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 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 315 void *fini_cb_arg; 316 317 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 318 }; 319 320 struct spdk_nvmf_tcp_control_msg { 321 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 322 }; 323 324 struct spdk_nvmf_tcp_control_msg_list { 325 void *msg_buf; 326 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 327 }; 328 329 struct spdk_nvmf_tcp_poll_group { 330 struct spdk_nvmf_transport_poll_group group; 331 struct spdk_sock_group *sock_group; 332 333 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 334 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 335 336 struct spdk_io_channel *accel_channel; 337 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 338 339 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 340 }; 341 342 struct spdk_nvmf_tcp_port { 343 const struct spdk_nvme_transport_id *trid; 344 struct spdk_sock *listen_sock; 345 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 346 }; 347 348 struct tcp_transport_opts { 349 bool c2h_success; 350 uint16_t control_msg_num; 351 uint32_t sock_priority; 352 }; 353 354 struct spdk_nvmf_tcp_transport { 355 struct spdk_nvmf_transport transport; 356 struct tcp_transport_opts tcp_opts; 357 358 struct spdk_nvmf_tcp_poll_group *next_pg; 359 360 struct spdk_poller *accept_poller; 361 pthread_mutex_t lock; 362 363 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 364 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 365 }; 366 367 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 368 { 369 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 370 spdk_json_decode_bool, true 371 }, 372 { 373 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 374 spdk_json_decode_uint16, true 375 }, 376 { 377 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 378 spdk_json_decode_uint32, true 379 }, 380 }; 381 382 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 383 struct spdk_nvmf_tcp_req *tcp_req); 384 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 385 386 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 387 struct spdk_nvmf_tcp_req *tcp_req); 388 389 static void 390 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 391 enum spdk_nvmf_tcp_req_state state) 392 { 393 struct spdk_nvmf_qpair *qpair; 394 struct spdk_nvmf_tcp_qpair *tqpair; 395 396 qpair = tcp_req->req.qpair; 397 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 398 399 assert(tqpair->state_cntr[tcp_req->state] > 0); 400 tqpair->state_cntr[tcp_req->state]--; 401 tqpair->state_cntr[state]++; 402 403 tcp_req->state = state; 404 } 405 406 static inline struct nvme_tcp_pdu * 407 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 408 { 409 assert(tcp_req->pdu_in_use == false); 410 411 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 412 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 413 414 return tcp_req->pdu; 415 } 416 417 static struct spdk_nvmf_tcp_req * 418 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 419 { 420 struct spdk_nvmf_tcp_req *tcp_req; 421 422 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 423 if (!tcp_req) { 424 return NULL; 425 } 426 427 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 428 tcp_req->h2c_offset = 0; 429 tcp_req->has_in_capsule_data = false; 430 tcp_req->req.dif_enabled = false; 431 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 432 433 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 434 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 435 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 436 return tcp_req; 437 } 438 439 static inline void 440 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 441 { 442 assert(!tcp_req->pdu_in_use); 443 444 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 445 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 446 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 447 } 448 449 static void 450 nvmf_tcp_request_free(void *cb_arg) 451 { 452 struct spdk_nvmf_tcp_transport *ttransport; 453 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 454 455 assert(tcp_req != NULL); 456 457 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 458 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 459 struct spdk_nvmf_tcp_transport, transport); 460 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 461 nvmf_tcp_req_process(ttransport, tcp_req); 462 } 463 464 static int 465 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 466 { 467 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 468 469 nvmf_tcp_request_free(tcp_req); 470 471 return 0; 472 } 473 474 static void 475 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 476 enum spdk_nvmf_tcp_req_state state) 477 { 478 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 479 480 assert(state != TCP_REQUEST_STATE_FREE); 481 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 482 if (state == tcp_req->state) { 483 nvmf_tcp_request_free(tcp_req); 484 } 485 } 486 } 487 488 static void 489 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 490 { 491 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 492 493 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 494 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 495 496 /* Wipe the requests waiting for buffer from the global list */ 497 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 498 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 499 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 500 spdk_nvmf_request, buf_link); 501 } 502 } 503 504 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 505 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 506 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 507 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 508 } 509 510 static void 511 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 512 { 513 int i; 514 struct spdk_nvmf_tcp_req *tcp_req; 515 516 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 517 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 518 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 519 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 520 if ((int)tcp_req->state == i) { 521 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 522 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 523 } 524 } 525 } 526 } 527 528 static void 529 _nvmf_tcp_qpair_destroy(void *_tqpair) 530 { 531 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 532 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 533 void *cb_arg = tqpair->fini_cb_arg; 534 int err = 0; 535 536 spdk_trace_record(TRACE_TCP_QP_DESTROY, 0, 0, (uintptr_t)tqpair); 537 538 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 539 540 err = spdk_sock_close(&tqpair->sock); 541 assert(err == 0); 542 nvmf_tcp_cleanup_all_states(tqpair); 543 544 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 545 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 546 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 547 tqpair->resource_count); 548 err++; 549 } 550 551 if (err > 0) { 552 nvmf_tcp_dump_qpair_req_contents(tqpair); 553 } 554 555 spdk_dma_free(tqpair->pdus); 556 free(tqpair->reqs); 557 spdk_free(tqpair->bufs); 558 free(tqpair); 559 560 if (cb_fn != NULL) { 561 cb_fn(cb_arg); 562 } 563 564 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 565 } 566 567 static void 568 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 569 { 570 /* Delay the destruction to make sure it isn't performed from the context of a sock 571 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 572 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 573 */ 574 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 575 } 576 577 static void 578 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 579 { 580 struct spdk_nvmf_tcp_transport *ttransport; 581 assert(w != NULL); 582 583 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 584 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 585 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 586 } 587 588 static int 589 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 590 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 591 { 592 struct spdk_nvmf_tcp_transport *ttransport; 593 594 assert(transport != NULL); 595 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 596 597 spdk_poller_unregister(&ttransport->accept_poller); 598 pthread_mutex_destroy(&ttransport->lock); 599 free(ttransport); 600 601 if (cb_fn) { 602 cb_fn(cb_arg); 603 } 604 return 0; 605 } 606 607 static int 608 nvmf_tcp_accept(void *ctx); 609 610 static struct spdk_nvmf_transport * 611 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 612 { 613 struct spdk_nvmf_tcp_transport *ttransport; 614 uint32_t sge_count; 615 uint32_t min_shared_buffers; 616 617 ttransport = calloc(1, sizeof(*ttransport)); 618 if (!ttransport) { 619 return NULL; 620 } 621 622 TAILQ_INIT(&ttransport->ports); 623 TAILQ_INIT(&ttransport->poll_groups); 624 625 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 626 627 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 628 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 629 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 630 if (opts->transport_specific != NULL && 631 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 632 SPDK_COUNTOF(tcp_transport_opts_decoder), 633 &ttransport->tcp_opts)) { 634 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 635 free(ttransport); 636 return NULL; 637 } 638 639 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 640 641 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 642 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 643 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 644 " in_capsule_data_size=%d, max_aq_depth=%d\n" 645 " num_shared_buffers=%d, c2h_success=%d,\n" 646 " dif_insert_or_strip=%d, sock_priority=%d\n" 647 " abort_timeout_sec=%d, control_msg_num=%hu\n", 648 opts->max_queue_depth, 649 opts->max_io_size, 650 opts->max_qpairs_per_ctrlr - 1, 651 opts->io_unit_size, 652 opts->in_capsule_data_size, 653 opts->max_aq_depth, 654 opts->num_shared_buffers, 655 ttransport->tcp_opts.c2h_success, 656 opts->dif_insert_or_strip, 657 ttransport->tcp_opts.sock_priority, 658 opts->abort_timeout_sec, 659 ttransport->tcp_opts.control_msg_num); 660 661 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 662 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 663 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 664 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 665 free(ttransport); 666 return NULL; 667 } 668 669 if (ttransport->tcp_opts.control_msg_num == 0 && 670 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 671 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 672 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 673 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 674 } 675 676 /* I/O unit size cannot be larger than max I/O size */ 677 if (opts->io_unit_size > opts->max_io_size) { 678 opts->io_unit_size = opts->max_io_size; 679 } 680 681 sge_count = opts->max_io_size / opts->io_unit_size; 682 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 683 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 684 free(ttransport); 685 return NULL; 686 } 687 688 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 689 if (min_shared_buffers > opts->num_shared_buffers) { 690 SPDK_ERRLOG("There are not enough buffers to satisfy " 691 "per-poll group caches for each thread. (%" PRIu32 ") " 692 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 693 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 694 free(ttransport); 695 return NULL; 696 } 697 698 pthread_mutex_init(&ttransport->lock, NULL); 699 700 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 701 opts->acceptor_poll_rate); 702 if (!ttransport->accept_poller) { 703 pthread_mutex_destroy(&ttransport->lock); 704 free(ttransport); 705 return NULL; 706 } 707 708 return &ttransport->transport; 709 } 710 711 static int 712 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 713 { 714 unsigned long long ull; 715 char *end = NULL; 716 717 ull = strtoull(trsvcid, &end, 10); 718 if (end == NULL || end == trsvcid || *end != '\0') { 719 return -1; 720 } 721 722 /* Valid TCP/IP port numbers are in [0, 65535] */ 723 if (ull > 65535) { 724 return -1; 725 } 726 727 return (int)ull; 