requests waiting for buffer from the global list */ 454 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 455 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 456 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 457 spdk_nvmf_request, buf_link); 458 } 459 } 460 461 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 462 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 463 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 464 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 465 } 466 467 static void 468 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 469 { 470 int i; 471 struct spdk_nvmf_tcp_req *tcp_req; 472 473 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 474 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 475 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 476 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 477 if ((int)tcp_req->state == i) { 478 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 479 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 480 } 481 } 482 } 483 } 484 485 static void 486 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 487 { 488 int err = 0; 489 490 spdk_trace_record(TRACE_TCP_QP_DESTROY, 0, 0, (uintptr_t)tqpair); 491 492 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 493 494 err = spdk_sock_close(&tqpair->sock); 495 assert(err == 0); 496 nvmf_tcp_cleanup_all_states(tqpair); 497 498 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 499 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 500 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 501 tqpair->resource_count); 502 err++; 503 } 504 505 if (err > 0) { 506 nvmf_tcp_dump_qpair_req_contents(tqpair); 507 } 508 509 spdk_dma_free(tqpair->pdus); 510 free(tqpair->reqs); 511 spdk_free(tqpair->bufs); 512 free(tqpair); 513 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 514 } 515 516 static void 517 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 518 { 519 struct spdk_nvmf_tcp_transport *ttransport; 520 assert(w != NULL); 521 522 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 523 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 524 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 525 } 526 527 static int 528 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 529 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 530 { 531 struct spdk_nvmf_tcp_transport *ttransport; 532 533 assert(transport != NULL); 534 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 535 536 pthread_mutex_destroy(&ttransport->lock); 537 free(ttransport); 538 539 if (cb_fn) { 540 cb_fn(cb_arg); 541 } 542 return 0; 543 } 544 545 static struct spdk_nvmf_transport * 546 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 547 { 548 struct spdk_nvmf_tcp_transport *ttransport; 549 uint32_t sge_count; 550 uint32_t min_shared_buffers; 551 552 ttransport = calloc(1, sizeof(*ttransport)); 553 if (!ttransport) { 554 return NULL; 555 } 556 557 TAILQ_INIT(&ttransport->ports); 558 559 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 560 561 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 562 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 563 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 564 if (opts->transport_specific != NULL && 565 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 566 SPDK_COUNTOF(tcp_transport_opts_decoder), 567 &ttransport->tcp_opts)) { 568 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 569 free(ttransport); 570 return NULL; 571 } 572 573 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 574 575 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 576 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 577 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 578 " in_capsule_data_size=%d, max_aq_depth=%d\n" 579 " num_shared_buffers=%d, c2h_success=%d,\n" 580 " dif_insert_or_strip=%d, sock_priority=%d\n" 581 " abort_timeout_sec=%d, control_msg_num=%hu\n", 582 opts->max_queue_depth, 583 opts->max_io_size, 584 opts->max_qpairs_per_ctrlr - 1, 585 opts->io_unit_size, 586 opts->in_capsule_data_size, 587 opts->max_aq_depth, 588 opts->num_shared_buffers, 589 ttransport->tcp_opts.c2h_success, 590 opts->dif_insert_or_strip, 591 ttransport->tcp_opts.sock_priority, 592 opts->abort_timeout_sec, 593 ttransport->tcp_opts.control_msg_num); 594 595 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 596 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 597 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 598 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 599 free(ttransport); 600 return NULL; 601 } 602 603 if (ttransport->tcp_opts.control_msg_num == 0 && 604 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 605 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 606 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 607 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 608 } 609 610 /* I/O unit size cannot be larger than max I/O size */ 611 if (opts->io_unit_size > opts->max_io_size) { 612 opts->io_unit_size = opts->max_io_size; 613 } 614 615 sge_count = opts->max_io_size / opts->io_unit_size; 616 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 617 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 618 free(ttransport); 619 return NULL; 620 } 621 622 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 623 if (min_shared_buffers > opts->num_shared_buffers) { 624 SPDK_ERRLOG("There are not enough buffers to satisfy " 625 "per-poll group caches for each thread. (%" PRIu32 ") " 626 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 627 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 628 free(ttransport); 629 return NULL; 630 } 631 632 pthread_mutex_init(&ttransport->lock, NULL); 633 634 return &ttransport->transport; 635 } 636 637 static int 638 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 639 { 640 unsigned long long ull; 641 char *end = NULL; 642 643 ull = strtoull(trsvcid, &end, 10); 644 if (end == NULL || end == trsvcid || *end != '\0') { 645 return -1; 646 } 647 648 /* Valid TCP/IP port numbers are in [0, 65535] */ 649 if (ull > 65535) { 650 return -1; 651 } 652 653 return (int)ull; 654 } 655 656 /** 657 * Canonicalize a listen address trid. 658 */ 659 static int 660 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 661 const struct spdk_nvme_transport_id *trid) 662 { 663 int trsvcid_int; 664 665 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 666 if (trsvcid_int < 0) { 667 return -EINVAL; 668 } 669 670 memset(canon_trid, 0, sizeof(*canon_trid)); 671 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 672 canon_trid->adrfam = trid->adrfam; 673 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 674 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 675 676 return 0; 677 } 678 679 /** 680 * Find an existing listening port. 681 * 682 * Caller must hold ttransport->lock. 683 */ 684 static struct spdk_nvmf_tcp_port * 685 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 686 const struct spdk_nvme_transport_id *trid) 687 { 688 struct spdk_nvme_transport_id canon_trid; 689 struct spdk_nvmf_tcp_port *port; 690 691 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 692 return NULL; 693 } 694 695 TAILQ_FOREACH(port, &ttransport->ports, link) { 696 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 697 return port; 698 } 699 } 700 701 return NULL; 702 } 703 704 static int 705 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 706 struct spdk_nvmf_listen_opts *listen_opts) 707 { 708 struct spdk_nvmf_tcp_transport *ttransport; 709 struct spdk_nvmf_tcp_port *port; 710 int trsvcid_int; 711 uint8_t adrfam; 712 struct spdk_sock_opts opts; 713 714 if (!strlen(trid->trsvcid)) { 715 SPDK_ERRLOG("Service id is required\n"); 716 return -EINVAL; 717 } 718 719 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 720 721 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 722 if (trsvcid_int < 0) { 723 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 724 return -EINVAL; 725 } 726 727 pthread_mutex_lock(&ttransport->lock); 728 port = calloc(1, sizeof(*port)); 729 if (!port) { 730 SPDK_ERRLOG("Port allocation failed\n"); 731 pthread_mutex_unlock(&ttransport->lock); 732 return -ENOMEM; 733 } 734 735 port->trid = trid; 736 opts.opts_size = sizeof(opts); 737 spdk_sock_get_default_opts(&opts); 738 opts.priority = ttransport->tcp_opts.sock_priority; 739 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 740 NULL, &opts); 741 if (port->listen_sock == NULL) { 742 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 743 trid->traddr, trsvcid_int, 744 spdk_strerror(errno), errno); 745 free(port); 746 pthread_mutex_unlock(&ttransport->lock); 747 return -errno; 748 } 749 750 if (spdk_sock_is_ipv4(port->listen_sock)) { 751 adrfam = SPDK_NVMF_ADRFAM_IPV4; 752 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 753 adrfam = SPDK_NVMF_ADRFAM_IPV6; 754 } else { 755 SPDK_ERRLOG("Unhandled socket type\n"); 756 adrfam = 0; 757 } 758 759 if (adrfam != trid->adrfam) { 760 SPDK_ERRLOG("Socket address family mismatch\n"); 761 spdk_sock_close(&port->listen_sock); 762 free(port); 763 pthread_mutex_unlock(&ttransport->lock); 764 return -EINVAL; 765 } 766 767 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 768 trid->traddr, trid->trsvcid); 769 770 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 771 pthread_mutex_unlock(&ttransport->lock); 772 return 0; 773 } 774 775 static void 776 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 777 const struct spdk_nvme_transport_id *trid) 778 { 779 struct spdk_nvmf_tcp_transport *ttransport; 780 struct spdk_nvmf_tcp_port *port; 781 782 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 783 784 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 785 trid->traddr, trid->trsvcid); 786 787 pthread_mutex_lock(&ttransport->lock); 788 port = nvmf_tcp_find_port(ttransport, trid); 789 if (port) { 790 TAILQ_REMOVE(&ttransport->ports, port, link); 791 