1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019 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.h" 40 #include "spdk/nvmf_spec.h" 41 #include "spdk/sock.h" 42 #include "spdk/string.h" 43 #include "spdk/trace.h" 44 #include "spdk/util.h" 45 46 #include "nvmf_internal.h" 47 #include "transport.h" 48 49 #include "spdk_internal/assert.h" 50 #include "spdk_internal/log.h" 51 #include "spdk_internal/nvme_tcp.h" 52 53 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 54 55 #define NVMF_TCP_PDU_MAX_H2C_DATA_SIZE 131072 56 #define NVMF_TCP_PDU_MAX_C2H_DATA_SIZE 131072 57 #define NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM 64 /* Maximal c2h_data pdu number for ecah tqpair */ 58 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 6 59 #define SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR 4 60 61 /* spdk nvmf related structure */ 62 enum spdk_nvmf_tcp_req_state { 63 64 /* The request is not currently in use */ 65 TCP_REQUEST_STATE_FREE = 0, 66 67 /* Initial state when request first received */ 68 TCP_REQUEST_STATE_NEW, 69 70 /* The request is queued until a data buffer is available. */ 71 TCP_REQUEST_STATE_NEED_BUFFER, 72 73 /* The request is currently transferring data from the host to the controller. */ 74 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 75 76 /* The request is ready to execute at the block device */ 77 TCP_REQUEST_STATE_READY_TO_EXECUTE, 78 79 /* The request is currently executing at the block device */ 80 TCP_REQUEST_STATE_EXECUTING, 81 82 /* The request finished executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTED, 84 85 /* The request is ready to send a completion */ 86 TCP_REQUEST_STATE_READY_TO_COMPLETE, 87 88 /* The request is currently transferring final pdus from the controller to the host. */ 89 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 90 91 /* The request completed and can be marked free. */ 92 TCP_REQUEST_STATE_COMPLETED, 93 94 /* Terminator */ 95 TCP_REQUEST_NUM_STATES, 96 }; 97 98 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 99 "Invalid PDU Header Field", 100 "PDU Sequence Error", 101 "Header Digiest Error", 102 "Data Transfer Out of Range", 103 "R2T Limit Exceeded", 104 "Unsupported parameter", 105 }; 106 107 #define OBJECT_NVMF_TCP_IO 0x80 108 109 #define TRACE_GROUP_NVMF_TCP 0x5 110 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 111 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 112 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 113 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 114 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 115 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 116 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 117 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 118 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 119 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 120 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 121 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 122 123 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 124 { 125 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 126 spdk_trace_register_description("TCP_REQ_NEW", 127 TRACE_TCP_REQUEST_STATE_NEW, 128 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 129 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 130 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 132 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 133 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 136 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_EXECUTING", 139 TRACE_TCP_REQUEST_STATE_EXECUTING, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_EXECUTED", 142 TRACE_TCP_REQUEST_STATE_EXECUTED, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 145 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 148 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_COMPLETED", 151 TRACE_TCP_REQUEST_STATE_COMPLETED, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_WRITE_START", 154 TRACE_TCP_FLUSH_WRITEBUF_START, 155 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 156 spdk_trace_register_description("TCP_WRITE_DONE", 157 TRACE_TCP_FLUSH_WRITEBUF_DONE, 158 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 159 spdk_trace_register_description("TCP_READ_DONE", 160 TRACE_TCP_READ_FROM_SOCKET_DONE, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 } 163 164 struct spdk_nvmf_tcp_req { 165 struct spdk_nvmf_request req; 166 struct spdk_nvme_cpl rsp; 167 struct spdk_nvme_cmd cmd; 168 169 /* In-capsule data buffer */ 170 uint8_t *buf; 171 172 bool has_incapsule_data; 173 174 /* transfer_tag */ 175 uint16_t ttag; 176 177 enum spdk_nvmf_tcp_req_state state; 178 179 /* 180 * next_expected_r2t_offset is used when we receive the h2c_data PDU. 181 */ 182 uint32_t next_expected_r2t_offset; 183 uint32_t r2tl_remain; 184 185 /* 186 * c2h_data_offset is used when we send the c2h_data PDU. 187 */ 188 uint32_t c2h_data_offset; 189 uint32_t c2h_data_pdu_num; 190 191 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 192 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 193 }; 194 195 struct nvme_tcp_pdu_recv_buf { 196 char *buf; 197 uint32_t off; 198 uint32_t size; 199 uint32_t remain_size; 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 struct spdk_poller *flush_poller; 208 209 enum nvme_tcp_pdu_recv_state recv_state; 210 enum nvme_tcp_qpair_state state; 211 212 struct nvme_tcp_pdu pdu_in_progress; 213 struct nvme_tcp_pdu_recv_buf pdu_recv_buf; 214 215 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 216 TAILQ_HEAD(, nvme_tcp_pdu) free_queue; 217 218 struct nvme_tcp_pdu *pdu; 219 struct nvme_tcp_pdu *pdu_pool; 220 uint16_t free_pdu_num; 221 222 /* Queues to track the requests in all states */ 223 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 224 /* Number of requests in each state */ 225 int32_t state_cntr[TCP_REQUEST_NUM_STATES]; 226 227 STAILQ_HEAD(, spdk_nvmf_tcp_req) queued_c2h_data_tcp_req; 228 229 uint8_t cpda; 230 231 /* Array of size "max_queue_depth * InCapsuleDataSize" containing 232 * buffers to be used for in capsule data. 233 */ 234 void *buf; 235 void *bufs; 236 struct spdk_nvmf_tcp_req *req; 237 struct spdk_nvmf_tcp_req *reqs; 238 239 bool host_hdgst_enable; 240 bool host_ddgst_enable; 241 242 243 /* The maximum number of I/O outstanding on this connection at one time */ 244 uint16_t max_queue_depth; 245 246 247 /** Specifies the maximum number of PDU-Data bytes per H2C Data Transfer PDU */ 248 uint32_t maxh2cdata; 249 250 uint32_t c2h_data_pdu_cnt; 251 252 /* IP address */ 253 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 254 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 255 256 /* IP port */ 257 uint16_t initiator_port; 258 uint16_t target_port; 259 260 /* Timer used to destroy qpair after detecting transport error issue if initiator does 261 * not close the connection. 262 */ 263 struct spdk_poller *timeout_poller; 264 265 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 266 }; 267 268 struct spdk_nvmf_tcp_poll_group { 269 struct spdk_nvmf_transport_poll_group group; 270 struct spdk_sock_group *sock_group; 271 272 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 273 }; 274 275 struct spdk_nvmf_tcp_port { 276 struct spdk_nvme_transport_id trid; 277 struct spdk_sock *listen_sock; 278 uint32_t ref; 279 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 280 }; 281 282 struct spdk_nvmf_tcp_transport { 283 struct spdk_nvmf_transport transport; 284 285 pthread_mutex_t lock; 286 287 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 288 }; 289 290 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 291 struct spdk_nvmf_tcp_req *tcp_req); 292 static void spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair); 293 294 static void 295 spdk_nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 296 enum spdk_nvmf_tcp_req_state state) 297 { 298 struct spdk_nvmf_qpair *qpair; 299 struct spdk_nvmf_tcp_qpair *tqpair; 300 301 qpair = tcp_req->req.qpair; 302 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 303 304 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 305 tqpair->state_cntr[tcp_req->state]--; 306 assert(tqpair->state_cntr[tcp_req->state] >= 0); 307 308 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 309 tqpair->state_cntr[state]++; 310 311 tcp_req->state = state; 312 } 313 314 static struct nvme_tcp_pdu * 315 spdk_nvmf_tcp_pdu_get(struct spdk_nvmf_tcp_qpair *tqpair) 316 { 317 struct nvme_tcp_pdu *pdu; 318 319 pdu = TAILQ_FIRST(&tqpair->free_queue); 320 if (!pdu) { 321 SPDK_ERRLOG("Unable to get PDU for tqpair=%p\n", tqpair); 322 abort(); 323 return NULL; 324 } 325 326 tqpair->free_pdu_num--; 327 TAILQ_REMOVE(&tqpair->free_queue, pdu, tailq); 328 memset(pdu, 0, sizeof(*pdu)); 329 pdu->ref = 1; 330 pdu->hdr = &pdu->hdr_mem; 331 332 return pdu; 333 } 334 335 static void 336 spdk_nvmf_tcp_pdu_put(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 337 { 338 if (!pdu) { 339 return; 340 } 341 342 assert(pdu->ref > 0); 343 344 pdu->ref--; 345 if (pdu->ref == 0) { 346 tqpair->free_pdu_num++; 347 TAILQ_INSERT_HEAD(&tqpair->free_queue, pdu, tailq); 348 } 349 } 350 351 static struct spdk_nvmf_tcp_req * 352 spdk_nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 353 { 354 struct spdk_nvmf_tcp_req *tcp_req; 355 356 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 357 if (!tcp_req) { 358 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 359 return NULL; 360 } 361 362 memset(&tcp_req->cmd, 0, sizeof(tcp_req->cmd)); 363 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 364 tcp_req->next_expected_r2t_offset = 0; 365 tcp_req->r2tl_remain = 0; 366 tcp_req->c2h_data_offset = 0; 367 tcp_req->has_incapsule_data = false; 368 memset(&tcp_req->req.dif, 0, sizeof(tcp_req->req.dif)); 369 370 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 371 return tcp_req; 372 } 373 374 static void 375 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 376 { 377 struct spdk_nvmf_tcp_transport *ttransport; 378 379 if (!tcp_req) { 380 return; 381 } 382 383 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req); 384 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 385 struct spdk_nvmf_tcp_transport, transport); 386 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 387 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 388 } 389 390 static int 391 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req) 392 { 393 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 394 395 nvmf_tcp_request_free(tcp_req); 396 397 return 0; 398 } 399 400 static void 401 spdk_nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 402 enum spdk_nvmf_tcp_req_state state) 403 { 404 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 405 406 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 407 nvmf_tcp_request_free(tcp_req); 408 } 409 } 410 411 static void 412 spdk_nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 413 { 414 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 415 struct nvme_tcp_pdu *pdu, *tmp_pdu; 416 417 /* Free the pdus in the send_queue */ 418 TAILQ_FOREACH_SAFE(pdu, &tqpair->send_queue, tailq, tmp_pdu) { 419 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 420 /* Also check the pdu type, we need to calculte the c2h_data_pdu_cnt later */ 421 if (pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_DATA) { 