728 } 729 730 /** 731 * Canonicalize a listen address trid. 732 */ 733 static int 734 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 735 const struct spdk_nvme_transport_id *trid) 736 { 737 int trsvcid_int; 738 739 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 740 if (trsvcid_int < 0) { 741 return -EINVAL; 742 } 743 744 memset(canon_trid, 0, sizeof(*canon_trid)); 745 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 746 canon_trid->adrfam = trid->adrfam; 747 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 748 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 749 750 return 0; 751 } 752 753 /** 754 * Find an existing listening port. 755 * 756 * Caller must hold ttransport->lock. 757 */ 758 static struct spdk_nvmf_tcp_port * 759 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 760 const struct spdk_nvme_transport_id *trid) 761 { 762 struct spdk_nvme_transport_id canon_trid; 763 struct spdk_nvmf_tcp_port *port; 764 765 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 766 return NULL; 767 } 768 769 TAILQ_FOREACH(port, &ttransport->ports, link) { 770 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 771 return port; 772 } 773 } 774 775 return NULL; 776 } 777 778 static int 779 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 780 struct spdk_nvmf_listen_opts *listen_opts) 781 { 782 struct spdk_nvmf_tcp_transport *ttransport; 783 struct spdk_nvmf_tcp_port *port; 784 int trsvcid_int; 785 uint8_t adrfam; 786 struct spdk_sock_opts opts; 787 788 if (!strlen(trid->trsvcid)) { 789 SPDK_ERRLOG("Service id is required\n"); 790 return -EINVAL; 791 } 792 793 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 794 795 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 796 if (trsvcid_int < 0) { 797 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 798 return -EINVAL; 799 } 800 801 pthread_mutex_lock(&ttransport->lock); 802 port = calloc(1, sizeof(*port)); 803 if (!port) { 804 SPDK_ERRLOG("Port allocation failed\n"); 805 pthread_mutex_unlock(&ttransport->lock); 806 return -ENOMEM; 807 } 808 809 port->trid = trid; 810 opts.opts_size = sizeof(opts); 811 spdk_sock_get_default_opts(&opts); 812 opts.priority = ttransport->tcp_opts.sock_priority; 813 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 814 NULL, &opts); 815 if (port->listen_sock == NULL) { 816 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 817 trid->traddr, trsvcid_int, 818 spdk_strerror(errno), errno); 819 free(port); 820 pthread_mutex_unlock(&ttransport->lock); 821 return -errno; 822 } 823 824 if (spdk_sock_is_ipv4(port->listen_sock)) { 825 adrfam = SPDK_NVMF_ADRFAM_IPV4; 826 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 827 adrfam = SPDK_NVMF_ADRFAM_IPV6; 828 } else { 829 SPDK_ERRLOG("Unhandled socket type\n"); 830 adrfam = 0; 831 } 832 833 if (adrfam != trid->adrfam) { 834 SPDK_ERRLOG("Socket address family mismatch\n"); 835 spdk_sock_close(&port->listen_sock); 836 free(port); 837 pthread_mutex_unlock(&ttransport->lock); 838 return -EINVAL; 839 } 840 841 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 842 trid->traddr, trid->trsvcid); 843 844 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 845 pthread_mutex_unlock(&ttransport->lock); 846 return 0; 847 } 848 849 static void 850 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 851 const struct spdk_nvme_transport_id *trid) 852 { 853 struct spdk_nvmf_tcp_transport *ttransport; 854 struct spdk_nvmf_tcp_port *port; 855 856 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 857 858 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 859 trid->traddr, trid->trsvcid); 860 861 pthread_mutex_lock(&ttransport->lock); 862 port = nvmf_tcp_find_port(ttransport, trid); 863 if (port) { 864 TAILQ_REMOVE(&ttransport->ports, port, link); 865 spdk_sock_close(&port->listen_sock); 866 free(port); 867 } 868 869 pthread_mutex_unlock(&ttransport->lock); 870 } 871 872 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 873 enum nvme_tcp_pdu_recv_state state); 874 875 static void 876 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 877 { 878 tqpair->state = state; 879 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 880 tqpair->state); 881 } 882 883 static void 884 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 885 { 886 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 887 888 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, 0, 0, (uintptr_t)tqpair); 889 890 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 891 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 892 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 893 spdk_poller_unregister(&tqpair->timeout_poller); 894 895 /* This will end up calling nvmf_tcp_close_qpair */ 896 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 897 } 898 } 899 900 static void 901 _mgmt_pdu_write_done(void *_tqpair, int err) 902 { 903 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 904 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 905 906 if (spdk_unlikely(err != 0)) { 907 nvmf_tcp_qpair_disconnect(tqpair); 908 return; 909 } 910 911 assert(pdu->cb_fn != NULL); 912 pdu->cb_fn(pdu->cb_arg); 913 } 914 915 static void 916 _req_pdu_write_done(void *req, int err) 917 { 918 struct spdk_nvmf_tcp_req *tcp_req = req; 919 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 920 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 921 922 assert(tcp_req->pdu_in_use); 923 tcp_req->pdu_in_use = false; 924 925 /* If the request is in a completed state, we're waiting for write completion to free it */ 926 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 927 nvmf_tcp_request_free(tcp_req); 928 return; 929 } 930 931 if (spdk_unlikely(err != 0)) { 932 nvmf_tcp_qpair_disconnect(tqpair); 933 return; 934 } 935 936 assert(pdu->cb_fn != NULL); 937 pdu->cb_fn(pdu->cb_arg); 938 } 939 940 static void 941 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 942 { 943 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 944 } 945 946 static void 947 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 948 { 949 uint32_t mapped_length = 0; 950 ssize_t rc; 951 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 952 953 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 954 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 955 &mapped_length); 956 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 957 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 958 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 959 if (rc == mapped_length) { 960 _pdu_write_done(pdu, 0); 961 } else { 962 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 963 _pdu_write_done(pdu, -1); 964 } 965 } else { 966 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 967 } 968 } 969 970 static void 971 data_crc32_accel_done(void *cb_arg, int status) 972 { 973 struct nvme_tcp_pdu *pdu = cb_arg; 974 975 if (spdk_unlikely(status)) { 976 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 977 _pdu_write_done(pdu, status); 978 return; 979 } 980 981 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 982 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 983 984 _tcp_write_pdu(pdu); 985 } 986 987 static void 988 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 989 { 990 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 991 uint32_t crc32c; 992 993 /* Data Digest */ 994 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 995 /* Only suport this limitated case for the first step */ 996 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 997 && tqpair->group)) { 998 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, 999 pdu->data_iov, pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1000 return; 1001 } 1002 1003 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1004 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 1005 } 1006 1007 _tcp_write_pdu(pdu); 1008 } 1009 1010 static void 1011 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1012 struct nvme_tcp_pdu *pdu, 1013 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1014 void *cb_arg) 1015 { 1016 int hlen; 1017 uint32_t crc32c; 1018 1019 assert(tqpair->pdu_in_progress != pdu); 1020 1021 hlen = pdu->hdr.common.hlen; 1022 pdu->cb_fn = cb_fn; 1023 pdu->cb_arg = cb_arg; 1024 1025 pdu->iov[0].iov_base = &pdu->hdr.raw; 1026 pdu->iov[0].iov_len = hlen; 1027 1028 /* Header Digest */ 1029 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1030 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1031 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1032 } 1033 1034 /* Data Digest */ 1035 pdu_data_crc32_compute(pdu); 1036 } 1037 1038 static void 1039 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1040 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1041 void *cb_arg) 1042 { 1043 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1044 1045 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1046 pdu->sock_req.cb_arg = tqpair; 1047 1048 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1049 } 1050 1051 static void 1052 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1053 struct spdk_nvmf_tcp_req *tcp_req, 1054 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1055 void *cb_arg) 1056 { 1057 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1058 1059 pdu->sock_req.cb_fn = _req_pdu_write_done; 1060 pdu->sock_req.cb_arg = tcp_req; 1061 1062 assert(!tcp_req->pdu_in_use); 1063 tcp_req->pdu_in_use = true; 1064 1065 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1066 } 1067 1068 static int 1069 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1070 { 1071 uint32_t i; 1072 struct spdk_nvmf_transport_opts *opts; 1073 uint32_t in_capsule_data_size; 1074 1075 opts = &tqpair->qpair.transport->opts; 1076 1077 in_capsule_data_size = opts->in_capsule_data_size; 1078 if (opts->dif_insert_or_strip) { 1079 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1080 } 1081 1082 tqpair->resource_count = opts->max_queue_depth; 1083 1084 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1085 if (!tqpair->reqs) { 1086 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1087 return -1; 1088 } 1089 1090 if (in_capsule_data_size) { 1091 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1092 NULL, SPDK_ENV_LCORE_ID_ANY, 1093 SPDK_MALLOC_DMA); 1094 if (!tqpair->bufs) { 1095 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1096 return -1; 1097 } 1098 } 1099 1100 /* Add additional 2 members, which will be used for mgmt_pdu and pdu_in_progress owned by the tqpair */ 1101 tqpair->pdus = spdk_dma_zmalloc((tqpair->resource_count + 2) * sizeof(*tqpair->pdus), 0x1000, NULL); 1102 if (!tqpair->pdus) { 1103 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1104 return -1; 1105 } 1106 1107 for (i = 0; i < tqpair->resource_count; i++) { 1108 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1109 1110 tcp_req->ttag = i + 1; 1111 tcp_req->req.qpair = &tqpair->qpair; 1112 1113 tcp_req->pdu = &tqpair->pdus[i]; 1114 tcp_req->pdu->qpair = tqpair; 1115 1116 /* Set up memory to receive commands */ 1117 if (tqpair->bufs) { 1118 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1119 } 1120 1121 /* Set the cmdn and rsp */ 1122 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1123 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1124 1125 tcp_req->req.stripped_data = NULL; 1126 1127 /* Initialize request state to FREE */ 1128 tcp_req->state = TCP_REQUEST_STATE_FREE; 1129 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1130 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1131 } 1132 1133 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1134 tqpair->mgmt_pdu->qpair = tqpair; 1135 tqpair->pdu_in_progress = &tqpair->pdus[i + 1]; 1136 1137 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1138 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1139 1140 return 0; 1141 } 1142 1143 static int 1144 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1145 { 1146 struct spdk_nvmf_tcp_qpair *tqpair; 1147 1148 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1149 1150 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1151 1152 spdk_trace_record(TRACE_TCP_QP_CREATE, 0, 0, (uintptr_t)tqpair); 1153 1154 /* Initialise request state queues of the qpair */ 1155 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1156 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1157 1158 tqpair->host_hdgst_enable = true; 1159 tqpair->host_ddgst_enable = true; 1160 1161 return 0; 1162 } 1163 1164 static int 1165 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1166 { 1167 int rc; 1168 1169 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, 0, 0, (uintptr_t)tqpair); 1170 1171 /* set low water mark */ 1172 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1173 if (rc != 0) { 1174 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1175 return rc; 1176 } 1177 1178 return 0; 1179 } 1180 1181 static void 1182 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1183 struct spdk_nvmf_tcp_port *port, 1184 struct spdk_sock *sock) 1185 { 1186 struct spdk_nvmf_tcp_qpair *tqpair; 1187 int rc; 1188 1189 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1190 port->trid->traddr, port->trid->trsvcid); 1191 1192 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1193 if (tqpair == NULL) { 1194 SPDK_ERRLOG("Could not allocate new connection.