spdk_sock_close(&port->listen_sock); 792 free(port); 793 } 794 795 pthread_mutex_unlock(&ttransport->lock); 796 } 797 798 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 799 enum nvme_tcp_pdu_recv_state state); 800 801 static void 802 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 803 { 804 tqpair->state = state; 805 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, 0, 0, (uintptr_t)tqpair, tqpair->state); 806 } 807 808 static void 809 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 810 { 811 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 812 813 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, 0, 0, (uintptr_t)tqpair); 814 815 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 816 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 817 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 818 spdk_poller_unregister(&tqpair->timeout_poller); 819 820 /* This will end up calling nvmf_tcp_close_qpair */ 821 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 822 } 823 } 824 825 static void 826 _pdu_write_done(void *_pdu, int err) 827 { 828 struct nvme_tcp_pdu *pdu = _pdu; 829 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 830 831 if (err != 0) { 832 nvmf_tcp_qpair_disconnect(tqpair); 833 return; 834 } 835 836 assert(pdu->cb_fn != NULL); 837 pdu->cb_fn(pdu->cb_arg); 838 } 839 840 static void 841 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 842 { 843 uint32_t mapped_length = 0; 844 ssize_t rc; 845 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 846 847 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 848 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 849 &mapped_length); 850 pdu->sock_req.cb_fn = _pdu_write_done; 851 pdu->sock_req.cb_arg = pdu; 852 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 853 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 854 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 855 if (rc == mapped_length) { 856 _pdu_write_done(pdu, 0); 857 } else { 858 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 859 _pdu_write_done(pdu, -1); 860 } 861 } else { 862 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 863 } 864 } 865 866 static void 867 data_crc32_accel_done(void *cb_arg, int status) 868 { 869 struct nvme_tcp_pdu *pdu = cb_arg; 870 871 if (spdk_unlikely(status)) { 872 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 873 _pdu_write_done(pdu, status); 874 return; 875 } 876 877 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 878 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 879 880 _tcp_write_pdu(pdu); 881 } 882 883 static void 884 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 885 { 886 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 887 uint32_t crc32c; 888 889 /* Data Digest */ 890 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 891 /* Only suport this limitated case for the first step */ 892 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 893 && tqpair->group)) { 894 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, 895 pdu->data_iov, pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 896 return; 897 } 898 899 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 900 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 901 } 902 903 _tcp_write_pdu(pdu); 904 } 905 906 static void 907 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 908 struct nvme_tcp_pdu *pdu, 909 nvme_tcp_qpair_xfer_complete_cb cb_fn, 910 void *cb_arg) 911 { 912 int hlen; 913 uint32_t crc32c; 914 915 assert(tqpair->pdu_in_progress != pdu); 916 917 hlen = pdu->hdr.common.hlen; 918 pdu->cb_fn = cb_fn; 919 pdu->cb_arg = cb_arg; 920 921 pdu->iov[0].iov_base = &pdu->hdr.raw; 922 pdu->iov[0].iov_len = hlen; 923 924 /* Header Digest */ 925 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 926 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 927 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 928 } 929 930 /* Data Digest */ 931 pdu_data_crc32_compute(pdu); 932 } 933 934 static int 935 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 936 { 937 uint32_t i; 938 struct spdk_nvmf_transport_opts *opts; 939 uint32_t in_capsule_data_size; 940 941 opts = &tqpair->qpair.transport->opts; 942 943 in_capsule_data_size = opts->in_capsule_data_size; 944 if (opts->dif_insert_or_strip) { 945 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 946 } 947 948 tqpair->resource_count = opts->max_queue_depth; 949 950 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 951 if (!tqpair->reqs) { 952 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 953 return -1; 954 } 955 956 if (in_capsule_data_size) { 957 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 958 NULL, SPDK_ENV_LCORE_ID_ANY, 959 SPDK_MALLOC_DMA); 960 if (!tqpair->bufs) { 961 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 962 return -1; 963 } 964 } 965 966 /* Add addtional 2 members, which will be used for mgmt_pdu and pdu_in_progress owned by the tqpair */ 967 tqpair->pdus = spdk_dma_zmalloc((tqpair->resource_count + 2) * sizeof(*tqpair->pdus), 0x1000, NULL); 968 if (!tqpair->pdus) { 969 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 970 return -1; 971 } 972 973 for (i = 0; i < tqpair->resource_count; i++) { 974 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 975 976 tcp_req->ttag = i + 1; 977 tcp_req->req.qpair = &tqpair->qpair; 978 979 tcp_req->pdu = &tqpair->pdus[i]; 980 tcp_req->pdu->qpair = tqpair; 981 982 /* Set up memory to receive commands */ 983 if (tqpair->bufs) { 984 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 985 } 986 987 /* Set the cmdn and rsp */ 988 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 989 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 990 991 /* Initialize request state to FREE */ 992 tcp_req->state = TCP_REQUEST_STATE_FREE; 993 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 994 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 995 } 996 997 tqpair->mgmt_pdu = &tqpair->pdus[i]; 998 tqpair->mgmt_pdu->qpair = tqpair; 999 tqpair->pdu_in_progress = &tqpair->pdus[i + 1]; 1000 1001 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1002 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1003 1004 return 0; 1005 } 1006 1007 static int 1008 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1009 { 1010 struct spdk_nvmf_tcp_qpair *tqpair; 1011 1012 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1013 1014 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1015 1016 spdk_trace_record(TRACE_TCP_QP_CREATE, 0, 0, (uintptr_t)tqpair); 1017 1018 /* Initialise request state queues of the qpair */ 1019 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1020 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1021 1022 tqpair->host_hdgst_enable = true; 1023 tqpair->host_ddgst_enable = true; 1024 1025 return 0; 1026 } 1027 1028 static int 1029 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1030 { 1031 int rc; 1032 1033 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, 0, 0, (uintptr_t)tqpair); 1034 1035 /* set low water mark */ 1036 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1037 if (rc != 0) { 1038 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1039 return rc; 1040 } 1041 1042 return 0; 1043 } 1044 1045 static void 1046 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1047 struct spdk_nvmf_tcp_port *port, 1048 struct spdk_sock *sock) 1049 { 1050 struct spdk_nvmf_tcp_qpair *tqpair; 1051 int rc; 1052 1053 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1054 port->trid->traddr, port->trid->trsvcid); 1055 1056 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1057 if (tqpair == NULL) { 1058 SPDK_ERRLOG("Could not allocate new connection.\n"); 1059 spdk_sock_close(&sock); 1060 return; 1061 } 1062 1063 tqpair->sock = sock; 1064 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1065 tqpair->port = port; 1066 tqpair->qpair.transport = transport; 1067 1068 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1069 sizeof(tqpair->target_addr), &tqpair->target_port, 1070 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1071 &tqpair->initiator_port); 1072 if (rc < 0) { 1073 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1074 nvmf_tcp_qpair_destroy(tqpair); 1075 return; 1076 } 1077 1078 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1079 } 1080 1081 static uint32_t 1082 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1083 { 1084 struct spdk_sock *sock; 1085 uint32_t count = 0; 1086 int i; 1087 1088 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1089 sock = spdk_sock_accept(port->listen_sock); 1090 if (sock == NULL) { 1091 break; 1092 } 1093 count++; 1094 nvmf_tcp_handle_connect(transport, port, sock); 1095 } 1096 1097 return count; 1098 } 1099 1100 static uint32_t 1101 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1102 { 1103 struct spdk_nvmf_tcp_transport *ttransport; 1104 struct spdk_nvmf_tcp_port *port; 1105 uint32_t count = 0; 1106 1107 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1108 1109 TAILQ_FOREACH(port, &ttransport->ports, link) { 1110 count += nvmf_tcp_port_accept(transport, port); 1111 } 1112 1113 return count; 1114 } 1115 1116 static void 1117 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1118 struct spdk_nvme_transport_id *trid, 1119 struct spdk_nvmf_discovery_log_page_entry *entry) 1120 { 1121 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1122 entry->adrfam = trid->adrfam; 1123 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1124 1125 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1126 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1127 1128 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1129 } 1130 1131 static struct spdk_nvmf_tcp_control_msg_list * 1132 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1133 { 1134 struct spdk_nvmf_tcp_control_msg_list *list; 1135 struct spdk_nvmf_tcp_control_msg *msg; 1136 uint16_t i; 1137 1138 list = calloc(1, sizeof(*list)); 1139 if (!list) { 