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/stdinc.h" 35 #include "spdk/crc32.h" 36 #include "spdk/endian.h" 37 #include "spdk/assert.h" 38 #include "spdk/thread.h" 39 #include "spdk/nvmf_transport.h" 40 #include "spdk/sock.h" 41 #include "spdk/string.h" 42 #include "spdk/trace.h" 43 #include "spdk/util.h" 44 45 #include "spdk_internal/assert.h" 46 #include "spdk_internal/log.h" 47 #include "spdk_internal/nvme_tcp.h" 48 49 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 50 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 6 51 #define SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR 4 52 53 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 54 55 /* spdk nvmf related structure */ 56 enum spdk_nvmf_tcp_req_state { 57 58 /* The request is not currently in use */ 59 TCP_REQUEST_STATE_FREE = 0, 60 61 /* Initial state when request first received */ 62 TCP_REQUEST_STATE_NEW, 63 64 /* The request is queued until a data buffer is available. */ 65 TCP_REQUEST_STATE_NEED_BUFFER, 66 67 /* The request is currently transferring data from the host to the controller. */ 68 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 69 70 /* The request is waiting for the R2T send acknowledgement. */ 71 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 72 73 /* The request is ready to execute at the block device */ 74 TCP_REQUEST_STATE_READY_TO_EXECUTE, 75 76 /* The request is currently executing at the block device */ 77 TCP_REQUEST_STATE_EXECUTING, 78 79 /* The request finished executing at the block device */ 80 TCP_REQUEST_STATE_EXECUTED, 81 82 /* The request is ready to send a completion */ 83 TCP_REQUEST_STATE_READY_TO_COMPLETE, 84 85 /* The request is currently transferring final pdus from the controller to the host. */ 86 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 87 88 /* The request completed and can be marked free. */ 89 TCP_REQUEST_STATE_COMPLETED, 90 91 /* Terminator */ 92 TCP_REQUEST_NUM_STATES, 93 }; 94 95 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 96 "Invalid PDU Header Field", 97 "PDU Sequence Error", 98 "Header Digiest Error", 99 "Data Transfer Out of Range", 100 "R2T Limit Exceeded", 101 "Unsupported parameter", 102 }; 103 104 #define OBJECT_NVMF_TCP_IO 0x80 105 106 #define TRACE_GROUP_NVMF_TCP 0x5 107 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 108 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 109 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 110 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 111 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 112 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 113 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 115 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 116 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 117 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 118 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 119 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 120 121 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 122 { 123 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 124 spdk_trace_register_description("TCP_REQ_NEW", 125 TRACE_TCP_REQUEST_STATE_NEW, 126 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 127 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 128 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 129 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 130 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 131 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 132 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 133 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 134 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 135 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 136 spdk_trace_register_description("TCP_REQ_EXECUTING", 137 TRACE_TCP_REQUEST_STATE_EXECUTING, 138 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 139 spdk_trace_register_description("TCP_REQ_EXECUTED", 140 TRACE_TCP_REQUEST_STATE_EXECUTED, 141 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 142 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 143 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 144 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 145 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 146 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 147 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 148 spdk_trace_register_description("TCP_REQ_COMPLETED", 149 TRACE_TCP_REQUEST_STATE_COMPLETED, 150 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 151 spdk_trace_register_description("TCP_WRITE_START", 152 TRACE_TCP_FLUSH_WRITEBUF_START, 153 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 154 spdk_trace_register_description("TCP_WRITE_DONE", 155 TRACE_TCP_FLUSH_WRITEBUF_DONE, 156 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 157 spdk_trace_register_description("TCP_READ_DONE", 158 TRACE_TCP_READ_FROM_SOCKET_DONE, 159 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 160 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 161 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 162 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 163 } 164 165 struct spdk_nvmf_tcp_req { 166 struct spdk_nvmf_request req; 167 struct spdk_nvme_cpl rsp; 168 struct spdk_nvme_cmd cmd; 169 170 /* A PDU that can be used for sending responses. This is 171 * not the incoming PDU! */ 172 struct nvme_tcp_pdu *pdu; 173 174 /* 175 * The PDU for a request may be used multiple times in serial over 176 * the request's lifetime. For example, first to send an R2T, then 177 * to send a completion. To catch mistakes where the PDU is used 178 * twice at the same time, add a debug flag here for init/fini. 179 */ 180 bool pdu_in_use; 181 182 /* In-capsule data buffer */ 183 uint8_t *buf; 184 185 bool has_incapsule_data; 186 187 /* transfer_tag */ 188 uint16_t ttag; 189 190 enum spdk_nvmf_tcp_req_state state; 191 192 /* 193 * h2c_offset is used when we receive the h2c_data PDU. 194 */ 195 uint32_t h2c_offset; 196 197 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 198 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 199 }; 200 201 struct spdk_nvmf_tcp_qpair { 202 struct spdk_nvmf_qpair qpair; 203 struct spdk_nvmf_tcp_poll_group *group; 204 struct spdk_nvmf_tcp_port *port; 205 struct spdk_sock *sock; 206 207 enum nvme_tcp_pdu_recv_state recv_state; 208 enum nvme_tcp_qpair_state state; 209 210 /* PDU being actively received */ 211 struct nvme_tcp_pdu pdu_in_progress; 212 uint32_t recv_buf_size; 213 214 /* This is a spare PDU used for sending special management 215 * operations. Primarily, this is used for the initial 216 * connection response and c2h termination request. */ 217 struct nvme_tcp_pdu mgmt_pdu; 218 219 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 220 221 /* Arrays of in-capsule buffers, requests, and pdus. 222 * Each array is 'resource_count' number of elements */ 223 void *bufs; 224 struct spdk_nvmf_tcp_req *reqs; 225 struct nvme_tcp_pdu *pdus; 226 uint32_t resource_count; 227 228 /* Queues to track the requests in all states */ 229 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 230 /* Number of requests in each state */ 231 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 232 233 uint8_t cpda; 234 235 bool host_hdgst_enable; 236 bool host_ddgst_enable; 237 238 /* IP address */ 239 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 240 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 241 242 /* IP port */ 243 uint16_t initiator_port; 244 uint16_t target_port; 245 246 /* Timer used to destroy qpair after detecting transport error issue if initiator does 247 * not close the connection. 248 */ 249 struct spdk_poller *timeout_poller; 250 251 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 252 }; 253 254 struct spdk_nvmf_tcp_poll_group { 255 struct spdk_nvmf_transport_poll_group group; 256 struct spdk_sock_group *sock_group; 257 258 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 259 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 260 }; 261 262 struct spdk_nvmf_tcp_port { 263 const struct spdk_nvme_transport_id *trid; 264 struct spdk_sock *listen_sock; 265 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 266 }; 267 268 struct spdk_nvmf_tcp_transport { 269 struct spdk_nvmf_transport transport; 270 271 pthread_mutex_t lock; 272 273 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 274 }; 275 276 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 277 struct spdk_nvmf_tcp_req *tcp_req); 278 279 static void 280 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 281 enum spdk_nvmf_tcp_req_state state) 282 { 283 struct spdk_nvmf_qpair *qpair; 284 struct spdk_nvmf_tcp_qpair *tqpair; 285 286 qpair = tcp_req->req.qpair; 287 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 288 289 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 290 assert(tqpair->state_cntr[tcp_req->state] > 0); 291 tqpair->state_cntr[tcp_req->state]--; 292 293 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 294 tqpair->state_cntr[state]++; 295 296 tcp_req->state = state; 297 } 298 299 static inline struct nvme_tcp_pdu * 300 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 301 { 302 assert(tcp_req->pdu_in_use == false); 303 tcp_req->pdu_in_use = true; 304 305 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 306 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 307 308 return tcp_req->pdu; 309 } 310 311 static inline void 312 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 313 { 314 tcp_req->pdu_in_use = false; 315 } 316 317 static struct spdk_nvmf_tcp_req * 318 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 319 { 320 struct spdk_nvmf_tcp_req *tcp_req; 321 322 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 323 if (!tcp_req) { 324 return NULL; 325 } 326 327 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 328 tcp_req->h2c_offset = 0; 329 tcp_req->has_incapsule_data = false; 330 tcp_req->req.dif.dif_insert_or_strip = false; 331 332 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 333 return tcp_req; 334 } 335 336 static void 337 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 338 { 339 struct spdk_nvmf_tcp_transport *ttransport; 340 341 assert(tcp_req != NULL); 342 343 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req); 344 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 345 struct spdk_nvmf_tcp_transport, transport); 346 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 347 nvmf_tcp_req_process(ttransport, tcp_req); 348 } 349 350 static int 351 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 352 { 353 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 354 355 nvmf_tcp_request_free(tcp_req); 356 357 return 0; 358 } 359 360 static void 361 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 362 enum spdk_nvmf_tcp_req_state state) 363 { 364 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 365 366 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 367 nvmf_tcp_request_free(tcp_req); 368 } 369 } 370 371 static void 372 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 373 { 374 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 375 376 assert(TAILQ_EMPTY(&tqpair->send_queue)); 377 378 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 379 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 380 381 /* Wipe the requests waiting for buffer from the global list */ 382 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 383 req_tmp) { 384 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 385 spdk_nvmf_request, buf_link); 386 } 387 388 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 389 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 390 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 391 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 392 } 393 394 static void 395 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 396 { 397 int i; 398 struct spdk_nvmf_tcp_req *tcp_req; 399 400 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 401 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 402 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 403 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 404 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 405 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 406 } 407 } 408 } 409 410 static void 411 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 412 { 413 int err = 0; 414 415 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 416 417 err = spdk_sock_close(&tqpair->sock); 418 assert(err == 0); 419 nvmf_tcp_cleanup_all_states(tqpair); 420 421 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 422 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 423 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 424 tqpair->resource_count); 425 err++; 426 } 427 428 if (err > 0) { 429 nvmf_tcp_dump_qpair_req_contents(tqpair); 430 } 431 432 spdk_dma_free(tqpair->pdus); 433 free(tqpair->reqs); 434 spdk_free(tqpair->bufs); 435 free(tqpair); 436 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n"); 437 } 438 439 static int 440 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 441 { 442 struct spdk_nvmf_tcp_transport *ttransport; 443 444 assert(transport != NULL); 445 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 446 447 pthread_mutex_destroy(&ttransport->lock); 448 free(ttransport); 449 return 0; 450 } 451 452 static struct spdk_nvmf_transport * 453 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 454 { 455 struct spdk_nvmf_tcp_transport *ttransport; 456 uint32_t sge_count; 457 uint32_t min_shared_buffers; 458 459 ttransport = calloc(1, sizeof(*ttransport)); 460 if (!ttransport) { 461 return NULL; 462 } 463 464 TAILQ_INIT(&ttransport->ports); 465 466 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 467 468 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 469 470 SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n" 471 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 472 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 473 " in_capsule_data_size=%d, max_aq_depth=%d\n" 474 " num_shared_buffers=%d, c2h_success=%d,\n" 475 " dif_insert_or_strip=%d, sock_priority=%d\n", 476 opts->max_queue_depth, 477 opts->max_io_size, 478 opts->max_qpairs_per_ctrlr - 1, 479 opts->io_unit_size, 480 opts->in_capsule_data_size, 481 opts->max_aq_depth, 482 opts->num_shared_buffers, 483 opts->c2h_success, 484 opts->dif_insert_or_strip, 485 opts->sock_priority); 486 487 if (opts->sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 488 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 489 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 490 opts->sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 491 free(ttransport); 492 return NULL; 493 } 494 495 /* I/O unit size cannot be larger than max I/O size */ 496 if (opts->io_unit_size > opts->max_io_size) { 497 opts->io_unit_size = opts->max_io_size; 498 } 499 500 sge_count = opts->max_io_size / opts->io_unit_size; 501 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 502 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 503 free(ttransport); 504 return NULL; 505 } 506 507 min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size; 508 if (min_shared_buffers > opts->num_shared_buffers) { 509 SPDK_ERRLOG("There are not enough buffers to satisfy" 510 "per-poll group caches for each thread. (%" PRIu32 ")" 511 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 512 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 513 nvmf_tcp_destroy(&ttransport->transport); 514 return NULL; 515 } 516 517 pthread_mutex_init(&ttransport->lock, NULL); 518 519 return &ttransport->transport; 520 } 521 522 static int 523 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 524 { 525 unsigned long long ull; 526 char *end = NULL; 527 528 ull = strtoull(trsvcid, &end, 10); 529 if (end == NULL || end == trsvcid || *end != '\0') { 530 return -1; 531 } 532 533 /* Valid TCP/IP port numbers are in [0, 65535] */ 534 if (ull > 65535) { 535 return -1; 536 } 537 538 return (int)ull; 539 } 540 541 /** 542 * Canonicalize a listen address trid. 543 */ 544 static int 545 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 546 const struct spdk_nvme_transport_id *trid) 547 { 548 int trsvcid_int; 549 550 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 551 if (trsvcid_int < 0) { 552 return -EINVAL; 553 } 554 555 memset(canon_trid, 0, sizeof(*canon_trid)); 556 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 557 canon_trid->adrfam = trid->adrfam; 558 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 559 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 560 561 return 0; 562 } 563 564 /** 565 * Find an existing listening port. 566 * 567 * Caller must hold ttransport->lock. 568 */ 569 static struct spdk_nvmf_tcp_port * 570 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 571 const struct spdk_nvme_transport_id *trid) 572 { 573 struct spdk_nvme_transport_id canon_trid; 574 struct spdk_nvmf_tcp_port *port; 575 576 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 577 return NULL; 578 } 579 580 TAILQ_FOREACH(port, &ttransport->ports, link) { 581 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 582 return port; 583 } 584 } 585 586 return NULL; 587 } 588 589 static int 590 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 591 const struct spdk_nvme_transport_id *trid) 592 { 593 struct spdk_nvmf_tcp_transport *ttransport; 594 struct spdk_nvmf_tcp_port *port; 595 int trsvcid_int; 596 uint8_t adrfam; 597 struct spdk_sock_opts opts; 598 599 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 600 601 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 602 if (trsvcid_int < 0) { 603 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 604 return -EINVAL; 605 } 606 607 pthread_mutex_lock(&ttransport->lock); 608 port = calloc(1, sizeof(*port)); 609 if (!port) { 610 SPDK_ERRLOG("Port allocation failed\n"); 611 pthread_mutex_unlock(&ttransport->lock); 612 return -ENOMEM; 613 } 614 615 port->trid = trid; 616 opts.opts_size = sizeof(opts); 617 spdk_sock_get_default_opts(&opts); 618 opts.priority = transport->opts.sock_priority; 619 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 620 NULL, &opts); 621 if (port->listen_sock == NULL) { 622 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 623 trid->traddr, trsvcid_int, 624 spdk_strerror(errno), errno); 625 free(port); 626 pthread_mutex_unlock(&ttransport->lock); 627 return -errno; 628 } 629 630 if (spdk_sock_is_ipv4(port->listen_sock)) { 631 adrfam = SPDK_NVMF_ADRFAM_IPV4; 632 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 633 adrfam = SPDK_NVMF_ADRFAM_IPV6; 634 } else { 635 SPDK_ERRLOG("Unhandled socket type\n"); 636 adrfam = 0; 637 } 638 639 if (adrfam != trid->adrfam) { 640 SPDK_ERRLOG("Socket address family mismatch\n"); 641 spdk_sock_close(&port->listen_sock); 642 free(port); 643 pthread_mutex_unlock(&ttransport->lock); 644 return -EINVAL; 645 } 646 647 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 648 trid->traddr, trid->trsvcid); 649 650 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 651 pthread_mutex_unlock(&ttransport->lock); 652 return 0; 653 } 654 655 static void 656 