422 assert(tqpair->c2h_data_pdu_cnt > 0); 423 tqpair->c2h_data_pdu_cnt--; 424 } 425 spdk_nvmf_tcp_pdu_put(tqpair, pdu); 426 } 427 428 while (!STAILQ_EMPTY(&tqpair->queued_c2h_data_tcp_req)) { 429 STAILQ_REMOVE_HEAD(&tqpair->queued_c2h_data_tcp_req, link); 430 } 431 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 432 433 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 434 435 /* Wipe the requests waiting for buffer from the global list */ 436 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 437 req_tmp) { 438 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 439 spdk_nvmf_request, buf_link); 440 } 441 442 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 443 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 444 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 445 } 446 447 static void 448 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 449 { 450 int i; 451 struct spdk_nvmf_tcp_req *tcp_req; 452 453 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 454 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 455 SPDK_ERRLOG("\tNum of requests in state[%d] = %d\n", i, tqpair->state_cntr[i]); 456 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 457 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 458 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 459 } 460 } 461 } 462 463 static void 464 spdk_nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 465 { 466 int err = 0; 467 468 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 469 470 spdk_poller_unregister(&tqpair->flush_poller); 471 spdk_sock_close(&tqpair->sock); 472 spdk_nvmf_tcp_cleanup_all_states(tqpair); 473 474 if (tqpair->free_pdu_num != (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 475 SPDK_ERRLOG("tqpair(%p) free pdu pool num is %u but should be %u\n", tqpair, 476 tqpair->free_pdu_num, 477 (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)); 478 err++; 479 } 480 481 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->max_queue_depth) { 482 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 483 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 484 tqpair->max_queue_depth); 485 err++; 486 } 487 488 if (tqpair->c2h_data_pdu_cnt != 0) { 489 SPDK_ERRLOG("tqpair(%p) free c2h_data_pdu cnt is %u but should be 0\n", tqpair, 490 tqpair->c2h_data_pdu_cnt); 491 err++; 492 } 493 494 if (err > 0) { 495 nvmf_tcp_dump_qpair_req_contents(tqpair); 496 } 497 free(tqpair->pdu); 498 free(tqpair->pdu_pool); 499 free(tqpair->req); 500 free(tqpair->reqs); 501 spdk_free(tqpair->buf); 502 spdk_free(tqpair->bufs); 503 free(tqpair->pdu_recv_buf.buf); 504 free(tqpair); 505 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n"); 506 } 507 508 static int 509 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 510 { 511 struct spdk_nvmf_tcp_transport *ttransport; 512 513 assert(transport != NULL); 514 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 515 516 pthread_mutex_destroy(&ttransport->lock); 517 free(ttransport); 518 return 0; 519 } 520 521 static struct spdk_nvmf_transport * 522 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 523 { 524 struct spdk_nvmf_tcp_transport *ttransport; 525 uint32_t sge_count; 526 uint32_t min_shared_buffers; 527 528 ttransport = calloc(1, sizeof(*ttransport)); 529 if (!ttransport) { 530 return NULL; 531 } 532 533 TAILQ_INIT(&ttransport->ports); 534 535 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 536 537 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 538 539 SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n" 540 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 541 " max_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 542 " in_capsule_data_size=%d, max_aq_depth=%d\n" 543 " num_shared_buffers=%d, c2h_success=%d,\n" 544 " dif_insert_or_strip=%d, sock_priority=%d\n", 545 opts->max_queue_depth, 546 opts->max_io_size, 547 opts->max_qpairs_per_ctrlr, 548 opts->io_unit_size, 549 opts->in_capsule_data_size, 550 opts->max_aq_depth, 551 opts->num_shared_buffers, 552 opts->c2h_success, 553 opts->dif_insert_or_strip, 554 opts->sock_priority); 555 556 if (opts->sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 557 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 558 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 559 opts->sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 560 free(ttransport); 561 return NULL; 562 } 563 564 /* I/O unit size cannot be larger than max I/O size */ 565 if (opts->io_unit_size > opts->max_io_size) { 566 opts->io_unit_size = opts->max_io_size; 567 } 568 569 sge_count = opts->max_io_size / opts->io_unit_size; 570 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 571 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 572 free(ttransport); 573 return NULL; 574 } 575 576 min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size; 577 if (min_shared_buffers > opts->num_shared_buffers) { 578 SPDK_ERRLOG("There are not enough buffers to satisfy" 579 "per-poll group caches for each thread. (%" PRIu32 ")" 580 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 581 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 582 spdk_nvmf_tcp_destroy(&ttransport->transport); 583 return NULL; 584 } 585 586 pthread_mutex_init(&ttransport->lock, NULL); 587 588 return &ttransport->transport; 589 } 590 591 static int 592 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid) 593 { 594 unsigned long long ull; 595 char *end = NULL; 596 597 ull = strtoull(trsvcid, &end, 10); 598 if (end == NULL || end == trsvcid || *end != '\0') { 599 return -1; 600 } 601 602 /* Valid TCP/IP port numbers are in [0, 65535] */ 603 if (ull > 65535) { 604 return -1; 605 } 606 607 return (int)ull; 608 } 609 610 /** 611 * Canonicalize a listen address trid. 612 */ 613 static int 614 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 615 const struct spdk_nvme_transport_id *trid) 616 { 617 int trsvcid_int; 618 619 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 620 if (trsvcid_int < 0) { 621 return -EINVAL; 622 } 623 624 memset(canon_trid, 0, sizeof(*canon_trid)); 625 canon_trid->trtype = SPDK_NVME_TRANSPORT_TCP; 626 canon_trid->adrfam = trid->adrfam; 627 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 628 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 629 630 return 0; 631 } 632 633 /** 634 * Find an existing listening port. 635 * 636 * Caller must hold ttransport->lock. 637 */ 638 static struct spdk_nvmf_tcp_port * 639 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 640 const struct spdk_nvme_transport_id *trid) 641 { 642 struct spdk_nvme_transport_id canon_trid; 643 struct spdk_nvmf_tcp_port *port; 644 645 if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 646 return NULL; 647 } 648 649 TAILQ_FOREACH(port, &ttransport->ports, link) { 650 if (spdk_nvme_transport_id_compare(&canon_trid, &port->trid) == 0) { 651 return port; 652 } 653 } 654 655 return NULL; 656 } 657 658 static int 659 spdk_nvmf_tcp_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 int trsvcid_int; 665 uint8_t adrfam; 666 667 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 668 669 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 670 if (trsvcid_int < 0) { 671 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 672 return -EINVAL; 673 } 674 675 pthread_mutex_lock(&ttransport->lock); 676 677 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 678 if (port) { 679 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Already listening on %s port %s\n", 680 trid->traddr, trid->trsvcid); 681 port->ref++; 682 pthread_mutex_unlock(&ttransport->lock); 683 return 0; 684 } 685 686 port = calloc(1, sizeof(*port)); 687 if (!port) { 688 SPDK_ERRLOG("Port allocation failed\n"); 689 free(port); 690 pthread_mutex_unlock(&ttransport->lock); 691 return -ENOMEM; 692 } 693 694 port->ref = 1; 695 696 if (_spdk_nvmf_tcp_canon_listen_trid(&port->trid, trid) != 0) { 697 SPDK_ERRLOG("Invalid traddr %s / trsvcid %s\n", 698 trid->traddr, trid->trsvcid); 699 free(port); 700 pthread_mutex_unlock(&ttransport->lock); 701 return -ENOMEM; 702 } 703 704 port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int); 705 if (port->listen_sock == NULL) { 706 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 707 trid->traddr, trsvcid_int, 708 spdk_strerror(errno), errno); 709 free(port); 710 pthread_mutex_unlock(&ttransport->lock); 711 return -errno; 712 } 713 714 if (spdk_sock_is_ipv4(port->listen_sock)) { 715 adrfam = SPDK_NVMF_ADRFAM_IPV4; 716 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 717 adrfam = SPDK_NVMF_ADRFAM_IPV6; 718 } else { 719 SPDK_ERRLOG("Unhandled socket type\n"); 720 adrfam = 0; 721 } 722 723 if (adrfam != trid->adrfam) { 724 SPDK_ERRLOG("Socket address family mismatch\n"); 725 spdk_sock_close(&port->listen_sock); 726 free(port); 727 pthread_mutex_unlock(&ttransport->lock); 728 return -EINVAL; 729 } 730 731 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %d ***\n", 732 trid->traddr, trsvcid_int); 733 734 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 735 pthread_mutex_unlock(&ttransport->lock); 736 737 return 0; 738 } 739 740 static int 741 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 742 const struct spdk_nvme_transport_id *trid) 743 { 744 struct spdk_nvmf_tcp_transport *ttransport; 745 struct spdk_nvmf_tcp_port *port; 746 int rc; 747 748 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 749 750 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 751 trid->traddr, trid->trsvcid); 752 753 pthread_mutex_lock(&ttransport->lock); 754 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 755 if (port) { 756 assert(port->ref > 0); 757 port->ref--; 758 if (port->ref == 0) { 759 TAILQ_REMOVE(&ttransport->ports, port, link); 760 spdk_sock_close(&port->listen_sock); 761 free(port); 762 } 763 rc = 0; 764 } else { 765 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Port not found\n"); 766 rc = -ENOENT; 767 } 768 pthread_mutex_unlock(&ttransport->lock); 769 770 return rc; 771 } 772 773 static int 774 spdk_nvmf_tcp_qpair_flush_pdus_internal(struct spdk_nvmf_tcp_qpair *tqpair) 775 { 776 const int array_size = 32; 777 struct iovec iovs[array_size]; 778 int iovcnt = 0; 779 int bytes = 0; 780 int total_length = 0; 781 uint32_t mapped_length = 0; 782 struct nvme_tcp_pdu *pdu; 783 int pdu_length; 784 TAILQ_HEAD(, nvme_tcp_pdu) completed_pdus_list; 785 786 pdu = TAILQ_FIRST(&tqpair->send_queue); 787 788 if (pdu == NULL) { 789 return 0; 790 } 791 792 /* 793 * Build up a list of iovecs for the first few PDUs in the 794 * tqpair 's send_queue. 795 */ 796 while (pdu != NULL && ((array_size - iovcnt) >= (2 + (int)pdu->data_iovcnt))) { 797 iovcnt += nvme_tcp_build_iovs(&iovs[iovcnt], 798 array_size - iovcnt, 799 pdu, 800 tqpair->host_hdgst_enable, 801 tqpair->host_ddgst_enable, 802 &mapped_length); 803 total_length += mapped_length; 804 pdu = TAILQ_NEXT(pdu, tailq); 805 } 806 807 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_START, 0, total_length, 0, iovcnt); 808 809 bytes = spdk_sock_writev(tqpair->sock, iovs, iovcnt); 810 if (bytes == -1) { 811 if (errno == EWOULDBLOCK || errno == EAGAIN) { 812 return 1; 813 } else { 814 SPDK_ERRLOG("spdk_sock_writev() failed, errno %d: %s\n", 815 errno, spdk_strerror(errno)); 816 return -1; 817 } 818 } 819 820 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_DONE, 0, bytes, 0, 0); 821 822 pdu = TAILQ_FIRST(&tqpair->send_queue); 823 824 /* 825 * Free any PDUs that were fully written. If a PDU was only 826 * partially written, update its writev_offset so that next 827 * time only the unwritten portion will be sent to writev(). 