1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * 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/log.h" 50 #include "spdk_internal/nvme_tcp.h" 51 52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 54 #define NVMF_TCP_PDU_MAX_H2C_DATA_SIZE 131072 55 #define NVMF_TCP_PDU_MAX_C2H_DATA_SIZE 131072 56 #define NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM 64 /* Maximal c2h_data pdu number for ecah tqpair */ 57 58 /* This is used to support the Linux kernel NVMe-oF initiator */ 59 #define LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 0 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 pending on r2t slots */ 74 TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 75 76 /* The request is currently transferring data from the host to the controller. */ 77 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 78 79 /* The request is ready to execute at the block device */ 80 TCP_REQUEST_STATE_READY_TO_EXECUTE, 81 82 /* The request is currently executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTING, 84 85 /* The request finished executing at the block device */ 86 TCP_REQUEST_STATE_EXECUTED, 87 88 /* The request is ready to send a completion */ 89 TCP_REQUEST_STATE_READY_TO_COMPLETE, 90 91 /* The request is currently transferring final pdus from the controller to the host. */ 92 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 93 94 /* The request completed and can be marked free. */ 95 TCP_REQUEST_STATE_COMPLETED, 96 97 /* Terminator */ 98 TCP_REQUEST_NUM_STATES, 99 }; 100 101 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 102 "Invalid PDU Header Field", 103 "PDU Sequence Error", 104 "Header Digiest Error", 105 "Data Transfer Out of Range", 106 "R2T Limit Exceeded", 107 "Unsupported parameter", 108 }; 109 110 #define OBJECT_NVMF_TCP_IO 0x80 111 112 #define TRACE_GROUP_NVMF_TCP 0x5 113 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 114 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 115 #define TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 116 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 117 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 118 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 119 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 120 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 121 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 122 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 123 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 124 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", "", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 132 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", "", 133 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_TX_PENDING_R2T", "", 136 TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", "", 139 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", "", 142 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTING", "", 145 TRACE_TCP_REQUEST_STATE_EXECUTING, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_EXECUTED", "", 148 TRACE_TCP_REQUEST_STATE_EXECUTED, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", "", 151 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_REQ_COMPLETING_INCAPSULE", "", 154 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 156 spdk_trace_register_description("TCP_REQ_COMPLETED", "", 157 TRACE_TCP_REQUEST_STATE_COMPLETED, 158 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 159 spdk_trace_register_description("TCP_FLUSH_WRITEBUF_START", "", 160 TRACE_TCP_FLUSH_WRITEBUF_START, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 spdk_trace_register_description("TCP_FLUSH_WRITEBUF_DONE", "", 163 TRACE_TCP_FLUSH_WRITEBUF_DONE, 164 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 165 } 166 167 struct spdk_nvmf_tcp_req { 168 struct spdk_nvmf_request req; 169 struct spdk_nvme_cpl rsp; 170 struct spdk_nvme_cmd cmd; 171 172 /* In-capsule data buffer */ 173 