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_enabled = 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 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 890 struct nvme_tcp_pdu *pdu, 891 nvme_tcp_qpair_xfer_complete_cb cb_fn, 892 void *cb_arg) 893 { 894 int hlen; 895 uint32_t crc32c; 896 897 assert(tqpair->pdu_in_progress != pdu); 898 899 hlen = pdu->hdr.common.hlen; 900 pdu->cb_fn = cb_fn; 901 pdu->cb_arg = cb_arg; 902 903 pdu->iov[0].iov_base = &pdu->hdr.raw; 904 pdu->iov[0].iov_len = hlen; 905 906 /* Header Digest */ 907 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 908 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 909 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 910 } 911 912 /* Data Digest */ 913 pdu_data_crc32_compute(pdu); 914 } 915 916 static int 917 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 918 { 919 uint32_t i; 920 struct spdk_nvmf_transport_opts *opts; 921 uint32_t in_capsule_data_size; 922 923 opts = &tqpair->qpair.transport->opts; 924 925 in_capsule_data_size = opts->in_capsule_data_size; 926 if (opts->dif_insert_or_strip) { 927 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 928 } 929 930 tqpair->resource_count = opts->max_queue_depth; 931 932 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 933 if (!tqpair->reqs) { 934 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 935 return -1; 936 } 937 938 if (in_capsule_data_size) { 939 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 940 NULL, SPDK_ENV_LCORE_ID_ANY, 941 SPDK_MALLOC_DMA); 942 if (!tqpair->bufs) { 943 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 944 return -1; 945 } 946 } 947 948 /* Add addtional 2 members, which will be used for mgmt_pdu and pdu_in_progress owned by the tqpair */ 949 tqpair->pdus = spdk_dma_zmalloc((tqpair->resource_count + 2) * sizeof(*tqpair->pdus), 0x1000, NULL); 950 if (!tqpair->pdus) { 951 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 952 return -1; 953 } 954 955 for (i = 0; i < tqpair->resource_count; i++) { 956 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 957 958 tcp_req->ttag = i + 1; 959 tcp_req->req.qpair = &tqpair->qpair; 960 961 tcp_req->pdu = &tqpair->pdus[i]; 962 tcp_req->pdu->qpair = tqpair; 963 964 /* Set up memory to receive commands */ 965 if (tqpair->bufs) { 966 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 967 } 968 969 /* Set the cmdn and rsp */ 970 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 971 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 972 973 /* Initialize request state to FREE */ 974 tcp_req->state = TCP_REQUEST_STATE_FREE; 975 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 976 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 977 } 978 979 tqpair->mgmt_pdu = &tqpair->pdus[i]; 980 tqpair->mgmt_pdu->qpair = tqpair; 981 tqpair->pdu_in_progress = &tqpair->pdus[i + 1]; 982 983 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 984 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 985 986 return 0; 987 } 988 989 static int 990 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 991 { 992 struct spdk_nvmf_tcp_qpair *tqpair; 993 994 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 995 996 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 997 998 /* Initialise request state queues of the qpair */ 999 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1000 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1001 1002 tqpair->host_hdgst_enable = true; 1003 tqpair->host_ddgst_enable = true; 1004 1005 return 0; 1006 } 1007 1008 static int 1009 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1010 { 1011 int rc; 1012 1013 /* set low water mark */ 1014 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1015 if (rc != 0) { 1016 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1017 return rc; 1018 } 1019 1020 return 0; 1021 } 1022 1023 static void 1024 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1025 struct spdk_nvmf_tcp_port *port, 1026 struct spdk_sock *sock) 1027 { 1028 struct spdk_nvmf_tcp_qpair *tqpair; 1029 int rc; 1030 1031 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1032 port->trid->traddr, port->trid->trsvcid); 1033 1034 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1035 if (tqpair == NULL) { 1036 SPDK_ERRLOG("Could not allocate new connection.\n"); 1037 spdk_sock_close(&sock); 1038 return; 1039 } 1040 1041 tqpair->sock = sock; 1042 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1043 tqpair->port = port; 1044 tqpair->qpair.transport = transport; 1045 1046 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1047 sizeof(tqpair->target_addr), &tqpair->target_port, 1048 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1049 &tqpair->initiator_port); 1050 if (rc < 0) { 1051 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1052 nvmf_tcp_qpair_destroy(tqpair); 1053 return; 1054 } 1055 1056 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1057 } 1058 1059 static uint32_t 1060 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1061 { 1062 struct spdk_sock *sock; 1063 uint32_t count = 0; 1064 int i; 1065 1066 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1067 sock = spdk_sock_accept(port->listen_sock); 1068 if (sock == NULL) { 1069 break; 1070 } 1071 count++; 1072 nvmf_tcp_handle_connect(transport, port, sock); 1073 } 1074 1075 return count; 1076 } 1077 1078 static uint32_t 1079 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1080 { 1081 struct spdk_nvmf_tcp_transport *ttransport; 1082 struct spdk_nvmf_tcp_port *port; 1083 uint32_t count = 0; 1084 1085 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1086 1087 TAILQ_FOREACH(port, &ttransport->ports, link) { 1088 count += nvmf_tcp_port_accept(transport, port); 1089 } 1090 1091 return count; 1092 } 1093 1094 static void 1095 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1096 struct spdk_nvme_transport_id *trid, 1097 struct spdk_nvmf_discovery_log_page_entry *entry) 1098 { 1099 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1100 entry->adrfam = trid->adrfam; 1101 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1102 1103 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1104 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1105 1106 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1107 } 1108 1109 static struct spdk_nvmf_tcp_control_msg_list * 1110 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1111 { 1112 struct spdk_nvmf_tcp_control_msg_list *list; 1113 struct spdk_nvmf_tcp_control_msg *msg; 1114 uint16_t i; 1115 1116 list = calloc(1, sizeof(*list)); 1117 if (!list) { 1118 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1119 return NULL; 1120 } 1121 1122 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1123 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1124 if (!list->msg_buf) { 1125 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1126 free(list); 1127 return NULL; 1128 } 1129 1130 STAILQ_INIT(&list->free_msgs); 1131 1132 for (i = 0; i < num_messages; i++) { 1133 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1134 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1135 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1136 } 1137 1138 return list; 1139 } 1140 1141 static void 1142 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1143 { 1144 if (!list) { 1145 return; 1146 } 1147 1148 spdk_free(list->msg_buf); 1149 free(list); 1150 } 1151 1152 static struct spdk_nvmf_transport_poll_group * 1153 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1154 { 1155 struct spdk_nvmf_tcp_transport *ttransport; 1156 struct spdk_nvmf_tcp_poll_group *tgroup; 1157 1158 tgroup = calloc(1, sizeof(*tgroup)); 1159 if (!tgroup) { 1160 return NULL; 1161 } 1162 1163 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1164 if (!tgroup->sock_group) { 1165 goto cleanup; 1166 } 1167 1168 TAILQ_INIT(&tgroup->qpairs); 1169 TAILQ_INIT(&tgroup->await_req); 1170 1171 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1172 1173 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1174 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1175 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1176 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1177 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1178 