1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/accel_engine.h" 35 #include "spdk/stdinc.h" 36 #include "spdk/crc32.h" 37 #include "spdk/endian.h" 38 #include "spdk/assert.h" 39 #include "spdk/thread.h" 40 #include "spdk/nvmf_transport.h" 41 #include "spdk/string.h" 42 #include "spdk/trace.h" 43 #include "spdk/util.h" 44 #include "spdk/log.h" 45 46 #include "spdk_internal/assert.h" 47 #include "spdk_internal/nvme_tcp.h" 48 #include "spdk_internal/sock.h" 49 50 #include "nvmf_internal.h" 51 52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 54 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 55 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 56 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 57 58 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 59 60 /* spdk nvmf related structure */ 61 enum spdk_nvmf_tcp_req_state { 62 63 /* The request is not currently in use */ 64 TCP_REQUEST_STATE_FREE = 0, 65 66 /* Initial state when request first received */ 67 TCP_REQUEST_STATE_NEW, 68 69 /* The request is queued until a data buffer is available. */ 70 TCP_REQUEST_STATE_NEED_BUFFER, 71 72 /* The request is currently transferring data from the host to the controller. */ 73 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 74 75 /* The request is waiting for the R2T send acknowledgement. */ 76 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 77 78 /* The request is ready to execute at the block device */ 79 TCP_REQUEST_STATE_READY_TO_EXECUTE, 80 81 /* The request is currently executing at the block device */ 82 TCP_REQUEST_STATE_EXECUTING, 83 84 /* The request finished executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTED, 86 87 /* The request is ready to send a completion */ 88 TCP_REQUEST_STATE_READY_TO_COMPLETE, 89 90 /* The request is currently transferring final pdus from the controller to the host. */ 91 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 92 93 /* The request completed and can be marked free. */ 94 TCP_REQUEST_STATE_COMPLETED, 95 96 /* Terminator */ 97 TCP_REQUEST_NUM_STATES, 98 }; 99 100 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 101 "Invalid PDU Header Field", 102 "PDU Sequence Error", 103 "Header Digiest Error", 104 "Data Transfer Out of Range", 105 "R2T Limit Exceeded", 106 "Unsupported parameter", 107 }; 108 109 #define OBJECT_NVMF_TCP_IO 0x80 110 111 #define TRACE_GROUP_NVMF_TCP 0x5 112 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 113 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 115 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 117 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 118 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 119 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 120 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 121 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 122 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 123 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 124 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 132 SPDK_TRACE_ARG_TYPE_INT, ""); 133 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 134 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 135 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 136 SPDK_TRACE_ARG_TYPE_INT, ""); 137 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 138 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 139 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 140 SPDK_TRACE_ARG_TYPE_INT, ""); 141 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 142 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 144 SPDK_TRACE_ARG_TYPE_INT, ""); 145 spdk_trace_register_description("TCP_REQ_EXECUTING", 146 TRACE_TCP_REQUEST_STATE_EXECUTING, 147 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 148 SPDK_TRACE_ARG_TYPE_INT, ""); 149 spdk_trace_register_description("TCP_REQ_EXECUTED", 150 TRACE_TCP_REQUEST_STATE_EXECUTED, 151 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 152 SPDK_TRACE_ARG_TYPE_INT, ""); 153 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 154 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 156 SPDK_TRACE_ARG_TYPE_INT, ""); 157 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 158 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 159 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 160 SPDK_TRACE_ARG_TYPE_INT, ""); 161 spdk_trace_register_description("TCP_REQ_COMPLETED", 162 TRACE_TCP_REQUEST_STATE_COMPLETED, 163 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 164 SPDK_TRACE_ARG_TYPE_INT, ""); 165 spdk_trace_register_description("TCP_WRITE_START", 166 TRACE_TCP_FLUSH_WRITEBUF_START, 167 OWNER_NONE, OBJECT_NONE, 0, 168 SPDK_TRACE_ARG_TYPE_INT, ""); 169 spdk_trace_register_description("TCP_WRITE_DONE", 170 TRACE_TCP_FLUSH_WRITEBUF_DONE, 171 OWNER_NONE, OBJECT_NONE, 0, 172 SPDK_TRACE_ARG_TYPE_INT, ""); 173 spdk_trace_register_description("TCP_READ_DONE", 174 TRACE_TCP_READ_FROM_SOCKET_DONE, 175 OWNER_NONE, OBJECT_NONE, 0, 176 SPDK_TRACE_ARG_TYPE_INT, ""); 177 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 178 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 179 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 180 SPDK_TRACE_ARG_TYPE_INT, ""); 181 } 182 183 struct spdk_nvmf_tcp_req { 184 struct spdk_nvmf_request req; 185 struct spdk_nvme_cpl rsp; 186 struct spdk_nvme_cmd cmd; 187 188 /* A PDU that can be used for sending responses. This is 189 * not the incoming PDU! */ 190 struct nvme_tcp_pdu *pdu; 191 192 /* In-capsule data buffer */ 193 uint8_t *buf; 194 /* 195 * The PDU for a request may be used multiple times in serial over 196 * the request's lifetime. For example, first to send an R2T, then 197 * to send a completion. To catch mistakes where the PDU is used 198 * twice at the same time, add a debug flag here for init/fini. 199 */ 200 bool pdu_in_use; 201 bool has_incapsule_data; 202 203 /* transfer_tag */ 204 uint16_t ttag; 205 206 enum spdk_nvmf_tcp_req_state state; 207 208 /* 209 * h2c_offset is used when we receive the h2c_data PDU. 210 */ 211 uint32_t h2c_offset; 212 213 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 214 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 215 }; 216 217 struct spdk_nvmf_tcp_qpair { 218 struct spdk_nvmf_qpair qpair; 219 struct spdk_nvmf_tcp_poll_group *group; 220 struct spdk_sock *sock; 221 222 enum nvme_tcp_pdu_recv_state recv_state; 223 enum nvme_tcp_qpair_state state; 224 225 /* PDU being actively received */ 226 struct nvme_tcp_pdu *pdu_in_progress; 227 228 /* Queues to track the requests in all states */ 229 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 230 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 231 232 /* Number of requests in each state */ 233 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 234 235 uint8_t cpda; 236 237 bool host_hdgst_enable; 238 bool host_ddgst_enable; 239 240 /* This is a spare PDU used for sending special management 241 * operations. Primarily, this is used for the initial 242 * connection response and c2h termination request. */ 243 struct nvme_tcp_pdu *mgmt_pdu; 244 245 /* Arrays of in-capsule buffers, requests, and pdus. 246 * Each array is 'resource_count' number of elements */ 247 void *bufs; 248 struct spdk_nvmf_tcp_req *reqs; 249 struct nvme_tcp_pdu *pdus; 250 uint32_t resource_count; 251 uint32_t recv_buf_size; 252 253 struct spdk_nvmf_tcp_port *port; 254 255 /* IP address */ 256 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 257 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 258 259 /* IP port */ 260 uint16_t initiator_port; 261 uint16_t target_port; 262 263 /* Timer used to destroy qpair after detecting transport error issue if initiator does 264 * not close the connection. 265 */ 266 struct spdk_poller *timeout_poller; 267 268 269 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 270 }; 271 272 struct spdk_nvmf_tcp_control_msg { 273 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 274 }; 275 276 struct spdk_nvmf_tcp_control_msg_list { 277 void *msg_buf; 278 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 279 }; 280 281 struct spdk_nvmf_tcp_poll_group { 282 struct spdk_nvmf_transport_poll_group group; 283 struct spdk_sock_group *sock_group; 284 285 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 286 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 287 288 struct spdk_io_channel *accel_channel; 289 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 290 }; 291 292 struct spdk_nvmf_tcp_port { 293 const struct spdk_nvme_transport_id *trid; 294 struct spdk_sock *listen_sock; 295 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 296 }; 297 298 struct tcp_transport_opts { 299 bool c2h_success; 300 uint16_t control_msg_num; 301 uint32_t sock_priority; 302 }; 303 304 struct spdk_nvmf_tcp_transport { 305 struct spdk_nvmf_transport transport; 306 struct tcp_transport_opts tcp_opts; 307 308 pthread_mutex_t lock; 309 310 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 311 }; 312 313 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 314 { 315 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 316 spdk_json_decode_bool, true 317 }, 318 { 319 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 320 spdk_json_decode_uint16, true 321 }, 322 { 323 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 324 spdk_json_decode_uint32, true 325 }, 326 }; 327 328 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 329 struct spdk_nvmf_tcp_req *tcp_req); 330 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 331 332 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 333 struct spdk_nvmf_tcp_req *tcp_req); 334 335 static void 336 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 337 enum spdk_nvmf_tcp_req_state state) 338 { 339 struct spdk_nvmf_qpair *qpair; 340 struct spdk_nvmf_tcp_qpair *tqpair; 341 342 qpair = tcp_req->req.qpair; 343 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 344 345 assert(tqpair->state_cntr[tcp_req->state] > 0); 346 tqpair->state_cntr[tcp_req->state]--; 347 tqpair->state_cntr[state]++; 348 349 tcp_req->state = state; 350 } 351 352 static inline struct nvme_tcp_pdu * 353 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 354 { 355 assert(tcp_req->pdu_in_use == false); 356 tcp_req->pdu_in_use = true; 357 358 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 359 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 360 361 return tcp_req->pdu; 362 } 363 364 static inline void 365 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 366 { 367 tcp_req->pdu_in_use = false; 368 } 369 370 static struct spdk_nvmf_tcp_req * 371 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 372 { 373 struct spdk_nvmf_tcp_req *tcp_req; 374 375 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 376 if (!tcp_req) { 377 return NULL; 378 } 379 380 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 381 tcp_req->h2c_offset = 0; 382 tcp_req->has_incapsule_data = false; 383 tcp_req->req.dif.dif_insert_or_strip = false; 384 385 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 386 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 387 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 388 return tcp_req; 389 } 390 391 static inline void 392 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 393 { 394 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 395 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 