1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (c) Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel_engine.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 19 #include "spdk_internal/assert.h" 20 #include "spdk_internal/nvme_tcp.h" 21 #include "spdk_internal/sock.h" 22 23 #include "nvmf_internal.h" 24 25 #include "spdk_internal/trace_defs.h" 26 27 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 28 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 29 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 30 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 31 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 32 33 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 34 35 /* spdk nvmf related structure */ 36 enum spdk_nvmf_tcp_req_state { 37 38 /* The request is not currently in use */ 39 TCP_REQUEST_STATE_FREE = 0, 40 41 /* Initial state when request first received */ 42 TCP_REQUEST_STATE_NEW = 1, 43 44 /* The request is queued until a data buffer is available. */ 45 TCP_REQUEST_STATE_NEED_BUFFER = 2, 46 47 /* The request is waiting for zcopy_start to finish */ 48 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 3, 49 50 /* The request has received a zero-copy buffer */ 51 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 4, 52 53 /* The request is currently transferring data from the host to the controller. */ 54 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 5, 55 56 /* The request is waiting for the R2T send acknowledgement. */ 57 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 6, 58 59 /* The request is ready to execute at the block device */ 60 TCP_REQUEST_STATE_READY_TO_EXECUTE = 7, 61 62 /* The request is currently executing at the block device */ 63 TCP_REQUEST_STATE_EXECUTING = 8, 64 65 /* The request is waiting for zcopy buffers to be commited */ 66 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 9, 67 68 /* The request finished executing at the block device */ 69 TCP_REQUEST_STATE_EXECUTED = 10, 70 71 /* The request is ready to send a completion */ 72 TCP_REQUEST_STATE_READY_TO_COMPLETE = 11, 73 74 /* The request is currently transferring final pdus from the controller to the host. */ 75 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 12, 76 77 /* The request is waiting for zcopy buffers to be released (without committing) */ 78 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 13, 79 80 /* The request completed and can be marked free. */ 81 TCP_REQUEST_STATE_COMPLETED = 14, 82 83 /* Terminator */ 84 TCP_REQUEST_NUM_STATES, 85 }; 86 87 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 88 "Invalid PDU Header Field", 89 "PDU Sequence Error", 90 "Header Digiest Error", 91 "Data Transfer Out of Range", 92 "R2T Limit Exceeded", 93 "Unsupported parameter", 94 }; 95 96 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 97 { 98 spdk_trace_register_owner(OWNER_NVMF_TCP, 't'); 99 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 100 spdk_trace_register_description("TCP_REQ_NEW", 101 TRACE_TCP_REQUEST_STATE_NEW, 102 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 103 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 104 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 105 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 106 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 107 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 108 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 109 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 110 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 111 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 112 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 113 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 114 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 115 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 116 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 117 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 118 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 119 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 120 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 121 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 122 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 123 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 124 spdk_trace_register_description("TCP_REQ_EXECUTING", 125 TRACE_TCP_REQUEST_STATE_EXECUTING, 126 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 127 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 128 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 129 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 130 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 131 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 132 spdk_trace_register_description("TCP_REQ_EXECUTED", 133 TRACE_TCP_REQUEST_STATE_EXECUTED, 134 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 135 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 136 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 137 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 138 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 139 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 140 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 141 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 142 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 143 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 144 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 145 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 146 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 147 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 148 spdk_trace_register_description("TCP_REQ_COMPLETED", 149 TRACE_TCP_REQUEST_STATE_COMPLETED, 150 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 151 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 152 spdk_trace_register_description("TCP_WRITE_START", 153 TRACE_TCP_FLUSH_WRITEBUF_START, 154 OWNER_NVMF_TCP, OBJECT_NONE, 0, 155 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 156 spdk_trace_register_description("TCP_WRITE_DONE", 157 TRACE_TCP_FLUSH_WRITEBUF_DONE, 158 OWNER_NVMF_TCP, OBJECT_NONE, 0, 159 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 160 spdk_trace_register_description("TCP_READ_DONE", 161 TRACE_TCP_READ_FROM_SOCKET_DONE, 162 OWNER_NVMF_TCP, OBJECT_NONE, 0, 163 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 164 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 165 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 166 OWNER_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 167 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 168 169 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 170 OWNER_NVMF_TCP, OBJECT_NONE, 0, 171 SPDK_TRACE_ARG_TYPE_INT, ""); 172 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 173 OWNER_NVMF_TCP, OBJECT_NONE, 0, 174 SPDK_TRACE_ARG_TYPE_INT, ""); 175 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 176 OWNER_NVMF_TCP, OBJECT_NONE, 0, 177 SPDK_TRACE_ARG_TYPE_INT, "state"); 178 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 179 OWNER_NVMF_TCP, OBJECT_NONE, 0, 180 SPDK_TRACE_ARG_TYPE_INT, ""); 181 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 182 OWNER_NVMF_TCP, OBJECT_NONE, 0, 183 SPDK_TRACE_ARG_TYPE_INT, ""); 184 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 185 OWNER_NVMF_TCP, OBJECT_NONE, 0, 186 SPDK_TRACE_ARG_TYPE_PTR, "qpair"); 187 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 188 OWNER_NVMF_TCP, OBJECT_NONE, 0, 189 SPDK_TRACE_ARG_TYPE_INT, "state"); 190 191 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 192 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 193 } 194 195 struct spdk_nvmf_tcp_req { 196 struct spdk_nvmf_request req; 197 struct spdk_nvme_cpl rsp; 198 struct spdk_nvme_cmd cmd; 199 200 /* A PDU that can be used for sending responses. This is 201 * not the incoming PDU! */ 202 struct nvme_tcp_pdu *pdu; 203 204 /* In-capsule data buffer */ 205 uint8_t *buf; 206 207 struct spdk_nvmf_tcp_req *fused_pair; 208 209 /* 210 * The PDU for a request may be used multiple times in serial over 211 * the request's lifetime. For example, first to send an R2T, then 212 * to send a completion. To catch mistakes where the PDU is used 213 * twice at the same time, add a debug flag here for init/fini. 214 */ 215 bool pdu_in_use; 216 bool has_in_capsule_data; 217 bool fused_failed; 218 219 /* transfer_tag */ 220 uint16_t ttag; 221 222 enum spdk_nvmf_tcp_req_state state; 223 224 /* 225 * h2c_offset is used when we receive the h2c_data PDU. 226 */ 227 uint32_t h2c_offset; 228 229 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 230 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 231 }; 232 233 struct spdk_nvmf_tcp_qpair { 234 struct spdk_nvmf_qpair qpair; 235 struct spdk_nvmf_tcp_poll_group *group; 236 struct spdk_sock *sock; 237 238 enum nvme_tcp_pdu_recv_state recv_state; 239 enum nvme_tcp_qpair_state state; 240 241 /* PDU being actively received */ 242 struct nvme_tcp_pdu *pdu_in_progress; 243 244 struct spdk_nvmf_tcp_req *fused_first; 245 246 /* Queues to track the requests in all states */ 247 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 248 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 249 250 /* Number of requests in each state */ 251 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 252 253 uint8_t cpda; 254 255 bool host_hdgst_enable; 256 bool host_ddgst_enable; 257 258 /* This is a spare PDU used for sending special management 259 * operations. Primarily, this is used for the initial 260 * connection response and c2h termination request. */ 261 struct nvme_tcp_pdu *mgmt_pdu; 262 263 /* Arrays of in-capsule buffers, requests, and pdus. 264 * Each array is 'resource_count' number of elements */ 265 void *bufs; 266 struct spdk_nvmf_tcp_req *reqs; 267 struct nvme_tcp_pdu *pdus; 268 uint32_t resource_count; 269 uint32_t recv_buf_size; 270 271 struct spdk_nvmf_tcp_port *port; 272 273 /* IP address */ 274 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 275 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 276 277 /* IP port */ 278 uint16_t initiator_port; 279 uint16_t target_port; 280 281 /* Timer used to destroy qpair after detecting transport error issue if initiator does 282 * not close the connection. 283 */ 284 struct spdk_poller *timeout_poller; 285 286 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 287 void *fini_cb_arg; 288 289 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 290 }; 291 292 struct spdk_nvmf_tcp_control_msg { 293 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 294 }; 295 296 struct spdk_nvmf_tcp_control_msg_list { 297 void *msg_buf; 298 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 299 }; 300 301 struct spdk_nvmf_tcp_poll_group { 302 struct spdk_nvmf_transport_poll_group group; 303 struct spdk_sock_group *sock_group; 304 305 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 306 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 307 308 struct spdk_io_channel *accel_channel; 309 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 310 311 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 312 }; 313 314 struct spdk_nvmf_tcp_port { 315 const struct spdk_nvme_transport_id *trid; 316 struct spdk_sock *listen_sock; 317 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 318 }; 319 320 struct tcp_transport_opts { 321 bool c2h_success; 322 uint16_t control_msg_num; 323 uint32_t sock_priority; 324 }; 325 326 struct spdk_nvmf_tcp_transport { 327 struct spdk_nvmf_transport transport; 328 struct tcp_transport_opts tcp_opts; 329 330 struct spdk_nvmf_tcp_poll_group *next_pg; 331 332 struct spdk_poller *accept_poller; 333 pthread_mutex_t lock; 334 335 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 336 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 337 }; 338 339 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 340 { 341 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 342 spdk_json_decode_bool, true 343 }, 344 { 345 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 346 spdk_json_decode_uint16, true 347 }, 348 { 349 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 350 spdk_json_decode_uint32, true 351 }, 352 }; 353 354 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 355 struct spdk_nvmf_tcp_req *tcp_req); 356 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 357 358 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 359 struct spdk_nvmf_tcp_req *tcp_req); 360 361 static void 362 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 363 enum spdk_nvmf_tcp_req_state state) 364 { 365 struct spdk_nvmf_qpair *qpair; 366 struct spdk_nvmf_tcp_qpair *tqpair; 367 368 qpair = tcp_req->req.qpair; 369 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 370 371 assert(tqpair->state_cntr[tcp_req->state] > 0); 372 tqpair->state_cntr[tcp_req->state]--; 373 tqpair->state_cntr[state]++; 374 375 tcp_req->state = state; 376 } 377 378 static inline struct nvme_tcp_pdu * 379 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 380 { 381 assert(tcp_req->pdu_in_use == false); 382 383 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 384 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 385 386 return tcp_req->pdu; 387 } 388 389 static struct spdk_nvmf_tcp_req * 390 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 391 { 392 struct spdk_nvmf_tcp_req *tcp_req; 393 394 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 395 if (!tcp_req) { 396 return NULL; 397 } 398 399 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 400 tcp_req->h2c_offset = 0; 401 tcp_req->has_in_capsule_data = false; 402 