\n"); 1195 spdk_sock_close(&sock); 1196 return; 1197 } 1198 1199 tqpair->sock = sock; 1200 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1201 tqpair->port = port; 1202 tqpair->qpair.transport = transport; 1203 1204 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1205 sizeof(tqpair->target_addr), &tqpair->target_port, 1206 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1207 &tqpair->initiator_port); 1208 if (rc < 0) { 1209 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1210 nvmf_tcp_qpair_destroy(tqpair); 1211 return; 1212 } 1213 1214 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1215 } 1216 1217 static uint32_t 1218 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1219 { 1220 struct spdk_sock *sock; 1221 uint32_t count = 0; 1222 int i; 1223 1224 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1225 sock = spdk_sock_accept(port->listen_sock); 1226 if (sock == NULL) { 1227 break; 1228 } 1229 count++; 1230 nvmf_tcp_handle_connect(transport, port, sock); 1231 } 1232 1233 return count; 1234 } 1235 1236 static int 1237 nvmf_tcp_accept(void *ctx) 1238 { 1239 struct spdk_nvmf_transport *transport = ctx; 1240 struct spdk_nvmf_tcp_transport *ttransport; 1241 struct spdk_nvmf_tcp_port *port; 1242 uint32_t count = 0; 1243 1244 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1245 1246 TAILQ_FOREACH(port, &ttransport->ports, link) { 1247 count += nvmf_tcp_port_accept(transport, port); 1248 } 1249 1250 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1251 } 1252 1253 static void 1254 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1255 struct spdk_nvme_transport_id *trid, 1256 struct spdk_nvmf_discovery_log_page_entry *entry) 1257 { 1258 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1259 entry->adrfam = trid->adrfam; 1260 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1261 1262 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1263 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1264 1265 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1266 } 1267 1268 static struct spdk_nvmf_tcp_control_msg_list * 1269 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1270 { 1271 struct spdk_nvmf_tcp_control_msg_list *list; 1272 struct spdk_nvmf_tcp_control_msg *msg; 1273 uint16_t i; 1274 1275 list = calloc(1, sizeof(*list)); 1276 if (!list) { 1277 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1278 return NULL; 1279 } 1280 1281 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1282 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1283 if (!list->msg_buf) { 1284 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1285 free(list); 1286 return NULL; 1287 } 1288 1289 STAILQ_INIT(&list->free_msgs); 1290 1291 for (i = 0; i < num_messages; i++) { 1292 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1293 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1294 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1295 } 1296 1297 return list; 1298 } 1299 1300 static void 1301 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1302 { 1303 if (!list) { 1304 return; 1305 } 1306 1307 spdk_free(list->msg_buf); 1308 free(list); 1309 } 1310 1311 static struct spdk_nvmf_transport_poll_group * 1312 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1313 struct spdk_nvmf_poll_group *group) 1314 { 1315 struct spdk_nvmf_tcp_transport *ttransport; 1316 struct spdk_nvmf_tcp_poll_group *tgroup; 1317 1318 tgroup = calloc(1, sizeof(*tgroup)); 1319 if (!tgroup) { 1320 return NULL; 1321 } 1322 1323 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1324 if (!tgroup->sock_group) { 1325 goto cleanup; 1326 } 1327 1328 TAILQ_INIT(&tgroup->qpairs); 1329 TAILQ_INIT(&tgroup->await_req); 1330 1331 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1332 1333 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1334 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1335 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1336 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1337 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1338 if (!tgroup->control_msg_list) { 1339 goto cleanup; 1340 } 1341 } 1342 1343 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1344 if (spdk_unlikely(!tgroup->accel_channel)) { 1345 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1346 goto cleanup; 1347 } 1348 1349 pthread_mutex_lock(&ttransport->lock); 1350 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1351 if (ttransport->next_pg == NULL) { 1352 ttransport->next_pg = tgroup; 1353 } 1354 pthread_mutex_unlock(&ttransport->lock); 1355 1356 return &tgroup->group; 1357 1358 cleanup: 1359 nvmf_tcp_poll_group_destroy(&tgroup->group); 1360 return NULL; 1361 } 1362 1363 static struct spdk_nvmf_transport_poll_group * 1364 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1365 { 1366 struct spdk_nvmf_tcp_transport *ttransport; 1367 struct spdk_nvmf_tcp_poll_group **pg; 1368 struct spdk_nvmf_tcp_qpair *tqpair; 1369 struct spdk_sock_group *group = NULL, *hint = NULL; 1370 int rc; 1371 1372 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1373 1374 pthread_mutex_lock(&ttransport->lock); 1375 1376 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1377 pthread_mutex_unlock(&ttransport->lock); 1378 return NULL; 1379 } 1380 1381 pg = &ttransport->next_pg; 1382 assert(*pg != NULL); 1383 hint = (*pg)->sock_group; 1384 1385 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1386 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1387 if (rc != 0) { 1388 pthread_mutex_unlock(&ttransport->lock); 1389 return NULL; 1390 } else if (group != NULL) { 1391 /* Optimal poll group was found */ 1392 pthread_mutex_unlock(&ttransport->lock); 1393 return spdk_sock_group_get_ctx(group); 1394 } 1395 1396 /* The hint was used for optimal poll group, advance next_pg. */ 1397 *pg = TAILQ_NEXT(*pg, link); 1398 if (*pg == NULL) { 1399 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1400 } 1401 1402 pthread_mutex_unlock(&ttransport->lock); 1403 return spdk_sock_group_get_ctx(hint); 1404 } 1405 1406 static void 1407 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1408 { 1409 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1410 struct spdk_nvmf_tcp_transport *ttransport; 1411 1412 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1413 spdk_sock_group_close(&tgroup->sock_group); 1414 if (tgroup->control_msg_list) { 1415 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1416 } 1417 1418 if (tgroup->accel_channel) { 1419 spdk_put_io_channel(tgroup->accel_channel); 1420 } 1421 1422 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1423 1424 pthread_mutex_lock(&ttransport->lock); 1425 next_tgroup = TAILQ_NEXT(tgroup, link); 1426 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1427 if (next_tgroup == NULL) { 1428 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1429 } 1430 if (ttransport->next_pg == tgroup) { 1431 ttransport->next_pg = next_tgroup; 1432 } 1433 pthread_mutex_unlock(&ttransport->lock); 1434 1435 free(tgroup); 1436 } 1437 1438 static void 1439 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1440 enum nvme_tcp_pdu_recv_state state) 1441 { 1442 if (tqpair->recv_state == state) { 1443 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1444 tqpair, state); 1445 return; 1446 } 1447 1448 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1449 /* When leaving the await req state, move the qpair to the main list */ 1450 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1451 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1452 } 1453 1454 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1455 tqpair->recv_state = state; 1456 1457 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.qid, 0, (uintptr_t)tqpair, 1458 tqpair->recv_state); 1459 1460 switch (state) { 1461 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1462 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1463 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1464 break; 1465 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1466 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1467 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1468 break; 1469 case NVME_TCP_PDU_RECV_STATE_ERROR: 1470 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1471 memset(tqpair->pdu_in_progress, 0, sizeof(*(tqpair->pdu_in_progress))); 1472 break; 1473 default: 1474 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1475 abort(); 1476 break; 1477 } 1478 } 1479 1480 static int 1481 nvmf_tcp_qpair_handle_timeout(void *ctx) 1482 { 1483 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1484 1485 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1486 1487 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1488 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1489 1490 nvmf_tcp_qpair_disconnect(tqpair); 1491 return SPDK_POLLER_BUSY; 1492 } 1493 1494 static void 1495 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1496 { 1497 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1498 1499 if (!tqpair->timeout_poller) { 1500 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1501 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1502 } 1503 } 1504 1505 static void 1506 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1507 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1508 { 1509 struct nvme_tcp_pdu *rsp_pdu; 1510 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1511 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1512 uint32_t copy_len; 1513 1514 rsp_pdu = tqpair->mgmt_pdu; 1515 1516 c2h_term_req = &rsp_pdu->hdr.term_req; 1517 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1518 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1519 1520 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1521 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1522 DSET32(&c2h_term_req->fei, error_offset); 1523 } 1524 1525 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1526 1527 /* Copy the error info into the buffer */ 1528 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1529 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1530 1531 /* Contain the header of the wrong received pdu */ 1532 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1533 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1534 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1535 } 1536 1537 static void 1538 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1539 struct spdk_nvmf_tcp_qpair *tqpair, 1540 struct nvme_tcp_pdu *pdu) 1541 { 1542 struct spdk_nvmf_tcp_req *tcp_req; 1543 1544 assert(pdu->psh_valid_bytes == pdu->psh_len); 1545 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1546 1547 tcp_req = nvmf_tcp_req_get(tqpair); 1548 if (!tcp_req) { 1549 /* Directly return and make the allocation retry again. This can happen if we're 1550 * using asynchronous writes to send the response to the host or when releasing 1551 * zero-copy buffers after a response has been sent. In both cases, the host might 1552 * receive the response before we've finished processing the request and is free to 1553 * send another one. 1554 */ 1555 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1556 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1557 return; 1558 } 1559 1560 /* The host sent more commands than the maximum queue depth. */ 1561 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1562 nvmf_tcp_qpair_disconnect(tqpair); 1563 return; 1564 } 1565 1566 pdu->req = tcp_req; 1567 