1140 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1141 return NULL; 1142 } 1143 1144 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1145 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1146 if (!list->msg_buf) { 1147 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1148 free(list); 1149 return NULL; 1150 } 1151 1152 STAILQ_INIT(&list->free_msgs); 1153 1154 for (i = 0; i < num_messages; i++) { 1155 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1156 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1157 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1158 } 1159 1160 return list; 1161 } 1162 1163 static void 1164 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1165 { 1166 if (!list) { 1167 return; 1168 } 1169 1170 spdk_free(list->msg_buf); 1171 free(list); 1172 } 1173 1174 static struct spdk_nvmf_transport_poll_group * 1175 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1176 { 1177 struct spdk_nvmf_tcp_transport *ttransport; 1178 struct spdk_nvmf_tcp_poll_group *tgroup; 1179 1180 tgroup = calloc(1, sizeof(*tgroup)); 1181 if (!tgroup) { 1182 return NULL; 1183 } 1184 1185 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1186 if (!tgroup->sock_group) { 1187 goto cleanup; 1188 } 1189 1190 TAILQ_INIT(&tgroup->qpairs); 1191 TAILQ_INIT(&tgroup->await_req); 1192 1193 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1194 1195 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1196 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1197 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1198 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1199 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1200 if (!tgroup->control_msg_list) { 1201 goto cleanup; 1202 } 1203 } 1204 1205 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1206 if (spdk_unlikely(!tgroup->accel_channel)) { 1207 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1208 goto cleanup; 1209 } 1210 1211 return &tgroup->group; 1212 1213 cleanup: 1214 nvmf_tcp_poll_group_destroy(&tgroup->group); 1215 return NULL; 1216 } 1217 1218 static struct spdk_nvmf_transport_poll_group * 1219 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1220 { 1221 struct spdk_nvmf_tcp_qpair *tqpair; 1222 struct spdk_sock_group *group = NULL; 1223 int rc; 1224 1225 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1226 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1227 if (!rc && group != NULL) { 1228 return spdk_sock_group_get_ctx(group); 1229 } 1230 1231 return NULL; 1232 } 1233 1234 static void 1235 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1236 { 1237 struct spdk_nvmf_tcp_poll_group *tgroup; 1238 1239 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1240 spdk_sock_group_close(&tgroup->sock_group); 1241 if (tgroup->control_msg_list) { 1242 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1243 } 1244 1245 if (tgroup->accel_channel) { 1246 spdk_put_io_channel(tgroup->accel_channel); 1247 } 1248 1249 free(tgroup); 1250 } 1251 1252 static void 1253 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1254 enum nvme_tcp_pdu_recv_state state) 1255 { 1256 if (tqpair->recv_state == state) { 1257 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1258 tqpair, state); 1259 return; 1260 } 1261 1262 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1263 /* When leaving the await req state, move the qpair to the main list */ 1264 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1265 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1266 } 1267 1268 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1269 tqpair->recv_state = state; 1270 1271 spdk_trace_record(TRACE_TCP_QP_RECV_STATE_CHANGE, 0, 0, (uintptr_t)tqpair, tqpair->recv_state); 1272 1273 switch (state) { 1274 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1275 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1276 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1277 break; 1278 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1279 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1280 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1281 break; 1282 case NVME_TCP_PDU_RECV_STATE_ERROR: 1283 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1284 memset(tqpair->pdu_in_progress, 0, sizeof(*(tqpair->pdu_in_progress))); 1285 break; 1286 default: 1287 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1288 abort(); 1289 break; 1290 } 1291 } 1292 1293 static int 1294 nvmf_tcp_qpair_handle_timeout(void *ctx) 1295 { 1296 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1297 1298 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1299 1300 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1301 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1302 1303 nvmf_tcp_qpair_disconnect(tqpair); 1304 return SPDK_POLLER_BUSY; 1305 } 1306 1307 static void 1308 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1309 { 1310 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1311 1312 if (!tqpair->timeout_poller) { 1313 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1314 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1315 } 1316 } 1317 1318 static void 1319 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1320 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1321 { 1322 struct nvme_tcp_pdu *rsp_pdu; 1323 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1324 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1325 uint32_t copy_len; 1326 1327 rsp_pdu = tqpair->mgmt_pdu; 1328 1329 c2h_term_req = &rsp_pdu->hdr.term_req; 1330 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1331 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1332 1333 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1334 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1335 DSET32(&c2h_term_req->fei, error_offset); 1336 } 1337 1338 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1339 1340 /* Copy the error info into the buffer */ 1341 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1342 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1343 1344 /* Contain the header of the wrong received pdu */ 1345 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1346 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1347 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1348 } 1349 1350 static void 1351 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1352 struct spdk_nvmf_tcp_qpair *tqpair, 1353 struct nvme_tcp_pdu *pdu) 1354 { 1355 struct spdk_nvmf_tcp_req *tcp_req; 1356 1357 assert(pdu->psh_valid_bytes == pdu->psh_len); 1358 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1359 1360 tcp_req = nvmf_tcp_req_get(tqpair); 1361 if (!tcp_req) { 1362 /* Directly return and make the allocation retry again */ 1363 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1364 return; 1365 } 1366 1367 /* The host sent more commands than the maximum queue depth. */ 1368 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1369 nvmf_tcp_qpair_disconnect(tqpair); 1370 return; 1371 } 1372 1373 pdu->req = tcp_req; 1374 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1375 nvmf_tcp_req_process(ttransport, tcp_req); 1376 } 1377 1378 static void 1379 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1380 struct spdk_nvmf_tcp_qpair *tqpair, 1381 struct nvme_tcp_pdu *pdu) 1382 { 1383 struct spdk_nvmf_tcp_req *tcp_req; 1384 struct spdk_nvme_tcp_cmd *capsule_cmd; 1385 uint32_t error_offset = 0; 1386 enum spdk_nvme_tcp_term_req_fes fes; 1387 struct spdk_nvme_cpl *rsp; 1388 1389 capsule_cmd = &pdu->hdr.capsule_cmd; 1390 tcp_req = pdu->req; 1391 assert(tcp_req != NULL); 1392 1393 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1394 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1395 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1396 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1397 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1398 goto err; 1399 } 1400 1401 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1402 1403 rsp = &tcp_req->req.rsp->nvme_cpl; 1404 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1405 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1406 } else { 1407 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1408 } 1409 1410 nvmf_tcp_req_process(ttransport, tcp_req); 1411 1412 return; 1413 err: 1414 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1415 } 1416 1417 static int 1418 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1419 enum spdk_nvmf_tcp_req_state state, 1420 uint16_t cid, uint16_t tag, 1421 struct spdk_nvmf_tcp_req **req) 1422 { 1423 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1424 1425 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1426 if (tcp_req->state != state) { 1427 continue; 1428 } 1429 1430 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1431 continue; 1432 } 1433 1434 if (tcp_req->ttag == tag) { 1435 *req = tcp_req; 1436 return 0; 1437 } 1438 1439 *req = NULL; 1440 return -1; 1441 } 1442 1443 /* Didn't find it, but not an error */ 1444 *req = NULL; 1445 return 0; 1446 } 1447 1448 static void 1449 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1450 struct spdk_nvmf_tcp_qpair *tqpair, 1451 struct nvme_tcp_pdu *pdu) 1452 { 1453 struct spdk_nvmf_tcp_req *tcp_req; 1454 uint32_t error_offset = 0; 1455 enum spdk_nvme_tcp_term_req_fes fes = 0; 1456 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1457 int rc; 1458 1459 h2c_data = &pdu->hdr.h2c_data; 1460 1461 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1462 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1463 1464 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1465 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1466 if (rc == 0 && tcp_req == NULL) { 1467 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1468 h2c_data->ttag, &tcp_req); 1469 } 1470 1471 if (!tcp_req) { 1472 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1473 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1474 