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/log.h" 47 #include "spdk_internal/nvme_tcp.h" 48 49 #include "nvmf_internal.h" 50 51 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 52 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 53 54 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 55 56 /* spdk nvmf related structure */ 57 enum spdk_nvmf_tcp_req_state { 58 59 /* The request is not currently in use */ 60 TCP_REQUEST_STATE_FREE = 0, 61 62 /* Initial state when request first received */ 63 TCP_REQUEST_STATE_NEW, 64 65 /* The request is queued until a data buffer is available. */ 66 TCP_REQUEST_STATE_NEED_BUFFER, 67 68 /* The request is currently transferring data from the host to the controller. */ 69 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 70 71 /* The request is waiting for the R2T send acknowledgement. */ 72 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 73 74 /* The request is ready to execute at the block device */ 75 TCP_REQUEST_STATE_READY_TO_EXECUTE, 76 77 /* The request is currently executing at the block device */ 78 TCP_REQUEST_STATE_EXECUTING, 79 80 /* The request finished executing at the block device */ 81 TCP_REQUEST_STATE_EXECUTED, 82 83 /* The request is ready to send a completion */ 84 TCP_REQUEST_STATE_READY_TO_COMPLETE, 85 86 /* The request is currently transferring final pdus from the controller to the host. */ 87 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 88 89 /* The request completed and can be marked free. */ 90 TCP_REQUEST_STATE_COMPLETED, 91 92 /* Terminator */ 93 TCP_REQUEST_NUM_STATES, 94 }; 95 96 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 97 "Invalid PDU Header Field", 98 "PDU Sequence Error", 99 "Header Digiest Error", 100 "Data Transfer Out of Range", 101 "R2T Limit Exceeded", 102 "Unsupported parameter", 103 }; 104 105 #define OBJECT_NVMF_TCP_IO 0x80 106 107 #define TRACE_GROUP_NVMF_TCP 0x5 108 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 109 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 110 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 111 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 112 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 113 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 114 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 115 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 116 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 117 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 118 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 119 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 120 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 121 122 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 123 { 124 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 125 spdk_trace_register_description("TCP_REQ_NEW", 126 TRACE_TCP_REQUEST_STATE_NEW, 127 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 128 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 129 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 130 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 131 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 132 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 133 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 134 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 135 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 136 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 137 spdk_trace_register_description("TCP_REQ_EXECUTING", 138 TRACE_TCP_REQUEST_STATE_EXECUTING, 139 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 140 spdk_trace_register_description("TCP_REQ_EXECUTED", 141 TRACE_TCP_REQUEST_STATE_EXECUTED, 142 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 143 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 144 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 145 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 146 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 147 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 148 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 149 spdk_trace_register_description("TCP_REQ_COMPLETED", 150 TRACE_TCP_REQUEST_STATE_COMPLETED, 151 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 152 spdk_trace_register_description("TCP_WRITE_START", 153 TRACE_TCP_FLUSH_WRITEBUF_START, 154 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 155 spdk_trace_register_description("TCP_WRITE_DONE", 156 TRACE_TCP_FLUSH_WRITEBUF_DONE, 157 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 158 spdk_trace_register_description("TCP_READ_DONE", 159 TRACE_TCP_READ_FROM_SOCKET_DONE, 160 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 161 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 162 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 163 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 164 } 165 166 struct spdk_nvmf_tcp_req { 167 struct spdk_nvmf_request req; 168 struct spdk_nvme_cpl rsp; 169 struct spdk_nvme_cmd cmd; 170 171 /* A PDU that can be used for sending responses. This is 172 * not the incoming PDU! */ 173 struct nvme_tcp_pdu *pdu; 174 175 /* 176 * The PDU for a request may be used multiple times in serial over 177 * the request's lifetime. For example, first to send an R2T, then 178 * to send a completion. To catch mistakes where the PDU is used 179 * twice at the same time, add a debug flag here for init/fini. 180 */ 181 bool pdu_in_use; 182 183 /* In-capsule data buffer */ 184 uint8_t *buf; 185 186 bool has_incapsule_data; 187 188 /* transfer_tag */ 189 uint16_t ttag; 190 191 enum spdk_nvmf_tcp_req_state state; 192 193 /* 194 * h2c_offset is used when we receive the h2c_data PDU. 195 */ 196 uint32_t h2c_offset; 197 198 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 199 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 200 }; 201 202 struct spdk_nvmf_tcp_qpair { 203 struct spdk_nvmf_qpair qpair; 204 struct spdk_nvmf_tcp_poll_group *group; 205 struct spdk_nvmf_tcp_port *port; 206 struct spdk_sock *sock; 207 208 enum nvme_tcp_pdu_recv_state recv_state; 209 enum nvme_tcp_qpair_state state; 210 211 /* PDU being actively received */ 212 struct nvme_tcp_pdu pdu_in_progress; 213 uint32_t recv_buf_size; 214 215 /* This is a spare PDU used for sending special management 216 * operations. Primarily, this is used for the initial 217 * connection response and c2h termination request. */ 218 struct nvme_tcp_pdu *mgmt_pdu; 219 220 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 221 222 /* Arrays of in-capsule buffers, requests, and pdus. 223 * Each array is 'resource_count' number of elements */ 224 void *bufs; 225 struct spdk_nvmf_tcp_req *reqs; 226 struct nvme_tcp_pdu *pdus; 227 uint32_t resource_count; 228 229 /* Queues to track the requests in all states */ 230 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 231 /* Number of requests in each state */ 232 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 233 234 uint8_t cpda; 235 236 bool host_hdgst_enable; 237 bool host_ddgst_enable; 238 239 /* IP address */ 240 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 241 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 242 243 /* IP port */ 244 uint16_t initiator_port; 245 uint16_t target_port; 246 247 /* Timer used to destroy qpair after detecting transport error issue if initiator does 248 * not close the connection. 249 */ 250 struct spdk_poller *timeout_poller; 251 252 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 253 }; 254 255 struct spdk_nvmf_tcp_poll_group { 256 struct spdk_nvmf_transport_poll_group group; 257 struct spdk_sock_group *sock_group; 258 259 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 260 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 261 }; 262 263 struct spdk_nvmf_tcp_port { 264 const struct spdk_nvme_transport_id *trid; 265 struct spdk_sock *listen_sock; 266 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 267 }; 268 269 struct spdk_nvmf_tcp_transport { 270 struct spdk_nvmf_transport transport; 271 272 pthread_mutex_t lock; 273 274 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 275 }; 276 277 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 278 struct spdk_nvmf_tcp_req *tcp_req); 279 280 static void 281 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 282 enum spdk_nvmf_tcp_req_state state) 283 { 284 struct spdk_nvmf_qpair *qpair; 285 struct spdk_nvmf_tcp_qpair *tqpair; 286 287 qpair = tcp_req->req.qpair; 288 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 289 290 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 291 assert(tqpair->state_cntr[tcp_req->state] > 0); 292 tqpair->state_cntr[tcp_req->state]--; 293 294 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 295 tqpair->state_cntr[state]++; 296 297 tcp_req->state = state; 298 } 299 300 static inline struct nvme_tcp_pdu * 301 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 302 { 303 assert(tcp_req->pdu_in_use == false); 304 tcp_req->pdu_in_use = true; 305 306 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 307 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 308 309 return tcp_req->pdu; 310 } 311 312 static inline void 313 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 314 { 315 tcp_req->pdu_in_use = false; 316 } 317 318 static struct spdk_nvmf_tcp_req * 319 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 320 { 321 struct spdk_nvmf_tcp_req *tcp_req; 322 323 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 324 if (!tcp_req) { 325 return NULL; 326 } 327 328 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 329 tcp_req->h2c_offset = 0; 330 tcp_req->has_incapsule_data = false; 331 tcp_req->req.dif.dif_insert_or_strip = false; 332 333 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 334 return tcp_req; 335 } 336 337 static void 338 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 339 { 340 struct spdk_nvmf_tcp_transport *ttransport; 341 342 assert(tcp_req != NULL); 343 344 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 