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.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 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 250 251 /* Number of requests in each state */ 252 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 253 254 uint8_t cpda; 255 256 bool host_hdgst_enable; 257 bool host_ddgst_enable; 258 259 /* This is a spare PDU used for sending special management 260 * operations. Primarily, this is used for the initial 261 * connection response and c2h termination request. */ 262 struct nvme_tcp_pdu *mgmt_pdu; 263 264 /* Arrays of in-capsule buffers, requests, and pdus. 265 * Each array is 'resource_count' number of elements */ 266 void *bufs; 267 struct spdk_nvmf_tcp_req *reqs; 268 struct nvme_tcp_pdu *pdus; 269 uint32_t resource_count; 270 uint32_t recv_buf_size; 271 272 struct spdk_nvmf_tcp_port *port; 273 274 /* IP address */ 275 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 276 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 277 278 /* IP port */ 279 uint16_t initiator_port; 280 uint16_t target_port; 281 282 /* Timer used to destroy qpair after detecting transport error issue if initiator does 283 * not close the connection. 284 */ 285 struct spdk_poller *timeout_poller; 286 287 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 288 void *fini_cb_arg; 289 290 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 291 }; 292 293 struct spdk_nvmf_tcp_control_msg { 294 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 295 }; 296 297 struct spdk_nvmf_tcp_control_msg_list { 298 void *msg_buf; 299 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 300 }; 301 302 struct spdk_nvmf_tcp_poll_group { 303 struct spdk_nvmf_transport_poll_group group; 304 struct spdk_sock_group *sock_group; 305 306 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 307 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 308 309 struct spdk_io_channel *accel_channel; 310 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 311 312 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 313 }; 314 315 struct spdk_nvmf_tcp_port { 316 const struct spdk_nvme_transport_id *trid; 317 struct spdk_sock *listen_sock; 318 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 319 }; 320 321 struct tcp_transport_opts { 322 bool c2h_success; 323 uint16_t control_msg_num; 324 uint32_t sock_priority; 325 }; 326 327 struct spdk_nvmf_tcp_transport { 328 struct spdk_nvmf_transport transport; 329 struct tcp_transport_opts tcp_opts; 330 331 struct spdk_nvmf_tcp_poll_group *next_pg; 332 333 struct spdk_poller *accept_poller; 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 inline 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 (spdk_unlikely(!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 free(ttransport); 575 576 if (cb_fn) { 577 cb_fn(cb_arg); 578 } 579 return 0; 580 } 581 582 static int nvmf_tcp_accept(void *ctx); 583 584 static struct spdk_nvmf_transport * 585 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 586 { 587 struct spdk_nvmf_tcp_transport *ttransport; 588 uint32_t sge_count; 589 uint32_t min_shared_buffers; 590 591 ttransport = calloc(1, sizeof(*ttransport)); 592 if (!ttransport) { 593 return NULL; 594 } 595 596 TAILQ_INIT(&ttransport->ports); 597 TAILQ_INIT(&ttransport->poll_groups); 598 599 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 600 601 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 602 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 603 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 604 if (opts->transport_specific != NULL && 605 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 606 SPDK_COUNTOF(tcp_transport_opts_decoder), 607 &ttransport->tcp_opts)) { 608 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 609 free(ttransport); 610 return NULL; 611 } 612 613 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 614 615 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 616 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 617 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 618 " in_capsule_data_size=%d, max_aq_depth=%d\n" 619 " num_shared_buffers=%d, c2h_success=%d,\n" 620 " dif_insert_or_strip=%d, sock_priority=%d\n" 621 " abort_timeout_sec=%d, control_msg_num=%hu\n", 622 opts->max_queue_depth, 623 opts->max_io_size, 624 opts->max_qpairs_per_ctrlr - 1, 625 opts->io_unit_size, 626 opts->in_capsule_data_size, 627 opts->max_aq_depth, 628 opts->num_shared_buffers, 629 ttransport->tcp_opts.c2h_success, 630 opts->dif_insert_or_strip, 631 ttransport->tcp_opts.sock_priority, 632 opts->abort_timeout_sec, 633 ttransport->tcp_opts.control_msg_num); 634 635 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 636 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 637 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 638 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 639 free(ttransport); 640 return NULL; 641 } 642 643 if (ttransport->tcp_opts.control_msg_num == 0 && 644 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 645 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 646 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 647 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 648 } 649 650 /* I/O unit size cannot be larger than max I/O size */ 651 if (opts->io_unit_size > opts->max_io_size) { 652 opts->io_unit_size = opts->max_io_size; 653 } 654 655 sge_count = opts->max_io_size / opts->io_unit_size; 656 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 657 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 658 free(ttransport); 659 return NULL; 660 } 661 662 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 663 if (min_shared_buffers > opts->num_shared_buffers) { 664 SPDK_ERRLOG("There are not enough buffers to satisfy " 665 "per-poll group caches for each thread. (%" PRIu32 ") " 666 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 667 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 668 free(ttransport); 669 return NULL; 670 } 671 672 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 673 opts->acceptor_poll_rate); 674 if (!ttransport->accept_poller) { 675 free(ttransport); 676 return NULL; 677 } 678 679 return &ttransport->transport; 680 } 681 682 static int 683 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 684 { 685 unsigned long long ull; 686 char *end = NULL; 687 688 ull = strtoull(trsvcid, &end, 10); 689 if (end == NULL || end == trsvcid || *end != '\0') { 690 return -1; 691 } 692 693 /* Valid TCP/IP port numbers are in [0, 65535] */ 694 if (ull > 65535) { 695 return -1; 696 } 697 698 return (int)ull; 699 } 700 701 /** 702 * Canonicalize a listen address trid. 703 */ 704 static int 705 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 706 const struct spdk_nvme_transport_id *trid) 707 { 708 int trsvcid_int; 709 710 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 711 if (trsvcid_int < 0) { 712 return -EINVAL; 713 } 714 715 memset(canon_trid, 0, sizeof(*canon_trid)); 716 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 717 canon_trid->adrfam = trid->adrfam; 718 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 719 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 720 721 return 0; 722 } 723 724 /** 725 * Find an existing listening port. 726 */ 727 static struct spdk_nvmf_tcp_port * 728 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 729 const struct spdk_nvme_transport_id *trid) 730 { 731 struct spdk_nvme_transport_id canon_trid; 732 struct spdk_nvmf_tcp_port *port; 733 734 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 735 return NULL; 736 } 737 738 TAILQ_FOREACH(port, &ttransport->ports, link) { 739 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 740 return port; 741 } 742 } 743 744 return NULL; 745 } 746 747 static int 748 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 749 struct spdk_nvmf_listen_opts *listen_opts) 750 { 751 struct spdk_nvmf_tcp_transport *ttransport; 752 struct spdk_nvmf_tcp_port *port; 753 int trsvcid_int; 754 uint8_t adrfam; 755 struct spdk_sock_opts opts; 756 757 if (!strlen(trid->trsvcid)) { 758 SPDK_ERRLOG("Service id is required\n"); 759 return -EINVAL; 760 } 761 762 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 763 764 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 765 if (trsvcid_int < 0) { 766 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 767 return -EINVAL; 768 } 769 770 port = calloc(1, sizeof(*port)); 771 if (!port) { 772 SPDK_ERRLOG("Port allocation failed\n"); 773 return -ENOMEM; 774 } 775 776 port->trid = trid; 777 opts.opts_size = sizeof(opts); 778 spdk_sock_get_default_opts(&opts); 779 opts.priority = ttransport->tcp_opts.sock_priority; 780 /* TODO: also add impl_opts like on the initiator side */ 781 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 782 NULL, &opts); 783 if (port->listen_sock == NULL) { 784 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 785 trid->traddr, trsvcid_int, 786 spdk_strerror(errno), errno); 787 free(port); 788 return -errno; 789 } 790 791 if (spdk_sock_is_ipv4(port->listen_sock)) { 792 adrfam = SPDK_NVMF_ADRFAM_IPV4; 793 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 794 adrfam = SPDK_NVMF_ADRFAM_IPV6; 795 } else { 796 SPDK_ERRLOG("Unhandled socket type\n"); 797 adrfam = 0; 798 } 799 800 if (adrfam != trid->adrfam) { 801 SPDK_ERRLOG("Socket address family mismatch\n"); 802 spdk_sock_close(&port->listen_sock); 803 free(port); 804 return -EINVAL; 805 } 806 807 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 808 trid->traddr, trid->trsvcid); 809 810 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 811 return 0; 812 } 813 814 static void 815 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 816 const struct spdk_nvme_transport_id *trid) 817 { 818 struct spdk_nvmf_tcp_transport *ttransport; 819 struct spdk_nvmf_tcp_port *port; 820 821 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 822 823 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 824 trid->traddr, trid->trsvcid); 825 826 port = nvmf_tcp_find_port(ttransport, trid); 827 if (port) { 828 TAILQ_REMOVE(&ttransport->ports, port, link); 829 spdk_sock_close(&port->listen_sock); 830 free(port); 831 } 832 } 833 834 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 835 enum nvme_tcp_pdu_recv_state state); 836 837 static void 838 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 839 { 840 tqpair->state = state; 841 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 842 tqpair->state); 843 } 844 845 static void 846 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 847 { 848 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 849 850 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, 0, 0, (uintptr_t)tqpair); 851 852 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 853 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 854 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 855 spdk_poller_unregister(&tqpair->timeout_poller); 856 857 /* This will end up calling nvmf_tcp_close_qpair */ 858 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 859 } 860 } 861 862 static void 863 _mgmt_pdu_write_done(void *_tqpair, int err) 864 { 865 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 866 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 867 868 if (spdk_unlikely(err != 0)) { 869 nvmf_tcp_qpair_disconnect(tqpair); 870 return; 871 } 872 873 assert(pdu->cb_fn != NULL); 874 pdu->cb_fn(pdu->cb_arg); 875 } 876 877 static void 878 _req_pdu_write_done(void *req, int err) 879 { 880 struct spdk_nvmf_tcp_req *tcp_req = req; 881 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 