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