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 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 277 struct spdk_nvmf_tcp_req *tcp_req); 278 279 static void 280 spdk_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 spdk_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 spdk_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 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 347 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 348 } 349 350 static int 351 spdk_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 spdk_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 spdk_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 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 379 spdk_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 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 389 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 390 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 391 spdk_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 spdk_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 spdk_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 spdk_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 spdk_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_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, 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 spdk_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 _spdk_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 _spdk_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 = _spdk_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 _spdk_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 (_spdk_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 spdk_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; 597 598 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 599 600 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 601 if (trsvcid_int < 0) { 602 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 603 return -EINVAL; 604 } 605 606 pthread_mutex_lock(&ttransport->lock); 607 port = calloc(1, sizeof(*port)); 608 if (!port) { 609 SPDK_ERRLOG("Port allocation failed\n"); 610 pthread_mutex_unlock(&ttransport->lock); 611 return -ENOMEM; 612 } 613 614 port->trid = trid; 615 port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int, NULL); 616 if (port->listen_sock == NULL) { 617 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 618 trid->traddr, trsvcid_int, 619 spdk_strerror(errno), errno); 620 free(port); 621 pthread_mutex_unlock(&ttransport->lock); 622 return -errno; 623 } 624 625 if (spdk_sock_is_ipv4(port->listen_sock)) { 626 adrfam = SPDK_NVMF_ADRFAM_IPV4; 627 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 628 adrfam = SPDK_NVMF_ADRFAM_IPV6; 629 } else { 630 SPDK_ERRLOG("Unhandled socket type\n"); 631 adrfam = 0; 632 } 633 634 if (adrfam != trid->adrfam) { 635 SPDK_ERRLOG("Socket address family mismatch\n"); 636 spdk_sock_close(&port->listen_sock); 637 free(port); 638 pthread_mutex_unlock(&ttransport->lock); 639 return -EINVAL; 640 } 641 642 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 643 trid->traddr, trid->trsvcid); 644 645 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 646 pthread_mutex_unlock(&ttransport->lock); 647 return 0; 648 } 649 650 static void 651 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 652 const struct spdk_nvme_transport_id *trid) 653 { 654 struct spdk_nvmf_tcp_transport *ttransport; 655 struct spdk_nvmf_tcp_port *port; 656 657 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 658 659 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 660 trid->traddr, trid->trsvcid); 661 662 pthread_mutex_lock(&ttransport->lock); 663 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 664 if (port) { 665 TAILQ_REMOVE(&ttransport->ports, port, link); 666 spdk_sock_close(&port->listen_sock); 667 free(port); 668 } 669 670 pthread_mutex_unlock(&ttransport->lock); 671 } 672 673 static void spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 674 enum nvme_tcp_pdu_recv_state state); 675 676 static void 677 spdk_nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 678 { 679 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Disconnecting qpair %p\n", tqpair); 680 681 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 682 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 683 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 684 spdk_poller_unregister(&tqpair->timeout_poller); 685 686 /* This will end up calling spdk_nvmf_tcp_close_qpair */ 687 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 688 } 689 } 690 691 static void 692 _pdu_write_done(void *_pdu, int err) 693 { 694 struct nvme_tcp_pdu *pdu = _pdu; 695 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 696 697 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 698 699 if (err != 0) { 700 spdk_nvmf_tcp_qpair_disconnect(tqpair); 701 return; 702 } 703 704 assert(pdu->cb_fn != NULL); 705 pdu->cb_fn(pdu->cb_arg); 706 } 707 708 static void 709 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 710 struct nvme_tcp_pdu *pdu, 711 nvme_tcp_qpair_xfer_complete_cb cb_fn, 712 void *cb_arg) 713 { 714 int hlen; 715 uint32_t crc32c; 716 uint32_t mapped_length = 0; 717 ssize_t rc; 718 719 assert(&tqpair->pdu_in_progress != pdu); 720 721 hlen = pdu->hdr.common.hlen; 722 723 /* Header Digest */ 724 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 725 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 726 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 727 } 728 729 /* Data Digest */ 730 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 731 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 732 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 733 } 734 735 pdu->cb_fn = cb_fn; 736 pdu->cb_arg = cb_arg; 737 738 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 739 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 740 &mapped_length); 741 pdu->sock_req.cb_fn = _pdu_write_done; 742 pdu->sock_req.cb_arg = pdu; 743 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 744 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 745 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 746 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 747 if (rc == mapped_length) { 748 _pdu_write_done(pdu, 0); 749 } else { 750 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 751 _pdu_write_done(pdu, -1); 752 } 753 } else { 754 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 755 } 756 } 757 758 static int 759 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 760 { 761 uint32_t i; 762 struct spdk_nvmf_transport_opts *opts; 763 uint32_t in_capsule_data_size; 764 765 opts = &tqpair->qpair.transport->opts; 766 767 in_capsule_data_size = opts->in_capsule_data_size; 768 if (opts->dif_insert_or_strip) { 769 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 770 } 771 772 tqpair->resource_count = opts->max_queue_depth; 773 774 tqpair->mgmt_pdu.qpair = tqpair; 775 776 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 777 if (!tqpair->reqs) { 778 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 779 return -1; 780 } 781 782 if (in_capsule_data_size) { 783 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 784 NULL, SPDK_ENV_LCORE_ID_ANY, 785 SPDK_MALLOC_DMA); 786 if (!tqpair->bufs) { 787 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 788 return -1; 789 } 790 } 791 792 tqpair->pdus = spdk_dma_malloc(tqpair->resource_count * sizeof(*tqpair->pdus), 0x1000, NULL); 793 if (!tqpair->pdus) { 794 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 795 return -1; 796 } 797 798 for (i = 0; i < tqpair->resource_count; i++) { 799 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 