882 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 883 884 assert(tcp_req->pdu_in_use); 885 tcp_req->pdu_in_use = false; 886 887 /* If the request is in a completed state, we're waiting for write completion to free it */ 888 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 889 nvmf_tcp_request_free(tcp_req); 890 return; 891 } 892 893 if (spdk_unlikely(err != 0)) { 894 nvmf_tcp_qpair_disconnect(tqpair); 895 return; 896 } 897 898 assert(pdu->cb_fn != NULL); 899 pdu->cb_fn(pdu->cb_arg); 900 } 901 902 static void 903 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 904 { 905 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 906 } 907 908 static void 909 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 910 { 911 uint32_t mapped_length = 0; 912 ssize_t rc; 913 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 914 915 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 916 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 917 &mapped_length); 918 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 919 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 920 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 921 if (rc == mapped_length) { 922 _pdu_write_done(pdu, 0); 923 } else { 924 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 925 _pdu_write_done(pdu, -1); 926 } 927 } else { 928 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 929 } 930 } 931 932 static void 933 data_crc32_accel_done(void *cb_arg, int status) 934 { 935 struct nvme_tcp_pdu *pdu = cb_arg; 936 937 if (spdk_unlikely(status)) { 938 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 939 _pdu_write_done(pdu, status); 940 return; 941 } 942 943 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 944 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 945 946 _tcp_write_pdu(pdu); 947 } 948 949 static void 950 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 951 { 952 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 953 int rc = 0; 954 955 /* Data Digest */ 956 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 957 /* Only suport this limitated case for the first step */ 958 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 959 && tqpair->group)) { 960 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 961 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 962 if (spdk_likely(rc == 0)) { 963 return; 964 } 965 } else { 966 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 967 } 968 data_crc32_accel_done(pdu, rc); 969 } else { 970 _tcp_write_pdu(pdu); 971 } 972 } 973 974 static void 975 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 976 struct nvme_tcp_pdu *pdu, 977 nvme_tcp_qpair_xfer_complete_cb cb_fn, 978 void *cb_arg) 979 { 980 int hlen; 981 uint32_t crc32c; 982 983 assert(tqpair->pdu_in_progress != pdu); 984 985 hlen = pdu->hdr.common.hlen; 986 pdu->cb_fn = cb_fn; 987 pdu->cb_arg = cb_arg; 988 989 pdu->iov[0].iov_base = &pdu->hdr.raw; 990 pdu->iov[0].iov_len = hlen; 991 992 /* Header Digest */ 993 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 994 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 995 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 996 } 997 998 /* Data Digest */ 999 pdu_data_crc32_compute(pdu); 1000 } 1001 1002 static void 1003 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1004 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1005 void *cb_arg) 1006 { 1007 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1008 1009 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1010 pdu->sock_req.cb_arg = tqpair; 1011 1012 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1013 } 1014 1015 static void 1016 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1017 struct spdk_nvmf_tcp_req *tcp_req, 1018 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1019 void *cb_arg) 1020 { 1021 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1022 1023 pdu->sock_req.cb_fn = _req_pdu_write_done; 1024 pdu->sock_req.cb_arg = tcp_req; 1025 1026 assert(!tcp_req->pdu_in_use); 1027 tcp_req->pdu_in_use = true; 1028 1029 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1030 } 1031 1032 static int 1033 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1034 { 1035 uint32_t i; 1036 struct spdk_nvmf_transport_opts *opts; 1037 uint32_t in_capsule_data_size; 1038 1039 opts = &tqpair->qpair.transport->opts; 1040 1041 in_capsule_data_size = opts->in_capsule_data_size; 1042 if (opts->dif_insert_or_strip) { 1043 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1044 } 1045 1046 tqpair->resource_count = opts->max_queue_depth; 1047 1048 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1049 if (!tqpair->reqs) { 1050 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1051 return -1; 1052 } 1053 1054 if (in_capsule_data_size) { 1055 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1056 NULL, SPDK_ENV_LCORE_ID_ANY, 1057 SPDK_MALLOC_DMA); 1058 if (!tqpair->bufs) { 1059 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1060 return -1; 1061 } 1062 } 1063 /* prepare memory space for receiving pdus and tcp_req */ 1064 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1065 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1066 NULL); 1067 if (!tqpair->pdus) { 1068 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1069 return -1; 1070 } 1071 1072 for (i = 0; i < tqpair->resource_count; i++) { 1073 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1074 1075 tcp_req->ttag = i + 1; 1076 tcp_req->req.qpair = &tqpair->qpair; 1077 1078 tcp_req->pdu = &tqpair->pdus[i]; 1079 tcp_req->pdu->qpair = tqpair; 1080 1081 /* Set up memory to receive commands */ 1082 if (tqpair->bufs) { 1083 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1084 } 1085 1086 /* Set the cmdn and rsp */ 1087 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1088 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1089 1090 tcp_req->req.stripped_data = NULL; 1091 1092 /* Initialize request state to FREE */ 1093 tcp_req->state = TCP_REQUEST_STATE_FREE; 1094 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1095 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1096 } 1097 1098 for (; i < 2 * tqpair->resource_count; i++) { 1099 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1100 1101 pdu->qpair = tqpair; 1102 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1103 } 1104 1105 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1106 tqpair->mgmt_pdu->qpair = tqpair; 1107 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1108 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1109 1110 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1111 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1112 1113 return 0; 1114 } 1115 1116 static int 1117 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1118 { 1119 struct spdk_nvmf_tcp_qpair *tqpair; 1120 1121 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1122 1123 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1124 1125 spdk_trace_record(TRACE_TCP_QP_CREATE, 0, 0, (uintptr_t)tqpair); 1126 1127 /* Initialise request state queues of the qpair */ 1128 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1129 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1130 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1131 1132 tqpair->host_hdgst_enable = true; 1133 tqpair->host_ddgst_enable = true; 1134 1135 return 0; 1136 } 1137 1138 static int 1139 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1140 { 1141 int rc; 1142 1143 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, 0, 0, (uintptr_t)tqpair); 1144 1145 /* set low water mark */ 1146 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1147 if (rc != 0) { 1148 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1149 return rc; 1150 } 1151 1152 return 0; 1153 } 1154 1155 static void 1156 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1157 struct spdk_nvmf_tcp_port *port, 1158 struct spdk_sock *sock) 1159 { 1160 struct spdk_nvmf_tcp_qpair *tqpair; 1161 int rc; 1162 1163 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1164 port->trid->traddr, port->trid->trsvcid); 1165 1166 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1167 if (tqpair == NULL) { 1168 SPDK_ERRLOG("Could not allocate new connection.\n"); 1169 spdk_sock_close(&sock); 1170 return; 1171 } 1172 1173 tqpair->sock = sock; 1174 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1175 tqpair->port = port; 1176 tqpair->qpair.transport = transport; 1177 1178 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1179 sizeof(tqpair->target_addr), &tqpair->target_port, 1180 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1181 &tqpair->initiator_port); 1182 if (rc < 0) { 1183 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1184 nvmf_tcp_qpair_destroy(tqpair); 1185 return; 1186 } 1187 1188 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1189 } 1190 1191 static uint32_t 1192 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1193 { 1194 struct spdk_sock *sock; 1195 uint32_t count = 0; 1196 int i; 1197 1198 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1199 sock = spdk_sock_accept(port->listen_sock); 1200 if (sock == NULL) { 1201 break; 1202 } 1203 count++; 1204 nvmf_tcp_handle_connect(transport, port, sock); 1205 } 1206 1207 return count; 1208 } 1209 1210 static int 1211 nvmf_tcp_accept(void *ctx) 1212 { 1213 struct spdk_nvmf_transport *transport = ctx; 1214 struct spdk_nvmf_tcp_transport *ttransport; 1215 struct spdk_nvmf_tcp_port *port; 1216 uint32_t count = 0; 1217 1218 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1219 1220 TAILQ_FOREACH(port, &ttransport->ports, link) { 1221 count += nvmf_tcp_port_accept(transport, port); 1222 } 1223 1224 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1225 } 1226 1227 static void 1228 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1229 struct spdk_nvme_transport_id *trid, 1230 struct spdk_nvmf_discovery_log_page_entry *entry) 1231 { 1232 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1233 entry->adrfam = trid->adrfam; 1234 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1235 1236 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1237 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1238 1239 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1240 } 1241 1242 static struct spdk_nvmf_tcp_control_msg_list * 1243 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1244 { 1245 struct spdk_nvmf_tcp_control_msg_list *list; 1246 struct spdk_nvmf_tcp_control_msg *msg; 1247 uint16_t i; 1248 1249 list = calloc(1, sizeof(*list)); 1250 if (!list) { 1251 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1252 return NULL; 1253 } 1254 1255 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1256 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1257 if (!list->msg_buf) { 1258 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1259 free(list); 1260 return NULL; 1261 } 1262 1263 STAILQ_INIT(&list->free_msgs); 1264 1265 for (i = 0; i < num_messages; i++) { 1266 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1267 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1268 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1269 } 1270 1271 return list; 1272 } 1273 1274 static void 1275 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1276 { 1277 if (!list) { 1278 return; 1279 } 1280 1281 spdk_free(list->msg_buf); 1282 free(list); 1283 } 1284 1285 static struct spdk_nvmf_transport_poll_group * 1286 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1287 struct spdk_nvmf_poll_group *group) 1288 { 1289 struct spdk_nvmf_tcp_transport *ttransport; 