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1568 nvmf_tcp_req_process(ttransport, tcp_req); 1569 } 1570 1571 static void 1572 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1573 struct spdk_nvmf_tcp_qpair *tqpair, 1574 struct nvme_tcp_pdu *pdu) 1575 { 1576 struct spdk_nvmf_tcp_req *tcp_req; 1577 struct spdk_nvme_tcp_cmd *capsule_cmd; 1578 uint32_t error_offset = 0; 1579 enum spdk_nvme_tcp_term_req_fes fes; 1580 struct spdk_nvme_cpl *rsp; 1581 1582 capsule_cmd = &pdu->hdr.capsule_cmd; 1583 tcp_req = pdu->req; 1584 assert(tcp_req != NULL); 1585 1586 /* Zero-copy requests don't support ICD */ 1587 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1588 1589 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1590 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1591 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1592 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1593 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1594 goto err; 1595 } 1596 1597 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1598 1599 rsp = &tcp_req->req.rsp->nvme_cpl; 1600 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1601 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1602 } else { 1603 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1604 } 1605 1606 nvmf_tcp_req_process(ttransport, tcp_req); 1607 1608 return; 1609 err: 1610 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1611 } 1612 1613 static int 1614 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1615 enum spdk_nvmf_tcp_req_state state, 1616 uint16_t cid, uint16_t tag, 1617 struct spdk_nvmf_tcp_req **req) 1618 { 1619 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1620 1621 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1622 if (tcp_req->state != state) { 1623 continue; 1624 } 1625 1626 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1627 continue; 1628 } 1629 1630 if (tcp_req->ttag == tag) { 1631 *req = tcp_req; 1632 return 0; 1633 } 1634 1635 *req = NULL; 1636 return -1; 1637 } 1638 1639 /* Didn't find it, but not an error */ 1640 *req = NULL; 1641 return 0; 1642 } 1643 1644 static void 1645 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1646 struct spdk_nvmf_tcp_qpair *tqpair, 1647 struct nvme_tcp_pdu *pdu) 1648 { 1649 struct spdk_nvmf_tcp_req *tcp_req; 1650 uint32_t error_offset = 0; 1651 enum spdk_nvme_tcp_term_req_fes fes = 0; 1652 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1653 int rc; 1654 1655 h2c_data = &pdu->hdr.h2c_data; 1656 1657 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1658 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1659 1660 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1661 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1662 if (rc == 0 && tcp_req == NULL) { 1663 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1664 h2c_data->ttag, &tcp_req); 1665 } 1666 1667 if (!tcp_req) { 1668 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1669 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1670 if (rc == 0) { 1671 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1672 } else { 1673 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1674 } 1675 goto err; 1676 } 1677 1678 if (tcp_req->h2c_offset != h2c_data->datao) { 1679 SPDK_DEBUGLOG(nvmf_tcp, 1680 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1681 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1682 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1683 goto err; 1684 } 1685 1686 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1687 SPDK_DEBUGLOG(nvmf_tcp, 1688 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1689 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1690 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1691 goto err; 1692 } 1693 1694 pdu->req = tcp_req; 1695 1696 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1697 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1698 } 1699 1700 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1701 h2c_data->datao, h2c_data->datal); 1702 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1703 return; 1704 1705 err: 1706 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1707 } 1708 1709 static void 1710 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1711 struct spdk_nvmf_tcp_qpair *tqpair) 1712 { 1713 struct nvme_tcp_pdu *rsp_pdu; 1714 struct spdk_nvme_tcp_rsp *capsule_resp; 1715 1716 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1717 1718 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1719 assert(rsp_pdu != NULL); 1720 1721 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1722 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1723 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1724 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1725 if (tqpair->host_hdgst_enable) { 1726 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1727 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1728 } 1729 1730 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1731 } 1732 1733 static void 1734 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1735 { 1736 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1737 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1738 struct spdk_nvmf_tcp_qpair, qpair); 1739 1740 assert(tqpair != NULL); 1741 1742 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1743 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1744 tcp_req->req.length); 1745 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1746 return; 1747 } 1748 1749 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1750 nvmf_tcp_request_free(tcp_req); 1751 } else { 1752 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1753 } 1754 } 1755 1756 static void 1757 nvmf_tcp_r2t_complete(void *cb_arg) 1758 { 1759 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1760 struct spdk_nvmf_tcp_transport *ttransport; 1761 1762 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1763 struct spdk_nvmf_tcp_transport, transport); 1764 1765 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1766 1767 if (tcp_req->h2c_offset == tcp_req->req.length) { 1768 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1769 nvmf_tcp_req_process(ttransport, tcp_req); 1770 } 1771 } 1772 1773 static void 1774 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1775 struct spdk_nvmf_tcp_req *tcp_req) 1776 { 1777 struct nvme_tcp_pdu *rsp_pdu; 1778 struct spdk_nvme_tcp_r2t_hdr *r2t; 1779 1780 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1781 assert(rsp_pdu != NULL); 1782 1783 r2t = &rsp_pdu->hdr.r2t; 1784 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1785 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1786 1787 if (tqpair->host_hdgst_enable) { 1788 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1789 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1790 } 1791 1792 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1793 r2t->ttag = tcp_req->ttag; 1794 r2t->r2to = tcp_req->h2c_offset; 1795 r2t->r2tl = tcp_req->req.length; 1796 1797 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1798 1799 SPDK_DEBUGLOG(nvmf_tcp, 1800 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1801 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1802 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1803 } 1804 1805 static void 1806 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1807 struct spdk_nvmf_tcp_qpair *tqpair, 1808 struct nvme_tcp_pdu *pdu) 1809 { 1810 struct spdk_nvmf_tcp_req *tcp_req; 1811 struct spdk_nvme_cpl *rsp; 1812 1813 tcp_req = pdu->req; 1814 assert(tcp_req != NULL); 1815 1816 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1817 1818 tcp_req->h2c_offset += pdu->data_len; 1819 1820 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1821 1822 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1823 * acknowledged before moving on. */ 1824 if (tcp_req->h2c_offset == tcp_req->req.length && 1825 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1826 /* After receiving all the h2c data, we need to check whether there is 1827 * transient transport error */ 1828 rsp = &tcp_req->req.rsp->nvme_cpl; 1829 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1830 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1831 } else { 1832 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1833 } 1834 nvmf_tcp_req_process(ttransport, tcp_req); 1835 } 1836 } 1837 1838 static void 1839 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1840 { 1841 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1842 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1843 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1844 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1845 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1846 DGET32(h2c_term_req->fei)); 1847 } 1848 } 1849 1850 static void 1851 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1852 struct nvme_tcp_pdu *pdu) 1853 { 1854 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1855 uint32_t error_offset = 0; 1856 enum spdk_nvme_tcp_term_req_fes fes; 1857 1858 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1859 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1860 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1861 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1862 goto end; 1863 } 1864 1865 /* set the data buffer */ 1866 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1867 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1868 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1869 return; 1870 end: 1871 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1872 } 1873 1874 static void 1875 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1876 struct nvme_tcp_pdu *pdu) 1877 { 1878 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1879 1880 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1881 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1882 } 1883 1884 static void 1885 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1886 struct spdk_nvmf_tcp_transport *ttransport) 1887 { 1888 struct nvme_tcp_pdu *pdu = tqpair->pdu_in_progress; 1889 1890 switch (pdu->hdr.common.pdu_type) { 1891 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1892 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1893 break; 1894 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1895 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1896 break; 1897 1898 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1899 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1900 break; 1901 1902 default: 1903 /* The code should not go to here */ 1904 SPDK_ERRLOG("The code should not go to here\n"); 1905 break; 1906 } 1907 } 1908 1909 static void 1910 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1911 struct spdk_nvmf_tcp_transport *ttransport) 1912 { 1913 int rc = 0; 1914 struct nvme_tcp_pdu *pdu; 1915 uint32_t crc32c; 1916 struct spdk_nvmf_tcp_req *tcp_req; 1917 struct spdk_nvme_cpl *rsp; 1918 1919 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1920 pdu = tqpair->pdu_in_progress; 1921 1922 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1923 /* check data digest if need */ 1924 if (pdu->ddgst_enable) { 1925 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1926 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1927 if (rc == 0) { 1928 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1929 tcp_req = pdu->req; 1930 assert(tcp_req != NULL); 1931 rsp = &tcp_req->req.rsp->nvme_cpl; 1932 rsp->status.sc = SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR; 1933 } 1934 } 1935 1936 _nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1937 } 1938 1939 static void 1940 nvmf_tcp_send_icresp_complete(void *cb_arg) 1941 { 1942 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1943 1944 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 1945 } 1946 1947 static void 1948 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1949 struct spdk_nvmf_tcp_qpair *tqpair, 1950 struct nvme_tcp_pdu *pdu) 1951 { 1952 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1953 struct nvme_tcp_pdu *rsp_pdu; 1954 struct spdk_nvme_tcp_ic_resp *ic_resp; 1955 uint32_t error_offset = 0; 1956 enum spdk_nvme_tcp_term_req_fes fes; 1957 1958 /* Only PFV 0 is defined currently */ 1959 if (ic_req->pfv != 0) { 1960 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1961 