800 801 tcp_req->ttag = i + 1; 802 tcp_req->req.qpair = &tqpair->qpair; 803 804 tcp_req->pdu = &tqpair->pdus[i]; 805 tcp_req->pdu->qpair = tqpair; 806 807 /* Set up memory to receive commands */ 808 if (tqpair->bufs) { 809 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 810 } 811 812 /* Set the cmdn and rsp */ 813 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 814 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 815 816 /* Initialize request state to FREE */ 817 tcp_req->state = TCP_REQUEST_STATE_FREE; 818 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 819 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 820 } 821 822 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 823 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 824 825 return 0; 826 } 827 828 static int 829 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 830 { 831 struct spdk_nvmf_tcp_qpair *tqpair; 832 int i; 833 834 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 835 836 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 837 838 TAILQ_INIT(&tqpair->send_queue); 839 840 /* Initialise request state queues of the qpair */ 841 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 842 TAILQ_INIT(&tqpair->state_queue[i]); 843 } 844 845 tqpair->host_hdgst_enable = true; 846 tqpair->host_ddgst_enable = true; 847 848 return 0; 849 } 850 851 static int 852 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 853 { 854 int rc; 855 856 /* set low water mark */ 857 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr)); 858 if (rc != 0) { 859 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 860 return rc; 861 } 862 863 return 0; 864 } 865 866 static void 867 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 868 struct spdk_nvmf_tcp_port *port, 869 struct spdk_sock *sock, 870 new_qpair_fn cb_fn, void *cb_arg) 871 { 872 struct spdk_nvmf_tcp_qpair *tqpair; 873 int rc; 874 875 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 876 port->trid->traddr, port->trid->trsvcid); 877 878 if (transport->opts.sock_priority) { 879 rc = spdk_sock_set_priority(sock, transport->opts.sock_priority); 880 if (rc) { 881 SPDK_ERRLOG("Failed to set the priority of the socket\n"); 882 spdk_sock_close(&sock); 883 return; 884 } 885 } 886 887 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 888 if (tqpair == NULL) { 889 SPDK_ERRLOG("Could not allocate new connection.\n"); 890 spdk_sock_close(&sock); 891 return; 892 } 893 894 tqpair->sock = sock; 895 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 896 tqpair->port = port; 897 tqpair->qpair.transport = transport; 898 899 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 900 sizeof(tqpair->target_addr), &tqpair->target_port, 901 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 902 &tqpair->initiator_port); 903 if (rc < 0) { 904 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 905 spdk_nvmf_tcp_qpair_destroy(tqpair); 906 return; 907 } 908 909 cb_fn(&tqpair->qpair, cb_arg); 910 } 911 912 static void 913 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port, 914 new_qpair_fn cb_fn, void *cb_arg) 915 { 916 struct spdk_sock *sock; 917 int i; 918 919 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 920 sock = spdk_sock_accept(port->listen_sock); 921 if (sock == NULL) { 922 break; 923 } 924 _spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn, cb_arg); 925 } 926 } 927 928 static void 929 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn, void *cb_arg) 930 { 931 struct spdk_nvmf_tcp_transport *ttransport; 932 struct spdk_nvmf_tcp_port *port; 933 934 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 935 936 TAILQ_FOREACH(port, &ttransport->ports, link) { 937 spdk_nvmf_tcp_port_accept(transport, port, cb_fn, cb_arg); 938 } 939 } 940 941 static void 942 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 943 struct spdk_nvme_transport_id *trid, 944 struct spdk_nvmf_discovery_log_page_entry *entry) 945 { 946 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 947 entry->adrfam = trid->adrfam; 948 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 949 950 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 951 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 952 953 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 954 } 955 956 static struct spdk_nvmf_transport_poll_group * 957 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 958 { 959 struct spdk_nvmf_tcp_poll_group *tgroup; 960 961 tgroup = calloc(1, sizeof(*tgroup)); 962 if (!tgroup) { 963 return NULL; 964 } 965 966 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 967 if (!tgroup->sock_group) { 968 goto cleanup; 969 } 970 971 TAILQ_INIT(&tgroup->qpairs); 972 TAILQ_INIT(&tgroup->await_req); 973 974 return &tgroup->group; 975 976 cleanup: 977 free(tgroup); 978 return NULL; 979 } 980 981 static struct spdk_nvmf_transport_poll_group * 982 spdk_nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 983 { 984 struct spdk_nvmf_tcp_qpair *tqpair; 985 struct spdk_sock_group *group = NULL; 986 int rc; 987 988 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 989 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 990 if (!rc && group != NULL) { 991 return spdk_sock_group_get_ctx(group); 992 } 993 994 return NULL; 995 } 996 997 static void 998 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 999 { 1000 struct spdk_nvmf_tcp_poll_group *tgroup; 1001 1002 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1003 spdk_sock_group_close(&tgroup->sock_group); 1004 1005 free(tgroup); 1006 } 1007 1008 static void 1009 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1010 enum nvme_tcp_pdu_recv_state state) 1011 { 1012 if (tqpair->recv_state == state) { 1013 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1014 tqpair, state); 1015 return; 1016 } 1017 1018 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1019 /* When leaving the await req state, move the qpair to the main list */ 1020 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1021 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1022 } 1023 1024 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1025 tqpair->recv_state = state; 1026 1027 switch (state) { 1028 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1029 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1030 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1031 break; 1032 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1033 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1034 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1035 break; 1036 case NVME_TCP_PDU_RECV_STATE_ERROR: 1037 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1038 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1039 break; 1040 default: 1041 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1042 abort(); 1043 break; 1044 } 1045 } 1046 1047 static int 1048 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx) 1049 { 1050 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1051 1052 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1053 1054 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1055 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1056 1057 spdk_nvmf_tcp_qpair_disconnect(tqpair); 1058 return 0; 1059 } 1060 1061 static void 1062 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1063 { 1064 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1065 1066 if (!tqpair->timeout_poller) { 1067 tqpair->timeout_poller = SPDK_POLLER_REGISTER(spdk_nvmf_tcp_qpair_handle_timeout, tqpair, 1068 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1069 } 1070 } 1071 1072 static void 1073 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1074 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1075 { 1076 struct nvme_tcp_pdu *rsp_pdu; 1077 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1078 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1079 uint32_t copy_len; 1080 1081 rsp_pdu = &tqpair->mgmt_pdu; 1082 1083 c2h_term_req = &rsp_pdu->hdr.term_req; 1084 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1085 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1086 1087 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1088 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1089 DSET32(&c2h_term_req->fei, error_offset); 1090 } 1091 1092 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1093 1094 /* Copy the error info into the buffer */ 1095 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1096 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1097 1098 /* Contain the header of the wrong received pdu */ 1099 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1100 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1101 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair); 1102 } 1103 1104 static void 1105 spdk_nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1106 struct spdk_nvmf_tcp_qpair *tqpair, 1107 struct nvme_tcp_pdu *pdu) 1108 { 1109 struct spdk_nvmf_tcp_req *tcp_req; 1110 1111 assert(pdu->psh_valid_bytes == pdu->psh_len); 1112 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1113 1114 tcp_req = spdk_nvmf_tcp_req_get(tqpair); 1115 if (!tcp_req) { 1116 /* Directly return and make the allocation retry again */ 1117 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1118 return; 1119 } 1120 1121 /* The host sent more commands than the maximum queue depth. */ 1122 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1123 spdk_nvmf_tcp_qpair_disconnect(tqpair); 1124 return; 1125 } 1126 1127 pdu->req = tcp_req; 1128 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1129 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1130 } 1131 1132 static void 1133 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1134 struct spdk_nvmf_tcp_qpair *tqpair, 1135 struct nvme_tcp_pdu *pdu) 1136 { 1137 struct spdk_nvmf_tcp_req *tcp_req; 1138 struct spdk_nvme_tcp_cmd *capsule_cmd; 1139 uint32_t error_offset = 0; 1140 enum spdk_nvme_tcp_term_req_fes fes; 1141 1142 capsule_cmd = &pdu->hdr.capsule_cmd; 1143 tcp_req = pdu->req; 1144 assert(tcp_req != NULL); 1145 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1146 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1147 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1148 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1149 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1150 goto err; 1151 } 1152 1153 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1154 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1155 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1156 1157 return; 1158 err: 1159 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1160 } 1161 1162 static int 1163 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1164 enum spdk_nvmf_tcp_req_state state, 1165 uint16_t cid, uint16_t tag, 1166 struct spdk_nvmf_tcp_req **req) 1167 { 1168 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1169 1170 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) { 1171 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1172 continue; 1173 } 1174 1175 if (tcp_req->ttag == tag) { 1176 *req = tcp_req; 1177 return 0; 1178 } 1179 1180 *req = NULL; 1181 return -1; 1182 } 1183 1184 /* Didn't find it, but not an error */ 1185 *req = NULL; 1186 return 0; 1187 } 1188 1189 static void 1190 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1191 struct spdk_nvmf_tcp_qpair *tqpair, 1192 struct nvme_tcp_pdu *pdu) 1193 { 1194 struct spdk_nvmf_tcp_req *tcp_req; 1195 uint32_t error_offset = 0; 1196 enum spdk_nvme_tcp_term_req_fes fes = 0; 1197 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1198 int rc; 1199 1200 h2c_data = &pdu->hdr.h2c_data; 1201 1202 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1203 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1204 1205 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1206 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1207 if (rc == 0 && tcp_req == NULL) { 1208 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1209 h2c_data->ttag, &tcp_req); 1210 } 1211 1212 if (!tcp_req) { 1213 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1214 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1215 if (rc == 0) { 1216 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1217 } else { 1218 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1219 } 1220 goto err; 1221 } 1222 1223 if (tcp_req->h2c_offset != h2c_data->datao) { 1224 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1225 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1226 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1227 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1228 goto err; 1229 } 1230 1231 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1232 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1233 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1234 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1235 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1236 goto err; 1237 } 1238 1239 pdu->req = tcp_req; 1240 1241 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1242 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1243 } 1244 1245 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1246 h2c_data->datao, h2c_data->datal); 1247 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1248 return; 1249 1250 err: 1251 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1252 } 1253 1254 static void 1255 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1256 { 1257 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1258 nvmf_tcp_request_free(tcp_req); 1259 } 1260 1261 static void 1262 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1263 struct spdk_nvmf_tcp_qpair *tqpair) 1264 { 1265 struct nvme_tcp_pdu *rsp_pdu; 1266 struct spdk_nvme_tcp_rsp *capsule_resp; 1267 1268 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1269 1270 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1271 assert(rsp_pdu != NULL); 1272 1273 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1274 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1275 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1276 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1277 if (tqpair->host_hdgst_enable) { 1278 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1279 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1280 } 1281 1282 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req); 1283 } 1284 1285 static void 1286 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1287 { 1288 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1289 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1290 struct spdk_nvmf_tcp_qpair, qpair); 1291 1292 assert(tqpair != NULL); 1293 if (tqpair->qpair.transport->opts.c2h_success) { 1294 nvmf_tcp_request_free(tcp_req); 1295 } else { 1296 nvmf_tcp_req_pdu_fini(tcp_req); 1297 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1298 } 1299 } 1300 1301 static void 1302 spdk_nvmf_tcp_r2t_complete(void *cb_arg) 1303 { 1304 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1305 struct spdk_nvmf_tcp_transport *ttransport; 1306 1307 nvmf_tcp_req_pdu_fini(tcp_req); 1308 1309 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1310 struct spdk_nvmf_tcp_transport, transport); 1311 1312 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1313 1314 if (tcp_req->h2c_offset == tcp_req->req.length) { 1315 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1316 