tcp_req->req.dif_enabled = false; 403 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 404 405 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 406 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 407 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 408 return tcp_req; 409 } 410 411 static inline void 412 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 413 { 414 assert(!tcp_req->pdu_in_use); 415 416 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 417 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 418 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 419 } 420 421 static void 422 nvmf_tcp_request_free(void *cb_arg) 423 { 424 struct spdk_nvmf_tcp_transport *ttransport; 425 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 426 427 assert(tcp_req != NULL); 428 429 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 430 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 431 struct spdk_nvmf_tcp_transport, transport); 432 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 433 nvmf_tcp_req_process(ttransport, tcp_req); 434 } 435 436 static int 437 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 438 { 439 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 440 441 nvmf_tcp_request_free(tcp_req); 442 443 return 0; 444 } 445 446 static void 447 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 448 enum spdk_nvmf_tcp_req_state state) 449 { 450 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 451 452 assert(state != TCP_REQUEST_STATE_FREE); 453 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 454 if (state == tcp_req->state) { 455 nvmf_tcp_request_free(tcp_req); 456 } 457 } 458 } 459 460 static void 461 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 462 { 463 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 464 465 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 466 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 467 468 /* Wipe the requests waiting for buffer from the global list */ 469 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 470 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 471 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 472 spdk_nvmf_request, buf_link); 473 } 474 } 475 476 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 477 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 478 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 479 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 480 } 481 482 static void 483 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 484 { 485 int i; 486 struct spdk_nvmf_tcp_req *tcp_req; 487 488 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 489 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 490 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 491 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 492 if ((int)tcp_req->state == i) { 493 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 494 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 495 } 496 } 497 } 498 } 499 500 static void 501 _nvmf_tcp_qpair_destroy(void *_tqpair) 502 { 503 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 504 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 505 void *cb_arg = tqpair->fini_cb_arg; 506 int err = 0; 507 508 spdk_trace_record(TRACE_TCP_QP_DESTROY, 0, 0, (uintptr_t)tqpair); 509 510 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 511 512 err = spdk_sock_close(&tqpair->sock); 513 assert(err == 0); 514 nvmf_tcp_cleanup_all_states(tqpair); 515 516 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 517 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 518 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 519 tqpair->resource_count); 520 err++; 521 } 522 523 if (err > 0) { 524 nvmf_tcp_dump_qpair_req_contents(tqpair); 525 } 526 527 /* The timeout poller might still be registered here if we close the qpair before host 528 * terminates the connection. 529 */ 530 spdk_poller_unregister(&tqpair->timeout_poller); 531 spdk_dma_free(tqpair->pdus); 532 free(tqpair->reqs); 533 spdk_free(tqpair->bufs); 534 free(tqpair); 535 536 if (cb_fn != NULL) { 537 cb_fn(cb_arg); 538 } 539 540 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 541 } 542 543 static void 544 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 545 { 546 /* Delay the destruction to make sure it isn't performed from the context of a sock 547 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 548 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 549 */ 550 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 551 } 552 553 static void 554 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 555 { 556 struct spdk_nvmf_tcp_transport *ttransport; 557 assert(w != NULL); 558 559 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 560 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 561 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 562 } 563 564 static int 565 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 566 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 567 { 568 struct spdk_nvmf_tcp_transport *ttransport; 569 570 assert(transport != NULL); 571 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 572 573 spdk_poller_unregister(&ttransport->accept_poller); 574 pthread_mutex_destroy(&ttransport->lock); 575 free(ttransport); 576 577 if (cb_fn) { 578 cb_fn(cb_arg); 579 } 580 return 0; 581 } 582 583 static int nvmf_tcp_accept(void *ctx); 584 585 static struct spdk_nvmf_transport * 586 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 587 { 588 struct spdk_nvmf_tcp_transport *ttransport; 589 uint32_t sge_count; 590 uint32_t min_shared_buffers; 591 592 ttransport = calloc(1, sizeof(*ttransport)); 593 if (!ttransport) { 594 return NULL; 595 } 596 597 TAILQ_INIT(&ttransport->ports); 598 TAILQ_INIT(&ttransport->poll_groups); 599 600 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 601 602 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 603 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 604 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 605 if (opts->transport_specific != NULL && 606 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 607 SPDK_COUNTOF(tcp_transport_opts_decoder), 608 &ttransport->tcp_opts)) { 609 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 610 free(ttransport); 611 return NULL; 612 } 613 614 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 615 616 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 617 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 618 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 619 " in_capsule_data_size=%d, max_aq_depth=%d\n" 620 " num_shared_buffers=%d, c2h_success=%d,\n" 621 " dif_insert_or_strip=%d, sock_priority=%d\n" 622 " abort_timeout_sec=%d, control_msg_num=%hu\n", 623 opts->max_queue_depth, 624 opts->max_io_size, 625 opts->max_qpairs_per_ctrlr - 1, 626 opts->io_unit_size, 627 opts->in_capsule_data_size, 628 opts->max_aq_depth, 629 opts->num_shared_buffers, 630 ttransport->tcp_opts.c2h_success, 631 opts->dif_insert_or_strip, 632 ttransport->tcp_opts.sock_priority, 633 opts->abort_timeout_sec, 634 ttransport->tcp_opts.control_msg_num); 635 636 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 637 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 638 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 639 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 640 free(ttransport); 641 return NULL; 642 } 643 644 if (ttransport->tcp_opts.control_msg_num == 0 && 645 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 646 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 647 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 648 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 649 } 650 651 /* I/O unit size cannot be larger than max I/O size */ 652 if (opts->io_unit_size > opts->max_io_size) { 653 opts->io_unit_size = opts->max_io_size; 654 } 655 656 sge_count = opts->max_io_size / opts->io_unit_size; 657 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 658 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 659 free(ttransport); 660 return NULL; 661 } 662 663 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 664 if (min_shared_buffers > opts->num_shared_buffers) { 665 SPDK_ERRLOG("There are not enough buffers to satisfy " 666 "per-poll group caches for each thread. (%" PRIu32 ") " 667 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 668 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 669 free(ttransport); 670 return NULL; 671 } 672 673 pthread_mutex_init(&ttransport->lock, NULL); 674 675 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 676 opts->acceptor_poll_rate); 677 if (!ttransport->accept_poller) { 678 pthread_mutex_destroy(&ttransport->lock); 679 free(ttransport); 680 return NULL; 681 } 682 683 return &ttransport->transport; 684 } 685 686 static int 687 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 688 { 689 unsigned long long ull; 690 char *end = NULL; 691 692 ull = strtoull(trsvcid, &end, 10); 693 if (end == NULL || end == trsvcid || *end != '\0') { 694 return -1; 695 } 696 697 /* Valid TCP/IP port numbers are in [0, 65535] */ 698 if (ull > 65535) { 699 return -1; 700 } 701 702 return (int)ull; 703 } 704 705 /** 706 * Canonicalize a listen address trid. 707 */ 708 static int 709 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 710 const struct spdk_nvme_transport_id *trid) 711 { 712 int trsvcid_int; 713 714 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 715 if (trsvcid_int < 0) { 716 return -EINVAL; 717 } 718 719 memset(canon_trid, 0, sizeof(*canon_trid)); 720 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 721 canon_trid->adrfam = trid->adrfam; 722 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 723 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 724 725 return 0; 726 } 727 728 /** 729 * Find an existing listening port. 730 * 731 * Caller must hold ttransport->lock. 732 */ 733 static struct spdk_nvmf_tcp_port * 734 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 735 const struct spdk_nvme_transport_id *trid) 736 { 737 struct spdk_nvme_transport_id canon_trid; 738 struct spdk_nvmf_tcp_port *port; 739 740 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 741 return NULL; 742 } 743 744 TAILQ_FOREACH(port, &ttransport->ports, link) { 745 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 746 return port; 747 } 748 } 749 750 return NULL; 751 } 752 753 static int 754 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 755 struct spdk_nvmf_listen_opts *listen_opts) 756 { 757 struct spdk_nvmf_tcp_transport *ttransport; 758 struct spdk_nvmf_tcp_port *port; 759 int trsvcid_int; 760 uint8_t adrfam; 761 struct spdk_sock_opts opts; 762 763 if (!strlen(trid->trsvcid)) { 764 SPDK_ERRLOG("Service id is required\n"); 765 return -EINVAL; 766 } 767 768 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 769 770 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 771 if (trsvcid_int < 0) { 772 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 773 return -EINVAL; 774 } 775 776 pthread_mutex_lock(&ttransport->lock); 777 port = calloc(1, sizeof(*port)); 778 if (!port) { 779 SPDK_ERRLOG("Port allocation failed\n"); 780 pthread_mutex_unlock(&ttransport->lock); 781 return -ENOMEM; 782 } 783 784 port->trid = trid; 785 opts.opts_size = sizeof(opts); 786 spdk_sock_get_default_opts(&opts); 787 opts.priority = ttransport->tcp_opts.sock_priority; 788 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 789 NULL, &opts); 790 if (port->listen_sock == NULL) { 791 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 792 trid->traddr, trsvcid_int, 793 spdk_strerror(errno), errno); 794 free(port); 795 pthread_mutex_unlock(&ttransport->lock); 796 return -errno; 797 } 798 799 if (spdk_sock_is_ipv4(port->listen_sock)) { 800 adrfam = SPDK_NVMF_ADRFAM_IPV4; 801 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 802 adrfam = SPDK_NVMF_ADRFAM_IPV6; 803 } else { 804 SPDK_ERRLOG("Unhandled socket type\n"); 805 adrfam = 0; 806 } 807 808 if (adrfam != trid->adrfam) { 809 SPDK_ERRLOG("Socket address family mismatch\n"); 810 spdk_sock_close(&port->listen_sock); 811 free(port); 812 pthread_mutex_unlock(&ttransport->lock); 813 return -EINVAL; 814 } 815 816 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 817 trid->traddr, trid->trsvcid); 818 819 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 820 pthread_mutex_unlock(&ttransport->lock); 821 return 0; 822 } 823 824 static void 825 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 826 const struct spdk_nvme_transport_id *trid) 827 { 828 struct spdk_nvmf_tcp_transport *ttransport; 829 struct spdk_nvmf_tcp_port *port; 830 831 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 832 833 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 834 trid->traddr, trid->trsvcid); 835 836 pthread_mutex_lock(&ttransport->lock); 837 port = nvmf_tcp_find_port(ttransport, trid); 838 if (port) { 839 TAILQ_REMOVE(&ttransport->ports, port, link); 840 spdk_sock_close(&port->listen_sock); 841 free(port); 842 } 843 844 pthread_mutex_unlock(&ttransport->lock); 845 } 846 847 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 848 enum nvme_tcp_pdu_recv_state state); 849 850 static void 851 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 852 { 853 tqpair->state = state; 854 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 855 tqpair->state); 856 } 857 858 static void 859 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 860 { 861 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 862 863 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, 0, 0, (uintptr_t)tqpair); 864 865 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 