396 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 397 } 398 399 static void 400 nvmf_tcp_request_free(void *cb_arg) 401 { 402 struct spdk_nvmf_tcp_transport *ttransport; 403 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 404 405 assert(tcp_req != NULL); 406 407 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 408 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 409 struct spdk_nvmf_tcp_transport, transport); 410 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 411 nvmf_tcp_req_process(ttransport, tcp_req); 412 } 413 414 static int 415 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 416 { 417 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 418 419 nvmf_tcp_request_free(tcp_req); 420 421 return 0; 422 } 423 424 static void 425 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 426 enum spdk_nvmf_tcp_req_state state) 427 { 428 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 429 430 assert(state != TCP_REQUEST_STATE_FREE); 431 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 432 if (state == tcp_req->state) { 433 nvmf_tcp_request_free(tcp_req); 434 } 435 } 436 } 437 438 static void 439 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 440 { 441 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 442 443 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 444 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 445 446 /* Wipe the requests waiting for buffer from the global list */ 447 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 448 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 449 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 450 spdk_nvmf_request, buf_link); 451 } 452 } 453 454 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 455 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 456 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 457 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 458 } 459 460 static void 461 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 462 { 463 int i; 464 struct spdk_nvmf_tcp_req *tcp_req; 465 466 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 467 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 468 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 469 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 470 if ((int)tcp_req->state == i) { 471 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 472 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 473 } 474 } 475 } 476 } 477 478 static void 479 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 480 { 481 int err = 0; 482 483 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 484 485 err = spdk_sock_close(&tqpair->sock); 486 assert(err == 0); 487 nvmf_tcp_cleanup_all_states(tqpair); 488 489 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 490 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 491 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 492 tqpair->resource_count); 493 err++; 494 } 495 496 if (err > 0) { 497 nvmf_tcp_dump_qpair_req_contents(tqpair); 498 } 499 500 spdk_dma_free(tqpair->pdus); 501 free(tqpair->reqs); 502 spdk_free(tqpair->bufs); 503 free(tqpair); 504 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 505 } 506 507 static void 508 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 509 { 510 struct spdk_nvmf_tcp_transport *ttransport; 511 assert(w != NULL); 512 513 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 514 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 515 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 516 } 517 518 static int 519 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 520 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 521 { 522 struct spdk_nvmf_tcp_transport *ttransport; 523 524 assert(transport != NULL); 525 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 526 527 pthread_mutex_destroy(&ttransport->lock); 528 free(ttransport); 529 530 if (cb_fn) { 531 cb_fn(cb_arg); 532 } 533 return 0; 534 } 535 536 static struct spdk_nvmf_transport * 537 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 538 { 539 struct spdk_nvmf_tcp_transport *ttransport; 540 uint32_t sge_count; 541 uint32_t min_shared_buffers; 542 543 ttransport = calloc(1, sizeof(*ttransport)); 544 if (!ttransport) { 545 return NULL; 546 } 547 548 TAILQ_INIT(&ttransport->ports); 549 550 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 551 552 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 553 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 554 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 555 if (opts->transport_specific != NULL && 556 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 557 SPDK_COUNTOF(tcp_transport_opts_decoder), 558 &ttransport->tcp_opts)) { 559 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 560 free(ttransport); 561 return NULL; 562 } 563 564 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 565 566 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 567 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 568 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 569 " in_capsule_data_size=%d, max_aq_depth=%d\n" 570 " num_shared_buffers=%d, c2h_success=%d,\n" 571 " dif_insert_or_strip=%d, sock_priority=%d\n" 572 " abort_timeout_sec=%d, control_msg_num=%hu\n", 573 opts->max_queue_depth, 574 opts->max_io_size, 575 opts->max_qpairs_per_ctrlr - 1, 576 opts->io_unit_size, 577 opts->in_capsule_data_size, 578 opts->max_aq_depth, 579 opts->num_shared_buffers, 580 ttransport->tcp_opts.c2h_success, 581 opts->dif_insert_or_strip, 582 ttransport->tcp_opts.sock_priority, 583 opts->abort_timeout_sec, 584 ttransport->tcp_opts.control_msg_num); 585 586 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 587 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 588 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 589 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 590 free(ttransport); 591 return NULL; 592 } 593 594 if (ttransport->tcp_opts.control_msg_num == 0 && 595 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 596 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 597 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 598 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 599 } 600 601 /* I/O unit size cannot be larger than max I/O size */ 602 if (opts->io_unit_size > opts->max_io_size) { 603 opts->io_unit_size = opts->max_io_size; 604 } 605 606 sge_count = opts->max_io_size / opts->io_unit_size; 607 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 608 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 609 free(ttransport); 610 return NULL; 611 } 612 613 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 614 if (min_shared_buffers > opts->num_shared_buffers) { 615 SPDK_ERRLOG("There are not enough buffers to satisfy" 616 "per-poll group caches for each thread. (%" PRIu32 ")" 617 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 618 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 619 free(ttransport); 620 return NULL; 621 } 622 623 pthread_mutex_init(&ttransport->lock, NULL); 624 625 return &ttransport->transport; 626 } 627 628 static int 629 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 630 { 631 unsigned long long ull; 632 char *end = NULL; 633 634 ull = strtoull(trsvcid, &end, 10); 635 if (end == NULL || end == trsvcid || *end != '\0') { 636 return -1; 637 } 638 639 /* Valid TCP/IP port numbers are in [0, 65535] */ 640 if (ull > 65535) { 641 return -1; 642 } 643 644 return (int)ull; 645 } 646 647 /** 648 * Canonicalize a listen address trid. 649 */ 650 static int 651 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 652 const struct spdk_nvme_transport_id *trid) 653 { 654 int trsvcid_int; 655 656 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 657 if (trsvcid_int < 0) { 658 return -EINVAL; 659 } 660 661 memset(canon_trid, 0, sizeof(*canon_trid)); 662 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 663 canon_trid->adrfam = trid->adrfam; 664 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 665 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 666 667 return 0; 668 } 669 670 /** 671 * Find an existing listening port. 672 * 673 * Caller must hold ttransport->lock. 674 */ 675 static struct spdk_nvmf_tcp_port * 676 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 677 const struct spdk_nvme_transport_id *trid) 678 { 679 struct spdk_nvme_transport_id canon_trid; 680 struct spdk_nvmf_tcp_port *port; 681 682 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 683 return NULL; 684 } 685 686 TAILQ_FOREACH(port, &ttransport->ports, link) { 687 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 688 return port; 689 } 690 } 691 692 return NULL; 693 } 694 695 static int 696 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 697 struct spdk_nvmf_listen_opts *listen_opts) 698 { 699 struct spdk_nvmf_tcp_transport *ttransport; 700 struct spdk_nvmf_tcp_port *port; 701 int trsvcid_int; 702 uint8_t adrfam; 703 struct spdk_sock_opts opts; 704 705 if (!strlen(trid->trsvcid)) { 706 SPDK_ERRLOG("Service id is required\n"); 707 return -EINVAL; 708 } 709 710 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 711 712 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 713 if (trsvcid_int < 0) { 714 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 715 return -EINVAL; 716 } 717 718 pthread_mutex_lock(&ttransport->lock); 719 port = calloc(1, sizeof(*port)); 720 if (!port) { 721 SPDK_ERRLOG("Port allocation failed\n"); 722 pthread_mutex_unlock(&ttransport->lock); 723 return -ENOMEM; 724 } 725 726 port->trid = trid; 727 opts.opts_size = sizeof(opts); 728 spdk_sock_get_default_opts(&opts); 729 opts.priority = ttransport->tcp_opts.sock_priority; 730 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 731 NULL, &opts); 732 if (port->listen_sock == NULL) { 733 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 734 trid->traddr, trsvcid_int, 735 spdk_strerror(errno), errno); 736 free(port); 737 pthread_mutex_unlock(&ttransport->lock); 738 return -errno; 739 } 740 741 if (spdk_sock_is_ipv4(port->listen_sock)) { 742 adrfam = SPDK_NVMF_ADRFAM_IPV4; 743 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 744 adrfam = SPDK_NVMF_ADRFAM_IPV6; 745 } else { 746 SPDK_ERRLOG("Unhandled socket type\n"); 747 adrfam = 0; 748 } 749 750 if (adrfam != trid->adrfam) { 751 SPDK_ERRLOG("Socket address family mismatch\n"); 752 spdk_sock_close(&port->listen_sock); 753 free(port); 754 pthread_mutex_unlock(&ttransport->lock); 755 return -EINVAL; 756 } 757 758 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 759 trid->traddr, trid->trsvcid); 760 761 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 762 pthread_mutex_unlock(&ttransport->lock); 763 return 0; 764 } 765 766 static void 767 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 768 const struct spdk_nvme_transport_id *trid) 769 { 770 struct spdk_nvmf_tcp_transport *ttransport; 771 struct spdk_nvmf_tcp_port *port; 772 773 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 774 775 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 776 trid->traddr, trid->trsvcid); 777 778 pthread_mutex_lock(&ttransport->lock); 779 port = nvmf_tcp_find_port(ttransport, trid); 780 if (port) { 781 TAILQ_REMOVE(&ttransport->ports, port, link); 782 spdk_sock_close(&port->listen_sock); 783 free(port); 784 } 785 786 pthread_mutex_unlock(&ttransport->lock); 787 } 788 789 