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1317 } 1318 } 1319 1320 static void 1321 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1322 struct spdk_nvmf_tcp_req *tcp_req) 1323 { 1324 struct nvme_tcp_pdu *rsp_pdu; 1325 struct spdk_nvme_tcp_r2t_hdr *r2t; 1326 1327 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1328 assert(rsp_pdu != NULL); 1329 1330 r2t = &rsp_pdu->hdr.r2t; 1331 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1332 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1333 1334 if (tqpair->host_hdgst_enable) { 1335 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1336 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1337 } 1338 1339 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1340 r2t->ttag = tcp_req->ttag; 1341 r2t->r2to = tcp_req->h2c_offset; 1342 r2t->r2tl = tcp_req->req.length; 1343 1344 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1345 1346 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1347 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1348 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1349 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_r2t_complete, tcp_req); 1350 } 1351 1352 static void 1353 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1354 struct spdk_nvmf_tcp_qpair *tqpair, 1355 struct nvme_tcp_pdu *pdu) 1356 { 1357 struct spdk_nvmf_tcp_req *tcp_req; 1358 1359 tcp_req = pdu->req; 1360 assert(tcp_req != NULL); 1361 1362 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1363 1364 tcp_req->h2c_offset += pdu->data_len; 1365 1366 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1367 1368 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1369 * acknowledged before moving on. */ 1370 if (tcp_req->h2c_offset == tcp_req->req.length && 1371 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1372 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1373 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1374 } 1375 } 1376 1377 static void 1378 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1379 { 1380 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1381 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1382 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1383 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1384 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1385 DGET32(h2c_term_req->fei)); 1386 } 1387 } 1388 1389 static void 1390 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1391 struct nvme_tcp_pdu *pdu) 1392 { 1393 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1394 uint32_t error_offset = 0; 1395 enum spdk_nvme_tcp_term_req_fes fes; 1396 1397 1398 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1399 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1400 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1401 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1402 goto end; 1403 } 1404 1405 /* set the data buffer */ 1406 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1407 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1408 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1409 return; 1410 end: 1411 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1412 } 1413 1414 static void 1415 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1416 struct nvme_tcp_pdu *pdu) 1417 { 1418 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1419 1420 spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1421 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1422 } 1423 1424 static void 1425 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1426 struct spdk_nvmf_tcp_transport *ttransport) 1427 { 1428 int rc = 0; 1429 struct nvme_tcp_pdu *pdu; 1430 uint32_t crc32c, error_offset = 0; 1431 enum spdk_nvme_tcp_term_req_fes fes; 1432 1433 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1434 pdu = &tqpair->pdu_in_progress; 1435 1436 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1437 /* check data digest if need */ 1438 if (pdu->ddgst_enable) { 1439 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1440 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1441 if (rc == 0) { 1442 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1443 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1444 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1445 return; 1446 1447 } 1448 } 1449 1450 switch (pdu->hdr.common.pdu_type) { 1451 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1452 spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1453 break; 1454 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1455 spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1456 break; 1457 1458 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1459 spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1460 break; 1461 1462 default: 1463 /* The code should not go to here */ 1464 SPDK_ERRLOG("The code should not go to here\n"); 1465 break; 1466 } 1467 } 1468 1469 static void 1470 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg) 1471 { 1472 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1473 1474 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1475 } 1476 1477 static void 1478 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1479 struct spdk_nvmf_tcp_qpair *tqpair, 1480 struct nvme_tcp_pdu *pdu) 1481 { 1482 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1483 struct nvme_tcp_pdu *rsp_pdu; 1484 struct spdk_nvme_tcp_ic_resp *ic_resp; 1485 uint32_t error_offset = 0; 1486 enum spdk_nvme_tcp_term_req_fes fes; 1487 1488 /* Only PFV 0 is defined currently */ 1489 if (ic_req->pfv != 0) { 1490 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1491 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1492 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1493 goto end; 1494 } 1495 1496 /* MAXR2T is 0's based */ 1497 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1498 1499 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1500 if (!tqpair->host_hdgst_enable) { 1501 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1502 } 1503 1504 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1505 if (!tqpair->host_ddgst_enable) { 1506 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1507 } 1508 1509 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1510 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1511 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1512 tqpair, 1513 tqpair->recv_buf_size); 1514 /* Not fatal. */ 1515 } 1516 1517 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1518 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1519 1520 rsp_pdu = &tqpair->mgmt_pdu; 1521 1522 ic_resp = &rsp_pdu->hdr.ic_resp; 1523 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1524 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1525 ic_resp->pfv = 0; 1526 ic_resp->cpda = tqpair->cpda; 1527 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1528 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1529 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1530 1531 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1532 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1533 1534 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1535 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair); 1536 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1537 return; 1538 end: 1539 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1540 } 1541 1542 static void 1543 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1544 struct spdk_nvmf_tcp_transport *ttransport) 1545 { 1546 struct nvme_tcp_pdu *pdu; 1547 int rc; 1548 uint32_t crc32c, error_offset = 0; 1549 enum spdk_nvme_tcp_term_req_fes fes; 1550 1551 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1552 