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 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 54 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 55 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 56 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 57 58 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 59 60 /* spdk nvmf related structure */ 61 enum spdk_nvmf_tcp_req_state { 62 63 /* The request is not currently in use */ 64 TCP_REQUEST_STATE_FREE = 0, 65 66 /* Initial state when request first received */ 67 TCP_REQUEST_STATE_NEW, 68 69 /* The request is queued until a data buffer is available. */ 70 TCP_REQUEST_STATE_NEED_BUFFER, 71 72 /* The request is currently transferring data from the host to the controller. */ 73 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 74 75 /* The request is waiting for the R2T send acknowledgement. */ 76 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 77 78 /* The request is ready to execute at the block device */ 79 TCP_REQUEST_STATE_READY_TO_EXECUTE, 80 81 /* The request is currently executing at the block device */ 82 TCP_REQUEST_STATE_EXECUTING, 83 84 /* The request finished executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTED, 86 87 /* The request is ready to send a completion */ 88 TCP_REQUEST_STATE_READY_TO_COMPLETE, 89 90 /* The request is currently transferring final pdus from the controller to the host. */ 91 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 92 93 /* The request completed and can be marked free. */ 94 TCP_REQUEST_STATE_COMPLETED, 95 96 /* Terminator */ 97 TCP_REQUEST_NUM_STATES, 98 }; 99 100 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 101 "Invalid PDU Header Field", 102 "PDU Sequence Error", 103 "Header Digiest Error", 104 "Data Transfer Out of Range", 105 "R2T Limit Exceeded", 106 "Unsupported parameter", 107 }; 108 109 #define OBJECT_NVMF_TCP_IO 0x80 110 111 #define TRACE_GROUP_NVMF_TCP 0x5 112 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 113 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 115 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 117 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 118 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 119 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 120 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 121 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 122 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 123 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 124 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 132 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 133 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 136 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 139 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_EXECUTING", 142 TRACE_TCP_REQUEST_STATE_EXECUTING, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTED", 145 TRACE_TCP_REQUEST_STATE_EXECUTED, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 148 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 151 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_REQ_COMPLETED", 154 TRACE_TCP_REQUEST_STATE_COMPLETED, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 156 spdk_trace_register_description("TCP_WRITE_START", 157 TRACE_TCP_FLUSH_WRITEBUF_START, 158 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 159 spdk_trace_register_description("TCP_WRITE_DONE", 160 TRACE_TCP_FLUSH_WRITEBUF_DONE, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 spdk_trace_register_description("TCP_READ_DONE", 163 TRACE_TCP_READ_FROM_SOCKET_DONE, 164 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 165 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 166 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 167 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 168 } 169 170 struct spdk_nvmf_tcp_req { 171 struct spdk_nvmf_request req; 172 struct spdk_nvme_cpl rsp; 173 struct spdk_nvme_cmd cmd; 174 175 /* A PDU that can be used for sending responses. This is 176 * not the incoming PDU! */ 177 struct nvme_tcp_pdu *pdu; 178 179 /* In-capsule data buffer */ 180 uint8_t *buf; 181 /* 182 * The PDU for a request may be used multiple times in serial over 183 * the request's lifetime. For example, first to send an R2T, then 184 * to send a completion. To catch mistakes where the PDU is used 185 * twice at the same time, add a debug flag here for init/fini. 186 */ 187 bool pdu_in_use; 188 bool has_incapsule_data; 189 190 /* transfer_tag */ 191 uint16_t ttag; 192 193 enum spdk_nvmf_tcp_req_state state; 194 195 /* 196 * h2c_offset is used when we receive the h2c_data PDU. 197 */ 198 uint32_t h2c_offset; 199 200 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 201 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 202 }; 203 204 struct spdk_nvmf_tcp_qpair { 205 struct spdk_nvmf_qpair qpair; 206 struct spdk_nvmf_tcp_poll_group *group; 207 struct spdk_sock *sock; 208 209 enum nvme_tcp_pdu_recv_state recv_state; 210 enum nvme_tcp_qpair_state state; 211 212 /* PDU being actively received */ 213 struct nvme_tcp_pdu pdu_in_progress; 214 215 /* Queues to track the requests in all states */ 216 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 217 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 218 219 /* Number of requests in each state */ 220 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 221 222 uint8_t cpda; 223 224 bool host_hdgst_enable; 225 bool host_ddgst_enable; 226 227 /* This is a spare PDU used for sending special management 228 * operations. Primarily, this is used for the initial 229 * connection response and c2h termination request. */ 230 struct nvme_tcp_pdu *mgmt_pdu; 231 232 /* Arrays of in-capsule buffers, requests, and pdus. 233 * Each array is 'resource_count' number of elements */ 234 void *bufs; 235 struct spdk_nvmf_tcp_req *reqs; 236 struct nvme_tcp_pdu *pdus; 237 uint32_t resource_count; 238 uint32_t recv_buf_size; 239 240 struct spdk_nvmf_tcp_port *port; 241 242 /* IP address */ 243 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 244 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 245 246 /* IP port */ 247 uint16_t initiator_port; 248 uint16_t target_port; 249 250 /* Timer used to destroy qpair after detecting transport error issue if initiator does 251 * not close the connection. 252 */ 253 struct spdk_poller *timeout_poller; 254 255 256 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 257 }; 258 259 struct spdk_nvmf_tcp_control_msg { 260 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 261 }; 262 263 struct spdk_nvmf_tcp_control_msg_list { 264 void *msg_buf; 265 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 266 }; 267 268 struct spdk_nvmf_tcp_poll_group { 269 struct spdk_nvmf_transport_poll_group group; 270 struct spdk_sock_group *sock_group; 271 272 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 273 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 274 275 struct spdk_io_channel *accel_channel; 276 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 277 }; 278 279 struct spdk_nvmf_tcp_port { 280 const struct spdk_nvme_transport_id *trid; 281 struct spdk_sock *listen_sock; 282 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 283 }; 284 285 struct tcp_transport_opts { 286 bool c2h_success; 287 uint16_t control_msg_num; 288 uint32_t sock_priority; 289 }; 290 291 struct spdk_nvmf_tcp_transport { 292 struct spdk_nvmf_transport transport; 293 struct tcp_transport_opts tcp_opts; 294 295 pthread_mutex_t lock; 296 297 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 298 }; 299 300 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 301 { 302 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 303 spdk_json_decode_bool, true 304 }, 305 { 306 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 307 spdk_json_decode_uint16, true 308 }, 309 { 310 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 311 spdk_json_decode_uint32, true 312 }, 313 }; 314 315 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 316 struct spdk_nvmf_tcp_req *tcp_req); 317 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 318 319 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 320 struct spdk_nvmf_tcp_req *tcp_req); 321 322 static void 323 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 324 enum spdk_nvmf_tcp_req_state state) 325 { 326 struct spdk_nvmf_qpair *qpair; 327 struct spdk_nvmf_tcp_qpair *tqpair; 328 329 qpair = tcp_req->req.qpair; 330 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 331 332 assert(tqpair->state_cntr[tcp_req->state] > 0); 333 tqpair->state_cntr[tcp_req->state]--; 334 tqpair->state_cntr[state]++; 335 336 tcp_req->state = state; 337 } 338 339 static inline struct nvme_tcp_pdu * 340 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 341 { 342 assert(tcp_req->pdu_in_use == false); 343 tcp_req->pdu_in_use = true; 344 345 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 346 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 347 348 return tcp_req->pdu; 349 } 350 351 static inline void 352 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 353 { 354 tcp_req->pdu_in_use = false; 355 } 356 357 static struct spdk_nvmf_tcp_req * 358 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 359 { 360 struct spdk_nvmf_tcp_req *tcp_req; 361 362 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 363 if (!tcp_req) { 364 return NULL; 365 } 366 367 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 368 tcp_req->h2c_offset = 0; 369 tcp_req->has_incapsule_data = false; 370 tcp_req->req.dif.dif_insert_or_strip = false; 371 372 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 373 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 374 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 375 return tcp_req; 376 } 377 378 static inline void 379 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 380 { 381 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 382 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 384 } 385 386 static void 387 nvmf_tcp_request_free(void *cb_arg) 388 { 389 struct spdk_nvmf_tcp_transport *ttransport; 390 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 391 392 assert(tcp_req != NULL); 393 394 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 395 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 396 struct spdk_nvmf_tcp_transport, transport); 397 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 398 nvmf_tcp_req_process(ttransport, tcp_req); 399 } 400 401 static int 402 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 403 { 404 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 405 406 nvmf_tcp_request_free(tcp_req); 407 408 return 0; 409 } 410 411 static void 412 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 413 enum spdk_nvmf_tcp_req_state state) 414 { 415 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 416 417 assert(state != TCP_REQUEST_STATE_FREE); 418 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 419 if (state == tcp_req->state) { 420 nvmf_tcp_request_free(tcp_req); 421 } 422 } 423 } 424 425 static void 426 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 