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 nvme_tcp_pdu_recv_buf { 202 char *buf; 203 uint32_t off; 204 uint32_t size; 205 uint32_t remain_size; 206 }; 207 208 struct spdk_nvmf_tcp_qpair { 209 struct spdk_nvmf_qpair qpair; 210 struct spdk_nvmf_tcp_poll_group *group; 211 struct spdk_nvmf_tcp_port *port; 212 struct spdk_sock *sock; 213 214 enum nvme_tcp_pdu_recv_state recv_state; 215 enum nvme_tcp_qpair_state state; 216 217 /* PDU being actively received */ 218 struct nvme_tcp_pdu pdu_in_progress; 219 struct nvme_tcp_pdu_recv_buf pdu_recv_buf; 220 221 /* This is a spare PDU used for sending special management 222 * operations. Primarily, this is used for the initial 223 * connection response and c2h termination request. */ 224 struct nvme_tcp_pdu mgmt_pdu; 225 226 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 227 228 /* Arrays of in-capsule buffers, requests, and pdus. 229 * Each array is 'resource_count' number of elements */ 230 void *bufs; 231 struct spdk_nvmf_tcp_req *reqs; 232 struct nvme_tcp_pdu *pdus; 233 uint32_t resource_count; 234 235 /* Queues to track the requests in all states */ 236 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 237 /* Number of requests in each state */ 238 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 239 240 uint8_t cpda; 241 242 bool host_hdgst_enable; 243 bool host_ddgst_enable; 244 245 /* IP address */ 246 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 247 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 248 249 /* IP port */ 250 uint16_t initiator_port; 251 uint16_t target_port; 252 253 /* Timer used to destroy qpair after detecting transport error issue if initiator does 254 * not close the connection. 255 */ 256 struct spdk_poller *timeout_poller; 257 258 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 259 }; 260 261 struct spdk_nvmf_tcp_poll_group { 262 struct spdk_nvmf_transport_poll_group group; 263 struct spdk_sock_group *sock_group; 264 265 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 266 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 267 }; 268 269 struct spdk_nvmf_tcp_port { 270 const struct spdk_nvme_transport_id *trid; 271 struct spdk_sock *listen_sock; 272 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 273 }; 274 275 struct spdk_nvmf_tcp_transport { 276 struct spdk_nvmf_transport transport; 277 278 pthread_mutex_t lock; 279 280 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 281 }; 282 283 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 284 struct spdk_nvmf_tcp_req *tcp_req); 285 286 static void 287 spdk_nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 288 enum spdk_nvmf_tcp_req_state state) 289 { 290 struct spdk_nvmf_qpair *qpair; 291 struct spdk_nvmf_tcp_qpair *tqpair; 292 293 qpair = tcp_req->req.qpair; 294 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 295 296 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 297 assert(tqpair->state_cntr[tcp_req->state] > 0); 298 tqpair->state_cntr[tcp_req->state]--; 299 300 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 301 tqpair->state_cntr[state]++; 302 303 tcp_req->state = state; 304 } 305 306 static inline struct nvme_tcp_pdu * 307 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 308 { 309 assert(tcp_req->pdu_in_use == false); 310 tcp_req->pdu_in_use = true; 311 312 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 313 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 314 tcp_req->pdu->hdr = &tcp_req->pdu->hdr_mem; 315 316 return tcp_req->pdu; 317 } 318 319 static inline void 320 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 321 { 322 tcp_req->pdu_in_use = false; 323 } 324 325 static struct spdk_nvmf_tcp_req * 326 spdk_nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 327 { 328 struct spdk_nvmf_tcp_req *tcp_req; 329 330 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 331 if (!tcp_req) { 332 return NULL; 333 } 334 335 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 336 tcp_req->h2c_offset = 0; 337 tcp_req->has_incapsule_data = false; 338 tcp_req->req.dif.dif_insert_or_strip = false; 339 340 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 341 return tcp_req; 342 } 343 344 static void 345 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 346 { 347 struct spdk_nvmf_tcp_transport *ttransport; 348 349 assert(tcp_req != NULL); 350 351 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req); 352 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 353 struct spdk_nvmf_tcp_transport, transport); 354 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 355 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 356 } 357 358 static int 359 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req) 360 { 361 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 362 363 nvmf_tcp_request_free(tcp_req); 364 365 return 0; 366 } 367 368 static void 369 spdk_nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 370 enum spdk_nvmf_tcp_req_state state) 371 { 372 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 373 374 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 375 nvmf_tcp_request_free(tcp_req); 376 } 377 } 378 379 static void 380 spdk_nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 381 { 382 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 383 384 assert(TAILQ_EMPTY(&tqpair->send_queue)); 385 386 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 387 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 388 389 /* Wipe the requests waiting for buffer from the global list */ 390 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 391 req_tmp) { 392 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 393 spdk_nvmf_request, buf_link); 394 } 395 396 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 397 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 398 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 399 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 400 } 401 402 static void 403 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 404 { 405 int i; 406 struct spdk_nvmf_tcp_req *tcp_req; 407 408 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 409 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 410 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 411 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 412 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 413 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 414 } 415 } 416 } 417 418 static void 419 spdk_nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 420 { 421 int err = 0; 422 423 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 424 425 err = spdk_sock_close(&tqpair->sock); 426 assert(err == 0); 427 spdk_nvmf_tcp_cleanup_all_states(tqpair); 428 429 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 430 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 431 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 432 tqpair->resource_count); 433 err++; 434 } 435 436 if (err > 0) { 437 nvmf_tcp_dump_qpair_req_contents(tqpair); 438 } 439 440 spdk_dma_free(tqpair->pdus); 441 free(tqpair->reqs); 442 spdk_free(tqpair->bufs); 443 free(tqpair->pdu_recv_buf.buf); 444 free(tqpair); 445 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n"); 446 } 447 448 static int 449 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 450 { 451 struct spdk_nvmf_tcp_transport *ttransport; 452 453 assert(transport != NULL); 454 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 455 456 pthread_mutex_destroy(&ttransport->lock); 457 free(ttransport); 458 return 0; 459 } 460 461 static struct spdk_nvmf_transport * 462 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 463 { 464 struct spdk_nvmf_tcp_transport *ttransport; 465 uint32_t sge_count; 466 uint32_t min_shared_buffers; 467 468 ttransport = calloc(1, sizeof(*ttransport)); 469 if (!ttransport) { 470 return NULL; 471 } 472 473 TAILQ_INIT(&ttransport->ports); 474 475 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 476 477 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 478 479 SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n" 480 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 481 " max_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 482 " in_capsule_data_size=%d, max_aq_depth=%d\n" 483 " num_shared_buffers=%d, c2h_success=%d,\n" 484 " dif_insert_or_strip=%d, sock_priority=%d\n", 485 opts->max_queue_depth, 486 opts->max_io_size, 487 opts->max_qpairs_per_ctrlr, 488 opts->io_unit_size, 489 opts->in_capsule_data_size, 490 opts->max_aq_depth, 491 opts->num_shared_buffers, 492 opts->c2h_success, 493 opts->dif_insert_or_strip, 494 opts->sock_priority); 495 496 if (opts->sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 497 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 498 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 499 opts->sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 500 free(ttransport); 501 return NULL; 502 } 503 504 /* I/O unit size cannot be larger than max I/O size */ 505 if (opts->io_unit_size > opts->max_io_size) { 506 opts->io_unit_size = opts->max_io_size; 507 } 508 509 sge_count = opts->max_io_size / opts->io_unit_size; 510 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 511 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 512 free(ttransport); 513 return NULL; 514 } 515 516 min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size; 517 if (min_shared_buffers > opts->num_shared_buffers) { 518 SPDK_ERRLOG("There are not enough buffers to satisfy" 519 "per-poll group caches for each thread. (%" PRIu32 ")" 520 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 521 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 522 spdk_nvmf_tcp_destroy(&ttransport->transport); 523 return NULL; 524 } 525 526 pthread_mutex_init(&ttransport->lock, NULL); 527 528 return &ttransport->transport; 529 } 530 531 static int 532 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid) 533 { 534 unsigned long long ull; 535 char *end = NULL; 536 537 ull = strtoull(trsvcid, &end, 10); 538 if (end == NULL || end == trsvcid || *end != '\0') { 539 return -1; 540 } 541 542 /* Valid TCP/IP port numbers are in [0, 65535] */ 543 if (ull > 65535) { 544 return -1; 545 } 546 547 return (int)ull; 548 } 549 550 /** 551 * Canonicalize a listen address trid. 552 */ 553 static int 554 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 555 const struct spdk_nvme_transport_id *trid) 556 { 557 int trsvcid_int; 558 559 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 560 if (trsvcid_int < 0) { 561 return -EINVAL; 562 } 563 564 memset(canon_trid, 0, sizeof(*canon_trid)); 565 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 566 canon_trid->adrfam = trid->adrfam; 567 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 568 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 569 570 return 0; 571 } 572 573 /** 574 * Find an existing listening port. 575 * 576 * Caller must hold ttransport->lock. 577 */ 578 static struct spdk_nvmf_tcp_port * 579 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 580 const struct spdk_nvme_transport_id *trid) 581 { 582 struct spdk_nvme_transport_id canon_trid; 583 struct spdk_nvmf_tcp_port *port; 584 585 if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 586 return NULL; 587 } 588 589 TAILQ_FOREACH(port, &ttransport->ports, link) { 590 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 591 return port; 592 } 593 } 594 595 return NULL; 596 } 597 598 static int 599 spdk_nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 600 const struct spdk_nvme_transport_id *trid, 601 spdk_nvmf_tgt_listen_done_fn cb_fn, 602 void *cb_arg) 603 { 604 struct spdk_nvmf_tcp_transport *ttransport; 605 struct spdk_nvmf_tcp_port *port; 606 int trsvcid_int; 607 uint8_t adrfam; 608 609 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 610 611 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 612 if (trsvcid_int < 0) { 613 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 614 return -EINVAL; 615 } 616 617 pthread_mutex_lock(&ttransport->lock); 618 port = calloc(1, sizeof(*port)); 619 if (!port) { 620 SPDK_ERRLOG("Port allocation failed\n"); 621 pthread_mutex_unlock(&ttransport->lock); 622 return -ENOMEM; 623 } 624 625 port->trid = trid; 626 port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int, NULL); 627 if (port->listen_sock == NULL) { 628 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 629 trid->traddr, trsvcid_int, 630 spdk_strerror(errno), errno); 631 free(port); 632 pthread_mutex_unlock(&ttransport->lock); 633 return -errno; 634 } 635 636 if (spdk_sock_is_ipv4(port->listen_sock)) { 637 adrfam = SPDK_NVMF_ADRFAM_IPV4; 638 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 639 adrfam = SPDK_NVMF_ADRFAM_IPV6; 640 } else { 641 SPDK_ERRLOG("Unhandled socket type\n"); 642 adrfam = 0; 643 } 644 645 if (adrfam != trid->adrfam) { 646 SPDK_ERRLOG("Socket address family mismatch\n"); 647 spdk_sock_close(&port->listen_sock); 648 free(port); 649 pthread_mutex_unlock(&ttransport->lock); 650 return -EINVAL; 651 } 652 653 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 654 trid->traddr, trid->trsvcid); 655 656 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 657 pthread_mutex_unlock(&ttransport->lock); 658 659 if (cb_fn != NULL) { 660 cb_fn(cb_arg, 0); 661 } 662 663 return 0; 664 } 665 666 static void 667 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 668 const struct spdk_nvme_transport_id *trid) 669 { 670 struct spdk_nvmf_tcp_transport *ttransport; 671 struct spdk_nvmf_tcp_port *port; 672 673 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 674 675 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 676 trid->traddr, trid->trsvcid); 677 678 pthread_mutex_lock(&ttransport->lock); 679 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 680 if (port) { 681 TAILQ_REMOVE(&ttransport->ports, port, link); 682 spdk_sock_close(&port->listen_sock); 683 free(port); 684 } 685 686 pthread_mutex_unlock(&ttransport->lock); 687 } 688 689 static void spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 690 enum nvme_tcp_pdu_recv_state state); 691 692 static void 693 spdk_nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 694 { 695 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Disconnecting qpair %p\n", tqpair); 696 697 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 698 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 699 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 700 spdk_poller_unregister(&tqpair->timeout_poller); 701 702 /* This will end up calling spdk_nvmf_tcp_close_qpair */ 703 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 704 } 705 } 706 707 static void 708 _pdu_write_done(void *_pdu, int err) 709 { 710 struct nvme_tcp_pdu *pdu = _pdu; 711 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 712 713 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 714 715 if (err != 0) { 716 spdk_nvmf_tcp_qpair_disconnect(tqpair); 717 return; 718 } 719 720 assert(pdu->cb_fn != NULL); 721 pdu->cb_fn(pdu->cb_arg); 722 } 723 724 static void 725 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 726 struct nvme_tcp_pdu *pdu, 727 nvme_tcp_qpair_xfer_complete_cb cb_fn, 728 void *cb_arg) 729 { 730 int hlen; 731 uint32_t crc32c; 732 uint32_t mapped_length = 0; 733 ssize_t rc; 734 735 assert(&tqpair->pdu_in_progress != pdu); 736 737 hlen = pdu->hdr->common.hlen; 738 739 /* Header Digest */ 740 if (g_nvme_tcp_hdgst[pdu->hdr->common.pdu_type] && tqpair->host_hdgst_enable) { 741 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 742 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr->raw + hlen, crc32c); 743 } 744 745 /* Data Digest */ 746 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr->common.pdu_type] && tqpair->host_ddgst_enable) { 747 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 748 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 749 } 750 751 pdu->cb_fn = cb_fn; 752 pdu->cb_arg = cb_arg; 753 754 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 755 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 756 &mapped_length); 757 pdu->sock_req.cb_fn = _pdu_write_done; 758 pdu->sock_req.cb_arg = pdu; 759 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 760 if (pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 761 pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 762 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 763 if (rc == mapped_length) { 764 _pdu_write_done(pdu, 0); 765 } else { 766 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 767 _pdu_write_done(pdu, -1); 768 } 769 } else { 770 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 771 } 772 } 773 774 static int 775 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 776 { 777 uint32_t i; 778 struct spdk_nvmf_transport_opts *opts; 779 uint32_t in_capsule_data_size; 780 781 opts = &tqpair->qpair.transport->opts; 782 783 in_capsule_data_size = opts->in_capsule_data_size; 784 if (opts->dif_insert_or_strip) { 785 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 786 } 787 788 tqpair->resource_count = opts->max_queue_depth; 789 790 tqpair->mgmt_pdu.qpair = tqpair; 791 tqpair->mgmt_pdu.hdr = &tqpair->mgmt_pdu.hdr_mem; 792 793 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 794 if (!tqpair->reqs) { 795 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 796 return -1; 797 } 798 799 if (in_capsule_data_size) { 800 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 801 NULL, SPDK_ENV_LCORE_ID_ANY, 802 SPDK_MALLOC_DMA); 803 if (!tqpair->bufs) { 804 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 805 return -1; 806 } 807 } 808 809 tqpair->pdus = spdk_dma_malloc(tqpair->resource_count * sizeof(*tqpair->pdus), 0x1000, NULL); 810 if (!tqpair->pdus) { 811 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 812 return -1; 813 } 814 815 for (i = 0; i < tqpair->resource_count; i++) { 816 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 817 818 tcp_req->ttag = i + 1; 819 tcp_req->req.qpair = &tqpair->qpair; 820 821 tcp_req->pdu = &tqpair->pdus[i]; 822 tcp_req->pdu->qpair = tqpair; 823 tcp_req->pdu->hdr = &tcp_req->pdu->hdr_mem; 824 825 /* Set up memory to receive commands */ 826 if (tqpair->bufs) { 827 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 828 } 829 830 /* Set the cmdn and rsp */ 831 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 832 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 833 834 /* Initialize request state to FREE */ 835 tcp_req->state = TCP_REQUEST_STATE_FREE; 836 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 837 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 838 } 839 840 tqpair->pdu_recv_buf.size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 841 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 842 tqpair->pdu_recv_buf.buf = calloc(1, tqpair->pdu_recv_buf.size); 843 if (!tqpair->pdu_recv_buf.buf) { 