345 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 346 struct spdk_nvmf_tcp_transport, transport); 347 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 348 nvmf_tcp_req_process(ttransport, tcp_req); 349 } 350 351 static int 352 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 353 { 354 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 355 356 nvmf_tcp_request_free(tcp_req); 357 358 return 0; 359 } 360 361 static void 362 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 363 enum spdk_nvmf_tcp_req_state state) 364 { 365 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 366 367 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 368 nvmf_tcp_request_free(tcp_req); 369 } 370 } 371 372 static void 373 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 374 { 375 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 376 377 assert(TAILQ_EMPTY(&tqpair->send_queue)); 378 379 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 380 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 381 382 /* Wipe the requests waiting for buffer from the global list */ 383 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 384 req_tmp) { 385 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 386 spdk_nvmf_request, buf_link); 387 } 388 389 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 390 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 391 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 392 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 393 } 394 395 static void 396 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 397 { 398 int i; 399 struct spdk_nvmf_tcp_req *tcp_req; 400 401 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 402 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 403 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 404 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 405 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 406 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 407 } 408 } 409 } 410 411 static void 412 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 413 { 414 int err = 0; 415 416 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 417 418 err = spdk_sock_close(&tqpair->sock); 419 assert(err == 0); 420 nvmf_tcp_cleanup_all_states(tqpair); 421 422 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 423 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 424 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 425 tqpair->resource_count); 426 err++; 427 } 428 429 if (err > 0) { 430 nvmf_tcp_dump_qpair_req_contents(tqpair); 431 } 432 433 spdk_dma_free(tqpair->pdus); 434 free(tqpair->reqs); 435 spdk_free(tqpair->bufs); 436 free(tqpair); 437 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 438 } 439 440 static int 441 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 442 { 443 struct spdk_nvmf_tcp_transport *ttransport; 444 445 assert(transport != NULL); 446 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 447 448 pthread_mutex_destroy(&ttransport->lock); 449 free(ttransport); 450 return 0; 451 } 452 453 static struct spdk_nvmf_transport * 454 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 455 { 456 struct spdk_nvmf_tcp_transport *ttransport; 457 uint32_t sge_count; 458 uint32_t min_shared_buffers; 459 460 ttransport = calloc(1, sizeof(*ttransport)); 461 if (!ttransport) { 462 return NULL; 463 } 464 465 TAILQ_INIT(&ttransport->ports); 466 467 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 468 469 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 470 471 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 472 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 473 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 474 " in_capsule_data_size=%d, max_aq_depth=%d\n" 475 " num_shared_buffers=%d, c2h_success=%d,\n" 476 " dif_insert_or_strip=%d, sock_priority=%d\n" 477 " abort_timeout_sec=%d\n", 478 opts->max_queue_depth, 479 opts->max_io_size, 480 opts->max_qpairs_per_ctrlr - 1, 481 opts->io_unit_size, 482 opts->in_capsule_data_size, 483 opts->max_aq_depth, 484 opts->num_shared_buffers, 485 opts->c2h_success, 486 opts->dif_insert_or_strip, 487 opts->sock_priority, 488 opts->abort_timeout_sec); 489 490 if (opts->sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 491 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 492 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 493 opts->sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 494 free(ttransport); 495 return NULL; 496 } 497 498 /* I/O unit size cannot be larger than max I/O size */ 499 if (opts->io_unit_size > opts->max_io_size) { 500 opts->io_unit_size = opts->max_io_size; 501 } 502 503 sge_count = opts->max_io_size / opts->io_unit_size; 504 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 505 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 506 free(ttransport); 507 return NULL; 508 } 509 510 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 511 if (min_shared_buffers > opts->num_shared_buffers) { 512 SPDK_ERRLOG("There are not enough buffers to satisfy" 513 "per-poll group caches for each thread. (%" PRIu32 ")" 514 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 515 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 516 nvmf_tcp_destroy(&ttransport->transport); 517 return NULL; 518 } 519 520 pthread_mutex_init(&ttransport->lock, NULL); 521 522 return &ttransport->transport; 523 } 524 525 static int 526 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 527 { 528 unsigned long long ull; 529 char *end = NULL; 530 531 ull = strtoull(trsvcid, &end, 10); 532 if (end == NULL || end == trsvcid || *end != '\0') { 533 return -1; 534 } 535 536 /* Valid TCP/IP port numbers are in [0, 65535] */ 537 if (ull > 65535) { 538 return -1; 539 } 540 541 return (int)ull; 542 } 543 544 /** 545 * Canonicalize a listen address trid. 546 */ 547 static int 548 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 549 const struct spdk_nvme_transport_id *trid) 550 { 551 int trsvcid_int; 552 553 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 554 if (trsvcid_int < 0) { 555 return -EINVAL; 556 } 557 558 memset(canon_trid, 0, sizeof(*canon_trid)); 559 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 560 canon_trid->adrfam = trid->adrfam; 561 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 562 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 563 564 return 0; 565 } 566 567 /** 568 * Find an existing listening port. 569 * 570 * Caller must hold ttransport->lock. 571 */ 572 static struct spdk_nvmf_tcp_port * 573 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 574 const struct spdk_nvme_transport_id *trid) 575 { 576 struct spdk_nvme_transport_id canon_trid; 577 struct spdk_nvmf_tcp_port *port; 578 579 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 580 return NULL; 581 } 582 583 TAILQ_FOREACH(port, &ttransport->ports, link) { 584 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 585 return port; 586 } 587 } 588 589 return NULL; 590 } 591 592 static int 593 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 594 const struct spdk_nvme_transport_id *trid) 595 { 596 struct spdk_nvmf_tcp_transport *ttransport; 597 struct spdk_nvmf_tcp_port *port; 598 int trsvcid_int; 599 uint8_t adrfam; 600 struct spdk_sock_opts opts; 601 602 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 603 604 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 605 if (trsvcid_int < 0) { 606 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 607 return -EINVAL; 608 } 609 610 pthread_mutex_lock(&ttransport->lock); 611 port = calloc(1, sizeof(*port)); 612 if (!port) { 613 SPDK_ERRLOG("Port allocation failed\n"); 614 pthread_mutex_unlock(&ttransport->lock); 615 return -ENOMEM; 616 } 617 618 port->trid = trid; 619 opts.opts_size = sizeof(opts); 620 spdk_sock_get_default_opts(&opts); 621 opts.priority = transport->opts.sock_priority; 622 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 623 NULL, &opts); 624 if (port->listen_sock == NULL) { 625 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 626 trid->traddr, trsvcid_int, 627 spdk_strerror(errno), errno); 628 free(port); 629 pthread_mutex_unlock(&ttransport->lock); 630 return -errno; 631 } 632 633 if (spdk_sock_is_ipv4(port->listen_sock)) { 634 adrfam = SPDK_NVMF_ADRFAM_IPV4; 635 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 636 adrfam = SPDK_NVMF_ADRFAM_IPV6; 637 } else { 638 SPDK_ERRLOG("Unhandled socket type\n"); 639 adrfam = 0; 640 } 641 642 if (adrfam != trid->adrfam) { 643 SPDK_ERRLOG("Socket address family mismatch\n"); 644 spdk_sock_close(&port->listen_sock); 645 free(port); 646 pthread_mutex_unlock(&ttransport->lock); 647 return -EINVAL; 648 } 649 650 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 651 trid->traddr, trid->trsvcid); 652 653 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 654 pthread_mutex_unlock(&ttransport->lock); 655 return 0; 656 } 657 658 static void 659 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 660 const struct spdk_nvme_transport_id *trid) 661 { 662 struct spdk_nvmf_tcp_transport *ttransport; 663 struct spdk_nvmf_tcp_port *port; 664 665 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 666 667 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 668 trid->traddr, trid->trsvcid); 669 670 pthread_mutex_lock(&ttransport->lock); 671 port = nvmf_tcp_find_port(ttransport, trid); 672 if (port) { 673 TAILQ_REMOVE(&ttransport->ports, port, link); 674 spdk_sock_close(&port->listen_sock); 675 free(port); 676 } 677 678 pthread_mutex_unlock(&ttransport->lock); 679 } 680 681 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 682 enum nvme_tcp_pdu_recv_state state); 683 684 static void 685 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 686 { 687 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 688 689 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 690 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 691 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 