427 { 428 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 429 430 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 431 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 432 433 /* Wipe the requests waiting for buffer from the global list */ 434 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 435 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 436 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 437 spdk_nvmf_request, buf_link); 438 } 439 } 440 441 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 442 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 443 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 444 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 445 } 446 447 static void 448 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 449 { 450 int i; 451 struct spdk_nvmf_tcp_req *tcp_req; 452 453 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 454 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 455 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 456 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 457 if ((int)tcp_req->state == i) { 458 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 459 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 460 } 461 } 462 } 463 } 464 465 static void 466 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 467 { 468 int err = 0; 469 470 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 471 472 err = spdk_sock_close(&tqpair->sock); 473 assert(err == 0); 474 nvmf_tcp_cleanup_all_states(tqpair); 475 476 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 477 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 478 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 479 tqpair->resource_count); 480 err++; 481 } 482 483 if (err > 0) { 484 nvmf_tcp_dump_qpair_req_contents(tqpair); 485 } 486 487 spdk_dma_free(tqpair->pdus); 488 free(tqpair->reqs); 489 spdk_free(tqpair->bufs); 490 free(tqpair); 491 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 492 } 493 494 static void 495 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 496 { 497 struct spdk_nvmf_tcp_transport *ttransport; 498 assert(w != NULL); 499 500 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 501 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 502 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 503 } 504 505 static int 506 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 507 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 508 { 509 struct spdk_nvmf_tcp_transport *ttransport; 510 511 assert(transport != NULL); 512 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 513 514 pthread_mutex_destroy(&ttransport->lock); 515 free(ttransport); 516 517 if (cb_fn) { 518 cb_fn(cb_arg); 519 } 520 return 0; 521 } 522 523 static struct spdk_nvmf_transport * 524 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 525 { 526 struct spdk_nvmf_tcp_transport *ttransport; 527 uint32_t sge_count; 528 uint32_t min_shared_buffers; 529 530 ttransport = calloc(1, sizeof(*ttransport)); 531 if (!ttransport) { 532 return NULL; 533 } 534 535 TAILQ_INIT(&ttransport->ports); 536 537 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 538 539 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 540 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 541 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 542 if (opts->transport_specific != NULL && 543 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 544 SPDK_COUNTOF(tcp_transport_opts_decoder), 545 &ttransport->tcp_opts)) { 546 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 547 free(ttransport); 548 return NULL; 549 } 550 551 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 552 553 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 554 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 555 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 556 " in_capsule_data_size=%d, max_aq_depth=%d\n" 557 " num_shared_buffers=%d, c2h_success=%d,\n" 558 " dif_insert_or_strip=%d, sock_priority=%d\n" 559 " abort_timeout_sec=%d, control_msg_num=%hu\n", 560 opts->max_queue_depth, 561 opts->max_io_size, 562 opts->max_qpairs_per_ctrlr - 1, 563 opts->io_unit_size, 564 opts->in_capsule_data_size, 565 opts->max_aq_depth, 566 opts->num_shared_buffers, 567 ttransport->tcp_opts.c2h_success, 568 opts->dif_insert_or_strip, 569 ttransport->tcp_opts.sock_priority, 570 opts->abort_timeout_sec, 571 ttransport->tcp_opts.control_msg_num); 572 573 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 574 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 575 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 576 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 577 free(ttransport); 578 return NULL; 579 } 580 581 if (ttransport->tcp_opts.control_msg_num == 0 && 582 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 583 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 584 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 585 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 586 } 587 588 /* I/O unit size cannot be larger than max I/O size */ 589 if (opts->io_unit_size > opts->max_io_size) { 590 opts->io_unit_size = opts->max_io_size; 591 } 592 593 sge_count = opts->max_io_size / opts->io_unit_size; 594 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 595 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 596 free(ttransport); 597 return NULL; 598 } 599 600 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 601 if (min_shared_buffers > opts->num_shared_buffers) { 602 SPDK_ERRLOG("There are not enough buffers to satisfy" 603 "per-poll group caches for each thread. (%" PRIu32 ")" 604 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 605 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 606 free(ttransport); 607 return NULL; 608 } 609 610 pthread_mutex_init(&ttransport->lock, NULL); 611 612 return &ttransport->transport; 613 } 614 615 static int 616 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 617 { 618 unsigned long long ull; 619 char *end = NULL; 620 621 ull = strtoull(trsvcid, &end, 10); 622 if (end == NULL || end == trsvcid || *end != '\0') { 623 return -1; 624 } 625 626 /* Valid TCP/IP port numbers are in [0, 65535] */ 627 if (ull > 65535) { 628 return -1; 629 } 630 631 return (int)ull; 632 } 633 634 /** 635 * Canonicalize a listen address trid. 636 */ 637 static int 638 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 639 const struct spdk_nvme_transport_id *trid) 640 { 641 int trsvcid_int; 642 643 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 644 if (trsvcid_int < 0) { 645 return -EINVAL; 646 } 647 648 memset(canon_trid, 0, sizeof(*canon_trid)); 649 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 650 canon_trid->adrfam = trid->adrfam; 651 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 652 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 653 654 return 0; 655 } 656 657 /** 658 * Find an existing listening port. 659 * 660 * Caller must hold ttransport->lock. 661 */ 662 static struct spdk_nvmf_tcp_port * 663 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 664 const struct spdk_nvme_transport_id *trid) 665 { 666 struct spdk_nvme_transport_id canon_trid; 667 struct spdk_nvmf_tcp_port *port; 668 669 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 670 return NULL; 671 } 672 673 TAILQ_FOREACH(port, &ttransport->ports, link) { 674 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 675 return port; 676 } 677 } 678 679 return NULL; 680 } 681 682 static int 683 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 684 struct spdk_nvmf_listen_opts *listen_opts) 685 { 686 struct spdk_nvmf_tcp_transport *ttransport; 687 struct spdk_nvmf_tcp_port *port; 688 int trsvcid_int; 689 uint8_t adrfam; 690 struct spdk_sock_opts opts; 691 692 if (!strlen(trid->trsvcid)) { 693 SPDK_ERRLOG("Service id is required\n"); 694 return -EINVAL; 695 } 696 697 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 698 699 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 700 if (trsvcid_int < 0) { 701 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 702 return -EINVAL; 703 } 704 705 pthread_mutex_lock(&ttransport->lock); 706 port = calloc(1, sizeof(*port)); 707 if (!port) { 708 SPDK_ERRLOG("Port allocation failed\n"); 709 pthread_mutex_unlock(&ttransport->lock); 710 return -ENOMEM; 711 } 712 713 port->trid = trid; 714 opts.opts_size = sizeof(opts); 715 spdk_sock_get_default_opts(&opts); 716 opts.priority = ttransport->tcp_opts.sock_priority; 717 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 718 NULL, &opts); 719 if (port->listen_sock == NULL) { 720 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 721 trid->traddr, trsvcid_int, 722 spdk_strerror(errno), errno); 723 free(port); 724 pthread_mutex_unlock(&ttransport->lock); 725 return -errno; 726 } 727 728 if (spdk_sock_is_ipv4(port->listen_sock)) { 729 adrfam = SPDK_NVMF_ADRFAM_IPV4; 730 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 731 adrfam = SPDK_NVMF_ADRFAM_IPV6; 732 } else { 733 SPDK_ERRLOG("Unhandled socket type\n"); 734 adrfam = 0; 735 } 736 737 if (adrfam != trid->adrfam) { 738 SPDK_ERRLOG("Socket address family mismatch\n"); 739 spdk_sock_close(&port->listen_sock); 740 free(port); 741 pthread_mutex_unlock(&ttransport->lock); 742 return -EINVAL; 743 } 744 745 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 746 trid->traddr, trid->trsvcid); 747 748 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 749 pthread_mutex_unlock(&ttransport->lock); 750 return 0; 751 } 752 753 static void 754 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 755 const struct spdk_nvme_transport_id *trid) 756 { 757 struct spdk_nvmf_tcp_transport *ttransport; 758 struct spdk_nvmf_tcp_port *port; 759 760 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 761 762 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 763 trid->traddr, trid->trsvcid); 764 765 pthread_mutex_lock(&ttransport->lock); 766 port = nvmf_tcp_find_port(ttransport, trid); 767 if (port) { 768 TAILQ_REMOVE(&ttransport->ports, port, link); 769 spdk_sock_close(&port->listen_sock); 770 free(port); 771 } 772 773 pthread_mutex_unlock(&ttransport->lock); 774 } 775 776 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 777 enum nvme_tcp_pdu_recv_state state); 778 779 static void 780 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 781 { 782 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 783 784 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 785 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 786 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 787 spdk_poller_unregister(&tqpair->timeout_poller); 788 789 /* This will end up calling nvmf_tcp_close_qpair */ 790 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 791 } 792 } 793 794 static void 795 _pdu_write_done(void *_pdu, int err) 796 { 797 struct nvme_tcp_pdu *pdu = _pdu; 798 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 799 800 if (err != 0) { 801 nvmf_tcp_qpair_disconnect(tqpair); 802 return; 803 } 804 805 assert(pdu->cb_fn != NULL); 806 pdu->cb_fn(pdu->cb_arg); 807 } 808 809 static void 810 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 