828 */ 829 TAILQ_INIT(&completed_pdus_list); 830 while (bytes > 0) { 831 pdu_length = pdu->hdr->common.plen - pdu->writev_offset; 832 if (bytes >= pdu_length) { 833 bytes -= pdu_length; 834 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 835 TAILQ_INSERT_TAIL(&completed_pdus_list, pdu, tailq); 836 pdu = TAILQ_FIRST(&tqpair->send_queue); 837 838 } else { 839 pdu->writev_offset += bytes; 840 bytes = 0; 841 } 842 } 843 844 while (!TAILQ_EMPTY(&completed_pdus_list)) { 845 pdu = TAILQ_FIRST(&completed_pdus_list); 846 TAILQ_REMOVE(&completed_pdus_list, pdu, tailq); 847 assert(pdu->cb_fn != NULL); 848 pdu->cb_fn(pdu->cb_arg); 849 spdk_nvmf_tcp_pdu_put(tqpair, pdu); 850 } 851 852 return TAILQ_EMPTY(&tqpair->send_queue) ? 0 : 1; 853 } 854 855 static int 856 spdk_nvmf_tcp_qpair_flush_pdus(void *_tqpair) 857 { 858 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 859 int rc; 860 861 if (tqpair->state == NVME_TCP_QPAIR_STATE_RUNNING) { 862 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 863 if (rc == 0 && tqpair->flush_poller != NULL) { 864 spdk_poller_unregister(&tqpair->flush_poller); 865 } else if (rc == 1 && tqpair->flush_poller == NULL) { 866 tqpair->flush_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_flush_pdus, 867 tqpair, 50); 868 } 869 } else { 870 /* 871 * If the tqpair state is not RUNNING, then 872 * keep trying to flush PDUs until our list is 873 * empty - to make sure all data is sent before 874 * closing the connection. 875 */ 876 do { 877 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 878 } while (rc == 1); 879 } 880 881 if (rc < 0 && tqpair->state < NVME_TCP_QPAIR_STATE_EXITING) { 882 /* 883 * If the poller has already started destruction of the tqpair, 884 * i.e. the socket read failed, then the connection state may already 885 * be EXITED. We don't want to set it back to EXITING in that case. 886 */ 887 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 888 } 889 890 return -1; 891 } 892 893 static void 894 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 895 struct nvme_tcp_pdu *pdu, 896 nvme_tcp_qpair_xfer_complete_cb cb_fn, 897 void *cb_arg) 898 { 899 int enable_digest; 900 int hlen; 901 uint32_t crc32c; 902 903 hlen = pdu->hdr->common.hlen; 904 enable_digest = 1; 905 if (pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 906 pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 907 /* this PDU should be sent without digest */ 908 enable_digest = 0; 909 } 910 911 /* Header Digest */ 912 if (enable_digest && tqpair->host_hdgst_enable) { 913 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 914 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr->raw + hlen, crc32c); 915 } 916 917 /* Data Digest */ 918 if (pdu->data_len > 0 && enable_digest && tqpair->host_ddgst_enable) { 919 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 920 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 921 } 922 923 pdu->cb_fn = cb_fn; 924 pdu->cb_arg = cb_arg; 925 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 926 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 927 } 928 929 static int 930 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair, uint16_t size) 931 { 932 int i; 933 struct spdk_nvmf_tcp_req *tcp_req; 934 struct spdk_nvmf_transport *transport = tqpair->qpair.transport; 935 uint32_t in_capsule_data_size; 936 937 in_capsule_data_size = transport->opts.in_capsule_data_size; 938 if (transport->opts.dif_insert_or_strip) { 939 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 940 } 941 942 if (!tqpair->qpair.sq_head_max) { 943 tqpair->req = calloc(1, sizeof(*tqpair->req)); 944 if (!tqpair->req) { 945 SPDK_ERRLOG("Unable to allocate req on tqpair=%p.\n", tqpair); 946 return -1; 947 } 948 949 if (in_capsule_data_size) { 950 tqpair->buf = spdk_zmalloc(in_capsule_data_size, 0x1000, 951 NULL, SPDK_ENV_LCORE_ID_ANY, 952 SPDK_MALLOC_DMA); 953 if (!tqpair->buf) { 954 SPDK_ERRLOG("Unable to allocate buf on tqpair=%p.\n", tqpair); 955 return -1; 956 } 957 } 958 959 tcp_req = tqpair->req; 960 tcp_req->ttag = 0; 961 tcp_req->req.qpair = &tqpair->qpair; 962 963 /* Set up memory to receive commands */ 964 if (tqpair->buf) { 965 tcp_req->buf = tqpair->buf; 966 } 967 968 /* Set the cmdn and rsp */ 969 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 970 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 971 972 /* Initialize request state to FREE */ 973 tcp_req->state = TCP_REQUEST_STATE_FREE; 974 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 975 976 tqpair->pdu = calloc(NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM + 1, sizeof(*tqpair->pdu)); 977 if (!tqpair->pdu) { 978 SPDK_ERRLOG("Unable to allocate pdu on tqpair=%p.\n", tqpair); 979 return -1; 980 } 981 982 for (i = 0; i < 1 + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; i++) { 983 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu[i], tailq); 984 } 985 986 tqpair->pdu_recv_buf.size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 987 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 988 tqpair->pdu_recv_buf.buf = calloc(1, tqpair->pdu_recv_buf.size); 989 if (!tqpair->pdu_recv_buf.buf) { 990 SPDK_ERRLOG("Unable to allocate the pdu recv buf on tqpair=%p with size=%d\n", tqpair, 991 tqpair->pdu_recv_buf.size); 992 return -1; 993 } 994 tqpair->pdu_in_progress.hdr = (union nvme_tcp_pdu_hdr *)tqpair->pdu_recv_buf.buf; 995 } else { 996 tqpair->reqs = calloc(size, sizeof(*tqpair->reqs)); 997 if (!tqpair->reqs) { 998 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 999 return -1; 1000 } 1001 1002 if (in_capsule_data_size) { 1003 tqpair->bufs = spdk_zmalloc(size * in_capsule_data_size, 0x1000, 1004 NULL, SPDK_ENV_LCORE_ID_ANY, 1005 SPDK_MALLOC_DMA); 1006 if (!tqpair->bufs) { 1007 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1008 return -1; 1009 } 1010 } 1011 1012 for (i = 0; i < size; i++) { 1013 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1014 1015 tcp_req->ttag = i + 1; 1016 tcp_req->req.qpair = &tqpair->qpair; 1017 1018 /* Set up memory to receive commands */ 1019 if (tqpair->bufs) { 1020 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1021 } 1022 1023 /* Set the cmdn and rsp */ 1024 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1025 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1026 1027 /* Initialize request state to FREE */ 1028 tcp_req->state = TCP_REQUEST_STATE_FREE; 1029 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 1030 } 1031 1032 tqpair->pdu_pool = calloc(size, sizeof(*tqpair->pdu_pool)); 1033 if (!tqpair->pdu_pool) { 1034 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1035 return -1; 1036 } 1037 1038 for (i = 0; i < size; i++) { 1039 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu_pool[i], tailq); 1040 } 1041 } 1042 1043 return 0; 1044 } 1045 1046 static int 1047 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1048 { 1049 struct spdk_nvmf_tcp_qpair *tqpair; 1050 int i; 1051 1052 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1053 1054 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 1055 1056 TAILQ_INIT(&tqpair->send_queue); 1057 TAILQ_INIT(&tqpair->free_queue); 1058 STAILQ_INIT(&tqpair->queued_c2h_data_tcp_req); 1059 1060 /* Initialise request state queues of the qpair */ 1061 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 1062 TAILQ_INIT(&tqpair->state_queue[i]); 1063 } 1064 1065 tqpair->host_hdgst_enable = true; 1066 tqpair->host_ddgst_enable = true; 1067 return 0; 1068 } 1069 1070 static int 1071 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1072 { 1073 1074 int rc; 1075 int buf_size; 1076 1077 /* set send buffer size */ 1078 buf_size = 2 * 1024 * 1024; 1079 rc = spdk_sock_set_sendbuf(tqpair->sock, buf_size); 1080 if (rc != 0) { 1081 SPDK_ERRLOG("spdk_sock_set_sendbuf failed\n"); 1082 return rc; 1083 } 1084 1085 /* set low water mark */ 1086 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr)); 1087 if (rc != 0) { 1088 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1089 return rc; 1090 } 1091 1092 return 0; 1093 } 1094 1095 static void 1096 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1097 struct spdk_nvmf_tcp_port *port, 1098 struct spdk_sock *sock, 1099 new_qpair_fn cb_fn, void *cb_arg) 1100 { 1101 struct spdk_nvmf_tcp_qpair *tqpair; 1102 int rc; 1103 1104 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 1105 port->trid.traddr, port->trid.trsvcid); 1106 1107 if (transport->opts.sock_priority) { 1108 rc = spdk_sock_set_priority(sock, transport->opts.sock_priority); 1109 if (rc) { 1110 SPDK_ERRLOG("Failed to set the priority of the socket\n"); 1111 spdk_sock_close(&sock); 1112 return; 1113 } 1114 } 1115 1116 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1117 if (tqpair == NULL) { 1118 SPDK_ERRLOG("Could not allocate new connection.\n"); 1119 spdk_sock_close(&sock); 1120 return; 1121 } 1122 1123 tqpair->sock = sock; 1124 tqpair->max_queue_depth = 1; 1125 tqpair->free_pdu_num = tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; 1126 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = tqpair->max_queue_depth; 1127 tqpair->port = port; 1128 tqpair->qpair.transport = transport; 1129 1130 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1131 sizeof(tqpair->target_addr), &tqpair->target_port, 1132 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1133 &tqpair->initiator_port); 1134 if (rc < 0) { 1135 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1136 spdk_nvmf_tcp_qpair_destroy(tqpair); 1137 return; 1138 } 1139 1140 cb_fn(&tqpair->qpair, cb_arg); 1141 } 1142 1143 static void 1144 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port, 1145 new_qpair_fn cb_fn, void *cb_arg) 1146 { 1147 struct spdk_sock *sock; 1148 int i; 1149 1150 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1151 sock = spdk_sock_accept(port->listen_sock); 1152 if (sock) { 1153 _spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn, cb_arg); 1154 } 1155 } 1156 } 1157 1158 static void 1159 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn, void *cb_arg) 1160 { 1161 struct spdk_nvmf_tcp_transport *ttransport; 1162 struct spdk_nvmf_tcp_port *port; 1163 1164 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1165 1166 TAILQ_FOREACH(port, &ttransport->ports, link) { 1167 spdk_nvmf_tcp_port_accept(transport, port, cb_fn, cb_arg); 1168 } 1169 } 1170 1171 static void 1172 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1173 struct spdk_nvme_transport_id *trid, 1174 struct spdk_nvmf_discovery_log_page_entry *entry) 1175 { 1176 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1177 entry->adrfam = trid->adrfam; 1178 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1179 1180 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1181 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1182 1183 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1184 } 1185 1186 static struct spdk_nvmf_transport_poll_group * 1187 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1188 { 1189 struct spdk_nvmf_tcp_poll_group *tgroup; 1190 1191 tgroup = calloc(1, sizeof(*tgroup)); 1192 if (!tgroup) { 1193 return NULL; 1194 } 1195 1196 