692 spdk_poller_unregister(&tqpair->timeout_poller); 693 694 /* This will end up calling nvmf_tcp_close_qpair */ 695 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 696 } 697 } 698 699 static void 700 _pdu_write_done(void *_pdu, int err) 701 { 702 struct nvme_tcp_pdu *pdu = _pdu; 703 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 704 705 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 706 707 if (err != 0) { 708 nvmf_tcp_qpair_disconnect(tqpair); 709 return; 710 } 711 712 assert(pdu->cb_fn != NULL); 713 pdu->cb_fn(pdu->cb_arg); 714 } 715 716 static void 717 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 718 struct nvme_tcp_pdu *pdu, 719 nvme_tcp_qpair_xfer_complete_cb cb_fn, 720 void *cb_arg) 721 { 722 int hlen; 723 uint32_t crc32c; 724 uint32_t mapped_length = 0; 725 ssize_t rc; 726 727 assert(&tqpair->pdu_in_progress != pdu); 728 729 hlen = pdu->hdr.common.hlen; 730 731 /* Header Digest */ 732 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 733 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 734 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 735 } 736 737 /* Data Digest */ 738 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 739 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 740 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 741 } 742 743 pdu->cb_fn = cb_fn; 744 pdu->cb_arg = cb_arg; 745 746 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 747 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 748 &mapped_length); 749 pdu->sock_req.cb_fn = _pdu_write_done; 750 pdu->sock_req.cb_arg = pdu; 751 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 752 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 753 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 754 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 755 if (rc == mapped_length) { 756 _pdu_write_done(pdu, 0); 757 } else { 758 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 759 _pdu_write_done(pdu, -1); 760 } 761 } else { 762 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 763 } 764 } 765 766 static int 767 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 768 { 769 uint32_t i; 770 struct spdk_nvmf_transport_opts *opts; 771 uint32_t in_capsule_data_size; 772 773 opts = &tqpair->qpair.transport->opts; 774 775 in_capsule_data_size = opts->in_capsule_data_size; 776 if (opts->dif_insert_or_strip) { 777 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 778 } 779 780 tqpair->resource_count = opts->max_queue_depth; 781 782 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 783 if (!tqpair->reqs) { 784 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 785 return -1; 786 } 787 788 if (in_capsule_data_size) { 789 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 790 NULL, SPDK_ENV_LCORE_ID_ANY, 791 SPDK_MALLOC_DMA); 792 if (!tqpair->bufs) { 793 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 794 return -1; 795 } 796 } 797 798 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 799 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 800 if (!tqpair->pdus) { 801 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 802 return -1; 803 } 804 805 for (i = 0; i < tqpair->resource_count; i++) { 806 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 807 808 tcp_req->ttag = i + 1; 809 tcp_req->req.qpair = &tqpair->qpair; 810 811 tcp_req->pdu = &tqpair->pdus[i]; 812 tcp_req->pdu->qpair = tqpair; 813 814 /* Set up memory to receive commands */ 815 if (tqpair->bufs) { 816 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 817 } 818 819 /* Set the cmdn and rsp */ 820 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 821 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 822 823 /* Initialize request state to FREE */ 824 tcp_req->state = TCP_REQUEST_STATE_FREE; 825 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 826 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 827 } 828 829 tqpair->mgmt_pdu = &tqpair->pdus[i]; 830 tqpair->mgmt_pdu->qpair = tqpair; 831 832 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 833 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 834 835 return 0; 836 } 837 838 static int 839 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 840 { 841 struct spdk_nvmf_tcp_qpair *tqpair; 842 int i; 843 844 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 845 846 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 847 848 TAILQ_INIT(&tqpair->send_queue); 849 850 /* Initialise request state queues of the qpair */ 851 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 852 TAILQ_INIT(&tqpair->state_queue[i]); 853 } 854 855 tqpair->host_hdgst_enable = true; 856 tqpair->host_ddgst_enable = true; 857 858 return 0; 859 } 860 861 static int 862 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 863 { 864 int rc; 865 866 /* set low water mark */ 867 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 868 if (rc != 0) { 869 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 870 return rc; 871 } 872 873 return 0; 874 } 875 876 static void 877 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 878 struct spdk_nvmf_tcp_port *port, 879 struct spdk_sock *sock) 880 { 881 struct spdk_nvmf_tcp_qpair *tqpair; 882 int rc; 883 884 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 885 port->trid->traddr, port->trid->trsvcid); 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 tqpair->qpair.trid = port->trid; 899 900 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 901 sizeof(tqpair->target_addr), &tqpair->target_port, 902 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 903 &tqpair->initiator_port); 904 if (rc < 0) { 905 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 906 nvmf_tcp_qpair_destroy(tqpair); 907 return; 908 } 909 910 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 911 } 912 913 static uint32_t 914 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 915 { 916 struct spdk_sock *sock; 917 uint32_t count = 0; 918 int i; 919 920 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 921 sock = spdk_sock_accept(port->listen_sock); 922 if (sock == NULL) { 923 break; 924 } 925 count++; 926 nvmf_tcp_handle_connect(transport, port, sock); 927 } 928 929 return count; 930 } 931 932 static uint32_t 933 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 934 { 935 struct spdk_nvmf_tcp_transport *ttransport; 936 struct spdk_nvmf_tcp_port *port; 937 uint32_t count = 0; 938 939 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 940 941 TAILQ_FOREACH(port, &ttransport->ports, link) { 942 count += nvmf_tcp_port_accept(transport, port); 943 } 944 945 return count; 946 } 947 948 static void 949 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 950 struct spdk_nvme_transport_id *trid, 951 struct spdk_nvmf_discovery_log_page_entry *entry) 952 { 953 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 954 entry->adrfam = trid->adrfam; 955 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 956 957 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 958 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 959 960 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 961 } 962 963 static struct spdk_nvmf_transport_poll_group * 964 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 965 { 966 struct spdk_nvmf_tcp_poll_group *tgroup; 967 968 tgroup = calloc(1, sizeof(*tgroup)); 969 if (!tgroup) { 970 return NULL; 971 } 972 973 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 974 if (!tgroup->sock_group) { 975 goto cleanup; 976 } 977 978 TAILQ_INIT(&tgroup->qpairs); 979 TAILQ_INIT(&tgroup->await_req); 980 981 return &tgroup->group; 982 983 cleanup: 984 free(tgroup); 985 return NULL; 986 } 987 988 static struct spdk_nvmf_transport_poll_group * 989 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 990 { 991 struct spdk_nvmf_tcp_qpair *tqpair; 992 struct spdk_sock_group *group = NULL; 993 int rc; 994 995 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 996 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 997 if (!rc && group != NULL) { 998 return spdk_sock_group_get_ctx(group); 999 } 1000 1001 return NULL; 1002 } 1003 1004 static void 1005 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1006 { 1007 struct spdk_nvmf_tcp_poll_group *tgroup; 1008 1009 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1010 spdk_sock_group_close(&tgroup->sock_group); 1011 1012 free(tgroup); 1013 } 1014 1015 static void 1016 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1017 enum nvme_tcp_pdu_recv_state state) 1018 { 1019 if (tqpair->recv_state == state) { 1020 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1021 tqpair, state); 1022 return; 1023 } 1024 1025 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1026 /* When leaving the await req state, move the qpair to the main list */ 1027 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1028 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1029 } 1030 1031 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1032 tqpair->recv_state = state; 1033 1034 switch (state) { 1035 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1036 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1037 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1038 break; 1039 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1040 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1041 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1042 break; 1043 case NVME_TCP_PDU_RECV_STATE_ERROR: 1044 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1045 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1046 break; 1047 default: 1048 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1049 abort(); 1050 break; 1051 } 1052 } 1053 1054 static int 1055 nvmf_tcp_qpair_handle_timeout(void *ctx) 1056 { 1057 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1058 1059 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1060 1061 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1062 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1063 