811 { 812 uint32_t mapped_length = 0; 813 ssize_t rc; 814 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 815 816 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 817 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 818 &mapped_length); 819 pdu->sock_req.cb_fn = _pdu_write_done; 820 pdu->sock_req.cb_arg = pdu; 821 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 822 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 823 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 824 if (rc == mapped_length) { 825 _pdu_write_done(pdu, 0); 826 } else { 827 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 828 _pdu_write_done(pdu, -1); 829 } 830 } else { 831 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 832 } 833 } 834 835 static void 836 data_crc32_accel_done(void *cb_arg, int status) 837 { 838 struct nvme_tcp_pdu *pdu = cb_arg; 839 840 if (spdk_unlikely(status)) { 841 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 842 _pdu_write_done(pdu, status); 843 return; 844 } 845 846 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 847 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 848 849 _tcp_write_pdu(pdu); 850 } 851 852 static void 853 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 854 { 855 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 856 uint32_t crc32c; 857 858 /* Data Digest */ 859 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 860 /* Only suport this limitated case for the first step */ 861 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 862 && tqpair->group)) { 863 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, 864 pdu->data_iov, pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 865 return; 866 } 867 868 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 869 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 870 } 871 872 _tcp_write_pdu(pdu); 873 } 874 875 static void 876 header_crc32_accel_done(void *cb_arg, int status) 877 { 878 struct nvme_tcp_pdu *pdu = cb_arg; 879 880 pdu->header_digest_crc32 ^= SPDK_CRC32C_XOR; 881 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, pdu->header_digest_crc32); 882 if (spdk_unlikely(status)) { 883 SPDK_ERRLOG("Failed to compute header digest on pdu=%p\n", pdu); 884 _pdu_write_done(pdu, status); 885 return; 886 } 887 888 pdu_data_crc32_compute(pdu); 889 } 890 891 static void 892 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 893 struct nvme_tcp_pdu *pdu, 894 nvme_tcp_qpair_xfer_complete_cb cb_fn, 895 void *cb_arg) 896 { 897 int hlen; 898 899 assert(&tqpair->pdu_in_progress != pdu); 900 901 hlen = pdu->hdr.common.hlen; 902 pdu->cb_fn = cb_fn; 903 pdu->cb_arg = cb_arg; 904 905 pdu->iov[0].iov_base = &pdu->hdr.raw; 906 pdu->iov[0].iov_len = hlen; 907 908 /* Header Digest */ 909 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable && tqpair->group) { 910 spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->header_digest_crc32, 911 pdu->iov, 1, 0, header_crc32_accel_done, pdu); 912 return; 913 } 914 915 pdu_data_crc32_compute(pdu); 916 } 917 918 static int 919 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 920 { 921 uint32_t i; 922 struct spdk_nvmf_transport_opts *opts; 923 uint32_t in_capsule_data_size; 924 925 opts = &tqpair->qpair.transport->opts; 926 927 in_capsule_data_size = opts->in_capsule_data_size; 928 if (opts->dif_insert_or_strip) { 929 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 930 } 931 932 tqpair->resource_count = opts->max_queue_depth; 933 934 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 935 if (!tqpair->reqs) { 936 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 937 return -1; 938 } 939 940 if (in_capsule_data_size) { 941 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 942 NULL, SPDK_ENV_LCORE_ID_ANY, 943 SPDK_MALLOC_DMA); 944 if (!tqpair->bufs) { 945 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 946 return -1; 947 } 948 } 949 950 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 951 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 952 if (!tqpair->pdus) { 953 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 954 return -1; 955 } 956 957 for (i = 0; i < tqpair->resource_count; i++) { 958 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 959 960 tcp_req->ttag = i + 1; 961 tcp_req->req.qpair = &tqpair->qpair; 962 963 tcp_req->pdu = &tqpair->pdus[i]; 964 tcp_req->pdu->qpair = tqpair; 965 966 /* Set up memory to receive commands */ 967 if (tqpair->bufs) { 968 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 969 } 970 971 /* Set the cmdn and rsp */ 972 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 973 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 974 975 /* Initialize request state to FREE */ 976 tcp_req->state = TCP_REQUEST_STATE_FREE; 977 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 978 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 979 } 980 981 tqpair->mgmt_pdu = &tqpair->pdus[i]; 982 tqpair->mgmt_pdu->qpair = tqpair; 983 984 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 985 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 986 987 return 0; 988 } 989 990 static int 991 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 992 { 993 struct spdk_nvmf_tcp_qpair *tqpair; 994 995 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 996 997 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 998 999 /* Initialise request state queues of the qpair */ 1000 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1001 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1002 1003 tqpair->host_hdgst_enable = true; 1004 tqpair->host_ddgst_enable = true; 1005 1006 return 0; 1007 } 1008 1009 static int 1010 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1011 { 1012 int rc; 1013 1014 /* set low water mark */ 1015 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 1016 if (rc != 0) { 1017 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1018 return rc; 1019 } 1020 1021 return 0; 1022 } 1023 1024 static void 1025 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1026 struct spdk_nvmf_tcp_port *port, 1027 struct spdk_sock *sock) 1028 { 1029 struct spdk_nvmf_tcp_qpair *tqpair; 1030 int rc; 1031 1032 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1033 port->trid->traddr, port->trid->trsvcid); 1034 1035 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1036 if (tqpair == NULL) { 1037 SPDK_ERRLOG("Could not allocate new connection.\n"); 1038 spdk_sock_close(&sock); 1039 return; 1040 } 1041 1042 tqpair->sock = sock; 1043 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1044 tqpair->port = port; 1045 tqpair->qpair.transport = transport; 1046 1047 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1048 sizeof(tqpair->target_addr), &tqpair->target_port, 1049 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1050 &tqpair->initiator_port); 1051 if (rc < 0) { 1052 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1053 nvmf_tcp_qpair_destroy(tqpair); 1054 return; 1055 } 1056 1057 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1058 } 1059 1060 static uint32_t 1061 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1062 { 1063 struct spdk_sock *sock; 1064 uint32_t count = 0; 1065 int i; 1066 1067 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1068 sock = spdk_sock_accept(port->listen_sock); 1069 if (sock == NULL) { 1070 break; 1071 } 1072 count++; 1073 nvmf_tcp_handle_connect(transport, port, sock); 1074 } 1075 1076 return count; 1077 } 1078 1079 static uint32_t 1080 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1081 { 1082 struct spdk_nvmf_tcp_transport *ttransport; 1083 struct spdk_nvmf_tcp_port *port; 1084 uint32_t count = 0; 1085 1086 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1087 1088 TAILQ_FOREACH(port, &ttransport->ports, link) { 1089 count += nvmf_tcp_port_accept(transport, port); 1090 } 1091 1092 return count; 1093 } 1094 1095 static void 1096 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1097 struct spdk_nvme_transport_id *trid, 1098 struct spdk_nvmf_discovery_log_page_entry *entry) 1099 { 1100 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1101 entry->adrfam = trid->adrfam; 1102 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1103 1104 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1105 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1106 1107 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1108 } 1109 1110 static struct spdk_nvmf_tcp_control_msg_list * 1111 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1112 { 1113 struct spdk_nvmf_tcp_control_msg_list *list; 1114 struct spdk_nvmf_tcp_control_msg *msg; 1115 uint16_t i; 1116 1117 list = calloc(1, sizeof(*list)); 1118 if (!list) { 1119 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1120 return NULL; 1121 } 1122 1123 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1124 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1125 if (!list->msg_buf) { 1126 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1127 free(list); 1128 return NULL; 1129 } 1130 1131 STAILQ_INIT(&list->free_msgs); 1132 1133 for (i = 0; i < num_messages; i++) { 1134 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1135 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1136 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1137 } 1138 1139 return list; 1140 } 1141 1142 static void 1143 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1144 { 1145 if (!list) { 1146 return; 1147 } 1148 1149 spdk_free(list->msg_buf); 1150 free(list); 1151 } 1152 1153 static struct spdk_nvmf_transport_poll_group * 1154 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1155 { 1156 struct spdk_nvmf_tcp_transport *ttransport; 1157 struct spdk_nvmf_tcp_poll_group *tgroup; 1158 1159 tgroup = calloc(1, sizeof(*tgroup)); 1160 if (!tgroup) { 1161 return NULL; 1162 } 1163 1164 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1165 if (!tgroup->sock_group) { 1166 goto cleanup; 1167 } 1168 1169 TAILQ_INIT(&tgroup->qpairs); 1170 TAILQ_INIT(&tgroup->await_req); 1171 1172 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1173 1174 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1175 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1176 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1177 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1178 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1179 if (!tgroup->control_msg_list) { 1180 goto cleanup; 1181 } 1182 } 1183 1184 tgroup->accel_channel = spdk_accel_engine_get_io_channel(); 1185 if (spdk_unlikely(!tgroup->accel_channel)) { 1186 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1187 goto cleanup; 1188 } 1189 1190 return &tgroup->group; 1191 1192 cleanup: 1193 nvmf_tcp_poll_group_destroy(&tgroup->group); 1194 return NULL; 1195 } 1196 1197 static struct spdk_nvmf_transport_poll_group * 1198 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1199 { 1200 struct spdk_nvmf_tcp_qpair *tqpair; 1201 struct spdk_sock_group *group = NULL; 1202 int rc; 1203 1204 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1205 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1206 if (!rc && group != NULL) { 1207 return