1290 struct spdk_nvmf_tcp_poll_group *tgroup; 1291 1292 tgroup = calloc(1, sizeof(*tgroup)); 1293 if (!tgroup) { 1294 return NULL; 1295 } 1296 1297 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1298 if (!tgroup->sock_group) { 1299 goto cleanup; 1300 } 1301 1302 TAILQ_INIT(&tgroup->qpairs); 1303 TAILQ_INIT(&tgroup->await_req); 1304 1305 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1306 1307 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1308 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1309 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1310 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1311 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1312 if (!tgroup->control_msg_list) { 1313 goto cleanup; 1314 } 1315 } 1316 1317 tgroup->accel_channel = spdk_accel_get_io_channel(); 1318 if (spdk_unlikely(!tgroup->accel_channel)) { 1319 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1320 goto cleanup; 1321 } 1322 1323 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1324 if (ttransport->next_pg == NULL) { 1325 ttransport->next_pg = tgroup; 1326 } 1327 1328 return &tgroup->group; 1329 1330 cleanup: 1331 nvmf_tcp_poll_group_destroy(&tgroup->group); 1332 return NULL; 1333 } 1334 1335 static struct spdk_nvmf_transport_poll_group * 1336 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1337 { 1338 struct spdk_nvmf_tcp_transport *ttransport; 1339 struct spdk_nvmf_tcp_poll_group **pg; 1340 struct spdk_nvmf_tcp_qpair *tqpair; 1341 struct spdk_sock_group *group = NULL, *hint = NULL; 1342 int rc; 1343 1344 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1345 1346 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1347 return NULL; 1348 } 1349 1350 pg = &ttransport->next_pg; 1351 assert(*pg != NULL); 1352 hint = (*pg)->sock_group; 1353 1354 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1355 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1356 if (rc != 0) { 1357 return NULL; 1358 } else if (group != NULL) { 1359 /* Optimal poll group was found */ 1360 return spdk_sock_group_get_ctx(group); 1361 } 1362 1363 /* The hint was used for optimal poll group, advance next_pg. */ 1364 *pg = TAILQ_NEXT(*pg, link); 1365 if (*pg == NULL) { 1366 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1367 } 1368 1369 return spdk_sock_group_get_ctx(hint); 1370 } 1371 1372 static void 1373 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1374 { 1375 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1376 struct spdk_nvmf_tcp_transport *ttransport; 1377 1378 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1379 spdk_sock_group_close(&tgroup->sock_group); 1380 if (tgroup->control_msg_list) { 1381 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1382 } 1383 1384 if (tgroup->accel_channel) { 1385 spdk_put_io_channel(tgroup->accel_channel); 1386 } 1387 1388 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1389 1390 next_tgroup = TAILQ_NEXT(tgroup, link); 1391 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1392 if (next_tgroup == NULL) { 1393 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1394 } 1395 if (ttransport->next_pg == tgroup) { 1396 ttransport->next_pg = next_tgroup; 1397 } 1398 1399 free(tgroup); 1400 } 1401 1402 static void 1403 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1404 enum nvme_tcp_pdu_recv_state state) 1405 { 1406 if (tqpair->recv_state == state) { 1407 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1408 tqpair, state); 1409 return; 1410 } 1411 1412 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1413 /* When leaving the await req state, move the qpair to the main list */ 1414 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1415 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1416 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1417 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1418 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1419 } 1420 1421 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1422 tqpair->recv_state = state; 1423 1424 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 1425 tqpair->recv_state); 1426 } 1427 1428 static int 1429 nvmf_tcp_qpair_handle_timeout(void *ctx) 1430 { 1431 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1432 1433 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1434 1435 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1436 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1437 1438 nvmf_tcp_qpair_disconnect(tqpair); 1439 return SPDK_POLLER_BUSY; 1440 } 1441 1442 static void 1443 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1444 { 1445 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1446 1447 if (!tqpair->timeout_poller) { 1448 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1449 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1450 } 1451 } 1452 1453 static void 1454 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1455 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1456 { 1457 struct nvme_tcp_pdu *rsp_pdu; 1458 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1459 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1460 uint32_t copy_len; 1461 1462 rsp_pdu = tqpair->mgmt_pdu; 1463 1464 c2h_term_req = &rsp_pdu->hdr.term_req; 1465 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1466 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1467 c2h_term_req->fes = fes; 1468 1469 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1470 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1471 DSET32(&c2h_term_req->fei, error_offset); 1472 } 1473 1474 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1475 1476 /* Copy the error info into the buffer */ 1477 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1478 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1479 1480 /* Contain the header of the wrong received pdu */ 1481 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1482 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1483 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1484 } 1485 1486 static void 1487 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1488 struct spdk_nvmf_tcp_qpair *tqpair, 1489 struct nvme_tcp_pdu *pdu) 1490 { 1491 struct spdk_nvmf_tcp_req *tcp_req; 1492 1493 assert(pdu->psh_valid_bytes == pdu->psh_len); 1494 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1495 1496 tcp_req = nvmf_tcp_req_get(tqpair); 1497 if (!tcp_req) { 1498 /* Directly return and make the allocation retry again. This can happen if we're 1499 * using asynchronous writes to send the response to the host or when releasing 1500 * zero-copy buffers after a response has been sent. In both cases, the host might 1501 * receive the response before we've finished processing the request and is free to 1502 * send another one. 1503 */ 1504 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1505 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1506 return; 1507 } 1508 1509 /* The host sent more commands than the maximum queue depth. */ 1510 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1511 nvmf_tcp_qpair_disconnect(tqpair); 1512 return; 1513 } 1514 1515 pdu->req = tcp_req; 1516 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1517 nvmf_tcp_req_process(ttransport, tcp_req); 1518 } 1519 1520 static void 1521 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1522 struct spdk_nvmf_tcp_qpair *tqpair, 1523 struct nvme_tcp_pdu *pdu) 1524 { 1525 struct spdk_nvmf_tcp_req *tcp_req; 1526 struct spdk_nvme_tcp_cmd *capsule_cmd; 1527 uint32_t error_offset = 0; 1528 enum spdk_nvme_tcp_term_req_fes fes; 1529 struct spdk_nvme_cpl *rsp; 1530 1531 capsule_cmd = &pdu->hdr.capsule_cmd; 1532 tcp_req = pdu->req; 1533 assert(tcp_req != NULL); 1534 1535 /* Zero-copy requests don't support ICD */ 1536 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1537 1538 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1539 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1540 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1541 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1542 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1543 goto err; 1544 } 1545 1546 rsp = &tcp_req->req.rsp->nvme_cpl; 1547 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1548 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1549 } else { 1550 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1551 } 1552 1553 nvmf_tcp_req_process(ttransport, tcp_req); 1554 1555 return; 1556 err: 1557 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1558 } 1559 1560 static void 1561 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1562 struct spdk_nvmf_tcp_qpair *tqpair, 1563 struct nvme_tcp_pdu *pdu) 1564 { 1565 struct spdk_nvmf_tcp_req *tcp_req; 1566 uint32_t error_offset = 0; 1567 enum spdk_nvme_tcp_term_req_fes fes = 0; 1568 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1569 1570 h2c_data = &pdu->hdr.h2c_data; 1571 1572 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1573 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1574 1575 if (h2c_data->ttag > tqpair->resource_count) { 1576 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1577 tqpair->resource_count); 1578 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1579 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1580 goto err; 1581 } 1582 1583 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1584 1585 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1586 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1587 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1588 tcp_req->state); 1589 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1590 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1591 goto err; 1592 } 1593 1594 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1595 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1596 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1597 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1598 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1599 goto err; 1600 } 1601 1602 if (tcp_req->h2c_offset != h2c_data->datao) { 1603 SPDK_DEBUGLOG(nvmf_tcp, 1604 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1605 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1606 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1607 goto err; 1608 } 1609 1610 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1611 SPDK_DEBUGLOG(nvmf_tcp, 1612 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1613 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1614 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1615 goto err; 1616 } 1617 1618 pdu->req = tcp_req; 1619 1620 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1621 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1622 } 1623 1624 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1625 h2c_data->datao, h2c_data->datal); 1626 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1627 return; 1628 1629 err: 1630 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1631 } 1632 1633 static void 1634 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1635 struct spdk_nvmf_tcp_qpair *tqpair) 1636 { 1637 struct nvme_tcp_pdu *rsp_pdu; 1638 struct spdk_nvme_tcp_rsp *capsule_resp; 1639 1640 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1641 1642 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1643 assert(rsp_pdu != NULL); 1644 1645 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1646 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1647 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1648 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1649 if (tqpair->host_hdgst_enable) { 1650 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1651 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1652 } 1653 1654 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1655 } 1656 1657 static void 1658 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1659 { 1660 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1661 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1662 struct spdk_nvmf_tcp_qpair, qpair); 1663 1664 assert(tqpair != NULL); 1665 1666 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1667 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1668 tcp_req->req.length); 1669 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1670 return; 1671 } 1672 1673 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1674 nvmf_tcp_request_free(tcp_req); 1675 } else { 1676 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1677 } 1678 } 1679 1680 static void 1681 nvmf_tcp_r2t_complete(void *cb_arg) 1682 { 1683 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1684 struct spdk_nvmf_tcp_transport *ttransport; 1685 1686 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1687 struct spdk_nvmf_tcp_transport, transport); 1688 1689 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1690 1691 if (tcp_req->h2c_offset == tcp_req->req.length) { 1692 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1693 nvmf_tcp_req_process(ttransport, tcp_req); 1694 } 1695 } 1696 1697 static void 1698 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1699 struct spdk_nvmf_tcp_req *tcp_req) 1700 { 1701 struct nvme_tcp_pdu *rsp_pdu; 1702 struct spdk_nvme_tcp_r2t_hdr *r2t; 1703 1704 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1705 assert(rsp_pdu != NULL); 1706 1707 r2t = &rsp_pdu->hdr.r2t; 1708 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1709 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1710 1711 if (tqpair->host_hdgst_enable) { 1712 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1713 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1714 } 1715 1716 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1717 r2t->ttag = tcp_req->ttag; 1718 r2t->r2to = tcp_req->h2c_offset; 1719 r2t->r2tl = tcp_req->req.length; 1720 1721 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1722 1723 SPDK_DEBUGLOG(nvmf_tcp, 1724 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1725 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1726 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1727 } 1728 1729 static void 1730 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1731 struct spdk_nvmf_tcp_qpair *tqpair, 1732 struct nvme_tcp_pdu *pdu) 1733 { 1734 struct spdk_nvmf_tcp_req *tcp_req; 1735 struct spdk_nvme_cpl *rsp; 1736 1737 tcp_req = pdu->req; 1738 assert(tcp_req != NULL); 1739 1740 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1741 1742 tcp_req->h2c_offset += pdu->data_len; 1743 1744 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1745 * acknowledged before moving on. */ 1746 if (tcp_req->h2c_offset == tcp_req->req.length && 1747 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1748 /* After receiving all the h2c data, we need to check whether there is 1749 * transient transport error */ 1750 rsp = &tcp_req->req.rsp->nvme_cpl; 1751 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1752 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1753 } else { 1754 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1755 } 1756 nvmf_tcp_req_process(ttransport, tcp_req); 1757 } 1758 } 1759 1760 static void 1761 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1762 { 1763 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1764 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1765 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1766 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1767 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1768 DGET32(h2c_term_req->fei)); 1769 } 1770 } 1771 1772 static void 1773 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1774 struct nvme_tcp_pdu *pdu) 1775 { 1776 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1777 uint32_t error_offset = 0; 1778 enum spdk_nvme_tcp_term_req_fes fes; 1779 1780 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1781 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1782 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1783 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1784 goto end; 1785 } 1786 1787 /* set the data buffer */ 1788 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1789 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1790 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1791 return; 1792 end: 1793 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1794 } 1795 1796 static void 1797 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1798 struct nvme_tcp_pdu *pdu) 1799 { 1800 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1801 1802 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1803 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1804 } 1805 1806 static void 1807 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 1808 { 1809 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1810 struct spdk_nvmf_tcp_transport, transport); 1811 1812 switch (pdu->hdr.common.pdu_type) { 1813 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1814 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1815 break; 1816 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1817 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1818 break; 1819 1820 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1821 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1822 break; 1823 1824 default: 1825 /* The code should not go to here */ 1826 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 1827 break; 1828 } 1829 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1830 } 1831 1832 static void 1833 data_crc32_calc_done(void *cb_arg, int status) 1834 { 1835 struct nvme_tcp_pdu *pdu = cb_arg; 1836 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1837 struct spdk_nvmf_tcp_req *tcp_req; 1838 struct spdk_nvme_cpl *rsp; 1839 1840 /* async crc32 calculation is failed and use direct calculation to check */ 1841 if (spdk_unlikely(status)) { 1842 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 1843 tqpair, pdu); 1844 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1845 } 1846 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1847 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 1848 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1849 tcp_req = pdu->req; 1850 assert(tcp_req != NULL); 1851 rsp = &tcp_req->req.rsp->nvme_cpl; 1852 rsp->status.sc = SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR; 1853 } 1854 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 1855 } 1856 1857 static void 1858 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 1859 { 1860 int rc = 0; 1861 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1862 tqpair->pdu_in_progress = NULL; 1863 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1864 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1865 /* check data digest if need */ 1866 if (pdu->ddgst_enable) { 1867 if (!pdu->dif_ctx && tqpair->group && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 1868 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1869 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 1870 if (spdk_likely(rc == 0)) { 1871 return; 1872 } 1873 } else { 1874 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1875 } 1876 data_crc32_calc_done(pdu, rc); 1877 } else { 1878 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 1879 } 1880 } 1881 1882 static void 1883 nvmf_tcp_send_icresp_complete(void *cb_arg) 1884 { 1885 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1886 1887 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 1888 } 1889 1890 static void 1891 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1892 struct spdk_nvmf_tcp_qpair *tqpair, 1893 struct nvme_tcp_pdu *pdu) 1894 { 1895 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1896 struct nvme_tcp_pdu *rsp_pdu; 1897 struct spdk_nvme_tcp_ic_resp *ic_resp; 1898 uint32_t error_offset = 0; 1899 enum spdk_nvme_tcp_term_req_fes fes; 1900 1901 /* Only PFV 0 is defined currently */ 1902 if (ic_req->pfv != 0) { 1903 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1904 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1905 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1906 goto end; 1907 } 1908 1909 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 1910 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 1911 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 1912 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1913 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 1914 goto end; 1915 } 1916 1917 /* MAXR2T is 0's based */ 1918 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1919 1920 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1921 if (!tqpair->host_hdgst_enable) { 1922 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1923 } 1924 1925 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1926 if (!tqpair->host_ddgst_enable) { 1927 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1928 } 1929 1930 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1931 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1932 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1933 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1934 tqpair, 1935 tqpair->recv_buf_size); 1936 /* Not fatal. */ 1937 } 1938 1939 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1940 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1941 1942 rsp_pdu = tqpair->mgmt_pdu; 1943 1944 ic_resp = &rsp_pdu->hdr.ic_resp; 1945 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1946 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1947 ic_resp->pfv = 0; 1948 ic_resp->cpda = tqpair->cpda; 1949 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1950 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1951 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1952 1953 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1954 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1955 1956 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 1957 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 1958 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1959 return; 1960 end: 1961 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1962 } 1963 1964 static void 1965 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1966 struct spdk_nvmf_tcp_transport *ttransport) 1967 { 1968 struct nvme_tcp_pdu *pdu; 1969 int rc; 1970 uint32_t crc32c, error_offset = 0; 1971 enum spdk_nvme_tcp_term_req_fes fes; 1972 1973 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1974 pdu = tqpair->pdu_in_progress; 1975 1976 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1977 pdu->hdr.common.pdu_type); 1978 /* check header digest if needed */ 1979 if (pdu->has_hdgst) { 1980 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1981 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1982 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1983 if (rc == 0) { 1984 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1985 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1986 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1987 return; 1988 1989 } 1990 } 1991 1992 switch (pdu->hdr.common.pdu_type) { 1993 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1994 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1995 