866 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 867 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 868 spdk_poller_unregister(&tqpair->timeout_poller); 869 870 /* This will end up calling nvmf_tcp_close_qpair */ 871 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 872 } 873 } 874 875 static void 876 _mgmt_pdu_write_done(void *_tqpair, int err) 877 { 878 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 879 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 880 881 if (spdk_unlikely(err != 0)) { 882 nvmf_tcp_qpair_disconnect(tqpair); 883 return; 884 } 885 886 assert(pdu->cb_fn != NULL); 887 pdu->cb_fn(pdu->cb_arg); 888 } 889 890 static void 891 _req_pdu_write_done(void *req, int err) 892 { 893 struct spdk_nvmf_tcp_req *tcp_req = req; 894 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 895 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 896 897 assert(tcp_req->pdu_in_use); 898 tcp_req->pdu_in_use = false; 899 900 /* If the request is in a completed state, we're waiting for write completion to free it */ 901 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 902 nvmf_tcp_request_free(tcp_req); 903 return; 904 } 905 906 if (spdk_unlikely(err != 0)) { 907 nvmf_tcp_qpair_disconnect(tqpair); 908 return; 909 } 910 911 assert(pdu->cb_fn != NULL); 912 pdu->cb_fn(pdu->cb_arg); 913 } 914 915 static void 916 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 917 { 918 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 919 } 920 921 static void 922 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 923 { 924 uint32_t mapped_length = 0; 925 ssize_t rc; 926 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 927 928 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 929 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 930 &mapped_length); 931 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 932 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 933 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 934 if (rc == mapped_length) { 935 _pdu_write_done(pdu, 0); 936 } else { 937 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 938 _pdu_write_done(pdu, -1); 939 } 940 } else { 941 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 942 } 943 } 944 945 static void 946 data_crc32_accel_done(void *cb_arg, int status) 947 { 948 struct nvme_tcp_pdu *pdu = cb_arg; 949 950 if (spdk_unlikely(status)) { 951 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 952 _pdu_write_done(pdu, status); 953 return; 954 } 955 956 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 957 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 958 959 _tcp_write_pdu(pdu); 960 } 961 962 static void 963 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 964 { 965 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 966 uint32_t crc32c; 967 968 /* Data Digest */ 969 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 970 /* Only suport this limitated case for the first step */ 971 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 972 && tqpair->group)) { 973 if (!spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 974 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu)) { 975 return; 976 } 977 } 978 979 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 980 crc32c = crc32c ^ SPDK_CRC32C_XOR; 981 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 982 } 983 984 _tcp_write_pdu(pdu); 985 } 986 987 static void 988 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 989 struct nvme_tcp_pdu *pdu, 990 nvme_tcp_qpair_xfer_complete_cb cb_fn, 991 void *cb_arg) 992 { 993 int hlen; 994 uint32_t crc32c; 995 996 assert(tqpair->pdu_in_progress != pdu); 997 998 hlen = pdu->hdr.common.hlen; 999 pdu->cb_fn = cb_fn; 1000 pdu->cb_arg = cb_arg; 1001 1002 pdu->iov[0].iov_base = &pdu->hdr.raw; 1003 pdu->iov[0].iov_len = hlen; 1004 1005 /* Header Digest */ 1006 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1007 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1008 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1009 } 1010 1011 /* Data Digest */ 1012 pdu_data_crc32_compute(pdu); 1013 } 1014 1015 static void 1016 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1017 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1018 void *cb_arg) 1019 { 1020 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1021 1022 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1023 pdu->sock_req.cb_arg = tqpair; 1024 1025 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1026 } 1027 1028 static void 1029 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1030 struct spdk_nvmf_tcp_req *tcp_req, 1031 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1032 void *cb_arg) 1033 { 1034 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1035 1036 pdu->sock_req.cb_fn = _req_pdu_write_done; 1037 pdu->sock_req.cb_arg = tcp_req; 1038 1039 assert(!tcp_req->pdu_in_use); 1040 tcp_req->pdu_in_use = true; 1041 1042 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1043 } 1044 1045 static int 1046 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1047 { 1048 uint32_t i; 1049 struct spdk_nvmf_transport_opts *opts; 1050 uint32_t in_capsule_data_size; 1051 1052 opts = &tqpair->qpair.transport->opts; 1053 1054 in_capsule_data_size = opts->in_capsule_data_size; 1055 if (opts->dif_insert_or_strip) { 1056 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1057 } 1058 1059 tqpair->resource_count = opts->max_queue_depth; 1060 1061 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1062 if (!tqpair->reqs) { 1063 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1064 return -1; 1065 } 1066 1067 if (in_capsule_data_size) { 1068 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1069 NULL, SPDK_ENV_LCORE_ID_ANY, 1070 SPDK_MALLOC_DMA); 1071 if (!tqpair->bufs) { 1072 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1073 return -1; 1074 } 1075 } 1076 1077 /* Add additional 2 members, which will be used for mgmt_pdu and pdu_in_progress owned by the tqpair */ 1078 tqpair->pdus = spdk_dma_zmalloc((tqpair->resource_count + 2) * sizeof(*tqpair->pdus), 0x1000, NULL); 1079 if (!tqpair->pdus) { 1080 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1081 return -1; 1082 } 1083 1084 for (i = 0; i < tqpair->resource_count; i++) { 1085 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1086 1087 tcp_req->ttag = i + 1; 1088 tcp_req->req.qpair = &tqpair->qpair; 1089 1090 tcp_req->pdu = &tqpair->pdus[i]; 1091 tcp_req->pdu->qpair = tqpair; 1092 1093 /* Set up memory to receive commands */ 1094 if (tqpair->bufs) { 1095 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1096 } 1097 1098 /* Set the cmdn and rsp */ 1099 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1100 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1101 1102 tcp_req->req.stripped_data = NULL; 1103 1104 /* Initialize request state to FREE */ 1105 tcp_req->state = TCP_REQUEST_STATE_FREE; 1106 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1107 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1108 } 1109 1110 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1111 tqpair->mgmt_pdu->qpair = tqpair; 1112 tqpair->pdu_in_progress = &tqpair->pdus[i + 1]; 1113 1114 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1115 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1116 1117 return 0; 1118 } 1119 1120 static int 1121 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1122 { 1123 struct spdk_nvmf_tcp_qpair *tqpair; 1124 1125 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1126 1127 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1128 1129 spdk_trace_record(TRACE_TCP_QP_CREATE, 0, 0, (uintptr_t)tqpair); 1130 1131 /* Initialise request state queues of the qpair */ 1132 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1133 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1134 1135 tqpair->host_hdgst_enable = true; 1136 tqpair->host_ddgst_enable = true; 1137 1138 return 0; 1139 } 1140 1141 static int 1142 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1143 { 1144 int rc; 1145 1146 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, 0, 0, (uintptr_t)tqpair); 1147 1148 /* set low water mark */ 1149 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1150 if (rc != 0) { 1151 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1152 return rc; 1153 } 1154 1155 return 0; 1156 } 1157 1158 static void 1159 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1160 struct spdk_nvmf_tcp_port *port, 1161 struct spdk_sock *sock) 1162 { 1163 struct spdk_nvmf_tcp_qpair *tqpair; 1164 int rc; 1165 1166 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1167 port->trid->traddr, port->trid->trsvcid); 1168 1169 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1170 if (tqpair == NULL) { 1171 SPDK_ERRLOG("Could not allocate new connection.\n"); 1172 spdk_sock_close(&sock); 1173 return; 1174 } 1175 1176 tqpair->sock = sock; 1177 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1178 tqpair->port = port; 1179 tqpair->qpair.transport = transport; 1180 1181 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1182 sizeof(tqpair->target_addr), &tqpair->target_port, 1183 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1184 &tqpair->initiator_port); 1185 if (rc < 0) { 1186 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1187 nvmf_tcp_qpair_destroy(tqpair); 1188 return; 1189 } 1190 1191 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1192 } 1193 1194 static uint32_t 1195 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1196 { 1197 struct spdk_sock *sock; 1198 uint32_t count = 0; 1199 int i; 1200 1201 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1202 sock = spdk_sock_accept(port->listen_sock); 1203 if (sock == NULL) { 1204 break; 1205 } 1206 count++; 1207 nvmf_tcp_handle_connect(transport, port, sock); 1208 } 1209 1210 return count; 1211 } 1212 1213 static int 1214 nvmf_tcp_accept(void *ctx) 1215 { 1216 struct spdk_nvmf_transport *transport = ctx; 1217 struct spdk_nvmf_tcp_transport *ttransport; 1218 struct spdk_nvmf_tcp_port *port; 1219 uint32_t count = 0; 1220 1221 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1222 1223 TAILQ_FOREACH(port, &ttransport->ports, link) { 1224 count += nvmf_tcp_port_accept(transport, port); 1225 } 1226 1227 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1228 } 1229 1230 static void 1231 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1232 struct spdk_nvme_transport_id *trid, 1233 struct spdk_nvmf_discovery_log_page_entry *entry) 1234 { 1235 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1236 entry->adrfam = trid->adrfam; 1237 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1238 1239 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1240 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1241 1242 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1243 } 1244 1245 static struct spdk_nvmf_tcp_control_msg_list * 1246 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1247 { 1248 struct spdk_nvmf_tcp_control_msg_list *list; 1249 struct spdk_nvmf_tcp_control_msg *msg; 1250 uint16_t i; 1251 1252 list = calloc(1, sizeof(*list)); 1253 if (!list) { 1254 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1255 return NULL; 1256 } 1257 1258 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1259 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1260 if (!list->msg_buf) { 1261 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1262 free(list); 1263 return NULL; 1264 } 1265 1266 STAILQ_INIT(&list->free_msgs); 1267 1268 for (i = 0; i < num_messages; i++) { 1269 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1270 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1271 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1272 } 1273 1274 return list; 1275 } 1276 1277 static void 1278 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1279 { 1280 if (!list) { 1281 return; 1282 } 1283 1284 spdk_free(list->msg_buf); 1285 free(list); 1286 } 1287 1288 static struct spdk_nvmf_transport_poll_group * 1289 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1290 struct spdk_nvmf_poll_group *group) 1291 { 1292 struct spdk_nvmf_tcp_transport *ttransport; 1293 struct spdk_nvmf_tcp_poll_group *tgroup; 1294 1295 tgroup = calloc(1, sizeof(*tgroup)); 1296 if (!tgroup) { 1297 return NULL; 1298 } 1299 1300 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1301 if (!tgroup->sock_group) { 1302 goto cleanup; 1303 } 1304 1305 TAILQ_INIT(&tgroup->qpairs); 1306 TAILQ_INIT(&tgroup->await_req); 1307 1308 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1309 1310 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1311 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1312 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1313 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1314 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1315 if (!tgroup->control_msg_list) { 1316 goto cleanup; 1317 } 1318 } 1319 1320 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1321 if (spdk_unlikely(!tgroup->accel_channel)) { 1322 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1323 goto cleanup; 1324 } 1325 1326 pthread_mutex_lock(&ttransport->lock); 1327 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1328 if (ttransport->next_pg == NULL) { 1329 ttransport->next_pg = tgroup; 1330 } 1331 pthread_mutex_unlock(&ttransport->lock); 1332 1333 return &tgroup->group; 1334 1335 cleanup: 1336 nvmf_tcp_poll_group_destroy(&tgroup->group); 1337 return NULL; 1338 } 1339 1340 static struct spdk_nvmf_transport_poll_group * 1341 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1342 { 1343 struct spdk_nvmf_tcp_transport *ttransport; 