spdk_sock_group_get_ctx(group); 1208 } 1209 1210 return NULL; 1211 } 1212 1213 static void 1214 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1215 { 1216 struct spdk_nvmf_tcp_poll_group *tgroup; 1217 1218 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1219 spdk_sock_group_close(&tgroup->sock_group); 1220 if (tgroup->control_msg_list) { 1221 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1222 } 1223 1224 if (tgroup->accel_channel) { 1225 spdk_put_io_channel(tgroup->accel_channel); 1226 } 1227 1228 free(tgroup); 1229 } 1230 1231 static void 1232 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1233 enum nvme_tcp_pdu_recv_state state) 1234 { 1235 if (tqpair->recv_state == state) { 1236 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1237 tqpair, state); 1238 return; 1239 } 1240 1241 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1242 /* When leaving the await req state, move the qpair to the main list */ 1243 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1244 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1245 } 1246 1247 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1248 tqpair->recv_state = state; 1249 1250 switch (state) { 1251 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1252 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1253 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1254 break; 1255 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1256 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1257 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1258 break; 1259 case NVME_TCP_PDU_RECV_STATE_ERROR: 1260 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1261 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1262 break; 1263 default: 1264 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1265 abort(); 1266 break; 1267 } 1268 } 1269 1270 static int 1271 nvmf_tcp_qpair_handle_timeout(void *ctx) 1272 { 1273 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1274 1275 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1276 1277 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1278 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1279 1280 nvmf_tcp_qpair_disconnect(tqpair); 1281 return SPDK_POLLER_BUSY; 1282 } 1283 1284 static void 1285 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1286 { 1287 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1288 1289 if (!tqpair->timeout_poller) { 1290 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1291 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1292 } 1293 } 1294 1295 static void 1296 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1297 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1298 { 1299 struct nvme_tcp_pdu *rsp_pdu; 1300 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1301 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1302 uint32_t copy_len; 1303 1304 rsp_pdu = tqpair->mgmt_pdu; 1305 1306 c2h_term_req = &rsp_pdu->hdr.term_req; 1307 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1308 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1309 1310 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1311 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1312 DSET32(&c2h_term_req->fei, error_offset); 1313 } 1314 1315 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1316 1317 /* Copy the error info into the buffer */ 1318 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1319 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1320 1321 /* Contain the header of the wrong received pdu */ 1322 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1323 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1324 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1325 } 1326 1327 static void 1328 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1329 struct spdk_nvmf_tcp_qpair *tqpair, 1330 struct nvme_tcp_pdu *pdu) 1331 { 1332 struct spdk_nvmf_tcp_req *tcp_req; 1333 1334 assert(pdu->psh_valid_bytes == pdu->psh_len); 1335 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1336 1337 tcp_req = nvmf_tcp_req_get(tqpair); 1338 if (!tcp_req) { 1339 /* Directly return and make the allocation retry again */ 1340 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1341 return; 1342 } 1343 1344 /* The host sent more commands than the maximum queue depth. */ 1345 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1346 nvmf_tcp_qpair_disconnect(tqpair); 1347 return; 1348 } 1349 1350 pdu->req = tcp_req; 1351 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1352 nvmf_tcp_req_process(ttransport, tcp_req); 1353 } 1354 1355 static void 1356 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1357 struct spdk_nvmf_tcp_qpair *tqpair, 1358 struct nvme_tcp_pdu *pdu) 1359 { 1360 struct spdk_nvmf_tcp_req *tcp_req; 1361 struct spdk_nvme_tcp_cmd *capsule_cmd; 1362 uint32_t error_offset = 0; 1363 enum spdk_nvme_tcp_term_req_fes fes; 1364 1365 capsule_cmd = &pdu->hdr.capsule_cmd; 1366 tcp_req = pdu->req; 1367 assert(tcp_req != NULL); 1368 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1369 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1370 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1371 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1372 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1373 goto err; 1374 } 1375 1376 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1377 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1378 nvmf_tcp_req_process(ttransport, tcp_req); 1379 1380 return; 1381 err: 1382 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1383 } 1384 1385 static int 1386 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1387 enum spdk_nvmf_tcp_req_state state, 1388 uint16_t cid, uint16_t tag, 1389 struct spdk_nvmf_tcp_req **req) 1390 { 1391 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1392 1393 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1394 if (tcp_req->state != state) { 1395 continue; 1396 } 1397 1398 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1399 continue; 1400 } 1401 1402 if (tcp_req->ttag == tag) { 1403 *req = tcp_req; 1404 return 0; 1405 } 1406 1407 *req = NULL; 1408 return -1; 1409 } 1410 1411 /* Didn't find it, but not an error */ 1412 *req = NULL; 1413 return 0; 1414 } 1415 1416 static void 1417 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1418 struct spdk_nvmf_tcp_qpair *tqpair, 1419 struct nvme_tcp_pdu *pdu) 1420 { 1421 struct spdk_nvmf_tcp_req *tcp_req; 1422 uint32_t error_offset = 0; 1423 enum spdk_nvme_tcp_term_req_fes fes = 0; 1424 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1425 int rc; 1426 1427 h2c_data = &pdu->hdr.h2c_data; 1428 1429 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1430 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1431 1432 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1433 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1434 if (rc == 0 && tcp_req == NULL) { 1435 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1436 h2c_data->ttag, &tcp_req); 1437 } 1438 1439 if (!tcp_req) { 1440 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1441 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1442 if (rc == 0) { 1443 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1444 } else { 1445 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1446 } 1447 goto err; 1448 } 1449 1450 if (tcp_req->h2c_offset != h2c_data->datao) { 1451 SPDK_DEBUGLOG(nvmf_tcp, 1452 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1453 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1454 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1455 goto err; 1456 } 1457 1458 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1459 SPDK_DEBUGLOG(nvmf_tcp, 1460 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1461 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1462 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1463 goto err; 1464 } 1465 1466 pdu->req = tcp_req; 1467 1468 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1469 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1470 } 1471 1472 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1473 h2c_data->datao, h2c_data->datal); 1474 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1475 return; 1476 1477 err: 1478 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1479 } 1480 1481 static void 1482 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1483 struct spdk_nvmf_tcp_qpair *tqpair) 1484 { 1485 struct nvme_tcp_pdu *rsp_pdu; 1486 struct spdk_nvme_tcp_rsp *capsule_resp; 1487 1488 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1489 1490 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1491 assert(rsp_pdu != NULL); 1492 1493 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1494 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1495 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1496 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1497 if (tqpair->host_hdgst_enable) { 1498 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1499 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1500 } 1501 1502 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1503 } 1504 1505 static void 1506 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1507 { 1508 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1509 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1510 struct spdk_nvmf_tcp_qpair, qpair); 1511 1512 assert(tqpair != NULL); 1513 1514 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1515 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1516 tcp_req->req.length); 1517 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1518 return; 1519 } 1520 1521 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1522 nvmf_tcp_request_free(tcp_req); 1523 } else { 1524 nvmf_tcp_req_pdu_fini(tcp_req); 1525 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1526 } 1527 } 1528 1529 static void 1530 nvmf_tcp_r2t_complete(void *cb_arg) 1531 { 1532 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1533 struct spdk_nvmf_tcp_transport *ttransport; 1534 1535 nvmf_tcp_req_pdu_fini(tcp_req); 1536 1537 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1538 struct spdk_nvmf_tcp_transport, transport); 1539 1540 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1541 1542 if (tcp_req->h2c_offset == tcp_req->req.length) { 1543 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1544 nvmf_tcp_req_process(ttransport, tcp_req); 1545 } 1546 } 1547 1548 static void 1549 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1550 struct spdk_nvmf_tcp_req *tcp_req) 1551 { 1552 struct nvme_tcp_pdu *rsp_pdu; 1553 struct spdk_nvme_tcp_r2t_hdr *r2t; 1554 1555 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1556 assert(rsp_pdu != NULL); 1557 1558 r2t = &rsp_pdu->hdr.r2t; 1559 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1560 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1561 1562 if (tqpair->host_hdgst_enable) { 1563 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1564 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1565 } 1566 1567 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1568 r2t->ttag = tcp_req->ttag; 1569 r2t->r2to = tcp_req->h2c_offset; 