if (rc == 0) { 1475 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1476 } else { 1477 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1478 } 1479 goto err; 1480 } 1481 1482 if (tcp_req->h2c_offset != h2c_data->datao) { 1483 SPDK_DEBUGLOG(nvmf_tcp, 1484 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1485 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1486 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1487 goto err; 1488 } 1489 1490 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1491 SPDK_DEBUGLOG(nvmf_tcp, 1492 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1493 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1494 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1495 goto err; 1496 } 1497 1498 pdu->req = tcp_req; 1499 1500 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1501 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1502 } 1503 1504 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1505 h2c_data->datao, h2c_data->datal); 1506 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1507 return; 1508 1509 err: 1510 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1511 } 1512 1513 static void 1514 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1515 struct spdk_nvmf_tcp_qpair *tqpair) 1516 { 1517 struct nvme_tcp_pdu *rsp_pdu; 1518 struct spdk_nvme_tcp_rsp *capsule_resp; 1519 1520 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1521 1522 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1523 assert(rsp_pdu != NULL); 1524 1525 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1526 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1527 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1528 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1529 if (tqpair->host_hdgst_enable) { 1530 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1531 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1532 } 1533 1534 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1535 } 1536 1537 static void 1538 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1539 { 1540 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1541 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1542 struct spdk_nvmf_tcp_qpair, qpair); 1543 1544 assert(tqpair != NULL); 1545 1546 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1547 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1548 tcp_req->req.length); 1549 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1550 return; 1551 } 1552 1553 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1554 nvmf_tcp_request_free(tcp_req); 1555 } else { 1556 nvmf_tcp_req_pdu_fini(tcp_req); 1557 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1558 } 1559 } 1560 1561 static void 1562 nvmf_tcp_r2t_complete(void *cb_arg) 1563 { 1564 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1565 struct spdk_nvmf_tcp_transport *ttransport; 1566 1567 nvmf_tcp_req_pdu_fini(tcp_req); 1568 1569 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1570 struct spdk_nvmf_tcp_transport, transport); 1571 1572 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1573 1574 if (tcp_req->h2c_offset == tcp_req->req.length) { 1575 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1576 nvmf_tcp_req_process(ttransport, tcp_req); 1577 } 1578 } 1579 1580 static void 1581 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1582 struct spdk_nvmf_tcp_req *tcp_req) 1583 { 1584 struct nvme_tcp_pdu *rsp_pdu; 1585 struct spdk_nvme_tcp_r2t_hdr *r2t; 1586 1587 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1588 assert(rsp_pdu != NULL); 1589 1590 r2t = &rsp_pdu->hdr.r2t; 1591 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1592 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1593 1594 if (tqpair->host_hdgst_enable) { 1595 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1596 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1597 } 1598 1599 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1600 r2t->ttag = tcp_req->ttag; 1601 r2t->r2to = tcp_req->h2c_offset; 1602 r2t->r2tl = tcp_req->req.length; 1603 1604 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1605 1606 SPDK_DEBUGLOG(nvmf_tcp, 1607 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1608 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1609 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1610 } 1611 1612 static void 1613 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1614 struct spdk_nvmf_tcp_qpair *tqpair, 1615 struct nvme_tcp_pdu *pdu) 1616 { 1617 struct spdk_nvmf_tcp_req *tcp_req; 1618 struct spdk_nvme_cpl *rsp; 1619 1620 tcp_req = pdu->req; 1621 assert(tcp_req != NULL); 1622 1623 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1624 1625 tcp_req->h2c_offset += pdu->data_len; 1626 1627 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1628 1629 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1630 * acknowledged before moving on. */ 1631 if (tcp_req->h2c_offset == tcp_req->req.length && 1632 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1633 /* After receving all the h2c data, we need to check whether there is 1634 * transient transport error */ 1635 rsp = &tcp_req->req.rsp->nvme_cpl; 1636 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1637 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1638 } else { 1639 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1640 } 1641 nvmf_tcp_req_process(ttransport, tcp_req); 1642 } 1643 } 1644 1645 static void 1646 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1647 { 1648 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1649 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1650 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1651 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1652 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1653 DGET32(h2c_term_req->fei)); 1654 } 1655 } 1656 1657 static void 1658 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1659 struct nvme_tcp_pdu *pdu) 1660 { 1661 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1662 uint32_t error_offset = 0; 1663 enum spdk_nvme_tcp_term_req_fes fes; 1664 1665 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1666 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1667 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1668 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1669 goto end; 1670 } 1671 1672 /* set the data buffer */ 1673 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1674 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1675 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1676 return; 1677 end: 1678 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1679 } 1680 1681 static void 1682 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1683 struct nvme_tcp_pdu *pdu) 1684 { 1685 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1686 1687 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1688 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1689 } 1690 1691 static void 1692 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1693 struct spdk_nvmf_tcp_transport *ttransport) 1694 { 1695 struct nvme_tcp_pdu *pdu = tqpair->pdu_in_progress; 1696 1697 switch (pdu->hdr.common.pdu_type) { 1698 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1699 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1700 break; 1701 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1702 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1703 break; 1704 1705 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1706 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1707 break; 1708 1709 default: 1710 /* The code should not go to here */ 1711 SPDK_ERRLOG("The code should not go to here\n"); 1712 break; 1713 } 1714 } 1715 1716 static void 1717 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1718 struct spdk_nvmf_tcp_transport *ttransport) 1719 { 1720 int rc = 0; 1721 struct nvme_tcp_pdu *pdu; 1722 uint32_t crc32c; 1723 struct spdk_nvmf_tcp_req *tcp_req; 1724 struct spdk_nvme_cpl *rsp; 1725 1726 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1727 pdu = tqpair->pdu_in_progress; 1728 1729 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1730 /* check data digest if need */ 1731 if (pdu->ddgst_enable) { 1732 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1733 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1734 if (rc == 0) { 1735 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1736 tcp_req = pdu->req; 1737 assert(tcp_req != NULL); 1738 rsp = &tcp_req->req.rsp->nvme_cpl; 1739 rsp->status.sc = SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR; 1740 } 1741 } 1742 1743 _nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1744 } 1745 1746 static void 1747 nvmf_tcp_send_icresp_complete(void *cb_arg) 1748 { 1749 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1750 1751 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 1752 } 1753 1754 static void 1755 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1756 struct spdk_nvmf_tcp_qpair *tqpair, 1757 struct nvme_tcp_pdu *pdu) 1758 { 1759 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1760 struct nvme_tcp_pdu *rsp_pdu; 1761 struct spdk_nvme_tcp_ic_resp *ic_resp; 1762 uint32_t error_offset = 0; 1763 enum spdk_nvme_tcp_term_req_fes fes; 1764 1765 /* Only PFV 0 is defined currently */ 1766 if (ic_req->pfv != 0) { 1767 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1768 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1769 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1770 goto end; 1771 } 1772 1773 /* MAXR2T is 0's based */ 1774 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1775 1776 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1777 if (!tqpair->host_hdgst_enable) { 1778 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1779 } 1780 1781 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1782 if (!tqpair->host_ddgst_enable) { 1783 