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1197 if (!tgroup->sock_group) { 1198 goto cleanup; 1199 } 1200 1201 TAILQ_INIT(&tgroup->qpairs); 1202 1203 return &tgroup->group; 1204 1205 cleanup: 1206 free(tgroup); 1207 return NULL; 1208 } 1209 1210 static struct spdk_nvmf_transport_poll_group * 1211 spdk_nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1212 { 1213 struct spdk_nvmf_tcp_qpair *tqpair; 1214 struct spdk_sock_group *group = NULL; 1215 int rc; 1216 1217 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1218 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1219 if (!rc && group != NULL) { 1220 return spdk_sock_group_get_ctx(group); 1221 } 1222 1223 return NULL; 1224 } 1225 1226 static void 1227 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1228 { 1229 struct spdk_nvmf_tcp_poll_group *tgroup; 1230 1231 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1232 spdk_sock_group_close(&tgroup->sock_group); 1233 1234 free(tgroup); 1235 } 1236 1237 static inline void 1238 spdk_nvmf_tcp_reset_pdu_in_process(struct spdk_nvmf_tcp_qpair *tqpair) 1239 { 1240 struct nvme_tcp_pdu_recv_buf *pdu_recv_buf = &tqpair->pdu_recv_buf; 1241 char *dst, *src; 1242 1243 if (spdk_unlikely((pdu_recv_buf->off + sizeof(union nvme_tcp_pdu_hdr)) > 1244 pdu_recv_buf->size)) { 1245 if (pdu_recv_buf->remain_size) { 1246 dst = pdu_recv_buf->buf; 1247 src = (char *)((void *)pdu_recv_buf->buf + pdu_recv_buf->off); 1248 1249 /* purpose: to avoid overlap copy, so do not use memcpy if there is overlap case */ 1250 memmove(dst, src, pdu_recv_buf->remain_size); 1251 } 1252 tqpair->pdu_recv_buf.off = 0; 1253 } else if (!pdu_recv_buf->remain_size) { 1254 tqpair->pdu_recv_buf.off = 0; 1255 } 1256 1257 tqpair->pdu_in_progress.hdr = (union nvme_tcp_pdu_hdr *)((void *)pdu_recv_buf->buf + 1258 pdu_recv_buf->off); 1259 } 1260 1261 static void 1262 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1263 enum nvme_tcp_pdu_recv_state state) 1264 { 1265 if (tqpair->recv_state == state) { 1266 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1267 tqpair, state); 1268 return; 1269 } 1270 1271 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1272 tqpair->recv_state = state; 1273 1274 switch (state) { 1275 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1276 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1277 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1278 break; 1279 case NVME_TCP_PDU_RECV_STATE_ERROR: 1280 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1281 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1282 spdk_nvmf_tcp_reset_pdu_in_process(tqpair); 1283 break; 1284 default: 1285 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1286 abort(); 1287 break; 1288 } 1289 } 1290 1291 static int 1292 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx) 1293 { 1294 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1295 1296 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1297 1298 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1299 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1300 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 1301 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair); 1302 spdk_poller_unregister(&tqpair->timeout_poller); 1303 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 1304 1305 return 0; 1306 } 1307 1308 static void 1309 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1310 { 1311 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1312 1313 if (!tqpair->timeout_poller) { 1314 tqpair->timeout_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_handle_timeout, tqpair, 1315 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1316 } 1317 } 1318 1319 static void 1320 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1321 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1322 { 1323 struct nvme_tcp_pdu *rsp_pdu; 1324 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1325 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1326 uint32_t copy_len; 1327 1328 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1329 if (!rsp_pdu) { 1330 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1331 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1332 return; 1333 } 1334 1335 c2h_term_req = &rsp_pdu->hdr->term_req; 1336 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1337 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1338 1339 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1340 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1341 DSET32(&c2h_term_req->fei, error_offset); 1342 } 1343 1344 copy_len = pdu->hdr->common.hlen; 1345 if (copy_len > SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE) { 1346 copy_len = SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE; 1347 } 1348 1349 /* Copy the error info into the buffer */ 1350 memcpy((uint8_t *)rsp_pdu->hdr->raw + c2h_term_req_hdr_len, pdu->hdr->raw, copy_len); 1351 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr->raw + c2h_term_req_hdr_len, copy_len); 1352 1353 /* Contain the header of the wrong received pdu */ 1354 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1355 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1356 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair); 1357 } 1358 1359 static void 1360 spdk_nvmf_tcp_capsule_cmd_hdr_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 = spdk_nvmf_tcp_req_get(tqpair); 1367 if (!tcp_req) { 1368 SPDK_ERRLOG("Cannot allocate tcp_req\n"); 1369 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1370 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1371 return; 1372 } 1373 1374 pdu->ctx = tcp_req; 1375 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 1376 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1377 return; 1378 } 1379 1380 static void 1381 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1382 struct spdk_nvmf_tcp_qpair *tqpair, 1383 struct nvme_tcp_pdu *pdu) 1384 { 1385 struct spdk_nvmf_tcp_req *tcp_req; 1386 struct spdk_nvme_tcp_cmd *capsule_cmd; 1387 uint32_t error_offset = 0; 1388 enum spdk_nvme_tcp_term_req_fes fes; 1389 1390 capsule_cmd = &pdu->hdr->capsule_cmd; 1391 tcp_req = pdu->ctx; 1392 assert(tcp_req != NULL); 1393 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1394 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1395 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1396 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1397 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1398 goto err; 1399 } 1400 1401 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1402 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1403 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1404 1405 return; 1406 err: 1407 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1408 } 1409 1410 static void 1411 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1412 struct spdk_nvmf_tcp_qpair *tqpair, 1413 struct nvme_tcp_pdu *pdu) 1414 { 1415 struct spdk_nvmf_tcp_req *tcp_req; 1416 uint32_t error_offset = 0; 1417 enum spdk_nvme_tcp_term_req_fes fes = 0; 1418 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1419 bool ttag_offset_error = false; 1420 1421 h2c_data = &pdu->hdr->h2c_data; 1422 1423 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1424 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1425 1426 /* According to the information in the pdu to find the req */ 1427 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER], 1428 state_link) { 1429 if ((tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid) && (tcp_req->ttag == h2c_data->ttag)) { 1430 break; 1431 } 1432 1433 if (!ttag_offset_error && (tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid)) { 1434 ttag_offset_error = true; 1435 } 1436 } 1437 1438 if (!tcp_req) { 1439 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1440 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1441 if (!ttag_offset_error) { 1442 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1443 } else { 1444 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1445 } 1446 goto err; 1447 } 1448 1449 if (tcp_req->next_expected_r2t_offset != h2c_data->datao) { 1450 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1451 "tcp_req(%p), tqpair=%p, expected_r2t_offset=%u, but data offset =%u\n", 1452 tcp_req, tqpair, tcp_req->next_expected_r2t_offset, h2c_data->datao); 1453 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1454 goto err; 1455 } 1456 1457 if (h2c_data->datal > tqpair->maxh2cdata) { 1458 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req(%p), tqpair=%p, datao=%u execeeds maxh2cdata size=%u\n", 1459 tcp_req, tqpair, h2c_data->datao, tqpair->maxh2cdata); 1460 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1461 goto err; 1462 } 1463 1464 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1465 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1466 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1467 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1468 fes = SPDK_NVME_TCP_TERM_REQ_FES_R2T_LIMIT_EXCEEDED; 1469 goto err; 1470 } 1471 1472 pdu->ctx = tcp_req; 1473 1474 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1475 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1476 } 1477 1478 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1479 h2c_data->datao, h2c_data->datal); 1480 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1481 return; 1482 1483 err: 1484 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1485 } 1486 1487 static void 1488 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1489 { 1490 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1491 nvmf_tcp_request_free(tcp_req); 1492 } 1493 1494 static void 1495 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1496 struct spdk_nvmf_tcp_qpair *tqpair) 1497 { 1498 struct nvme_tcp_pdu *rsp_pdu; 1499 struct spdk_nvme_tcp_rsp *capsule_resp; 1500 1501 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1502 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1503 if (!rsp_pdu) { 1504 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1505 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1506 return; 1507 } 1508 1509 capsule_resp = &rsp_pdu->hdr->capsule_resp; 1510 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1511 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1512 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1513 if (tqpair->host_hdgst_enable) { 1514 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1515 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1516 } 1517 1518 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req); 1519 } 1520 1521 static void 1522 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1523 { 1524 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1525 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1526 struct spdk_nvmf_tcp_qpair, qpair); 1527 1528 assert(tqpair != NULL); 1529 assert(tcp_req->c2h_data_pdu_num > 0); 1530 tcp_req->c2h_data_pdu_num--; 1531 if (!tcp_req->c2h_data_pdu_num) { 1532 if (tqpair->qpair.transport->opts.c2h_success) { 1533 nvmf_tcp_request_free(tcp_req); 1534 } else { 1535 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1536 } 1537 } 1538 1539 tqpair->c2h_data_pdu_cnt--; 1540 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 1541 } 1542 1543 static void 1544 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1545 struct