844 SPDK_ERRLOG("Unable to allocate the pdu recv buf on tqpair=%p with size=%d\n", tqpair, 845 tqpair->pdu_recv_buf.size); 846 return -1; 847 } 848 tqpair->pdu_in_progress.hdr = (union nvme_tcp_pdu_hdr *)tqpair->pdu_recv_buf.buf; 849 850 return 0; 851 } 852 853 static int 854 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 855 { 856 struct spdk_nvmf_tcp_qpair *tqpair; 857 int i; 858 859 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 860 861 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 862 863 TAILQ_INIT(&tqpair->send_queue); 864 865 /* Initialise request state queues of the qpair */ 866 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 867 TAILQ_INIT(&tqpair->state_queue[i]); 868 } 869 870 tqpair->host_hdgst_enable = true; 871 tqpair->host_ddgst_enable = true; 872 return 0; 873 } 874 875 static int 876 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 877 { 878 int rc; 879 880 /* set low water mark */ 881 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr)); 882 if (rc != 0) { 883 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 884 return rc; 885 } 886 887 return 0; 888 } 889 890 static void 891 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 892 struct spdk_nvmf_tcp_port *port, 893 struct spdk_sock *sock, 894 new_qpair_fn cb_fn, void *cb_arg) 895 { 896 struct spdk_nvmf_tcp_qpair *tqpair; 897 int rc; 898 899 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 900 port->trid->traddr, port->trid->trsvcid); 901 902 if (transport->opts.sock_priority) { 903 rc = spdk_sock_set_priority(sock, transport->opts.sock_priority); 904 if (rc) { 905 SPDK_ERRLOG("Failed to set the priority of the socket\n"); 906 spdk_sock_close(&sock); 907 return; 908 } 909 } 910 911 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 912 if (tqpair == NULL) { 913 SPDK_ERRLOG("Could not allocate new connection.\n"); 914 spdk_sock_close(&sock); 915 return; 916 } 917 918 tqpair->sock = sock; 919 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 920 tqpair->port = port; 921 tqpair->qpair.transport = transport; 922 923 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 924 sizeof(tqpair->target_addr), &tqpair->target_port, 925 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 926 &tqpair->initiator_port); 927 if (rc < 0) { 928 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 929 spdk_nvmf_tcp_qpair_destroy(tqpair); 930 return; 931 } 932 933 cb_fn(&tqpair->qpair, cb_arg); 934 } 935 936 static void 937 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port, 938 new_qpair_fn cb_fn, void *cb_arg) 939 { 940 struct spdk_sock *sock; 941 int i; 942 943 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 944 sock = spdk_sock_accept(port->listen_sock); 945 if (sock == NULL) { 946 break; 947 } 948 _spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn, cb_arg); 949 } 950 } 951 952 static void 953 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn, void *cb_arg) 954 { 955 struct spdk_nvmf_tcp_transport *ttransport; 956 struct spdk_nvmf_tcp_port *port; 957 958 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 959 960 TAILQ_FOREACH(port, &ttransport->ports, link) { 961 spdk_nvmf_tcp_port_accept(transport, port, cb_fn, cb_arg); 962 } 963 } 964 965 static void 966 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 967 struct spdk_nvme_transport_id *trid, 968 struct spdk_nvmf_discovery_log_page_entry *entry) 969 { 970 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 971 entry->adrfam = trid->adrfam; 972 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 973 974 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 975 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 976 977 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 978 } 979 980 static struct spdk_nvmf_transport_poll_group * 981 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 982 { 983 struct spdk_nvmf_tcp_poll_group *tgroup; 984 985 tgroup = calloc(1, sizeof(*tgroup)); 986 if (!tgroup) { 987 return NULL; 988 } 989 990 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 991 if (!tgroup->sock_group) { 992 goto cleanup; 993 } 994 995 TAILQ_INIT(&tgroup->qpairs); 996 TAILQ_INIT(&tgroup->await_req); 997 998 return &tgroup->group; 999 1000 cleanup: 1001 free(tgroup); 1002 return NULL; 1003 } 1004 1005 static struct spdk_nvmf_transport_poll_group * 1006 spdk_nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1007 { 1008 struct spdk_nvmf_tcp_qpair *tqpair; 1009 struct spdk_sock_group *group = NULL; 1010 int rc; 1011 1012 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1013 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1014 if (!rc && group != NULL) { 1015 return spdk_sock_group_get_ctx(group); 1016 } 1017 1018 return NULL; 1019 } 1020 1021 static void 1022 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1023 { 1024 struct spdk_nvmf_tcp_poll_group *tgroup; 1025 1026 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1027 spdk_sock_group_close(&tgroup->sock_group); 1028 1029 free(tgroup); 1030 } 1031 1032 static inline void 1033 spdk_nvmf_tcp_reset_pdu_in_process(struct spdk_nvmf_tcp_qpair *tqpair) 1034 { 1035 struct nvme_tcp_pdu_recv_buf *pdu_recv_buf = &tqpair->pdu_recv_buf; 1036 char *dst, *src; 1037 1038 if (spdk_unlikely((pdu_recv_buf->off + sizeof(union nvme_tcp_pdu_hdr)) > 1039 pdu_recv_buf->size)) { 1040 if (pdu_recv_buf->remain_size) { 1041 dst = pdu_recv_buf->buf; 1042 src = (char *)((void *)pdu_recv_buf->buf + pdu_recv_buf->off); 1043 1044 /* purpose: to avoid overlap copy, so do not use memcpy if there is overlap case */ 1045 memmove(dst, src, pdu_recv_buf->remain_size); 1046 } 1047 tqpair->pdu_recv_buf.off = 0; 1048 } else if (!pdu_recv_buf->remain_size) { 1049 tqpair->pdu_recv_buf.off = 0; 1050 } 1051 1052 tqpair->pdu_in_progress.hdr = (union nvme_tcp_pdu_hdr *)((void *)pdu_recv_buf->buf + 1053 pdu_recv_buf->off); 1054 } 1055 1056 static void 1057 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1058 enum nvme_tcp_pdu_recv_state state) 1059 { 1060 if (tqpair->recv_state == state) { 1061 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1062 tqpair, state); 1063 return; 1064 } 1065 1066 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1067 /* When leaving the await req state, move the qpair to the main list */ 1068 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1069 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1070 } 1071 1072 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1073 tqpair->recv_state = state; 1074 1075 switch (state) { 1076 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1077 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1078 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1079 break; 1080 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1081 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1082 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1083 break; 1084 case NVME_TCP_PDU_RECV_STATE_ERROR: 1085 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1086 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1087 spdk_nvmf_tcp_reset_pdu_in_process(tqpair); 1088 break; 1089 default: 1090 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1091 abort(); 1092 break; 1093 } 1094 } 1095 1096 static int 1097 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx) 1098 { 1099 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1100 1101 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1102 1103 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1104 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1105 1106 spdk_nvmf_tcp_qpair_disconnect(tqpair); 1107 return 0; 1108 } 1109 1110 static void 1111 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1112 { 1113 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1114 1115 if (!tqpair->timeout_poller) { 1116 tqpair->timeout_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_handle_timeout, tqpair, 1117 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1118 } 1119 } 1120 1121 static void 1122 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1123 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1124 { 1125 struct nvme_tcp_pdu *rsp_pdu; 1126 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1127 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1128 uint32_t copy_len; 1129 1130 rsp_pdu = &tqpair->mgmt_pdu; 1131 1132 c2h_term_req = &rsp_pdu->hdr->term_req; 1133 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1134 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1135 1136 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1137 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1138 DSET32(&c2h_term_req->fei, error_offset); 1139 } 1140 1141 copy_len = spdk_min(pdu->hdr->common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1142 1143 /* Copy the error info into the buffer */ 1144 memcpy((uint8_t *)rsp_pdu->hdr->raw + c2h_term_req_hdr_len, pdu->hdr->raw, copy_len); 1145 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr->raw + c2h_term_req_hdr_len, copy_len); 1146 1147 /* Contain the header of the wrong received pdu */ 1148 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1149 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1150 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair); 1151 } 1152 1153 static void 1154 spdk_nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1155 struct spdk_nvmf_tcp_qpair *tqpair, 1156 struct nvme_tcp_pdu *pdu) 