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 790 enum nvme_tcp_pdu_recv_state state); 791 792 static void 793 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 794 { 795 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 796 797 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 798 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 799 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 800 spdk_poller_unregister(&tqpair->timeout_poller); 801 802 /* This will end up calling nvmf_tcp_close_qpair */ 803 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 804 } 805 } 806 807 static void 808 _pdu_write_done(void *_pdu, int err) 809 { 810 struct nvme_tcp_pdu *pdu = _pdu; 811 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 812 813 if (err != 0) { 814 nvmf_tcp_qpair_disconnect(tqpair); 815 return; 816 } 817 818 assert(pdu->cb_fn != NULL); 819 pdu->cb_fn(pdu->cb_arg); 820 } 821 822 static void 823 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 824 { 825 uint32_t mapped_length = 0; 826 ssize_t rc; 827 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 828 829 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 830 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 831 &mapped_length); 832 pdu->sock_req.cb_fn = _pdu_write_done; 833 pdu->sock_req.cb_arg = pdu; 834 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 835 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 836 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 837 if (rc == mapped_length) { 838 _pdu_write_done(pdu, 0); 839 } else { 840 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 841 _pdu_write_done(pdu, -1); 842 } 843 } else { 844 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 845 } 846 } 847 848 static void 849 data_crc32_accel_done(void *cb_arg, int status) 850 { 851 struct nvme_tcp_pdu *pdu = cb_arg; 852 853 if (spdk_unlikely(status)) { 854 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 855 _pdu_write_done(pdu, status); 856 return; 857 } 858 859 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 860 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 861 862 _tcp_write_pdu(pdu); 863 } 864 865 static void 866 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 867 { 868 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 869 uint32_t crc32c; 870 871 /* Data Digest */ 872 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 873 /* Only suport this limitated case for the first step */ 874 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 875 && tqpair->group)) { 876 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, 877 pdu->data_iov, pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 878 return; 879 } 880 881 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 882 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 883 } 884 885 _tcp_write_pdu(pdu); 886 } 887 888 static void 889 header_crc32_accel_done(void *cb_arg, int status) 890 { 891 struct nvme_tcp_pdu *pdu = cb_arg; 892 893 pdu->header_digest_crc32 ^= SPDK_CRC32C_XOR; 894 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, pdu->header_digest_crc32); 895 if (spdk_unlikely(status)) { 896 SPDK_ERRLOG("Failed to compute header digest on pdu=%p\n", pdu); 897 _pdu_write_done(pdu, status); 898 return; 899 } 900 901 pdu_data_crc32_compute(pdu); 902 } 903 904 static void 905 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 906 struct nvme_tcp_pdu *pdu, 907 nvme_tcp_qpair_xfer_complete_cb cb_fn, 908 void *cb_arg) 909 { 910 int hlen; 911 912 assert(tqpair->pdu_in_progress != pdu); 913 914 hlen = pdu->hdr.common.hlen; 915 pdu->cb_fn = cb_fn; 916 pdu->cb_arg = cb_arg; 917 918 pdu->iov[0].iov_base = &pdu->hdr.raw; 919 pdu->iov[0].iov_len = hlen; 920 921 /* Header Digest */ 922 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable && tqpair->group) { 923 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->header_digest_crc32, 924 pdu->iov, 1, 0, header_crc32_accel_done, pdu); 925 return; 926 } 927 928 pdu_data_crc32_compute(pdu); 929 } 930 931 static int 932 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 933 { 934 uint32_t i; 935 struct spdk_nvmf_transport_opts *opts; 936 uint32_t in_capsule_data_size; 937 938 opts = &tqpair->qpair.transport->opts; 939 940 in_capsule_data_size = opts->in_capsule_data_size; 941 if (opts->dif_insert_or_strip) { 942 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 943 } 944 945 tqpair->resource_count = opts->max_queue_depth; 946 947 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 948 if (!tqpair->reqs) { 949 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 950 return -1; 951 } 952 953 if (in_capsule_data_size) { 954 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 955 NULL, SPDK_ENV_LCORE_ID_ANY, 956 SPDK_MALLOC_DMA); 957 if (!tqpair->bufs) { 958 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 959 return -1; 960 } 961 } 962 963 /* Add addtional 2 members, which will be used for mgmt_pdu and pdu_in_progress owned by the tqpair */ 964 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 2) * sizeof(*tqpair->pdus), 0x1000, NULL); 965 if (!tqpair->pdus) { 966 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 967 return -1; 968 } 969 970 for (i = 0; i < tqpair->resource_count; i++) { 971 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 972 973 tcp_req->ttag = i + 1; 974 tcp_req->req.qpair = &tqpair->qpair; 975 976 tcp_req->pdu = &tqpair->pdus[i]; 977 tcp_req->pdu->qpair = tqpair; 978 979 /* Set up memory to receive commands */ 980 if (tqpair->bufs) { 981 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 982 } 983 984 /* Set the cmdn and rsp */ 985 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 986 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 987 988 /* Initialize request state to FREE */ 989 tcp_req->state = TCP_REQUEST_STATE_FREE; 990 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 991 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 992 } 993 994 tqpair->mgmt_pdu = &tqpair->pdus[i]; 995 tqpair->mgmt_pdu->qpair = tqpair; 996 tqpair->pdu_in_progress = &tqpair->pdus[i + 1]; 997 998 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 999 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1000 1001 return 0; 1002 } 1003 1004 static int 1005 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1006 { 1007 struct spdk_nvmf_tcp_qpair *tqpair; 1008 1009 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1010 1011 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1012 1013 /* Initialise request state queues of the qpair */ 1014 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1015 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1016 1017 tqpair->host_hdgst_enable = true; 1018 tqpair->host_ddgst_enable = true; 1019 1020 return 0; 1021 } 1022 1023 static int 1024 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1025 { 1026 int rc; 1027 1028 /* set low water mark */ 1029 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1030 if (rc != 0) { 1031 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1032 return rc; 1033 } 1034 1035 return 0; 1036 } 1037 1038 static void 1039 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1040 struct spdk_nvmf_tcp_port *port, 1041 struct spdk_sock *sock) 1042 { 1043 struct spdk_nvmf_tcp_qpair *tqpair; 1044 int rc; 1045 1046 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1047 port->trid->traddr, port->trid->trsvcid); 1048 1049 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1050 if (tqpair == NULL) { 1051 SPDK_ERRLOG("Could not allocate new connection.\n"); 1052 spdk_sock_close(&sock); 1053 return; 1054 } 1055 1056 tqpair->sock = sock; 1057 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1058 tqpair->port = port; 1059 tqpair->qpair.transport = transport; 1060 1061 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1062 sizeof(tqpair->target_addr), &tqpair->target_port, 1063 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1064 &tqpair->initiator_port); 1065 if (rc < 0) { 1066 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1067 nvmf_tcp_qpair_destroy(tqpair); 1068 return; 1069 } 1070 1071 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1072 } 1073 1074 static uint32_t 1075 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1076 { 1077 struct spdk_sock *sock; 1078 uint32_t count = 0; 1079 int i; 1080 1081 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1082 sock = spdk_sock_accept(port->listen_sock); 1083 if (sock == NULL) { 1084 break; 1085 } 1086 count++; 1087 nvmf_tcp_handle_connect(transport, port, sock); 1088 } 1089 1090 return count; 1091 } 1092 1093 static uint32_t 1094 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1095 { 1096 struct spdk_nvmf_tcp_transport *ttransport; 1097 struct spdk_nvmf_tcp_port *port; 1098 uint32_t count = 0; 1099 1100 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1101 1102 TAILQ_FOREACH(port, &ttransport->ports, link) { 1103 count += nvmf_tcp_port_accept(transport, port); 1104 } 1105 1106 return count; 1107 } 1108 1109 static void 1110 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1111 struct spdk_nvme_transport_id *trid, 1112 struct spdk_nvmf_discovery_log_page_entry *entry) 1113 { 1114 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1115 entry->adrfam = trid->adrfam; 1116 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1117 1118 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1119 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1120 1121 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1122 } 1123 1124 static struct spdk_nvmf_tcp_control_msg_list * 1125 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1126 { 1127 struct spdk_nvmf_tcp_control_msg_list *list; 1128 struct spdk_nvmf_tcp_control_msg *msg; 1129 uint16_t i; 1130 1131 list = calloc(1, sizeof(*list)); 1132 if (!list) { 1133 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1134 return NULL; 1135 } 1136 1137 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1138 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1139 if (!list->msg_buf) { 1140 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1141 free(list); 1142 return NULL; 1143 } 1144 1145 STAILQ_INIT(&list->free_msgs); 1146 1147 for (i = 0; i < num_messages; i++) { 1148 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1149 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1150 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1151 } 1152 1153 return list; 1154 } 1155 1156 static void 1157 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1158 { 1159 if (!list) { 1160 return; 1161 } 1162 1163 spdk_free(list->msg_buf); 1164 free(list); 1165 } 1166 1167 static struct spdk_nvmf_transport_poll_group * 1168 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1169 { 1170 struct spdk_nvmf_tcp_transport *ttransport; 1171 struct spdk_nvmf_tcp_poll_group *tgroup; 1172 1173 tgroup = calloc(1, sizeof(*tgroup)); 1174 if (!tgroup) { 1175 return NULL; 1176 } 1177 1178 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1179 if (!tgroup->sock_group) { 1180 goto cleanup; 1181 } 1182 1183 TAILQ_INIT(&tgroup->qpairs); 1184 TAILQ_INIT(&tgroup->await_req); 1185 1186 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1187 1188 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1189 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1190 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1191 