if (!tgroup->control_msg_list) { 1179 goto cleanup; 1180 } 1181 } 1182 1183 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1184 if (spdk_unlikely(!tgroup->accel_channel)) { 1185 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1186 goto cleanup; 1187 } 1188 1189 return &tgroup->group; 1190 1191 cleanup: 1192 nvmf_tcp_poll_group_destroy(&tgroup->group); 1193 return NULL; 1194 } 1195 1196 static struct spdk_nvmf_transport_poll_group * 1197 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1198 { 1199 struct spdk_nvmf_tcp_qpair *tqpair; 1200 struct spdk_sock_group *group = NULL; 1201 int rc; 1202 1203 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1204 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1205 if (!rc && group != NULL) { 1206 return spdk_sock_group_get_ctx(group); 1207 } 1208 1209 return NULL; 1210 } 1211 1212 static void 1213 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1214 { 1215 struct spdk_nvmf_tcp_poll_group *tgroup; 1216 1217 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1218 spdk_sock_group_close(&tgroup->sock_group); 1219 if (tgroup->control_msg_list) { 1220 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1221 } 1222 1223 if (tgroup->accel_channel) { 1224 spdk_put_io_channel(tgroup->accel_channel); 1225 } 1226 1227 free(tgroup); 1228 } 1229 1230 static void 1231 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1232 enum nvme_tcp_pdu_recv_state state) 1233 { 1234 if (tqpair->recv_state == state) { 1235 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1236 tqpair, state); 1237 return; 1238 } 1239 1240 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1241 /* When leaving the await req state, move the qpair to the main list */ 1242 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1243 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1244 } 1245 1246 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1247 tqpair->recv_state = state; 1248 1249 switch (state) { 1250 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1251 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1252 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1253 break; 1254 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1255 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1256 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1257 break; 1258 case NVME_TCP_PDU_RECV_STATE_ERROR: 1259 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1260 memset(tqpair->pdu_in_progress, 0, sizeof(*(tqpair->pdu_in_progress))); 1261 break; 1262 default: 1263 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1264 abort(); 1265 break; 1266 } 1267 } 1268 1269 static int 1270 nvmf_tcp_qpair_handle_timeout(void *ctx) 1271 { 1272 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1273 1274 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1275 1276 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1277 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1278 1279 nvmf_tcp_qpair_disconnect(tqpair); 1280 return SPDK_POLLER_BUSY; 1281 } 1282 1283 static void 1284 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1285 { 1286 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1287 1288 if (!tqpair->timeout_poller) { 1289 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1290 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1291 } 1292 } 1293 1294 static void 1295 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1296 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1297 { 1298 struct nvme_tcp_pdu *rsp_pdu; 1299 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1300 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1301 uint32_t copy_len; 1302 1303 rsp_pdu = tqpair->mgmt_pdu; 1304 1305 c2h_term_req = &rsp_pdu->hdr.term_req; 1306 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1307 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1308 1309 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1310 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1311 DSET32(&c2h_term_req->fei, error_offset); 1312 } 1313 1314 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1315 1316 /* Copy the error info into the buffer */ 1317 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1318 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1319 1320 /* Contain the header of the wrong received pdu */ 1321 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1322 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1323 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1324 } 1325 1326 static void 1327 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1328 struct spdk_nvmf_tcp_qpair *tqpair, 1329 struct nvme_tcp_pdu *pdu) 1330 { 1331 struct spdk_nvmf_tcp_req *tcp_req; 1332 1333 assert(pdu->psh_valid_bytes == pdu->psh_len); 1334 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1335 1336 tcp_req = nvmf_tcp_req_get(tqpair); 1337 if (!tcp_req) { 1338 /* Directly return and make the allocation retry again */ 1339 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1340 return; 1341 } 1342 1343 /* The host sent more commands than the maximum queue depth. */ 1344 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1345 nvmf_tcp_qpair_disconnect(tqpair); 1346 return; 1347 } 1348 1349 pdu->req = tcp_req; 1350 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1351 nvmf_tcp_req_process(ttransport, tcp_req); 1352 } 1353 1354 static void 1355 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1356 struct spdk_nvmf_tcp_qpair *tqpair, 1357 struct nvme_tcp_pdu *pdu) 1358 { 1359 struct spdk_nvmf_tcp_req *tcp_req; 1360 struct spdk_nvme_tcp_cmd *capsule_cmd; 1361 uint32_t error_offset = 0; 1362 enum spdk_nvme_tcp_term_req_fes fes; 1363 struct spdk_nvme_cpl *rsp; 1364 1365 capsule_cmd = &pdu->hdr.capsule_cmd; 1366 tcp_req = pdu->req; 1367 assert(tcp_req != NULL); 1368 1369 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1370 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1371 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1372 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1373 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1374 goto err; 1375 } 1376 1377 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1378 1379 rsp = &tcp_req->req.rsp->nvme_cpl; 1380 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1381 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1382 } else { 1383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1384 } 1385 1386 nvmf_tcp_req_process(ttransport, tcp_req); 1387 1388 return; 1389 err: 1390 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1391 } 1392 1393 static int 1394 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1395 enum spdk_nvmf_tcp_req_state state, 1396 uint16_t cid, uint16_t tag, 1397 struct spdk_nvmf_tcp_req **req) 1398 { 1399 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1400 1401 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1402 if (tcp_req->state != state) { 1403 continue; 1404 } 1405 1406 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1407 continue; 1408 } 1409 1410 if (tcp_req->ttag == tag) { 1411 *req = tcp_req; 1412 return 0; 1413 } 1414 1415 *req = NULL; 1416 return -1; 1417 } 1418 1419 /* Didn't find it, but not an error */ 1420 *req = NULL; 1421 return 0; 1422 } 1423 1424 static void 1425 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1426 struct spdk_nvmf_tcp_qpair *tqpair, 1427 struct nvme_tcp_pdu *pdu) 1428 { 1429 struct spdk_nvmf_tcp_req *tcp_req; 1430 uint32_t error_offset = 0; 1431 enum spdk_nvme_tcp_term_req_fes fes = 0; 1432 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1433 int rc; 1434 1435 h2c_data = &pdu->hdr.h2c_data; 1436 1437 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1438 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1439 1440 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1441 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1442 if (rc == 0 && tcp_req == NULL) { 1443 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1444 h2c_data->ttag, &tcp_req); 1445 } 1446 1447 if (!tcp_req) { 1448 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1449 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1450 if (rc == 0) { 1451 