1344 struct spdk_nvmf_tcp_poll_group **pg; 1345 struct spdk_nvmf_tcp_qpair *tqpair; 1346 struct spdk_sock_group *group = NULL, *hint = NULL; 1347 int rc; 1348 1349 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1350 1351 pthread_mutex_lock(&ttransport->lock); 1352 1353 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1354 pthread_mutex_unlock(&ttransport->lock); 1355 return NULL; 1356 } 1357 1358 pg = &ttransport->next_pg; 1359 assert(*pg != NULL); 1360 hint = (*pg)->sock_group; 1361 1362 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1363 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1364 if (rc != 0) { 1365 pthread_mutex_unlock(&ttransport->lock); 1366 return NULL; 1367 } else if (group != NULL) { 1368 /* Optimal poll group was found */ 1369 pthread_mutex_unlock(&ttransport->lock); 1370 return spdk_sock_group_get_ctx(group); 1371 } 1372 1373 /* The hint was used for optimal poll group, advance next_pg. */ 1374 *pg = TAILQ_NEXT(*pg, link); 1375 if (*pg == NULL) { 1376 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1377 } 1378 1379 pthread_mutex_unlock(&ttransport->lock); 1380 return spdk_sock_group_get_ctx(hint); 1381 } 1382 1383 static void 1384 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1385 { 1386 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1387 struct spdk_nvmf_tcp_transport *ttransport; 1388 1389 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1390 spdk_sock_group_close(&tgroup->sock_group); 1391 if (tgroup->control_msg_list) { 1392 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1393 } 1394 1395 if (tgroup->accel_channel) { 1396 spdk_put_io_channel(tgroup->accel_channel); 1397 } 1398 1399 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1400 1401 pthread_mutex_lock(&ttransport->lock); 1402 next_tgroup = TAILQ_NEXT(tgroup, link); 1403 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1404 if (next_tgroup == NULL) { 1405 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1406 } 1407 if (ttransport->next_pg == tgroup) { 1408 ttransport->next_pg = next_tgroup; 1409 } 1410 pthread_mutex_unlock(&ttransport->lock); 1411 1412 free(tgroup); 1413 } 1414 1415 static void 1416 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1417 enum nvme_tcp_pdu_recv_state state) 1418 { 1419 if (tqpair->recv_state == state) { 1420 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1421 tqpair, state); 1422 return; 1423 } 1424 1425 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1426 /* When leaving the await req state, move the qpair to the main list */ 1427 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1428 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1429 } 1430 1431 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1432 tqpair->recv_state = state; 1433 1434 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 1435 tqpair->recv_state); 1436 1437 switch (state) { 1438 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1439 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1440 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1441 break; 1442 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1443 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1444 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1445 break; 1446 case NVME_TCP_PDU_RECV_STATE_ERROR: 1447 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1448 memset(tqpair->pdu_in_progress, 0, sizeof(*(tqpair->pdu_in_progress))); 1449 break; 1450 default: 1451 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1452 abort(); 1453 break; 1454 } 1455 } 1456 1457 static int 1458 nvmf_tcp_qpair_handle_timeout(void *ctx) 1459 { 1460 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1461 1462 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1463 1464 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1465 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1466 1467 nvmf_tcp_qpair_disconnect(tqpair); 1468 return SPDK_POLLER_BUSY; 1469 } 1470 1471 static void 1472 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1473 { 1474 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1475 1476 if (!tqpair->timeout_poller) { 1477 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1478 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1479 } 1480 } 1481 1482 static void 1483 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1484 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1485 { 1486 struct nvme_tcp_pdu *rsp_pdu; 1487 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1488 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1489 uint32_t copy_len; 1490 1491 rsp_pdu = tqpair->mgmt_pdu; 1492 1493 c2h_term_req = &rsp_pdu->hdr.term_req; 1494 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1495 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1496 1497 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1498 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1499 DSET32(&c2h_term_req->fei, error_offset); 1500 } 1501 1502 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1503 1504 /* Copy the error info into the buffer */ 1505 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1506 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1507 1508 /* Contain the header of the wrong received pdu */ 1509 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1510 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1511 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1512 } 1513 1514 static void 1515 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1516 struct spdk_nvmf_tcp_qpair *tqpair, 1517 struct nvme_tcp_pdu *pdu) 1518 { 1519 struct spdk_nvmf_tcp_req *tcp_req; 1520 1521 assert(pdu->psh_valid_bytes == pdu->psh_len); 1522 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1523 1524 tcp_req = nvmf_tcp_req_get(tqpair); 1525 if (!tcp_req) { 1526 /* Directly return and make the allocation retry again. This can happen if we're 1527 * using asynchronous writes to send the response to the host or when releasing 1528 * zero-copy buffers after a response has been sent. In both cases, the host might 1529 * receive the response before we've finished processing the request and is free to 1530 * send another one. 1531 */ 1532 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1533 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1534 return; 1535 } 1536 1537 /* The host sent more commands than the maximum queue depth. */ 1538 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1539 nvmf_tcp_qpair_disconnect(tqpair); 1540 return; 1541 } 1542 1543 pdu->req = tcp_req; 1544 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1545 nvmf_tcp_req_process(ttransport, tcp_req); 1546 } 1547 1548 static void 1549 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1550 struct spdk_nvmf_tcp_qpair *tqpair, 1551 struct nvme_tcp_pdu *pdu) 1552 { 1553 struct spdk_nvmf_tcp_req *tcp_req; 1554 struct spdk_nvme_tcp_cmd *capsule_cmd; 1555 uint32_t error_offset = 0; 1556 enum spdk_nvme_tcp_term_req_fes fes; 1557 struct spdk_nvme_cpl *rsp; 1558 1559 capsule_cmd = &pdu->hdr.capsule_cmd; 1560 tcp_req = pdu->req; 1561 assert(tcp_req != NULL); 1562 1563 /* Zero-copy requests don't support ICD */ 1564 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1565 1566 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1567 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1568 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1569 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1570 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1571 goto err; 1572 } 1573 1574 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1575 1576 rsp = &tcp_req->req.rsp->nvme_cpl; 1577 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1578 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1579 } else { 1580 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1581 } 1582 1583 nvmf_tcp_req_process(ttransport, tcp_req); 1584 1585 return; 1586 err: 1587 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1588 } 1589 1590 static int 1591 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1592 enum spdk_nvmf_tcp_req_state state, 1593 uint16_t cid, uint16_t tag, 1594 struct spdk_nvmf_tcp_req **req) 1595 { 1596 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1597 1598 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1599 if (tcp_req->state != state) { 1600 continue; 1601 } 1602 1603 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1604 continue; 1605 } 1606 1607 if (tcp_req->ttag == tag) { 1608 *req = tcp_req; 1609 return 0; 1610 } 1611 1612 *req = NULL; 1613 return -1; 1614 } 1615 1616 /* Didn't find it, but not an error */ 1617 *req = NULL; 1618 return 0; 1619 } 1620 1621 static void 1622 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1623 struct spdk_nvmf_tcp_qpair *tqpair, 1624 struct nvme_tcp_pdu *pdu) 1625 { 1626 struct spdk_nvmf_tcp_req *tcp_req; 1627 uint32_t error_offset = 0; 1628 enum spdk_nvme_tcp_term_req_fes fes = 0; 1629 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1630 int rc; 1631 1632 h2c_data = &pdu->hdr.h2c_data; 1633 1634 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1635 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1636 1637 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1638 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1639 if (rc == 0 && tcp_req == NULL) { 1640 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1641 h2c_data->ttag, &tcp_req); 1642 } 1643 1644 if (!tcp_req) { 1645 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1646 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1647 if (rc == 0) { 1648 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1649 } else { 1650 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1651 } 1652 goto err; 1653 } 1654 1655 if (tcp_req->h2c_offset != h2c_data->datao) { 1656 SPDK_DEBUGLOG(nvmf_tcp, 1657 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1658 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1659 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1660 goto err; 1661 } 1662 1663 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1664 SPDK_DEBUGLOG(nvmf_tcp, 1665 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1666 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1667 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1668 goto err; 1669 } 1670 1671 pdu->req = tcp_req; 1672 1673 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1674 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1675 } 1676 1677 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1678 h2c_data->datao, h2c_data->datal); 1679 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1680 return; 1681 1682 err: 1683 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1684 } 1685 1686 static void 1687 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1688 struct spdk_nvmf_tcp_qpair *tqpair) 1689 { 1690 struct nvme_tcp_pdu *rsp_pdu; 1691 struct spdk_nvme_tcp_rsp *capsule_resp; 1692 1693 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1694 1695 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1696 assert(rsp_pdu != NULL); 1697 1698 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1699 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1700 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1701 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1702 if (tqpair->host_hdgst_enable) { 1703 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1704 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1705 } 1706 1707 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1708 } 1709 1710 static void 1711 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1712 { 1713 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1714 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1715 struct spdk_nvmf_tcp_qpair, qpair); 1716 1717 assert(tqpair != NULL); 1718 1719 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1720 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1721 tcp_req->req.length); 1722 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1723 return; 1724 } 1725 1726 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1727 nvmf_tcp_request_free(tcp_req); 1728 } else { 1729 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1730 } 1731 } 1732 1733 static void 1734 nvmf_tcp_r2t_complete(void *cb_arg) 1735 { 1736 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1737 struct spdk_nvmf_tcp_transport *ttransport; 1738 1739 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1740 struct spdk_nvmf_tcp_transport, transport); 1741 1742 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1743 1744 if (tcp_req->h2c_offset == tcp_req->req.length) { 1745 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1746 nvmf_tcp_req_process(ttransport, tcp_req); 1747 } 1748 } 1749 1750 static void 1751 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1752 struct spdk_nvmf_tcp_req *tcp_req) 1753 { 1754 struct nvme_tcp_pdu *rsp_pdu; 1755 struct spdk_nvme_tcp_r2t_hdr *r2t; 1756 1757 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1758 assert(rsp_pdu != NULL); 1759 1760 r2t = &rsp_pdu->hdr.r2t; 1761 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1762 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1763 1764 if (tqpair->host_hdgst_enable) { 1765 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1766 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1767 } 1768 1769 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1770 r2t->ttag = tcp_req->ttag; 1771 r2t->r2to = tcp_req->h2c_offset; 1772 r2t->r2tl = tcp_req->req.length; 1773 1774 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1775 1776 SPDK_DEBUGLOG(nvmf_tcp, 1777 