break; 1996 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1997 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1998 break; 1999 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2000 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2001 break; 2002 2003 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2004 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2005 break; 2006 2007 default: 2008 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2009 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2010 error_offset = 1; 2011 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2012 break; 2013 } 2014 } 2015 2016 static void 2017 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2018 { 2019 struct nvme_tcp_pdu *pdu; 2020 uint32_t error_offset = 0; 2021 enum spdk_nvme_tcp_term_req_fes fes; 2022 uint8_t expected_hlen, pdo; 2023 bool plen_error = false, pdo_error = false; 2024 2025 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2026 pdu = tqpair->pdu_in_progress; 2027 assert(pdu); 2028 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2029 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2030 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2031 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2032 goto err; 2033 } 2034 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2035 if (pdu->hdr.common.plen != expected_hlen) { 2036 plen_error = true; 2037 } 2038 } else { 2039 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2040 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2041 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2042 goto err; 2043 } 2044 2045 switch (pdu->hdr.common.pdu_type) { 2046 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2047 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2048 pdo = pdu->hdr.common.pdo; 2049 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2050 pdo_error = true; 2051 break; 2052 } 2053 2054 if (pdu->hdr.common.plen < expected_hlen) { 2055 plen_error = true; 2056 } 2057 break; 2058 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2059 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2060 pdo = pdu->hdr.common.pdo; 2061 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2062 pdo_error = true; 2063 break; 2064 } 2065 if (pdu->hdr.common.plen < expected_hlen) { 2066 plen_error = true; 2067 } 2068 break; 2069 2070 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2071 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2072 if ((pdu->hdr.common.plen <= expected_hlen) || 2073 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2074 plen_error = true; 2075 } 2076 break; 2077 2078 default: 2079 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2080 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2081 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2082 goto err; 2083 } 2084 } 2085 2086 if (pdu->hdr.common.hlen != expected_hlen) { 2087 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2088 pdu->hdr.common.pdu_type, 2089 expected_hlen, pdu->hdr.common.hlen, tqpair); 2090 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2091 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2092 goto err; 2093 } else if (pdo_error) { 2094 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2095 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2096 } else if (plen_error) { 2097 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2098 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2099 goto err; 2100 } else { 2101 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2102 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2103 return; 2104 } 2105 err: 2106 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2107 } 2108 2109 static int 2110 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2111 { 2112 int rc = 0; 2113 struct nvme_tcp_pdu *pdu; 2114 enum nvme_tcp_pdu_recv_state prev_state; 2115 uint32_t data_len; 2116 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2117 struct spdk_nvmf_tcp_transport, transport); 2118 2119 /* The loop here is to allow for several back-to-back state changes. */ 2120 do { 2121 prev_state = tqpair->recv_state; 2122 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2123 2124 pdu = tqpair->pdu_in_progress; 2125 assert(pdu || tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2126 switch (tqpair->recv_state) { 2127 /* Wait for the common header */ 2128 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2129 if (!pdu) { 2130 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2131 if (spdk_unlikely(!pdu)) { 2132 return NVME_TCP_PDU_IN_PROGRESS; 2133 } 2134 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2135 tqpair->pdu_in_progress = pdu; 2136 } 2137 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2138 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2139 /* FALLTHROUGH */ 2140 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2141 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2142 return rc; 2143 } 2144 2145 rc = nvme_tcp_read_data(tqpair->sock, 2146 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2147 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2148 if (rc < 0) { 2149 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2150 return NVME_TCP_PDU_FATAL; 2151 } else if (rc > 0) { 2152 pdu->ch_valid_bytes += rc; 2153 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2154 } 2155 2156 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2157 return NVME_TCP_PDU_IN_PROGRESS; 2158 } 2159 2160 /* The command header of this PDU has now been read from the socket. */ 2161 nvmf_tcp_pdu_ch_handle(tqpair); 2162 break; 2163 /* Wait for the pdu specific header */ 2164 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2165 rc = nvme_tcp_read_data(tqpair->sock, 2166 pdu->psh_len - pdu->psh_valid_bytes, 2167 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2168 if (rc < 0) { 2169 return NVME_TCP_PDU_FATAL; 2170 } else if (rc > 0) { 2171 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2172 pdu->psh_valid_bytes += rc; 2173 } 2174 2175 if (pdu->psh_valid_bytes < pdu->psh_len) { 2176 return NVME_TCP_PDU_IN_PROGRESS; 2177 } 2178 2179 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2180 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2181 break; 2182 /* Wait for the req slot */ 2183 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2184 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2185 break; 2186 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2187 /* check whether the data is valid, if not we just return */ 2188 if (!pdu->data_len) { 2189 return NVME_TCP_PDU_IN_PROGRESS; 2190 } 2191 2192 data_len = pdu->data_len; 2193 /* data digest */ 2194 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2195 tqpair->host_ddgst_enable)) { 2196 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2197 pdu->ddgst_enable = true; 2198 } 2199 2200 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2201 if (rc < 0) { 2202 return NVME_TCP_PDU_FATAL; 2203 } 2204 pdu->rw_offset += rc; 2205 2206 if (pdu->rw_offset < data_len) { 2207 return NVME_TCP_PDU_IN_PROGRESS; 2208 } 2209 2210 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2211 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2212 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2213 pdu->dif_ctx) != 0) { 2214 SPDK_ERRLOG("DIF generate failed\n"); 2215 return NVME_TCP_PDU_FATAL; 2216 } 2217 2218 /* All of this PDU has now been read from the socket. */ 2219 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2220 break; 2221 case NVME_TCP_PDU_RECV_STATE_ERROR: 2222 if (!spdk_sock_is_connected(tqpair->sock)) { 2223 return NVME_TCP_PDU_FATAL; 2224 } 2225 break; 2226 default: 2227 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2228 abort(); 2229 break; 2230 } 2231 } while (tqpair->recv_state != prev_state); 2232 2233 return rc; 2234 } 2235 2236 static inline void * 2237 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2238 { 2239 struct spdk_nvmf_tcp_control_msg *msg; 2240 2241 assert(list); 2242 2243 msg = STAILQ_FIRST(&list->free_msgs); 2244 if (!msg) { 2245 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2246 return NULL; 2247 } 2248 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2249 return msg; 2250 } 2251 2252 static inline void 2253 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2254 { 2255 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2256 2257 assert(list); 2258 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2259 } 2260 2261 static int 2262 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2263 struct spdk_nvmf_transport *transport, 2264 struct spdk_nvmf_transport_poll_group *group) 2265 { 2266 struct spdk_nvmf_request *req = &tcp_req->req; 2267 struct spdk_nvme_cmd *cmd; 2268 struct spdk_nvme_cpl *rsp; 2269 struct spdk_nvme_sgl_descriptor *sgl; 2270 struct spdk_nvmf_tcp_poll_group *tgroup; 2271 uint32_t length; 2272 2273 cmd = &req->cmd->nvme_cmd; 2274 rsp = &req->rsp->nvme_cpl; 2275 sgl = &cmd->dptr.sgl1; 2276 2277 length = sgl->unkeyed.length; 2278 2279 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2280 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2281 if (length > transport->opts.max_io_size) { 2282 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2283 length, transport->opts.max_io_size); 2284 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2285 return -1; 2286 } 2287 2288 /* fill request length and populate iovs */ 2289 req->length = length; 2290 2291 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2292 2293 if (spdk_unlikely(req->dif_enabled)) { 2294 req->dif.orig_length = length; 2295 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2296 req->dif.elba_length = length; 2297 } 2298 2299 if (nvmf_ctrlr_use_zcopy(req)) { 2300 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2301 req->data_from_pool = false; 2302 return 0; 2303 } 2304 2305 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2306 /* No available buffers. Queue this request up. */ 2307 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2308 tcp_req); 2309 return 0; 2310 } 2311 2312 /* backward compatible */ 2313 req->data = req->iov[0].iov_base; 2314 2315 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2316 tcp_req, req->iovcnt, req->data); 2317 2318 return 0; 2319 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2320 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2321 uint64_t offset = sgl->address; 2322 uint32_t max_len = transport->opts.in_capsule_data_size; 2323 assert(tcp_req->has_in_capsule_data); 2324 2325 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2326 offset, length); 2327 2328 if (offset > max_len) { 2329 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2330 offset, max_len); 2331 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2332 return -1; 2333 } 2334 max_len -= (uint32_t)offset; 2335 2336 if (spdk_unlikely(length > max_len)) { 2337 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2338 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2339 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2340 2341 /* Get a buffer from dedicated list */ 2342 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2343 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2344 assert(tgroup->control_msg_list); 2345 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2346 if (!req->data) { 2347 /* No available buffers. Queue this request up. */ 2348 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2349 return 0; 2350 } 2351 } else { 2352 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2353 length, max_len); 2354 