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1452 } else { 1453 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1454 } 1455 goto err; 1456 } 1457 1458 if (tcp_req->h2c_offset != h2c_data->datao) { 1459 SPDK_DEBUGLOG(nvmf_tcp, 1460 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1461 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1462 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1463 goto err; 1464 } 1465 1466 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1467 SPDK_DEBUGLOG(nvmf_tcp, 1468 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1469 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1470 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1471 goto err; 1472 } 1473 1474 pdu->req = tcp_req; 1475 1476 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1477 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1478 } 1479 1480 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1481 h2c_data->datao, h2c_data->datal); 1482 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1483 return; 1484 1485 err: 1486 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1487 } 1488 1489 static void 1490 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1491 struct spdk_nvmf_tcp_qpair *tqpair) 1492 { 1493 struct nvme_tcp_pdu *rsp_pdu; 1494 struct spdk_nvme_tcp_rsp *capsule_resp; 1495 1496 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1497 1498 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1499 assert(rsp_pdu != NULL); 1500 1501 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1502 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1503 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1504 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1505 if (tqpair->host_hdgst_enable) { 1506 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1507 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1508 } 1509 1510 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1511 } 1512 1513 static void 1514 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1515 { 1516 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1517 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1518 struct spdk_nvmf_tcp_qpair, qpair); 1519 1520 assert(tqpair != NULL); 1521 1522 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1523 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1524 tcp_req->req.length); 1525 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1526 return; 1527 } 1528 1529 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1530 nvmf_tcp_request_free(tcp_req); 1531 } else { 1532 nvmf_tcp_req_pdu_fini(tcp_req); 1533 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1534 } 1535 } 1536 1537 static void 1538 nvmf_tcp_r2t_complete(void *cb_arg) 1539 { 1540 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1541 struct spdk_nvmf_tcp_transport *ttransport; 1542 1543 nvmf_tcp_req_pdu_fini(tcp_req); 1544 1545 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1546 struct spdk_nvmf_tcp_transport, transport); 1547 1548 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1549 1550 if (tcp_req->h2c_offset == tcp_req->req.length) { 1551 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1552 nvmf_tcp_req_process(ttransport, tcp_req); 1553 } 1554 } 1555 1556 static void 1557 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1558 struct spdk_nvmf_tcp_req *tcp_req) 1559 { 1560 struct nvme_tcp_pdu *rsp_pdu; 1561 struct spdk_nvme_tcp_r2t_hdr *r2t; 1562 1563 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1564 assert(rsp_pdu != NULL); 1565 1566 r2t = &rsp_pdu->hdr.r2t; 1567 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1568 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1569 1570 if (tqpair->host_hdgst_enable) { 1571 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1572 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1573 } 1574 1575 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1576 r2t->ttag = tcp_req->ttag; 1577 r2t->r2to = tcp_req->h2c_offset; 1578 r2t->r2tl = tcp_req->req.length; 1579 1580 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1581 1582 SPDK_DEBUGLOG(nvmf_tcp, 1583 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1584 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1585 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1586 } 1587 1588 static void 1589 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1590 struct spdk_nvmf_tcp_qpair *tqpair, 1591 struct nvme_tcp_pdu *pdu) 1592 { 1593 struct spdk_nvmf_tcp_req *tcp_req; 1594 struct spdk_nvme_cpl *rsp; 1595 1596 tcp_req = pdu->req; 1597 assert(tcp_req != NULL); 1598 1599 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1600 1601 tcp_req->h2c_offset += pdu->data_len; 1602 1603 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1604 1605 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1606 * acknowledged before moving on. */ 1607 if (tcp_req->h2c_offset == tcp_req->req.length && 1608 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1609 /* After receving all the h2c data, we need to check whether there is 1610 * transient transport error */ 1611 rsp = &tcp_req->req.rsp->nvme_cpl; 1612 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1613 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1614 } else { 1615 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1616 } 1617 nvmf_tcp_req_process(ttransport, tcp_req); 1618 } 1619 } 1620 1621 static void 1622 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1623 { 1624 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1625 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1626 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1627 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1628 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1629 DGET32(h2c_term_req->fei)); 1630 } 1631 } 1632 1633 static void 1634 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1635 struct nvme_tcp_pdu *pdu) 1636 { 1637 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1638 uint32_t error_offset = 0; 1639 enum spdk_nvme_tcp_term_req_fes fes; 1640 1641 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1642 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1643 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1644 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1645 goto end; 1646 } 1647 1648 /* set the data buffer */ 1649 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1650 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1651 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1652 return; 1653 end: 1654 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1655 } 1656 1657 static void 1658 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1659 struct nvme_tcp_pdu *pdu) 1660 { 1661 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1662 1663 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1664 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1665 } 1666 1667 static void 1668 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1669 struct spdk_nvmf_tcp_transport *ttransport) 1670 { 1671 struct nvme_tcp_pdu *pdu = tqpair->pdu_in_progress; 1672 1673 switch (pdu->hdr.common.pdu_type) { 1674 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1675 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1676 break; 1677 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1678 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1679 break; 1680 1681 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1682 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1683 break; 1684 1685 default: 1686 /* The code should not go to here */ 1687 SPDK_ERRLOG("The code should not go to here\n"); 1688 break; 1689 } 1690 } 1691 1692 static void 1693 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1694 struct spdk_nvmf_tcp_transport *ttransport) 1695 { 1696 int rc = 0; 1697 struct nvme_tcp_pdu *pdu; 1698 uint32_t crc32c; 1699 struct spdk_nvmf_tcp_req *tcp_req; 1700 struct spdk_nvme_cpl *rsp; 1701 1702 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1703 pdu = tqpair->pdu_in_progress; 1704 1705 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1706 /* check data digest if need */ 1707 if (pdu->ddgst_enable) { 1708 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1709 