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1778 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1779 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1780 } 1781 1782 static void 1783 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1784 struct spdk_nvmf_tcp_qpair *tqpair, 1785 struct nvme_tcp_pdu *pdu) 1786 { 1787 struct spdk_nvmf_tcp_req *tcp_req; 1788 struct spdk_nvme_cpl *rsp; 1789 1790 tcp_req = pdu->req; 1791 assert(tcp_req != NULL); 1792 1793 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1794 1795 tcp_req->h2c_offset += pdu->data_len; 1796 1797 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1798 1799 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1800 * acknowledged before moving on. */ 1801 if (tcp_req->h2c_offset == tcp_req->req.length && 1802 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1803 /* After receiving all the h2c data, we need to check whether there is 1804 * transient transport error */ 1805 rsp = &tcp_req->req.rsp->nvme_cpl; 1806 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1807 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1808 } else { 1809 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1810 } 1811 nvmf_tcp_req_process(ttransport, tcp_req); 1812 } 1813 } 1814 1815 static void 1816 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1817 { 1818 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1819 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1820 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1821 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1822 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1823 DGET32(h2c_term_req->fei)); 1824 } 1825 } 1826 1827 static void 1828 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1829 struct nvme_tcp_pdu *pdu) 1830 { 1831 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1832 uint32_t error_offset = 0; 1833 enum spdk_nvme_tcp_term_req_fes fes; 1834 1835 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1836 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1837 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1838 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1839 goto end; 1840 } 1841 1842 /* set the data buffer */ 1843 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1844 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1845 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1846 return; 1847 end: 1848 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1849 } 1850 1851 static void 1852 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1853 struct nvme_tcp_pdu *pdu) 1854 { 1855 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1856 1857 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1858 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1859 } 1860 1861 static void 1862 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1863 struct spdk_nvmf_tcp_transport *ttransport) 1864 { 1865 struct nvme_tcp_pdu *pdu = tqpair->pdu_in_progress; 1866 1867 switch (pdu->hdr.common.pdu_type) { 1868 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1869 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1870 break; 1871 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1872 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1873 break; 1874 1875 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1876 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1877 break; 1878 1879 default: 1880 /* The code should not go to here */ 1881 SPDK_ERRLOG("The code should not go to here\n"); 1882 break; 1883 } 1884 } 1885 1886 static void 1887 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1888 struct spdk_nvmf_tcp_transport *ttransport) 1889 { 1890 int rc = 0; 1891 struct nvme_tcp_pdu *pdu; 1892 uint32_t crc32c; 1893 struct spdk_nvmf_tcp_req *tcp_req; 1894 struct spdk_nvme_cpl *rsp; 1895 1896 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1897 pdu = tqpair->pdu_in_progress; 1898 1899 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1900 /* check data digest if need */ 1901 if (pdu->ddgst_enable) { 1902 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1903 crc32c = crc32c ^ SPDK_CRC32C_XOR; 1904 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1905 if (rc == 0) { 1906 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1907 tcp_req = pdu->req; 1908 assert(tcp_req != NULL); 1909 rsp = &tcp_req->req.rsp->nvme_cpl; 1910 rsp->status.sc = SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR; 1911 } 1912 } 1913 1914 _nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1915 } 1916 1917 static void 1918 nvmf_tcp_send_icresp_complete(void *cb_arg) 1919 { 1920 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1921 1922 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 1923 } 1924 1925 static void 1926 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1927 struct spdk_nvmf_tcp_qpair *tqpair, 1928 struct nvme_tcp_pdu *pdu) 1929 { 1930 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1931 struct nvme_tcp_pdu *rsp_pdu; 1932 struct spdk_nvme_tcp_ic_resp *ic_resp; 1933 uint32_t error_offset = 0; 1934 enum spdk_nvme_tcp_term_req_fes fes; 1935 1936 /* Only PFV 0 is defined currently */ 1937 if (ic_req->pfv != 0) { 1938 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1939 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1940 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1941 goto end; 1942 } 1943 1944 /* MAXR2T is 0's based */ 1945 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1946 1947 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1948 if (!tqpair->host_hdgst_enable) { 1949 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1950 } 1951 1952 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1953 if (!tqpair->host_ddgst_enable) { 1954 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1955 } 1956 1957 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1958 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1959 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1960 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1961 tqpair, 1962 tqpair->recv_buf_size); 1963 /* Not fatal. */ 1964 } 1965 1966 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1967 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1968 1969 rsp_pdu = tqpair->mgmt_pdu; 1970 1971 ic_resp = &rsp_pdu->hdr.ic_resp; 1972 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1973 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1974 ic_resp->pfv = 0; 1975 ic_resp->cpda = tqpair->cpda; 1976 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1977 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1978 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1979 1980 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1981 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1982 1983 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 1984 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 1985 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1986 return; 1987 end: 1988 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1989 } 1990 1991 static void 1992 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1993 struct spdk_nvmf_tcp_transport *ttransport) 1994 { 1995 struct nvme_tcp_pdu *pdu; 1996 int rc; 1997 uint32_t crc32c, error_offset = 0; 1998 enum spdk_nvme_tcp_term_req_fes fes; 1999 2000 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2001 pdu = tqpair->pdu_in_progress; 2002 2003 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2004 pdu->hdr.common.pdu_type); 2005 /* check header digest if needed */ 2006 if (pdu->has_hdgst) { 2007 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2008 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2009 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2010 if (rc == 0) { 2011 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2012 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2013 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2014 return; 2015 2016 } 2017 } 2018 2019 switch (pdu->hdr.common.pdu_type) { 2020 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2021 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2022 break; 2023 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2024 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2025 break; 2026 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2027 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2028 break; 2029 2030 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2031 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2032 break; 2033 2034 default: 2035 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2036 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2037 error_offset = 1; 2038 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2039 break; 2040 } 2041 } 2042 2043 static void 2044 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2045 { 2046 struct nvme_tcp_pdu *pdu; 2047 uint32_t error_offset = 0; 2048 enum spdk_nvme_tcp_term_req_fes fes; 2049 uint8_t expected_hlen, pdo; 2050 bool plen_error = false, pdo_error = false; 2051 2052 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2053 pdu = tqpair->pdu_in_progress; 2054 2055 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2056 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2057 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2058 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2059 goto err; 2060 } 2061 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2062 if (pdu->hdr.common.plen != expected_hlen) { 2063 plen_error = true; 2064 } 2065 } else { 2066 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2067 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2068 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2069 goto err; 2070 } 2071 2072 switch (pdu->hdr.common.pdu_type) { 2073 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2074 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2075 pdo = pdu->hdr.common.pdo; 2076 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2077 pdo_error = true; 2078 break; 2079 } 2080 2081 if (pdu->hdr.common.plen < expected_hlen) { 2082 plen_error = true; 2083 } 2084 break; 2085 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2086 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2087 pdo = pdu->hdr.common.pdo; 2088 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2089 pdo_error = true; 2090 break; 2091 } 2092 if (pdu->hdr.common.plen < expected_hlen) { 2093 plen_error = true; 2094 } 2095 break; 2096 2097 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2098 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2099 if ((pdu->hdr.common.plen <= expected_hlen) || 2100 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2101 plen_error = true; 2102 } 2103 break; 2104 2105 default: 2106 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2107 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2108 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2109 goto err; 2110 } 2111 } 2112 2113 if (pdu->hdr.common.hlen != expected_hlen) { 2114 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2115 pdu->hdr.common.pdu_type, 2116 expected_hlen, pdu->hdr.common.hlen, tqpair); 2117 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2118 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2119 goto err; 2120 } else if (pdo_error) { 2121 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2122 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2123 } else if (plen_error) { 2124 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2125 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2126 goto err; 2127 } else { 2128 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2129 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2130 return; 2131 } 2132 err: 2133 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2134 } 2135 2136 static int 2137 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 2138 int read_len) 2139 { 2140 int rc; 2141 2142 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 2143 read_offset, read_len, pdu->dif_ctx); 2144 if (rc != 0) { 2145 SPDK_ERRLOG("DIF generate failed\n"); 2146 } 2147 2148 return rc; 2149 } 2150 2151 static int 2152 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2153 { 2154 int rc = 0; 2155 struct nvme_tcp_pdu *pdu; 2156 enum nvme_tcp_pdu_recv_state prev_state; 2157 uint32_t data_len; 2158 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2159 struct spdk_nvmf_tcp_transport, transport); 2160 2161 /* The loop here is to allow for several back-to-back state changes. */ 2162 do { 2163 prev_state = tqpair->recv_state; 2164 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2165 2166 pdu = tqpair->pdu_in_progress; 2167 switch (tqpair->recv_state) { 2168 /* Wait for the common header */ 2169 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2170 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2171 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2172 return rc; 2173 } 2174 2175 rc = nvme_tcp_read_data(tqpair->sock, 2176 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2177 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2178 if (rc < 0) { 2179 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2180 return NVME_TCP_PDU_FATAL; 2181 } else if (rc > 0) { 2182 pdu->ch_valid_bytes += rc; 2183 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2184 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 2185 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2186 } 2187 } 2188 2189 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2190 return NVME_TCP_PDU_IN_PROGRESS; 2191 } 2192 2193 /* The command header of this