uint8_t *buf; 174 175 bool data_from_pool; 176 void *buffers[SPDK_NVMF_MAX_SGL_ENTRIES]; 177 178 /* transfer_tag */ 179 uint16_t ttag; 180 181 /* 182 * next_expected_r2t_offset is used when we receive the h2c_data PDU. 183 */ 184 uint32_t next_expected_r2t_offset; 185 uint32_t r2tl_remain; 186 187 /* 188 * c2h_data_offset is used when we send the c2h_data PDU. 189 */ 190 uint32_t c2h_data_offset; 191 uint32_t c2h_data_pdu_num; 192 193 enum spdk_nvmf_tcp_req_state state; 194 bool has_incapsule_data; 195 196 TAILQ_ENTRY(spdk_nvmf_tcp_req) link; 197 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 198 }; 199 200 struct spdk_nvmf_tcp_qpair { 201 struct spdk_nvmf_qpair qpair; 202 struct spdk_nvmf_tcp_poll_group *group; 203 struct spdk_nvmf_tcp_port *port; 204 struct spdk_sock *sock; 205 struct spdk_poller *flush_poller; 206 207 enum nvme_tcp_pdu_recv_state recv_state; 208 enum nvme_tcp_qpair_state state; 209 210 struct nvme_tcp_pdu pdu_in_progress; 211 212 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 213 TAILQ_HEAD(, nvme_tcp_pdu) free_queue; 214 215 struct nvme_tcp_pdu *pdu; 216 struct nvme_tcp_pdu *pdu_pool; 217 uint16_t free_pdu_num; 218 219 /* Queues to track the requests in all states */ 220 TAILQ_HEAD(, spdk_nvmf_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 221 /* Number of requests in each state */ 222 int32_t state_cntr[TCP_REQUEST_NUM_STATES]; 223 224 uint32_t maxr2t; 225 uint32_t pending_r2t; 226 TAILQ_HEAD(, spdk_nvmf_tcp_req) queued_c2h_data_tcp_req; 227 228 uint8_t cpda; 229 230 /* Array of size "max_queue_depth * InCapsuleDataSize" containing 231 * buffers to be used for in capsule data. 232 */ 233 void *buf; 234 void *bufs; 235 struct spdk_nvmf_tcp_req *req; 236 struct spdk_nvmf_tcp_req *reqs; 237 238 bool host_hdgst_enable; 239 bool host_ddgst_enable; 240 241 242 /* The maximum number of I/O outstanding on this connection at one time */ 243 uint16_t max_queue_depth; 244 245 246 /** Specifies the maximum number of PDU-Data bytes per H2C Data Transfer PDU */ 247 uint32_t maxh2cdata; 248 249 uint32_t c2h_data_pdu_cnt; 250 251 /* IP address */ 252 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 253 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 254 255 /* IP port */ 256 uint16_t initiator_port; 257 uint16_t target_port; 258 259 /* Timer used to destroy qpair after detecting transport error issue if initiator does 260 * not close the connection. 261 */ 262 struct spdk_poller *timeout_poller; 263 264 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 265 }; 266 267 struct spdk_nvmf_tcp_poll_group { 268 struct spdk_nvmf_transport_poll_group group; 269 struct spdk_sock_group *sock_group; 270 271 /* Requests that are waiting to obtain a data buffer */ 272 TAILQ_HEAD(, spdk_nvmf_tcp_req) pending_data_buf_queue; 273 274 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 275 }; 276 277 struct spdk_nvmf_tcp_port { 278 struct spdk_nvme_transport_id trid; 279 struct spdk_sock *listen_sock; 280 uint32_t ref; 281 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 282 }; 283 284 struct spdk_nvmf_tcp_transport { 285 struct spdk_nvmf_transport transport; 286 287 pthread_mutex_t lock; 288 289 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 290 }; 291 292 static void spdk_nvmf_tcp_qpair_process_pending(struct spdk_nvmf_tcp_transport *ttransport, 293 struct spdk_nvmf_tcp_qpair *tqpair); 294 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 295 struct spdk_nvmf_tcp_req *tcp_req); 296 static void spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair); 297 298 static void 299 spdk_nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 300 enum spdk_nvmf_tcp_req_state state) 301 { 302 struct spdk_nvmf_qpair *qpair; 303 struct spdk_nvmf_tcp_qpair *tqpair; 304 305 qpair = tcp_req->req.qpair; 306 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 307 308 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 309 tqpair->state_cntr[tcp_req->state]--; 