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1710 if (rc == 0) { 1711 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1712 tcp_req = pdu->req; 1713 assert(tcp_req != NULL); 1714 rsp = &tcp_req->req.rsp->nvme_cpl; 1715 rsp->status.sc = SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR; 1716 } 1717 } 1718 1719 _nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1720 } 1721 1722 static void 1723 nvmf_tcp_send_icresp_complete(void *cb_arg) 1724 { 1725 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1726 1727 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1728 } 1729 1730 static void 1731 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1732 struct spdk_nvmf_tcp_qpair *tqpair, 1733 struct nvme_tcp_pdu *pdu) 1734 { 1735 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1736 struct nvme_tcp_pdu *rsp_pdu; 1737 struct spdk_nvme_tcp_ic_resp *ic_resp; 1738 uint32_t error_offset = 0; 1739 enum spdk_nvme_tcp_term_req_fes fes; 1740 1741 /* Only PFV 0 is defined currently */ 1742 if (ic_req->pfv != 0) { 1743 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1744 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1745 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1746 goto end; 1747 } 1748 1749 /* MAXR2T is 0's based */ 1750 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1751 1752 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1753 if (!tqpair->host_hdgst_enable) { 1754 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1755 } 1756 1757 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1758 if (!tqpair->host_ddgst_enable) { 1759 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1760 } 1761 1762 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1763 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1764 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1765 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1766 tqpair, 1767 tqpair->recv_buf_size); 1768 /* Not fatal. */ 1769 } 1770 1771 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1772 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1773 1774 rsp_pdu = tqpair->mgmt_pdu; 1775 1776 ic_resp = &rsp_pdu->hdr.ic_resp; 1777 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1778 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1779 ic_resp->pfv = 0; 1780 ic_resp->cpda = tqpair->cpda; 1781 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1782 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1783 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1784 1785 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1786 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1787 1788 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1789 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1790 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1791 return; 1792 end: 1793 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1794 } 1795 1796 static void 1797 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1798 struct spdk_nvmf_tcp_transport *ttransport) 1799 { 1800 struct nvme_tcp_pdu *pdu; 1801 int rc; 1802 uint32_t crc32c, error_offset = 0; 1803 enum spdk_nvme_tcp_term_req_fes fes; 1804 1805 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1806 pdu = tqpair->pdu_in_progress; 1807 1808 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1809 pdu->hdr.common.pdu_type); 1810 /* check header digest if needed */ 1811 if (pdu->has_hdgst) { 1812 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1813 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1814 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1815 if (rc == 0) { 1816 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1817 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1818 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1819 return; 1820 1821 } 1822 } 1823 1824 switch (pdu->hdr.common.pdu_type) { 1825 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1826 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1827 break; 1828 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1829 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1830 break; 1831 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1832 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1833 break; 1834 1835 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1836 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1837 break; 1838 1839 default: 1840 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 1841 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1842 error_offset = 1; 1843 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1844 break; 1845 } 1846 } 1847 1848 static void 1849 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1850 { 1851 struct nvme_tcp_pdu *pdu; 1852 uint32_t error_offset = 0; 1853 enum spdk_nvme_tcp_term_req_fes fes; 1854 uint8_t expected_hlen, pdo; 1855 bool plen_error = false, pdo_error = false; 1856 1857 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1858 pdu = tqpair->pdu_in_progress; 1859 1860 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1861 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1862 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1863 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1864 goto err; 1865 } 1866 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1867 if (pdu->hdr.common.plen != expected_hlen) { 1868 plen_error = true; 1869 } 1870 } else { 1871 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1872 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1873 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1874 goto err; 1875 } 1876 1877 switch (pdu->hdr.common.pdu_type) { 1878 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1879 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1880 pdo = pdu->hdr.common.pdo; 1881 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1882 pdo_error = true; 1883 break; 1884 } 1885 1886 if (pdu->hdr.common.plen < expected_hlen) { 1887 plen_error = true; 1888 } 1889 break; 1890 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1891 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1892 pdo = pdu->hdr.common.pdo; 1893 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1894 pdo_error = true; 1895 break; 1896 } 1897 if (pdu->hdr.common.plen < expected_hlen) { 1898 plen_error = true; 1899 } 1900 break; 1901 1902 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1903 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1904 if ((pdu->hdr.common.plen <= expected_hlen) || 1905 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1906 plen_error = true; 1907 } 1908 break; 1909 1910 default: 1911 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1912 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1913 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1914 goto err; 1915 } 1916 } 1917 1918 if (pdu->hdr.common.hlen != expected_hlen) { 1919 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1920 pdu->hdr.common.pdu_type, 1921 expected_hlen, pdu->hdr.common.hlen, tqpair); 1922 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1923 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1924 goto err; 1925 } else if (pdo_error) { 1926 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1927 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1928 } else if (plen_error) { 1929 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1930 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1931 goto err; 1932 } else { 1933 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1934 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1935 return; 1936 } 1937 err: 1938 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1939 } 1940 1941 static int 1942 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1943 int read_len) 1944 { 1945 int rc; 1946 1947 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1948 read_offset, read_len, pdu->dif_ctx); 1949 if (rc != 0) { 1950 SPDK_ERRLOG("DIF generate failed\n"); 1951 } 1952 1953 return rc; 1954 } 1955 1956 static int 1957 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1958 { 1959 int rc = 0; 1960 struct nvme_tcp_pdu *pdu; 1961 enum nvme_tcp_pdu_recv_state prev_state; 1962 uint32_t data_len; 1963 