spdk_nvmf_tcp_req *tcp_req) 1546 { 1547 struct nvme_tcp_pdu *rsp_pdu; 1548 struct spdk_nvme_tcp_r2t_hdr *r2t; 1549 1550 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1551 if (!rsp_pdu) { 1552 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1553 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1554 return; 1555 } 1556 1557 r2t = &rsp_pdu->hdr->r2t; 1558 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1559 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1560 1561 if (tqpair->host_hdgst_enable) { 1562 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1563 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1564 } 1565 1566 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1567 r2t->ttag = tcp_req->ttag; 1568 r2t->r2to = tcp_req->next_expected_r2t_offset; 1569 r2t->r2tl = spdk_min(tcp_req->req.length - tcp_req->next_expected_r2t_offset, tqpair->maxh2cdata); 1570 tcp_req->r2tl_remain = r2t->r2tl; 1571 1572 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1573 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1574 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1575 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, NULL); 1576 } 1577 1578 static void 1579 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1580 struct spdk_nvmf_tcp_qpair *tqpair, 1581 struct nvme_tcp_pdu *pdu) 1582 { 1583 struct spdk_nvmf_tcp_req *tcp_req; 1584 1585 tcp_req = pdu->ctx; 1586 assert(tcp_req != NULL); 1587 1588 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1589 1590 tcp_req->next_expected_r2t_offset += pdu->data_len; 1591 tcp_req->r2tl_remain -= pdu->data_len; 1592 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1593 1594 if (!tcp_req->r2tl_remain) { 1595 if (tcp_req->next_expected_r2t_offset == tcp_req->req.length) { 1596 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1597 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1598 } else { 1599 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Send r2t pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 1600 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 1601 } 1602 } 1603 } 1604 1605 static void 1606 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1607 { 1608 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1609 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1610 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1611 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1612 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1613 DGET32(h2c_term_req->fei)); 1614 } 1615 } 1616 1617 static void 1618 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1619 struct nvme_tcp_pdu *pdu) 1620 { 1621 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr->term_req; 1622 uint32_t error_offset = 0; 1623 enum spdk_nvme_tcp_term_req_fes fes; 1624 1625 1626 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1627 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1628 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1629 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1630 goto end; 1631 } 1632 1633 /* set the data buffer */ 1634 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr->raw + h2c_term_req->common.hlen, 1635 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1636 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1637 return; 1638 end: 1639 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1640 return; 1641 } 1642 1643 static void 1644 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1645 struct nvme_tcp_pdu *pdu) 1646 { 1647 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr->term_req; 1648 1649 spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1650 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1651 return; 1652 } 1653 1654 static void 1655 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1656 { 1657 int rc = 0; 1658 struct nvme_tcp_pdu *pdu; 1659 uint32_t crc32c, error_offset = 0; 1660 enum spdk_nvme_tcp_term_req_fes fes; 1661 struct spdk_nvmf_tcp_transport *ttransport; 1662 1663 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1664 pdu = &tqpair->pdu_in_progress; 1665 1666 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1667 /* check data digest if need */ 1668 if (pdu->ddgst_enable) { 1669 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1670 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1671 if (rc == 0) { 1672 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1673 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1674 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1675 return; 1676 1677 } 1678 } 1679 1680 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1681 switch (pdu->hdr->common.pdu_type) { 1682 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1683 spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1684 break; 1685 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1686 spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1687 break; 1688 1689 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1690 spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1691 break; 1692 1693 default: 1694 /* The code should not go to here */ 1695 SPDK_ERRLOG("The code should not go to here\n"); 1696 break; 1697 } 1698 } 1699 1700 static void 1701 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg) 1702 { 1703 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1704 1705 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1706 } 1707 1708 static void 1709 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1710 struct spdk_nvmf_tcp_qpair *tqpair, 1711 struct nvme_tcp_pdu *pdu) 1712 { 1713 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr->ic_req; 1714 struct nvme_tcp_pdu *rsp_pdu; 1715 struct spdk_nvme_tcp_ic_resp *ic_resp; 1716 uint32_t error_offset = 0; 1717 enum spdk_nvme_tcp_term_req_fes fes; 1718 1719 /* Only PFV 0 is defined currently */ 1720 if (ic_req->pfv != 0) { 1721 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1722 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1723 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1724 goto end; 1725 } 1726 1727 /* MAXR2T is 0's based */ 1728 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1729 1730 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1731 if (!tqpair->host_hdgst_enable) { 1732 tqpair->pdu_recv_buf.size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1733 } 1734 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1735 if (!tqpair->host_ddgst_enable) { 1736 tqpair->pdu_recv_buf.size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1737 } 1738 1739 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1740 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1741 1742 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1743 if (!rsp_pdu) { 1744 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1745 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1746 return; 1747 } 1748 1749 ic_resp = &rsp_pdu->hdr->ic_resp; 1750 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1751 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1752 ic_resp->pfv = 0; 1753 ic_resp->cpda = tqpair->cpda; 1754 tqpair->maxh2cdata = spdk_min(NVMF_TCP_PDU_MAX_H2C_DATA_SIZE, 1755 ttransport->transport.opts.io_unit_size); 1756 ic_resp->maxh2cdata = tqpair->maxh2cdata; 1757 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1758 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1759 1760 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1761 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1762 1763 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair); 1764 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1765 return; 1766 end: 1767 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1768 return; 1769 } 1770 1771 static void 1772 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1773 { 1774 struct nvme_tcp_pdu *pdu; 1775 int rc; 1776 uint32_t crc32c, error_offset = 0; 1777 enum spdk_nvme_tcp_term_req_fes fes; 1778 struct spdk_nvmf_tcp_transport *ttransport; 1779 1780 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1781 pdu = &tqpair->pdu_in_progress; 1782 1783 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1784 pdu->hdr->common.pdu_type); 1785 /* check header digest if needed */ 1786 if (pdu->has_hdgst) { 1787 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1788 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1789 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr->raw + pdu->hdr->common.hlen, crc32c); 1790 if (rc == 0) { 1791 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1792 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1793 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1794 return; 1795 1796 } 1797 } 1798 1799 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1800 switch (pdu->hdr->common.pdu_type) { 1801 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1802 spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1803 break; 1804 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1805 spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1806 break; 1807 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1808 spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1809 break; 1810 1811 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1812 spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1813 break; 1814 1815 default: 1816 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr->common.pdu_type); 1817 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1818 error_offset = 1; 1819 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1820 break; 1821 } 1822 } 1823 1824 static void 1825 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1826 { 1827 struct nvme_tcp_pdu *pdu; 1828 uint32_t error_offset = 0; 1829 enum spdk_nvme_tcp_term_req_fes fes; 1830 uint8_t expected_hlen, pdo; 1831 bool plen_error = false, pdo_error = false; 1832 1833 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1834 pdu = &tqpair->pdu_in_progress; 1835 1836 if (pdu->hdr->common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1837 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1838 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1839 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1840 goto err; 1841 } 1842 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1843 if (pdu->hdr->common.plen != expected_hlen) { 1844 plen_error = true; 1845 } 1846 } else { 1847 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1848 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1849 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1850 goto err; 1851 } 1852 1853 switch (pdu->hdr->common.pdu_type) { 1854 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1855 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1856 pdo = pdu->hdr->common.pdo; 1857 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1858 pdo_error = true; 1859 break; 1860 } 1861 1862 if (pdu->hdr->common.plen < expected_hlen) { 1863 plen_error = true; 1864 } 1865 break; 1866 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1867 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1868 pdo = pdu->hdr->common.pdo; 1869 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1870 pdo_error = true; 1871 break; 1872 } 1873 if (pdu->hdr->common.plen < expected_hlen) { 1874 plen_error = true; 1875 } 1876 break; 1877 1878 