1064 nvmf_tcp_qpair_disconnect(tqpair); 1065 return SPDK_POLLER_BUSY; 1066 } 1067 1068 static void 1069 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1070 { 1071 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1072 1073 if (!tqpair->timeout_poller) { 1074 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1075 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1076 } 1077 } 1078 1079 static void 1080 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1081 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1082 { 1083 struct nvme_tcp_pdu *rsp_pdu; 1084 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1085 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1086 uint32_t copy_len; 1087 1088 rsp_pdu = tqpair->mgmt_pdu; 1089 1090 c2h_term_req = &rsp_pdu->hdr.term_req; 1091 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1092 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1093 1094 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1095 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1096 DSET32(&c2h_term_req->fei, error_offset); 1097 } 1098 1099 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1100 1101 /* Copy the error info into the buffer */ 1102 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1103 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1104 1105 /* Contain the header of the wrong received pdu */ 1106 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1107 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1108 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1109 } 1110 1111 static void 1112 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1113 struct spdk_nvmf_tcp_qpair *tqpair, 1114 struct nvme_tcp_pdu *pdu) 1115 { 1116 struct spdk_nvmf_tcp_req *tcp_req; 1117 1118 assert(pdu->psh_valid_bytes == pdu->psh_len); 1119 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1120 1121 tcp_req = nvmf_tcp_req_get(tqpair); 1122 if (!tcp_req) { 1123 /* Directly return and make the allocation retry again */ 1124 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1125 return; 1126 } 1127 1128 /* The host sent more commands than the maximum queue depth. */ 1129 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1130 nvmf_tcp_qpair_disconnect(tqpair); 1131 return; 1132 } 1133 1134 pdu->req = tcp_req; 1135 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1136 nvmf_tcp_req_process(ttransport, tcp_req); 1137 } 1138 1139 static void 1140 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1141 struct spdk_nvmf_tcp_qpair *tqpair, 1142 struct nvme_tcp_pdu *pdu) 1143 { 1144 struct spdk_nvmf_tcp_req *tcp_req; 1145 struct spdk_nvme_tcp_cmd *capsule_cmd; 1146 uint32_t error_offset = 0; 1147 enum spdk_nvme_tcp_term_req_fes fes; 1148 1149 capsule_cmd = &pdu->hdr.capsule_cmd; 1150 tcp_req = pdu->req; 1151 assert(tcp_req != NULL); 1152 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1153 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1154 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1155 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1156 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1157 goto err; 1158 } 1159 1160 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1161 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1162 nvmf_tcp_req_process(ttransport, tcp_req); 1163 1164 return; 1165 err: 1166 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1167 } 1168 1169 static int 1170 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1171 enum spdk_nvmf_tcp_req_state state, 1172 uint16_t cid, uint16_t tag, 1173 struct spdk_nvmf_tcp_req **req) 1174 { 1175 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1176 1177 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) { 1178 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1179 continue; 1180 } 1181 1182 if (tcp_req->ttag == tag) { 1183 *req = tcp_req; 1184 return 0; 1185 } 1186 1187 *req = NULL; 1188 return -1; 1189 } 1190 1191 /* Didn't find it, but not an error */ 1192 *req = NULL; 1193 return 0; 1194 } 1195 1196 static void 1197 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1198 struct spdk_nvmf_tcp_qpair *tqpair, 1199 struct nvme_tcp_pdu *pdu) 1200 { 1201 struct spdk_nvmf_tcp_req *tcp_req; 1202 uint32_t error_offset = 0; 1203 enum spdk_nvme_tcp_term_req_fes fes = 0; 1204 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1205 int rc; 1206 1207 h2c_data = &pdu->hdr.h2c_data; 1208 1209 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1210 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1211 1212 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1213 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1214 if (rc == 0 && tcp_req == NULL) { 1215 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1216 h2c_data->ttag, &tcp_req); 1217 } 1218 1219 if (!tcp_req) { 1220 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1221 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1222 if (rc == 0) { 1223 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1224 } else { 1225 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1226 } 1227 goto err; 1228 } 1229 1230 if (tcp_req->h2c_offset != h2c_data->datao) { 1231 SPDK_DEBUGLOG(nvmf_tcp, 1232 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1233 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1234 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1235 goto err; 1236 } 1237 1238 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1239 SPDK_DEBUGLOG(nvmf_tcp, 1240 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1241 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1242 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1243 goto err; 1244 } 1245 1246 pdu->req = tcp_req; 1247 1248 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1249 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1250 } 1251 1252 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1253 h2c_data->datao, h2c_data->datal); 1254 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1255 return; 1256 1257 err: 1258 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1259 } 1260 1261 static void 1262 nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1263 { 1264 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1265 nvmf_tcp_request_free(tcp_req); 1266 } 1267 1268 static void 1269 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1270 struct spdk_nvmf_tcp_qpair *tqpair) 1271 { 1272 struct nvme_tcp_pdu *rsp_pdu; 1273 struct spdk_nvme_tcp_rsp *capsule_resp; 1274 1275 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1276 1277 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1278 assert(rsp_pdu != NULL); 1279 1280 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1281 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1282 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1283 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1284 if (tqpair->host_hdgst_enable) { 1285 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1286 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1287 } 1288 1289 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_cmd_complete, tcp_req); 1290 } 1291 1292 static void 1293 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1294 { 1295 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1296 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1297 struct spdk_nvmf_tcp_qpair, qpair); 1298 1299 assert(tqpair != NULL); 1300 if (tqpair->qpair.transport->opts.c2h_success) { 1301 nvmf_tcp_request_free(tcp_req); 1302 } else { 1303 nvmf_tcp_req_pdu_fini(tcp_req); 1304 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1305 } 1306 } 1307 1308 static void 1309 nvmf_tcp_r2t_complete(void *cb_arg) 1310 { 1311 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1312 struct spdk_nvmf_tcp_transport *ttransport; 1313 1314 nvmf_tcp_req_pdu_fini(tcp_req); 1315 1316 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1317 struct spdk_nvmf_tcp_transport, transport); 1318 1319 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1320 1321 if (tcp_req->h2c_offset == tcp_req->req.length) { 1322 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1323 nvmf_tcp_req_process(ttransport, tcp_req); 1324 } 1325 } 1326 1327 static void 1328 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1329 struct spdk_nvmf_tcp_req *tcp_req) 1330 { 1331 struct nvme_tcp_pdu *rsp_pdu; 1332 struct spdk_nvme_tcp_r2t_hdr *r2t; 1333 1334 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1335 assert(rsp_pdu != NULL); 1336 1337 r2t = &rsp_pdu->hdr.r2t; 1338 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1339 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1340 1341 if (tqpair->host_hdgst_enable) { 1342 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1343 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1344 } 1345 1346 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1347 r2t->ttag = tcp_req->ttag; 1348 r2t->r2to = tcp_req->h2c_offset; 1349 r2t->r2tl = tcp_req->req.length; 1350 1351 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1352 1353 SPDK_DEBUGLOG(nvmf_tcp, 1354 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1355 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1356 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1357 } 1358 1359 static void 1360 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1361 struct spdk_nvmf_tcp_qpair *tqpair, 1362 struct nvme_tcp_pdu *pdu) 1363 { 1364 struct spdk_nvmf_tcp_req *tcp_req; 1365 1366 tcp_req = pdu->req; 1367 assert(tcp_req != NULL); 1368 1369 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1370 1371 tcp_req->h2c_offset += pdu->data_len; 1372 1373 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1374 1375 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1376 * acknowledged before moving on. */ 1377 if (tcp_req->h2c_offset == tcp_req->req.length && 1378 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1379 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1380 nvmf_tcp_req_process(ttransport, tcp_req); 