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2355 return -1; 2356 } 2357 } else { 2358 req->data = tcp_req->buf; 2359 } 2360 2361 req->length = length; 2362 req->data_from_pool = false; 2363 2364 if (spdk_unlikely(req->dif_enabled)) { 2365 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2366 req->dif.elba_length = length; 2367 } 2368 2369 req->iov[0].iov_base = req->data; 2370 req->iov[0].iov_len = length; 2371 req->iovcnt = 1; 2372 2373 return 0; 2374 } 2375 2376 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2377 sgl->generic.type, sgl->generic.subtype); 2378 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2379 return -1; 2380 } 2381 2382 static inline enum spdk_nvme_media_error_status_code 2383 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2384 enum spdk_nvme_media_error_status_code result; 2385 2386 switch (err_type) 2387 { 2388 case SPDK_DIF_REFTAG_ERROR: 2389 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2390 break; 2391 case SPDK_DIF_APPTAG_ERROR: 2392 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2393 break; 2394 case SPDK_DIF_GUARD_ERROR: 2395 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2396 break; 2397 default: 2398 SPDK_UNREACHABLE(); 2399 break; 2400 } 2401 2402 return result; 2403 } 2404 2405 static void 2406 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2407 struct spdk_nvmf_tcp_req *tcp_req) 2408 { 2409 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2410 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2411 struct nvme_tcp_pdu *rsp_pdu; 2412 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2413 uint32_t plen, pdo, alignment; 2414 int rc; 2415 2416 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2417 2418 rsp_pdu = tcp_req->pdu; 2419 assert(rsp_pdu != NULL); 2420 2421 c2h_data = &rsp_pdu->hdr.c2h_data; 2422 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2423 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2424 2425 if (tqpair->host_hdgst_enable) { 2426 plen += SPDK_NVME_TCP_DIGEST_LEN; 2427 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2428 } 2429 2430 /* set the psh */ 2431 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2432 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2433 c2h_data->datao = tcp_req->pdu->rw_offset; 2434 2435 /* set the padding */ 2436 rsp_pdu->padding_len = 0; 2437 pdo = plen; 2438 if (tqpair->cpda) { 2439 alignment = (tqpair->cpda + 1) << 2; 2440 if (plen % alignment != 0) { 2441 pdo = (plen + alignment) / alignment * alignment; 2442 rsp_pdu->padding_len = pdo - plen; 2443 plen = pdo; 2444 } 2445 } 2446 2447 c2h_data->common.pdo = pdo; 2448 plen += c2h_data->datal; 2449 if (tqpair->host_ddgst_enable) { 2450 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2451 plen += SPDK_NVME_TCP_DIGEST_LEN; 2452 } 2453 2454 c2h_data->common.plen = plen; 2455 2456 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2457 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2458 } 2459 2460 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2461 c2h_data->datao, c2h_data->datal); 2462 2463 2464 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2465 /* Need to send the capsule response if response is not all 0 */ 2466 if (ttransport->tcp_opts.c2h_success && 2467 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2468 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2469 } 2470 2471 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2472 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2473 struct spdk_dif_error err_blk = {}; 2474 uint32_t mapped_length = 0; 2475 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2476 uint32_t ddgst_len = 0; 2477 2478 if (tqpair->host_ddgst_enable) { 2479 /* Data digest consumes additional iov entry */ 2480 available_iovs--; 2481 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2482 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2483 c2h_data->common.plen -= ddgst_len; 2484 } 2485 /* Temp call to estimate if data can be described by limited number of iovs. 2486 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2487 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2488 false, &mapped_length); 2489 2490 if (mapped_length != c2h_data->common.plen) { 2491 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2492 SPDK_DEBUGLOG(nvmf_tcp, 2493 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2494 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2495 c2h_data->common.plen = mapped_length; 2496 2497 /* Rebuild pdu->data_iov since data length is changed */ 2498 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2499 c2h_data->datal); 2500 2501 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2502 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2503 } 2504 2505 c2h_data->common.plen += ddgst_len; 2506 2507 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2508 2509 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2510 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2511 if (rc != 0) { 2512 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2513 err_blk.err_type, err_blk.err_offset); 2514 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2515 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2516 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2517 return; 2518 } 2519 } 2520 2521 rsp_pdu->rw_offset += c2h_data->datal; 2522 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2523 } 2524 2525 static void 2526 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2527 struct spdk_nvmf_tcp_req *tcp_req) 2528 { 2529 nvmf_tcp_req_pdu_init(tcp_req); 2530 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2531 } 2532 2533 static int 2534 request_transfer_out(struct spdk_nvmf_request *req) 2535 { 2536 struct spdk_nvmf_tcp_req *tcp_req; 2537 struct spdk_nvmf_qpair *qpair; 2538 struct spdk_nvmf_tcp_qpair *tqpair; 2539 struct spdk_nvme_cpl *rsp; 2540 2541 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2542 2543 qpair = req->qpair; 2544 rsp = &req->rsp->nvme_cpl; 2545 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2546 2547 /* Advance our sq_head pointer */ 2548 if (qpair->sq_head == qpair->sq_head_max) { 2549 qpair->sq_head = 0; 2550 } else { 2551 qpair->sq_head++; 2552 } 2553 rsp->sqhd = qpair->sq_head; 2554 2555 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2556 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2557 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2558 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2559 } else { 2560 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2561 } 2562 2563 return 0; 2564 } 2565 2566 static void 2567 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2568 struct spdk_nvmf_tcp_qpair *tqpair, 2569 struct spdk_nvmf_tcp_req *tcp_req) 2570 { 2571 enum spdk_nvme_cmd_fuse last, next; 2572 2573 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2574 next = tcp_req->cmd.fuse; 2575 2576 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2577 2578 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2579 return; 2580 } 2581 2582 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2583 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2584 /* This is a valid pair of fused commands. Point them at each other 2585 * so they can be submitted consecutively once ready to be executed. 2586 */ 2587 tqpair->fused_first->fused_pair = tcp_req; 2588 tcp_req->fused_pair = tqpair->fused_first; 2589 tqpair->fused_first = NULL; 2590 return; 2591 } else { 2592 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2593 tqpair->fused_first->fused_failed = true; 2594 2595 /* 2596 * If the last req is in READY_TO_EXECUTE state, then call 2597 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2598 */ 2599 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2600 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2601 } 2602 2603 tqpair->fused_first = NULL; 2604 } 2605 } 2606 2607 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2608 /* Set tqpair->fused_first here so that we know to check that the next request 2609 * is a SECOND (and to fail this one if it isn't). 2610 */ 2611 tqpair->fused_first = tcp_req; 2612 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2613 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2614 tcp_req->fused_failed = true; 2615 } 2616 } 2617 2618 static bool 2619 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2620 struct spdk_nvmf_tcp_req *tcp_req) 2621 { 2622 struct spdk_nvmf_tcp_qpair *tqpair; 2623 int rc; 2624 uint32_t plen; 2625 struct nvme_tcp_pdu *pdu; 2626 enum spdk_nvmf_tcp_req_state prev_state; 2627 bool progress = false; 2628 struct spdk_nvmf_transport *transport = &ttransport->transport; 2629 struct spdk_nvmf_transport_poll_group *group; 2630 struct spdk_nvmf_tcp_poll_group *tgroup; 2631 2632 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2633 group = &tqpair->group->group; 2634 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2635 2636 /* If the qpair is not active, we need to abort the outstanding requests. */ 2637 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2638 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2639 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2640 } 2641 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2642 } 2643 2644 /* The loop here is to allow for several back-to-back state changes. */ 2645 do { 2646 prev_state = tcp_req->state; 2647 2648 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2649 tqpair); 2650 2651 switch (tcp_req->state) { 2652 case TCP_REQUEST_STATE_FREE: 2653 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2654 * to escape this state. */ 2655 break; 2656 case TCP_REQUEST_STATE_NEW: 2657 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, tqpair); 2658 2659 /* copy the cmd from the receive pdu */ 2660 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2661 2662 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2663 tcp_req->req.dif_enabled = true; 2664 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2665 } 2666 2667 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2668 2669 /* The next state transition depends on the data transfer needs of this request. */ 2670 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2671 2672 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2673 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2674 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 2675 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2676 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2677 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2678 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2679 break; 2680 } 2681 2682 /* If no data to transfer, ready to execute. */ 2683 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2684 /* Reset the tqpair receiving pdu state */ 2685 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2686 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2687 break; 2688 } 2689 2690 pdu = tqpair->pdu_in_progress; 2691 plen = pdu->hdr.common.hlen; 2692 if (tqpair->host_hdgst_enable) { 2693 plen += SPDK_NVME_TCP_DIGEST_LEN; 2694 } 2695 if (pdu->hdr.common.plen != plen) { 2696 tcp_req->has_in_capsule_data = true; 2697 } else { 2698 /* Data is transmitted by C2H PDUs */ 2699 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2700 } 2701 2702 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2703 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2704 break; 2705 case TCP_REQUEST_STATE_NEED_BUFFER: 2706 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2707 tqpair); 2708 2709 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2710 