310 assert(tqpair->state_cntr[tcp_req->state] >= 0); 311 312 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 313 tqpair->state_cntr[state]++; 314 315 tcp_req->state = state; 316 } 317 318 static struct nvme_tcp_pdu * 319 spdk_nvmf_tcp_pdu_get(struct spdk_nvmf_tcp_qpair *tqpair) 320 { 321 struct nvme_tcp_pdu *pdu; 322 323 pdu = TAILQ_FIRST(&tqpair->free_queue); 324 if (!pdu) { 325 SPDK_ERRLOG("Unable to get PDU for tqpair=%p\n", tqpair); 326 abort(); 327 return NULL; 328 } 329 330 tqpair->free_pdu_num--; 331 TAILQ_REMOVE(&tqpair->free_queue, pdu, tailq); 332 memset(pdu, 0, sizeof(*pdu)); 333 pdu->ref = 1; 334 335 return pdu; 336 } 337 338 static void 339 spdk_nvmf_tcp_pdu_put(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 340 { 341 if (!pdu) { 342 return; 343 } 344 345 assert(pdu->ref > 0); 346 347 pdu->ref--; 348 if (pdu->ref == 0) { 349 tqpair->free_pdu_num++; 350 TAILQ_INSERT_HEAD(&tqpair->free_queue, pdu, tailq); 351 } 352 } 353 354 static struct spdk_nvmf_tcp_req * 355 spdk_nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 356 { 357 struct spdk_nvmf_tcp_req *tcp_req; 358 359 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 360 if (!tcp_req) { 361 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 362 return NULL; 363 } 364 365 memset(&tcp_req->cmd, 0, sizeof(tcp_req->cmd)); 366 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 367 tcp_req->next_expected_r2t_offset = 0; 368 tcp_req->r2tl_remain = 0; 369 tcp_req->c2h_data_offset = 0; 370 tcp_req->has_incapsule_data = false; 371 372 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 373 return tcp_req; 374 } 375 376 static void 377 nvmf_tcp_request_free(struct spdk_nvmf_tcp_req *tcp_req) 378 { 379 struct spdk_nvmf_tcp_transport *ttransport; 380 381 if (!tcp_req) { 382 return; 383 } 384 385 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req); 386 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 387 struct spdk_nvmf_tcp_transport, transport); 388 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 389 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 390 } 391 392 static int 393 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req) 394 { 395 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 396 397 nvmf_tcp_request_free(tcp_req); 398 399 return 0; 400 } 401 402 static void 403 spdk_nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 404 enum spdk_nvmf_tcp_req_state state) 405 { 406 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 407 408 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 409 nvmf_tcp_request_free(tcp_req); 410 } 411 } 412 413 static void 414 spdk_nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 415 { 416 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 417 struct nvme_tcp_pdu *pdu, *tmp_pdu; 418 419 /* Free the pdus in the send_queue */ 420 TAILQ_FOREACH_SAFE(pdu, &tqpair->send_queue, tailq, tmp_pdu) { 421 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 422 /* Also check the pdu type, we need to calculte the c2h_data_pdu_cnt later */ 423 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_DATA) { 424 assert(tqpair->c2h_data_pdu_cnt > 0); 425 tqpair->c2h_data_pdu_cnt--; 426 } 427 spdk_nvmf_tcp_pdu_put(tqpair, pdu); 428 } 429 430 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->queued_c2h_data_tcp_req, link, req_tmp) { 431 TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 432 } 433 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 434 435 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 436 437 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T); 438 439 /* Wipe the requests waiting for buffer from the global list */ 440 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], state_link, 441 req_tmp) { 442 TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link); 443 } 444 445 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 446 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 447 