1381 } 1382 } 1383 1384 static void 1385 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1386 { 1387 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1388 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1389 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1390 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1391 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1392 DGET32(h2c_term_req->fei)); 1393 } 1394 } 1395 1396 static void 1397 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1398 struct nvme_tcp_pdu *pdu) 1399 { 1400 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1401 uint32_t error_offset = 0; 1402 enum spdk_nvme_tcp_term_req_fes fes; 1403 1404 1405 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1406 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1407 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1408 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1409 goto end; 1410 } 1411 1412 /* set the data buffer */ 1413 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1414 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1415 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1416 return; 1417 end: 1418 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1419 } 1420 1421 static void 1422 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1423 struct nvme_tcp_pdu *pdu) 1424 { 1425 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1426 1427 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1428 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1429 } 1430 1431 static void 1432 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1433 struct spdk_nvmf_tcp_transport *ttransport) 1434 { 1435 int rc = 0; 1436 struct nvme_tcp_pdu *pdu; 1437 uint32_t crc32c, error_offset = 0; 1438 enum spdk_nvme_tcp_term_req_fes fes; 1439 1440 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1441 pdu = &tqpair->pdu_in_progress; 1442 1443 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1444 /* check data digest if need */ 1445 if (pdu->ddgst_enable) { 1446 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1447 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1448 if (rc == 0) { 1449 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1450 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1451 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1452 return; 1453 1454 } 1455 } 1456 1457 switch (pdu->hdr.common.pdu_type) { 1458 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1459 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1460 break; 1461 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1462 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1463 break; 1464 1465 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1466 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1467 break; 1468 1469 default: 1470 /* The code should not go to here */ 1471 SPDK_ERRLOG("The code should not go to here\n"); 1472 break; 1473 } 1474 } 1475 1476 static void 1477 nvmf_tcp_send_icresp_complete(void *cb_arg) 1478 { 1479 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1480 1481 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1482 } 1483 1484 static void 1485 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1486 struct spdk_nvmf_tcp_qpair *tqpair, 1487 struct nvme_tcp_pdu *pdu) 1488 { 1489 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1490 struct nvme_tcp_pdu *rsp_pdu; 1491 struct spdk_nvme_tcp_ic_resp *ic_resp; 1492 uint32_t error_offset = 0; 1493 enum spdk_nvme_tcp_term_req_fes fes; 1494 1495 /* Only PFV 0 is defined currently */ 1496 if (ic_req->pfv != 0) { 1497 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1498 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1499 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1500 goto end; 1501 } 1502 1503 /* MAXR2T is 0's based */ 1504 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1505 1506 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1507 if (!tqpair->host_hdgst_enable) { 1508 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1509 } 1510 1511 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1512 if (!tqpair->host_ddgst_enable) { 1513 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1514 } 1515 1516 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1517 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1518 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1519 tqpair, 1520 tqpair->recv_buf_size); 1521 /* Not fatal. */ 1522 } 1523 1524 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1525 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1526 1527 rsp_pdu = tqpair->mgmt_pdu; 1528 1529 ic_resp = &rsp_pdu->hdr.ic_resp; 1530 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1531 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1532 ic_resp->pfv = 0; 1533 ic_resp->cpda = tqpair->cpda; 1534 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1535 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1536 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1537 1538 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1539 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1540 1541 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1542 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1543 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1544 return; 1545 end: 1546 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1547 } 1548 1549 static void 1550 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1551 struct spdk_nvmf_tcp_transport *ttransport) 1552 { 1553 struct nvme_tcp_pdu *pdu; 1554 int rc; 1555 uint32_t crc32c, error_offset = 0; 1556 enum spdk_nvme_tcp_term_req_fes fes; 1557 1558 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1559 pdu = &tqpair->pdu_in_progress; 1560 1561 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1562 pdu->hdr.common.pdu_type); 1563 /* check header digest if needed */ 1564 if (pdu->has_hdgst) { 1565 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1566 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1567 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1568 if (rc == 0) { 1569 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1570 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1571 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1572 return; 1573 1574 } 1575 } 1576 1577 switch (pdu->hdr.common.pdu_type) { 1578 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1579 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1580 break; 1581 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1582 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1583 break; 1584 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1585 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1586 break; 1587 1588 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1589 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1590 break; 1591 1592 default: 1593 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1594 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1595 error_offset = 1; 1596 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1597 break; 1598 } 1599 } 1600 1601 static void 1602 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1603 { 1604 struct nvme_tcp_pdu *pdu; 1605 uint32_t error_offset = 0; 1606 enum spdk_nvme_tcp_term_req_fes fes; 1607 uint8_t expected_hlen, pdo; 1608 bool plen_error = false, pdo_error = false; 1609 1610 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1611 pdu = &tqpair->pdu_in_progress; 1612 1613 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1614 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1615 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1616 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1617 goto err; 1618 } 1619 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1620 if (pdu->hdr.common.plen != expected_hlen) { 1621 plen_error = true; 1622 } 1623 } else { 1624 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1625 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1626 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1627 goto err; 1628 } 1629 1630 switch (pdu->hdr.common.pdu_type) { 1631 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1632 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1633 pdo = pdu->hdr.common.pdo; 1634 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1635 pdo_error = true; 1636 break; 1637 } 1638 1639 if (pdu->hdr.common.plen < expected_hlen) { 1640 plen_error = true; 1641 } 1642 break; 1643 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1644 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1645 pdo = pdu->hdr.common.pdo; 1646 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1647 pdo_error = true; 1648 break; 1649 } 1650 if (pdu->hdr.common.plen < expected_hlen) { 1651 plen_error = true; 1652 } 1653 break; 1654 1655 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1656 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1657 if ((pdu->hdr.common.plen <= expected_hlen) || 1658 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1659 plen_error = true; 1660 } 1661 break; 1662 1663 default: 1664 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1665 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1666 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1667 goto err; 1668 } 1669 } 1670 1671 if (pdu->hdr.common.hlen != expected_hlen) { 1672 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1673 pdu->hdr.common.pdu_type, 1674 expected_hlen, pdu->hdr.common.hlen, tqpair); 1675 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1676 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1677 goto err; 1678 } else if (pdo_error) { 1679 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1680 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1681 } else if (plen_error) { 1682 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1683 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1684 goto err; 1685 } else { 1686 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1687 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1688 return; 1689 } 1690 err: 1691 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1692 } 1693 1694 static int 1695 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1696 int read_len) 1697 { 1698 int rc; 1699 1700 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1701 read_offset, read_len, pdu->dif_ctx); 1702 if (rc != 0) { 1703 SPDK_ERRLOG("DIF generate failed\n"); 1704 } 1705 1706 return rc; 1707 } 1708 1709 static int 1710 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1711 { 1712 int rc = 0; 1713 struct nvme_tcp_pdu *pdu; 1714 enum nvme_tcp_pdu_recv_state prev_state; 1715 uint32_t data_len; 1716 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1717 struct spdk_nvmf_tcp_transport, transport); 1718 1719 /* The loop here is to allow for several back-to-back state changes. */ 1720 do { 1721 prev_state = tqpair->recv_state; 1722 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1723 1724 pdu = &tqpair->pdu_in_progress; 1725 switch (tqpair->recv_state) { 1726 /* Wait for the common header */ 1727 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1728 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1729 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1730 return rc; 1731 } 1732 1733 rc = nvme_tcp_read_data(tqpair->sock, 1734 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1735 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1736 if (rc < 0) { 1737 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1738 return NVME_TCP_PDU_FATAL; 1739 } else if (rc > 0) { 1740 pdu->ch_valid_bytes += rc; 1741 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1742 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1743 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1744 } 1745 } 1746 1747 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1748 return NVME_TCP_PDU_IN_PROGRESS; 1749 } 1750 1751 /* The command header of this PDU has now been read from the socket. */ 1752 nvmf_tcp_pdu_ch_handle(tqpair); 1753 break; 1754 /* Wait for the pdu specific header */ 1755 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1756 rc = nvme_tcp_read_data(tqpair->sock, 1757 pdu->psh_len - pdu->psh_valid_bytes, 1758 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1759 if (rc < 0) { 1760 return NVME_TCP_PDU_FATAL; 1761 } else if (rc > 0) { 1762 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1763 0, rc, 0, 0); 1764 pdu->psh_valid_bytes += rc; 1765 } 1766 1767 if (pdu->psh_valid_bytes < pdu->psh_len) { 1768 return NVME_TCP_PDU_IN_PROGRESS; 1769 } 1770 1771 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1772 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1773 break; 1774 /* Wait for the req slot */ 1775 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1776 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1777 break; 1778 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1779 /* check whether the data is valid, if not we just return */ 1780 if (!pdu->data_len) { 1781 return NVME_TCP_PDU_IN_PROGRESS; 1782 } 1783 1784 data_len = pdu->data_len; 1785 /* data digest */ 1786 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1787 tqpair->host_ddgst_enable)) { 1788 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1789 pdu->ddgst_enable = true; 1790 } 1791 1792 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1793 if (rc < 0) { 1794 return NVME_TCP_PDU_FATAL; 1795 } 1796 pdu->readv_offset += rc; 1797 1798 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1799 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1800 if (rc != 0) { 1801 return NVME_TCP_PDU_FATAL; 1802 } 1803 } 1804 1805 if (pdu->readv_offset < data_len) { 1806 return NVME_TCP_PDU_IN_PROGRESS; 1807 } 1808 1809 /* All of this PDU has now been read from the socket. */ 1810 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1811 break; 1812 case NVME_TCP_PDU_RECV_STATE_ERROR: 1813 if (!spdk_sock_is_connected(tqpair->sock)) { 1814 return NVME_TCP_PDU_FATAL; 1815 } 1816 break; 1817 default: 1818 assert(0); 1819 SPDK_ERRLOG("code should not come to here"); 1820 break; 1821 } 1822 } while (tqpair->recv_state != prev_state); 1823 1824 return rc; 1825 } 1826 1827 static int 1828 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 1829 struct spdk_nvmf_transport *transport, 1830 struct spdk_nvmf_transport_poll_group *group) 1831 { 1832 struct spdk_nvmf_request *req = &tcp_req->req; 1833 struct spdk_nvme_cmd *cmd; 1834 struct spdk_nvme_cpl *rsp; 1835 struct spdk_nvme_sgl_descriptor *sgl; 1836 uint32_t length; 1837 1838 cmd = &req->cmd->nvme_cmd; 1839 rsp = &req->rsp->nvme_cpl; 1840 sgl = &cmd->dptr.sgl1; 1841 1842 length = sgl->unkeyed.length; 1843 1844 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 1845 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 1846 if (length > transport->opts.max_io_size) { 1847 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 1848 length, transport->opts.max_io_size); 1849 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1850 return -1; 1851 } 1852 1853 /* fill request length and populate iovs */ 1854 req->length = length; 1855 1856 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 1857 1858 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1859 req->dif.orig_length = length; 1860 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1861 req->dif.elba_length = length; 1862 } 1863 1864 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 1865 /* No available buffers. Queue this request up. */ 1866 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 1867 tcp_req); 1868 return 0; 1869 } 1870 1871 /* backward compatible */ 1872 req->data = req->iov[0].iov_base; 1873 1874 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 1875 tcp_req, req->iovcnt, req->data); 1876 1877 return 0; 1878 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 1879 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 1880 uint64_t offset = sgl->address; 1881 uint32_t max_len = transport->opts.in_capsule_data_size; 1882 1883 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 1884 offset, length); 1885 1886 if (offset > max_len) { 1887 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 1888 offset, max_len); 1889 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 1890 return -1; 1891 } 1892 max_len -= (uint32_t)offset; 1893 1894 if (length > max_len) { 1895 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 1896 length, max_len); 1897 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1898 return -1; 1899 } 1900 1901 req->data = tcp_req->buf + offset; 1902 req->data_from_pool = false; 1903 req->length = length; 1904 1905 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1906 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1907 req->dif.elba_length = length; 1908 } 1909 1910 req->iov[0].iov_base = req->data; 1911 req->iov[0].iov_len = length; 1912 req->iovcnt = 1; 1913 1914 return 0; 1915 } 1916 1917 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 1918 sgl->generic.type, sgl->generic.subtype); 1919 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 1920 return -1; 1921 } 1922 1923 static inline enum spdk_nvme_media_error_status_code 1924 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 1925 enum spdk_nvme_media_error_status_code result; 1926 1927 switch (err_type) 1928 { 1929 case SPDK_DIF_REFTAG_ERROR: 1930 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 1931 break; 1932 case SPDK_DIF_APPTAG_ERROR: 1933 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 1934 break; 1935 case SPDK_DIF_GUARD_ERROR: 1936 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 1937 break; 1938 default: 1939 SPDK_UNREACHABLE(); 1940 break; 1941 } 1942 1943 return result; 1944 } 1945 1946 static void 1947 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 1948 struct spdk_nvmf_tcp_req *tcp_req) 1949 { 1950 struct nvme_tcp_pdu *rsp_pdu; 1951 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 1952 uint32_t plen, pdo, alignment; 1953 int rc; 1954 1955 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1956 1957 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1958 assert(rsp_pdu != NULL); 1959 1960 c2h_data = &rsp_pdu->hdr.c2h_data; 1961 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 1962 plen = c2h_data->common.hlen = sizeof(*c2h_data); 1963 1964 if (tqpair->host_hdgst_enable) { 1965 plen += SPDK_NVME_TCP_DIGEST_LEN; 1966 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1967 } 1968 1969 /* set the psh */ 1970 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1971 c2h_data->datal = tcp_req->req.length; 1972 c2h_data->datao = 0; 1973 1974 /* set the padding */ 1975 rsp_pdu->padding_len = 0; 1976 pdo = plen; 1977 if (tqpair->cpda) { 1978 alignment = (tqpair->cpda + 1) << 2; 1979 if (alignment > plen) { 1980 rsp_pdu->padding_len = alignment - plen; 1981 pdo = plen = alignment; 1982 } 1983 } 1984 1985 c2h_data->common.pdo = pdo; 1986 plen += c2h_data->datal; 1987 if (tqpair->host_ddgst_enable) { 1988 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 1989 plen += SPDK_NVME_TCP_DIGEST_LEN; 1990 } 1991 1992 c2h_data->common.plen = plen; 1993 1994 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1995 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1996 } 1997 1998 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1999 c2h_data->datao, c2h_data->datal); 2000 2001 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2002 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2003 struct spdk_dif_error err_blk = {}; 2004 2005 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2006 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2007 if (rc != 0) { 2008 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2009 err_blk.err_type, err_blk.err_offset); 2010 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2011 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2012 nvmf_tcp_req_pdu_fini(tcp_req); 2013 