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 657 const struct spdk_nvme_transport_id *trid) 658 { 659 struct spdk_nvmf_tcp_transport *ttransport; 660 struct spdk_nvmf_tcp_port *port; 661 662 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 663 664 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 665 trid->traddr, trid->trsvcid); 666 667 pthread_mutex_lock(&ttransport->lock); 668 port = nvmf_tcp_find_port(ttransport, trid); 669 if (port) { 670 TAILQ_REMOVE(&ttransport->ports, port, link); 671 spdk_sock_close(&port->listen_sock); 672 free(port); 673 } 674 675 pthread_mutex_unlock(&ttransport->lock); 676 } 677 678 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 679 enum nvme_tcp_pdu_recv_state state); 680 681 static void 682 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 683 { 684 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Disconnecting qpair %p\n", tqpair); 685 686 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 687 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 688 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 689 spdk_poller_unregister(&tqpair->timeout_poller); 690 691 /* This will end up calling nvmf_tcp_close_qpair */ 692 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 693 } 694 } 695 696 static void 697 _pdu_write_done(void *_pdu, int err) 698 { 699 struct nvme_tcp_pdu *pdu = _pdu; 700 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 701 702 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 703 704 if (err != 0) { 705 nvmf_tcp_qpair_disconnect(tqpair); 706 return; 707 } 708 709 assert(pdu->cb_fn != NULL); 710 pdu->cb_fn(pdu->cb_arg); 711 } 712 713 static void 714 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 715 struct nvme_tcp_pdu *pdu, 716 nvme_tcp_qpair_xfer_complete_cb cb_fn, 717 void *cb_arg) 718 { 719 int hlen; 720 uint32_t crc32c; 721 uint32_t mapped_length = 0; 722 ssize_t rc; 723 724 assert(&tqpair->pdu_in_progress != pdu); 725 726 hlen = pdu->hdr.common.hlen; 727 728 /* Header Digest */ 729 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 730 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 731 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 732 } 733 734 /* Data Digest */ 735 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 736 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 737 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 738 } 739 740 pdu->cb_fn = cb_fn; 741 pdu->cb_arg = cb_arg; 742 743 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 744 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 745 &mapped_length); 746 pdu->sock_req.cb_fn = _pdu_write_done; 747 pdu->sock_req.cb_arg = pdu; 748 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 749 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 750 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 751 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 752 if (rc == mapped_length) { 753 _pdu_write_done(pdu, 0); 754 } else { 755 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 756 _pdu_write_done(pdu, -1); 757 } 758 } else { 759 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 760 } 761 } 762 763 static int 764 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 765 { 766 uint32_t i; 767 struct spdk_nvmf_transport_opts *opts; 768 uint32_t in_capsule_data_size; 769 770 opts = &tqpair->qpair.transport->opts; 771 772 in_capsule_data_size = opts->in_capsule_data_size; 773 if (opts->dif_insert_or_strip) { 774 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 775 } 776 777 tqpair->resource_count = opts->max_queue_depth; 778 779 tqpair->mgmt_pdu.qpair = tqpair; 780 781 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 782 if (!tqpair->reqs) { 783 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 784 return -1; 785 } 786 787 if (in_capsule_data_size) { 788 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 789 NULL, SPDK_ENV_LCORE_ID_ANY, 790 SPDK_MALLOC_DMA); 791 if (!tqpair->bufs) { 792 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 793 return -1; 794 } 795 } 796 797 tqpair->pdus = spdk_dma_malloc(tqpair->resource_count * sizeof(*tqpair->pdus), 0x1000, NULL); 798 if (!tqpair->pdus) { 799 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 800 return -1; 801 } 802 803 for (i = 0; i < tqpair->resource_count; i++) { 804 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 805 806 tcp_req->ttag = i + 1; 807 tcp_req->req.qpair = &tqpair->qpair; 808 809 tcp_req->pdu = &tqpair->pdus[i]; 810 tcp_req->pdu->qpair = tqpair; 811 812 /* Set up memory to receive commands */ 813 if (tqpair->bufs) { 814 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 815 } 816 817 /* Set the cmdn and rsp */ 818 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 819 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 820 821 /* Initialize request state to FREE */ 822 tcp_req->state = TCP_REQUEST_STATE_FREE; 823 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 824 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 825 } 826 827 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 828 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 829 830 return 0; 831 } 832 833 static int 834 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 835 { 836 struct spdk_nvmf_tcp_qpair *tqpair; 837 int i; 838 839 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 840 841 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 842 843 TAILQ_INIT(&tqpair->send_queue); 844 845 /* Initialise request state queues of the qpair */ 846 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 847 TAILQ_INIT(&tqpair->state_queue[i]); 848 } 849 850 tqpair->host_hdgst_enable = true; 851 tqpair->host_ddgst_enable = true; 852 853 return 0; 854 } 855 856 static int 857 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 858 { 859 int rc; 860 861 /* set low water mark */ 862 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 863 if (rc != 0) { 864 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 865 return rc; 866 } 867 868 return 0; 869 } 870 871 static void 872 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 873 struct spdk_nvmf_tcp_port *port, 874 struct spdk_sock *sock) 875 { 876 struct spdk_nvmf_tcp_qpair *tqpair; 877 int rc; 878 879 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 880 port->trid->traddr, port->trid->trsvcid); 881 882 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 883 if (tqpair == NULL) { 884 SPDK_ERRLOG("Could not allocate new connection.\n"); 885 spdk_sock_close(&sock); 886 return; 887 } 888 889 tqpair->sock = sock; 890 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 891 tqpair->port = port; 892 tqpair->qpair.transport = transport; 893 894 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 895 sizeof(tqpair->target_addr), &tqpair->target_port, 896 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 897 &tqpair->initiator_port); 898 if (rc < 0) { 899 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 900 nvmf_tcp_qpair_destroy(tqpair); 901 return; 902 } 903 904 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 905 } 906 907 static uint32_t 908 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 909 { 910 struct spdk_sock *sock; 911 uint32_t count = 0; 912 int i; 913 914 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 915 sock = spdk_sock_accept(port->listen_sock); 916 if (sock == NULL) { 917 break; 918 } 919 count++; 920 nvmf_tcp_handle_connect(transport, port, sock); 921 } 922 923 return count; 924 } 925 926 static uint32_t 927 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 928 { 929 struct spdk_nvmf_tcp_transport *ttransport; 930 struct spdk_nvmf_tcp_port *port; 931 uint32_t count = 0; 932 933 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 934 935 TAILQ_FOREACH(port, &ttransport->ports, link) { 936 count += nvmf_tcp_port_accept(transport, port); 937 } 938 939 return count; 940 } 941 942 static void 943 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 944 struct spdk_nvme_transport_id *trid, 945 struct spdk_nvmf_discovery_log_page_entry *entry) 946 { 947 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 948 entry->adrfam = trid->adrfam; 949 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 950 951 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 952 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 953 954 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 955 } 956 957 static struct spdk_nvmf_transport_poll_group * 958 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 959 { 960 struct spdk_nvmf_tcp_poll_group *tgroup; 961 962 tgroup = calloc(1, sizeof(*tgroup)); 963 if (!tgroup) { 964 return NULL; 965 } 966 967 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 968 if (!tgroup->sock_group) { 969 goto cleanup; 970 } 971 972 TAILQ_INIT(&tgroup->qpairs); 973 TAILQ_INIT(&tgroup->await_req); 974 975 return &tgroup->group; 976 977 cleanup: 978 free(tgroup); 979 return NULL; 980 } 981 982 static struct spdk_nvmf_transport_poll_group * 983 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 984 { 985 struct spdk_nvmf_tcp_qpair *tqpair; 986 struct spdk_sock_group *group = NULL; 987 int rc; 988 989 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 990 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 991 if (!rc && group != NULL) { 992 return spdk_sock_group_get_ctx(group); 993 } 994 995 return NULL; 996 } 997 998 static void 999 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1000 { 1001 struct spdk_nvmf_tcp_poll_group *tgroup; 1002 1003 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1004 spdk_sock_group_close(&tgroup->sock_group); 1005 1006 free(tgroup); 1007 } 1008 1009 static void 1010 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1011 enum nvme_tcp_pdu_recv_state state) 1012 { 1013 if (tqpair->recv_state == state) { 1014 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1015 tqpair, state); 1016 return; 1017 } 1018 1019 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1020 /* When leaving the await req state, move the qpair to the main list */ 1021 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1022 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1023 } 1024 1025 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1026 tqpair->recv_state = state; 1027 1028 switch (state) { 1029 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1030 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1031 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1032 break; 1033 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1034 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1035 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1036 break; 1037 case NVME_TCP_PDU_RECV_STATE_ERROR: 1038 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1039 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1040 break; 1041 default: 1042 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1043 abort(); 1044 break; 1045 } 1046 } 1047 1048 static int 1049 nvmf_tcp_qpair_handle_timeout(void *ctx) 1050 { 1051 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1052 1053 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1054 1055 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1056 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1057 1058 nvmf_tcp_qpair_disconnect(tqpair); 1059 return SPDK_POLLER_BUSY; 1060 } 1061 1062 static void 1063 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1064 { 1065 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1066 1067 if (!tqpair->timeout_poller) { 1068 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1069 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1070 } 1071 } 1072 1073 static void 1074 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1075 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1076 { 1077 struct nvme_tcp_pdu *rsp_pdu; 1078 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1079 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1080 uint32_t copy_len; 1081 1082 rsp_pdu = &tqpair->mgmt_pdu; 1083 1084 c2h_term_req = &rsp_pdu->hdr.term_req; 1085 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1086 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1087 1088 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1089 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1090 DSET32(&c2h_term_req->fei, error_offset); 1091 } 1092 1093 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1094 1095 /* Copy the error info into the buffer */ 1096 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1097 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1098 1099 /* Contain the header of the wrong received pdu */ 1100 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1101 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1102 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1103 } 1104 1105 static void 1106 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1107 struct spdk_nvmf_tcp_qpair *tqpair, 1108 struct nvme_tcp_pdu *pdu) 1109 { 1110 struct spdk_nvmf_tcp_req *tcp_req; 1111 1112 assert(pdu->psh_valid_bytes == pdu->psh_len); 1113 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1114 1115 tcp_req = nvmf_tcp_req_get(tqpair); 1116 if (!tcp_req) { 1117 /* Directly return and make the allocation retry again */ 1118 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1119 return; 1120 } 1121 1122 /* The host sent more commands than the maximum queue depth. */ 1123 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1124 nvmf_tcp_qpair_disconnect(tqpair); 1125 return; 1126 } 1127 1128 pdu->req = tcp_req; 1129 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1130 nvmf_tcp_req_process(ttransport, tcp_req); 1131 } 1132 1133 static void 1134 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1135 struct spdk_nvmf_tcp_qpair *tqpair, 1136 struct nvme_tcp_pdu *pdu) 1137 { 1138 struct spdk_nvmf_tcp_req *tcp_req; 1139 struct spdk_nvme_tcp_cmd *capsule_cmd; 1140 uint32_t error_offset = 0; 1141 enum spdk_nvme_tcp_term_req_fes fes; 1142 1143 capsule_cmd = &pdu->hdr.capsule_cmd; 1144 tcp_req = pdu->req; 1145 assert(tcp_req != NULL); 1146 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1147 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1148 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1149 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1150 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1151 goto err; 1152 } 1153 1154 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1155 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1156 nvmf_tcp_req_process(ttransport, tcp_req); 1157 1158 return; 1159 err: 1160 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1161 } 1162 1163 static int 1164 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1165 enum spdk_nvmf_tcp_req_state state, 1166 uint16_t cid, uint16_t tag, 1167 struct spdk_nvmf_tcp_req **req) 1168 { 1169 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1170 1171 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) { 1172 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1173 continue; 1174 } 1175 1176 if (tcp_req->ttag == tag) { 1177 *req = tcp_req; 1178 return 0; 1179 } 1180 1181 *req = NULL; 1182 return -1; 1183 } 1184 1185 /* Didn't find it, but not an error */ 1186 *req = NULL; 1187 return 0; 1188 } 1189 1190 static void 1191 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1192 struct spdk_nvmf_tcp_qpair *tqpair, 1193 struct nvme_tcp_pdu *pdu) 1194 { 1195 struct spdk_nvmf_tcp_req *tcp_req; 1196 uint32_t error_offset = 0; 1197 enum spdk_nvme_tcp_term_req_fes fes = 0; 1198 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1199 int rc; 1200 1201 h2c_data = &pdu->hdr.h2c_data; 1202 1203 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1204 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1205 1206 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1207 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1208 if (rc == 0 && tcp_req == NULL) { 1209 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1210 h2c_data->ttag, &tcp_req); 1211 } 1212 1213 if (!tcp_req) { 1214 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1215 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1216 if (rc == 0) { 1217 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1218 } else { 1219 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1220 } 1221 goto err; 1222 } 1223 1224 if (tcp_req->h2c_offset != h2c_data->datao) { 1225 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1226 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1227 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1228 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1229 goto err; 1230 } 1231 1232 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1233 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1234 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1235 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1236 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1237 goto err; 1238 } 1239 1240 pdu->req = tcp_req; 1241 1242 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1243 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1244 } 1245 1246 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1247 h2c_data->datao, h2c_data->datal); 1248 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1249 return; 1250 1251 err: 1252 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1253 } 1254 1255 static void 1256 nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1257 { 1258 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1259 nvmf_tcp_request_free(tcp_req); 1260 } 1261 1262 static void 1263 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1264 struct spdk_nvmf_tcp_qpair *tqpair) 1265 { 1266 struct nvme_tcp_pdu *rsp_pdu; 1267 struct spdk_nvme_tcp_rsp *capsule_resp; 1268 1269 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1270 1271 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1272 assert(rsp_pdu != NULL); 1273 1274 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1275 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1276 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1277 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1278 if (tqpair->host_hdgst_enable) { 1279 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1280 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1281 } 1282 1283 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_cmd_complete, tcp_req); 1284 } 1285 1286 static void 1287 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1288 { 1289 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1290 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1291 struct spdk_nvmf_tcp_qpair, qpair); 1292 1293 assert(tqpair != NULL); 1294 if (tqpair->qpair.transport->opts.c2h_success) { 1295 nvmf_tcp_request_free(tcp_req); 1296 } else { 1297 nvmf_tcp_req_pdu_fini(tcp_req); 1298 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1299 } 1300 } 1301 1302 static void 1303 nvmf_tcp_r2t_complete(void *cb_arg) 1304 { 1305 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1306 struct spdk_nvmf_tcp_transport *ttransport; 1307 1308 nvmf_tcp_req_pdu_fini(tcp_req); 1309 1310 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1311 struct spdk_nvmf_tcp_transport, transport); 