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 448 } 449 450 static void 451 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 452 { 453 int i; 454 struct spdk_nvmf_tcp_req *tcp_req; 455 456 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 457 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 458 SPDK_ERRLOG("\tNum of requests in state[%d] = %d\n", i, tqpair->state_cntr[i]); 459 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[i], state_link) { 460 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->data_from_pool); 461 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 462 } 463 } 464 } 465 466 static void 467 spdk_nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 468 { 469 int err = 0; 470 471 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 472 473 spdk_poller_unregister(&tqpair->flush_poller); 474 spdk_sock_close(&tqpair->sock); 475 spdk_nvmf_tcp_cleanup_all_states(tqpair); 476 477 if (tqpair->free_pdu_num != (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 478 SPDK_ERRLOG("tqpair(%p) free pdu pool num is %u but should be %u\n", tqpair, 479 tqpair->free_pdu_num, 480 (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)); 481 err++; 482 } 483 484 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->max_queue_depth) { 485 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 486 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 487 tqpair->max_queue_depth); 488 err++; 489 } 490 491 if (tqpair->c2h_data_pdu_cnt != 0) { 492 SPDK_ERRLOG("tqpair(%p) free c2h_data_pdu cnt is %u but should be 0\n", tqpair, 493 tqpair->c2h_data_pdu_cnt); 494 err++; 495 } 496 497 if (err > 0) { 498 nvmf_tcp_dump_qpair_req_contents(tqpair); 499 } 500 free(tqpair->pdu); 501 free(tqpair->pdu_pool); 502 free(tqpair->req); 503 free(tqpair->reqs); 504 spdk_dma_free(tqpair->buf); 505 spdk_dma_free(tqpair->bufs); 506 free(tqpair); 507 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n"); 508 } 509 510 static int 511 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 512 { 513 struct spdk_nvmf_tcp_transport *ttransport; 514 515 assert(transport != NULL); 516 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 517 518 pthread_mutex_destroy(&ttransport->lock); 519 free(ttransport); 520 return 0; 521 } 522 523 static struct spdk_nvmf_transport * 524 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 525 { 526 struct spdk_nvmf_tcp_transport *ttransport; 527 uint32_t sge_count; 528 uint32_t min_shared_buffers; 529 530 ttransport = calloc(1, sizeof(*ttransport)); 531 if (!ttransport) { 532 return NULL; 533 } 534 535 TAILQ_INIT(&ttransport->ports); 536 537 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 538 539 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 540 541 SPDK_INFOLOG(SPDK_LOG_NVMF_TCP, "*** TCP Transport Init ***\n" 542 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 543 " max_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 544 " in_capsule_data_size=%d, max_aq_depth=%d\n" 545 " num_shared_buffers=%d\n", 546 opts->max_queue_depth, 547 opts->max_io_size, 548 opts->max_qpairs_per_ctrlr, 549 opts->io_unit_size, 550 opts->in_capsule_data_size, 551 opts->max_aq_depth, 552 opts->num_shared_buffers); 553 554 /* I/O unit size cannot be larger than max I/O size */ 555 if (opts->io_unit_size > opts->max_io_size) { 556 opts->io_unit_size = opts->max_io_size; 557 } 558 559 sge_count = opts->max_io_size / opts->io_unit_size; 560 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 561 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 562 free(ttransport); 563 return NULL; 564 } 565 566 min_shared_buffers = spdk_thread_get_count() * opts->buf_cache_size; 567 if (min_shared_buffers > opts->num_shared_buffers) { 568 SPDK_ERRLOG("There are not enough buffers to satisfy" 569 "per-poll group caches for each thread. (%" PRIu32 ")" 570 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 571 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 572 