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1192 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1193 if (!tgroup->control_msg_list) { 1194 goto cleanup; 1195 } 1196 } 1197 1198 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1199 if (spdk_unlikely(!tgroup->accel_channel)) { 1200 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1201 goto cleanup; 1202 } 1203 1204 return &tgroup->group; 1205 1206 cleanup: 1207 nvmf_tcp_poll_group_destroy(&tgroup->group); 1208 return NULL; 1209 } 1210 1211 static struct spdk_nvmf_transport_poll_group * 1212 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1213 { 1214 struct spdk_nvmf_tcp_qpair *tqpair; 1215 struct spdk_sock_group *group = NULL; 1216 int rc; 1217 1218 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1219 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1220 if (!rc && group != NULL) { 1221 return spdk_sock_group_get_ctx(group); 1222 } 1223 1224 return NULL; 1225 } 1226 1227 static void 1228 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1229 { 1230 struct spdk_nvmf_tcp_poll_group *tgroup; 1231 1232 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1233 spdk_sock_group_close(&tgroup->sock_group); 1234 if (tgroup->control_msg_list) { 1235 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1236 } 1237 1238 if (tgroup->accel_channel) { 1239 spdk_put_io_channel(tgroup->accel_channel); 1240 } 1241 1242 free(tgroup); 1243 } 1244 1245 static void 1246 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1247 enum nvme_tcp_pdu_recv_state state) 1248 { 1249 if (tqpair->recv_state == state) { 1250 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1251 tqpair, state); 1252 return; 1253 } 1254 1255 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1256 /* When leaving the await req state, move the qpair to the main list */ 1257 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1258 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1259 } 1260 1261 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1262 tqpair->recv_state = state; 1263 1264 switch (state) { 1265 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1266 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1267 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1268 break; 1269 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1270 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1271 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1272 break; 1273 case NVME_TCP_PDU_RECV_STATE_ERROR: 1274 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1275 memset(tqpair->pdu_in_progress, 0, sizeof(*(tqpair->pdu_in_progress))); 1276 break; 1277 default: 1278 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1279 abort(); 1280 break; 1281 } 1282 } 1283 1284 static int 1285 nvmf_tcp_qpair_handle_timeout(void *ctx) 1286 { 1287 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1288 1289 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1290 1291 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1292 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1293 1294 nvmf_tcp_qpair_disconnect(tqpair); 1295 return SPDK_POLLER_BUSY; 1296 } 1297 1298 static void 1299 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1300 { 1301 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1302 1303 if (!tqpair->timeout_poller) { 1304 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1305 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1306 } 1307 } 1308 1309 static void 1310 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1311 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1312 { 1313 struct nvme_tcp_pdu *rsp_pdu; 1314 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1315 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1316 uint32_t copy_len; 1317 1318 rsp_pdu = tqpair->mgmt_pdu; 1319 1320 c2h_term_req = &rsp_pdu->hdr.term_req; 1321 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1322 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1323 1324 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1325 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1326 DSET32(&c2h_term_req->fei, error_offset); 1327 } 1328 1329 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1330 1331 /* Copy the error info into the buffer */ 1332 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1333 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1334 1335 /* Contain the header of the wrong received pdu */ 1336 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1337 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1338 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1339 } 1340 1341 static void 1342 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1343 struct spdk_nvmf_tcp_qpair *tqpair, 1344 struct nvme_tcp_pdu *pdu) 1345 { 1346 struct spdk_nvmf_tcp_req *tcp_req; 1347 1348 assert(pdu->psh_valid_bytes == pdu->psh_len); 1349 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1350 1351 tcp_req = nvmf_tcp_req_get(tqpair); 1352 if (!tcp_req) { 1353 /* Directly return and make the allocation retry again */ 1354 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1355 return; 1356 } 1357 1358 /* The host sent more commands than the maximum queue depth. */ 1359 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1360 nvmf_tcp_qpair_disconnect(tqpair); 1361 return; 1362 } 1363 1364 pdu->req = tcp_req; 1365 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1366 nvmf_tcp_req_process(ttransport, tcp_req); 1367 } 1368 1369 static void 1370 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1371 struct spdk_nvmf_tcp_qpair *tqpair, 1372 struct nvme_tcp_pdu *pdu) 1373 { 1374 struct spdk_nvmf_tcp_req *tcp_req; 1375 struct spdk_nvme_tcp_cmd *capsule_cmd; 1376 uint32_t error_offset = 0; 1377 enum spdk_nvme_tcp_term_req_fes fes; 1378 1379 capsule_cmd = &pdu->hdr.capsule_cmd; 1380 tcp_req = pdu->req; 1381 assert(tcp_req != NULL); 1382 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1383 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1384 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1385 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1386 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1387 goto err; 1388 } 1389 1390 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1391 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1392 nvmf_tcp_req_process(ttransport, tcp_req); 1393 1394 return; 1395 err: 1396 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1397 } 1398 1399 static int 1400 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1401 enum spdk_nvmf_tcp_req_state state, 1402 uint16_t cid, uint16_t tag, 1403 struct spdk_nvmf_tcp_req **req) 1404 { 1405 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1406 1407 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1408 if (tcp_req->state != state) { 1409 continue; 1410 } 1411 1412 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1413 continue; 1414 } 1415 1416 if (tcp_req->ttag == tag) { 1417 *req = tcp_req; 1418 return 0; 1419 } 1420 1421 *req = NULL; 1422 return -1; 1423 } 1424 1425 /* Didn't find it, but not an error */ 1426 *req = NULL; 1427 return 0; 1428 } 1429 1430 static void 1431 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1432 struct spdk_nvmf_tcp_qpair *tqpair, 1433 struct nvme_tcp_pdu *pdu) 1434 { 1435 struct spdk_nvmf_tcp_req *tcp_req; 1436 uint32_t error_offset = 0; 1437 enum spdk_nvme_tcp_term_req_fes fes = 0; 1438 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1439 int rc; 1440 1441 h2c_data = &pdu->hdr.h2c_data; 1442 1443 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1444 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1445 1446 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1447 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1448 if (rc == 0 && tcp_req == NULL) { 1449 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1450 h2c_data->ttag, &tcp_req); 1451 } 1452 1453 if (!tcp_req) { 1454 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1455 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1456 if (rc == 0) { 1457 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1458 } else { 1459 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1460 } 1461 goto err; 1462 } 1463 1464 if (tcp_req->h2c_offset != h2c_data->datao) { 1465 SPDK_DEBUGLOG(nvmf_tcp, 1466 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1467 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1468 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1469 goto err; 1470 } 1471 1472 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1473 SPDK_DEBUGLOG(nvmf_tcp, 1474 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1475 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1476 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1477 goto err; 1478 } 1479 1480 pdu->req = tcp_req; 1481 1482 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1483 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1484 } 1485 1486 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1487 h2c_data->datao, h2c_data->datal); 1488 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1489 return; 1490 1491 err: 1492 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1493 } 1494 1495 static void 1496 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1497 struct spdk_nvmf_tcp_qpair *tqpair) 1498 { 1499 struct nvme_tcp_pdu *rsp_pdu; 1500 struct spdk_nvme_tcp_rsp *capsule_resp; 1501 1502 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1503 1504 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1505 assert(rsp_pdu != NULL); 1506 1507 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1508 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1509 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1510 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1511 if (tqpair->host_hdgst_enable) { 1512 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1513 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1514 } 1515 1516 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1517 } 1518 1519 static void 1520 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1521 { 1522 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1523 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1524 struct spdk_nvmf_tcp_qpair, qpair); 1525 1526 assert(tqpair != NULL); 1527 1528 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1529 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1530 tcp_req->req.length); 1531 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1532 return; 1533 } 1534 1535 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1536 nvmf_tcp_request_free(tcp_req); 1537 } else { 1538 nvmf_tcp_req_pdu_fini(tcp_req); 1539 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1540 } 1541 } 1542 1543 static void 1544 nvmf_tcp_r2t_complete(void *cb_arg) 1545 { 1546 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1547 struct spdk_nvmf_tcp_transport *ttransport; 1548 1549 nvmf_tcp_req_pdu_fini(tcp_req); 1550 1551 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1552 struct spdk_nvmf_tcp_transport, transport); 1553 1554 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1555 1556 if (tcp_req->h2c_offset == tcp_req->req.length) { 1557 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1558 nvmf_tcp_req_process(ttransport, tcp_req); 1559 } 1560 } 1561 1562 static void 1563 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1564 