1312 1313 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1314 1315 if (tcp_req->h2c_offset == tcp_req->req.length) { 1316 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1317 nvmf_tcp_req_process(ttransport, tcp_req); 1318 } 1319 } 1320 1321 static void 1322 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1323 struct spdk_nvmf_tcp_req *tcp_req) 1324 { 1325 struct nvme_tcp_pdu *rsp_pdu; 1326 struct spdk_nvme_tcp_r2t_hdr *r2t; 1327 1328 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1329 assert(rsp_pdu != NULL); 1330 1331 r2t = &rsp_pdu->hdr.r2t; 1332 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1333 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1334 1335 if (tqpair->host_hdgst_enable) { 1336 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1337 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1338 } 1339 1340 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1341 r2t->ttag = tcp_req->ttag; 1342 r2t->r2to = tcp_req->h2c_offset; 1343 r2t->r2tl = tcp_req->req.length; 1344 1345 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1346 1347 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1348 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1349 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1350 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1351 } 1352 1353 static void 1354 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1355 struct spdk_nvmf_tcp_qpair *tqpair, 1356 struct nvme_tcp_pdu *pdu) 1357 { 1358 struct spdk_nvmf_tcp_req *tcp_req; 1359 1360 tcp_req = pdu->req; 1361 assert(tcp_req != NULL); 1362 1363 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1364 1365 tcp_req->h2c_offset += pdu->data_len; 1366 1367 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1368 1369 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1370 * acknowledged before moving on. */ 1371 if (tcp_req->h2c_offset == tcp_req->req.length && 1372 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1373 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1374 nvmf_tcp_req_process(ttransport, tcp_req); 1375 } 1376 } 1377 1378 static void 1379 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1380 { 1381 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1382 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1383 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1384 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1385 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1386 DGET32(h2c_term_req->fei)); 1387 } 1388 } 1389 1390 static void 1391 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1392 struct nvme_tcp_pdu *pdu) 1393 { 1394 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1395 uint32_t error_offset = 0; 1396 enum spdk_nvme_tcp_term_req_fes fes; 1397 1398 1399 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1400 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1401 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1402 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1403 goto end; 1404 } 1405 1406 /* set the data buffer */ 1407 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1408 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1409 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1410 return; 1411 end: 1412 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1413 } 1414 1415 static void 1416 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1417 struct nvme_tcp_pdu *pdu) 1418 { 1419 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1420 1421 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1422 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1423 } 1424 1425 static void 1426 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1427 struct spdk_nvmf_tcp_transport *ttransport) 1428 { 1429 int rc = 0; 1430 struct nvme_tcp_pdu *pdu; 1431 uint32_t crc32c, error_offset = 0; 1432 enum spdk_nvme_tcp_term_req_fes fes; 1433 1434 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1435 pdu = &tqpair->pdu_in_progress; 1436 1437 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1438 /* check data digest if need */ 1439 if (pdu->ddgst_enable) { 1440 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1441 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1442 if (rc == 0) { 1443 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1444 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1445 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1446 return; 1447 1448 } 1449 } 1450 1451 switch (pdu->hdr.common.pdu_type) { 1452 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1453 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1454 break; 1455 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1456 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1457 break; 1458 1459 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1460 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1461 break; 1462 1463 default: 1464 /* The code should not go to here */ 1465 SPDK_ERRLOG("The code should not go to here\n"); 1466 break; 1467 } 1468 } 1469 1470 static void 1471 nvmf_tcp_send_icresp_complete(void *cb_arg) 1472 { 1473 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1474 1475 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1476 } 1477 1478 static void 1479 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1480 struct spdk_nvmf_tcp_qpair *tqpair, 1481 struct nvme_tcp_pdu *pdu) 1482 { 1483 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1484 struct nvme_tcp_pdu *rsp_pdu; 1485 struct spdk_nvme_tcp_ic_resp *ic_resp; 1486 uint32_t error_offset = 0; 1487 enum spdk_nvme_tcp_term_req_fes fes; 1488 1489 /* Only PFV 0 is defined currently */ 1490 if (ic_req->pfv != 0) { 1491 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1492 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1493 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1494 goto end; 1495 } 1496 1497 /* MAXR2T is 0's based */ 1498 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1499 1500 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1501 if (!tqpair->host_hdgst_enable) { 1502 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1503 } 1504 1505 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1506 if (!tqpair->host_ddgst_enable) { 1507 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1508 } 1509 1510 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1511 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1512 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1513 tqpair, 1514 tqpair->recv_buf_size); 1515 /* Not fatal. */ 1516 } 1517 1518 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1519 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1520 1521 rsp_pdu = &tqpair->mgmt_pdu; 1522 1523 ic_resp = &rsp_pdu->hdr.ic_resp; 1524 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1525 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1526 ic_resp->pfv = 0; 1527 ic_resp->cpda = tqpair->cpda; 1528 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1529 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1530 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1531 1532 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1533 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1534 1535 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1536 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1537 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1538 return; 1539 end: 1540 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1541 } 1542 1543 static void 1544 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1545 struct spdk_nvmf_tcp_transport *ttransport) 1546 { 1547 struct nvme_tcp_pdu *pdu; 1548 int rc; 1549 uint32_t crc32c, error_offset = 0; 1550 enum spdk_nvme_tcp_term_req_fes fes; 1551 1552 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1553 pdu = &tqpair->pdu_in_progress; 1554 1555 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1556 pdu->hdr.common.pdu_type); 1557 /* check header digest if needed */ 1558 if (pdu->has_hdgst) { 1559 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1560 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1561 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1562 if (rc == 0) { 1563 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1564 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1565 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1566 return; 1567 1568 } 1569 } 1570 1571 switch (pdu->hdr.common.pdu_type) { 1572 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1573 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1574 break; 1575 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1576 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1577 break; 1578 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1579 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1580 break; 1581 1582 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1583 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1584 break; 1585 1586 default: 1587 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1588 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1589 error_offset = 1; 1590 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1591 break; 1592 } 1593 } 1594 1595 static void 1596 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1597 { 1598 struct nvme_tcp_pdu *pdu; 1599 uint32_t error_offset = 0; 1600 enum spdk_nvme_tcp_term_req_fes fes; 1601 uint8_t expected_hlen, pdo; 1602 bool plen_error = false, pdo_error = false; 1603 1604 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1605 pdu = &tqpair->pdu_in_progress; 