spdk_nvmf_tcp_destroy(&ttransport->transport); 573 return NULL; 574 } 575 576 pthread_mutex_init(&ttransport->lock, NULL); 577 578 return &ttransport->transport; 579 } 580 581 static int 582 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid) 583 { 584 unsigned long long ull; 585 char *end = NULL; 586 587 ull = strtoull(trsvcid, &end, 10); 588 if (end == NULL || end == trsvcid || *end != '\0') { 589 return -1; 590 } 591 592 /* Valid TCP/IP port numbers are in [0, 65535] */ 593 if (ull > 65535) { 594 return -1; 595 } 596 597 return (int)ull; 598 } 599 600 /** 601 * Canonicalize a listen address trid. 602 */ 603 static int 604 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 605 const struct spdk_nvme_transport_id *trid) 606 { 607 int trsvcid_int; 608 609 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 610 if (trsvcid_int < 0) { 611 return -EINVAL; 612 } 613 614 memset(canon_trid, 0, sizeof(*canon_trid)); 615 canon_trid->trtype = SPDK_NVME_TRANSPORT_TCP; 616 canon_trid->adrfam = trid->adrfam; 617 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 618 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 619 620 return 0; 621 } 622 623 /** 624 * Find an existing listening port. 625 * 626 * Caller must hold ttransport->lock. 627 */ 628 static struct spdk_nvmf_tcp_port * 629 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 630 const struct spdk_nvme_transport_id *trid) 631 { 632 struct spdk_nvme_transport_id canon_trid; 633 struct spdk_nvmf_tcp_port *port; 634 635 if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 636 return NULL; 637 } 638 639 TAILQ_FOREACH(port, &ttransport->ports, link) { 640 if (spdk_nvme_transport_id_compare(&canon_trid, &port->trid) == 0) { 641 return port; 642 } 643 } 644 645 return NULL; 646 } 647 648 static int 649 spdk_nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 650 const struct spdk_nvme_transport_id *trid) 651 { 652 struct spdk_nvmf_tcp_transport *ttransport; 653 struct spdk_nvmf_tcp_port *port; 654 int trsvcid_int; 655 uint8_t adrfam; 656 657 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 658 659 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 660 if (trsvcid_int < 0) { 661 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 662 return -EINVAL; 663 } 664 665 pthread_mutex_lock(&ttransport->lock); 666 667 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 668 if (port) { 669 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Already listening on %s port %s\n", 670 trid->traddr, trid->trsvcid); 671 port->ref++; 672 pthread_mutex_unlock(&ttransport->lock); 673 return 0; 674 } 675 676 port = calloc(1, sizeof(*port)); 677 if (!port) { 678 SPDK_ERRLOG("Port allocation failed\n"); 679 free(port); 680 pthread_mutex_unlock(&ttransport->lock); 681 return -ENOMEM; 682 } 683 684 port->ref = 1; 685 686 if (_spdk_nvmf_tcp_canon_listen_trid(&port->trid, trid) != 0) { 687 SPDK_ERRLOG("Invalid traddr %s / trsvcid %s\n", 688 trid->traddr, trid->trsvcid); 689 free(port); 690 pthread_mutex_unlock(&ttransport->lock); 691 return -ENOMEM; 692 } 693 694 port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int); 695 if (port->listen_sock == NULL) { 696 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 697 trid->traddr, trsvcid_int, 698 spdk_strerror(errno), errno); 699 free(port); 700 pthread_mutex_unlock(&ttransport->lock); 701 return -errno; 702 } 703 704 if (spdk_sock_is_ipv4(port->listen_sock)) { 705 adrfam = SPDK_NVMF_ADRFAM_IPV4; 706 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 707 adrfam = SPDK_NVMF_ADRFAM_IPV6; 708 } else { 709 SPDK_ERRLOG("Unhandled socket type\n"); 710 adrfam = 0; 711 } 712 713 if (adrfam != trid->adrfam) { 714 SPDK_ERRLOG("Socket address family mismatch\n"); 715 spdk_sock_close(&port->listen_sock); 716 free(port); 717 pthread_mutex_unlock(&ttransport->lock); 718 return -EINVAL; 719 } 720 721 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %d ***\n", 722 trid->traddr, trsvcid_int); 723 724 