pdu = &tqpair->pdu_in_progress; 1553 1554 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1555 pdu->hdr.common.pdu_type); 1556 /* check header digest if needed */ 1557 if (pdu->has_hdgst) { 1558 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1559 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1560 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1561 if (rc == 0) { 1562 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1563 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1564 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1565 return; 1566 1567 } 1568 } 1569 1570 switch (pdu->hdr.common.pdu_type) { 1571 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1572 spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1573 break; 1574 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1575 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1576 break; 1577 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1578 spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1579 break; 1580 1581 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1582 spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1583 break; 1584 1585 default: 1586 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1587 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1588 error_offset = 1; 1589 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1590 break; 1591 } 1592 } 1593 1594 static void 1595 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1596 { 1597 struct nvme_tcp_pdu *pdu; 1598 uint32_t error_offset = 0; 1599 enum spdk_nvme_tcp_term_req_fes fes; 1600 uint8_t expected_hlen, pdo; 1601 bool plen_error = false, pdo_error = false; 1602 1603 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1604 pdu = &tqpair->pdu_in_progress; 1605 1606 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1607 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1608 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1609 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1610 goto err; 1611 } 1612 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1613 if (pdu->hdr.common.plen != expected_hlen) { 1614 plen_error = true; 1615 } 1616 } else { 1617 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1618 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1619 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1620 goto err; 1621 } 1622 1623 switch (pdu->hdr.common.pdu_type) { 1624 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1625 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1626 pdo = pdu->hdr.common.pdo; 1627 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1628 pdo_error = true; 1629 break; 1630 } 1631 1632 if (pdu->hdr.common.plen < expected_hlen) { 1633 plen_error = true; 1634 } 1635 break; 1636 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1637 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1638 pdo = pdu->hdr.common.pdo; 1639 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1640 pdo_error = true; 1641 break; 1642 } 1643 if (pdu->hdr.common.plen < expected_hlen) { 1644 plen_error = true; 1645 } 1646 break; 1647 1648 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1649 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1650 if ((pdu->hdr.common.plen <= expected_hlen) || 1651 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1652 plen_error = true; 1653 } 1654 break; 1655 1656 default: 1657 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1658 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1659 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1660 goto err; 1661 } 1662 } 1663 1664 if (pdu->hdr.common.hlen != expected_hlen) { 1665 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1666 pdu->hdr.common.pdu_type, 1667 expected_hlen, pdu->hdr.common.hlen, tqpair); 1668 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1669 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1670 goto err; 1671 } else if (pdo_error) { 1672 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1673 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1674 } else if (plen_error) { 1675 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1676 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1677 goto err; 1678 } else { 1679 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1680 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1681 return; 1682 } 1683 err: 1684 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1685 } 1686 1687 static int 1688 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1689 int read_len) 1690 { 1691 int rc; 1692 1693 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1694 read_offset, read_len, pdu->dif_ctx); 1695 if (rc != 0) { 1696 SPDK_ERRLOG("DIF generate failed\n"); 1697 } 1698 1699 return rc; 1700 } 1701 1702 static int 1703 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1704 { 1705 int rc = 0; 1706 struct nvme_tcp_pdu *pdu; 1707 enum nvme_tcp_pdu_recv_state prev_state; 1708 uint32_t data_len; 1709 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1710 struct spdk_nvmf_tcp_transport, transport); 1711 1712 /* The loop here is to allow for several back-to-back state changes. */ 1713 do { 1714 prev_state = tqpair->recv_state; 1715 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1716 1717 pdu = &tqpair->pdu_in_progress; 1718 switch (tqpair->recv_state) { 1719 /* Wait for the common header */ 1720 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1721 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1722 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1723 return rc; 1724 } 1725 1726 rc = nvme_tcp_read_data(tqpair->sock, 1727 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1728 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1729 if (rc < 0) { 1730 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair); 1731 return NVME_TCP_PDU_FATAL; 1732 } else if (rc > 0) { 1733 pdu->ch_valid_bytes += rc; 1734 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1735 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1736 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1737 } 1738 } 1739 1740 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1741 return NVME_TCP_PDU_IN_PROGRESS; 1742 } 1743 1744 /* The command header of this PDU has now been read from the socket. */ 1745 spdk_nvmf_tcp_pdu_ch_handle(tqpair); 1746 break; 1747 /* Wait for the pdu specific header */ 1748 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1749 rc = nvme_tcp_read_data(tqpair->sock, 1750 pdu->psh_len - pdu->psh_valid_bytes, 1751 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1752 if (rc < 0) { 1753 return NVME_TCP_PDU_FATAL; 1754 } else if (rc > 0) { 1755 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1756 0, rc, 0, 0); 1757 pdu->psh_valid_bytes += rc; 1758 } 1759 1760 if (pdu->psh_valid_bytes < pdu->psh_len) { 1761 return NVME_TCP_PDU_IN_PROGRESS; 1762 } 1763 1764 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1765 spdk_nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1766 break; 1767 /* Wait for the req slot */ 1768 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1769 spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1770 break; 1771 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1772 /* check whether the data is valid, if not we just return */ 1773 if (!pdu->data_len) { 1774 return NVME_TCP_PDU_IN_PROGRESS; 1775 } 1776 1777 data_len = pdu->data_len; 1778 /* data digest */ 1779 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1780 tqpair->host_ddgst_enable)) { 1781 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1782 pdu->ddgst_enable = true; 1783 } 1784 1785 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1786 