1606 1607 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1608 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1609 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1610 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1611 goto err; 1612 } 1613 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1614 if (pdu->hdr.common.plen != expected_hlen) { 1615 plen_error = true; 1616 } 1617 } else { 1618 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1619 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1620 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1621 goto err; 1622 } 1623 1624 switch (pdu->hdr.common.pdu_type) { 1625 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1626 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1627 pdo = pdu->hdr.common.pdo; 1628 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1629 pdo_error = true; 1630 break; 1631 } 1632 1633 if (pdu->hdr.common.plen < expected_hlen) { 1634 plen_error = true; 1635 } 1636 break; 1637 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1638 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1639 pdo = pdu->hdr.common.pdo; 1640 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1641 pdo_error = true; 1642 break; 1643 } 1644 if (pdu->hdr.common.plen < expected_hlen) { 1645 plen_error = true; 1646 } 1647 break; 1648 1649 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1650 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1651 if ((pdu->hdr.common.plen <= expected_hlen) || 1652 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1653 plen_error = true; 1654 } 1655 break; 1656 1657 default: 1658 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1659 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1660 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1661 goto err; 1662 } 1663 } 1664 1665 if (pdu->hdr.common.hlen != expected_hlen) { 1666 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1667 pdu->hdr.common.pdu_type, 1668 expected_hlen, pdu->hdr.common.hlen, tqpair); 1669 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1670 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1671 goto err; 1672 } else if (pdo_error) { 1673 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1674 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1675 } else if (plen_error) { 1676 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1677 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1678 goto err; 1679 } else { 1680 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1681 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1682 return; 1683 } 1684 err: 1685 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1686 } 1687 1688 static int 1689 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1690 int read_len) 1691 { 1692 int rc; 1693 1694 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1695 read_offset, read_len, pdu->dif_ctx); 1696 if (rc != 0) { 1697 SPDK_ERRLOG("DIF generate failed\n"); 1698 } 1699 1700 return rc; 1701 } 1702 1703 static int 1704 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1705 { 1706 int rc = 0; 1707 struct nvme_tcp_pdu *pdu; 1708 enum nvme_tcp_pdu_recv_state prev_state; 1709 uint32_t data_len; 1710 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1711 struct spdk_nvmf_tcp_transport, transport); 1712 1713 /* The loop here is to allow for several back-to-back state changes. */ 1714 do { 1715 prev_state = tqpair->recv_state; 1716 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1717 1718 pdu = &tqpair->pdu_in_progress; 1719 switch (tqpair->recv_state) { 1720 /* Wait for the common header */ 1721 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1722 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1723 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1724 return rc; 1725 } 1726 1727 rc = nvme_tcp_read_data(tqpair->sock, 1728 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1729 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1730 if (rc < 0) { 1731 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair); 1732 return NVME_TCP_PDU_FATAL; 1733 } else if (rc > 0) { 1734 pdu->ch_valid_bytes += rc; 1735 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1736 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1737 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1738 } 1739 } 1740 1741 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1742 return NVME_TCP_PDU_IN_PROGRESS; 1743 } 1744 1745 /* The command header of this PDU has now been read from the socket. */ 1746 nvmf_tcp_pdu_ch_handle(tqpair); 1747 break; 1748 /* Wait for the pdu specific header */ 1749 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1750 rc = nvme_tcp_read_data(tqpair->sock, 1751 pdu->psh_len - pdu->psh_valid_bytes, 1752 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1753 if (rc < 0) { 1754 return NVME_TCP_PDU_FATAL; 1755 } else if (rc > 0) { 1756 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1757 0, rc, 0, 0); 1758 pdu->psh_valid_bytes += rc; 1759 } 1760 1761 if (pdu->psh_valid_bytes < pdu->psh_len) { 1762 return NVME_TCP_PDU_IN_PROGRESS; 1763 } 1764 1765 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1766 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1767 break; 1768 /* Wait for the req slot */ 1769 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1770 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1771 break; 1772 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1773 /* check whether the data is valid, if not we just return */ 1774 if (!pdu->data_len) { 1775 return NVME_TCP_PDU_IN_PROGRESS; 1776 } 1777 1778 data_len = pdu->data_len; 1779 /* data digest */ 1780 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1781 tqpair->host_ddgst_enable)) { 1782 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1783 pdu->ddgst_enable = true; 1784 } 1785 1786 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1787 if (rc < 0) { 1788 return NVME_TCP_PDU_FATAL; 1789 } 1790 pdu->readv_offset += rc; 1791 1792 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1793 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1794 if (rc != 0) { 1795 return NVME_TCP_PDU_FATAL; 1796 } 1797 } 1798 1799 if (pdu->readv_offset < data_len) { 1800 return NVME_TCP_PDU_IN_PROGRESS; 1801 } 1802 1803 /* All of this PDU has now been read from the socket. */ 1804 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1805 break; 1806 case NVME_TCP_PDU_RECV_STATE_ERROR: 1807 if (!spdk_sock_is_connected(tqpair->sock)) { 1808 return NVME_TCP_PDU_FATAL; 1809 } 1810 break; 1811 default: 1812 assert(0); 1813 SPDK_ERRLOG("code should not come to here"); 1814 break; 1815 } 1816 } while (tqpair->recv_state != prev_state); 1817 1818 return rc; 1819 } 1820 1821 static int 1822 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 1823 struct spdk_nvmf_transport *transport, 1824 struct spdk_nvmf_transport_poll_group *group) 1825 { 1826 struct spdk_nvmf_request *req = &tcp_req->req; 1827 struct spdk_nvme_cmd *cmd; 1828 struct spdk_nvme_cpl *rsp; 1829 struct spdk_nvme_sgl_descriptor *sgl; 1830 uint32_t length; 1831 1832 cmd = &req->cmd->nvme_cmd; 1833 rsp = &req->rsp->nvme_cpl; 1834 sgl = &cmd->dptr.sgl1; 1835 1836 length = sgl->unkeyed.length; 1837 1838 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 1839 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 1840 if (length > transport->opts.max_io_size) { 1841 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 1842 length, transport->opts.max_io_size); 1843 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1844 return -1; 1845 } 1846 1847 /* fill request length and populate iovs */ 1848 req->length = length; 1849 1850 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", length); 1851 1852 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1853 req->dif.orig_length = length; 1854 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1855 req->dif.elba_length = length; 1856 } 1857 1858 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 1859 /* No available buffers. Queue this request up. */ 1860 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", 1861 tcp_req); 1862 return 0; 1863 } 1864 1865 /* backward compatible */ 1866 req->data = req->iov[0].iov_base; 1867 1868 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 1869 tcp_req, req->iovcnt, req->data); 1870 1871 return 0; 1872 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 1873 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 1874 uint64_t offset = sgl->address; 1875 uint32_t max_len = transport->opts.in_capsule_data_size; 1876 1877 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 1878 offset, length); 1879 1880 if (offset > max_len) { 1881 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 1882 offset, max_len); 1883 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 1884 return -1; 1885 } 1886 max_len -= (uint32_t)offset; 1887 1888 if (length > max_len) { 1889 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 1890 length, max_len); 1891 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1892 return -1; 1893 } 1894 1895 req->data = tcp_req->buf + offset; 1896 req->data_from_pool = false; 1897 req->length = length; 1898 1899 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1900 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1901 req->dif.elba_length = length; 1902 } 1903 1904 req->iov[0].iov_base = req->data; 1905 req->iov[0].iov_len = length; 1906 req->iovcnt = 1; 1907 1908 return 0; 1909 } 1910 1911 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 1912 sgl->generic.type, sgl->generic.subtype); 1913 