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 725 pthread_mutex_unlock(&ttransport->lock); 726 727 return 0; 728 } 729 730 static int 731 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 732 const struct spdk_nvme_transport_id *trid) 733 { 734 struct spdk_nvmf_tcp_transport *ttransport; 735 struct spdk_nvmf_tcp_port *port; 736 int rc; 737 738 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 739 740 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 741 trid->traddr, trid->trsvcid); 742 743 pthread_mutex_lock(&ttransport->lock); 744 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 745 if (port) { 746 assert(port->ref > 0); 747 port->ref--; 748 if (port->ref == 0) { 749 TAILQ_REMOVE(&ttransport->ports, port, link); 750 spdk_sock_close(&port->listen_sock); 751 free(port); 752 } 753 rc = 0; 754 } else { 755 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Port not found\n"); 756 rc = -ENOENT; 757 } 758 pthread_mutex_unlock(&ttransport->lock); 759 760 return rc; 761 } 762 763 static int 764 spdk_nvmf_tcp_qpair_flush_pdus_internal(struct spdk_nvmf_tcp_qpair *tqpair) 765 { 766 const int array_size = 32; 767 struct iovec iovec_array[array_size]; 768 struct iovec *iov = iovec_array; 769 int iovec_cnt = 0; 770 int bytes = 0; 771 int total_length = 0; 772 uint32_t mapped_length; 773 struct nvme_tcp_pdu *pdu; 774 int pdu_length; 775 TAILQ_HEAD(, nvme_tcp_pdu) completed_pdus_list; 776 struct spdk_nvmf_tcp_transport *ttransport; 777 778 pdu = TAILQ_FIRST(&tqpair->send_queue); 779 780 if (pdu == NULL) { 781 return 0; 782 } 783 784 /* 785 * Build up a list of iovecs for the first few PDUs in the 786 * tqpair 's send_queue. 787 */ 788 while (pdu != NULL && ((array_size - iovec_cnt) >= 3)) { 789 iovec_cnt += nvme_tcp_build_iovecs(&iovec_array[iovec_cnt], 790 array_size - iovec_cnt, 791 pdu, 792 tqpair->host_hdgst_enable, 793 tqpair->host_ddgst_enable, 794 &mapped_length); 795 total_length += mapped_length; 796 pdu = TAILQ_NEXT(pdu, tailq); 797 } 798 799 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_START, 0, total_length, 0, iovec_cnt); 800 801 bytes = spdk_sock_writev(tqpair->sock, iov, iovec_cnt); 802 if (bytes == -1) { 803 if (errno == EWOULDBLOCK || errno == EAGAIN) { 804 return 1; 805 } else { 806 SPDK_ERRLOG("spdk_sock_writev() failed, errno %d: %s\n", 807 errno, spdk_strerror(errno)); 808 return -1; 809 } 810 } 811 812 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_DONE, 0, bytes, 0, 0); 813 814 pdu = TAILQ_FIRST(&tqpair->send_queue); 815 816 /* 817 * Free any PDUs that were fully written. If a PDU was only 818 * partially written, update its writev_offset so that next 819 * time only the unwritten portion will be sent to writev(). 820 */ 821 TAILQ_INIT(&completed_pdus_list); 822 while (bytes > 0) { 823 pdu_length = pdu->hdr.common.plen - pdu->writev_offset; 824 if (bytes >= pdu_length) { 825 bytes -= pdu_length; 826 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 827 TAILQ_INSERT_TAIL(&completed_pdus_list, pdu, tailq); 828 pdu = TAILQ_FIRST(&tqpair->send_queue); 829 830 } else { 831 pdu->writev_offset += bytes; 832 bytes = 0; 833 } 834 } 835 836 while (!TAILQ_EMPTY(&completed_pdus_list)) { 837 pdu = TAILQ_FIRST(&completed_pdus_list); 838 TAILQ_REMOVE(&completed_pdus_list, pdu, tailq); 839 assert(pdu->cb_fn != NULL); 840 pdu->cb_fn(pdu->cb_arg); 841 spdk_nvmf_tcp_pdu_put(tqpair, pdu); 842 } 843 844 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 845 spdk_nvmf_tcp_qpair_process_pending(ttransport, tqpair); 846 847 return TAILQ_EMPTY(&tqpair->send_queue) ? 0 : 1; 848 } 849 850 static int 851 spdk_nvmf_tcp_qpair_flush_pdus(void *_tqpair) 852 { 853 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 854 int rc; 855 856 if (tqpair->state == NVME_TCP_QPAIR_STATE_RUNNING) { 857 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 858 if (rc == 0 && tqpair->flush_poller != NULL) { 859 spdk_poller_unregister(&tqpair->flush_poller); 860 } else if (rc == 1 && tqpair->flush_poller == NULL) { 861 tqpair->flush_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_flush_pdus, 862 tqpair, 50); 863 } 864 } else { 865 /* 866 * If the tqpair state is not RUNNING, then 867 * keep trying to flush PDUs until our list is 868 * empty - to make sure all data is sent before 869 * closing the connection. 