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1964 struct spdk_nvmf_tcp_transport, transport); 1965 1966 /* The loop here is to allow for several back-to-back state changes. */ 1967 do { 1968 prev_state = tqpair->recv_state; 1969 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1970 1971 pdu = tqpair->pdu_in_progress; 1972 switch (tqpair->recv_state) { 1973 /* Wait for the common header */ 1974 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1975 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1976 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1977 return rc; 1978 } 1979 1980 rc = nvme_tcp_read_data(tqpair->sock, 1981 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1982 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1983 if (rc < 0) { 1984 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1985 return NVME_TCP_PDU_FATAL; 1986 } else if (rc > 0) { 1987 pdu->ch_valid_bytes += rc; 1988 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0); 1989 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1990 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1991 } 1992 } 1993 1994 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1995 return NVME_TCP_PDU_IN_PROGRESS; 1996 } 1997 1998 /* The command header of this PDU has now been read from the socket. */ 1999 nvmf_tcp_pdu_ch_handle(tqpair); 2000 break; 2001 /* Wait for the pdu specific header */ 2002 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2003 rc = nvme_tcp_read_data(tqpair->sock, 2004 pdu->psh_len - pdu->psh_valid_bytes, 2005 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2006 if (rc < 0) { 2007 return NVME_TCP_PDU_FATAL; 2008 } else if (rc > 0) { 2009 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0); 2010 pdu->psh_valid_bytes += rc; 2011 } 2012 2013 if (pdu->psh_valid_bytes < pdu->psh_len) { 2014 return NVME_TCP_PDU_IN_PROGRESS; 2015 } 2016 2017 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2018 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2019 break; 2020 /* Wait for the req slot */ 2021 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2022 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2023 break; 2024 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2025 /* check whether the data is valid, if not we just return */ 2026 if (!pdu->data_len) { 2027 return NVME_TCP_PDU_IN_PROGRESS; 2028 } 2029 2030 data_len = pdu->data_len; 2031 /* data digest */ 2032 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2033 tqpair->host_ddgst_enable)) { 2034 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2035 pdu->ddgst_enable = true; 2036 } 2037 2038 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2039 if (rc < 0) { 2040 return NVME_TCP_PDU_FATAL; 2041 } 2042 pdu->rw_offset += rc; 2043 2044 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2045 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2046 if (rc != 0) { 2047 return NVME_TCP_PDU_FATAL; 2048 } 2049 } 2050 2051 if (pdu->rw_offset < data_len) { 2052 return NVME_TCP_PDU_IN_PROGRESS; 2053 } 2054 2055 /* All of this PDU has now been read from the socket. */ 2056 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2057 break; 2058 case NVME_TCP_PDU_RECV_STATE_ERROR: 2059 if (!spdk_sock_is_connected(tqpair->sock)) { 2060 return NVME_TCP_PDU_FATAL; 2061 } 2062 break; 2063 default: 2064 assert(0); 2065 SPDK_ERRLOG("code should not come to here"); 2066 break; 2067 } 2068 } while (tqpair->recv_state != prev_state); 2069 2070 return rc; 2071 } 2072 2073 static inline void * 2074 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2075 { 2076 struct spdk_nvmf_tcp_control_msg *msg; 2077 2078 assert(list); 2079 2080 msg = STAILQ_FIRST(&list->free_msgs); 2081 if (!msg) { 2082 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2083 return NULL; 2084 } 2085 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2086 return msg; 2087 } 2088 2089 static inline void 2090 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2091 { 2092 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2093 2094 assert(list); 2095 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2096 } 2097 2098 static int 2099 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2100 struct spdk_nvmf_transport *transport, 2101 struct spdk_nvmf_transport_poll_group *group) 2102 { 2103 struct spdk_nvmf_request *req = &tcp_req->req; 2104 struct spdk_nvme_cmd *cmd; 2105 struct spdk_nvme_cpl *rsp; 2106 struct spdk_nvme_sgl_descriptor *sgl; 2107 struct spdk_nvmf_tcp_poll_group *tgroup; 2108 uint32_t length; 2109 2110 cmd = &req->cmd->nvme_cmd; 2111 rsp = &req->rsp->nvme_cpl; 2112 sgl = &cmd->dptr.sgl1; 2113 2114 length = sgl->unkeyed.length; 2115 2116 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2117 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2118 if (length > transport->opts.max_io_size) { 2119 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2120 length, transport->opts.max_io_size); 2121 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2122 return -1; 2123 } 2124 2125 /* fill request length and populate iovs */ 2126 req->length = length; 2127 2128 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2129 2130 if (spdk_unlikely(req->dif_enabled)) { 2131 req->dif.orig_length = length; 2132 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2133 req->dif.elba_length = length; 2134 } 2135 2136 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2137 /* No available buffers. Queue this request up. */ 2138 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2139 tcp_req); 2140 return 0; 2141 } 2142 2143 /* backward compatible */ 2144 req->data = req->iov[0].iov_base; 2145 2146 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2147 tcp_req, req->iovcnt, req->data); 2148 2149 return 0; 2150 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2151 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2152 uint64_t offset = sgl->address; 2153 uint32_t max_len = transport->opts.in_capsule_data_size; 2154 assert(tcp_req->has_incapsule_data); 2155 2156 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2157 offset, length); 2158 2159 if (offset > max_len) { 2160 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2161 offset, max_len); 2162 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2163 return -1; 2164 } 2165 max_len -= (uint32_t)offset; 2166 2167 if (spdk_unlikely(length > max_len)) { 2168 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2169 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2170 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2171 2172 /* Get a buffer from dedicated list */ 2173 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2174 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2175 assert(tgroup->control_msg_list); 2176 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2177 if (!req->data) { 2178 /* No available buffers. Queue this request up. */ 2179 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2180 return 0; 2181 } 2182 } else { 2183 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2184 length, max_len); 2185 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2186 return -1; 2187 } 2188 } else { 2189 req->data = tcp_req->buf; 2190 } 2191 2192 req->length = length; 2193 req->data_from_pool = false; 2194 2195 if (spdk_unlikely(req->dif_enabled)) { 2196 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2197 req->dif.elba_length = length; 2198 } 2199 2200 req->iov[0].iov_base = req->data; 2201 req->iov[0].iov_len = length; 2202 req->iovcnt = 1; 2203 2204 return 0; 2205 } 2206 2207 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2208 sgl->generic.type, sgl->generic.subtype); 2209 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2210 return -1; 2211 } 2212 2213 static inline enum spdk_nvme_media_error_status_code 2214 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2215 enum spdk_nvme_media_error_status_code result; 2216 2217 switch (err_type) 2218 { 2219 case SPDK_DIF_REFTAG_ERROR: 2220 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2221 break; 2222 case SPDK_DIF_APPTAG_ERROR: 2223 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2224 break; 2225 case SPDK_DIF_GUARD_ERROR: 2226 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2227 break; 2228 