struct spdk_nvmf_tcp_req *tcp_req) 1565 { 1566 struct nvme_tcp_pdu *rsp_pdu; 1567 struct spdk_nvme_tcp_r2t_hdr *r2t; 1568 1569 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1570 assert(rsp_pdu != NULL); 1571 1572 r2t = &rsp_pdu->hdr.r2t; 1573 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1574 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1575 1576 if (tqpair->host_hdgst_enable) { 1577 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1578 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1579 } 1580 1581 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1582 r2t->ttag = tcp_req->ttag; 1583 r2t->r2to = tcp_req->h2c_offset; 1584 r2t->r2tl = tcp_req->req.length; 1585 1586 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1587 1588 SPDK_DEBUGLOG(nvmf_tcp, 1589 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1590 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1591 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1592 } 1593 1594 static void 1595 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1596 struct spdk_nvmf_tcp_qpair *tqpair, 1597 struct nvme_tcp_pdu *pdu) 1598 { 1599 struct spdk_nvmf_tcp_req *tcp_req; 1600 1601 tcp_req = pdu->req; 1602 assert(tcp_req != NULL); 1603 1604 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1605 1606 tcp_req->h2c_offset += pdu->data_len; 1607 1608 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1609 1610 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1611 * acknowledged before moving on. */ 1612 if (tcp_req->h2c_offset == tcp_req->req.length && 1613 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1614 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1615 nvmf_tcp_req_process(ttransport, tcp_req); 1616 } 1617 } 1618 1619 static void 1620 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1621 { 1622 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1623 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1624 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1625 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1626 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1627 DGET32(h2c_term_req->fei)); 1628 } 1629 } 1630 1631 static void 1632 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1633 struct nvme_tcp_pdu *pdu) 1634 { 1635 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1636 uint32_t error_offset = 0; 1637 enum spdk_nvme_tcp_term_req_fes fes; 1638 1639 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1640 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1641 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1642 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1643 goto end; 1644 } 1645 1646 /* set the data buffer */ 1647 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1648 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1649 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1650 return; 1651 end: 1652 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1653 } 1654 1655 static void 1656 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1657 struct nvme_tcp_pdu *pdu) 1658 { 1659 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1660 1661 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1662 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1663 } 1664 1665 static void 1666 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1667 struct spdk_nvmf_tcp_transport *ttransport) 1668 { 1669 struct nvme_tcp_pdu *pdu = tqpair->pdu_in_progress; 1670 1671 switch (pdu->hdr.common.pdu_type) { 1672 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1673 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1674 break; 1675 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1676 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1677 break; 1678 1679 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1680 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1681 break; 1682 1683 default: 1684 /* The code should not go to here */ 1685 SPDK_ERRLOG("The code should not go to here\n"); 1686 break; 1687 } 1688 } 1689 1690 static void 1691 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1692 struct spdk_nvmf_tcp_transport *ttransport) 1693 { 1694 int rc = 0; 1695 struct nvme_tcp_pdu *pdu; 1696 uint32_t crc32c, error_offset = 0; 1697 enum spdk_nvme_tcp_term_req_fes fes; 1698 1699 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1700 pdu = tqpair->pdu_in_progress; 1701 1702 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1703 /* check data digest if need */ 1704 if (pdu->ddgst_enable) { 1705 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1706 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1707 if (rc == 0) { 1708 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1709 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1710 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1711 return; 1712 1713 } 1714 } 1715 1716 _nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1717 } 1718 1719 static void 1720 nvmf_tcp_send_icresp_complete(void *cb_arg) 1721 { 1722 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1723 1724 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1725 } 1726 1727 static void 1728 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1729 struct spdk_nvmf_tcp_qpair *tqpair, 1730 struct nvme_tcp_pdu *pdu) 1731 { 1732 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1733 struct nvme_tcp_pdu *rsp_pdu; 1734 struct spdk_nvme_tcp_ic_resp *ic_resp; 1735 uint32_t error_offset = 0; 1736 enum spdk_nvme_tcp_term_req_fes fes; 1737 1738 /* Only PFV 0 is defined currently */ 1739 if (ic_req->pfv != 0) { 1740 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1741 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1742 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1743 goto end; 1744 } 1745 1746 /* MAXR2T is 0's based */ 1747 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1748 1749 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1750 if (!tqpair->host_hdgst_enable) { 1751 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1752 } 1753 1754 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1755 if (!tqpair->host_ddgst_enable) { 1756 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1757 } 1758 1759 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1760 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1761 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1762 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1763 tqpair, 1764 tqpair->recv_buf_size); 1765 /* Not fatal. */ 1766 } 1767 1768 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1769 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1770 1771 rsp_pdu = tqpair->mgmt_pdu; 1772 1773 ic_resp = &rsp_pdu->hdr.ic_resp; 1774 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1775 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1776 ic_resp->pfv = 0; 1777 ic_resp->cpda = tqpair->cpda; 1778 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1779 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1780 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1781 1782 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1783 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1784 1785 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1786 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1787 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1788 return; 1789 end: 1790 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1791 } 1792 1793 static void 1794 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1795 struct spdk_nvmf_tcp_transport *ttransport) 1796 { 1797 struct nvme_tcp_pdu *pdu; 1798 int rc; 1799 uint32_t crc32c, error_offset = 0; 1800 enum spdk_nvme_tcp_term_req_fes fes; 1801 1802 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1803 pdu = tqpair->pdu_in_progress; 1804 1805 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1806 pdu->hdr.common.pdu_type); 1807 /* check header digest if needed */ 1808 if (pdu->has_hdgst) { 1809 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1810 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1811 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1812 if (rc == 0) { 1813 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1814 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1815 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1816 return; 1817 1818 } 1819 } 1820 1821 switch (pdu->hdr.common.pdu_type) { 1822 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1823 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1824 break; 1825 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1826 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1827 break; 1828 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1829 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1830 break; 1831 1832 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1833 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1834 break; 1835 1836 default: 1837 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 1838 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1839 error_offset = 1; 1840 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1841 break; 1842 } 1843 } 1844 1845 static void 1846 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1847 { 1848 struct nvme_tcp_pdu *pdu; 1849 uint32_t error_offset = 0; 1850 enum spdk_nvme_tcp_term_req_fes fes; 1851 uint8_t expected_hlen, pdo; 1852 bool plen_error = false, pdo_error = false; 1853 1854 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1855 pdu = tqpair->pdu_in_progress; 1856 1857 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1858 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1859 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1860 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1861 goto err; 1862 } 1863 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1864 if (pdu->hdr.common.plen != expected_hlen) { 1865 plen_error = true; 1866 } 1867 } else { 1868 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1869 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1870 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1871 goto err; 1872 } 1873 1874 switch (pdu->hdr.common.pdu_type) { 1875 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1876 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1877 pdo = pdu->hdr.common.pdo; 1878 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1879 pdo_error = true; 1880 break; 1881 } 1882 1883 if (pdu->hdr.common.plen < expected_hlen) { 1884 plen_error = true; 1885 } 1886 break; 1887 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1888 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1889 pdo = pdu->hdr.common.pdo; 1890 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1891 pdo_error = true; 1892 break; 1893 } 1894 if (pdu->hdr.common.plen < expected_hlen) { 1895 plen_error = true; 1896 } 1897 break; 1898 1899 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1900 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1901 if ((pdu->hdr.common.plen <= expected_hlen) || 1902 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1903 plen_error = true; 1904 } 1905 break; 1906 1907 default: 1908 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1909 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1910 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1911 