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1962 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1963 goto end; 1964 } 1965 1966 /* MAXR2T is 0's based */ 1967 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1968 1969 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1970 if (!tqpair->host_hdgst_enable) { 1971 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1972 } 1973 1974 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1975 if (!tqpair->host_ddgst_enable) { 1976 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1977 } 1978 1979 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1980 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1981 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1982 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1983 tqpair, 1984 tqpair->recv_buf_size); 1985 /* Not fatal. */ 1986 } 1987 1988 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1989 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1990 1991 rsp_pdu = tqpair->mgmt_pdu; 1992 1993 ic_resp = &rsp_pdu->hdr.ic_resp; 1994 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1995 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1996 ic_resp->pfv = 0; 1997 ic_resp->cpda = tqpair->cpda; 1998 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1999 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2000 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2001 2002 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2003 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2004 2005 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2006 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2007 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2008 return; 2009 end: 2010 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2011 } 2012 2013 static void 2014 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2015 struct spdk_nvmf_tcp_transport *ttransport) 2016 { 2017 struct nvme_tcp_pdu *pdu; 2018 int rc; 2019 uint32_t crc32c, error_offset = 0; 2020 enum spdk_nvme_tcp_term_req_fes fes; 2021 2022 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2023 pdu = tqpair->pdu_in_progress; 2024 2025 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2026 pdu->hdr.common.pdu_type); 2027 /* check header digest if needed */ 2028 if (pdu->has_hdgst) { 2029 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2030 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2031 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2032 if (rc == 0) { 2033 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2034 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2035 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2036 return; 2037 2038 } 2039 } 2040 2041 switch (pdu->hdr.common.pdu_type) { 2042 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2043 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2044 break; 2045 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2046 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2047 break; 2048 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2049 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2050 break; 2051 2052 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2053 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2054 break; 2055 2056 default: 2057 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2058 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2059 error_offset = 1; 2060 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2061 break; 2062 } 2063 } 2064 2065 static void 2066 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2067 { 2068 struct nvme_tcp_pdu *pdu; 2069 uint32_t error_offset = 0; 2070 enum spdk_nvme_tcp_term_req_fes fes; 2071 uint8_t expected_hlen, pdo; 2072 bool plen_error = false, pdo_error = false; 2073 2074 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2075 pdu = tqpair->pdu_in_progress; 2076 2077 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2078 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2079 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2080 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2081 goto err; 2082 } 2083 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2084 if (pdu->hdr.common.plen != expected_hlen) { 2085 plen_error = true; 2086 } 2087 } else { 2088 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2089 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2090 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2091 goto err; 2092 } 2093 2094 switch (pdu->hdr.common.pdu_type) { 2095 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2096 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2097 pdo = pdu->hdr.common.pdo; 2098 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2099 pdo_error = true; 2100 break; 2101 } 2102 2103 if (pdu->hdr.common.plen < expected_hlen) { 2104 plen_error = true; 2105 } 2106 break; 2107 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2108 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2109 pdo = pdu->hdr.common.pdo; 2110 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2111 pdo_error = true; 2112 break; 2113 } 2114 if (pdu->hdr.common.plen < expected_hlen) { 2115 plen_error = true; 2116 } 2117 break; 2118 2119 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2120 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2121 if ((pdu->hdr.common.plen <= expected_hlen) || 2122 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2123 plen_error = true; 2124 } 2125 break; 2126 2127 default: 2128 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2129 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2130 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2131 goto err; 2132 } 2133 } 2134 2135 if (pdu->hdr.common.hlen != expected_hlen) { 2136 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2137 pdu->hdr.common.pdu_type, 2138 expected_hlen, pdu->hdr.common.hlen, tqpair); 2139 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2140 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2141 goto err; 2142 } else if (pdo_error) { 2143 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2144 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2145 } else if (plen_error) { 2146 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2147 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2148 goto err; 2149 } else { 2150 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2151 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2152 return; 2153 } 2154 err: 2155 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2156 } 2157 2158 static int 2159 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 2160 int read_len) 2161 { 2162 int rc; 2163 2164 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 2165 read_offset, read_len, pdu->dif_ctx); 2166 if (rc != 0) { 2167 SPDK_ERRLOG("DIF generate failed\n"); 2168 } 2169 2170 return rc; 2171 } 2172 2173 static int 2174 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2175 { 2176 int rc = 0; 2177 struct nvme_tcp_pdu *pdu; 2178 enum nvme_tcp_pdu_recv_state prev_state; 2179 uint32_t data_len; 2180 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2181 struct spdk_nvmf_tcp_transport, transport); 2182 2183 /* The loop here is to allow for several back-to-back state changes. */ 2184 do { 2185 prev_state = tqpair->recv_state; 2186 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2187 2188 pdu = tqpair->pdu_in_progress; 2189 switch (tqpair->recv_state) { 2190 /* Wait for the common header */ 2191 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2192 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2193 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2194 return rc; 2195 } 2196 2197 rc = nvme_tcp_read_data(tqpair->sock, 2198 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2199 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2200 if (rc < 0) { 2201 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2202 return NVME_TCP_PDU_FATAL; 2203 } else if (rc > 0) { 2204 pdu->ch_valid_bytes += rc; 2205 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2206 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 2207 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2208 } 2209 } 2210 2211 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2212 return NVME_TCP_PDU_IN_PROGRESS; 2213 } 2214 2215 /* The command header of this PDU has now been read from the socket. */ 2216 nvmf_tcp_pdu_ch_handle(tqpair); 2217 break; 2218 /* Wait for the pdu specific header */ 2219 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2220 rc = nvme_tcp_read_data(tqpair->sock, 2221 pdu->psh_len - pdu->psh_valid_bytes, 2222 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2223 if (rc < 0) { 2224 return NVME_TCP_PDU_FATAL; 2225 } else if (rc > 0) { 2226 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.qid, rc, 0, tqpair); 2227 pdu->psh_valid_bytes += rc; 2228 } 2229 2230 if (pdu->psh_valid_bytes < pdu->psh_len) { 2231 return NVME_TCP_PDU_IN_PROGRESS; 2232 } 2233 2234 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2235 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2236 break; 2237 /* Wait for the req slot */ 2238 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2239 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2240 break; 2241 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2242 /* check whether the data is valid, if not we just return */ 2243 if (!pdu->data_len) { 2244 return NVME_TCP_PDU_IN_PROGRESS; 2245 } 2246 2247 data_len = pdu->data_len; 2248 /* data digest */ 2249 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2250 tqpair->host_ddgst_enable)) { 2251 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2252 pdu->ddgst_enable = true; 2253 } 2254 2255 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2256 if (rc < 0) { 2257 return NVME_TCP_PDU_FATAL; 2258 } 2259 pdu->rw_offset += rc; 2260 2261 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2262 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2263 if (rc != 0) { 2264 return NVME_TCP_PDU_FATAL; 2265 } 2266 } 2267 2268 if (pdu->rw_offset < data_len) { 2269 return NVME_TCP_PDU_IN_PROGRESS; 2270 } 2271 2272 /* All of this PDU has now been read from the socket. */ 2273 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2274 break; 2275 case NVME_TCP_PDU_RECV_STATE_ERROR: 2276 if (!spdk_sock_is_connected(tqpair->sock)) { 2277 return NVME_TCP_PDU_FATAL; 2278 } 2279 break; 2280 default: 2281 assert(0); 2282 SPDK_ERRLOG("code should not come to here"); 2283 break; 2284 } 2285 } while (tqpair->recv_state != prev_state); 2286 2287 return rc; 2288 } 2289 2290 static inline void * 2291 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2292 { 2293 struct spdk_nvmf_tcp_control_msg *msg; 2294 2295 assert(list); 2296 2297 msg = STAILQ_FIRST(&list->free_msgs); 2298 if (!msg) { 2299 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2300 return NULL; 2301 } 2302 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2303 return msg; 2304 } 2305 2306 static inline void 2307 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2308 { 2309 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2310 2311 assert(list); 2312 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2313 } 2314 2315 static int 2316 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2317 struct spdk_nvmf_transport *transport, 2318 struct spdk_nvmf_transport_poll_group *group) 2319 { 2320 struct spdk_nvmf_request *req = &tcp_req->req; 2321 struct spdk_nvme_cmd *cmd; 2322 struct spdk_nvme_cpl *rsp; 2323 struct spdk_nvme_sgl_descriptor *sgl; 2324 struct spdk_nvmf_tcp_poll_group *tgroup; 2325 uint32_t length; 2326 2327 cmd = &req->cmd->nvme_cmd; 2328 rsp = &req->rsp->nvme_cpl; 2329 sgl = &cmd->dptr.sgl1; 2330 2331 length = sgl->unkeyed.length; 2332 2333 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2334 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2335 if (length > transport->opts.max_io_size) { 2336 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2337 length, transport->opts.max_io_size); 2338 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2339 return -1; 2340 } 2341 2342 /* fill request length and populate iovs */ 2343 req->length = length; 2344 2345 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2346 2347 if (spdk_unlikely(req->dif_enabled)) { 2348 req->dif.orig_length = length; 2349 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2350 req->dif.elba_length = length; 2351 } 2352 2353 if (nvmf_ctrlr_use_zcopy(req)) { 2354 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2355 req->data_from_pool = false; 2356 return 0; 2357 } 2358 2359 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2360 /* No available buffers. Queue this request up. */ 2361 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2362 tcp_req); 2363 return 0; 2364 } 2365 2366 /* backward compatible */ 2367 req->data = req->iov[0].iov_base; 2368 2369 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2370 tcp_req, req->iovcnt, req->data); 2371 2372 return 0; 2373 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2374 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2375 uint64_t offset = sgl->address; 2376 uint32_t max_len = transport->opts.in_capsule_data_size; 2377 assert(tcp_req->has_in_capsule_data); 2378 2379 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2380 offset, length); 2381 2382 if (offset > max_len) { 2383 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2384 offset, max_len); 2385 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2386 return -1; 2387 } 2388 max_len -= (uint32_t)offset; 2389 2390 if (spdk_unlikely(length > max_len)) { 2391 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2392 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2393 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2394 2395 /* Get a buffer from dedicated list */ 2396 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2397 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2398 assert(tgroup->control_msg_list); 2399 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2400 if (!req->data) { 2401 /* No available buffers. Queue this request up. */ 2402 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2403 return 0; 2404 } 2405 } else { 2406 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2407 length, max_len); 2408 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2409 return -1; 2410 } 2411 } else { 2412 req->data = tcp_req->buf; 2413 } 2414 2415 req->length = length; 2416 req->data_from_pool = false; 2417 2418 if (spdk_unlikely(req->dif_enabled)) { 2419 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2420 req->dif.elba_length = length; 2421 } 2422 2423 req->iov[0].iov_base = req->data; 2424 req->iov[0].iov_len = length; 2425 req->iovcnt = 1; 2426 2427 return 0; 2428 } 2429 2430 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2431 sgl->generic.type, sgl->generic.subtype); 2432 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2433 return -1; 2434 } 2435 2436 static inline enum spdk_nvme_media_error_status_code 2437 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2438 enum spdk_nvme_media_error_status_code result; 2439 2440 switch (err_type) 2441 { 2442 case SPDK_DIF_REFTAG_ERROR: 2443 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2444 break; 2445 case SPDK_DIF_APPTAG_ERROR: 2446 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2447 break; 2448 case SPDK_DIF_GUARD_ERROR: 2449 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2450 break; 2451 default: 2452 SPDK_UNREACHABLE(); 2453 break; 2454 } 2455 2456 return result; 2457 } 2458 2459 static void 2460 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2461 struct spdk_nvmf_tcp_req *tcp_req) 2462 { 2463 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2464 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2465 struct nvme_tcp_pdu *rsp_pdu; 2466 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2467 uint32_t plen, pdo, alignment; 2468 int rc; 2469 2470 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2471 2472 rsp_pdu = tcp_req->pdu; 2473 assert(rsp_pdu != NULL); 2474 2475 c2h_data = &rsp_pdu->hdr.c2h_data; 2476 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2477 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2478 2479 if (tqpair->host_hdgst_enable) { 2480 plen += SPDK_NVME_TCP_DIGEST_LEN; 2481 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2482 } 2483 2484 /* set the psh */ 2485 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2486 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2487 c2h_data->datao = tcp_req->pdu->rw_offset; 2488 2489 /* set the padding */ 2490 rsp_pdu->padding_len = 0; 2491 pdo = plen; 2492 if (tqpair->cpda) { 2493 alignment = (tqpair->cpda + 1) << 2; 2494 if (plen % alignment != 0) { 2495 pdo = (plen + alignment) / alignment * alignment; 2496 rsp_pdu->padding_len = pdo - plen; 2497 plen = pdo; 2498 } 2499 } 2500 2501 c2h_data->common.pdo = pdo; 2502 plen += c2h_data->datal; 2503 if (tqpair->host_ddgst_enable) { 2504 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2505 plen += SPDK_NVME_TCP_DIGEST_LEN; 2506 } 2507 2508 c2h_data->common.plen = plen; 2509 2510 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2511 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2512 } 2513 2514 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2515 c2h_data->datao, c2h_data->datal); 2516 2517 2518 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2519 /* Need to send the capsule response if response is not all 0 */ 2520 if (ttransport->tcp_opts.c2h_success && 2521 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2522 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2523 } 2524 2525 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2526 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2527 struct spdk_dif_error err_blk = {}; 2528 uint32_t mapped_length = 0; 2529 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2530 uint32_t ddgst_len = 0; 2531 2532 if (tqpair->host_ddgst_enable) { 2533 /* Data digest consumes additional iov entry */ 2534 available_iovs--; 2535 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2536 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2537 c2h_data->common.plen -= ddgst_len; 2538 } 2539 /* Temp call to estimate if data can be described by limited number of iovs. 2540 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2541 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2542 false, &mapped_length); 2543 2544 if (mapped_length != c2h_data->common.plen) { 2545 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2546 SPDK_DEBUGLOG(nvmf_tcp, 2547 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2548 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2549 c2h_data->common.plen = mapped_length; 2550 2551 /* Rebuild pdu->data_iov since data length is changed */ 2552 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2553 c2h_data->datal); 2554 2555 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2556 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2557 } 2558 2559 c2h_data->common.plen += ddgst_len; 2560 2561 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2562 2563 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2564 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2565 if (rc != 0) { 2566 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2567 err_blk.err_type, err_blk.err_offset); 2568 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2569 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2570 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2571 return; 2572 } 2573 } 2574 2575 rsp_pdu->rw_offset += c2h_data->datal; 2576 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2577 } 2578 2579 static void 2580 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2581 struct spdk_nvmf_tcp_req *tcp_req) 2582 { 2583 nvmf_tcp_req_pdu_init(tcp_req); 2584 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2585 } 2586 2587 static int 2588 request_transfer_out(struct spdk_nvmf_request *req) 2589 { 2590 struct spdk_nvmf_tcp_req *tcp_req; 2591 struct spdk_nvmf_qpair *qpair; 2592 struct spdk_nvmf_tcp_qpair *tqpair; 2593 struct spdk_nvme_cpl *rsp; 2594 2595 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2596 2597 qpair = req->qpair; 2598 rsp = &req->rsp->nvme_cpl; 2599 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2600 2601 /* Advance our sq_head pointer */ 2602 if (qpair->sq_head == qpair->sq_head_max) { 2603 qpair->sq_head = 0; 2604 } else { 2605 qpair->sq_head++; 2606 } 2607 rsp->sqhd = qpair->sq_head; 2608 2609 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2610 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2611 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2612 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2613 } else { 2614 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2615 } 2616 2617 return 0; 2618 } 2619 2620 static void 2621 nvmf_tcp_set_in_capsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2622 struct spdk_nvmf_tcp_req *tcp_req) 2623 { 2624 struct nvme_tcp_pdu *pdu; 2625 uint32_t plen = 0; 2626 2627 pdu = tqpair->pdu_in_progress; 2628 plen = pdu->hdr.common.hlen; 2629 2630 if (tqpair->host_hdgst_enable) { 2631 plen += SPDK_NVME_TCP_DIGEST_LEN; 2632 } 2633 2634 if (pdu->hdr.common.plen != plen) { 2635 tcp_req->has_in_capsule_data = true; 2636 } 2637 } 2638 2639 static void 2640 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2641 struct spdk_nvmf_tcp_qpair *tqpair, 2642 struct spdk_nvmf_tcp_req *tcp_req) 2643 { 2644 enum spdk_nvme_cmd_fuse last, next; 2645 2646 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2647 next = tcp_req->cmd.fuse; 2648 2649 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2650 2651 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2652 return; 2653 } 2654 2655 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2656 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2657 /* This is a valid pair of fused commands. Point them at each other 2658 * so they can be submitted consecutively once ready to be executed. 2659 */ 2660 tqpair->fused_first->fused_pair = tcp_req; 2661 tcp_req->fused_pair = tqpair->fused_first; 2662 tqpair->fused_first = NULL; 2663 return; 2664 } else { 2665 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2666 tqpair->fused_first->fused_failed = true; 2667 2668 /* 2669 * If the last req is in READY_TO_EXECUTE state, then call 2670 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2671 */ 2672 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2673 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2674 } 2675 2676 tqpair->fused_first = NULL; 2677 } 2678 } 2679 2680 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2681 /* Set tqpair->fused_first here so that we know to check that the next request 2682 * is a SECOND (and to fail this one if it isn't). 