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2014 return; 2015 } 2016 } 2017 2018 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2019 if (tqpair->qpair.transport->opts.c2h_success) { 2020 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2021 } 2022 2023 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2024 } 2025 2026 static int 2027 request_transfer_out(struct spdk_nvmf_request *req) 2028 { 2029 struct spdk_nvmf_tcp_req *tcp_req; 2030 struct spdk_nvmf_qpair *qpair; 2031 struct spdk_nvmf_tcp_qpair *tqpair; 2032 struct spdk_nvme_cpl *rsp; 2033 2034 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2035 2036 qpair = req->qpair; 2037 rsp = &req->rsp->nvme_cpl; 2038 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2039 2040 /* Advance our sq_head pointer */ 2041 if (qpair->sq_head == qpair->sq_head_max) { 2042 qpair->sq_head = 0; 2043 } else { 2044 qpair->sq_head++; 2045 } 2046 rsp->sqhd = qpair->sq_head; 2047 2048 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2049 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2050 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2051 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2052 } else { 2053 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2054 } 2055 2056 return 0; 2057 } 2058 2059 static void 2060 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2061 struct spdk_nvmf_tcp_req *tcp_req) 2062 { 2063 struct nvme_tcp_pdu *pdu; 2064 uint32_t plen = 0; 2065 2066 pdu = &tqpair->pdu_in_progress; 2067 plen = pdu->hdr.common.hlen; 2068 2069 if (tqpair->host_hdgst_enable) { 2070 plen += SPDK_NVME_TCP_DIGEST_LEN; 2071 } 2072 2073 if (pdu->hdr.common.plen != plen) { 2074 tcp_req->has_incapsule_data = true; 2075 } 2076 } 2077 2078 static bool 2079 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2080 struct spdk_nvmf_tcp_req *tcp_req) 2081 { 2082 struct spdk_nvmf_tcp_qpair *tqpair; 2083 int rc; 2084 enum spdk_nvmf_tcp_req_state prev_state; 2085 bool progress = false; 2086 struct spdk_nvmf_transport *transport = &ttransport->transport; 2087 struct spdk_nvmf_transport_poll_group *group; 2088 2089 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2090 group = &tqpair->group->group; 2091 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2092 2093 /* If the qpair is not active, we need to abort the outstanding requests. */ 2094 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2095 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2096 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2097 } 2098 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2099 } 2100 2101 /* The loop here is to allow for several back-to-back state changes. */ 2102 do { 2103 prev_state = tcp_req->state; 2104 2105 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2106 tqpair); 2107 2108 switch (tcp_req->state) { 2109 case TCP_REQUEST_STATE_FREE: 2110 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2111 * to escape this state. */ 2112 break; 2113 case TCP_REQUEST_STATE_NEW: 2114 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2115 2116 /* copy the cmd from the receive pdu */ 2117 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2118 2119 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2120 tcp_req->req.dif.dif_insert_or_strip = true; 2121 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2122 } 2123 2124 /* The next state transition depends on the data transfer needs of this request. */ 2125 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2126 2127 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2128 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2129 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2130 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2131 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2132 break; 2133 } 2134 2135 /* If no data to transfer, ready to execute. */ 2136 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2137 /* Reset the tqpair receving pdu state */ 2138 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2139 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2140 break; 2141 } 2142 2143 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2144 2145 if (!tcp_req->has_incapsule_data) { 2146 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2147 } 2148 2149 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2150 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2151 break; 2152 case TCP_REQUEST_STATE_NEED_BUFFER: 2153 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2154 2155 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2156 2157 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2158 SPDK_DEBUGLOG(nvmf_tcp, 2159 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2160 tcp_req, tqpair); 2161 /* This request needs to wait in line to obtain a buffer */ 2162 break; 2163 } 2164 2165 /* Try to get a data buffer */ 2166 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2167 if (rc < 0) { 2168 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2169 /* Reset the tqpair receving pdu state */ 2170 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2171 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2172 break; 2173 } 2174 2175 if (!tcp_req->req.data) { 2176 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2177 tcp_req, tqpair); 2178 /* No buffers available. */ 2179 break; 2180 } 2181 2182 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2183 2184 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2185 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2186 if (tcp_req->req.data_from_pool) { 2187 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2188 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2189 } else { 2190 struct nvme_tcp_pdu *pdu; 2191 2192 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2193 2194 pdu = &tqpair->pdu_in_progress; 2195 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2196 tqpair); 2197 /* No need to send r2t, contained in the capsuled data */ 2198 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2199 0, tcp_req->req.length); 2200 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2201 } 2202 break; 2203 } 2204 2205 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2206 break; 2207 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2208 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2209 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2210 break; 2211 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2212 2213 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2214 (uintptr_t)tcp_req, 0); 2215 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2216 * to escape this state. */ 2217 break; 2218 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2219 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2220 2221 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2222 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2223 tcp_req->req.length = tcp_req->req.dif.elba_length; 2224 } 2225 2226 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2227 spdk_nvmf_request_exec(&tcp_req->req); 2228 break; 2229 case TCP_REQUEST_STATE_EXECUTING: 2230 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2231 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2232 * to escape this state. */ 2233 break; 2234 case TCP_REQUEST_STATE_EXECUTED: 2235 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2236 2237 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2238 tcp_req->req.length = tcp_req->req.dif.orig_length; 2239 } 2240 2241 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2242 break; 2243 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2244 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2245 rc = request_transfer_out(&tcp_req->req); 2246 assert(rc == 0); /* No good way to handle this currently */ 2247 break; 2248 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2249 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2250 (uintptr_t)tcp_req, 2251 0); 2252 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2253 * to escape this state. */ 2254 break; 2255 case TCP_REQUEST_STATE_COMPLETED: 2256 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2257 if (tcp_req->req.data_from_pool) { 2258 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2259 } 2260 tcp_req->req.length = 0; 2261 tcp_req->req.iovcnt = 0; 2262 tcp_req->req.data = NULL; 2263 2264 nvmf_tcp_req_pdu_fini(tcp_req); 2265 2266 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2267 break; 2268 case TCP_REQUEST_NUM_STATES: 2269 default: 2270 assert(0); 2271 break; 2272 } 2273 2274 if (tcp_req->state != prev_state) { 2275 progress = true; 2276 } 2277 } while (tcp_req->state != prev_state); 2278 2279 return progress; 2280 } 2281 2282 static void 2283 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2284 { 2285 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2286 int rc; 2287 2288 assert(tqpair != NULL); 2289 rc = nvmf_tcp_sock_process(tqpair); 2290 2291 /* If there was a new socket error, disconnect */ 2292 if (rc < 0) { 2293 nvmf_tcp_qpair_disconnect(tqpair); 2294 } 2295 } 2296 2297 static int 2298 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2299 struct spdk_nvmf_qpair *qpair) 2300 { 2301 struct spdk_nvmf_tcp_poll_group *tgroup; 2302 struct spdk_nvmf_tcp_qpair *tqpair; 2303 int rc; 2304 2305 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2306 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2307 2308 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2309 nvmf_tcp_sock_cb, tqpair); 2310 if (rc != 0) { 2311 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2312 spdk_strerror(errno), errno); 2313 return -1; 2314 } 2315 2316 rc = nvmf_tcp_qpair_sock_init(tqpair); 2317 if (rc != 0) { 2318 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2319 return -1; 2320 } 2321 2322 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2323 if (rc < 0) { 2324 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2325 return -1; 2326 } 2327 2328 