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1784 } 1785 1786 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1787 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1788 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1789 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1790 tqpair, 1791 tqpair->recv_buf_size); 1792 /* Not fatal. */ 1793 } 1794 1795 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1796 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1797 1798 rsp_pdu = tqpair->mgmt_pdu; 1799 1800 ic_resp = &rsp_pdu->hdr.ic_resp; 1801 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1802 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1803 ic_resp->pfv = 0; 1804 ic_resp->cpda = tqpair->cpda; 1805 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1806 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1807 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1808 1809 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1810 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1811 1812 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 1813 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1814 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1815 return; 1816 end: 1817 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1818 } 1819 1820 static void 1821 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1822 struct spdk_nvmf_tcp_transport *ttransport) 1823 { 1824 struct nvme_tcp_pdu *pdu; 1825 int rc; 1826 uint32_t crc32c, error_offset = 0; 1827 enum spdk_nvme_tcp_term_req_fes fes; 1828 1829 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1830 pdu = tqpair->pdu_in_progress; 1831 1832 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1833 pdu->hdr.common.pdu_type); 1834 /* check header digest if needed */ 1835 if (pdu->has_hdgst) { 1836 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1837 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1838 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1839 if (rc == 0) { 1840 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1841 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1842 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1843 return; 1844 1845 } 1846 } 1847 1848 switch (pdu->hdr.common.pdu_type) { 1849 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1850 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1851 break; 1852 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1853 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1854 break; 1855 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1856 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1857 break; 1858 1859 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1860 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1861 break; 1862 1863 default: 1864 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 1865 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1866 error_offset = 1; 1867 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1868 break; 1869 } 1870 } 1871 1872 static void 1873 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1874 { 1875 struct nvme_tcp_pdu *pdu; 1876 uint32_t error_offset = 0; 1877 enum spdk_nvme_tcp_term_req_fes fes; 1878 uint8_t expected_hlen, pdo; 1879 bool plen_error = false, pdo_error = false; 1880 1881 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1882 pdu = tqpair->pdu_in_progress; 1883 1884 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1885 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1886 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1887 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1888 goto err; 1889 } 1890 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1891 if (pdu->hdr.common.plen != expected_hlen) { 1892 plen_error = true; 1893 } 1894 } else { 1895 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1896 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1897 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1898 goto err; 1899 } 1900 1901 switch (pdu->hdr.common.pdu_type) { 1902 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1903 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1904 pdo = pdu->hdr.common.pdo; 1905 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1906 pdo_error = true; 1907 break; 1908 } 1909 1910 if (pdu->hdr.common.plen < expected_hlen) { 1911 plen_error = true; 1912 } 1913 break; 1914 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1915 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1916 pdo = pdu->hdr.common.pdo; 1917 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1918 pdo_error = true; 1919 break; 1920 } 1921 if (pdu->hdr.common.plen < expected_hlen) { 1922 plen_error = true; 1923 } 1924 break; 1925 1926 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1927 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1928 if ((pdu->hdr.common.plen <= expected_hlen) || 1929 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1930 plen_error = true; 1931 } 1932 break; 1933 1934 default: 1935 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1936 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1937 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1938 goto err; 1939 } 1940 } 1941 1942 if (pdu->hdr.common.hlen != expected_hlen) { 1943 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1944 pdu->hdr.common.pdu_type, 1945 expected_hlen, pdu->hdr.common.hlen, tqpair); 1946 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1947 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1948 goto err; 1949 } else if (pdo_error) { 1950 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1951 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1952 } else if (plen_error) { 1953 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1954 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1955 goto err; 1956 } else { 1957 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1958 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1959 return; 1960 } 1961 err: 1962 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1963 } 1964 1965 static int 1966 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1967 int read_len) 1968 { 1969 int rc; 1970 1971 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1972 read_offset, read_len, pdu->dif_ctx); 1973 if (rc != 0) { 1974 SPDK_ERRLOG("DIF generate failed\n"); 1975 } 1976 1977 return rc; 1978 } 1979 1980 static int 1981 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1982 { 1983 int rc = 0; 1984 struct nvme_tcp_pdu *pdu; 1985 enum nvme_tcp_pdu_recv_state prev_state; 1986 uint32_t data_len; 1987 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1988 struct spdk_nvmf_tcp_transport, transport); 1989 1990 /* The loop here is to allow for several back-to-back state changes. */ 1991 do { 1992 prev_state = tqpair->recv_state; 1993 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1994 1995 pdu = tqpair->pdu_in_progress; 1996 switch (tqpair->recv_state) { 1997 /* Wait for the common header */ 1998 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1999 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2000 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2001 return rc; 2002 } 2003 2004 rc = nvme_tcp_read_data(tqpair->sock, 2005 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2006 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2007 if (rc < 0) { 2008 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2009 return NVME_TCP_PDU_FATAL; 2010 } else if (rc > 0) { 2011 pdu->ch_valid_bytes += rc; 2012 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, tqpair); 2013 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 2014 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2015 } 2016 } 2017 2018 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2019 return NVME_TCP_PDU_IN_PROGRESS; 2020 } 2021 2022 /* The command header of this PDU has now been read from the socket. */ 2023 nvmf_tcp_pdu_ch_handle(tqpair); 2024 break; 2025 /* Wait for the pdu specific header */ 2026 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2027 rc = nvme_tcp_read_data(tqpair->sock, 2028 pdu->psh_len - pdu->psh_valid_bytes, 2029 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2030 if (rc < 0) { 2031 return NVME_TCP_PDU_FATAL; 2032 } else if (rc > 0) { 2033 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, tqpair); 2034 pdu->psh_valid_bytes += rc; 2035 } 2036 2037 if (pdu->psh_valid_bytes < pdu->psh_len) { 2038 return NVME_TCP_PDU_IN_PROGRESS; 2039 } 2040 2041 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2042 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2043 break; 2044 /* Wait for the req slot */ 2045 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2046 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2047 break; 2048 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2049 /* check whether the data is valid, if not we just return */ 2050 if (!pdu->data_len) { 2051 return NVME_TCP_PDU_IN_PROGRESS; 2052 } 2053 2054 data_len = pdu->data_len; 2055 /* data digest */ 2056 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2057 tqpair->host_ddgst_enable)) { 2058 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2059 pdu->ddgst_enable = true; 2060 } 2061 2062 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2063 if (rc < 0) { 2064 return NVME_TCP_PDU_FATAL; 2065 } 2066 pdu->rw_offset += rc; 2067 2068 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2069 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2070 if (rc != 0) { 2071 return NVME_TCP_PDU_FATAL; 2072 } 2073 } 2074 2075 if (pdu->rw_offset < data_len) { 2076 return NVME_TCP_PDU_IN_PROGRESS; 2077 } 2078 2079 /* All of this PDU has now been read from the socket. */ 2080 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2081 break; 2082 case NVME_TCP_PDU_RECV_STATE_ERROR: 2083 if (!spdk_sock_is_connected(tqpair->sock)) { 2084 return NVME_TCP_PDU_FATAL; 2085 } 2086 break; 2087 default: 