2683 */ 2684 tqpair->fused_first = tcp_req; 2685 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2686 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2687 tcp_req->fused_failed = true; 2688 } 2689 } 2690 2691 static bool 2692 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2693 struct spdk_nvmf_tcp_req *tcp_req) 2694 { 2695 struct spdk_nvmf_tcp_qpair *tqpair; 2696 int rc; 2697 enum spdk_nvmf_tcp_req_state prev_state; 2698 bool progress = false; 2699 struct spdk_nvmf_transport *transport = &ttransport->transport; 2700 struct spdk_nvmf_transport_poll_group *group; 2701 struct spdk_nvmf_tcp_poll_group *tgroup; 2702 2703 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2704 group = &tqpair->group->group; 2705 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2706 2707 /* If the qpair is not active, we need to abort the outstanding requests. */ 2708 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2709 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2710 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2711 } 2712 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2713 } 2714 2715 /* The loop here is to allow for several back-to-back state changes. */ 2716 do { 2717 prev_state = tcp_req->state; 2718 2719 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2720 tqpair); 2721 2722 switch (tcp_req->state) { 2723 case TCP_REQUEST_STATE_FREE: 2724 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2725 * to escape this state. */ 2726 break; 2727 case TCP_REQUEST_STATE_NEW: 2728 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, tqpair); 2729 2730 /* copy the cmd from the receive pdu */ 2731 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2732 2733 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2734 tcp_req->req.dif_enabled = true; 2735 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2736 } 2737 2738 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2739 2740 /* The next state transition depends on the data transfer needs of this request. */ 2741 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2742 2743 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2744 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2745 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 2746 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2747 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2748 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2749 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2750 break; 2751 } 2752 2753 /* If no data to transfer, ready to execute. */ 2754 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2755 /* Reset the tqpair receiving pdu state */ 2756 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2757 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2758 break; 2759 } 2760 2761 nvmf_tcp_set_in_capsule_data(tqpair, tcp_req); 2762 2763 if (!tcp_req->has_in_capsule_data) { 2764 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2765 } 2766 2767 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2768 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2769 break; 2770 case TCP_REQUEST_STATE_NEED_BUFFER: 2771 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2772 tqpair); 2773 2774 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2775 2776 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2777 SPDK_DEBUGLOG(nvmf_tcp, 2778 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2779 tcp_req, tqpair); 2780 /* This request needs to wait in line to obtain a buffer */ 2781 break; 2782 } 2783 2784 /* Try to get a data buffer */ 2785 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2786 if (rc < 0) { 2787 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2788 /* Reset the tqpair receiving pdu state */ 2789 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2790 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2791 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2792 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2793 break; 2794 } 2795 2796 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2797 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2798 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2799 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2800 tcp_req->req.length = tcp_req->req.dif.elba_length; 2801 } 2802 2803 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2804 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2805 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2806 break; 2807 } 2808 2809 if (!tcp_req->req.data) { 2810 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2811 tcp_req, tqpair); 2812 /* No buffers available. */ 2813 break; 2814 } 2815 2816 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2817 2818 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2819 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2820 if (tcp_req->req.data_from_pool) { 2821 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2822 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2823 } else { 2824 struct nvme_tcp_pdu *pdu; 2825 2826 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2827 2828 pdu = tqpair->pdu_in_progress; 2829 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2830 tqpair); 2831 /* No need to send r2t, contained in the capsuled data */ 2832 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2833 0, tcp_req->req.length); 2834 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2835 } 2836 break; 2837 } 2838 2839 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2840 break; 2841 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 2842 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.qid, 0, 2843 (uintptr_t)tcp_req, tqpair); 2844 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 2845 * to escape this state. */ 2846 break; 2847 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 2848 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.qid, 0, 2849 (uintptr_t)tcp_req, tqpair); 2850 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 2851 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 2852 tcp_req, tqpair); 2853 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2854 break; 2855 } 2856 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2857 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2858 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2859 } else { 2860 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2861 } 2862 break; 2863 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2864 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2865 tqpair); 2866 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2867 break; 2868 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2869 2870 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.qid, 0, 2871 (uintptr_t)tcp_req, tqpair); 2872 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2873 * to escape this state. */ 2874 break; 2875 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2876 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.qid, 0, 2877 (uintptr_t)tcp_req, tqpair); 2878 2879 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2880 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2881 tcp_req->req.length = tcp_req->req.dif.elba_length; 2882 } 2883 2884 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 2885 if (tcp_req->fused_failed) { 2886 /* This request failed FUSED semantics. Fail it immediately, without 2887 * even sending it to the target layer. 2888 */ 2889 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2890 tcp_req->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 2891 tcp_req->req.rsp->nvme_cpl.cid = tcp_req->req.cmd->nvme_cmd.cid; 2892 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2893 break; 2894 } 2895 2896 if (tcp_req->fused_pair == NULL || 2897 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2898 /* This request is ready to execute, but either we don't know yet if it's 2899 * valid - i.e. this is a FIRST but we haven't received the next request yet), 2900 * or the other request of this fused pair isn't ready to execute. So 2901 * break here and this request will get processed later either when the 2902 * other request is ready or we find that this request isn't valid. 2903 */ 2904 break; 2905 } 2906 } 2907 2908 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2909 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2910 /* If we get to this point, and this request is a fused command, we know that 2911 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 2912 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 2913 * request, and the other request of the fused pair, in the correct order. 2914 * Also clear the ->fused_pair pointers on both requests, since after this point 2915 * we no longer need to maintain the relationship between these two requests. 2916 */ 2917 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 2918 assert(tcp_req->fused_pair != NULL); 2919 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2920 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2921 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2922 tcp_req->fused_pair->fused_pair = NULL; 2923 tcp_req->fused_pair = NULL; 2924 } 2925 spdk_nvmf_request_exec(&tcp_req->req); 2926 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 2927 assert(tcp_req->fused_pair != NULL); 2928 assert(tcp_req->fused_pair->fused_pair == tcp_req); 2929 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 2930 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 2931 tcp_req->fused_pair->fused_pair = NULL; 2932 tcp_req->fused_pair = NULL; 2933 } 2934 } else { 2935 /* For zero-copy, only requests with data coming from host to the 2936 * controller can end up here. */ 2937 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 2938 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 2939 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 2940 } 2941 2942 break; 2943 case TCP_REQUEST_STATE_EXECUTING: 2944 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2945 tqpair); 2946 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2947 * to escape this state. */ 2948 break; 2949 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 2950 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.qid, 0, 2951 (uintptr_t)tcp_req, tqpair); 2952 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2953 * to escape this state. */ 2954 break; 2955 case TCP_REQUEST_STATE_EXECUTED: 2956 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2957 tqpair); 2958 2959 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2960 tcp_req->req.length = tcp_req->req.dif.orig_length; 2961 } 2962 2963 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2964 break; 2965 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2966 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.qid, 0, 2967 (uintptr_t)tcp_req, tqpair); 2968 rc = request_transfer_out(&tcp_req->req); 2969 assert(rc == 0); /* No good way to handle this currently */ 2970 break; 2971 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2972 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.qid, 0, 2973 (uintptr_t)tcp_req, tqpair); 2974 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2975 * to escape this state. */ 2976 break; 2977 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 2978 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.qid, 0, 2979 (uintptr_t)tcp_req, tqpair); 2980 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2981 * to escape this state. */ 2982 break; 2983 case TCP_REQUEST_STATE_COMPLETED: 2984 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.qid, 0, (uintptr_t)tcp_req, 2985 tqpair); 2986 /* If there's an outstanding PDU sent to the host, the request is completed 2987 * due to the qpair being disconnected. We must delay the completion until 2988 * that write is done to avoid freeing the request twice. */ 2989 if (spdk_unlikely(tcp_req->pdu_in_use)) { 2990 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 2991 "write on req=%p\n", tcp_req); 2992 /* This can only happen for zcopy requests */ 2993 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 2994 assert(tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 2995 break; 2996 } 2997 2998 if (tcp_req->req.data_from_pool) { 2999 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3000 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3001 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3002 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3003 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3004 assert(tgroup->control_msg_list); 3005 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3006 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 3007 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3008 /* If the request has an unreleased zcopy bdev_io, it's either a 3009 * read, a failed write, or the qpair is being disconnected */ 3010 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3011 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3012 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3013 tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 3014 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3015 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3016 break; 3017 } 3018 tcp_req->req.length = 0; 3019 tcp_req->req.iovcnt = 0; 3020 tcp_req->req.data = NULL; 3021 tcp_req->fused_failed = false; 3022 if (tcp_req->fused_pair) { 3023 /* This req was part of a valid fused pair, but failed before it got to 3024 * READ_TO_EXECUTE state. This means we need to fail the other request 3025 * in the pair, because it is no longer part of a valid pair. If the pair 3026 * already reached READY_TO_EXECUTE state, we need to kick it. 3027 */ 3028 tcp_req->fused_pair->fused_failed = true; 3029 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3030 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3031 } 3032 tcp_req->fused_pair = NULL; 3033 } 3034 3035 nvmf_tcp_req_put(tqpair, tcp_req); 3036 break; 3037 case TCP_REQUEST_NUM_STATES: 3038 default: 3039 assert(0); 3040 break; 3041 } 3042 3043 if (tcp_req->state != prev_state) { 3044 progress = true; 3045 } 3046 } while (tcp_req->state != prev_state); 3047 3048 return progress; 3049 } 3050 3051 static void 3052 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3053 { 3054 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3055 int rc; 3056 3057 assert(tqpair != NULL); 3058 rc = nvmf_tcp_sock_process(tqpair); 3059 3060 /* If there was a new socket error, disconnect */ 3061 if (rc < 0) { 3062 nvmf_tcp_qpair_disconnect(tqpair); 3063 } 3064 } 3065 3066 static int 3067 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3068 struct spdk_nvmf_qpair *qpair) 3069 { 3070 struct spdk_nvmf_tcp_poll_group *tgroup; 3071 struct spdk_nvmf_tcp_qpair *tqpair; 3072 int rc; 3073 3074 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3075 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3076 3077 rc = nvmf_tcp_qpair_sock_init(tqpair); 3078 if (rc != 0) { 3079 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3080 return -1; 3081 } 3082 3083 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3084 if (rc < 0) { 3085 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3086 return -1; 3087 } 3088 3089 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3090 if (rc < 0) { 3091 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3092 return -1; 3093 } 3094 3095 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3096 nvmf_tcp_sock_cb, tqpair); 3097 if (rc != 0) { 3098 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3099 spdk_strerror(errno), errno); 3100 return -1; 3101 } 3102 3103 tqpair->group = tgroup; 3104 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3105 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3106 3107 return 0; 3108 } 3109 3110 static int 3111 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3112 struct spdk_nvmf_qpair *qpair) 3113 { 3114 struct spdk_nvmf_tcp_poll_group *tgroup; 3115 struct spdk_nvmf_tcp_qpair *tqpair; 3116 int rc; 3117 3118 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3119 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3120 3121 assert(tqpair->group == tgroup); 3122 3123 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3124 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3125 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 3126 } else { 3127 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3128 } 3129 3130 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3131 if (rc != 0) { 3132 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3133 spdk_strerror(errno), errno); 3134 } 3135 3136 return rc; 3137 } 3138 3139 static int 3140 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3141 { 3142 struct spdk_nvmf_tcp_transport *ttransport; 3143 struct spdk_nvmf_tcp_req *tcp_req; 3144 3145 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3146 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3147 3148 switch (tcp_req->state) { 3149 case TCP_REQUEST_STATE_EXECUTING: 3150 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3151 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3152 break; 3153 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3154 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3155 break; 3156 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3157 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3158 break; 3159 default: 3160 assert(0 && "Unexpected request state"); 3161 break; 3162 } 3163 3164 nvmf_tcp_req_process(ttransport, tcp_req); 3165 3166 return 0; 3167 } 3168 3169 static void 3170 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3171 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3172 { 3173 struct spdk_nvmf_tcp_qpair *tqpair; 3174 3175 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3176 3177 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3178 3179 assert(tqpair->fini_cb_fn == NULL); 3180 tqpair->fini_cb_fn = cb_fn; 3181 tqpair->fini_cb_arg = cb_arg; 3182 3183 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3184 nvmf_tcp_qpair_destroy(tqpair); 3185 } 3186 3187 static int 3188 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3189 { 3190 struct spdk_nvmf_tcp_poll_group *tgroup; 3191 int rc; 3192 struct spdk_nvmf_request *req, *req_tmp; 3193 struct spdk_nvmf_tcp_req *tcp_req; 3194 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3195 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3196 struct spdk_nvmf_tcp_transport, transport); 3197 3198 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3199 3200 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3201 return 0; 3202 } 3203 3204 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3205 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3206 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3207 break; 3208 } 3209 } 3210 3211 rc = spdk_sock_group_poll(tgroup->sock_group); 3212 if (rc < 0) { 3213 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3214 } 3215 3216 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3217 nvmf_tcp_sock_process(tqpair); 3218 } 3219 3220 return rc; 3221 } 3222 3223 static int 3224 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3225 struct spdk_nvme_transport_id *trid, bool peer) 3226 { 3227 struct spdk_nvmf_tcp_qpair *tqpair; 3228 uint16_t port; 3229 3230 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3231 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3232 3233 if (peer) { 3234 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3235 port = tqpair->initiator_port; 3236 } else { 3237 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3238 port = tqpair->target_port; 3239 } 3240 3241 if (spdk_sock_is_ipv4(tqpair->sock)) { 3242 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3243 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3244 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3245 } else { 3246 return -1; 3247 } 3248 3249 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3250 return 0; 3251 } 3252 3253 static int 3254 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3255 struct spdk_nvme_transport_id *trid) 3256 { 3257 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3258 } 3259 3260 static int 3261 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3262 struct spdk_nvme_transport_id *trid) 3263 { 3264 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3265 } 3266 3267 static int 3268 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3269 struct spdk_nvme_transport_id *trid) 3270 { 3271 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3272 } 3273 3274 static void 3275 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3276 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3277 { 3278 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3279 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3280 tcp_req_to_abort->req.rsp->nvme_cpl.cid = tcp_req_to_abort->req.cmd->nvme_cmd.cid; 3281 3282 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3283 3284 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3285 } 3286 3287 static int 3288 _nvmf_tcp_qpair_abort_request(void *ctx) 3289 { 3290 struct spdk_nvmf_request *req = ctx; 3291 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3292 struct spdk_nvmf_tcp_req, req); 3293 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3294 struct spdk_nvmf_tcp_qpair, qpair); 3295 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3296 struct spdk_nvmf_tcp_transport, transport); 3297 int rc; 3298 3299 spdk_poller_unregister(&req->poller); 3300 3301 switch (tcp_req_to_abort->state) { 3302 case TCP_REQUEST_STATE_EXECUTING: 3303 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3304 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3305 rc = nvmf_ctrlr_abort_request(req); 3306 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3307 return SPDK_POLLER_BUSY; 3308 } 3309 break; 3310 3311 case TCP_REQUEST_STATE_NEED_BUFFER: 3312 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3313 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3314 3315 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3316 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3317 break; 3318 3319 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3320 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3321 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3322 break; 3323 3324 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3325 if (spdk_get_ticks() < req->timeout_tsc) { 3326 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3327 return SPDK_POLLER_BUSY; 3328 } 3329 break; 3330 3331 default: 3332 break; 3333 } 3334 3335 spdk_nvmf_request_complete(req); 3336 return SPDK_POLLER_BUSY; 3337 } 3338 3339 static void 3340 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3341 struct spdk_nvmf_request *req) 3342 { 3343 struct spdk_nvmf_tcp_qpair *tqpair; 3344 struct spdk_nvmf_tcp_transport *ttransport; 3345 struct spdk_nvmf_transport *transport; 3346 uint16_t cid; 3347 uint32_t i; 3348 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3349 3350 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3351 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3352 transport = &ttransport->transport; 3353 3354 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3355 3356 for (i = 0; i < tqpair->resource_count; i++) { 3357 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3358 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3359 tcp_req_to_abort = &tqpair->reqs[i]; 3360 break; 3361 } 3362 } 3363 3364 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, qpair->qid, 0, (uintptr_t)req, tqpair); 3365 3366 if (tcp_req_to_abort == NULL) { 3367 spdk_nvmf_request_complete(req); 3368 return; 3369 } 3370 3371 req->req_to_abort = &tcp_req_to_abort->req; 3372 req->timeout_tsc = spdk_get_ticks() + 3373 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3374 req->poller = NULL; 3375 3376 _nvmf_tcp_qpair_abort_request(req); 3377 } 3378 3379 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 3380 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 3381 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 3382 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 3383 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 3384 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 3385 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 3386 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 3387 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 3388 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 3389 3390 static void 3391 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3392 { 3393 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 3394 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3395 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3396 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3397 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3398 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 3399 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3400 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3401 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3402 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3403 opts->transport_specific = NULL; 3404 } 3405 3406 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3407 .name = "TCP", 3408 .type = SPDK_NVME_TRANSPORT_TCP, 3409 .opts_init = nvmf_tcp_opts_init, 3410 .create = nvmf_tcp_create, 3411 .dump_opts = nvmf_tcp_dump_opts, 3412 .destroy = nvmf_tcp_destroy, 3413 3414 .listen = nvmf_tcp_listen, 3415 .stop_listen = nvmf_tcp_stop_listen, 3416 3417 .listener_discover = nvmf_tcp_discover, 3418 3419 .poll_group_create = nvmf_tcp_poll_group_create, 3420 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3421 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3422 .poll_group_add = nvmf_tcp_poll_group_add, 3423 .poll_group_remove = nvmf_tcp_poll_group_remove, 3424 .poll_group_poll = nvmf_tcp_poll_group_poll, 3425 3426 .req_free = nvmf_tcp_req_free, 3427 .req_complete = nvmf_tcp_req_complete, 3428 3429 .qpair_fini = nvmf_tcp_close_qpair, 3430 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3431 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3432 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3433 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3434 }; 3435 3436 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3437 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3438