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1879 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1880 if ((pdu->hdr->common.plen <= expected_hlen) || 1881 (pdu->hdr->common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1882 plen_error = true; 1883 } 1884 break; 1885 1886 default: 1887 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr->common.pdu_type); 1888 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1889 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1890 goto err; 1891 } 1892 } 1893 1894 if (pdu->hdr->common.hlen != expected_hlen) { 1895 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1896 pdu->hdr->common.pdu_type, 1897 expected_hlen, pdu->hdr->common.hlen, tqpair); 1898 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1899 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1900 goto err; 1901 } else if (pdo_error) { 1902 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1903 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1904 } else if (plen_error) { 1905 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1906 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1907 goto err; 1908 } else { 1909 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1910 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1911 return; 1912 } 1913 err: 1914 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1915 } 1916 1917 static int 1918 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1919 int read_len) 1920 { 1921 int rc; 1922 1923 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1924 read_offset, read_len, pdu->dif_ctx); 1925 if (rc != 0) { 1926 SPDK_ERRLOG("DIF generate failed\n"); 1927 } 1928 1929 return rc; 1930 } 1931 1932 static int 1933 nvme_tcp_recv_buf_read(struct spdk_sock *sock, struct nvme_tcp_pdu_recv_buf *pdu_recv_buf) 1934 { 1935 int rc; 1936 1937 rc = nvme_tcp_read_data(sock, pdu_recv_buf->size - pdu_recv_buf->off, 1938 (void *)pdu_recv_buf->buf + pdu_recv_buf->off); 1939 if (rc < 0) { 1940 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect sock=%p\n", sock); 1941 } else if (rc > 0) { 1942 pdu_recv_buf->remain_size = rc; 1943 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1944 } 1945 1946 return rc; 1947 } 1948 1949 static uint32_t 1950 nvme_tcp_read_data_from_pdu_recv_buf(struct nvme_tcp_pdu_recv_buf *pdu_recv_buf, 1951 uint32_t expected_size, 1952 char *dst) 1953 { 1954 uint32_t size; 1955 1956 assert(pdu_recv_buf->remain_size > 0); 1957 size = spdk_min(expected_size, pdu_recv_buf->remain_size); 1958 if (dst) { 1959 memcpy(dst, (void *)pdu_recv_buf->buf + pdu_recv_buf->off, size); 1960 } 1961 pdu_recv_buf->off += size; 1962 pdu_recv_buf->remain_size -= size; 1963 1964 1965 return size; 1966 } 1967 1968 static int 1969 nvme_tcp_read_payload_data_from_pdu_recv_buf(struct nvme_tcp_pdu_recv_buf *pdu_recv_buf, 1970 struct nvme_tcp_pdu *pdu) 1971 { 1972 struct iovec iov[NVME_TCP_MAX_SGL_DESCRIPTORS + 1]; 1973 int iovcnt, i; 1974 uint32_t size = 0; 1975 void *dst; 1976 1977 assert(pdu_recv_buf->remain_size > 0); 1978 iovcnt = nvme_tcp_build_payload_iovs(iov, NVME_TCP_MAX_SGL_DESCRIPTORS + 1, pdu, 1979 pdu->ddgst_enable, NULL); 1980 assert(iovcnt >= 0); 1981 for (i = 0; i < iovcnt; i++) { 1982 if (!pdu_recv_buf->remain_size) { 1983 break; 1984 } 1985 1986 dst = NULL; 1987 if (pdu->hdr->common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) { 1988 dst = iov[i].iov_base; 1989 } 1990 size += nvme_tcp_read_data_from_pdu_recv_buf(pdu_recv_buf, iov[i].iov_len, dst); 1991 } 1992 1993 return size; 1994 } 1995 1996 static int 1997 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1998 { 1999 int rc = 0; 2000 struct nvme_tcp_pdu *pdu; 2001 enum nvme_tcp_pdu_recv_state prev_state; 2002 uint32_t data_len; 2003 2004 /* The loop here is to allow for several back-to-back state changes. */ 2005 do { 2006 prev_state = tqpair->recv_state; 2007 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2008 2009 pdu = &tqpair->pdu_in_progress; 2010 switch (tqpair->recv_state) { 2011 /* Wait for the common header */ 2012 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2013 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2014 if (!tqpair->pdu_recv_buf.remain_size) { 2015 rc = nvme_tcp_recv_buf_read(tqpair->sock, &tqpair->pdu_recv_buf); 2016 if (rc <= 0) { 2017 return rc; 2018 } 2019 } 2020 rc = nvme_tcp_read_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, 2021 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2022 NULL); 2023 pdu->ch_valid_bytes += rc; 2024 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 2025 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2026 } 2027 2028 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2029 return NVME_TCP_PDU_IN_PROGRESS; 2030 } 2031 2032 /* The command header of this PDU has now been read from the socket. */ 2033 spdk_nvmf_tcp_pdu_ch_handle(tqpair); 2034 break; 2035 /* Wait for the pdu specific header */ 2036 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2037 if (!tqpair->pdu_recv_buf.remain_size) { 2038 rc = nvme_tcp_recv_buf_read(tqpair->sock, &tqpair->pdu_recv_buf); 2039 if (rc <= 0) { 2040 return rc; 2041 } 2042 } 2043 2044 rc = nvme_tcp_read_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, 2045 pdu->psh_len - pdu->psh_valid_bytes, 2046 NULL); 2047 pdu->psh_valid_bytes += rc; 2048 if (pdu->psh_valid_bytes < pdu->psh_len) { 2049 return NVME_TCP_PDU_IN_PROGRESS; 2050 } 2051 2052 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2053 spdk_nvmf_tcp_pdu_psh_handle(tqpair); 2054 break; 2055 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2056 /* check whether the data is valid, if not we just return */ 2057 if (!pdu->data_len) { 2058 return NVME_TCP_PDU_IN_PROGRESS; 2059 } 2060 2061 data_len = pdu->data_len; 2062 /* data digest */ 2063 if (spdk_unlikely((pdu->hdr->common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2064 tqpair->host_ddgst_enable)) { 2065 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2066 pdu->ddgst_enable = true; 2067 } 2068 2069 if (tqpair->pdu_recv_buf.remain_size) { 2070 rc = nvme_tcp_read_payload_data_from_pdu_recv_buf(&tqpair->pdu_recv_buf, pdu); 2071 pdu->readv_offset += rc; 2072 } 2073 2074 if (pdu->readv_offset < data_len) { 2075 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2076 if (rc < 0) { 2077 return NVME_TCP_PDU_IN_PROGRESS; 2078 } 2079 pdu->readv_offset += rc; 2080 } 2081 2082 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 2083 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 2084 if (rc != 0) { 2085 return NVME_TCP_PDU_FATAL; 2086 } 2087 } 2088 2089 if (pdu->readv_offset < data_len) { 2090 return NVME_TCP_PDU_IN_PROGRESS; 2091 } 2092 2093 /* All of this PDU has now been read from the socket. */ 2094 spdk_nvmf_tcp_pdu_payload_handle(tqpair); 2095 break; 2096 case NVME_TCP_PDU_RECV_STATE_ERROR: 2097 /* Check whether the connection is closed. Each time, we only read 1 byte every time */ 2098 rc = nvme_tcp_read_data(tqpair->sock, 1, (void *)&pdu->hdr->common); 2099 if (rc < 0) { 2100 return NVME_TCP_PDU_FATAL; 2101 } 2102 break; 2103 default: 2104 assert(0); 2105 SPDK_ERRLOG("code should not come to here"); 2106 break; 2107 } 2108 } while (tqpair->recv_state != prev_state); 2109 2110 return rc; 2111 } 2112 2113 static enum spdk_nvme_data_transfer 2114 spdk_nvmf_tcp_req_get_xfer(struct spdk_nvmf_tcp_req *tcp_req) { 2115 enum spdk_nvme_data_transfer xfer; 2116 struct spdk_nvme_cmd *cmd = &tcp_req->req.cmd->nvme_cmd; 2117 struct spdk_nvme_sgl_descriptor *sgl = &cmd->dptr.sgl1; 2118 2119 /* Figure out data transfer direction */ 2120 if (cmd->opc == SPDK_NVME_OPC_FABRIC) 2121 { 2122 xfer = spdk_nvme_opc_get_data_transfer(tcp_req->req.cmd->nvmf_cmd.fctype); 2123 } else 2124 { 2125 xfer = spdk_nvme_opc_get_data_transfer(cmd->opc); 2126 2127 /* Some admin commands are special cases */ 2128 if ((tcp_req->req.qpair->qid == 0) && 2129 ((cmd->opc == SPDK_NVME_OPC_GET_FEATURES) || 2130 (cmd->opc == SPDK_NVME_OPC_SET_FEATURES))) { 2131 switch (cmd->cdw10 & 0xff) { 2132 case SPDK_NVME_FEAT_LBA_RANGE_TYPE: 2133 case SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION: 2134 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 2135 break; 2136 default: 2137 xfer = SPDK_NVME_DATA_NONE; 2138 } 2139 } 2140 } 2141 2142 if (xfer == SPDK_NVME_DATA_NONE) 2143 { 2144 return xfer; 2145 } 2146 2147 /* Even for commands that may transfer data, they could have specified 0 length. 2148 * We want those to show up with xfer SPDK_NVME_DATA_NONE. 2149 */ 2150 switch (sgl->generic.type) 2151 { 2152 case SPDK_NVME_SGL_TYPE_DATA_BLOCK: 2153 case SPDK_NVME_SGL_TYPE_BIT_BUCKET: 2154 case SPDK_NVME_SGL_TYPE_SEGMENT: 2155 case SPDK_NVME_SGL_TYPE_LAST_SEGMENT: 2156 case SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK: 2157 if (sgl->unkeyed.length == 0) { 2158 xfer = SPDK_NVME_DATA_NONE; 2159 } 2160 break; 2161 case SPDK_NVME_SGL_TYPE_KEYED_DATA_BLOCK: 2162 if (sgl->keyed.length == 0) { 2163 xfer = SPDK_NVME_DATA_NONE; 2164 } 2165 break; 2166 } 2167 2168 return xfer; 2169 } 2170 2171 static int 2172 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2173 struct spdk_nvmf_transport *transport, 2174 struct spdk_nvmf_transport_poll_group *group) 2175 { 2176 struct spdk_nvmf_request *req = &tcp_req->req; 2177 struct spdk_nvme_cmd *cmd; 2178 struct spdk_nvme_cpl *rsp; 2179 struct spdk_nvme_sgl_descriptor *sgl; 2180 uint32_t length; 2181 2182 cmd = &req->cmd->nvme_cmd; 2183 rsp = &req->rsp->nvme_cpl; 2184 sgl = &cmd->dptr.sgl1; 2185 2186 length = sgl->unkeyed.length; 2187 2188 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2189 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2190 if (length > transport->opts.max_io_size) { 2191 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2192 length, transport->opts.max_io_size); 2193 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2194 return -1; 2195 } 2196 2197 /* fill request length and populate iovs */ 2198 req->length = length; 2199 2200 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", length); 2201 2202 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2203 req->dif.orig_length = length; 2204 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2205 req->dif.elba_length = length; 2206 } 2207 2208 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2209 /* No available buffers. Queue this request up. */ 2210 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", 2211 tcp_req); 2212 return 0; 2213 } 2214 2215 /* backward compatible */ 2216 req->data = req->iov[0].iov_base; 2217 2218 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 2219 tcp_req, req->iovcnt, req->data); 2220 2221 return 0; 2222 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2223 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2224 uint64_t offset = sgl->address; 2225 uint32_t max_len = transport->opts.in_capsule_data_size; 2226 2227 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2228 offset, length); 2229 2230 if (offset > max_len) { 2231 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2232 offset, max_len); 2233 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2234 return -1; 2235 } 2236 max_len -= (uint32_t)offset; 2237 2238 if (length > max_len) { 2239 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2240 length, max_len); 2241 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2242 return -1; 