default: 2229 SPDK_UNREACHABLE(); 2230 break; 2231 } 2232 2233 return result; 2234 } 2235 2236 static void 2237 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2238 struct spdk_nvmf_tcp_req *tcp_req) 2239 { 2240 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2241 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2242 struct nvme_tcp_pdu *rsp_pdu; 2243 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2244 uint32_t plen, pdo, alignment; 2245 int rc; 2246 2247 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2248 2249 rsp_pdu = tcp_req->pdu; 2250 assert(rsp_pdu != NULL); 2251 assert(tcp_req->pdu_in_use); 2252 2253 c2h_data = &rsp_pdu->hdr.c2h_data; 2254 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2255 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2256 2257 if (tqpair->host_hdgst_enable) { 2258 plen += SPDK_NVME_TCP_DIGEST_LEN; 2259 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2260 } 2261 2262 /* set the psh */ 2263 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2264 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2265 c2h_data->datao = tcp_req->pdu->rw_offset; 2266 2267 /* set the padding */ 2268 rsp_pdu->padding_len = 0; 2269 pdo = plen; 2270 if (tqpair->cpda) { 2271 alignment = (tqpair->cpda + 1) << 2; 2272 if (plen % alignment != 0) { 2273 pdo = (plen + alignment) / alignment * alignment; 2274 rsp_pdu->padding_len = pdo - plen; 2275 plen = pdo; 2276 } 2277 } 2278 2279 c2h_data->common.pdo = pdo; 2280 plen += c2h_data->datal; 2281 if (tqpair->host_ddgst_enable) { 2282 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2283 plen += SPDK_NVME_TCP_DIGEST_LEN; 2284 } 2285 2286 c2h_data->common.plen = plen; 2287 2288 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2289 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2290 } 2291 2292 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2293 c2h_data->datao, c2h_data->datal); 2294 2295 2296 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2297 /* Need to send the capsule response if response is not all 0 */ 2298 if (ttransport->tcp_opts.c2h_success && 2299 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2300 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2301 } 2302 2303 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2304 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2305 struct spdk_dif_error err_blk = {}; 2306 uint32_t mapped_length = 0; 2307 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2308 uint32_t ddgst_len = 0; 2309 2310 if (tqpair->host_ddgst_enable) { 2311 /* Data digest consumes additional iov entry */ 2312 available_iovs--; 2313 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2314 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2315 c2h_data->common.plen -= ddgst_len; 2316 } 2317 /* Temp call to estimate if data can be described by limited number of iovs. 2318 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2319 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2320 false, &mapped_length); 2321 2322 if (mapped_length != c2h_data->common.plen) { 2323 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2324 SPDK_DEBUGLOG(nvmf_tcp, 2325 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2326 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2327 c2h_data->common.plen = mapped_length; 2328 2329 /* Rebuild pdu->data_iov since data length is changed */ 2330 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2331 c2h_data->datal); 2332 2333 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2334 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2335 } 2336 2337 c2h_data->common.plen += ddgst_len; 2338 2339 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2340 2341 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2342 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2343 if (rc != 0) { 2344 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2345 err_blk.err_type, err_blk.err_offset); 2346 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2347 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2348 nvmf_tcp_req_pdu_fini(tcp_req); 2349 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2350 return; 2351 } 2352 } 2353 2354 rsp_pdu->rw_offset += c2h_data->datal; 2355 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2356 } 2357 2358 static void 2359 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2360 struct spdk_nvmf_tcp_req *tcp_req) 2361 { 2362 nvmf_tcp_req_pdu_init(tcp_req); 2363 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2364 } 2365 2366 static int 2367 request_transfer_out(struct spdk_nvmf_request *req) 2368 { 2369 struct spdk_nvmf_tcp_req *tcp_req; 2370 struct spdk_nvmf_qpair *qpair; 2371 struct spdk_nvmf_tcp_qpair *tqpair; 2372 struct spdk_nvme_cpl *rsp; 2373 2374 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2375 2376 qpair = req->qpair; 2377 rsp = &req->rsp->nvme_cpl; 2378 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2379 2380 /* Advance our sq_head pointer */ 2381 if (qpair->sq_head == qpair->sq_head_max) { 2382 qpair->sq_head = 0; 2383 } else { 2384 qpair->sq_head++; 2385 } 2386 rsp->sqhd = qpair->sq_head; 2387 2388 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2389 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2390 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2391 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2392 } else { 2393 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2394 } 2395 2396 return 0; 2397 } 2398 2399 static void 2400 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2401 struct spdk_nvmf_tcp_req *tcp_req) 2402 { 2403 struct nvme_tcp_pdu *pdu; 2404 uint32_t plen = 0; 2405 2406 pdu = tqpair->pdu_in_progress; 2407 plen = pdu->hdr.common.hlen; 2408 2409 if (tqpair->host_hdgst_enable) { 2410 plen += SPDK_NVME_TCP_DIGEST_LEN; 2411 } 2412 2413 if (pdu->hdr.common.plen != plen) { 2414 tcp_req->has_incapsule_data = true; 2415 } 2416 } 2417 2418 static bool 2419 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2420 struct spdk_nvmf_tcp_req *tcp_req) 2421 { 2422 struct spdk_nvmf_tcp_qpair *tqpair; 2423 int rc; 2424 enum spdk_nvmf_tcp_req_state prev_state; 2425 bool progress = false; 2426 struct spdk_nvmf_transport *transport = &ttransport->transport; 2427 struct spdk_nvmf_transport_poll_group *group; 2428 struct spdk_nvmf_tcp_poll_group *tgroup; 2429 2430 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2431 group = &tqpair->group->group; 2432 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2433 2434 /* If the qpair is not active, we need to abort the outstanding requests. */ 2435 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2436 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2437 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2438 } 2439 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2440 } 2441 2442 /* The loop here is to allow for several back-to-back state changes. */ 2443 do { 2444 prev_state = tcp_req->state; 2445 2446 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2447 tqpair); 2448 2449 switch (tcp_req->state) { 2450 case TCP_REQUEST_STATE_FREE: 2451 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2452 * to escape this state. */ 2453 break; 2454 case TCP_REQUEST_STATE_NEW: 2455 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req); 2456 2457 /* copy the cmd from the receive pdu */ 2458 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2459 2460 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2461 tcp_req->req.dif_enabled = true; 2462 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2463 } 2464 2465 /* The next state transition depends on the data transfer needs of this request. */ 2466 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2467 2468 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2469 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2470 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 2471 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2472 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2473 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2474 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2475 break; 2476 } 2477 2478 /* If no data to transfer, ready to execute. */ 2479 