2088 assert(0); 2089 SPDK_ERRLOG("code should not come to here"); 2090 break; 2091 } 2092 } while (tqpair->recv_state != prev_state); 2093 2094 return rc; 2095 } 2096 2097 static inline void * 2098 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2099 { 2100 struct spdk_nvmf_tcp_control_msg *msg; 2101 2102 assert(list); 2103 2104 msg = STAILQ_FIRST(&list->free_msgs); 2105 if (!msg) { 2106 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2107 return NULL; 2108 } 2109 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2110 return msg; 2111 } 2112 2113 static inline void 2114 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2115 { 2116 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2117 2118 assert(list); 2119 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2120 } 2121 2122 static int 2123 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2124 struct spdk_nvmf_transport *transport, 2125 struct spdk_nvmf_transport_poll_group *group) 2126 { 2127 struct spdk_nvmf_request *req = &tcp_req->req; 2128 struct spdk_nvme_cmd *cmd; 2129 struct spdk_nvme_cpl *rsp; 2130 struct spdk_nvme_sgl_descriptor *sgl; 2131 struct spdk_nvmf_tcp_poll_group *tgroup; 2132 uint32_t length; 2133 2134 cmd = &req->cmd->nvme_cmd; 2135 rsp = &req->rsp->nvme_cpl; 2136 sgl = &cmd->dptr.sgl1; 2137 2138 length = sgl->unkeyed.length; 2139 2140 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2141 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2142 if (length > transport->opts.max_io_size) { 2143 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2144 length, transport->opts.max_io_size); 2145 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2146 return -1; 2147 } 2148 2149 /* fill request length and populate iovs */ 2150 req->length = length; 2151 2152 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2153 2154 if (spdk_unlikely(req->dif_enabled)) { 2155 req->dif.orig_length = length; 2156 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2157 req->dif.elba_length = length; 2158 } 2159 2160 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2161 /* No available buffers. Queue this request up. */ 2162 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2163 tcp_req); 2164 return 0; 2165 } 2166 2167 /* backward compatible */ 2168 req->data = req->iov[0].iov_base; 2169 2170 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2171 tcp_req, req->iovcnt, req->data); 2172 2173 return 0; 2174 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2175 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2176 uint64_t offset = sgl->address; 2177 uint32_t max_len = transport->opts.in_capsule_data_size; 2178 assert(tcp_req->has_incapsule_data); 2179 2180 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2181 offset, length); 2182 2183 if (offset > max_len) { 2184 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2185 offset, max_len); 2186 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2187 return -1; 2188 } 2189 max_len -= (uint32_t)offset; 2190 2191 if (spdk_unlikely(length > max_len)) { 2192 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2193 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2194 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2195 2196 /* Get a buffer from dedicated list */ 2197 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2198 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2199 assert(tgroup->control_msg_list); 2200 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2201 if (!req->data) { 2202 /* No available buffers. Queue this request up. */ 2203 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2204 return 0; 2205 } 2206 } else { 2207 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2208 length, max_len); 2209 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2210 return -1; 2211 } 2212 } else { 2213 req->data = tcp_req->buf; 2214 } 2215 2216 req->length = length; 2217 req->data_from_pool = false; 2218 2219 if (spdk_unlikely(req->dif_enabled)) { 2220 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2221 req->dif.elba_length = length; 2222 } 2223 2224 req->iov[0].iov_base = req->data; 2225 req->iov[0].iov_len = length; 2226 req->iovcnt = 1; 2227 2228 return 0; 2229 } 2230 2231 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2232 sgl->generic.type, sgl->generic.subtype); 2233 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2234 return -1; 2235 } 2236 2237 static inline enum spdk_nvme_media_error_status_code 2238 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2239 enum spdk_nvme_media_error_status_code result; 2240 2241 switch (err_type) 2242 { 2243 case SPDK_DIF_REFTAG_ERROR: 2244 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2245 break; 2246 case SPDK_DIF_APPTAG_ERROR: 2247 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2248 break; 2249 case SPDK_DIF_GUARD_ERROR: 2250 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2251 break; 2252 default: 2253 SPDK_UNREACHABLE(); 2254 break; 2255 } 2256 2257 return result; 2258 } 2259 2260 static void 2261 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2262 struct spdk_nvmf_tcp_req *tcp_req) 2263 { 2264 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2265 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2266 struct nvme_tcp_pdu *rsp_pdu; 2267 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2268 uint32_t plen, pdo, alignment; 2269 int rc; 2270 2271 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2272 2273 rsp_pdu = tcp_req->pdu; 2274 assert(rsp_pdu != NULL); 2275 assert(tcp_req->pdu_in_use); 2276 2277 c2h_data = &rsp_pdu->hdr.c2h_data; 2278 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2279 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2280 2281 if (tqpair->host_hdgst_enable) { 2282 plen += SPDK_NVME_TCP_DIGEST_LEN; 2283 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2284 } 2285 2286 /* set the psh */ 2287 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2288 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2289 c2h_data->datao = tcp_req->pdu->rw_offset; 2290 2291 /* set the padding */ 2292 rsp_pdu->padding_len = 0; 2293 pdo = plen; 2294 if (tqpair->cpda) { 2295 alignment = (tqpair->cpda + 1) << 2; 2296 if (plen % alignment != 0) { 2297 pdo = (plen + alignment) / alignment * alignment; 2298 rsp_pdu->padding_len = pdo - plen; 2299 plen = pdo; 2300 } 2301 } 2302 2303 c2h_data->common.pdo = pdo; 2304 plen += c2h_data->datal; 2305 if (tqpair->host_ddgst_enable) { 2306 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2307 plen += SPDK_NVME_TCP_DIGEST_LEN; 2308 } 2309 2310 c2h_data->common.plen = plen; 2311 2312 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2313 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2314 } 2315 2316 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2317 c2h_data->datao, c2h_data->datal); 2318 2319 2320 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2321 /* Need to send the capsule response if response is not all 0 */ 2322 if (ttransport->tcp_opts.c2h_success && 2323 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2324 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2325 } 2326 2327 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2328 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2329 struct spdk_dif_error err_blk = {}; 2330 uint32_t mapped_length = 0; 2331 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2332 uint32_t ddgst_len = 0; 2333 2334 if (tqpair->host_ddgst_enable) { 2335 /* Data digest consumes additional iov entry */ 2336 available_iovs--; 2337 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2338 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2339 c2h_data->common.plen -= ddgst_len; 2340 } 2341 /* Temp call to estimate if data can be described by limited number of iovs. 2342 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2343 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2344 false, &mapped_length); 2345 2346 if (mapped_length != c2h_data->common.plen) { 2347 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2348 SPDK_DEBUGLOG(nvmf_tcp, 2349 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2350 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2351 c2h_data->common.plen = mapped_length; 2352 2353 /* Rebuild pdu->data_iov since data length is changed */ 2354 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2355 c2h_data->datal); 2356 2357 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2358 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2359 } 2360 2361 c2h_data->common.plen += ddgst_len; 2362 2363 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2364 2365 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2366 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2367 if (rc != 0) { 2368 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2369 err_blk.err_type, err_blk.err_offset); 2370 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2371 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2372 nvmf_tcp_req_pdu_fini(tcp_req); 2373 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2374 return; 2375 } 2376 } 2377 2378 rsp_pdu->rw_offset += c2h_data->datal; 2379 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2380 } 2381 2382 static void 2383 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2384 struct spdk_nvmf_tcp_req *tcp_req) 2385 { 2386 nvmf_tcp_req_pdu_init(tcp_req); 2387 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2388 } 2389 2390 static int 2391 request_transfer_out(struct spdk_nvmf_request *req) 2392 { 2393 struct spdk_nvmf_tcp_req *tcp_req; 2394 struct spdk_nvmf_qpair *qpair; 2395 struct spdk_nvmf_tcp_qpair *tqpair; 2396 struct spdk_nvme_cpl *rsp; 2397 2398 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2399 2400 qpair = req->qpair; 2401 rsp = &req->rsp->nvme_cpl; 2402 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2403 2404 /* Advance our sq_head pointer */ 2405 if (qpair->sq_head == qpair->sq_head_max) { 2406 qpair->sq_head = 0; 2407 } else { 2408 qpair->sq_head++; 2409 } 2410 rsp->sqhd = qpair->sq_head; 2411 2412 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2413 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2414 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2415 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2416 } else { 2417 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2418 } 2419 2420 return 0; 2421 } 2422 2423 static void 2424 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2425 struct spdk_nvmf_tcp_req *tcp_req) 2426 { 2427 struct nvme_tcp_pdu *pdu; 2428 uint32_t plen = 0; 2429 2430 pdu = tqpair->pdu_in_progress; 2431 plen = pdu->hdr.common.hlen; 2432 2433 if (tqpair->host_hdgst_enable) { 2434 plen += SPDK_NVME_TCP_DIGEST_LEN; 2435 } 2436 2437 if (pdu->hdr.common.plen != plen) { 2438 tcp_req->has_incapsule_data = true; 2439 } 2440 } 2441 2442 static bool 2443 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2444 struct spdk_nvmf_tcp_req *tcp_req) 2445 { 2446 struct spdk_nvmf_tcp_qpair *tqpair; 2447 int rc; 2448 enum spdk_nvmf_tcp_req_state prev_state; 2449 bool progress = false; 2450 struct spdk_nvmf_transport *transport = &ttransport->transport; 2451 struct spdk_nvmf_transport_poll_group *group; 2452 struct spdk_nvmf_tcp_poll_group *tgroup; 2453 2454 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2455 group = &tqpair->group->group; 2456 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2457 2458 /* If the qpair is not active, we need to abort the outstanding requests. */ 2459 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2460 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2461 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2462 } 2463 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2464 } 2465 2466 /* The loop here is to allow for several back-to-back state changes. */ 2467 do { 2468 prev_state = tcp_req->state; 2469 2470 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2471 tqpair); 2472 2473 switch (tcp_req->state) { 2474 case TCP_REQUEST_STATE_FREE: 2475 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2476 * to escape this state. */ 2477 break; 2478 case TCP_REQUEST_STATE_NEW: 2479 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, tqpair); 2480 2481 /* copy the cmd from the receive pdu */ 2482 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2483 2484 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2485 tcp_req->req.dif_enabled = true; 2486 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2487 } 2488 2489 /* The next state transition depends on the data transfer needs of this request. */ 2490 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2491 2492 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2493 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2494 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 2495 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2496 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2497 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2498 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2499 break; 2500 } 2501 2502 /* If no data to transfer, ready to execute. */ 2503 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2504 /* Reset the tqpair receving pdu state */ 2505 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2506 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2507 break; 2508 } 2509 2510 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2511 2512 if (!tcp_req->has_incapsule_data) { 2513 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2514 } 2515 2516 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2517 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2518 break; 2519 case TCP_REQUEST_STATE_NEED_BUFFER: 2520 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, tqpair); 2521 2522 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2523 2524 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2525 SPDK_DEBUGLOG(nvmf_tcp, 2526 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2527 tcp_req, tqpair); 2528 /* This request needs to wait in line to obtain a buffer */ 2529 break; 2530 } 2531 2532 /* Try to get a data buffer */ 2533 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2534 if (rc < 0) { 2535 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2536 /* Reset the tqpair receving pdu state */ 2537 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2538 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2539 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2540 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2541 break; 2542 } 2543 2544 if (!tcp_req->req.data) { 2545 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2546 tcp_req, tqpair); 2547 /* No buffers available. */ 2548 break; 2549 } 2550 2551 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2552 2553 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2554 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2555 if (tcp_req->req.data_from_pool) { 2556 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2557 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2558 } else { 2559 struct nvme_tcp_pdu *pdu; 2560 2561 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2562 2563 pdu = tqpair->pdu_in_progress; 2564 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2565 tqpair); 2566 /* No need to send r2t, contained in the capsuled data */ 2567 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2568 0, tcp_req->req.length); 2569 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2570 } 2571 break; 2572 } 2573 2574 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2575 break; 2576 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2577 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 2578 tqpair); 2579 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2580 break; 2581 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2582 2583 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2584 (uintptr_t)tcp_req, tqpair); 2585 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2586 * to escape this state. */ 2587 break; 2588 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2589 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 2590 tqpair); 2591 2592 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2593 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2594 tcp_req->req.length = tcp_req->req.dif.elba_length; 2595 } 2596 2597 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2598 spdk_nvmf_request_exec(&tcp_req->req); 2599 break; 2600 case TCP_REQUEST_STATE_EXECUTING: 2601 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, tqpair); 2602 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2603 * to escape this state. */ 2604 break; 2605 case TCP_REQUEST_STATE_EXECUTED: 2606 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, tqpair); 2607 2608 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2609 tcp_req->req.length = tcp_req->req.dif.orig_length; 2610 } 2611 2612 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2613 break; 2614 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2615 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 2616 tqpair); 2617 rc = request_transfer_out(&tcp_req->req); 2618 assert(rc == 0); /* No good way to handle this currently */ 2619 break; 2620 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2621 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2622 (uintptr_t)tcp_req, tqpair); 2623 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2624 * to escape this state. */ 2625 break; 2626 case TCP_REQUEST_STATE_COMPLETED: 2627 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, tqpair); 2628 if (tcp_req->req.data_from_pool) { 2629 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2630 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2631 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2632 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2633 assert(tgroup->control_msg_list); 2634 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2635 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2636 } 2637 tcp_req->req.length = 0; 2638 tcp_req->req.iovcnt = 0; 2639 tcp_req->req.data = NULL; 2640 2641 nvmf_tcp_req_pdu_fini(tcp_req); 2642 2643 nvmf_tcp_req_put(tqpair, tcp_req); 2644 break; 2645 case TCP_REQUEST_NUM_STATES: 2646 default: 2647 assert(0); 2648 break; 2649 } 2650 2651 if (tcp_req->state != prev_state) { 2652 progress = true; 2653 } 2654 } while (tcp_req->state != prev_state); 2655 2656 return progress; 2657 } 2658 2659 static void 2660 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2661 { 2662 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2663 int rc; 2664 2665 assert(tqpair != NULL); 2666 rc = nvmf_tcp_sock_process(tqpair); 2667 2668 /* If there was a new socket error, disconnect */ 2669 if (rc < 0) { 2670 nvmf_tcp_qpair_disconnect(tqpair); 2671 } 2672 } 2673 2674 static int 2675 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2676 struct spdk_nvmf_qpair *qpair) 2677 { 2678 struct spdk_nvmf_tcp_poll_group *tgroup; 2679 struct spdk_nvmf_tcp_qpair *tqpair; 2680 int rc; 2681 2682 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2683 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2684 2685 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2686 nvmf_tcp_sock_cb, tqpair); 2687 if (rc != 0) { 2688 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2689 spdk_strerror(errno), errno); 2690 return -1; 2691 } 2692 2693 rc = nvmf_tcp_qpair_sock_init(tqpair); 2694 if (rc != 0) { 2695 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2696 return -1; 2697 } 2698 2699 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2700 if (rc < 0) { 2701 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2702 return -1; 2703 } 2704 2705 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2706 if (rc < 0) { 2707 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2708 