goto err; 1912 } 1913 } 1914 1915 if (pdu->hdr.common.hlen != expected_hlen) { 1916 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1917 pdu->hdr.common.pdu_type, 1918 expected_hlen, pdu->hdr.common.hlen, tqpair); 1919 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1920 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1921 goto err; 1922 } else if (pdo_error) { 1923 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1924 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1925 } else if (plen_error) { 1926 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1927 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1928 goto err; 1929 } else { 1930 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1931 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1932 return; 1933 } 1934 err: 1935 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1936 } 1937 1938 static int 1939 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1940 int read_len) 1941 { 1942 int rc; 1943 1944 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1945 read_offset, read_len, pdu->dif_ctx); 1946 if (rc != 0) { 1947 SPDK_ERRLOG("DIF generate failed\n"); 1948 } 1949 1950 return rc; 1951 } 1952 1953 static int 1954 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1955 { 1956 int rc = 0; 1957 struct nvme_tcp_pdu *pdu; 1958 enum nvme_tcp_pdu_recv_state prev_state; 1959 uint32_t data_len; 1960 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1961 struct spdk_nvmf_tcp_transport, transport); 1962 1963 /* The loop here is to allow for several back-to-back state changes. */ 1964 do { 1965 prev_state = tqpair->recv_state; 1966 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1967 1968 pdu = tqpair->pdu_in_progress; 1969 switch (tqpair->recv_state) { 1970 /* Wait for the common header */ 1971 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1972 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1973 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1974 return rc; 1975 } 1976 1977 rc = nvme_tcp_read_data(tqpair->sock, 1978 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1979 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1980 if (rc < 0) { 1981 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1982 return NVME_TCP_PDU_FATAL; 1983 } else if (rc > 0) { 1984 pdu->ch_valid_bytes += rc; 1985 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1986 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1987 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1988 } 1989 } 1990 1991 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1992 return NVME_TCP_PDU_IN_PROGRESS; 1993 } 1994 1995 /* The command header of this PDU has now been read from the socket. */ 1996 nvmf_tcp_pdu_ch_handle(tqpair); 1997 break; 1998 /* Wait for the pdu specific header */ 1999 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2000 rc = nvme_tcp_read_data(tqpair->sock, 2001 pdu->psh_len - pdu->psh_valid_bytes, 2002 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2003 if (rc < 0) { 2004 return NVME_TCP_PDU_FATAL; 2005 } else if (rc > 0) { 2006 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 2007 0, rc, 0, 0); 2008 pdu->psh_valid_bytes += rc; 2009 } 2010 2011 if (pdu->psh_valid_bytes < pdu->psh_len) { 2012 return NVME_TCP_PDU_IN_PROGRESS; 2013 } 2014 2015 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2016 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2017 break; 2018 /* Wait for the req slot */ 2019 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2020 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2021 break; 2022 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2023 /* check whether the data is valid, if not we just return */ 2024 if (!pdu->data_len) { 2025 return NVME_TCP_PDU_IN_PROGRESS; 2026 } 2027 2028 data_len = pdu->data_len; 2029 /* data digest */ 2030 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2031 tqpair->host_ddgst_enable)) { 2032 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2033 pdu->ddgst_enable = true; 2034 } 2035 2036 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2037 if (rc < 0) { 2038 return NVME_TCP_PDU_FATAL; 2039 } 2040 pdu->rw_offset += rc; 2041 2042 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2043 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2044 if (rc != 0) { 2045 return NVME_TCP_PDU_FATAL; 2046 } 2047 } 2048 2049 if (pdu->rw_offset < data_len) { 2050 return NVME_TCP_PDU_IN_PROGRESS; 2051 } 2052 2053 /* All of this PDU has now been read from the socket. */ 2054 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2055 break; 2056 case NVME_TCP_PDU_RECV_STATE_ERROR: 2057 if (!spdk_sock_is_connected(tqpair->sock)) { 2058 return NVME_TCP_PDU_FATAL; 2059 } 2060 break; 2061 default: 2062 assert(0); 2063 SPDK_ERRLOG("code should not come to here"); 2064 break; 2065 } 2066 } while (tqpair->recv_state != prev_state); 2067 2068 return rc; 2069 } 2070 2071 static inline void * 2072 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2073 { 2074 struct spdk_nvmf_tcp_control_msg *msg; 2075 2076 assert(list); 2077 2078 msg = STAILQ_FIRST(&list->free_msgs); 2079 if (!msg) { 2080 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2081 return NULL; 2082 } 2083 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2084 return msg; 2085 } 2086 2087 static inline void 2088 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2089 { 2090 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2091 2092 assert(list); 2093 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2094 } 2095 2096 static int 2097 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2098 struct spdk_nvmf_transport *transport, 2099 struct spdk_nvmf_transport_poll_group *group) 2100 { 2101 struct spdk_nvmf_request *req = &tcp_req->req; 2102 struct spdk_nvme_cmd *cmd; 2103 struct spdk_nvme_cpl *rsp; 2104 struct spdk_nvme_sgl_descriptor *sgl; 2105 struct spdk_nvmf_tcp_poll_group *tgroup; 2106 uint32_t length; 2107 2108 cmd = &req->cmd->nvme_cmd; 2109 rsp = &req->rsp->nvme_cpl; 2110 sgl = &cmd->dptr.sgl1; 2111 2112 length = sgl->unkeyed.length; 2113 2114 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2115 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2116 if (length > transport->opts.max_io_size) { 2117 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2118 length, transport->opts.max_io_size); 2119 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2120 return -1; 2121 } 2122 2123 /* fill request length and populate iovs */ 2124 req->length = length; 2125 2126 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2127 2128 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2129 req->dif.orig_length = length; 2130 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2131 req->dif.elba_length = length; 2132 } 2133 2134 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2135 /* No available buffers. Queue this request up. */ 2136 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2137 tcp_req); 2138 return 0; 2139 } 2140 2141 /* backward compatible */ 2142 req->data = req->iov[0].iov_base; 2143 2144 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2145 tcp_req, req->iovcnt, req->data); 2146 2147 return 0; 2148 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2149 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2150 uint64_t offset = sgl->address; 2151 uint32_t max_len = transport->opts.in_capsule_data_size; 2152 assert(tcp_req->has_incapsule_data); 2153 2154 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2155 offset, length); 2156 2157 if (offset > max_len) { 2158 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2159 offset, max_len); 2160 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2161 return -1; 2162 } 2163 max_len -= (uint32_t)offset; 2164 2165 if (spdk_unlikely(length > max_len)) { 2166 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2167 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2168 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2169 2170 /* Get a buffer from dedicated list */ 2171 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2172 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2173 assert(tgroup->control_msg_list); 2174 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2175 if (!req->data) { 2176 /* No available buffers. Queue this request up. */ 2177 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2178 return 0; 2179 } 2180 } else { 2181 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2182 length, max_len); 2183 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2184 return -1; 2185 } 2186 } else { 2187 req->data = tcp_req->buf; 2188 } 2189 2190 req->length = length; 2191 req->data_from_pool = false; 2192 2193 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2194 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2195 req->dif.elba_length = length; 2196 } 2197 2198 req->iov[0].iov_base = req->data; 2199 req->iov[0].iov_len = length; 2200 req->iovcnt = 1; 2201 2202 return 0; 2203 } 2204 2205 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2206 sgl->generic.type, sgl->generic.subtype); 2207 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2208 return -1; 2209 } 2210 2211 static inline enum spdk_nvme_media_error_status_code 2212 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2213 enum spdk_nvme_media_error_status_code result; 2214 2215 switch (err_type) 2216 { 2217 case SPDK_DIF_REFTAG_ERROR: 2218 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2219 break; 2220 case SPDK_DIF_APPTAG_ERROR: 2221 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2222 break; 2223 case SPDK_DIF_GUARD_ERROR: 2224 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2225 break; 2226 default: 2227 SPDK_UNREACHABLE(); 2228 break; 2229 } 2230 2231 return result; 2232 } 2233 2234 static void 2235 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2236 struct spdk_nvmf_tcp_req *tcp_req) 2237 { 2238 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2239 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2240 struct nvme_tcp_pdu *rsp_pdu; 2241 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2242 uint32_t plen, pdo, alignment; 2243 int rc; 2244 2245 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2246 2247 rsp_pdu = tcp_req->pdu; 2248 assert(rsp_pdu != NULL); 2249 assert(tcp_req->pdu_in_use); 2250 2251 c2h_data = &rsp_pdu->hdr.c2h_data; 2252 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2253 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2254 2255 if (tqpair->host_hdgst_enable) { 2256 plen += SPDK_NVME_TCP_DIGEST_LEN; 2257 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2258 } 2259 2260 /* set the psh */ 2261 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2262 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2263 c2h_data->datao = tcp_req->pdu->rw_offset; 2264 2265 /* set the padding */ 2266 rsp_pdu->padding_len = 0; 2267 pdo = plen; 2268 if (tqpair->cpda) { 2269 alignment = (tqpair->cpda + 1) << 2; 2270 if (plen % alignment != 0) { 2271 pdo = (plen + alignment) / alignment * alignment; 2272 rsp_pdu->padding_len = pdo - plen; 2273 plen = pdo; 2274 } 2275 } 2276 2277 c2h_data->common.pdo = pdo; 2278 plen += c2h_data->datal; 2279 if (tqpair->host_ddgst_enable) { 2280 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2281 plen += SPDK_NVME_TCP_DIGEST_LEN; 2282 } 2283 2284 c2h_data->common.plen = plen; 2285 