if (rc < 0) { 1787 return NVME_TCP_PDU_IN_PROGRESS; 1788 } 1789 pdu->readv_offset += rc; 1790 1791 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1792 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1793 if (rc != 0) { 1794 return NVME_TCP_PDU_FATAL; 1795 } 1796 } 1797 1798 if (pdu->readv_offset < data_len) { 1799 return NVME_TCP_PDU_IN_PROGRESS; 1800 } 1801 1802 /* All of this PDU has now been read from the socket. */ 1803 spdk_nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1804 break; 1805 case NVME_TCP_PDU_RECV_STATE_ERROR: 1806 if (!spdk_sock_is_connected(tqpair->sock)) { 1807 return NVME_TCP_PDU_FATAL; 1808 } 1809 break; 1810 default: 1811 assert(0); 1812 SPDK_ERRLOG("code should not come to here"); 1813 break; 1814 } 1815 } while (tqpair->recv_state != prev_state); 1816 1817 return rc; 1818 } 1819 1820 static int 1821 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 1822 struct spdk_nvmf_transport *transport, 1823 struct spdk_nvmf_transport_poll_group *group) 1824 { 1825 struct spdk_nvmf_request *req = &tcp_req->req; 1826 struct spdk_nvme_cmd *cmd; 1827 struct spdk_nvme_cpl *rsp; 1828 struct spdk_nvme_sgl_descriptor *sgl; 1829 uint32_t length; 1830 1831 cmd = &req->cmd->nvme_cmd; 1832 rsp = &req->rsp->nvme_cpl; 1833 sgl = &cmd->dptr.sgl1; 1834 1835 length = sgl->unkeyed.length; 1836 1837 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 1838 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 1839 if (length > transport->opts.max_io_size) { 1840 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 1841 length, transport->opts.max_io_size); 1842 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1843 return -1; 1844 } 1845 1846 /* fill request length and populate iovs */ 1847 req->length = length; 1848 1849 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", length); 1850 1851 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1852 req->dif.orig_length = length; 1853 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1854 req->dif.elba_length = length; 1855 } 1856 1857 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 1858 /* No available buffers. Queue this request up. */ 1859 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", 1860 tcp_req); 1861 return 0; 1862 } 1863 1864 /* backward compatible */ 1865 req->data = req->iov[0].iov_base; 1866 1867 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 1868 tcp_req, req->iovcnt, req->data); 1869 1870 return 0; 1871 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 1872 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 1873 uint64_t offset = sgl->address; 1874 uint32_t max_len = transport->opts.in_capsule_data_size; 1875 1876 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 1877 offset, length); 1878 1879 if (offset > max_len) { 1880 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 1881 offset, max_len); 1882 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 1883 return -1; 1884 } 1885 max_len -= (uint32_t)offset; 1886 1887 if (length > max_len) { 1888 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 1889 length, max_len); 1890 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1891 return -1; 1892 } 1893 1894 req->data = tcp_req->buf + offset; 1895 req->data_from_pool = false; 1896 req->length = length; 1897 1898 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1899 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1900 req->dif.elba_length = length; 1901 } 1902 1903 req->iov[0].iov_base = req->data; 1904 req->iov[0].iov_len = length; 1905 req->iovcnt = 1; 1906 1907 return 0; 1908 } 1909 1910 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 1911 sgl->generic.type, sgl->generic.subtype); 1912 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 1913 return -1; 1914 } 1915 1916 static inline enum spdk_nvme_media_error_status_code 1917 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 1918 enum spdk_nvme_media_error_status_code result; 1919 1920 switch (err_type) 1921 { 1922 case SPDK_DIF_REFTAG_ERROR: 1923 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 1924 break; 1925 case SPDK_DIF_APPTAG_ERROR: 1926 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 1927 break; 1928 case SPDK_DIF_GUARD_ERROR: 1929 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 1930 break; 1931 default: 1932 SPDK_UNREACHABLE(); 1933 break; 1934 } 1935 1936 return result; 1937 } 1938 1939 static void 1940 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 1941 struct spdk_nvmf_tcp_req *tcp_req) 1942 { 1943 struct nvme_tcp_pdu *rsp_pdu; 1944 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 1945 uint32_t plen, pdo, alignment; 1946 int rc; 1947 1948 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1949 1950 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1951 assert(rsp_pdu != NULL); 1952 1953 c2h_data = &rsp_pdu->hdr.c2h_data; 1954 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 1955 plen = c2h_data->common.hlen = sizeof(*c2h_data); 1956 1957 if (tqpair->host_hdgst_enable) { 1958 plen += SPDK_NVME_TCP_DIGEST_LEN; 1959 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1960 } 1961 1962 /* set the psh */ 1963 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1964 c2h_data->datal = tcp_req->req.length; 1965 c2h_data->datao = 0; 1966 1967 /* set the padding */ 1968 rsp_pdu->padding_len = 0; 1969 pdo = plen; 1970 if (tqpair->cpda) { 1971 alignment = (tqpair->cpda + 1) << 2; 1972 if (alignment > plen) { 1973 rsp_pdu->padding_len = alignment - plen; 1974 pdo = plen = alignment; 1975 } 1976 } 1977 1978 c2h_data->common.pdo = pdo; 1979 plen += c2h_data->datal; 1980 if (tqpair->host_ddgst_enable) { 1981 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 1982 plen += SPDK_NVME_TCP_DIGEST_LEN; 1983 } 1984 1985 c2h_data->common.plen = plen; 1986 1987 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1988 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1989 } 1990 1991 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1992 c2h_data->datao, c2h_data->datal); 1993 1994 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1995 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 1996 struct spdk_dif_error err_blk = {}; 1997 1998 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 1999 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2000 if (rc != 0) { 2001 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2002 err_blk.err_type, err_blk.err_offset); 2003 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2004 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2005 nvmf_tcp_req_pdu_fini(tcp_req); 2006 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2007 return; 2008 } 2009 } 2010 2011 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2012 if (tqpair->qpair.transport->opts.c2h_success) { 2013 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2014 } 2015 2016 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2017 } 2018 2019 static int 2020 request_transfer_out(struct spdk_nvmf_request *req) 2021 { 2022 struct spdk_nvmf_tcp_req *tcp_req; 2023 struct spdk_nvmf_qpair *qpair; 2024 struct spdk_nvmf_tcp_qpair *tqpair; 2025 struct spdk_nvme_cpl *rsp; 2026 2027 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2028 2029 qpair = req->qpair; 2030 rsp = &req->rsp->nvme_cpl; 2031 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2032 2033 /* Advance our sq_head pointer */ 2034 if (qpair->sq_head == qpair->sq_head_max) { 2035 qpair->sq_head = 0; 2036 } else { 2037 qpair->sq_head++; 2038 } 2039 rsp->sqhd = qpair->sq_head; 2040 2041 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2042 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2043 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2044 spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2045 } else { 2046 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2047 } 2048 2049 return 0; 2050 } 2051 2052 static void 2053 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2054 struct spdk_nvmf_tcp_req *tcp_req) 2055 { 2056 struct nvme_tcp_pdu *pdu; 2057 uint32_t plen = 0; 2058 2059 pdu = &tqpair->pdu_in_progress; 2060 plen = pdu->hdr.common.hlen; 2061 2062 if (tqpair->host_hdgst_enable) { 2063 plen += SPDK_NVME_TCP_DIGEST_LEN; 2064 } 2065 2066 if (pdu->hdr.common.plen != plen) { 2067 tcp_req->has_incapsule_data = true; 2068 } 2069 } 2070 2071 static bool 2072 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2073 struct spdk_nvmf_tcp_req *tcp_req) 2074 { 2075 struct spdk_nvmf_tcp_qpair *tqpair; 2076 int rc; 2077 enum spdk_nvmf_tcp_req_state prev_state; 2078 bool progress = false; 2079 struct spdk_nvmf_transport *transport = &ttransport->transport; 2080 struct spdk_nvmf_transport_poll_group *group; 2081 2082 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2083 group = &tqpair->group->group; 2084 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2085 2086 /* The loop here is to allow for several back-to-back state changes. */ 2087 do { 2088 prev_state = tcp_req->state; 2089 2090 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2091 tqpair); 2092 2093 switch (tcp_req->state) { 2094 case TCP_REQUEST_STATE_FREE: 2095 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2096 * to escape this state. */ 2097 break; 2098 case TCP_REQUEST_STATE_NEW: 2099 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2100 2101 /* copy the cmd from the receive pdu */ 2102 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2103 2104 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2105 tcp_req->req.dif.dif_insert_or_strip = true; 2106 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2107 } 2108 2109 /* The next state transition depends on the data transfer needs of this request. */ 2110 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2111 2112 /* If no data to transfer, ready to execute. */ 2113 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2114 /* Reset the tqpair receving pdu state */ 2115 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2116 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2117 break; 2118 } 2119 2120 spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2121 2122 if (!tcp_req->has_incapsule_data) { 2123 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2124 } 2125 2126 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2127 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2128 break; 2129 case TCP_REQUEST_STATE_NEED_BUFFER: 2130 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2131 2132 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2133 2134 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2135 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2136 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2137 tcp_req, tqpair); 2138 /* This request needs to wait in line to obtain a buffer */ 2139 break; 2140 } 2141 2142 /* Try to get a data buffer */ 2143 rc = spdk_nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2144 if (rc < 0) { 2145 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2146 /* Reset the tqpair receving pdu state */ 2147 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2148 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2149 break; 2150 } 2151 2152 if (!tcp_req->req.data) { 2153 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2154 tcp_req, tqpair); 2155 /* No buffers available. */ 2156 break; 2157 } 2158 2159 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2160 2161 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2162 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2163 if (tcp_req->req.data_from_pool) { 2164 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2165 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2166 } else { 2167 struct nvme_tcp_pdu *pdu; 2168 2169 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2170 2171 pdu = &tqpair->pdu_in_progress; 2172 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2173 tqpair); 2174 /* No need to send r2t, contained in the capsuled data */ 2175 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2176 0, tcp_req->req.length); 2177 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2178 } 2179 break; 2180 } 2181 2182 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2183 break; 2184 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2185 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2186 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2187 break; 2188 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2189 2190 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2191 (uintptr_t)tcp_req, 0); 2192 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2193 * to escape this state. */ 2194 break; 2195 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2196 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2197 2198 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2199 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2200 tcp_req->req.length = tcp_req->req.dif.elba_length; 2201 } 2202 2203 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2204 spdk_nvmf_request_exec(&tcp_req->req); 2205 break; 2206 case TCP_REQUEST_STATE_EXECUTING: 2207 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2208 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2209 * to escape this state. */ 2210 break; 2211 case TCP_REQUEST_STATE_EXECUTED: 2212 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2213 2214 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2215 tcp_req->req.length = tcp_req->req.dif.orig_length; 2216 } 2217 2218 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2219 break; 2220 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2221 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2222 rc = request_transfer_out(&tcp_req->req); 2223 assert(rc == 0); /* No good way to handle this currently */ 2224 break; 2225 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2226 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2227 (uintptr_t)tcp_req, 2228 0); 2229 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2230 * to escape this state. */ 2231 break; 2232 case TCP_REQUEST_STATE_COMPLETED: 2233 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2234 if (tcp_req->req.data_from_pool) { 2235 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2236 } 2237 tcp_req->req.length = 0; 2238 tcp_req->req.iovcnt = 0; 2239 tcp_req->req.data = NULL; 2240 2241 nvmf_tcp_req_pdu_fini(tcp_req); 2242 2243 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2244 break; 2245 case TCP_REQUEST_NUM_STATES: 2246 default: 2247 assert(0); 2248 break; 2249 } 2250 2251 if (tcp_req->state != prev_state) { 2252 progress = true; 2253 } 2254 } while (tcp_req->state != prev_state); 2255 2256 return progress; 2257 } 2258 2259 static void 2260 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2261 { 2262 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2263 int rc; 2264 2265 assert(tqpair != NULL); 2266 rc = spdk_nvmf_tcp_sock_process(tqpair); 2267 2268 /* If there was a new socket error, disconnect */ 2269 if (rc < 0) { 2270 spdk_nvmf_tcp_qpair_disconnect(tqpair); 2271 } 2272 } 2273 2274 static int 2275 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2276 struct spdk_nvmf_qpair *qpair) 2277 { 2278 struct spdk_nvmf_tcp_poll_group *tgroup; 2279 struct spdk_nvmf_tcp_qpair *tqpair; 2280 int rc; 2281 2282 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2283 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2284 2285 