return -1; 2709 } 2710 2711 tqpair->group = tgroup; 2712 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 2713 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2714 2715 return 0; 2716 } 2717 2718 static int 2719 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2720 struct spdk_nvmf_qpair *qpair) 2721 { 2722 struct spdk_nvmf_tcp_poll_group *tgroup; 2723 struct spdk_nvmf_tcp_qpair *tqpair; 2724 int rc; 2725 2726 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2727 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2728 2729 assert(tqpair->group == tgroup); 2730 2731 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2732 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2733 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2734 } else { 2735 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2736 } 2737 2738 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2739 if (rc != 0) { 2740 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2741 spdk_strerror(errno), errno); 2742 } 2743 2744 return rc; 2745 } 2746 2747 static int 2748 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2749 { 2750 struct spdk_nvmf_tcp_transport *ttransport; 2751 struct spdk_nvmf_tcp_req *tcp_req; 2752 2753 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2754 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2755 2756 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2757 nvmf_tcp_req_process(ttransport, tcp_req); 2758 2759 return 0; 2760 } 2761 2762 static void 2763 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2764 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2765 { 2766 struct spdk_nvmf_tcp_qpair *tqpair; 2767 2768 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2769 2770 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2771 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 2772 nvmf_tcp_qpair_destroy(tqpair); 2773 2774 if (cb_fn) { 2775 cb_fn(cb_arg); 2776 } 2777 } 2778 2779 static int 2780 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2781 { 2782 struct spdk_nvmf_tcp_poll_group *tgroup; 2783 int rc; 2784 struct spdk_nvmf_request *req, *req_tmp; 2785 struct spdk_nvmf_tcp_req *tcp_req; 2786 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2787 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2788 struct spdk_nvmf_tcp_transport, transport); 2789 2790 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2791 2792 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2793 return 0; 2794 } 2795 2796 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2797 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2798 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2799 break; 2800 } 2801 } 2802 2803 rc = spdk_sock_group_poll(tgroup->sock_group); 2804 if (rc < 0) { 2805 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2806 } 2807 2808 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2809 nvmf_tcp_sock_process(tqpair); 2810 } 2811 2812 return rc; 2813 } 2814 2815 static int 2816 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2817 struct spdk_nvme_transport_id *trid, bool peer) 2818 { 2819 struct spdk_nvmf_tcp_qpair *tqpair; 2820 uint16_t port; 2821 2822 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2823 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2824 2825 if (peer) { 2826 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2827 port = tqpair->initiator_port; 2828 } else { 2829 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2830 port = tqpair->target_port; 2831 } 2832 2833 if (spdk_sock_is_ipv4(tqpair->sock)) { 2834 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2835 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2836 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2837 } else { 2838 return -1; 2839 } 2840 2841 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2842 return 0; 2843 } 2844 2845 static int 2846 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2847 struct spdk_nvme_transport_id *trid) 2848 { 2849 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2850 } 2851 2852 static int 2853 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2854 struct spdk_nvme_transport_id *trid) 2855 { 2856 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2857 } 2858 2859 static int 2860 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2861 struct spdk_nvme_transport_id *trid) 2862 { 2863 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2864 } 2865 2866 static void 2867 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2868 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2869 { 2870 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2871 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2872 2873 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2874 2875 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2876 } 2877 2878 static int 2879 _nvmf_tcp_qpair_abort_request(void *ctx) 2880 { 2881 struct spdk_nvmf_request *req = ctx; 2882 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2883 struct spdk_nvmf_tcp_req, req); 2884 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2885 struct spdk_nvmf_tcp_qpair, qpair); 2886 int rc; 2887 2888 spdk_poller_unregister(&req->poller); 2889 2890 switch (tcp_req_to_abort->state) { 2891 case TCP_REQUEST_STATE_EXECUTING: 2892 rc = nvmf_ctrlr_abort_request(req); 2893 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2894 return SPDK_POLLER_BUSY; 2895 } 2896 break; 2897 2898 case TCP_REQUEST_STATE_NEED_BUFFER: 2899 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2900 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2901 2902 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2903 break; 2904 2905 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2906 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2907 break; 2908 2909 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2910 if (spdk_get_ticks() < req->timeout_tsc) { 2911 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2912 return SPDK_POLLER_BUSY; 2913 } 2914 break; 2915 2916 default: 2917 break; 2918 } 2919 2920 spdk_nvmf_request_complete(req); 2921 return SPDK_POLLER_BUSY; 2922 } 2923 2924 static void 2925 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2926 struct spdk_nvmf_request *req) 2927 { 2928 struct spdk_nvmf_tcp_qpair *tqpair; 2929 struct spdk_nvmf_tcp_transport *ttransport; 2930 struct spdk_nvmf_transport *transport; 2931 uint16_t cid; 2932 uint32_t i; 2933 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2934 2935 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2936 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2937 transport = &ttransport->transport; 2938 2939 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2940 2941 for (i = 0; i < tqpair->resource_count; i++) { 2942 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2943 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2944 tcp_req_to_abort = &tqpair->reqs[i]; 2945 break; 2946 } 2947 } 2948 2949 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, 0, 0, (uintptr_t)req, tqpair); 2950 2951 if (tcp_req_to_abort == NULL) { 2952 spdk_nvmf_request_complete(req); 2953 return; 2954 } 2955 2956 req->req_to_abort = &tcp_req_to_abort->req; 2957 req->timeout_tsc = spdk_get_ticks() + 2958 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2959 req->poller = NULL; 2960 2961 _nvmf_tcp_qpair_abort_request(req); 2962 } 2963 2964 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2965 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2966 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2967 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2968 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2969 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2970 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2971 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2972 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2973 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2974 2975 static void 2976 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2977 { 2978 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2979 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2980 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2981 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2982 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2983 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2984 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2985 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2986 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2987 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2988 opts->transport_specific = NULL; 2989 } 2990 2991 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2992 .name = "TCP", 2993 .type = SPDK_NVME_TRANSPORT_TCP, 2994 .opts_init = nvmf_tcp_opts_init, 2995 .create = nvmf_tcp_create, 2996 .dump_opts = nvmf_tcp_dump_opts, 2997 .destroy = nvmf_tcp_destroy, 2998 2999 .listen = nvmf_tcp_listen, 3000 .stop_listen = nvmf_tcp_stop_listen, 3001 .accept = nvmf_tcp_accept, 3002 3003 .listener_discover = nvmf_tcp_discover, 3004 3005 .poll_group_create = nvmf_tcp_poll_group_create, 3006 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3007 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3008 .poll_group_add = nvmf_tcp_poll_group_add, 3009 .poll_group_remove = nvmf_tcp_poll_group_remove, 3010 .poll_group_poll = nvmf_tcp_poll_group_poll, 3011 3012 .req_free = nvmf_tcp_req_free, 3013 .req_complete = nvmf_tcp_req_complete, 3014 3015 .qpair_fini = nvmf_tcp_close_qpair, 3016 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3017 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3018 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3019 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3020 }; 3021 3022 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3023 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3024