1570 r2t->r2tl = tcp_req->req.length; 1571 1572 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1573 1574 SPDK_DEBUGLOG(nvmf_tcp, 1575 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1576 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1577 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1578 } 1579 1580 static void 1581 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1582 struct spdk_nvmf_tcp_qpair *tqpair, 1583 struct nvme_tcp_pdu *pdu) 1584 { 1585 struct spdk_nvmf_tcp_req *tcp_req; 1586 1587 tcp_req = pdu->req; 1588 assert(tcp_req != NULL); 1589 1590 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1591 1592 tcp_req->h2c_offset += pdu->data_len; 1593 1594 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1595 1596 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1597 * acknowledged before moving on. */ 1598 if (tcp_req->h2c_offset == tcp_req->req.length && 1599 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1600 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1601 nvmf_tcp_req_process(ttransport, tcp_req); 1602 } 1603 } 1604 1605 static void 1606 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1607 { 1608 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1609 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1610 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1611 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1612 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1613 DGET32(h2c_term_req->fei)); 1614 } 1615 } 1616 1617 static void 1618 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1619 struct nvme_tcp_pdu *pdu) 1620 { 1621 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1622 uint32_t error_offset = 0; 1623 enum spdk_nvme_tcp_term_req_fes fes; 1624 1625 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1626 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1627 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1628 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1629 goto end; 1630 } 1631 1632 /* set the data buffer */ 1633 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1634 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1635 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1636 return; 1637 end: 1638 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1639 } 1640 1641 static void 1642 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1643 struct nvme_tcp_pdu *pdu) 1644 { 1645 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1646 1647 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1648 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1649 } 1650 1651 static void 1652 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1653 struct spdk_nvmf_tcp_transport *ttransport) 1654 { 1655 int rc = 0; 1656 struct nvme_tcp_pdu *pdu; 1657 uint32_t crc32c, error_offset = 0; 1658 enum spdk_nvme_tcp_term_req_fes fes; 1659 1660 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1661 pdu = &tqpair->pdu_in_progress; 1662 1663 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1664 /* check data digest if need */ 1665 if (pdu->ddgst_enable) { 1666 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1667 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1668 if (rc == 0) { 1669 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1670 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1671 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1672 return; 1673 1674 } 1675 } 1676 1677 switch (pdu->hdr.common.pdu_type) { 1678 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1679 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1680 break; 1681 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1682 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1683 break; 1684 1685 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1686 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1687 break; 1688 1689 default: 1690 /* The code should not go to here */ 1691 SPDK_ERRLOG("The code should not go to here\n"); 1692 break; 1693 } 1694 } 1695 1696 static void 1697 nvmf_tcp_send_icresp_complete(void *cb_arg) 1698 { 1699 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1700 1701 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1702 } 1703 1704 static void 1705 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1706 struct spdk_nvmf_tcp_qpair *tqpair, 1707 struct nvme_tcp_pdu *pdu) 1708 { 1709 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1710 struct nvme_tcp_pdu *rsp_pdu; 1711 struct spdk_nvme_tcp_ic_resp *ic_resp; 1712 uint32_t error_offset = 0; 1713 enum spdk_nvme_tcp_term_req_fes fes; 1714 1715 /* Only PFV 0 is defined currently */ 1716 if (ic_req->pfv != 0) { 1717 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1718 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1719 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1720 goto end; 1721 } 1722 1723 /* MAXR2T is 0's based */ 1724 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1725 1726 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1727 if (!tqpair->host_hdgst_enable) { 1728 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1729 } 1730 1731 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1732 if (!tqpair->host_ddgst_enable) { 1733 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1734 } 1735 1736 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1737 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1738 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1739 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1740 tqpair, 1741 tqpair->recv_buf_size); 1742 /* Not fatal. */ 1743 } 1744 1745 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1746 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1747 1748 rsp_pdu = tqpair->mgmt_pdu; 1749 1750 ic_resp = &rsp_pdu->hdr.ic_resp; 1751 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1752 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1753 ic_resp->pfv = 0; 1754 ic_resp->cpda = tqpair->cpda; 1755 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1756 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1757 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1758 1759 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1760 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1761 1762 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1763 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1764 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1765 return; 1766 end: 1767 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1768 } 1769 1770 static void 1771 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1772 struct spdk_nvmf_tcp_transport *ttransport) 1773 { 1774 struct nvme_tcp_pdu *pdu; 1775 int rc; 1776 uint32_t crc32c, error_offset = 0; 1777 enum spdk_nvme_tcp_term_req_fes fes; 1778 1779 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1780 pdu = &tqpair->pdu_in_progress; 1781 1782 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1783 pdu->hdr.common.pdu_type); 1784 /* check header digest if needed */ 1785 if (pdu->has_hdgst) { 1786 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1787 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1788 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1789 if (rc == 0) { 1790 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1791 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1792 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1793 return; 1794 1795 } 1796 } 1797 1798 switch (pdu->hdr.common.pdu_type) { 1799 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1800 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1801 break; 1802 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1803 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1804 break; 1805 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1806 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1807 break; 1808 1809 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1810 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1811 break; 1812 1813 default: 1814 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1815 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1816 error_offset = 1; 1817 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1818 break; 1819 } 1820 } 1821 1822 static void 1823 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1824 { 1825 struct nvme_tcp_pdu *pdu; 1826 uint32_t error_offset = 0; 1827 enum spdk_nvme_tcp_term_req_fes fes; 1828 uint8_t expected_hlen, pdo; 1829 bool plen_error = false, pdo_error = false; 1830 1831 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1832 pdu = &tqpair->pdu_in_progress; 1833 1834 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1835 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1836 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1837 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1838 goto err; 1839 } 1840 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1841 if (pdu->hdr.common.plen != expected_hlen) { 1842 plen_error = true; 1843 } 1844 } else { 1845 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1846 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1847 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1848 goto err; 1849 } 1850 1851 switch (pdu->hdr.common.pdu_type) { 1852 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1853 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1854 pdo = pdu->hdr.common.pdo; 1855 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1856 pdo_error = true; 1857 break; 1858 } 1859 1860 if (pdu->hdr.common.plen < expected_hlen) { 1861 plen_error = true; 1862 } 1863 break; 1864 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1865 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1866 pdo = pdu->hdr.common.pdo; 1867 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 1868 pdo_error = true; 1869 break; 1870 } 1871 if (pdu->hdr.common.plen < expected_hlen) { 1872 plen_error = true; 1873 } 1874 break; 1875 1876 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1877 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1878 if ((pdu->hdr.common.plen <= expected_hlen) || 1879 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1880 plen_error = true; 1881 } 1882 break; 1883 1884 default: 1885 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1886 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1887 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1888 goto err; 1889 } 1890 } 1891 1892 if (pdu->hdr.common.hlen != expected_hlen) { 1893 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1894 pdu->hdr.common.pdu_type, 1895 expected_hlen, pdu->hdr.common.hlen, tqpair); 1896 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1897 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1898 goto err; 1899 } else if (pdo_error) { 1900 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1901 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1902 } else if (plen_error) { 1903 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1904 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1905 goto err; 1906 } else { 1907 