2711 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2712 SPDK_DEBUGLOG(nvmf_tcp, 2713 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2714 tcp_req, tqpair); 2715 /* This request needs to wait in line to obtain a buffer */ 2716 break; 2717 } 2718 2719 /* Try to get a data buffer */ 2720 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2721 if (rc < 0) { 2722 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2723 /* Reset the tqpair receiving pdu state */ 2724 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2725 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2726 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2727 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2728 break; 2729 } 2730 2731 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2732 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2733 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2734 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2735 tcp_req->req.length = tcp_req->req.dif.elba_length; 2736 } 2737 2738 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2739 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2740 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2741 break; 2742 } 2743 2744 if (!tcp_req->req.data) { 2745 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2746 tcp_req, tqpair); 2747 /* No buffers available. */ 2748 break; 2749 } 2750 2751 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2752 2753 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2754 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2755 if (tcp_req->req.data_from_pool) { 2756 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2757 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2758 } else { 2759 struct nvme_tcp_pdu *pdu; 2760 2761 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2762 2763 pdu = tqpair->pdu_in_progress; 2764 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2765 tqpair); 2766 /* No need to send r2t, contained in the capsuled data */ 2767 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2768 0, tcp_req->req.length); 2769 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2770 } 2771 break; 2772 } 2773 2774 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2775 break; 2776 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 2777 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.qid, 0, 2778 (uintptr_t)tcp_req, tqpair); 2779 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 2780 * to escape this state. */ 2781 break; 2782 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 2783 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.qid, 0, 2784 (uintptr_t)tcp_req, tqpair); 2785 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 2786 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 2787 tcp_req, tqpair); 2788 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2789 break; 2790 } 2791 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2792 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2793 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2794 } else { 2795 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2796 } 2797 break; 2798 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2799 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2800 tqpair); 2801 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2802 break; 2803 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2804 2805 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.qid, 0, 2806 (uintptr_t)tcp_req, tqpair); 2807 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2808 * to escape this state. */ 2809 break; 2810 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2811 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.qid, 0, 2812 (uintptr_t)tcp_req, tqpair); 2813 2814 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2815 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2816 tcp_req->req.length = tcp_req->req.dif.elba_length; 2817 } 2818 2819 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 2820 if (tcp_req->fused_failed) { 2821 /* This request failed FUSED semantics. Fail it immediately, without 2822 * even sending it to the target layer. 2823 */ 2824 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2825 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 2826 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2827 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2828 break; 2829 } 2830 2831 if (tcp_req->fused_pair == NULL || 2832 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2833 /* This request is ready to execute, but either we don't know yet if it's 2834 * valid - i.e. this is a FIRST but we haven't received the next request yet), 2835 * or the other request of this fused pair isn't ready to execute. So 2836 * break here and this request will get processed later either when the 2837 * other request is ready or we find that this request isn't valid. 2838 */ 2839 break; 2840 } 2841 } 2842 2843 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2844 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2845 /* If we get to this point, and this request is a fused command, we know that 2846 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 2847 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 2848 * request, and the other request of the fused pair, in the correct order. 2849 * Also clear the ->fused_pair pointers on both requests, since after this point 2850 * we no longer need to maintain the relationship between these two requests. 2851 */ 2852 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 2853 assert(tcp_req->fused_pair != NULL); 2854 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2855 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2856 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2857 tcp_req->fused_pair->fused_pair = NULL; 2858 tcp_req->fused_pair = NULL; 2859 } 2860 spdk_nvmf_request_exec(&tcp_req->req); 2861 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 2862 assert(tcp_req->fused_pair != NULL); 2863 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2864 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2865 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2866 tcp_req->fused_pair->fused_pair = NULL; 2867 tcp_req->fused_pair = NULL; 2868 } 2869 } else { 2870 /* For zero-copy, only requests with data coming from host to the 2871 * controller can end up here. */ 2872 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 2873 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 2874 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 2875 } 2876 2877 break; 2878 case TCP_REQUEST_STATE_EXECUTING: 2879 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2880 tqpair); 2881 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2882 * to escape this state. */ 2883 break; 2884 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 2885 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.qid, 0, 2886 (uintptr_t)tcp_req, tqpair); 2887 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2888 * to escape this state. */ 2889 break; 2890 case TCP_REQUEST_STATE_EXECUTED: 2891 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2892 tqpair); 2893 2894 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2895 tcp_req->req.length = tcp_req->req.dif.orig_length; 2896 } 2897 2898 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2899 break; 2900 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2901 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.qid, 0, 2902 (uintptr_t)tcp_req, tqpair); 2903 rc = request_transfer_out(&tcp_req->req); 2904 assert(rc == 0); /* No good way to handle this currently */ 2905 break; 2906 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2907 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.qid, 0, 2908 (uintptr_t)tcp_req, tqpair); 2909 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2910 * to escape this state. */ 2911 break; 2912 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 2913 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.qid, 0, 2914 (uintptr_t)tcp_req, tqpair); 2915 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2916 * to escape this state. */ 2917 break; 2918 case TCP_REQUEST_STATE_COMPLETED: 2919 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2920 tqpair); 2921 /* If there's an outstanding PDU sent to the host, the request is completed 2922 * due to the qpair being disconnected. We must delay the completion until 2923 * that write is done to avoid freeing the request twice. */ 2924 if (spdk_unlikely(tcp_req->pdu_in_use)) { 2925 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 2926 "write on req=%p\n", tcp_req); 2927 /* This can only happen for zcopy requests */ 2928 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 2929 assert(tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 2930 break; 2931 } 2932 2933 if (tcp_req->req.data_from_pool) { 2934 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2935 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 2936 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2937 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2938 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2939 assert(tgroup->control_msg_list); 2940 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2941 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2942 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 2943 /* If the request has an unreleased zcopy bdev_io, it's either a 2944 * read, a failed write, or the qpair is being disconnected */ 2945 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 2946 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 2947 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 2948 tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 2949 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 2950 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 2951 break; 2952 } 2953 tcp_req->req.length = 0; 2954 tcp_req->req.iovcnt = 0; 2955 tcp_req->req.data = NULL; 2956 tcp_req->fused_failed = false; 2957 if (tcp_req->fused_pair) { 2958 /* This req was part of a valid fused pair, but failed before it got to 2959 * READ_TO_EXECUTE state. This means we need to fail the other request 2960 * in the pair, because it is no longer part of a valid pair. If the pair 2961 * already reached READY_TO_EXECUTE state, we need to kick it. 2962 */ 2963 tcp_req->fused_pair->fused_failed = true; 2964 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2965 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 2966 } 2967 tcp_req->fused_pair = NULL; 2968 } 2969 2970 nvmf_tcp_req_put(tqpair, tcp_req); 2971 break; 2972 case TCP_REQUEST_NUM_STATES: 2973 default: 2974 assert(0); 2975 break; 2976 } 2977 2978 if (tcp_req->state != prev_state) { 2979 progress = true; 2980 } 2981 } while (tcp_req->state != prev_state); 2982 2983 return progress; 2984 } 2985 2986 static void 2987 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2988 { 2989 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2990 int rc; 2991 2992 assert(tqpair != NULL); 2993 rc = nvmf_tcp_sock_process(tqpair); 2994 2995 /* If there was a new socket error, disconnect */ 2996 if (rc < 0) { 2997 nvmf_tcp_qpair_disconnect(tqpair); 2998 } 2999 } 3000 3001 static int 3002 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3003 struct spdk_nvmf_qpair *qpair) 3004 { 3005 struct spdk_nvmf_tcp_poll_group *tgroup; 3006 struct spdk_nvmf_tcp_qpair *tqpair; 3007 int rc; 3008 3009 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3010 