PDU has now been read from the socket. */ 2194 nvmf_tcp_pdu_ch_handle(tqpair); 2195 break; 2196 /* Wait for the pdu specific header */ 2197 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2198 rc = nvme_tcp_read_data(tqpair->sock, 2199 pdu->psh_len - pdu->psh_valid_bytes, 2200 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2201 if (rc < 0) { 2202 return NVME_TCP_PDU_FATAL; 2203 } else if (rc > 0) { 2204 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2205 pdu->psh_valid_bytes += rc; 2206 } 2207 2208 if (pdu->psh_valid_bytes < pdu->psh_len) { 2209 return NVME_TCP_PDU_IN_PROGRESS; 2210 } 2211 2212 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2213 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2214 break; 2215 /* Wait for the req slot */ 2216 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2217 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2218 break; 2219 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2220 /* check whether the data is valid, if not we just return */ 2221 if (!pdu->data_len) { 2222 return NVME_TCP_PDU_IN_PROGRESS; 2223 } 2224 2225 data_len = pdu->data_len; 2226 /* data digest */ 2227 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2228 tqpair->host_ddgst_enable)) { 2229 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2230 pdu->ddgst_enable = true; 2231 } 2232 2233 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2234 if (rc < 0) { 2235 return NVME_TCP_PDU_FATAL; 2236 } 2237 pdu->rw_offset += rc; 2238 2239 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2240 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2241 if (rc != 0) { 2242 return NVME_TCP_PDU_FATAL; 2243 } 2244 } 2245 2246 if (pdu->rw_offset < data_len) { 2247 return NVME_TCP_PDU_IN_PROGRESS; 2248 } 2249 2250 /* All of this PDU has now been read from the socket. */ 2251 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2252 break; 2253 case NVME_TCP_PDU_RECV_STATE_ERROR: 2254 if (!spdk_sock_is_connected(tqpair->sock)) { 2255 return NVME_TCP_PDU_FATAL; 2256 } 2257 break; 2258 default: 2259 assert(0); 2260 SPDK_ERRLOG("code should not come to here"); 2261 break; 2262 } 2263 } while (tqpair->recv_state != prev_state); 2264 2265 return rc; 2266 } 2267 2268 static inline void * 2269 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2270 { 2271 struct spdk_nvmf_tcp_control_msg *msg; 2272 2273 assert(list); 2274 2275 msg = STAILQ_FIRST(&list->free_msgs); 2276 if (!msg) { 2277 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2278 return NULL; 2279 } 2280 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2281 return msg; 2282 } 2283 2284 static inline void 2285 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2286 { 2287 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2288 2289 assert(list); 2290 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2291 } 2292 2293 static int 2294 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2295 struct spdk_nvmf_transport *transport, 2296 struct spdk_nvmf_transport_poll_group *group) 2297 { 2298 struct spdk_nvmf_request *req = &tcp_req->req; 2299 struct spdk_nvme_cmd *cmd; 2300 struct spdk_nvme_cpl *rsp; 2301 struct spdk_nvme_sgl_descriptor *sgl; 2302 struct spdk_nvmf_tcp_poll_group *tgroup; 2303 uint32_t length; 2304 2305 cmd = &req->cmd->nvme_cmd; 2306 rsp = &req->rsp->nvme_cpl; 2307 sgl = &cmd->dptr.sgl1; 2308 2309 length = sgl->unkeyed.length; 2310 2311 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2312 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2313 if (length > transport->opts.max_io_size) { 2314 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2315 length, transport->opts.max_io_size); 2316 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2317 return -1; 2318 } 2319 2320 /* fill request length and populate iovs */ 2321 req->length = length; 2322 2323 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2324 2325 if (spdk_unlikely(req->dif_enabled)) { 2326 req->dif.orig_length = length; 2327 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2328 req->dif.elba_length = length; 2329 } 2330 2331 if (nvmf_ctrlr_use_zcopy(req)) { 2332 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2333 req->data_from_pool = false; 2334 return 0; 2335 } 2336 2337 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2338 /* No available buffers. Queue this request up. */ 2339 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2340 tcp_req); 2341 return 0; 2342 } 2343 2344 /* backward compatible */ 2345 req->data = req->iov[0].iov_base; 2346 2347 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2348 tcp_req, req->iovcnt, req->data); 2349 2350 return 0; 2351 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2352 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2353 uint64_t offset = sgl->address; 2354 uint32_t max_len = transport->opts.in_capsule_data_size; 2355 assert(tcp_req->has_in_capsule_data); 2356 2357 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2358 offset, length); 2359 2360 if (offset > max_len) { 2361 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2362 offset, max_len); 2363 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2364 return -1; 2365 } 2366 max_len -= (uint32_t)offset; 2367 2368 if (spdk_unlikely(length > max_len)) { 2369 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2370 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2371 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2372 2373 /* Get a buffer from dedicated list */ 2374 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2375 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2376 assert(tgroup->control_msg_list); 2377 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2378 if (!req->data) { 2379 /* No available buffers. Queue this request up. */ 2380 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2381 return 0; 2382 } 2383 } else { 2384 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2385 length, max_len); 2386 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2387 return -1; 2388 } 2389 } else { 2390 req->data = tcp_req->buf; 2391 } 2392 2393 req->length = length; 2394 req->data_from_pool = false; 2395 2396 if (spdk_unlikely(req->dif_enabled)) { 2397 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2398 req->dif.elba_length = length; 2399 } 2400 2401 req->iov[0].iov_base = req->data; 2402 req->iov[0].iov_len = length; 2403 req->iovcnt = 1; 2404 2405 return 0; 2406 } 2407 2408 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2409 sgl->generic.type, sgl->generic.subtype); 2410 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2411 return -1; 2412 } 2413 2414 static inline enum spdk_nvme_media_error_status_code 2415 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2416 enum spdk_nvme_media_error_status_code result; 2417 2418 switch (err_type) 2419 { 2420 case SPDK_DIF_REFTAG_ERROR: 2421 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2422 break; 2423 case SPDK_DIF_APPTAG_ERROR: 2424 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2425 break; 2426 case SPDK_DIF_GUARD_ERROR: 2427 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2428 break; 2429 default: 2430 SPDK_UNREACHABLE(); 2431 break; 2432 } 2433 2434 return result; 2435 } 2436 2437 static void 2438 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2439 struct spdk_nvmf_tcp_req *tcp_req) 2440 { 2441 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2442 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2443 struct nvme_tcp_pdu *rsp_pdu; 2444 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2445 uint32_t plen, pdo, alignment; 2446 int rc; 2447 2448 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2449 2450 rsp_pdu = tcp_req->pdu; 2451 assert(rsp_pdu != NULL); 2452 2453 c2h_data = &rsp_pdu->hdr.c2h_data; 2454 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2455 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2456 2457 if (tqpair->host_hdgst_enable) { 2458 plen += SPDK_NVME_TCP_DIGEST_LEN; 2459 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2460 } 2461 2462 /* set the psh */ 2463 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2464 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2465 c2h_data->datao = tcp_req->pdu->rw_offset; 2466 2467 /* set the padding */ 2468 rsp_pdu->padding_len = 0; 2469 pdo = plen; 2470 if (tqpair->cpda) { 2471 alignment = (tqpair->cpda + 1) << 2; 2472 if (plen % alignment != 0) { 2473 pdo = (plen + alignment) / alignment * alignment; 2474 rsp_pdu->padding_len = pdo - plen; 2475 plen = pdo; 2476 } 2477 } 2478 2479 c2h_data->common.pdo = pdo; 2480 plen += c2h_data->datal; 2481 if (tqpair->host_ddgst_enable) { 2482 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2483 plen += SPDK_NVME_TCP_DIGEST_LEN; 2484 } 2485 2486 c2h_data->common.plen = plen; 2487 2488 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2489 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2490 } 2491 2492 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2493 c2h_data->datao, c2h_data->datal); 2494 2495 2496 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2497 /* Need to send the capsule response if response is not all 0 */ 2498 if (ttransport->tcp_opts.c2h_success && 2499 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2500 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2501 } 2502 2503 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2504 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2505 struct spdk_dif_error err_blk = {}; 2506 uint32_t mapped_length = 0; 2507 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2508 uint32_t ddgst_len = 0; 2509 2510 if (tqpair->host_ddgst_enable) { 2511 /* Data digest consumes additional iov entry */ 2512 available_iovs--; 2513 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2514 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2515 c2h_data->common.plen -= ddgst_len; 2516 } 2517 /* Temp call to estimate if data can be described by limited number of iovs. 2518 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2519 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2520 false, &mapped_length); 2521 2522 if (mapped_length != c2h_data->common.plen) { 2523 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2524 SPDK_DEBUGLOG(nvmf_tcp, 2525 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2526 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2527 c2h_data->common.plen = mapped_length; 2528 2529 /* Rebuild pdu->data_iov since data length is changed */ 2530 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2531 c2h_data->datal); 2532 2533 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2534 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2535 } 2536 2537 c2h_data->common.plen += ddgst_len; 2538 2539 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2540 2541 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2542 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2543 if (rc != 0) { 2544 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2545 err_blk.err_type, err_blk.err_offset); 2546 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2547 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2548 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2549 return; 2550 } 2551 } 2552 2553 rsp_pdu->rw_offset += c2h_data->datal; 2554 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2555 } 2556 2557 static void 2558 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2559 struct spdk_nvmf_tcp_req *tcp_req) 2560 { 2561 nvmf_tcp_req_pdu_init(tcp_req); 2562 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2563 } 2564 2565 static int 2566 request_transfer_out(struct spdk_nvmf_request *req) 2567 { 2568 struct spdk_nvmf_tcp_req *tcp_req; 2569 struct spdk_nvmf_qpair *qpair; 2570 struct spdk_nvmf_tcp_qpair *tqpair; 2571 struct spdk_nvme_cpl *rsp; 2572 2573 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2574 2575 qpair = req->qpair; 2576 rsp = &req->rsp->nvme_cpl; 2577 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2578 2579 /* Advance our sq_head pointer */ 2580 if (qpair->sq_head == qpair->sq_head_max) { 2581 qpair->sq_head = 0; 2582 } else { 2583 qpair->sq_head++; 2584 } 2585 rsp->sqhd = qpair->sq_head; 2586 2587 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2588 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2589 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2590 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2591 } else { 2592 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2593 } 2594 2595 return 0; 2596 } 2597 2598 static void 2599 nvmf_tcp_set_in_capsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2600 struct spdk_nvmf_tcp_req *tcp_req) 2601 { 2602 struct nvme_tcp_pdu *pdu; 2603 uint32_t plen = 0; 2604 2605 pdu = tqpair->pdu_in_progress; 2606 plen = pdu->hdr.common.hlen; 2607 2608 if (tqpair->host_hdgst_enable) { 2609 plen += SPDK_NVME_TCP_DIGEST_LEN; 2610 } 2611 2612 if (pdu->hdr.common.plen != plen) { 2613 tcp_req->has_in_capsule_data = true; 2614 } 2615 } 2616 2617 static void 2618 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2619 struct spdk_nvmf_tcp_qpair *tqpair, 2620 struct spdk_nvmf_tcp_req *tcp_req) 2621 { 2622 enum spdk_nvme_cmd_fuse last, next; 2623 2624 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2625 next = tcp_req->cmd.fuse; 2626 2627 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2628 2629 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2630 return; 2631 } 2632 2633 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2634 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2635 /* This is a valid pair of fused commands. Point them at each other 2636 * so they can be submitted consecutively once ready to be executed. 