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2329 if (rc < 0) { 2330 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2331 return -1; 2332 } 2333 2334 tqpair->group = tgroup; 2335 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2336 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2337 2338 return 0; 2339 } 2340 2341 static int 2342 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2343 struct spdk_nvmf_qpair *qpair) 2344 { 2345 struct spdk_nvmf_tcp_poll_group *tgroup; 2346 struct spdk_nvmf_tcp_qpair *tqpair; 2347 int rc; 2348 2349 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2350 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2351 2352 assert(tqpair->group == tgroup); 2353 2354 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2355 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2356 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2357 } else { 2358 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2359 } 2360 2361 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2362 if (rc != 0) { 2363 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2364 spdk_strerror(errno), errno); 2365 } 2366 2367 return rc; 2368 } 2369 2370 static int 2371 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2372 { 2373 struct spdk_nvmf_tcp_transport *ttransport; 2374 struct spdk_nvmf_tcp_req *tcp_req; 2375 2376 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2377 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2378 2379 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2380 nvmf_tcp_req_process(ttransport, tcp_req); 2381 2382 return 0; 2383 } 2384 2385 static void 2386 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2387 { 2388 struct spdk_nvmf_tcp_qpair *tqpair; 2389 2390 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2391 2392 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2393 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2394 nvmf_tcp_qpair_destroy(tqpair); 2395 } 2396 2397 static int 2398 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2399 { 2400 struct spdk_nvmf_tcp_poll_group *tgroup; 2401 int rc; 2402 struct spdk_nvmf_request *req, *req_tmp; 2403 struct spdk_nvmf_tcp_req *tcp_req; 2404 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2405 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2406 struct spdk_nvmf_tcp_transport, transport); 2407 2408 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2409 2410 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2411 return 0; 2412 } 2413 2414 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2415 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2416 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2417 break; 2418 } 2419 } 2420 2421 rc = spdk_sock_group_poll(tgroup->sock_group); 2422 if (rc < 0) { 2423 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2424 } 2425 2426 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2427 nvmf_tcp_sock_process(tqpair); 2428 } 2429 2430 return rc; 2431 } 2432 2433 static int 2434 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2435 struct spdk_nvme_transport_id *trid, bool peer) 2436 { 2437 struct spdk_nvmf_tcp_qpair *tqpair; 2438 uint16_t port; 2439 2440 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2441 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2442 2443 if (peer) { 2444 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2445 port = tqpair->initiator_port; 2446 } else { 2447 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2448 port = tqpair->target_port; 2449 } 2450 2451 if (spdk_sock_is_ipv4(tqpair->sock)) { 2452 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2453 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2454 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2455 } else { 2456 return -1; 2457 } 2458 2459 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2460 return 0; 2461 } 2462 2463 static int 2464 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2465 struct spdk_nvme_transport_id *trid) 2466 { 2467 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2468 } 2469 2470 static int 2471 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2472 struct spdk_nvme_transport_id *trid) 2473 { 2474 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2475 } 2476 2477 static int 2478 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2479 struct spdk_nvme_transport_id *trid) 2480 { 2481 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2482 } 2483 2484 static void 2485 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2486 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2487 { 2488 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2489 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2490 2491 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2492 2493 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2494 } 2495 2496 static int 2497 _nvmf_tcp_qpair_abort_request(void *ctx) 2498 { 2499 struct spdk_nvmf_request *req = ctx; 2500 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2501 struct spdk_nvmf_tcp_req, req); 2502 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2503 struct spdk_nvmf_tcp_qpair, qpair); 2504 int rc; 2505 2506 spdk_poller_unregister(&req->poller); 2507 2508 switch (tcp_req_to_abort->state) { 2509 case TCP_REQUEST_STATE_EXECUTING: 2510 rc = nvmf_ctrlr_abort_request(req); 2511 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2512 return SPDK_POLLER_BUSY; 2513 } 2514 break; 2515 2516 case TCP_REQUEST_STATE_NEED_BUFFER: 2517 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2518 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2519 2520 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2521 break; 2522 2523 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2524 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2525 break; 2526 2527 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2528 if (spdk_get_ticks() < req->timeout_tsc) { 2529 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2530 return SPDK_POLLER_BUSY; 2531 } 2532 break; 2533 2534 default: 2535 break; 2536 } 2537 2538 spdk_nvmf_request_complete(req); 2539 return SPDK_POLLER_BUSY; 2540 } 2541 2542 static void 2543 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2544 struct spdk_nvmf_request *req) 2545 { 2546 struct spdk_nvmf_tcp_qpair *tqpair; 2547 struct spdk_nvmf_tcp_transport *ttransport; 2548 struct spdk_nvmf_transport *transport; 2549 uint16_t cid; 2550 uint32_t i; 2551 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2552 2553 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2554 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2555 transport = &ttransport->transport; 2556 2557 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2558 2559 for (i = 0; i < tqpair->resource_count; i++) { 2560 tcp_req_to_abort = &tqpair->reqs[i]; 2561 2562 if (tcp_req_to_abort->state != TCP_REQUEST_STATE_FREE && 2563 tcp_req_to_abort->req.cmd->nvme_cmd.cid == cid) { 2564 break; 2565 } 2566 } 2567 2568 if (tcp_req_to_abort == NULL) { 2569 spdk_nvmf_request_complete(req); 2570 return; 2571 } 2572 2573 req->req_to_abort = &tcp_req_to_abort->req; 2574 req->timeout_tsc = spdk_get_ticks() + 2575 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2576 req->poller = NULL; 2577 2578 _nvmf_tcp_qpair_abort_request(req); 2579 } 2580 2581 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2582 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2583 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2584 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2585 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2586 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2587 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2588 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2589 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 2590 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2591 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 2592 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2593 2594 static void 2595 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2596 { 2597 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2598 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2599 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2600 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2601 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2602 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2603 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2604 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2605 opts->c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 2606 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2607 opts->sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 2608 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2609 } 2610 2611 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2612 .name = "TCP", 2613 .type = SPDK_NVME_TRANSPORT_TCP, 2614 .opts_init = nvmf_tcp_opts_init, 2615 .create = nvmf_tcp_create, 2616 .destroy = nvmf_tcp_destroy, 2617 2618 .listen = nvmf_tcp_listen, 2619 .stop_listen = nvmf_tcp_stop_listen, 2620 .accept = nvmf_tcp_accept, 2621 2622 .listener_discover = nvmf_tcp_discover, 2623 2624 .poll_group_create = nvmf_tcp_poll_group_create, 2625 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2626 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2627 .poll_group_add = nvmf_tcp_poll_group_add, 2628 .poll_group_remove = nvmf_tcp_poll_group_remove, 2629 .poll_group_poll = nvmf_tcp_poll_group_poll, 2630 2631 .req_free = nvmf_tcp_req_free, 2632 .req_complete = nvmf_tcp_req_complete, 2633 2634 .qpair_fini = nvmf_tcp_close_qpair, 2635 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2636 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2637 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2638 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2639 }; 2640 2641 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2642 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2643