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3011 3012 rc = nvmf_tcp_qpair_sock_init(tqpair); 3013 if (rc != 0) { 3014 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3015 return -1; 3016 } 3017 3018 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3019 if (rc < 0) { 3020 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3021 return -1; 3022 } 3023 3024 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3025 if (rc < 0) { 3026 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3027 return -1; 3028 } 3029 3030 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3031 nvmf_tcp_sock_cb, tqpair); 3032 if (rc != 0) { 3033 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3034 spdk_strerror(errno), errno); 3035 return -1; 3036 } 3037 3038 tqpair->group = tgroup; 3039 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3040 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3041 3042 return 0; 3043 } 3044 3045 static int 3046 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3047 struct spdk_nvmf_qpair *qpair) 3048 { 3049 struct spdk_nvmf_tcp_poll_group *tgroup; 3050 struct spdk_nvmf_tcp_qpair *tqpair; 3051 int rc; 3052 3053 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3054 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3055 3056 assert(tqpair->group == tgroup); 3057 3058 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3059 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3060 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 3061 } else { 3062 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3063 } 3064 3065 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3066 if (rc != 0) { 3067 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3068 spdk_strerror(errno), errno); 3069 } 3070 3071 return rc; 3072 } 3073 3074 static int 3075 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3076 { 3077 struct spdk_nvmf_tcp_transport *ttransport; 3078 struct spdk_nvmf_tcp_req *tcp_req; 3079 3080 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3081 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3082 3083 switch (tcp_req->state) { 3084 case TCP_REQUEST_STATE_EXECUTING: 3085 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3086 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3087 break; 3088 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3089 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3090 break; 3091 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3092 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3093 break; 3094 default: 3095 assert(0 && "Unexpected request state"); 3096 break; 3097 } 3098 3099 nvmf_tcp_req_process(ttransport, tcp_req); 3100 3101 return 0; 3102 } 3103 3104 static void 3105 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3106 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3107 { 3108 struct spdk_nvmf_tcp_qpair *tqpair; 3109 3110 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3111 3112 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3113 3114 assert(tqpair->fini_cb_fn == NULL); 3115 tqpair->fini_cb_fn = cb_fn; 3116 tqpair->fini_cb_arg = cb_arg; 3117 3118 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3119 nvmf_tcp_qpair_destroy(tqpair); 3120 } 3121 3122 static int 3123 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3124 { 3125 struct spdk_nvmf_tcp_poll_group *tgroup; 3126 int rc; 3127 struct spdk_nvmf_request *req, *req_tmp; 3128 struct spdk_nvmf_tcp_req *tcp_req; 3129 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3130 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3131 struct spdk_nvmf_tcp_transport, transport); 3132 3133 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3134 3135 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3136 return 0; 3137 } 3138 3139 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3140 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3141 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3142 break; 3143 } 3144 } 3145 3146 rc = spdk_sock_group_poll(tgroup->sock_group); 3147 if (rc < 0) { 3148 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3149 } 3150 3151 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3152 nvmf_tcp_sock_process(tqpair); 3153 } 3154 3155 return rc; 3156 } 3157 3158 static int 3159 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3160 struct spdk_nvme_transport_id *trid, bool peer) 3161 { 3162 struct spdk_nvmf_tcp_qpair *tqpair; 3163 uint16_t port; 3164 3165 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3166 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3167 3168 if (peer) { 3169 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3170 port = tqpair->initiator_port; 3171 } else { 3172 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3173 port = tqpair->target_port; 3174 } 3175 3176 if (spdk_sock_is_ipv4(tqpair->sock)) { 3177 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3178 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3179 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3180 } else { 3181 return -1; 3182 } 3183 3184 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3185 return 0; 3186 } 3187 3188 static int 3189 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3190 struct spdk_nvme_transport_id *trid) 3191 { 3192 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3193 } 3194 3195 static int 3196 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3197 struct spdk_nvme_transport_id *trid) 3198 { 3199 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3200 } 3201 3202 static int 3203 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3204 struct spdk_nvme_transport_id *trid) 3205 { 3206 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3207 } 3208 3209 static void 3210 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3211 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3212 { 3213 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3214 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3215 tcp_req_to_abort->req.rsp->nvme_cpl.cid = tcp_req_to_abort->req.cmd->nvme_cmd.cid; 3216 3217 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3218 3219 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3220 } 3221 3222 static int 3223 _nvmf_tcp_qpair_abort_request(void *ctx) 3224 { 3225 struct spdk_nvmf_request *req = ctx; 3226 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3227 struct spdk_nvmf_tcp_req, req); 3228 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3229 struct spdk_nvmf_tcp_qpair, qpair); 3230 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3231 struct spdk_nvmf_tcp_transport, transport); 3232 int rc; 3233 3234 spdk_poller_unregister(&req->poller); 3235 3236 switch (tcp_req_to_abort->state) { 3237 case TCP_REQUEST_STATE_EXECUTING: 3238 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3239 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3240 rc = nvmf_ctrlr_abort_request(req); 3241 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3242 return SPDK_POLLER_BUSY; 3243 } 3244 break; 3245 3246 case TCP_REQUEST_STATE_NEED_BUFFER: 3247 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3248 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3249 3250 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3251 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3252 break; 3253 3254 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3255 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3256 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3257 break; 3258 3259 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3260 if (spdk_get_ticks() < req->timeout_tsc) { 3261 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3262 return SPDK_POLLER_BUSY; 3263 } 3264 break; 3265 3266 default: 3267 break; 3268 } 3269 3270 spdk_nvmf_request_complete(req); 3271 return SPDK_POLLER_BUSY; 3272 } 3273 3274 static void 3275 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3276 struct spdk_nvmf_request *req) 3277 { 3278 struct spdk_nvmf_tcp_qpair *tqpair; 3279 struct spdk_nvmf_tcp_transport *ttransport; 3280 struct spdk_nvmf_transport *transport; 3281 uint16_t cid; 3282 uint32_t i; 3283 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3284 3285 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3286 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3287 transport = &ttransport->transport; 3288 3289 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3290 3291 for (i = 0; i < tqpair->resource_count; i++) { 3292 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3293 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3294 tcp_req_to_abort = &tqpair->reqs[i]; 3295 break; 3296 } 3297 } 3298 3299 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, qpair->qid, 0, (uintptr_t)req, tqpair); 3300 3301 if (tcp_req_to_abort == NULL) { 3302 spdk_nvmf_request_complete(req); 3303 return; 3304 } 3305 3306 req->req_to_abort = &tcp_req_to_abort->req; 3307 req->timeout_tsc = spdk_get_ticks() + 3308 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3309 req->poller = NULL; 3310 3311 _nvmf_tcp_qpair_abort_request(req); 3312 } 3313 3314 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 3315 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 3316 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 3317 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 3318 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 3319 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 3320 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 3321 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 3322 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 3323 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 3324 3325 static void 3326 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3327 { 3328 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 3329 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3330 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3331 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3332 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3333 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 3334 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3335 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3336 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3337 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3338 opts->transport_specific = NULL; 3339 } 3340 3341 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3342 .name = "TCP", 3343 .type = SPDK_NVME_TRANSPORT_TCP, 3344 .opts_init = nvmf_tcp_opts_init, 3345 .create = nvmf_tcp_create, 3346 .dump_opts = nvmf_tcp_dump_opts, 3347 .destroy = nvmf_tcp_destroy, 3348 3349 .listen = nvmf_tcp_listen, 3350 .stop_listen = nvmf_tcp_stop_listen, 3351 3352 .listener_discover = nvmf_tcp_discover, 3353 3354 .poll_group_create = nvmf_tcp_poll_group_create, 3355 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3356 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3357 .poll_group_add = nvmf_tcp_poll_group_add, 3358 .poll_group_remove = nvmf_tcp_poll_group_remove, 3359 .poll_group_poll = nvmf_tcp_poll_group_poll, 3360 3361 .req_free = nvmf_tcp_req_free, 3362 .req_complete = nvmf_tcp_req_complete, 3363 3364 .qpair_fini = nvmf_tcp_close_qpair, 3365 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3366 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3367 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3368 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3369 }; 3370 3371 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3372 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3373