2637 */ 2638 tqpair->fused_first->fused_pair = tcp_req; 2639 tcp_req->fused_pair = tqpair->fused_first; 2640 tqpair->fused_first = NULL; 2641 return; 2642 } else { 2643 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2644 tqpair->fused_first->fused_failed = true; 2645 2646 /* 2647 * If the last req is in READY_TO_EXECUTE state, then call 2648 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2649 */ 2650 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2651 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2652 } 2653 2654 tqpair->fused_first = NULL; 2655 } 2656 } 2657 2658 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2659 /* Set tqpair->fused_first here so that we know to check that the next request 2660 * is a SECOND (and to fail this one if it isn't). 2661 */ 2662 tqpair->fused_first = tcp_req; 2663 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2664 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2665 tcp_req->fused_failed = true; 2666 } 2667 } 2668 2669 static bool 2670 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2671 struct spdk_nvmf_tcp_req *tcp_req) 2672 { 2673 struct spdk_nvmf_tcp_qpair *tqpair; 2674 int rc; 2675 enum spdk_nvmf_tcp_req_state prev_state; 2676 bool progress = false; 2677 struct spdk_nvmf_transport *transport = &ttransport->transport; 2678 struct spdk_nvmf_transport_poll_group *group; 2679 struct spdk_nvmf_tcp_poll_group *tgroup; 2680 2681 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2682 group = &tqpair->group->group; 2683 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2684 2685 /* If the qpair is not active, we need to abort the outstanding requests. */ 2686 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2687 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2688 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2689 } 2690 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2691 } 2692 2693 /* The loop here is to allow for several back-to-back state changes. */ 2694 do { 2695 prev_state = tcp_req->state; 2696 2697 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2698 tqpair); 2699 2700 switch (tcp_req->state) { 2701 case TCP_REQUEST_STATE_FREE: 2702 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2703 * to escape this state. */ 2704 break; 2705 case TCP_REQUEST_STATE_NEW: 2706 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, tqpair); 2707 2708 /* copy the cmd from the receive pdu */ 2709 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2710 2711 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2712 tcp_req->req.dif_enabled = true; 2713 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2714 } 2715 2716 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2717 2718 /* The next state transition depends on the data transfer needs of this request. */ 2719 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2720 2721 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2722 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2723 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 2724 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2725 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2726 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2727 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2728 break; 2729 } 2730 2731 /* If no data to transfer, ready to execute. */ 2732 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2733 /* Reset the tqpair receiving pdu state */ 2734 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2735 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2736 break; 2737 } 2738 2739 nvmf_tcp_set_in_capsule_data(tqpair, tcp_req); 2740 2741 if (!tcp_req->has_in_capsule_data) { 2742 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2743 } 2744 2745 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2746 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2747 break; 2748 case TCP_REQUEST_STATE_NEED_BUFFER: 2749 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2750 tqpair); 2751 2752 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2753 2754 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2755 SPDK_DEBUGLOG(nvmf_tcp, 2756 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2757 tcp_req, tqpair); 2758 /* This request needs to wait in line to obtain a buffer */ 2759 break; 2760 } 2761 2762 /* Try to get a data buffer */ 2763 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2764 if (rc < 0) { 2765 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2766 /* Reset the tqpair receiving pdu state */ 2767 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2768 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2769 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2770 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2771 break; 2772 } 2773 2774 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2775 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2776 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2777 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2778 tcp_req->req.length = tcp_req->req.dif.elba_length; 2779 } 2780 2781 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2782 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2783 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2784 break; 2785 } 2786 2787 if (!tcp_req->req.data) { 2788 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2789 tcp_req, tqpair); 2790 /* No buffers available. */ 2791 break; 2792 } 2793 2794 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2795 2796 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2797 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2798 if (tcp_req->req.data_from_pool) { 2799 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2800 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2801 } else { 2802 struct nvme_tcp_pdu *pdu; 2803 2804 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2805 2806 pdu = tqpair->pdu_in_progress; 2807 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2808 tqpair); 2809 /* No need to send r2t, contained in the capsuled data */ 2810 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2811 0, tcp_req->req.length); 2812 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2813 } 2814 break; 2815 } 2816 2817 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2818 break; 2819 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 2820 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.qid, 0, 2821 (uintptr_t)tcp_req, tqpair); 2822 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 2823 * to escape this state. */ 2824 break; 2825 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 2826 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.qid, 0, 2827 (uintptr_t)tcp_req, tqpair); 2828 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 2829 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 2830 tcp_req, tqpair); 2831 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2832 break; 2833 } 2834 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2835 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2836 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2837 } else { 2838 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2839 } 2840 break; 2841 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2842 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2843 tqpair); 2844 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2845 break; 2846 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2847 2848 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.qid, 0, 2849 (uintptr_t)tcp_req, tqpair); 2850 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2851 * to escape this state. */ 2852 break; 2853 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2854 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.qid, 0, 2855 (uintptr_t)tcp_req, tqpair); 2856 2857 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2858 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2859 tcp_req->req.length = tcp_req->req.dif.elba_length; 2860 } 2861 2862 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 2863 if (tcp_req->fused_failed) { 2864 /* This request failed FUSED semantics. Fail it immediately, without 2865 * even sending it to the target layer. 2866 */ 2867 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2868 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 2869 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2870 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2871 break; 2872 } 2873 2874 if (tcp_req->fused_pair == NULL || 2875 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2876 /* This request is ready to execute, but either we don't know yet if it's 2877 * valid - i.e. this is a FIRST but we haven't received the next request yet), 2878 * or the other request of this fused pair isn't ready to execute. So 2879 * break here and this request will get processed later either when the 2880 * other request is ready or we find that this request isn't valid. 2881 */ 2882 break; 2883 } 2884 } 2885 2886 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2887 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2888 /* If we get to this point, and this request is a fused command, we know that 2889 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 2890 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 2891 * request, and the other request of the fused pair, in the correct order. 2892 * Also clear the ->fused_pair pointers on both requests, since after this point 2893 * we no longer need to maintain the relationship between these two requests. 2894 */ 2895 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 2896 assert(tcp_req->fused_pair != NULL); 2897 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2898 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2899 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2900 tcp_req->fused_pair->fused_pair = NULL; 2901 tcp_req->fused_pair = NULL; 2902 } 2903 spdk_nvmf_request_exec(&tcp_req->req); 2904 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 2905 assert(tcp_req->fused_pair != NULL); 2906 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2907 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2908 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2909 tcp_req->fused_pair->fused_pair = NULL; 2910 tcp_req->fused_pair = NULL; 2911 } 2912 } else { 2913 /* For zero-copy, only requests with data coming from host to the 2914 * controller can end up here. */ 2915 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 2916 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 2917 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 2918 } 2919 2920 break; 2921 case TCP_REQUEST_STATE_EXECUTING: 2922 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2923 tqpair); 2924 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2925 * to escape this state. */ 2926 break; 2927 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 2928 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.qid, 0, 2929 (uintptr_t)tcp_req, tqpair); 2930 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2931 * to escape this state. */ 2932 break; 2933 case TCP_REQUEST_STATE_EXECUTED: 2934 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2935 tqpair); 2936 2937 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2938 tcp_req->req.length = tcp_req->req.dif.orig_length; 2939 } 2940 2941 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2942 break; 2943 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2944 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.qid, 0, 2945 (uintptr_t)tcp_req, tqpair); 2946 rc = request_transfer_out(&tcp_req->req); 2947 assert(rc == 0); /* No good way to handle this currently */ 2948 break; 2949 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2950 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.qid, 0, 2951 (uintptr_t)tcp_req, tqpair); 2952 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2953 * to escape this state. */ 2954 break; 2955 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 2956 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.qid, 0, 2957 (uintptr_t)tcp_req, tqpair); 2958 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2959 * to escape this state. */ 2960 break; 2961 case TCP_REQUEST_STATE_COMPLETED: 2962 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2963 tqpair); 2964 /* If there's an outstanding PDU sent to the host, the request is completed 2965 * due to the qpair being disconnected. We must delay the completion until 2966 * that write is done to avoid freeing the request twice. */ 2967 if (spdk_unlikely(tcp_req->pdu_in_use)) { 2968 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 2969 "write on req=%p\n", tcp_req); 2970 /* This can only happen for zcopy requests */ 2971 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 2972 assert(tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 2973 break; 2974 } 2975 2976 if (tcp_req->req.data_from_pool) { 2977 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2978 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 2979 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2980 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2981 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2982 assert(tgroup->control_msg_list); 2983 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2984 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2985 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 2986 /* If the request has an unreleased zcopy bdev_io, it's either a 2987 * read, a