1157 { 1158 struct spdk_nvmf_tcp_req *tcp_req; 1159 1160 assert(pdu->psh_valid_bytes == pdu->psh_len); 1161 assert(pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1162 1163 tcp_req = spdk_nvmf_tcp_req_get(tqpair); 1164 if (!tcp_req) { 1165 /* Directly return and make the allocation retry again */ 1166 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1167 return; 1168 } 1169 1170 /* The host sent more commands than the maximum queue depth. */ 1171 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1172 spdk_nvmf_tcp_qpair_disconnect(tqpair); 1173 return; 1174 } 1175 1176 pdu->req = tcp_req; 1177 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1178 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1179 } 1180 1181 static void 1182 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1183 struct spdk_nvmf_tcp_qpair *tqpair, 1184 struct nvme_tcp_pdu *pdu) 1185 { 1186 struct spdk_nvmf_tcp_req *tcp_req; 1187 struct spdk_nvme_tcp_cmd *capsule_cmd; 1188 uint32_t error_offset = 0; 1189 enum spdk_nvme_tcp_term_req_fes fes; 1190 1191 capsule_cmd = &pdu->hdr->capsule_cmd; 1192 tcp_req = pdu->req; 1193 assert(tcp_req != NULL); 1194 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1195 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1196 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1197 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1198 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1199 goto err; 1200 } 1201 1202 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1203 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1204 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1205 1206 return; 1207 err: 1208 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1209 } 1210 1211 static int 1212 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1213 enum spdk_nvmf_tcp_req_state state, 1214 uint16_t cid, uint16_t tag, 1215 struct spdk_nvmf_tcp_req **req) 1216 { 1217 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1218 1219 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[state], state_link) { 1220 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1221 continue; 1222 } 1223 1224 if (tcp_req->ttag == tag) { 1225 *req = tcp_req; 1226 return 0; 1227 } 1228 1229 *req = NULL; 1230 return -1; 1231 } 1232 1233 /* Didn't find it, but not an error */ 1234 *req = NULL; 1235 return 0; 1236 } 1237 1238 static void 1239 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1240 struct spdk_nvmf_tcp_qpair *tqpair, 1241 struct nvme_tcp_pdu *pdu) 1242 { 1243 struct spdk_nvmf_tcp_req *tcp_req; 1244 uint32_t error_offset = 0; 1245 enum spdk_nvme_tcp_term_req_fes fes = 0; 1246 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1247 int rc; 1248 1249 h2c_data = &pdu->hdr->h2c_data; 1250 1251 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1252 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1253 1254 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1255 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1256 if (rc == 0 && tcp_req == NULL) { 1257 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1258 h2c_data->ttag, &tcp_req); 1259 } 1260 1261 if (!tcp_req) { 1262 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1263 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1264 if (rc == 0) { 1265 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1266 } else { 1267 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1268 } 1269 goto err; 1270 } 1271 1272 if (tcp_req->h2c_offset != h2c_data->datao) { 1273 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1274 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1275 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1276 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1277 goto err; 1278 } 1279 1280 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1281 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1282 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1283 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1284 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1285 goto err; 1286 } 1287 1288 pdu->req = tcp_req; 1289 1290 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1291 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1292 } 1293 1294 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1295 h2c_data->datao, h2c_data->datal); 1296 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1297 return; 1298 1299 err: 1300 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1301 } 1302 1303 static void 1304 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1305 { 1306 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1307 nvmf_tcp_request_free(tcp_req); 1308 } 1309 1310 static void 1311 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1312 struct spdk_nvmf_tcp_qpair *tqpair) 1313 { 1314 struct nvme_tcp_pdu *rsp_pdu; 1315 struct spdk_nvme_tcp_rsp *capsule_resp; 1316 1317 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1318 1319 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1320 assert(rsp_pdu != NULL); 1321 1322 capsule_resp = &rsp_pdu->hdr->capsule_resp; 1323 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1324 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1325 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1326 if (tqpair->host_hdgst_enable) { 1327 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1328 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1329 } 1330 1331 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req); 1332 } 1333 1334 static void 1335 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1336 { 1337 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1338 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1339 struct spdk_nvmf_tcp_qpair, qpair); 1340 1341 assert(tqpair != NULL); 1342 if (tqpair->qpair.transport->opts.c2h_success) { 1343 nvmf_tcp_request_free(tcp_req); 1344 } else { 1345 nvmf_tcp_req_pdu_fini(tcp_req); 1346 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1347 } 1348 } 1349 1350 static void 1351 spdk_nvmf_tcp_r2t_complete(void *cb_arg) 1352 { 1353 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1354 struct spdk_nvmf_tcp_transport *ttransport; 1355 1356 nvmf_tcp_req_pdu_fini(tcp_req); 1357 1358 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1359 struct spdk_nvmf_tcp_transport, transport); 1360 1361 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1362 1363 if (tcp_req->h2c_offset == tcp_req->req.length) { 1364 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1365 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1366 } 1367 } 1368 1369 static void 1370 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1371 struct spdk_nvmf_tcp_req *tcp_req) 1372 { 1373 struct nvme_tcp_pdu *rsp_pdu; 1374 struct spdk_nvme_tcp_r2t_hdr *r2t; 1375 1376 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1377 assert(rsp_pdu != NULL); 1378 1379 r2t = &rsp_pdu->hdr->r2t; 1380 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1381 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1382 1383 if (tqpair->host_hdgst_enable) { 1384 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1385 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1386 } 1387 1388 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1389 r2t->ttag = tcp_req->ttag; 1390 r2t->r2to = tcp_req->h2c_offset; 1391 r2t->r2tl = tcp_req->req.length; 1392 1393 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1394 1395 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1396 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1397 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1398 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_r2t_complete, tcp_req); 1399 } 1400 1401 static void 1402 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1403 struct spdk_nvmf_tcp_qpair *tqpair, 1404 struct nvme_tcp_pdu *pdu) 1405 { 1406 struct spdk_nvmf_tcp_req *tcp_req; 1407 1408 tcp_req = pdu->req; 1409 assert(tcp_req != NULL); 1410 1411 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1412 1413 tcp_req->h2c_offset += pdu->data_len; 1414 1415 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1416 1417 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1418 * acknowledged before moving on. */ 1419 if (tcp_req->h2c_offset == tcp_req->req.length && 1420 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1421 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1422 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1423 } 1424 } 1425 1426 static void 1427 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1428 { 1429 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1430 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1431 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1432 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1433 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1434 DGET32(h2c_term_req->fei)); 1435 } 1436 } 1437 1438 static void 1439 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1440 struct nvme_tcp_pdu *pdu) 1441 { 1442 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr->term_req; 1443 uint32_t error_offset = 0; 1444 enum spdk_nvme_tcp_term_req_fes fes; 1445 1446 1447 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1448 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1449 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1450 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1451 goto end; 1452 } 1453 1454 /* set the data buffer */ 1455 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr->raw + h2c_term_req->common.hlen, 1456 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1457 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1458 return; 1459 end: 1460 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1461 } 1462 1463 static void 1464 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1465 struct nvme_tcp_pdu *pdu) 1466 { 1467 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr->term_req; 1468 1469 spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1470 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1471 } 1472 1473 static void 1474 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1475 struct spdk_nvmf_tcp_transport *ttransport) 1476 { 1477 int rc = 0; 1478 struct nvme_tcp_pdu *pdu; 1479 uint32_t crc32c, error_offset = 0; 1480 enum spdk_nvme_tcp_term_req_fes fes; 1481 1482 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1483 pdu = &tqpair->pdu_in_progress; 1484 1485 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1486 /* check data digest if need */ 1487 if (pdu->ddgst_enable) { 1488 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1489 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1490 if (rc == 0) { 1491 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1492 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1493 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1494 return; 1495 1496 } 1497 } 1498 1499 switch (pdu->hdr->common.pdu_type) { 1500 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1501 spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1502 break; 1503 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1504 spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1505 break; 1506 1507 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1508 spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1509 break; 1510 1511 default: 1512 /* The code should not go to here */ 1513 SPDK_ERRLOG("The code should not go to here\n"); 1514 break; 1515 } 1516 } 1517 1518 static void 1519 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg) 1520 { 1521 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1522 1523 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1524 } 1525 1526 static void 1527 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1528 struct spdk_nvmf_tcp_qpair *tqpair, 1529 struct nvme_tcp_pdu *pdu) 1530 { 1531 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr->ic_req; 1532 struct nvme_tcp_pdu *rsp_pdu; 1533 struct spdk_nvme_tcp_ic_resp *ic_resp; 1534 uint32_t error_offset = 0; 1535 enum spdk_nvme_tcp_term_req_fes fes; 1536 1537 /* Only PFV 0 is defined currently */ 1538 if (ic_req->pfv != 0) { 1539 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1540 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1541 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1542 goto end; 1543 } 1544 1545 /* MAXR2T is 0's based */ 1546 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1547 1548 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1549 if (!tqpair->host_hdgst_enable) { 1550 tqpair->pdu_recv_buf.size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1551 } 1552 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1553 if (!tqpair->host_ddgst_enable) { 1554 tqpair->pdu_recv_buf.size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1555 } 1556 1557 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1558 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1559 1560 rsp_pdu = &tqpair->mgmt_pdu; 1561 1562 ic_resp = &rsp_pdu->hdr->ic_resp; 1563 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1564 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1565 ic_resp->pfv = 0; 1566 ic_resp->cpda = tqpair->cpda; 1567 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1568 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1569 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1570 1571 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1572 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1573 1574 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1575 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair); 1576 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1577 return; 1578 end: 1579 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1580 } 1581 1582 static void 1583 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1584 struct spdk_nvmf_tcp_transport *ttransport) 1585 { 1586 struct nvme_tcp_pdu *pdu; 1587 int rc; 1588 uint32_t crc32c, error_offset = 0; 1589 enum spdk_nvme_tcp_term_req_fes fes; 1590 1591 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1592 pdu = &tqpair->pdu_in_progress; 1593 1594 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1595 pdu->hdr->common.pdu_type); 1596 /* check header digest if needed */ 1597 if (pdu->has_hdgst) { 1598 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1599 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1600 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr->raw + pdu->hdr->common.hlen, crc32c); 1601 if (rc == 0) { 1602 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1603 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1604 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1605 return; 1606 1607 } 1608 } 1609 1610 switch (pdu->hdr->common.pdu_type) { 1611 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1612 spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1613 break; 1614 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1615 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1616 break; 1617 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1618 spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1619 break; 1620 1621 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1622 spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1623 break; 1624 1625 default: 1626 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr->common.pdu_type); 1627 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1628 error_offset = 1; 1629 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1630 break; 1631 } 1632 } 1633 1634 static void 1635 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1636 { 1637 struct nvme_tcp_pdu *pdu; 1638 uint32_t error_offset = 0; 1639 enum spdk_nvme_tcp_term_req_fes fes; 1640 uint8_t expected_hlen, pdo; 1641 bool plen_error = false, pdo_error = false; 1642 1643 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1644 pdu = &tqpair->pdu_in_progress; 1645 1646 if (pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1647 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1648 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1649 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1650 goto err; 1651 } 1652 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1653 if (pdu->hdr->common.plen != expected_hlen) { 1654 plen_error = true; 1655 } 1656 } else { 1657 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1658 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1659 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1660 goto err; 1661 } 1662 1663 switch (pdu->hdr->common.pdu_type) { 1664 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1665 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1666 pdo = pdu->hdr->common.pdo; 1667 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1668 pdo_error = true; 1669 break; 1670 } 1671 1672 if (pdu->hdr->common.plen < expected_hlen) { 1673 plen_error = true; 1674 } 1675 break; 1676 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1677 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1678 pdo = pdu->hdr->common.pdo; 1679 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1680 pdo_error = true; 1681 break; 1682 } 1683 if (pdu->hdr->common.plen < expected_hlen) { 1684 plen_error = true; 1685 } 1686 break; 1687 1688 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1689 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1690 if ((pdu->hdr->common.plen <= expected_hlen) || 1691 (pdu->hdr->common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1692 plen_error = true; 1693 } 1694 break; 1695 1696 default: 1697 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr->common.pdu_type); 1698 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1699 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1700 goto err; 1701 } 1702 } 1703 1704 if (pdu->hdr->common.hlen != expected_hlen) { 1705 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1706 pdu->hdr->common.pdu_type, 1707 expected_hlen, pdu->hdr->common.hlen, tqpair); 1708 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1709 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1710 goto err; 1711 } else if (pdo_error) { 1712 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1713 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1714 } else if (plen_error) { 1715 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1716 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1717 goto err; 1718 } else { 1719 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1720 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1721 return; 1722 } 1723 err: 1724 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1725 } 1726 1727 static int 1728 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1729 int read_len) 1730 { 1731 int rc; 1732 1733 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1734 