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 1914 return -1; 1915 } 1916 1917 static inline enum spdk_nvme_media_error_status_code 1918 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 1919 enum spdk_nvme_media_error_status_code result; 1920 1921 switch (err_type) 1922 { 1923 case SPDK_DIF_REFTAG_ERROR: 1924 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 1925 break; 1926 case SPDK_DIF_APPTAG_ERROR: 1927 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 1928 break; 1929 case SPDK_DIF_GUARD_ERROR: 1930 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 1931 break; 1932 default: 1933 SPDK_UNREACHABLE(); 1934 break; 1935 } 1936 1937 return result; 1938 } 1939 1940 static void 1941 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 1942 struct spdk_nvmf_tcp_req *tcp_req) 1943 { 1944 struct nvme_tcp_pdu *rsp_pdu; 1945 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 1946 uint32_t plen, pdo, alignment; 1947 int rc; 1948 1949 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1950 1951 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1952 assert(rsp_pdu != NULL); 1953 1954 c2h_data = &rsp_pdu->hdr.c2h_data; 1955 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 1956 plen = c2h_data->common.hlen = sizeof(*c2h_data); 1957 1958 if (tqpair->host_hdgst_enable) { 1959 plen += SPDK_NVME_TCP_DIGEST_LEN; 1960 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1961 } 1962 1963 /* set the psh */ 1964 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1965 c2h_data->datal = tcp_req->req.length; 1966 c2h_data->datao = 0; 1967 1968 /* set the padding */ 1969 rsp_pdu->padding_len = 0; 1970 pdo = plen; 1971 if (tqpair->cpda) { 1972 alignment = (tqpair->cpda + 1) << 2; 1973 if (alignment > plen) { 1974 rsp_pdu->padding_len = alignment - plen; 1975 pdo = plen = alignment; 1976 } 1977 } 1978 1979 c2h_data->common.pdo = pdo; 1980 plen += c2h_data->datal; 1981 if (tqpair->host_ddgst_enable) { 1982 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 1983 plen += SPDK_NVME_TCP_DIGEST_LEN; 1984 } 1985 1986 c2h_data->common.plen = plen; 1987 1988 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1989 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1990 } 1991 1992 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1993 c2h_data->datao, c2h_data->datal); 1994 1995 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1996 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 1997 struct spdk_dif_error err_blk = {}; 1998 1999 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2000 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2001 if (rc != 0) { 2002 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2003 err_blk.err_type, err_blk.err_offset); 2004 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2005 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2006 nvmf_tcp_req_pdu_fini(tcp_req); 2007 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2008 return; 2009 } 2010 } 2011 2012 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2013 if (tqpair->qpair.transport->opts.c2h_success) { 2014 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2015 } 2016 2017 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2018 } 2019 2020 static int 2021 request_transfer_out(struct spdk_nvmf_request *req) 2022 { 2023 struct spdk_nvmf_tcp_req *tcp_req; 2024 struct spdk_nvmf_qpair *qpair; 2025 struct spdk_nvmf_tcp_qpair *tqpair; 2026 struct spdk_nvme_cpl *rsp; 2027 2028 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2029 2030 qpair = req->qpair; 2031 rsp = &req->rsp->nvme_cpl; 2032 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2033 2034 /* Advance our sq_head pointer */ 2035 if (qpair->sq_head == qpair->sq_head_max) { 2036 qpair->sq_head = 0; 2037 } else { 2038 qpair->sq_head++; 2039 } 2040 rsp->sqhd = qpair->sq_head; 2041 2042 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2043 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2044 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2045 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2046 } else { 2047 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2048 } 2049 2050 return 0; 2051 } 2052 2053 static void 2054 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2055 struct spdk_nvmf_tcp_req *tcp_req) 2056 { 2057 struct nvme_tcp_pdu *pdu; 2058 uint32_t plen = 0; 2059 2060 pdu = &tqpair->pdu_in_progress; 2061 plen = pdu->hdr.common.hlen; 2062 2063 if (tqpair->host_hdgst_enable) { 2064 plen += SPDK_NVME_TCP_DIGEST_LEN; 2065 } 2066 2067 if (pdu->hdr.common.plen != plen) { 2068 tcp_req->has_incapsule_data = true; 2069 } 2070 } 2071 2072 static bool 2073 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2074 struct spdk_nvmf_tcp_req *tcp_req) 2075 { 2076 struct spdk_nvmf_tcp_qpair *tqpair; 2077 int rc; 2078 enum spdk_nvmf_tcp_req_state prev_state; 2079 bool progress = false; 2080 struct spdk_nvmf_transport *transport = &ttransport->transport; 2081 struct spdk_nvmf_transport_poll_group *group; 2082 2083 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2084 group = &tqpair->group->group; 2085 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2086 2087 /* The loop here is to allow for several back-to-back state changes. */ 2088 do { 2089 prev_state = tcp_req->state; 2090 2091 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2092 tqpair); 2093 2094 switch (tcp_req->state) { 2095 case TCP_REQUEST_STATE_FREE: 2096 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2097 * to escape this state. */ 2098 break; 2099 case TCP_REQUEST_STATE_NEW: 2100 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2101 2102 /* copy the cmd from the receive pdu */ 2103 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2104 2105 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2106 tcp_req->req.dif.dif_insert_or_strip = true; 2107 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2108 } 2109 2110 /* The next state transition depends on the data transfer needs of this request. */ 2111 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2112 2113 /* If no data to transfer, ready to execute. */ 2114 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2115 /* Reset the tqpair receving pdu state */ 2116 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2117 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2118 break; 2119 } 2120 2121 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2122 2123 if (!tcp_req->has_incapsule_data) { 2124 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2125 } 2126 2127 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2128 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2129 break; 2130 case TCP_REQUEST_STATE_NEED_BUFFER: 2131 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2132 2133 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2134 2135 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2136 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2137 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2138 tcp_req, tqpair); 2139 /* This request needs to wait in line to obtain a buffer */ 2140 break; 2141 } 2142 2143 /* Try to get a data buffer */ 2144 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2145 if (rc < 0) { 2146 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2147 /* Reset the tqpair receving pdu state */ 2148 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2149 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2150 break; 2151 } 2152 2153 if (!tcp_req->req.data) { 2154 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2155 tcp_req, tqpair); 2156 /* No buffers available. */ 2157 break; 2158 } 2159 2160 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2161 2162 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2163 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2164 if (tcp_req->req.data_from_pool) { 2165 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2166 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2167 } else { 2168 struct nvme_tcp_pdu *pdu; 2169 2170 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2171 2172 pdu = &tqpair->pdu_in_progress; 2173 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2174 tqpair); 2175 /* No need to send r2t, contained in the capsuled data */ 2176 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2177 0, tcp_req->req.length); 2178 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2179 } 2180 break; 2181 } 2182 2183 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2184 break; 2185 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2186 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2187 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2188 break; 2189 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2190 2191 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2192 (uintptr_t)tcp_req, 0); 2193 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2194 * to escape this state. */ 2195 break; 2196 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2197 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2198 2199 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2200 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2201 tcp_req->req.length = tcp_req->req.dif.elba_length; 2202 } 2203 2204 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2205 spdk_nvmf_request_exec(&tcp_req->req); 2206 break; 2207 case TCP_REQUEST_STATE_EXECUTING: 2208 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2209 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2210 * to escape this state. */ 2211 break; 2212 case TCP_REQUEST_STATE_EXECUTED: 2213 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2214 