failed write, or the qpair is being disconnected */ 2988 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 2989 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 2990 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 2991 tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 2992 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 2993 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 2994 break; 2995 } 2996 tcp_req->req.length = 0; 2997 tcp_req->req.iovcnt = 0; 2998 tcp_req->req.data = NULL; 2999 tcp_req->fused_failed = false; 3000 if (tcp_req->fused_pair) { 3001 /* This req was part of a valid fused pair, but failed before it got to 3002 * READ_TO_EXECUTE state. This means we need to fail the other request 3003 * in the pair, because it is no longer part of a valid pair. If the pair 3004 * already reached READY_TO_EXECUTE state, we need to kick it. 3005 */ 3006 tcp_req->fused_pair->fused_failed = true; 3007 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3008 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3009 } 3010 tcp_req->fused_pair = NULL; 3011 } 3012 3013 nvmf_tcp_req_put(tqpair, tcp_req); 3014 break; 3015 case TCP_REQUEST_NUM_STATES: 3016 default: 3017 assert(0); 3018 break; 3019 } 3020 3021 if (tcp_req->state != prev_state) { 3022 progress = true; 3023 } 3024 } while (tcp_req->state != prev_state); 3025 3026 return progress; 3027 } 3028 3029 static void 3030 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3031 { 3032 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3033 int rc; 3034 3035 assert(tqpair != NULL); 3036 rc = nvmf_tcp_sock_process(tqpair); 3037 3038 /* If there was a new socket error, disconnect */ 3039 if (rc < 0) { 3040 nvmf_tcp_qpair_disconnect(tqpair); 3041 } 3042 } 3043 3044 static int 3045 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3046 struct spdk_nvmf_qpair *qpair) 3047 { 3048 struct spdk_nvmf_tcp_poll_group *tgroup; 3049 struct spdk_nvmf_tcp_qpair *tqpair; 3050 int rc; 3051 3052 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3053 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3054 3055 rc = nvmf_tcp_qpair_sock_init(tqpair); 3056 if (rc != 0) { 3057 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3058 return -1; 3059 } 3060 3061 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3062 if (rc < 0) { 3063 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3064 return -1; 3065 } 3066 3067 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3068 if (rc < 0) { 3069 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3070 return -1; 3071 } 3072 3073 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3074 nvmf_tcp_sock_cb, tqpair); 3075 if (rc != 0) { 3076 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3077 spdk_strerror(errno), errno); 3078 return -1; 3079 } 3080 3081 tqpair->group = tgroup; 3082 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3083 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3084 3085 return 0; 3086 } 3087 3088 static int 3089 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3090 struct spdk_nvmf_qpair *qpair) 3091 { 3092 struct spdk_nvmf_tcp_poll_group *tgroup; 3093 struct spdk_nvmf_tcp_qpair *tqpair; 3094 int rc; 3095 3096 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3097 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3098 3099 assert(tqpair->group == tgroup); 3100 3101 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3102 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3103 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 3104 } else { 3105 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3106 } 3107 3108 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3109 if (rc != 0) { 3110 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3111 spdk_strerror(errno), errno); 3112 } 3113 3114 return rc; 3115 } 3116 3117 static int 3118 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3119 { 3120 struct spdk_nvmf_tcp_transport *ttransport; 3121 struct spdk_nvmf_tcp_req *tcp_req; 3122 3123 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3124 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3125 3126 switch (tcp_req->state) { 3127 case TCP_REQUEST_STATE_EXECUTING: 3128 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3129 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3130 break; 3131 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3132 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3133 break; 3134 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3135 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3136 break; 3137 default: 3138 assert(0 && "Unexpected request state"); 3139 break; 3140 } 3141 3142 nvmf_tcp_req_process(ttransport, tcp_req); 3143 3144 return 0; 3145 } 3146 3147 static void 3148 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3149 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3150 { 3151 struct spdk_nvmf_tcp_qpair *tqpair; 3152 3153 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3154 3155 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3156 3157 assert(tqpair->fini_cb_fn == NULL); 3158 tqpair->fini_cb_fn = cb_fn; 3159 tqpair->fini_cb_arg = cb_arg; 3160 3161 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3162 nvmf_tcp_qpair_destroy(tqpair); 3163 } 3164 3165 static int 3166 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3167 { 3168 struct spdk_nvmf_tcp_poll_group *tgroup; 3169 int rc; 3170 struct spdk_nvmf_request *req, *req_tmp; 3171 struct spdk_nvmf_tcp_req *tcp_req; 3172 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3173 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3174 struct spdk_nvmf_tcp_transport, transport); 3175 3176 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3177 3178 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3179 return 0; 3180 } 3181 3182 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3183 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3184 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3185 break; 3186 } 3187 } 3188 3189 rc = spdk_sock_group_poll(tgroup->sock_group); 3190 if (rc < 0) { 3191 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3192 } 3193 3194 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3195 nvmf_tcp_sock_process(tqpair); 3196 } 3197 3198 return rc; 3199 } 3200 3201 static int 3202 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3203 struct spdk_nvme_transport_id *trid, bool peer) 3204 { 3205 struct spdk_nvmf_tcp_qpair *tqpair; 3206 uint16_t port; 3207 3208 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3209 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3210 3211 if (peer) { 3212 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3213 port = tqpair->initiator_port; 3214 } else { 3215 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3216 port = tqpair->target_port; 3217 } 3218 3219 if (spdk_sock_is_ipv4(tqpair->sock)) { 3220 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3221 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3222 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3223 } else { 3224 return -1; 3225 } 3226 3227 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3228 return 0; 3229 } 3230 3231 static int 3232 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3233 struct spdk_nvme_transport_id *trid) 3234 { 3235 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3236 } 3237 3238 static int 3239 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3240 struct spdk_nvme_transport_id *trid) 3241 { 3242 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3243 } 3244 3245 static int 3246 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3247 struct spdk_nvme_transport_id *trid) 3248 { 3249 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3250 } 3251 3252 static void 3253 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3254 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3255 { 3256 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3257 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3258 tcp_req_to_abort->req.rsp->nvme_cpl.cid = tcp_req_to_abort->req.cmd->nvme_cmd.cid; 3259 3260 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3261 3262 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3263 } 3264 3265 static int 3266 _nvmf_tcp_qpair_abort_request(void *ctx) 3267 { 3268 struct spdk_nvmf_request *req = ctx; 3269 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3270 struct spdk_nvmf_tcp_req, req); 3271 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3272 struct spdk_nvmf_tcp_qpair, qpair); 3273 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3274 struct spdk_nvmf_tcp_transport, transport); 3275 int rc; 3276 3277 spdk_poller_unregister(&req->poller); 3278 3279 switch (tcp_req_to_abort->state) { 3280 case TCP_REQUEST_STATE_EXECUTING: 3281 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3282 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3283 rc = nvmf_ctrlr_abort_request(req); 3284 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3285 return SPDK_POLLER_BUSY; 3286 } 3287 break; 3288 3289 case TCP_REQUEST_STATE_NEED_BUFFER: 3290 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3291 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3292 3293 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3294 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3295 break; 3296 3297 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3298 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3299 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3300 break; 3301 3302 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3303 if (spdk_get_ticks() < req->timeout_tsc) { 3304 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3305 return SPDK_POLLER_BUSY; 3306 } 3307 break; 3308 3309 default: 3310 break; 3311 } 3312 3313 spdk_nvmf_request_complete(req); 3314 return SPDK_POLLER_BUSY; 3315 } 3316 3317 static void 3318 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3319 struct spdk_nvmf_request *req) 3320 { 3321 struct spdk_nvmf_tcp_qpair *tqpair; 3322 struct spdk_nvmf_tcp_transport *ttransport; 3323 struct spdk_nvmf_transport *transport; 3324 uint16_t cid; 3325 uint32_t i; 3326 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3327 3328 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3329 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3330 transport = &ttransport->transport; 3331 3332 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3333 3334 for (i = 0; i < tqpair->resource_count; i++) { 3335 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3336 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3337 tcp_req_to_abort = &tqpair->reqs[i]; 3338 break; 3339 } 3340 } 3341 3342 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, qpair->qid, 0, (uintptr_t)req, tqpair); 3343 3344 if (tcp_req_to_abort == NULL) { 3345 spdk_nvmf_request_complete(req); 3346 return; 3347 } 3348 3349 req->req_to_abort = &tcp_req_to_abort->req; 3350 req->timeout_tsc = spdk_get_ticks() + 3351 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3352 req->poller = NULL; 3353 3354 _nvmf_tcp_qpair_abort_request(req); 3355 } 3356 3357 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 3358 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 3359 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 3360 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 3361 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 3362 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 3363 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 3364 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 3365 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 3366 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 3367 3368 static void 3369 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3370 { 3371 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 3372 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3373 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3374 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3375 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3376 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 3377 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3378 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3379 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3380 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3381 opts->transport_specific = NULL; 3382 } 3383 3384 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3385 .name = "TCP", 3386 .type = SPDK_NVME_TRANSPORT_TCP, 3387 .opts_init = nvmf_tcp_opts_init, 3388 .create = nvmf_tcp_create, 3389 .dump_opts = nvmf_tcp_dump_opts, 3390 .destroy = nvmf_tcp_destroy, 3391 3392 .listen = nvmf_tcp_listen, 3393 .stop_listen = nvmf_tcp_stop_listen, 3394 3395 .listener_discover = nvmf_tcp_discover, 3396 3397 .poll_group_create = nvmf_tcp_poll_group_create, 3398 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3399 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3400 .poll_group_add = nvmf_tcp_poll_group_add, 3401 .poll_group_remove = nvmf_tcp_poll_group_remove, 3402 .poll_group_poll = nvmf_tcp_poll_group_poll, 3403 3404 .req_free = nvmf_tcp_req_free, 3405 .req_complete = nvmf_tcp_req_complete, 3406 3407 .qpair_fini = nvmf_tcp_close_qpair, 3408 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3409 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3410 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3411 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3412 }; 3413 3414 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3415 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3416