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2480 /* Reset the tqpair receving pdu state */ 2481 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2482 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2483 break; 2484 } 2485 2486 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2487 2488 if (!tcp_req->has_incapsule_data) { 2489 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2490 } 2491 2492 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2493 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2494 break; 2495 case TCP_REQUEST_STATE_NEED_BUFFER: 2496 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req); 2497 2498 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2499 2500 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2501 SPDK_DEBUGLOG(nvmf_tcp, 2502 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2503 tcp_req, tqpair); 2504 /* This request needs to wait in line to obtain a buffer */ 2505 break; 2506 } 2507 2508 /* Try to get a data buffer */ 2509 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2510 if (rc < 0) { 2511 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2512 /* Reset the tqpair receving pdu state */ 2513 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2514 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2515 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2516 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2517 break; 2518 } 2519 2520 if (!tcp_req->req.data) { 2521 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2522 tcp_req, tqpair); 2523 /* No buffers available. */ 2524 break; 2525 } 2526 2527 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2528 2529 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2530 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2531 if (tcp_req->req.data_from_pool) { 2532 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2533 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2534 } else { 2535 struct nvme_tcp_pdu *pdu; 2536 2537 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2538 2539 pdu = tqpair->pdu_in_progress; 2540 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2541 tqpair); 2542 /* No need to send r2t, contained in the capsuled data */ 2543 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2544 0, tcp_req->req.length); 2545 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2546 } 2547 break; 2548 } 2549 2550 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2551 break; 2552 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2553 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req); 2554 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2555 break; 2556 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2557 2558 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2559 (uintptr_t)tcp_req); 2560 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2561 * to escape this state. */ 2562 break; 2563 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2564 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req); 2565 2566 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2567 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2568 tcp_req->req.length = tcp_req->req.dif.elba_length; 2569 } 2570 2571 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2572 spdk_nvmf_request_exec(&tcp_req->req); 2573 break; 2574 case TCP_REQUEST_STATE_EXECUTING: 2575 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req); 2576 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2577 * to escape this state. */ 2578 break; 2579 case TCP_REQUEST_STATE_EXECUTED: 2580 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req); 2581 2582 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2583 tcp_req->req.length = tcp_req->req.dif.orig_length; 2584 } 2585 2586 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2587 break; 2588 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2589 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req); 2590 rc = request_transfer_out(&tcp_req->req); 2591 assert(rc == 0); /* No good way to handle this currently */ 2592 break; 2593 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2594 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2595 (uintptr_t)tcp_req); 2596 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2597 * to escape this state. */ 2598 break; 2599 case TCP_REQUEST_STATE_COMPLETED: 2600 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req); 2601 if (tcp_req->req.data_from_pool) { 2602 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2603 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2604 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2605 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2606 assert(tgroup->control_msg_list); 2607 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2608 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2609 } 2610 tcp_req->req.length = 0; 2611 tcp_req->req.iovcnt = 0; 2612 tcp_req->req.data = NULL; 2613 2614 nvmf_tcp_req_pdu_fini(tcp_req); 2615 2616 nvmf_tcp_req_put(tqpair, tcp_req); 2617 break; 2618 case TCP_REQUEST_NUM_STATES: 2619 default: 2620 assert(0); 2621 break; 2622 } 2623 2624 if (tcp_req->state != prev_state) { 2625 progress = true; 2626 } 2627 } while (tcp_req->state != prev_state); 2628 2629 return progress; 2630 } 2631 2632 static void 2633 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2634 { 2635 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2636 int rc; 2637 2638 assert(tqpair != NULL); 2639 rc = nvmf_tcp_sock_process(tqpair); 2640 2641 /* If there was a new socket error, disconnect */ 2642 if (rc < 0) { 2643 nvmf_tcp_qpair_disconnect(tqpair); 2644 } 2645 } 2646 2647 static int 2648 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2649 struct spdk_nvmf_qpair *qpair) 2650 { 2651 struct spdk_nvmf_tcp_poll_group *tgroup; 2652 struct spdk_nvmf_tcp_qpair *tqpair; 2653 int rc; 2654 2655 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2656 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2657 2658 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2659 nvmf_tcp_sock_cb, tqpair); 2660 if (rc != 0) { 2661 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2662 spdk_strerror(errno), errno); 2663 return -1; 2664 } 2665 2666 rc = nvmf_tcp_qpair_sock_init(tqpair); 2667 if (rc != 0) { 2668 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2669 return -1; 2670 } 2671 2672 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2673 if (rc < 0) { 2674 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2675 return -1; 2676 } 2677 2678 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2679 if (rc < 0) { 2680 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2681 return -1; 2682 } 2683 2684 tqpair->group = tgroup; 2685 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2686 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2687 2688 return 0; 2689 } 2690 2691 static int 2692 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2693 struct spdk_nvmf_qpair *qpair) 2694 { 2695 struct spdk_nvmf_tcp_poll_group *tgroup; 2696 struct spdk_nvmf_tcp_qpair *tqpair; 2697 int rc; 2698 2699 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2700 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2701 2702 assert(tqpair->group == tgroup); 2703 2704 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2705 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2706 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2707 } else { 2708 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2709 } 2710 2711 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2712 if (rc != 0) { 2713 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2714 spdk_strerror(errno), errno); 2715 } 2716 2717 return rc; 2718 } 2719 2720 static int 2721 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2722 { 2723 struct spdk_nvmf_tcp_transport *ttransport; 2724 struct spdk_nvmf_tcp_req *tcp_req; 2725 2726 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2727 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2728 2729 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2730 nvmf_tcp_req_process(ttransport, tcp_req); 2731 2732 return 0; 2733 } 2734 2735 static void 2736 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2737 