2243 } 2244 2245 req->data = tcp_req->buf + offset; 2246 req->data_from_pool = false; 2247 req->length = length; 2248 2249 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2250 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2251 req->dif.elba_length = length; 2252 } 2253 2254 req->iov[0].iov_base = req->data; 2255 req->iov[0].iov_len = length; 2256 req->iovcnt = 1; 2257 2258 return 0; 2259 } 2260 2261 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2262 sgl->generic.type, sgl->generic.subtype); 2263 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2264 return -1; 2265 } 2266 2267 static inline enum spdk_nvme_media_error_status_code 2268 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2269 enum spdk_nvme_media_error_status_code result; 2270 2271 switch (err_type) 2272 { 2273 case SPDK_DIF_REFTAG_ERROR: 2274 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2275 break; 2276 case SPDK_DIF_APPTAG_ERROR: 2277 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2278 break; 2279 case SPDK_DIF_GUARD_ERROR: 2280 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2281 break; 2282 default: 2283 SPDK_UNREACHABLE(); 2284 break; 2285 } 2286 2287 return result; 2288 } 2289 2290 static void 2291 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2292 struct spdk_nvmf_tcp_req *tcp_req) 2293 { 2294 struct nvme_tcp_pdu *rsp_pdu; 2295 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2296 uint32_t plen, pdo, alignment; 2297 int rc; 2298 2299 assert(tcp_req == STAILQ_FIRST(&tqpair->queued_c2h_data_tcp_req)); 2300 2301 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2302 2303 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 2304 assert(rsp_pdu != NULL); 2305 2306 c2h_data = &rsp_pdu->hdr->c2h_data; 2307 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2308 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2309 2310 if (tqpair->host_hdgst_enable) { 2311 plen += SPDK_NVME_TCP_DIGEST_LEN; 2312 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2313 } 2314 2315 /* set the psh */ 2316 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2317 c2h_data->datal = spdk_min(NVMF_TCP_PDU_MAX_C2H_DATA_SIZE, 2318 tcp_req->req.length - tcp_req->c2h_data_offset); 2319 c2h_data->datao = tcp_req->c2h_data_offset; 2320 2321 /* set the padding */ 2322 rsp_pdu->padding_len = 0; 2323 pdo = plen; 2324 if (tqpair->cpda) { 2325 alignment = (tqpair->cpda + 1) << 2; 2326 if (alignment > plen) { 2327 rsp_pdu->padding_len = alignment - plen; 2328 pdo = plen = alignment; 2329 } 2330 } 2331 2332 c2h_data->common.pdo = pdo; 2333 plen += c2h_data->datal; 2334 if (tqpair->host_ddgst_enable) { 2335 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2336 plen += SPDK_NVME_TCP_DIGEST_LEN; 2337 } 2338 2339 c2h_data->common.plen = plen; 2340 2341 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2342 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2343 } 2344 2345 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2346 c2h_data->datao, c2h_data->datal); 2347 2348 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2349 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2350 struct spdk_dif_error err_blk = {}; 2351 2352 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2353 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2354 if (rc != 0) { 2355 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2356 err_blk.err_type, err_blk.err_offset); 2357 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2358 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2359 STAILQ_REMOVE_HEAD(&tqpair->queued_c2h_data_tcp_req, link); 2360 spdk_nvmf_tcp_pdu_put(tqpair, rsp_pdu); 2361 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2362 return; 2363 } 2364 } 2365 2366 tcp_req->c2h_data_offset += c2h_data->datal; 2367 if (tcp_req->c2h_data_offset == tcp_req->req.length) { 2368 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Last pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 2369 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2370 if (tqpair->qpair.transport->opts.c2h_success) { 2371 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2372 } 2373 STAILQ_REMOVE_HEAD(&tqpair->queued_c2h_data_tcp_req, link); 2374 } 2375 2376 tqpair->c2h_data_pdu_cnt += 1; 2377 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2378 } 2379 2380 static int 2381 spdk_nvmf_tcp_calc_c2h_data_pdu_num(struct spdk_nvmf_tcp_req *tcp_req) 2382 { 2383 return (tcp_req->req.length + NVMF_TCP_PDU_MAX_C2H_DATA_SIZE - 1) / 2384 NVMF_TCP_PDU_MAX_C2H_DATA_SIZE; 2385 } 2386 2387 static void 2388 spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair) 2389 { 2390 struct spdk_nvmf_tcp_req *tcp_req; 2391 2392 while (!STAILQ_EMPTY(&tqpair->queued_c2h_data_tcp_req) && 2393 (tqpair->c2h_data_pdu_cnt < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 2394 tcp_req = STAILQ_FIRST(&tqpair->queued_c2h_data_tcp_req); 2395 spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2396 } 2397 } 2398 2399 static void 2400 spdk_nvmf_tcp_queue_c2h_data(struct spdk_nvmf_tcp_req *tcp_req, 2401 struct spdk_nvmf_tcp_qpair *tqpair) 2402 { 2403 tcp_req->c2h_data_pdu_num = spdk_nvmf_tcp_calc_c2h_data_pdu_num(tcp_req); 2404 2405 assert(tcp_req->c2h_data_pdu_num < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM); 2406 2407 STAILQ_INSERT_TAIL(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 2408 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 2409 } 2410 2411 static int 2412 request_transfer_out(struct spdk_nvmf_request *req) 2413 { 2414 struct spdk_nvmf_tcp_req *tcp_req; 2415 struct spdk_nvmf_qpair *qpair; 2416 struct spdk_nvmf_tcp_qpair *tqpair; 2417 struct spdk_nvme_cpl *rsp; 2418 2419 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2420 2421 qpair = req->qpair; 2422 rsp = &req->rsp->nvme_cpl; 2423 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2424 2425 /* Advance our sq_head pointer */ 2426 if (qpair->sq_head == qpair->sq_head_max) { 2427 qpair->sq_head = 0; 2428 } else { 2429 qpair->sq_head++; 2430 } 2431 rsp->sqhd = qpair->sq_head; 2432 2433 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2434 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2435 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && 2436 req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2437 spdk_nvmf_tcp_queue_c2h_data(tcp_req, tqpair); 2438 } else { 2439 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2440 } 2441 2442 return 0; 2443 } 2444 2445 static void 2446 spdk_nvmf_tcp_pdu_set_buf_from_req(struct spdk_nvmf_tcp_qpair *tqpair, 2447 struct spdk_nvmf_tcp_req *tcp_req) 2448 { 2449 struct nvme_tcp_pdu *pdu; 2450 2451 if (tcp_req->req.data_from_pool) { 2452 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Will send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2453 tcp_req->next_expected_r2t_offset = 0; 2454 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2455 } else { 2456 pdu = &tqpair->pdu_in_progress; 2457 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2458 tqpair); 2459 /* No need to send r2t, contained in the capsuled data */ 2460 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2461 0, tcp_req->req.length); 2462 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2463 } 2464 } 2465 2466 static void 2467 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2468 struct spdk_nvmf_tcp_req *tcp_req) 2469 { 2470 struct nvme_tcp_pdu *pdu; 2471 uint32_t plen = 0; 2472 2473 pdu = &tqpair->pdu_in_progress; 2474 plen = pdu->hdr->common.hlen; 2475 2476 if (tqpair->host_hdgst_enable) { 2477 plen += SPDK_NVME_TCP_DIGEST_LEN; 2478 } 2479 2480 if (pdu->hdr->common.plen != plen) { 2481 tcp_req->has_incapsule_data = true; 2482 } 2483 } 2484 2485 static bool 2486 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2487 struct spdk_nvmf_tcp_req *tcp_req) 2488 { 2489 struct spdk_nvmf_tcp_qpair *tqpair; 2490 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2491 int rc; 2492 enum spdk_nvmf_tcp_req_state prev_state; 2493 bool progress = false; 2494 struct spdk_nvmf_transport *transport = &ttransport->transport; 2495 struct spdk_nvmf_transport_poll_group *group; 2496 2497 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2498 group = &tqpair->group->group; 2499 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2500 2501 /* The loop here is to allow for several back-to-back state changes. */ 2502 do { 2503 prev_state = tcp_req->state; 2504 2505 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2506 tqpair); 2507 2508 switch (tcp_req->state) { 2509 case TCP_REQUEST_STATE_FREE: 2510 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2511 * to escape this state. */ 2512 break; 2513 case TCP_REQUEST_STATE_NEW: 2514 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2515 2516 /* copy the cmd from the receive pdu */ 2517 tcp_req->cmd = tqpair->pdu_in_progress.hdr->capsule_cmd.ccsqe; 2518 2519 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2520 tcp_req->req.dif.dif_insert_or_strip = true; 2521 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2522 } 2523 2524 /* The next state transition depends on the data transfer needs of this request. */ 2525 tcp_req->req.xfer = spdk_nvmf_tcp_req_get_xfer(tcp_req); 2526 2527 /* If no data to transfer, ready to execute. */ 2528 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2529 /* Reset the tqpair receving pdu state */ 2530 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2531 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2532 break; 2533 } 2534 2535 spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2536 2537 if (!tcp_req->has_incapsule_data) { 2538 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2539 } 2540 2541 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2542 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2543 break; 2544 case TCP_REQUEST_STATE_NEED_BUFFER: 2545 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2546 2547 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2548 2549 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2550 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2551 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2552 tcp_req, tqpair); 2553 /* This request needs to wait in line to obtain a buffer */ 2554 break; 2555 } 2556 2557 /* Try to get a data buffer */ 2558 rc = spdk_nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2559 if (rc < 0) { 2560 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2561 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2562 /* Reset the tqpair receving pdu state */ 2563 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2564 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2565 break; 2566 } 2567 2568 if (!tcp_req->req.data) { 2569 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2570 tcp_req, tqpair); 2571 /* No buffers available. */ 2572 break; 2573 } 2574 2575 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2576 2577 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2578 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2579 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2580 spdk_nvmf_tcp_pdu_set_buf_from_req(tqpair, tcp_req); 2581 break; 2582 } 2583 2584 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2585 break; 2586 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2587 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2588 (uintptr_t)tcp_req, 