2286 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2287 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2288 } 2289 2290 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2291 c2h_data->datao, c2h_data->datal); 2292 2293 2294 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2295 /* Need to send the capsule response if response is not all 0 */ 2296 if (ttransport->tcp_opts.c2h_success && 2297 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2298 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2299 } 2300 2301 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2302 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2303 struct spdk_dif_error err_blk = {}; 2304 uint32_t mapped_length = 0; 2305 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2306 uint32_t ddgst_len = 0; 2307 2308 if (tqpair->host_ddgst_enable) { 2309 /* Data digest consumes additional iov entry */ 2310 available_iovs--; 2311 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2312 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2313 c2h_data->common.plen -= ddgst_len; 2314 } 2315 /* Temp call to estimate if data can be described by limited number of iovs. 2316 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2317 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2318 false, &mapped_length); 2319 2320 if (mapped_length != c2h_data->common.plen) { 2321 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2322 SPDK_DEBUGLOG(nvmf_tcp, 2323 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2324 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2325 c2h_data->common.plen = mapped_length; 2326 2327 /* Rebuild pdu->data_iov since data length is changed */ 2328 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2329 c2h_data->datal); 2330 2331 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2332 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2333 } 2334 2335 c2h_data->common.plen += ddgst_len; 2336 2337 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2338 2339 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2340 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2341 if (rc != 0) { 2342 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2343 err_blk.err_type, err_blk.err_offset); 2344 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2345 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2346 nvmf_tcp_req_pdu_fini(tcp_req); 2347 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2348 return; 2349 } 2350 } 2351 2352 rsp_pdu->rw_offset += c2h_data->datal; 2353 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2354 } 2355 2356 static void 2357 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2358 struct spdk_nvmf_tcp_req *tcp_req) 2359 { 2360 nvmf_tcp_req_pdu_init(tcp_req); 2361 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2362 } 2363 2364 static int 2365 request_transfer_out(struct spdk_nvmf_request *req) 2366 { 2367 struct spdk_nvmf_tcp_req *tcp_req; 2368 struct spdk_nvmf_qpair *qpair; 2369 struct spdk_nvmf_tcp_qpair *tqpair; 2370 struct spdk_nvme_cpl *rsp; 2371 2372 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2373 2374 qpair = req->qpair; 2375 rsp = &req->rsp->nvme_cpl; 2376 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2377 2378 /* Advance our sq_head pointer */ 2379 if (qpair->sq_head == qpair->sq_head_max) { 2380 qpair->sq_head = 0; 2381 } else { 2382 qpair->sq_head++; 2383 } 2384 rsp->sqhd = qpair->sq_head; 2385 2386 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2387 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2388 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2389 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2390 } else { 2391 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2392 } 2393 2394 return 0; 2395 } 2396 2397 static void 2398 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2399 struct spdk_nvmf_tcp_req *tcp_req) 2400 { 2401 struct nvme_tcp_pdu *pdu; 2402 uint32_t plen = 0; 2403 2404 pdu = tqpair->pdu_in_progress; 2405 plen = pdu->hdr.common.hlen; 2406 2407 if (tqpair->host_hdgst_enable) { 2408 plen += SPDK_NVME_TCP_DIGEST_LEN; 2409 } 2410 2411 if (pdu->hdr.common.plen != plen) { 2412 tcp_req->has_incapsule_data = true; 2413 } 2414 } 2415 2416 static bool 2417 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2418 struct spdk_nvmf_tcp_req *tcp_req) 2419 { 2420 struct spdk_nvmf_tcp_qpair *tqpair; 2421 int rc; 2422 enum spdk_nvmf_tcp_req_state prev_state; 2423 bool progress = false; 2424 struct spdk_nvmf_transport *transport = &ttransport->transport; 2425 struct spdk_nvmf_transport_poll_group *group; 2426 struct spdk_nvmf_tcp_poll_group *tgroup; 2427 2428 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2429 group = &tqpair->group->group; 2430 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2431 2432 /* If the qpair is not active, we need to abort the outstanding requests. */ 2433 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2434 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2435 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2436 } 2437 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2438 } 2439 2440 /* The loop here is to allow for several back-to-back state changes. */ 2441 do { 2442 prev_state = tcp_req->state; 2443 2444 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2445 tqpair); 2446 2447 switch (tcp_req->state) { 2448 case TCP_REQUEST_STATE_FREE: 2449 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2450 * to escape this state. */ 2451 break; 2452 case TCP_REQUEST_STATE_NEW: 2453 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2454 2455 /* copy the cmd from the receive pdu */ 2456 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2457 2458 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2459 tcp_req->req.dif.dif_insert_or_strip = true; 2460 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2461 } 2462 2463 /* The next state transition depends on the data transfer needs of this request. */ 2464 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2465 2466 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2467 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2468 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2469 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2470 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2471 break; 2472 } 2473 2474 /* If no data to transfer, ready to execute. */ 2475 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2476 /* Reset the tqpair receving pdu state */ 2477 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2478 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2479 break; 2480 } 2481 2482 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2483 2484 if (!tcp_req->has_incapsule_data) { 2485 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2486 } 2487 2488 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2489 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2490 break; 2491 case TCP_REQUEST_STATE_NEED_BUFFER: 2492 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2493 2494 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2495 2496 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2497 SPDK_DEBUGLOG(nvmf_tcp, 2498 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2499 tcp_req, tqpair); 2500 /* This request needs to wait in line to obtain a buffer */ 2501 break; 2502 } 2503 2504 /* Try to get a data buffer */ 2505 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2506 if (rc < 0) { 2507 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2508 /* Reset the tqpair receving pdu state */ 2509 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2510 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2511 break; 2512 } 2513 2514 if (!tcp_req->req.data) { 2515 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2516 tcp_req, tqpair); 2517 /* No buffers available. */ 2518 break; 2519 } 2520 2521 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2522 2523 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2524 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2525 if (tcp_req->req.data_from_pool) { 2526 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2527 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2528 } else { 2529 struct nvme_tcp_pdu *pdu; 2530 2531 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2532 2533 pdu = tqpair->pdu_in_progress; 2534 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2535 tqpair); 2536 /* No need to send r2t, contained in the capsuled data */ 2537 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2538 0, tcp_req->req.length); 2539 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2540 } 2541 break; 2542 } 2543 2544 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2545 break; 2546 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2547 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2548 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2549 break; 2550 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2551 2552 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2553 (uintptr_t)tcp_req, 0); 2554 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2555 * to escape this state. */ 2556 break; 2557 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2558 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2559 2560 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2561 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2562 tcp_req->req.length = tcp_req->req.dif.elba_length; 2563 } 2564 2565 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2566 spdk_nvmf_request_exec(&tcp_req->req); 2567 break; 2568 case TCP_REQUEST_STATE_EXECUTING: 2569 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2570 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2571 * to escape this state. */ 2572 break; 2573 case TCP_REQUEST_STATE_EXECUTED: 2574 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2575 2576 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2577 tcp_req->req.length = tcp_req->req.dif.orig_length; 2578 } 2579 2580 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2581 break; 2582 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2583 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2584 rc = request_transfer_out(&tcp_req->req); 2585 assert(rc == 0); /* No good way to handle this currently */ 2586 break; 2587 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2588 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2589 (uintptr_t)tcp_req, 2590 0); 2591 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2592 * to escape this state. */ 2593 break; 2594 case TCP_REQUEST_STATE_COMPLETED: 2595 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2596 if (tcp_req->req.data_from_pool) { 2597 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2598 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2599 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2600 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2601 assert(tgroup->control_msg_list); 2602 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2603 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2604 } 2605 tcp_req->req.length = 0; 2606 tcp_req->req.iovcnt = 0; 2607 tcp_req->req.data = NULL; 2608 2609 nvmf_tcp_req_pdu_fini(tcp_req); 2610 2611 nvmf_tcp_req_put(tqpair, tcp_req); 2612 break; 2613 case TCP_REQUEST_NUM_STATES: 2614 default: 2615 assert(0); 