read_offset, read_len, pdu->dif_ctx); 1735 if (rc != 0) { 1736 SPDK_ERRLOG("DIF generate failed\n"); 1737 } 1738 1739 return rc; 1740 } 1741 1742 static int 1743 nvme_tcp_recv_buf_read(struct spdk_sock *sock, struct nvme_tcp_pdu_recv_buf *pdu_recv_buf) 1744 { 1745 int rc; 1746 1747 rc = nvme_tcp_read_data(sock, pdu_recv_buf->size - pdu_recv_buf->off, 1748 (void *)pdu_recv_buf->buf + pdu_recv_buf->off); 1749 if (rc < 0) { 1750 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect sock=%p\n", sock); 1751 } else if (rc > 0) { 1752 pdu_recv_buf->remain_size = rc; 1753 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1754 } 1755 1756 return rc; 1757 } 1758 1759 static uint32_t 1760 nvme_tcp_read_data_from_pdu_recv_buf(struct nvme_tcp_pdu_recv_buf *pdu_recv_buf, 1761 uint32_t expected_size, 1762 char *dst) 1763 { 1764 uint32_t size; 1765 1766 assert(pdu_recv_buf->remain_size > 0); 1767 size = spdk_min(expected_size, pdu_recv_buf->remain_size); 1768 if (dst) { 1769 memcpy(dst, (void *)pdu_recv_buf->buf + pdu_recv_buf->off, size); 1770 } 1771 pdu_recv_buf->off += size; 1772 pdu_recv_buf->remain_size -= size; 1773 1774 1775 return size; 1776 } 1777 1778 static int 1779 nvme_tcp_read_payload_data_from_pdu_recv_buf(struct nvme_tcp_pdu_recv_buf *pdu_recv_buf, 1780 struct nvme_tcp_pdu *pdu) 1781 { 1782 struct iovec iov[NVME_TCP_MAX_SGL_DESCRIPTORS + 1]; 1783 int iovcnt, i; 1784 uint32_t size = 0; 1785 void *dst; 1786 1787 assert(pdu_recv_buf->remain_size > 0); 1788 iovcnt = nvme_tcp_build_payload_iovs(iov, NVME_TCP_MAX_SGL_DESCRIPTORS + 1, pdu, 1789 pdu->ddgst_enable, NULL); 1790 assert(iovcnt >= 0); 1791 for (i = 0; i < iovcnt; i++) { 1792 if (!pdu_recv_buf->remain_size) { 1793 break; 1794 } 1795 1796 dst = NULL; 1797 if (pdu->hdr->common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) { 1798 dst = iov[i].iov_base; 1799 } 1800 size += nvme_tcp_read_data_from_pdu_recv_buf(pdu_recv_buf, iov[i].iov_len, dst); 1801 } 1802 1803 return size; 1804 } 1805 1806 static int 1807 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1808 { 1809 int rc = 0; 1810 struct nvme_tcp_pdu *pdu; 1811 enum nvme_tcp_pdu_recv_state prev_state; 1812 uint32_t data_len; 1813 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1814 struct spdk_nvmf_tcp_transport, transport); 1815 1816 /* The loop here is to allow for several back-to-back state changes. */ 1817 do { 1818 prev_state = tqpair->recv_state; 1819 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1820 1821 pdu = &tqpair->pdu_in_progress; 1822 switch (tqpair->recv_state) { 1823 /* Wait for the common header */ 1824 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1825 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1826 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1827 return rc; 1828 } 1829 1830 if (!tqpair->pdu_recv_buf.remain_size) { 1831 rc = nvme_tcp_recv_buf_read(tqpair->sock, &tqpair->pdu_recv_buf); 1832 if (rc <= 0) { 1833 return rc; 1834 } 1835 } 1836 rc = nvme_tcp_read_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, 1837 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1838 NULL); 1839 pdu->ch_valid_bytes += rc; 1840 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1841 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1842 } 1843 1844 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1845 return NVME_TCP_PDU_IN_PROGRESS; 1846 } 1847 1848 /* The command header of this PDU has now been read from the socket. */ 1849 spdk_nvmf_tcp_pdu_ch_handle(tqpair); 1850 break; 1851 /* Wait for the pdu specific header */ 1852 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1853 if (!tqpair->pdu_recv_buf.remain_size) { 1854 rc = nvme_tcp_recv_buf_read(tqpair->sock, &tqpair->pdu_recv_buf); 1855 if (rc <= 0) { 1856 return rc; 1857 } 1858 } 1859 1860 rc = nvme_tcp_read_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, 1861 pdu->psh_len - pdu->psh_valid_bytes, 1862 NULL); 1863 pdu->psh_valid_bytes += rc; 1864 if (pdu->psh_valid_bytes < pdu->psh_len) { 1865 return NVME_TCP_PDU_IN_PROGRESS; 1866 } 1867 1868 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1869 spdk_nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1870 break; 1871 /* Wait for the req slot */ 1872 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1873 spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1874 break; 1875 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1876 /* check whether the data is valid, if not we just return */ 1877 if (!pdu->data_len) { 1878 return NVME_TCP_PDU_IN_PROGRESS; 1879 } 1880 1881 data_len = pdu->data_len; 1882 /* data digest */ 1883 if (spdk_unlikely((pdu->hdr->common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1884 tqpair->host_ddgst_enable)) { 1885 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1886 pdu->ddgst_enable = true; 1887 } 1888 1889 if (tqpair->pdu_recv_buf.remain_size) { 1890 rc = nvme_tcp_read_payload_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, pdu); 1891 pdu->readv_offset += rc; 1892 } 1893 1894 if (pdu->readv_offset < data_len) { 1895 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1896 if (rc < 0) { 1897 return NVME_TCP_PDU_IN_PROGRESS; 1898 } 1899 pdu->readv_offset += rc; 1900 } 1901 1902 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1903 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1904 if (rc != 0) { 1905 return NVME_TCP_PDU_FATAL; 1906 } 1907 } 1908 1909 if (pdu->readv_offset < data_len) { 1910 return NVME_TCP_PDU_IN_PROGRESS; 1911 } 1912 1913 /* All of this PDU has now been read from the socket. */ 1914 spdk_nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1915 break; 1916 case NVME_TCP_PDU_RECV_STATE_ERROR: 1917 if (!spdk_sock_is_connected(tqpair->sock)) { 1918 return NVME_TCP_PDU_FATAL; 1919 } 1920 break; 1921 default: 1922 assert(0); 1923 SPDK_ERRLOG("code should not come to here"); 1924 break; 1925 } 1926 } while (tqpair->recv_state != prev_state); 1927 1928 return rc; 1929 } 1930 1931 static int 1932 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 1933 struct spdk_nvmf_transport *transport, 1934 struct spdk_nvmf_transport_poll_group *group) 1935 { 1936 struct spdk_nvmf_request *req = &tcp_req->req; 1937 struct spdk_nvme_cmd *cmd; 1938 struct spdk_nvme_cpl *rsp; 1939 struct spdk_nvme_sgl_descriptor *sgl; 1940 uint32_t length; 1941 1942 cmd = &req->cmd->nvme_cmd; 1943 rsp = &req->rsp->nvme_cpl; 1944 sgl = &cmd->dptr.sgl1; 1945 1946 length = sgl->unkeyed.length; 1947 1948 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 1949 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 1950 if (length > transport->opts.max_io_size) { 1951 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 1952 length, transport->opts.max_io_size); 1953 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1954 return -1; 1955 } 1956 1957 /* fill request length and populate iovs */ 1958 req->length = length; 1959 1960 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", length); 1961 1962 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 1963 req->dif.orig_length = length; 1964 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 1965 req->dif.elba_length = length; 1966 } 1967 1968 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 1969 /* No available buffers. Queue this request up. */ 1970 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", 1971 tcp_req); 1972 return 0; 1973 } 1974 1975 /* backward compatible */ 1976 req->data = req->iov[0].iov_base; 1977 1978 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 1979 tcp_req, req->iovcnt, req->data); 1980 1981 return 0; 1982 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 1983 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 1984 uint64_t offset = sgl->address; 1985 uint32_t max_len = transport->opts.in_capsule_data_size; 1986 1987 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 1988 offset, length); 1989 1990 if (offset > max_len) { 1991 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 1992 offset, max_len); 1993 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 1994 return -1; 1995 } 1996 max_len -= (uint32_t)offset; 1997 1998 if (length > max_len) { 1999 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2000 length, max_len); 2001 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2002 return -1; 2003 } 2004 2005 req->data = tcp_req->buf + offset; 2006 req->data_from_pool = false; 2007 req->length = length; 2008 2009 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2010 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2011 req->dif.elba_length = length; 2012 } 2013 2014 req->iov[0].iov_base = req->data; 2015 req->iov[0].iov_len = length; 2016 req->iovcnt = 1; 2017 2018 return 0; 2019 } 2020 2021 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2022 sgl->generic.type, sgl->generic.subtype); 2023 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2024 return -1; 2025 } 2026 2027 static inline enum spdk_nvme_media_error_status_code 2028 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2029 enum spdk_nvme_media_error_status_code result; 2030 2031 switch (err_type) 2032 { 2033 case SPDK_DIF_REFTAG_ERROR: 2034 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2035 break; 2036 case SPDK_DIF_APPTAG_ERROR: 2037 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2038 break; 2039 case SPDK_DIF_GUARD_ERROR: 2040 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2041 break; 2042 