2215 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2216 tcp_req->req.length = tcp_req->req.dif.orig_length; 2217 } 2218 2219 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2220 break; 2221 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2222 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2223 rc = request_transfer_out(&tcp_req->req); 2224 assert(rc == 0); /* No good way to handle this currently */ 2225 break; 2226 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2227 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2228 (uintptr_t)tcp_req, 2229 0); 2230 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2231 * to escape this state. */ 2232 break; 2233 case TCP_REQUEST_STATE_COMPLETED: 2234 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2235 if (tcp_req->req.data_from_pool) { 2236 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2237 } 2238 tcp_req->req.length = 0; 2239 tcp_req->req.iovcnt = 0; 2240 tcp_req->req.data = NULL; 2241 2242 nvmf_tcp_req_pdu_fini(tcp_req); 2243 2244 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2245 break; 2246 case TCP_REQUEST_NUM_STATES: 2247 default: 2248 assert(0); 2249 break; 2250 } 2251 2252 if (tcp_req->state != prev_state) { 2253 progress = true; 2254 } 2255 } while (tcp_req->state != prev_state); 2256 2257 return progress; 2258 } 2259 2260 static void 2261 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2262 { 2263 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2264 int rc; 2265 2266 assert(tqpair != NULL); 2267 rc = nvmf_tcp_sock_process(tqpair); 2268 2269 /* If there was a new socket error, disconnect */ 2270 if (rc < 0) { 2271 nvmf_tcp_qpair_disconnect(tqpair); 2272 } 2273 } 2274 2275 static int 2276 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2277 struct spdk_nvmf_qpair *qpair) 2278 { 2279 struct spdk_nvmf_tcp_poll_group *tgroup; 2280 struct spdk_nvmf_tcp_qpair *tqpair; 2281 int rc; 2282 2283 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2284 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2285 2286 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2287 nvmf_tcp_sock_cb, tqpair); 2288 if (rc != 0) { 2289 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2290 spdk_strerror(errno), errno); 2291 return -1; 2292 } 2293 2294 rc = nvmf_tcp_qpair_sock_init(tqpair); 2295 if (rc != 0) { 2296 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2297 return -1; 2298 } 2299 2300 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2301 if (rc < 0) { 2302 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2303 return -1; 2304 } 2305 2306 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2307 if (rc < 0) { 2308 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2309 return -1; 2310 } 2311 2312 tqpair->group = tgroup; 2313 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2314 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2315 2316 return 0; 2317 } 2318 2319 static int 2320 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2321 struct spdk_nvmf_qpair *qpair) 2322 { 2323 struct spdk_nvmf_tcp_poll_group *tgroup; 2324 struct spdk_nvmf_tcp_qpair *tqpair; 2325 int rc; 2326 2327 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2328 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2329 2330 assert(tqpair->group == tgroup); 2331 2332 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2333 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2334 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2335 } else { 2336 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2337 } 2338 2339 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2340 if (rc != 0) { 2341 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2342 spdk_strerror(errno), errno); 2343 } 2344 2345 return rc; 2346 } 2347 2348 static int 2349 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2350 { 2351 struct spdk_nvmf_tcp_transport *ttransport; 2352 struct spdk_nvmf_tcp_req *tcp_req; 2353 2354 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2355 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2356 2357 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2358 nvmf_tcp_req_process(ttransport, tcp_req); 2359 2360 return 0; 2361 } 2362 2363 static void 2364 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2365 { 2366 struct spdk_nvmf_tcp_qpair *tqpair; 2367 2368 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Qpair: %p\n", qpair); 2369 2370 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2371 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2372 nvmf_tcp_qpair_destroy(tqpair); 2373 } 2374 2375 static int 2376 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2377 { 2378 struct spdk_nvmf_tcp_poll_group *tgroup; 2379 int rc; 2380 struct spdk_nvmf_request *req, *req_tmp; 2381 struct spdk_nvmf_tcp_req *tcp_req; 2382 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2383 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2384 struct spdk_nvmf_tcp_transport, transport); 2385 2386 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2387 2388 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2389 return 0; 2390 } 2391 2392 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2393 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2394 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2395 break; 2396 } 2397 } 2398 2399 rc = spdk_sock_group_poll(tgroup->sock_group); 2400 if (rc < 0) { 2401 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2402 } 2403 2404 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2405 nvmf_tcp_sock_process(tqpair); 2406 } 2407 2408 return rc; 2409 } 2410 2411 static int 2412 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2413 struct spdk_nvme_transport_id *trid, bool peer) 2414 { 2415 struct spdk_nvmf_tcp_qpair *tqpair; 2416 uint16_t port; 2417 2418 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2419 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2420 2421 if (peer) { 2422 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2423 port = tqpair->initiator_port; 2424 } else { 2425 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2426 port = tqpair->target_port; 2427 } 2428 2429 if (spdk_sock_is_ipv4(tqpair->sock)) { 2430 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2431 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2432 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2433 } else { 2434 return -1; 2435 } 2436 2437 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2438 return 0; 2439 } 2440 2441 static int 2442 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2443 struct spdk_nvme_transport_id *trid) 2444 { 2445 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2446 } 2447 2448 static int 2449 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2450 struct spdk_nvme_transport_id *trid) 2451 { 2452 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2453 } 2454 2455 static int 2456 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2457 struct spdk_nvme_transport_id *trid) 2458 { 2459 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2460 } 2461 2462 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2463 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2464 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2465 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2466 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2467 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2468 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2469 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2470 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 2471 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2472 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 2473 2474 static void 2475 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2476 { 2477 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2478 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2479 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2480 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2481 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2482 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2483 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2484 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2485 opts->c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 2486 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2487 opts->sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 2488 } 2489 2490 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2491 .name = "TCP", 2492 .type = SPDK_NVME_TRANSPORT_TCP, 2493 .opts_init = nvmf_tcp_opts_init, 2494 .create = nvmf_tcp_create, 2495 .destroy = nvmf_tcp_destroy, 2496 2497 .listen = nvmf_tcp_listen, 2498 .stop_listen = nvmf_tcp_stop_listen, 2499 .accept = nvmf_tcp_accept, 2500 2501 .listener_discover = nvmf_tcp_discover, 2502 2503 .poll_group_create = nvmf_tcp_poll_group_create, 2504 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2505 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2506 .poll_group_add = nvmf_tcp_poll_group_add, 2507 .poll_group_remove = nvmf_tcp_poll_group_remove, 2508 .poll_group_poll = nvmf_tcp_poll_group_poll, 2509 2510 .req_free = nvmf_tcp_req_free, 2511 .req_complete = nvmf_tcp_req_complete, 2512 2513 .qpair_fini = nvmf_tcp_close_qpair, 2514 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2515 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2516 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2517 }; 2518 2519 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2520 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2521