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2738 { 2739 struct spdk_nvmf_tcp_qpair *tqpair; 2740 2741 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2742 2743 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2744 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2745 nvmf_tcp_qpair_destroy(tqpair); 2746 2747 if (cb_fn) { 2748 cb_fn(cb_arg); 2749 } 2750 } 2751 2752 static int 2753 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2754 { 2755 struct spdk_nvmf_tcp_poll_group *tgroup; 2756 int rc; 2757 struct spdk_nvmf_request *req, *req_tmp; 2758 struct spdk_nvmf_tcp_req *tcp_req; 2759 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2760 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2761 struct spdk_nvmf_tcp_transport, transport); 2762 2763 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2764 2765 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2766 return 0; 2767 } 2768 2769 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2770 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2771 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2772 break; 2773 } 2774 } 2775 2776 rc = spdk_sock_group_poll(tgroup->sock_group); 2777 if (rc < 0) { 2778 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2779 } 2780 2781 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2782 nvmf_tcp_sock_process(tqpair); 2783 } 2784 2785 return rc; 2786 } 2787 2788 static int 2789 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2790 struct spdk_nvme_transport_id *trid, bool peer) 2791 { 2792 struct spdk_nvmf_tcp_qpair *tqpair; 2793 uint16_t port; 2794 2795 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2796 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2797 2798 if (peer) { 2799 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2800 port = tqpair->initiator_port; 2801 } else { 2802 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2803 port = tqpair->target_port; 2804 } 2805 2806 if (spdk_sock_is_ipv4(tqpair->sock)) { 2807 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2808 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2809 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2810 } else { 2811 return -1; 2812 } 2813 2814 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2815 return 0; 2816 } 2817 2818 static int 2819 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2820 struct spdk_nvme_transport_id *trid) 2821 { 2822 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2823 } 2824 2825 static int 2826 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2827 struct spdk_nvme_transport_id *trid) 2828 { 2829 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2830 } 2831 2832 static int 2833 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2834 struct spdk_nvme_transport_id *trid) 2835 { 2836 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2837 } 2838 2839 static void 2840 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2841 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2842 { 2843 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2844 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2845 2846 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2847 2848 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2849 } 2850 2851 static int 2852 _nvmf_tcp_qpair_abort_request(void *ctx) 2853 { 2854 struct spdk_nvmf_request *req = ctx; 2855 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2856 struct spdk_nvmf_tcp_req, req); 2857 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2858 struct spdk_nvmf_tcp_qpair, qpair); 2859 int rc; 2860 2861 spdk_poller_unregister(&req->poller); 2862 2863 switch (tcp_req_to_abort->state) { 2864 case TCP_REQUEST_STATE_EXECUTING: 2865 rc = nvmf_ctrlr_abort_request(req); 2866 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2867 return SPDK_POLLER_BUSY; 2868 } 2869 break; 2870 2871 case TCP_REQUEST_STATE_NEED_BUFFER: 2872 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2873 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2874 2875 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2876 break; 2877 2878 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2879 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2880 break; 2881 2882 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2883 if (spdk_get_ticks() < req->timeout_tsc) { 2884 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2885 return SPDK_POLLER_BUSY; 2886 } 2887 break; 2888 2889 default: 2890 break; 2891 } 2892 2893 spdk_nvmf_request_complete(req); 2894 return SPDK_POLLER_BUSY; 2895 } 2896 2897 static void 2898 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2899 struct spdk_nvmf_request *req) 2900 { 2901 struct spdk_nvmf_tcp_qpair *tqpair; 2902 struct spdk_nvmf_tcp_transport *ttransport; 2903 struct spdk_nvmf_transport *transport; 2904 uint16_t cid; 2905 uint32_t i; 2906 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2907 2908 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2909 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2910 transport = &ttransport->transport; 2911 2912 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2913 2914 for (i = 0; i < tqpair->resource_count; i++) { 2915 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2916 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2917 tcp_req_to_abort = &tqpair->reqs[i]; 2918 break; 2919 } 2920 } 2921 2922 if (tcp_req_to_abort == NULL) { 2923 spdk_nvmf_request_complete(req); 2924 return; 2925 } 2926 2927 req->req_to_abort = &tcp_req_to_abort->req; 2928 req->timeout_tsc = spdk_get_ticks() + 2929 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2930 req->poller = NULL; 2931 2932 _nvmf_tcp_qpair_abort_request(req); 2933 } 2934 2935 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2936 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2937 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2938 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2939 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2940 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2941 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2942 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2943 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2944 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2945 2946 static void 2947 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2948 { 2949 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2950 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2951 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2952 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2953 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2954 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2955 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2956 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2957 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2958 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2959 opts->transport_specific = NULL; 2960 } 2961 2962 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2963 .name = "TCP", 2964 .type = SPDK_NVME_TRANSPORT_TCP, 2965 .opts_init = nvmf_tcp_opts_init, 2966 .create = nvmf_tcp_create, 2967 .dump_opts = nvmf_tcp_dump_opts, 2968 .destroy = nvmf_tcp_destroy, 2969 2970 .listen = nvmf_tcp_listen, 2971 .stop_listen = nvmf_tcp_stop_listen, 2972 .accept = nvmf_tcp_accept, 2973 2974 .listener_discover = nvmf_tcp_discover, 2975 2976 .poll_group_create = nvmf_tcp_poll_group_create, 2977 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2978 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2979 .poll_group_add = nvmf_tcp_poll_group_add, 2980 .poll_group_remove = nvmf_tcp_poll_group_remove, 2981 .poll_group_poll = nvmf_tcp_poll_group_poll, 2982 2983 .req_free = nvmf_tcp_req_free, 2984 .req_complete = nvmf_tcp_req_complete, 2985 2986 .qpair_fini = nvmf_tcp_close_qpair, 2987 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2988 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2989 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2990 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2991 }; 2992 2993 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2994 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2995