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