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1908 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1909 return; 1910 } 1911 err: 1912 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1913 } 1914 1915 static int 1916 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1917 int read_len) 1918 { 1919 int rc; 1920 1921 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1922 read_offset, read_len, pdu->dif_ctx); 1923 if (rc != 0) { 1924 SPDK_ERRLOG("DIF generate failed\n"); 1925 } 1926 1927 return rc; 1928 } 1929 1930 static int 1931 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1932 { 1933 int rc = 0; 1934 struct nvme_tcp_pdu *pdu; 1935 enum nvme_tcp_pdu_recv_state prev_state; 1936 uint32_t data_len; 1937 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1938 struct spdk_nvmf_tcp_transport, transport); 1939 1940 /* The loop here is to allow for several back-to-back state changes. */ 1941 do { 1942 prev_state = tqpair->recv_state; 1943 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1944 1945 pdu = &tqpair->pdu_in_progress; 1946 switch (tqpair->recv_state) { 1947 /* Wait for the common header */ 1948 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1949 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1950 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1951 return rc; 1952 } 1953 1954 rc = nvme_tcp_read_data(tqpair->sock, 1955 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1956 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1957 if (rc < 0) { 1958 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1959 return NVME_TCP_PDU_FATAL; 1960 } else if (rc > 0) { 1961 pdu->ch_valid_bytes += rc; 1962 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1963 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1964 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1965 } 1966 } 1967 1968 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1969 return NVME_TCP_PDU_IN_PROGRESS; 1970 } 1971 1972 /* The command header of this PDU has now been read from the socket. */ 1973 nvmf_tcp_pdu_ch_handle(tqpair); 1974 break; 1975 /* Wait for the pdu specific header */ 1976 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1977 rc = nvme_tcp_read_data(tqpair->sock, 1978 pdu->psh_len - pdu->psh_valid_bytes, 1979 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1980 if (rc < 0) { 1981 return NVME_TCP_PDU_FATAL; 1982 } else if (rc > 0) { 1983 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1984 0, rc, 0, 0); 1985 pdu->psh_valid_bytes += rc; 1986 } 1987 1988 if (pdu->psh_valid_bytes < pdu->psh_len) { 1989 return NVME_TCP_PDU_IN_PROGRESS; 1990 } 1991 1992 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1993 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1994 break; 1995 /* Wait for the req slot */ 1996 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1997 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1998 break; 1999 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2000 /* check whether the data is valid, if not we just return */ 2001 if (!pdu->data_len) { 2002 return NVME_TCP_PDU_IN_PROGRESS; 2003 } 2004 2005 data_len = pdu->data_len; 2006 /* data digest */ 2007 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2008 tqpair->host_ddgst_enable)) { 2009 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2010 pdu->ddgst_enable = true; 2011 } 2012 2013 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2014 if (rc < 0) { 2015 return NVME_TCP_PDU_FATAL; 2016 } 2017 pdu->rw_offset += rc; 2018 2019 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2020 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 2021 if (rc != 0) { 2022 return NVME_TCP_PDU_FATAL; 2023 } 2024 } 2025 2026 if (pdu->rw_offset < data_len) { 2027 return NVME_TCP_PDU_IN_PROGRESS; 2028 } 2029 2030 /* All of this PDU has now been read from the socket. */ 2031 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2032 break; 2033 case NVME_TCP_PDU_RECV_STATE_ERROR: 2034 if (!spdk_sock_is_connected(tqpair->sock)) { 2035 return NVME_TCP_PDU_FATAL; 2036 } 2037 break; 2038 default: 2039 assert(0); 2040 SPDK_ERRLOG("code should not come to here"); 2041 break; 2042 } 2043 } while (tqpair->recv_state != prev_state); 2044 2045 return rc; 2046 } 2047 2048 static inline void * 2049 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2050 { 2051 struct spdk_nvmf_tcp_control_msg *msg; 2052 2053 assert(list); 2054 2055 msg = STAILQ_FIRST(&list->free_msgs); 2056 if (!msg) { 2057 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2058 return NULL; 2059 } 2060 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2061 return msg; 2062 } 2063 2064 static inline void 2065 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2066 { 2067 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2068 2069 assert(list); 2070 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2071 } 2072 2073 static int 2074 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2075 struct spdk_nvmf_transport *transport, 2076 struct spdk_nvmf_transport_poll_group *group) 2077 { 2078 struct spdk_nvmf_request *req = &tcp_req->req; 2079 struct spdk_nvme_cmd *cmd; 2080 struct spdk_nvme_cpl *rsp; 2081 struct spdk_nvme_sgl_descriptor *sgl; 2082 struct spdk_nvmf_tcp_poll_group *tgroup; 2083 uint32_t length; 2084 2085 cmd = &req->cmd->nvme_cmd; 2086 rsp = &req->rsp->nvme_cpl; 2087 sgl = &cmd->dptr.sgl1; 2088 2089 length = sgl->unkeyed.length; 2090 2091 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2092 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2093 if (length > transport->opts.max_io_size) { 2094 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2095 length, transport->opts.max_io_size); 2096 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2097 return -1; 2098 } 2099 2100 /* fill request length and populate iovs */ 2101 req->length = length; 2102 2103 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2104 2105 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2106 req->dif.orig_length = length; 2107 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2108 req->dif.elba_length = length; 2109 } 2110 2111 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2112 /* No available buffers. Queue this request up. */ 2113 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2114 tcp_req); 2115 return 0; 2116 } 2117 2118 /* backward compatible */ 2119 req->data = req->iov[0].iov_base; 2120 2121 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2122 tcp_req, req->iovcnt, req->data); 2123 2124 return 0; 2125 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2126 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2127 uint64_t offset = sgl->address; 2128 uint32_t max_len = transport->opts.in_capsule_data_size; 2129 assert(tcp_req->has_incapsule_data); 2130 2131 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2132 offset, length); 2133 2134 if (offset > max_len) { 2135 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2136 offset, max_len); 2137 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2138 return -1; 2139 } 2140 max_len -= (uint32_t)offset; 2141 2142 if (spdk_unlikely(length > max_len)) { 2143 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2144 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2145 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2146 2147 /* Get a buffer from dedicated list */ 2148 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2149 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2150 assert(tgroup->control_msg_list); 2151 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2152 if (!req->data) { 2153 /* No available buffers. Queue this request up. */ 2154 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2155 return 0; 2156 } 2157 } else { 2158 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2159 length, max_len); 2160 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2161 return -1; 2162 } 2163 } else { 2164 req->data = tcp_req->buf; 2165 } 2166 2167 req->length = length; 2168 req->data_from_pool = false; 2169 2170 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2171 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2172 req->dif.elba_length = length; 2173 } 2174 2175 req->iov[0].iov_base = req->data; 2176 req->iov[0].iov_len = length; 2177 req->iovcnt = 1; 2178 2179 return 0; 2180 } 2181 2182 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2183 sgl->generic.type, sgl->generic.subtype); 2184 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2185 return -1; 2186 } 2187 2188 static inline enum spdk_nvme_media_error_status_code 2189 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2190 enum spdk_nvme_media_error_status_code result; 2191 2192 switch (err_type) 2193 { 2194 case SPDK_DIF_REFTAG_ERROR: 2195 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2196 break; 2197 case SPDK_DIF_APPTAG_ERROR: 2198 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2199 break; 2200 case SPDK_DIF_GUARD_ERROR: 2201 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2202 break; 2203 default: 2204 SPDK_UNREACHABLE(); 2205 break; 2206 } 2207 2208 return result; 2209 } 2210 2211 static void 2212 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2213 struct spdk_nvmf_tcp_req *tcp_req) 2214 { 2215 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2216 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2217 struct nvme_tcp_pdu *rsp_pdu; 2218 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2219 uint32_t plen, pdo, alignment; 2220 int rc; 2221 2222 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2223 2224 rsp_pdu = tcp_req->pdu; 2225 assert(rsp_pdu != NULL); 2226 assert(tcp_req->pdu_in_use); 2227 2228 c2h_data = &rsp_pdu->hdr.c2h_data; 2229 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2230 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2231 2232 if (tqpair->host_hdgst_enable) { 2233 plen += SPDK_NVME_TCP_DIGEST_LEN; 2234 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2235 } 2236 2237 /* set the psh */ 2238 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2239 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2240 c2h_data->datao = tcp_req->pdu->rw_offset; 2241 2242 /* set the padding */ 2243 rsp_pdu->padding_len = 0; 2244 pdo = plen; 2245 if (tqpair->cpda) { 2246 alignment = (tqpair->cpda + 1) << 2; 2247 if (plen % alignment != 0) { 2248 pdo = (plen + alignment) / alignment * alignment; 2249 rsp_pdu->padding_len = pdo - plen; 2250 plen = pdo; 2251 } 2252 } 2253 2254 c2h_data->common.pdo = pdo; 2255 plen += c2h_data->datal; 2256 if (tqpair->host_ddgst_enable) { 2257 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2258 plen += SPDK_NVME_TCP_DIGEST_LEN; 2259 } 2260 2261 c2h_data->common.plen = plen; 2262 2263 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2264 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2265 } 2266 2267 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2268 c2h_data->datao, c2h_data->datal); 2269 2270 2271 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2272 /* Need to send the capsule response if response is not all 0 */ 2273 if (ttransport->tcp_opts.c2h_success && 2274 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2275 