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2286 spdk_nvmf_tcp_sock_cb, tqpair); 2287 if (rc != 0) { 2288 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2289 spdk_strerror(errno), errno); 2290 return -1; 2291 } 2292 2293 rc = spdk_nvmf_tcp_qpair_sock_init(tqpair); 2294 if (rc != 0) { 2295 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2296 return -1; 2297 } 2298 2299 rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair); 2300 if (rc < 0) { 2301 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2302 return -1; 2303 } 2304 2305 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair); 2306 if (rc < 0) { 2307 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2308 return -1; 2309 } 2310 2311 tqpair->group = tgroup; 2312 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2313 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2314 2315 return 0; 2316 } 2317 2318 static int 2319 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2320 struct spdk_nvmf_qpair *qpair) 2321 { 2322 struct spdk_nvmf_tcp_poll_group *tgroup; 2323 struct spdk_nvmf_tcp_qpair *tqpair; 2324 int rc; 2325 2326 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2327 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2328 2329 assert(tqpair->group == tgroup); 2330 2331 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2332 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2333 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2334 } else { 2335 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2336 } 2337 2338 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2339 if (rc != 0) { 2340 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2341 spdk_strerror(errno), errno); 2342 } 2343 2344 return rc; 2345 } 2346 2347 static int 2348 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2349 { 2350 struct spdk_nvmf_tcp_transport *ttransport; 2351 struct spdk_nvmf_tcp_req *tcp_req; 2352 2353 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2354 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2355 2356 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2357 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 2358 2359 return 0; 2360 } 2361 2362 static void 2363 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2364 { 2365 struct spdk_nvmf_tcp_qpair *tqpair; 2366 2367 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Qpair: %p\n", qpair); 2368 2369 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2370 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2371 spdk_nvmf_tcp_qpair_destroy(tqpair); 2372 } 2373 2374 static int 2375 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2376 { 2377 struct spdk_nvmf_tcp_poll_group *tgroup; 2378 int rc; 2379 struct spdk_nvmf_request *req, *req_tmp; 2380 struct spdk_nvmf_tcp_req *tcp_req; 2381 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2382 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2383 struct spdk_nvmf_tcp_transport, transport); 2384 2385 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2386 2387 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2388 return 0; 2389 } 2390 2391 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2392 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2393 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2394 break; 2395 } 2396 } 2397 2398 rc = spdk_sock_group_poll(tgroup->sock_group); 2399 if (rc < 0) { 2400 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2401 } 2402 2403 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2404 spdk_nvmf_tcp_sock_process(tqpair); 2405 } 2406 2407 return rc; 2408 } 2409 2410 static int 2411 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2412 struct spdk_nvme_transport_id *trid, bool peer) 2413 { 2414 struct spdk_nvmf_tcp_qpair *tqpair; 2415 uint16_t port; 2416 2417 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2418 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2419 2420 if (peer) { 2421 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2422 port = tqpair->initiator_port; 2423 } else { 2424 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2425 port = tqpair->target_port; 2426 } 2427 2428 if (spdk_sock_is_ipv4(tqpair->sock)) { 2429 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2430 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2431 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2432 } else { 2433 return -1; 2434 } 2435 2436 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2437 return 0; 2438 } 2439 2440 static int 2441 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2442 struct spdk_nvme_transport_id *trid) 2443 { 2444 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2445 } 2446 2447 static int 2448 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2449 struct spdk_nvme_transport_id *trid) 2450 { 2451 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2452 } 2453 2454 static int 2455 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2456 struct spdk_nvme_transport_id *trid) 2457 { 2458 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2459 } 2460 2461 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2462 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2463 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2464 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2465 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2466 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2467 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2468 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2469 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 2470 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2471 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 2472 2473 static void 2474 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2475 { 2476 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2477 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2478 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2479 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2480 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2481 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2482 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2483 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2484 opts->c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 2485 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2486 opts->sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 2487 } 2488 2489 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2490 .name = "TCP", 2491 .type = SPDK_NVME_TRANSPORT_TCP, 2492 .opts_init = spdk_nvmf_tcp_opts_init, 2493 .create = spdk_nvmf_tcp_create, 2494 .destroy = spdk_nvmf_tcp_destroy, 2495 2496 .listen = spdk_nvmf_tcp_listen, 2497 .stop_listen = spdk_nvmf_tcp_stop_listen, 2498 .accept = spdk_nvmf_tcp_accept, 2499 2500 .listener_discover = spdk_nvmf_tcp_discover, 2501 2502 .poll_group_create = spdk_nvmf_tcp_poll_group_create, 2503 .get_optimal_poll_group = spdk_nvmf_tcp_get_optimal_poll_group, 2504 .poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy, 2505 .poll_group_add = spdk_nvmf_tcp_poll_group_add, 2506 .poll_group_remove = spdk_nvmf_tcp_poll_group_remove, 2507 .poll_group_poll = spdk_nvmf_tcp_poll_group_poll, 2508 2509 .req_free = spdk_nvmf_tcp_req_free, 2510 .req_complete = spdk_nvmf_tcp_req_complete, 2511 2512 .qpair_fini = spdk_nvmf_tcp_close_qpair, 2513 .qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid, 2514 .qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid, 2515 .qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid, 2516 }; 2517 2518 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2519 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2520