default: 2043 SPDK_UNREACHABLE(); 2044 break; 2045 } 2046 2047 return result; 2048 } 2049 2050 static void 2051 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2052 struct spdk_nvmf_tcp_req *tcp_req) 2053 { 2054 struct nvme_tcp_pdu *rsp_pdu; 2055 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2056 uint32_t plen, pdo, alignment; 2057 int rc; 2058 2059 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2060 2061 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2062 assert(rsp_pdu != NULL); 2063 2064 c2h_data = &rsp_pdu->hdr->c2h_data; 2065 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2066 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2067 2068 if (tqpair->host_hdgst_enable) { 2069 plen += SPDK_NVME_TCP_DIGEST_LEN; 2070 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2071 } 2072 2073 /* set the psh */ 2074 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2075 c2h_data->datal = tcp_req->req.length; 2076 c2h_data->datao = 0; 2077 2078 /* set the padding */ 2079 rsp_pdu->padding_len = 0; 2080 pdo = plen; 2081 if (tqpair->cpda) { 2082 alignment = (tqpair->cpda + 1) << 2; 2083 if (alignment > plen) { 2084 rsp_pdu->padding_len = alignment - plen; 2085 pdo = plen = alignment; 2086 } 2087 } 2088 2089 c2h_data->common.pdo = pdo; 2090 plen += c2h_data->datal; 2091 if (tqpair->host_ddgst_enable) { 2092 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2093 plen += SPDK_NVME_TCP_DIGEST_LEN; 2094 } 2095 2096 c2h_data->common.plen = plen; 2097 2098 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2099 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2100 } 2101 2102 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2103 c2h_data->datao, c2h_data->datal); 2104 2105 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2106 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2107 struct spdk_dif_error err_blk = {}; 2108 2109 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2110 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2111 if (rc != 0) { 2112 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2113 err_blk.err_type, err_blk.err_offset); 2114 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2115 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2116 nvmf_tcp_req_pdu_fini(tcp_req); 2117 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2118 return; 2119 } 2120 } 2121 2122 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2123 if (tqpair->qpair.transport->opts.c2h_success) { 2124 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2125 } 2126 2127 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2128 } 2129 2130 static int 2131 request_transfer_out(struct spdk_nvmf_request *req) 2132 { 2133 struct spdk_nvmf_tcp_req *tcp_req; 2134 struct spdk_nvmf_qpair *qpair; 2135 struct spdk_nvmf_tcp_qpair *tqpair; 2136 struct spdk_nvme_cpl *rsp; 2137 2138 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2139 2140 qpair = req->qpair; 2141 rsp = &req->rsp->nvme_cpl; 2142 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2143 2144 /* Advance our sq_head pointer */ 2145 if (qpair->sq_head == qpair->sq_head_max) { 2146 qpair->sq_head = 0; 2147 } else { 2148 qpair->sq_head++; 2149 } 2150 rsp->sqhd = qpair->sq_head; 2151 2152 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2153 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2154 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2155 spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2156 } else { 2157 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2158 } 2159 2160 return 0; 2161 } 2162 2163 static void 2164 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2165 struct spdk_nvmf_tcp_req *tcp_req) 2166 { 2167 struct nvme_tcp_pdu *pdu; 2168 uint32_t plen = 0; 2169 2170 pdu = &tqpair->pdu_in_progress; 2171 plen = pdu->hdr->common.hlen; 2172 2173 if (tqpair->host_hdgst_enable) { 2174 plen += SPDK_NVME_TCP_DIGEST_LEN; 2175 } 2176 2177 if (pdu->hdr->common.plen != plen) { 2178 tcp_req->has_incapsule_data = true; 2179 } 2180 } 2181 2182 static bool 2183 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2184 struct spdk_nvmf_tcp_req *tcp_req) 2185 { 2186 struct spdk_nvmf_tcp_qpair *tqpair; 2187 int rc; 2188 enum spdk_nvmf_tcp_req_state prev_state; 2189 bool progress = false; 2190 struct spdk_nvmf_transport *transport = &ttransport->transport; 2191 struct spdk_nvmf_transport_poll_group *group; 2192 2193 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2194 group = &tqpair->group->group; 2195 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2196 2197 /* The loop here is to allow for several back-to-back state changes. */ 2198 do { 2199 prev_state = tcp_req->state; 2200 2201 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2202 tqpair); 2203 2204 switch (tcp_req->state) { 2205 case TCP_REQUEST_STATE_FREE: 2206 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2207 * to escape this state. */ 2208 break; 2209 case TCP_REQUEST_STATE_NEW: 2210 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2211 2212 /* copy the cmd from the receive pdu */ 2213 tcp_req->cmd = tqpair->pdu_in_progress.hdr->capsule_cmd.ccsqe; 2214 2215 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2216 tcp_req->req.dif.dif_insert_or_strip = true; 2217 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2218 } 2219 2220 /* The next state transition depends on the data transfer needs of this request. */ 2221 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2222 2223 /* If no data to transfer, ready to execute. */ 2224 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2225 /* Reset the tqpair receving pdu state */ 2226 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2227 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2228 break; 2229 } 2230 2231 spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2232 2233 if (!tcp_req->has_incapsule_data) { 2234 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2235 } 2236 2237 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2238 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2239 break; 2240 case TCP_REQUEST_STATE_NEED_BUFFER: 2241 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2242 2243 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2244 2245 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2246 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2247 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2248 tcp_req, tqpair); 2249 /* This request needs to wait in line to obtain a buffer */ 2250 break; 2251 } 2252 2253 /* Try to get a data buffer */ 2254 rc = spdk_nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2255 if (rc < 0) { 2256 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2257 /* Reset the tqpair receving pdu state */ 2258 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2259 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2260 break; 2261 } 2262 2263 if (!tcp_req->req.data) { 2264 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2265 tcp_req, tqpair); 2266 /* No buffers available. */ 2267 break; 2268 } 2269 2270 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2271 2272 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2273 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2274 if (tcp_req->req.data_from_pool) { 2275 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2276 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2277 } else { 2278 struct nvme_tcp_pdu *pdu; 2279 2280 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2281 2282 pdu = &tqpair->pdu_in_progress; 2283 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2284 tqpair); 2285 /* No need to send r2t, contained in the capsuled data */ 2286 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2287 0, tcp_req->req.length); 2288 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2289 } 2290 break; 2291 } 2292 2293 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2294 break; 2295 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2296 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2297 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2298 break; 2299 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2300 2301 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2302 (uintptr_t)tcp_req, 0); 2303 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2304 * to escape this state. */ 2305 break; 2306 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2307 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2308 2309 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2310 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2311 tcp_req->req.length = tcp_req->req.dif.elba_length; 2312 } 2313 2314 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2315 spdk_nvmf_request_exec(&tcp_req->req); 2316 break; 2317 case TCP_REQUEST_STATE_EXECUTING: 2318 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2319 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2320 * to escape this state. */ 2321 break; 2322 case TCP_REQUEST_STATE_EXECUTED: 2323 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2324 2325 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2326 tcp_req->req.length = tcp_req->req.dif.orig_length; 2327 } 2328 2329 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2330 break; 2331 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2332 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2333 rc = request_transfer_out(&tcp_req->req); 