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2276 } 2277 2278 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2279 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2280 struct spdk_dif_error err_blk = {}; 2281 uint32_t mapped_length = 0; 2282 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2283 uint32_t ddgst_len = 0; 2284 2285 if (tqpair->host_ddgst_enable) { 2286 /* Data digest consumes additional iov entry */ 2287 available_iovs--; 2288 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2289 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2290 c2h_data->common.plen -= ddgst_len; 2291 } 2292 /* Temp call to estimate if data can be described by limited number of iovs. 2293 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2294 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2295 false, &mapped_length); 2296 2297 if (mapped_length != c2h_data->common.plen) { 2298 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2299 SPDK_DEBUGLOG(nvmf_tcp, 2300 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2301 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2302 c2h_data->common.plen = mapped_length; 2303 2304 /* Rebuild pdu->data_iov since data length is changed */ 2305 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2306 c2h_data->datal); 2307 2308 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2309 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2310 } 2311 2312 c2h_data->common.plen += ddgst_len; 2313 2314 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2315 2316 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2317 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2318 if (rc != 0) { 2319 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2320 err_blk.err_type, err_blk.err_offset); 2321 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2322 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2323 nvmf_tcp_req_pdu_fini(tcp_req); 2324 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2325 return; 2326 } 2327 } 2328 2329 rsp_pdu->rw_offset += c2h_data->datal; 2330 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2331 } 2332 2333 static void 2334 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2335 struct spdk_nvmf_tcp_req *tcp_req) 2336 { 2337 nvmf_tcp_req_pdu_init(tcp_req); 2338 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2339 } 2340 2341 static int 2342 request_transfer_out(struct spdk_nvmf_request *req) 2343 { 2344 struct spdk_nvmf_tcp_req *tcp_req; 2345 struct spdk_nvmf_qpair *qpair; 2346 struct spdk_nvmf_tcp_qpair *tqpair; 2347 struct spdk_nvme_cpl *rsp; 2348 2349 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2350 2351 qpair = req->qpair; 2352 rsp = &req->rsp->nvme_cpl; 2353 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2354 2355 /* Advance our sq_head pointer */ 2356 if (qpair->sq_head == qpair->sq_head_max) { 2357 qpair->sq_head = 0; 2358 } else { 2359 qpair->sq_head++; 2360 } 2361 rsp->sqhd = qpair->sq_head; 2362 2363 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2364 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2365 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2366 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2367 } else { 2368 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2369 } 2370 2371 return 0; 2372 } 2373 2374 static void 2375 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2376 struct spdk_nvmf_tcp_req *tcp_req) 2377 { 2378 struct nvme_tcp_pdu *pdu; 2379 uint32_t plen = 0; 2380 2381 pdu = &tqpair->pdu_in_progress; 2382 plen = pdu->hdr.common.hlen; 2383 2384 if (tqpair->host_hdgst_enable) { 2385 plen += SPDK_NVME_TCP_DIGEST_LEN; 2386 } 2387 2388 if (pdu->hdr.common.plen != plen) { 2389 tcp_req->has_incapsule_data = true; 2390 } 2391 } 2392 2393 static bool 2394 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2395 struct spdk_nvmf_tcp_req *tcp_req) 2396 { 2397 struct spdk_nvmf_tcp_qpair *tqpair; 2398 int rc; 2399 enum spdk_nvmf_tcp_req_state prev_state; 2400 bool progress = false; 2401 struct spdk_nvmf_transport *transport = &ttransport->transport; 2402 struct spdk_nvmf_transport_poll_group *group; 2403 struct spdk_nvmf_tcp_poll_group *tgroup; 2404 2405 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2406 group = &tqpair->group->group; 2407 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2408 2409 /* If the qpair is not active, we need to abort the outstanding requests. */ 2410 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2411 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2412 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2413 } 2414 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2415 } 2416 2417 /* The loop here is to allow for several back-to-back state changes. */ 2418 do { 2419 prev_state = tcp_req->state; 2420 2421 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2422 tqpair); 2423 2424 switch (tcp_req->state) { 2425 case TCP_REQUEST_STATE_FREE: 2426 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2427 * to escape this state. */ 2428 break; 2429 case TCP_REQUEST_STATE_NEW: 2430 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2431 2432 /* copy the cmd from the receive pdu */ 2433 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2434 2435 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2436 tcp_req->req.dif.dif_insert_or_strip = true; 2437 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2438 } 2439 2440 /* The next state transition depends on the data transfer needs of this request. */ 2441 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2442 2443 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2444 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2445 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2446 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2447 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2448 break; 2449 } 2450 2451 /* If no data to transfer, ready to execute. */ 2452 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2453 /* Reset the tqpair receving pdu state */ 2454 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2455 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2456 break; 2457 } 2458 2459 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2460 2461 if (!tcp_req->has_incapsule_data) { 2462 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2463 } 2464 2465 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2466 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2467 break; 2468 case TCP_REQUEST_STATE_NEED_BUFFER: 2469 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2470 2471 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2472 2473 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2474 SPDK_DEBUGLOG(nvmf_tcp, 2475 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2476 tcp_req, tqpair); 2477 /* This request needs to wait in line to obtain a buffer */ 2478 break; 2479 } 2480 2481 /* Try to get a data buffer */ 2482 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2483 if (rc < 0) { 2484 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2485 /* Reset the tqpair receving pdu state */ 2486 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2487 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2488 break; 2489 } 2490 2491 if (!tcp_req->req.data) { 2492 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2493 tcp_req, tqpair); 2494 /* No buffers available. */ 2495 break; 2496 } 2497 2498 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2499 2500 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2501 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2502 if (tcp_req->req.data_from_pool) { 2503 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2504 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2505 } else { 2506 struct nvme_tcp_pdu *pdu; 2507 2508 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2509 2510 pdu = &tqpair->pdu_in_progress; 2511 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2512 tqpair); 2513 /* No need to send r2t, contained in the capsuled data */ 2514 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2515 0, tcp_req->req.length); 2516 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2517 } 2518 break; 2519 } 2520 2521 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2522 break; 2523 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2524 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2525 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2526 break; 2527 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2528 2529 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2530 (uintptr_t)tcp_req, 0); 2531 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2532 * to escape this state. */ 2533 break; 2534 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2535 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2536 2537 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2538 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2539 tcp_req->req.length = tcp_req->req.dif.elba_length; 2540 } 2541 2542 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2543 spdk_nvmf_request_exec(&tcp_req->req); 2544 break; 2545 case TCP_REQUEST_STATE_EXECUTING: 2546 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2547 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2548 * to escape this state. */ 2549 break; 2550 case TCP_REQUEST_STATE_EXECUTED: 2551 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2552 2553 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2554 tcp_req->req.length = tcp_req->req.dif.orig_length; 2555 } 2556 2557 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2558 break; 2559 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2560 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2561 rc = request_transfer_out(&tcp_req->req); 2562 assert(rc == 0); /* No good way to handle this currently */ 2563 break; 2564 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2565 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2566 (uintptr_t)tcp_req, 2567 0); 2568 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2569 * to escape this state. */ 2570 break; 2571 case TCP_REQUEST_STATE_COMPLETED: 2572 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2573 if (tcp_req->req.data_from_pool) { 2574 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2575 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2576 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2577 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2578 assert(tgroup->control_msg_list); 2579 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2580 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2581 } 2582 tcp_req->req.length = 0; 2583 tcp_req->req.iovcnt = 0; 2584 tcp_req->req.data = NULL; 2585 2586 nvmf_tcp_req_pdu_fini(tcp_req); 2587 2588 nvmf_tcp_req_put(tqpair, tcp_req); 2589 break; 2590 case TCP_REQUEST_NUM_STATES: 2591 default: 2592 assert(0); 2593 break; 2594 } 2595 2596 if (tcp_req->state != prev_state) { 2597 progress = true; 2598 } 2599 } while (tcp_req->state != prev_state); 2600 2601 return progress; 2602 } 2603 2604 static void 2605 