2616 break; 2617 } 2618 2619 if (tcp_req->state != prev_state) { 2620 progress = true; 2621 } 2622 } while (tcp_req->state != prev_state); 2623 2624 return progress; 2625 } 2626 2627 static void 2628 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2629 { 2630 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2631 int rc; 2632 2633 assert(tqpair != NULL); 2634 rc = nvmf_tcp_sock_process(tqpair); 2635 2636 /* If there was a new socket error, disconnect */ 2637 if (rc < 0) { 2638 nvmf_tcp_qpair_disconnect(tqpair); 2639 } 2640 } 2641 2642 static int 2643 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2644 struct spdk_nvmf_qpair *qpair) 2645 { 2646 struct spdk_nvmf_tcp_poll_group *tgroup; 2647 struct spdk_nvmf_tcp_qpair *tqpair; 2648 int rc; 2649 2650 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2651 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2652 2653 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2654 nvmf_tcp_sock_cb, tqpair); 2655 if (rc != 0) { 2656 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2657 spdk_strerror(errno), errno); 2658 return -1; 2659 } 2660 2661 rc = nvmf_tcp_qpair_sock_init(tqpair); 2662 if (rc != 0) { 2663 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2664 return -1; 2665 } 2666 2667 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2668 if (rc < 0) { 2669 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2670 return -1; 2671 } 2672 2673 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2674 if (rc < 0) { 2675 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2676 return -1; 2677 } 2678 2679 tqpair->group = tgroup; 2680 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2681 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2682 2683 return 0; 2684 } 2685 2686 static int 2687 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2688 struct spdk_nvmf_qpair *qpair) 2689 { 2690 struct spdk_nvmf_tcp_poll_group *tgroup; 2691 struct spdk_nvmf_tcp_qpair *tqpair; 2692 int rc; 2693 2694 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2695 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2696 2697 assert(tqpair->group == tgroup); 2698 2699 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2700 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2701 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2702 } else { 2703 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2704 } 2705 2706 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2707 if (rc != 0) { 2708 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2709 spdk_strerror(errno), errno); 2710 } 2711 2712 return rc; 2713 } 2714 2715 static int 2716 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2717 { 2718 struct spdk_nvmf_tcp_transport *ttransport; 2719 struct spdk_nvmf_tcp_req *tcp_req; 2720 2721 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2722 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2723 2724 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2725 nvmf_tcp_req_process(ttransport, tcp_req); 2726 2727 return 0; 2728 } 2729 2730 static void 2731 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2732 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2733 { 2734 struct spdk_nvmf_tcp_qpair *tqpair; 2735 2736 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2737 2738 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2739 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2740 nvmf_tcp_qpair_destroy(tqpair); 2741 2742 if (cb_fn) { 2743 cb_fn(cb_arg); 2744 } 2745 } 2746 2747 static int 2748 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2749 { 2750 struct spdk_nvmf_tcp_poll_group *tgroup; 2751 int rc; 2752 struct spdk_nvmf_request *req, *req_tmp; 2753 struct spdk_nvmf_tcp_req *tcp_req; 2754 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2755 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2756 struct spdk_nvmf_tcp_transport, transport); 2757 2758 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2759 2760 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2761 return 0; 2762 } 2763 2764 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2765 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2766 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2767 break; 2768 } 2769 } 2770 2771 rc = spdk_sock_group_poll(tgroup->sock_group); 2772 if (rc < 0) { 2773 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2774 } 2775 2776 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2777 nvmf_tcp_sock_process(tqpair); 2778 } 2779 2780 return rc; 2781 } 2782 2783 static int 2784 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2785 struct spdk_nvme_transport_id *trid, bool peer) 2786 { 2787 struct spdk_nvmf_tcp_qpair *tqpair; 2788 uint16_t port; 2789 2790 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2791 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2792 2793 if (peer) { 2794 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2795 port = tqpair->initiator_port; 2796 } else { 2797 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2798 port = tqpair->target_port; 2799 } 2800 2801 if (spdk_sock_is_ipv4(tqpair->sock)) { 2802 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2803 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2804 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2805 } else { 2806 return -1; 2807 } 2808 2809 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2810 return 0; 2811 } 2812 2813 static int 2814 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2815 struct spdk_nvme_transport_id *trid) 2816 { 2817 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2818 } 2819 2820 static int 2821 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2822 struct spdk_nvme_transport_id *trid) 2823 { 2824 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2825 } 2826 2827 static int 2828 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2829 struct spdk_nvme_transport_id *trid) 2830 { 2831 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2832 } 2833 2834 static void 2835 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2836 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2837 { 2838 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2839 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2840 2841 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2842 2843 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2844 } 2845 2846 static int 2847 _nvmf_tcp_qpair_abort_request(void *ctx) 2848 { 2849 struct spdk_nvmf_request *req = ctx; 2850 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2851 struct spdk_nvmf_tcp_req, req); 2852 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2853 struct spdk_nvmf_tcp_qpair, qpair); 2854 int rc; 2855 2856 spdk_poller_unregister(&req->poller); 2857 2858 switch (tcp_req_to_abort->state) { 2859 case TCP_REQUEST_STATE_EXECUTING: 2860 rc = nvmf_ctrlr_abort_request(req); 2861 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2862 return SPDK_POLLER_BUSY; 2863 } 2864 break; 2865 2866 case TCP_REQUEST_STATE_NEED_BUFFER: 2867 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2868 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2869 2870 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2871 break; 2872 2873 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2874 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2875 break; 2876 2877 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2878 if (spdk_get_ticks() < req->timeout_tsc) { 2879 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2880 return SPDK_POLLER_BUSY; 2881 } 2882 break; 2883 2884 default: 2885 break; 2886 } 2887 2888 spdk_nvmf_request_complete(req); 2889 return SPDK_POLLER_BUSY; 2890 } 2891 2892 static void 2893 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2894 struct spdk_nvmf_request *req) 2895 { 2896 struct spdk_nvmf_tcp_qpair *tqpair; 2897 struct spdk_nvmf_tcp_transport *ttransport; 2898 struct spdk_nvmf_transport *transport; 2899 uint16_t cid; 2900 uint32_t i; 2901 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2902 2903 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2904 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2905 transport = &ttransport->transport; 2906 2907 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2908 2909 for (i = 0; i < tqpair->resource_count; i++) { 2910 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2911 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2912 tcp_req_to_abort = &tqpair->reqs[i]; 2913 break; 2914 } 2915 } 2916 2917 if (tcp_req_to_abort == NULL) { 2918 spdk_nvmf_request_complete(req); 2919 return; 2920 } 2921 2922 req->req_to_abort = &tcp_req_to_abort->req; 2923 req->timeout_tsc = spdk_get_ticks() + 2924 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2925 req->poller = NULL; 2926 2927 _nvmf_tcp_qpair_abort_request(req); 2928 } 2929 2930 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2931 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2932 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2933 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2934 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2935 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2936 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2937 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2938 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2939 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2940 2941 static void 2942 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2943 { 2944 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2945 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2946 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2947 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2948 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2949 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2950 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2951 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2952 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2953 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2954 opts->transport_specific = NULL; 2955 } 2956 2957 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2958 .name = "TCP", 2959 .type = SPDK_NVME_TRANSPORT_TCP, 2960 .opts_init = nvmf_tcp_opts_init, 2961 .create = nvmf_tcp_create, 2962 .dump_opts = nvmf_tcp_dump_opts, 2963 .destroy = nvmf_tcp_destroy, 2964 2965 .listen = nvmf_tcp_listen, 2966 .stop_listen = nvmf_tcp_stop_listen, 2967 .accept = nvmf_tcp_accept, 2968 2969 .listener_discover = nvmf_tcp_discover, 2970 2971 .poll_group_create = nvmf_tcp_poll_group_create, 2972 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2973 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2974 .poll_group_add = nvmf_tcp_poll_group_add, 2975 .poll_group_remove = nvmf_tcp_poll_group_remove, 2976 .poll_group_poll = nvmf_tcp_poll_group_poll, 2977 2978 .req_free = nvmf_tcp_req_free, 2979 .req_complete = nvmf_tcp_req_complete, 2980 2981 .qpair_fini = nvmf_tcp_close_qpair, 2982 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2983 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2984 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2985 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2986 }; 2987 2988 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2989 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2990