0); 2589 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2590 * to escape this state. */ 2591 break; 2592 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2593 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2594 2595 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2596 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2597 tcp_req->req.length = tcp_req->req.dif.elba_length; 2598 } 2599 2600 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2601 spdk_nvmf_request_exec(&tcp_req->req); 2602 break; 2603 case TCP_REQUEST_STATE_EXECUTING: 2604 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2605 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2606 * to escape this state. */ 2607 break; 2608 case TCP_REQUEST_STATE_EXECUTED: 2609 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2610 2611 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2612 tcp_req->req.length = tcp_req->req.dif.orig_length; 2613 } 2614 2615 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2616 break; 2617 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2618 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2619 rc = request_transfer_out(&tcp_req->req); 2620 assert(rc == 0); /* No good way to handle this currently */ 2621 break; 2622 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2623 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2624 (uintptr_t)tcp_req, 2625 0); 2626 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2627 * to escape this state. */ 2628 break; 2629 case TCP_REQUEST_STATE_COMPLETED: 2630 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2631 if (tcp_req->req.data_from_pool) { 2632 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2633 } 2634 tcp_req->req.length = 0; 2635 tcp_req->req.iovcnt = 0; 2636 tcp_req->req.data = NULL; 2637 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2638 break; 2639 case TCP_REQUEST_NUM_STATES: 2640 default: 2641 assert(0); 2642 break; 2643 } 2644 2645 if (tcp_req->state != prev_state) { 2646 progress = true; 2647 } 2648 } while (tcp_req->state != prev_state); 2649 2650 return progress; 2651 } 2652 2653 static void 2654 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2655 { 2656 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2657 int rc; 2658 2659 assert(tqpair != NULL); 2660 rc = spdk_nvmf_tcp_sock_process(tqpair); 2661 2662 /* check the following two factors: 2663 * rc: The socket is closed 2664 * State of tqpair: The tqpair is in EXITING state due to internal error 2665 */ 2666 if ((rc < 0) || (tqpair->state == NVME_TCP_QPAIR_STATE_EXITING)) { 2667 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2668 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 2669 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair); 2670 spdk_poller_unregister(&tqpair->timeout_poller); 2671 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 2672 } 2673 } 2674 2675 static int 2676 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2677 struct spdk_nvmf_qpair *qpair) 2678 { 2679 struct spdk_nvmf_tcp_poll_group *tgroup; 2680 struct spdk_nvmf_tcp_qpair *tqpair; 2681 int rc; 2682 2683 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2684 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2685 2686 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2687 spdk_nvmf_tcp_sock_cb, tqpair); 2688 if (rc != 0) { 2689 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2690 spdk_strerror(errno), errno); 2691 spdk_nvmf_tcp_qpair_destroy(tqpair); 2692 return -1; 2693 } 2694 2695 rc = spdk_nvmf_tcp_qpair_sock_init(tqpair); 2696 if (rc != 0) { 2697 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2698 spdk_nvmf_tcp_qpair_destroy(tqpair); 2699 return -1; 2700 } 2701 2702 rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair); 2703 if (rc < 0) { 2704 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2705 spdk_nvmf_tcp_qpair_destroy(tqpair); 2706 return -1; 2707 } 2708 2709 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, 1); 2710 if (rc < 0) { 2711 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2712 spdk_nvmf_tcp_qpair_destroy(tqpair); 2713 return -1; 2714 } 2715 2716 tqpair->group = tgroup; 2717 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2718 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2719 2720 return 0; 2721 } 2722 2723 static int 2724 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2725 struct spdk_nvmf_qpair *qpair) 2726 { 2727 struct spdk_nvmf_tcp_poll_group *tgroup; 2728 struct spdk_nvmf_tcp_qpair *tqpair; 2729 int rc; 2730 2731 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2732 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2733 2734 assert(tqpair->group == tgroup); 2735 2736 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2737 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2738 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2739 if (rc != 0) { 2740 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2741 spdk_strerror(errno), errno); 2742 } 2743 2744 return rc; 2745 } 2746 2747 static int 2748 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2749 { 2750 struct spdk_nvmf_tcp_transport *ttransport; 2751 struct spdk_nvmf_tcp_req *tcp_req; 2752 2753 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2754 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2755 2756 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2757 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 2758 2759 return 0; 2760 } 2761 2762 static void 2763 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2764 { 2765 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2766 2767 spdk_nvmf_tcp_qpair_destroy(SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair)); 2768 } 2769 2770 static int 2771 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2772 { 2773 struct spdk_nvmf_tcp_poll_group *tgroup; 2774 int rc; 2775 struct spdk_nvmf_request *req, *req_tmp; 2776 struct spdk_nvmf_tcp_req *tcp_req; 2777 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2778 struct spdk_nvmf_tcp_transport, transport); 2779 2780 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2781 2782 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs))) { 2783 return 0; 2784 } 2785 2786 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2787 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2788 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2789 break; 2790 } 2791 } 2792 2793 rc = spdk_sock_group_poll(tgroup->sock_group); 2794 if (rc < 0) { 2795 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2796 } 2797 2798 return rc; 2799 } 2800 2801 static int 2802 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2803 struct spdk_nvme_transport_id *trid, bool peer) 2804 { 2805 struct spdk_nvmf_tcp_qpair *tqpair; 2806 uint16_t port; 2807 2808 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2809 trid->trtype = SPDK_NVME_TRANSPORT_TCP; 2810 2811 if (peer) { 2812 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2813 port = tqpair->initiator_port; 2814 } else { 2815 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2816 port = tqpair->target_port; 2817 } 2818 2819 if (spdk_sock_is_ipv4(tqpair->sock)) { 2820 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2821 } else if (spdk_sock_is_ipv4(tqpair->sock)) { 2822 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2823 } else { 2824 return -1; 2825 } 2826 2827 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2828 return 0; 2829 } 2830 2831 static int 2832 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2833 struct spdk_nvme_transport_id *trid) 2834 { 2835 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2836 } 2837 2838 static int 2839 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2840 struct spdk_nvme_transport_id *trid) 2841 { 2842 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2843 } 2844 2845 static int 2846 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2847 struct spdk_nvme_transport_id *trid) 2848 { 2849 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2850 } 2851 2852 static int 2853 spdk_nvmf_tcp_qpair_set_sq_size(struct spdk_nvmf_qpair *qpair) 2854 { 2855 struct spdk_nvmf_tcp_qpair *tqpair; 2856 int rc; 2857 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2858 2859 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, tqpair->qpair.sq_head_max); 2860 if (!rc) { 2861 tqpair->max_queue_depth += tqpair->qpair.sq_head_max; 2862 tqpair->free_pdu_num += tqpair->qpair.sq_head_max; 2863 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] += tqpair->qpair.sq_head_max; 2864 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The queue depth=%u for tqpair=%p\n", 2865 tqpair->max_queue_depth, tqpair); 2866 } 2867 2868 return rc; 2869 2870 } 2871 2872 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2873 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2874 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2875 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2876 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2877 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2878 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2879 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2880 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 2881 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2882 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 2883 2884 static void 2885 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2886 { 2887 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2888 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2889 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2890 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2891 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2892 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2893 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2894 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2895 opts->c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 2896 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2897 opts->sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 2898 } 2899 2900 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2901 .type = SPDK_NVME_TRANSPORT_TCP, 2902 .opts_init = spdk_nvmf_tcp_opts_init, 2903 .create = spdk_nvmf_tcp_create, 2904 .destroy = spdk_nvmf_tcp_destroy, 2905 2906 .listen = spdk_nvmf_tcp_listen, 2907 .stop_listen = spdk_nvmf_tcp_stop_listen, 2908 .accept = spdk_nvmf_tcp_accept, 2909 2910 .listener_discover = spdk_nvmf_tcp_discover, 2911 2912 .poll_group_create = spdk_nvmf_tcp_poll_group_create, 2913 .get_optimal_poll_group = spdk_nvmf_tcp_get_optimal_poll_group, 2914 .poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy, 2915 .poll_group_add = spdk_nvmf_tcp_poll_group_add, 2916 .poll_group_remove = spdk_nvmf_tcp_poll_group_remove, 2917 .poll_group_poll = spdk_nvmf_tcp_poll_group_poll, 2918 2919 .req_free = spdk_nvmf_tcp_req_free, 2920 .req_complete = spdk_nvmf_tcp_req_complete, 2921 2922 .qpair_fini = spdk_nvmf_tcp_close_qpair, 2923 .qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid, 2924 .qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid, 2925 .qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid, 2926 .qpair_set_sqsize = spdk_nvmf_tcp_qpair_set_sq_size, 2927 }; 2928 2929 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2930