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2606 { 2607 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2608 int rc; 2609 2610 assert(tqpair != NULL); 2611 rc = nvmf_tcp_sock_process(tqpair); 2612 2613 /* If there was a new socket error, disconnect */ 2614 if (rc < 0) { 2615 nvmf_tcp_qpair_disconnect(tqpair); 2616 } 2617 } 2618 2619 static int 2620 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2621 struct spdk_nvmf_qpair *qpair) 2622 { 2623 struct spdk_nvmf_tcp_poll_group *tgroup; 2624 struct spdk_nvmf_tcp_qpair *tqpair; 2625 int rc; 2626 2627 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2628 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2629 2630 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2631 nvmf_tcp_sock_cb, tqpair); 2632 if (rc != 0) { 2633 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2634 spdk_strerror(errno), errno); 2635 return -1; 2636 } 2637 2638 rc = nvmf_tcp_qpair_sock_init(tqpair); 2639 if (rc != 0) { 2640 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2641 return -1; 2642 } 2643 2644 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2645 if (rc < 0) { 2646 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2647 return -1; 2648 } 2649 2650 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2651 if (rc < 0) { 2652 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2653 return -1; 2654 } 2655 2656 tqpair->group = tgroup; 2657 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2658 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2659 2660 return 0; 2661 } 2662 2663 static int 2664 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2665 struct spdk_nvmf_qpair *qpair) 2666 { 2667 struct spdk_nvmf_tcp_poll_group *tgroup; 2668 struct spdk_nvmf_tcp_qpair *tqpair; 2669 int rc; 2670 2671 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2672 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2673 2674 assert(tqpair->group == tgroup); 2675 2676 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2677 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2678 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2679 } else { 2680 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2681 } 2682 2683 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2684 if (rc != 0) { 2685 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2686 spdk_strerror(errno), errno); 2687 } 2688 2689 return rc; 2690 } 2691 2692 static int 2693 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2694 { 2695 struct spdk_nvmf_tcp_transport *ttransport; 2696 struct spdk_nvmf_tcp_req *tcp_req; 2697 2698 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2699 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2700 2701 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2702 nvmf_tcp_req_process(ttransport, tcp_req); 2703 2704 return 0; 2705 } 2706 2707 static void 2708 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2709 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2710 { 2711 struct spdk_nvmf_tcp_qpair *tqpair; 2712 2713 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2714 2715 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2716 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2717 nvmf_tcp_qpair_destroy(tqpair); 2718 2719 if (cb_fn) { 2720 cb_fn(cb_arg); 2721 } 2722 } 2723 2724 static int 2725 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2726 { 2727 struct spdk_nvmf_tcp_poll_group *tgroup; 2728 int rc; 2729 struct spdk_nvmf_request *req, *req_tmp; 2730 struct spdk_nvmf_tcp_req *tcp_req; 2731 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2732 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2733 struct spdk_nvmf_tcp_transport, transport); 2734 2735 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2736 2737 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2738 return 0; 2739 } 2740 2741 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2742 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2743 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2744 break; 2745 } 2746 } 2747 2748 rc = spdk_sock_group_poll(tgroup->sock_group); 2749 if (rc < 0) { 2750 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2751 } 2752 2753 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2754 nvmf_tcp_sock_process(tqpair); 2755 } 2756 2757 return rc; 2758 } 2759 2760 static int 2761 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2762 struct spdk_nvme_transport_id *trid, bool peer) 2763 { 2764 struct spdk_nvmf_tcp_qpair *tqpair; 2765 uint16_t port; 2766 2767 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2768 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2769 2770 if (peer) { 2771 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2772 port = tqpair->initiator_port; 2773 } else { 2774 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2775 port = tqpair->target_port; 2776 } 2777 2778 if (spdk_sock_is_ipv4(tqpair->sock)) { 2779 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2780 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2781 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2782 } else { 2783 return -1; 2784 } 2785 2786 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2787 return 0; 2788 } 2789 2790 static int 2791 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2792 struct spdk_nvme_transport_id *trid) 2793 { 2794 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2795 } 2796 2797 static int 2798 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2799 struct spdk_nvme_transport_id *trid) 2800 { 2801 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2802 } 2803 2804 static int 2805 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2806 struct spdk_nvme_transport_id *trid) 2807 { 2808 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2809 } 2810 2811 static void 2812 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2813 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2814 { 2815 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2816 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2817 2818 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2819 2820 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2821 } 2822 2823 static int 2824 _nvmf_tcp_qpair_abort_request(void *ctx) 2825 { 2826 struct spdk_nvmf_request *req = ctx; 2827 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2828 struct spdk_nvmf_tcp_req, req); 2829 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2830 struct spdk_nvmf_tcp_qpair, qpair); 2831 int rc; 2832 2833 spdk_poller_unregister(&req->poller); 2834 2835 switch (tcp_req_to_abort->state) { 2836 case TCP_REQUEST_STATE_EXECUTING: 2837 rc = nvmf_ctrlr_abort_request(req); 2838 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2839 return SPDK_POLLER_BUSY; 2840 } 2841 break; 2842 2843 case TCP_REQUEST_STATE_NEED_BUFFER: 2844 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2845 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2846 2847 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2848 break; 2849 2850 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2851 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2852 break; 2853 2854 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2855 if (spdk_get_ticks() < req->timeout_tsc) { 2856 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2857 return SPDK_POLLER_BUSY; 2858 } 2859 break; 2860 2861 default: 2862 break; 2863 } 2864 2865 spdk_nvmf_request_complete(req); 2866 return SPDK_POLLER_BUSY; 2867 } 2868 2869 static void 2870 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2871 struct spdk_nvmf_request *req) 2872 { 2873 struct spdk_nvmf_tcp_qpair *tqpair; 2874 struct spdk_nvmf_tcp_transport *ttransport; 2875 struct spdk_nvmf_transport *transport; 2876 uint16_t cid; 2877 uint32_t i; 2878 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2879 2880 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2881 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2882 transport = &ttransport->transport; 2883 2884 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2885 2886 for (i = 0; i < tqpair->resource_count; i++) { 2887 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2888 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2889 tcp_req_to_abort = &tqpair->reqs[i]; 2890 break; 2891 } 2892 } 2893 2894 if (tcp_req_to_abort == NULL) { 2895 spdk_nvmf_request_complete(req); 2896 return; 2897 } 2898 2899 req->req_to_abort = &tcp_req_to_abort->req; 2900 req->timeout_tsc = spdk_get_ticks() + 2901 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2902 req->poller = NULL; 2903 2904 _nvmf_tcp_qpair_abort_request(req); 2905 } 2906 2907 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2908 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2909 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2910 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2911 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2912 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2913 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2914 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2915 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2916 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2917 2918 static void 2919 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2920 { 2921 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2922 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2923 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2924 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2925 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2926 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2927 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2928 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2929 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2930 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2931 opts->transport_specific = NULL; 2932 } 2933 2934 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2935 .name = "TCP", 2936 .type = SPDK_NVME_TRANSPORT_TCP, 2937 .opts_init = nvmf_tcp_opts_init, 2938 .create = nvmf_tcp_create, 2939 .dump_opts = nvmf_tcp_dump_opts, 2940 .destroy = nvmf_tcp_destroy, 2941 2942 .listen = nvmf_tcp_listen, 2943 .stop_listen = nvmf_tcp_stop_listen, 2944 .accept = nvmf_tcp_accept, 2945 2946 .listener_discover = nvmf_tcp_discover, 2947 2948 .poll_group_create = nvmf_tcp_poll_group_create, 2949 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2950 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2951 .poll_group_add = nvmf_tcp_poll_group_add, 2952 .poll_group_remove = nvmf_tcp_poll_group_remove, 2953 .poll_group_poll = nvmf_tcp_poll_group_poll, 2954 2955 .req_free = nvmf_tcp_req_free, 2956 .req_complete = nvmf_tcp_req_complete, 2957 2958 .qpair_fini = nvmf_tcp_close_qpair, 2959 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2960 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2961 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2962 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2963 }; 2964 2965 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2966 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2967