2334 assert(rc == 0); /* No good way to handle this currently */ 2335 break; 2336 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2337 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2338 (uintptr_t)tcp_req, 2339 0); 2340 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2341 * to escape this state. */ 2342 break; 2343 case TCP_REQUEST_STATE_COMPLETED: 2344 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2345 if (tcp_req->req.data_from_pool) { 2346 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2347 } 2348 tcp_req->req.length = 0; 2349 tcp_req->req.iovcnt = 0; 2350 tcp_req->req.data = NULL; 2351 2352 nvmf_tcp_req_pdu_fini(tcp_req); 2353 2354 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2355 break; 2356 case TCP_REQUEST_NUM_STATES: 2357 default: 2358 assert(0); 2359 break; 2360 } 2361 2362 if (tcp_req->state != prev_state) { 2363 progress = true; 2364 } 2365 } while (tcp_req->state != prev_state); 2366 2367 return progress; 2368 } 2369 2370 static void 2371 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2372 { 2373 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2374 int rc; 2375 2376 assert(tqpair != NULL); 2377 rc = spdk_nvmf_tcp_sock_process(tqpair); 2378 2379 /* If there was a new socket error, disconnect */ 2380 if (rc < 0) { 2381 spdk_nvmf_tcp_qpair_disconnect(tqpair); 2382 } 2383 } 2384 2385 static int 2386 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2387 struct spdk_nvmf_qpair *qpair) 2388 { 2389 struct spdk_nvmf_tcp_poll_group *tgroup; 2390 struct spdk_nvmf_tcp_qpair *tqpair; 2391 int rc; 2392 2393 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2394 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2395 2396 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2397 spdk_nvmf_tcp_sock_cb, tqpair); 2398 if (rc != 0) { 2399 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2400 spdk_strerror(errno), errno); 2401 return -1; 2402 } 2403 2404 rc = spdk_nvmf_tcp_qpair_sock_init(tqpair); 2405 if (rc != 0) { 2406 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2407 return -1; 2408 } 2409 2410 rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair); 2411 if (rc < 0) { 2412 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2413 return -1; 2414 } 2415 2416 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair); 2417 if (rc < 0) { 2418 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2419 return -1; 2420 } 2421 2422 tqpair->group = tgroup; 2423 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2424 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2425 2426 return 0; 2427 } 2428 2429 static int 2430 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2431 struct spdk_nvmf_qpair *qpair) 2432 { 2433 struct spdk_nvmf_tcp_poll_group *tgroup; 2434 struct spdk_nvmf_tcp_qpair *tqpair; 2435 int rc; 2436 2437 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2438 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2439 2440 assert(tqpair->group == tgroup); 2441 2442 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2443 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2444 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2445 } else { 2446 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2447 } 2448 2449 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2450 if (rc != 0) { 2451 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2452 spdk_strerror(errno), errno); 2453 } 2454 2455 return rc; 2456 } 2457 2458 static int 2459 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2460 { 2461 struct spdk_nvmf_tcp_transport *ttransport; 2462 struct spdk_nvmf_tcp_req *tcp_req; 2463 2464 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2465 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2466 2467 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2468 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 2469 2470 return 0; 2471 } 2472 2473 static void 2474 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2475 { 2476 struct spdk_nvmf_tcp_qpair *tqpair; 2477 2478 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Qpair: %p\n", qpair); 2479 2480 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2481 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2482 spdk_nvmf_tcp_qpair_destroy(tqpair); 2483 } 2484 2485 static int 2486 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2487 { 2488 struct spdk_nvmf_tcp_poll_group *tgroup; 2489 int rc; 2490 struct spdk_nvmf_request *req, *req_tmp; 2491 struct spdk_nvmf_tcp_req *tcp_req; 2492 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2493 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2494 struct spdk_nvmf_tcp_transport, transport); 2495 2496 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2497 2498 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2499 return 0; 2500 } 2501 2502 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2503 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2504 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2505 break; 2506 } 2507 } 2508 2509 rc = spdk_sock_group_poll(tgroup->sock_group); 2510 if (rc < 0) { 2511 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2512 } 2513 2514 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2515 spdk_nvmf_tcp_sock_process(tqpair); 2516 } 2517 2518 return rc; 2519 } 2520 2521 static int 2522 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2523 struct spdk_nvme_transport_id *trid, bool peer) 2524 { 2525 struct spdk_nvmf_tcp_qpair *tqpair; 2526 uint16_t port; 2527 2528 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2529 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2530 2531 if (peer) { 2532 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2533 port = tqpair->initiator_port; 2534 } else { 2535 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2536 port = tqpair->target_port; 2537 } 2538 2539 if (spdk_sock_is_ipv4(tqpair->sock)) { 2540 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2541 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2542 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2543 } else { 2544 return -1; 2545 } 2546 2547 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2548 return 0; 2549 } 2550 2551 static int 2552 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2553 struct spdk_nvme_transport_id *trid) 2554 { 2555 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2556 } 2557 2558 static int 2559 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2560 struct spdk_nvme_transport_id *trid) 2561 { 2562 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2563 } 2564 2565 static int 2566 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2567 struct spdk_nvme_transport_id *trid) 2568 { 2569 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2570 } 2571 2572 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2573 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2574 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2575 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2576 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2577 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2578 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2579 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2580 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 2581 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2582 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 2583 2584 static void 2585 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2586 { 2587 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2588 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2589 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2590 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2591 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2592 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2593 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2594 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2595 opts->c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 2596 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2597 opts->sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 2598 } 2599 2600 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2601 .name = "TCP", 2602 .type = SPDK_NVME_TRANSPORT_TCP, 2603 .opts_init = spdk_nvmf_tcp_opts_init, 2604 .create = spdk_nvmf_tcp_create, 2605 .destroy = spdk_nvmf_tcp_destroy, 2606 2607 .listen = spdk_nvmf_tcp_listen, 2608 .stop_listen = spdk_nvmf_tcp_stop_listen, 2609 .accept = spdk_nvmf_tcp_accept, 2610 2611 .listener_discover = spdk_nvmf_tcp_discover, 2612 2613 .poll_group_create = spdk_nvmf_tcp_poll_group_create, 2614 .get_optimal_poll_group = spdk_nvmf_tcp_get_optimal_poll_group, 2615 .poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy, 2616 .poll_group_add = spdk_nvmf_tcp_poll_group_add, 2617 .poll_group_remove = spdk_nvmf_tcp_poll_group_remove, 2618 .poll_group_poll = spdk_nvmf_tcp_poll_group_poll, 2619 2620 .req_free = spdk_nvmf_tcp_req_free, 2621 .req_complete = spdk_nvmf_tcp_req_complete, 2622 2623 .qpair_fini = spdk_nvmf_tcp_close_qpair, 2624 .qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid, 2625 .qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid, 2626 .qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid, 2627 }; 2628 2629 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2630 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2631