870 */ 871 do { 872 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 873 } while (rc == 1); 874 } 875 876 if (rc < 0 && tqpair->state < NVME_TCP_QPAIR_STATE_EXITING) { 877 /* 878 * If the poller has already started destruction of the tqpair, 879 * i.e. the socket read failed, then the connection state may already 880 * be EXITED. We don't want to set it back to EXITING in that case. 881 */ 882 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 883 } 884 885 return -1; 886 } 887 888 static void 889 spdk_nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 890 struct nvme_tcp_pdu *pdu, 891 nvme_tcp_qpair_xfer_complete_cb cb_fn, 892 void *cb_arg) 893 { 894 int enable_digest; 895 int hlen; 896 uint32_t crc32c; 897 898 hlen = pdu->hdr.common.hlen; 899 enable_digest = 1; 900 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 901 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 902 /* this PDU should be sent without digest */ 903 enable_digest = 0; 904 } 905 906 /* Header Digest */ 907 if (enable_digest && tqpair->host_hdgst_enable) { 908 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 909 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 910 } 911 912 /* Data Digest */ 913 if (pdu->data_len > 0 && enable_digest && tqpair->host_ddgst_enable) { 914 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 915 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 916 } 917 918 pdu->cb_fn = cb_fn; 919 pdu->cb_arg = cb_arg; 920 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 921 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 922 } 923 924 static int 925 spdk_nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair, uint16_t size) 926 { 927 int i; 928 struct spdk_nvmf_tcp_req *tcp_req; 929 struct spdk_nvmf_transport *transport = tqpair->qpair.transport; 930 struct spdk_nvmf_tcp_transport *ttransport; 931 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 932 933 if (!tqpair->qpair.sq_head_max) { 934 tqpair->req = calloc(1, sizeof(*tqpair->req)); 935 if (!tqpair->req) { 936 SPDK_ERRLOG("Unable to allocate req on tqpair=%p.\n", tqpair); 937 return -1; 938 } 939 940 if (transport->opts.in_capsule_data_size) { 941 tqpair->buf = spdk_dma_zmalloc(ttransport->transport.opts.in_capsule_data_size, 0x1000, NULL); 942 if (!tqpair->buf) { 943 SPDK_ERRLOG("Unable to allocate buf on tqpair=%p.\n", tqpair); 944 return -1; 945 } 946 } 947 948 tcp_req = tqpair->req; 949 tcp_req->ttag = 0; 950 tcp_req->req.qpair = &tqpair->qpair; 951 952 /* Set up memory to receive commands */ 953 if (tqpair->buf) { 954 tcp_req->buf = tqpair->buf; 955 } 956 957 /* Set the cmdn and rsp */ 958 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 959 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 960 961 /* Initialize request state to FREE */ 962 tcp_req->state = TCP_REQUEST_STATE_FREE; 963 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 964 965 tqpair->pdu = calloc(NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM + 1, sizeof(*tqpair->pdu)); 966 if (!tqpair->pdu) { 967 SPDK_ERRLOG("Unable to allocate pdu on tqpair=%p.\n", tqpair); 968 return -1; 969 } 970 971 for (i = 0; i < 1 + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; i++) { 972 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu[i], tailq); 973 } 974 975 } else { 976 tqpair->reqs = calloc(size, sizeof(*tqpair->reqs)); 977 if (!tqpair->reqs) { 978 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 979 return -1; 980 } 981 982 if (transport->opts.in_capsule_data_size) { 983 tqpair->bufs = spdk_dma_zmalloc(size * transport->opts.in_capsule_data_size, 984 0x1000, NULL); 985 if (!tqpair->bufs) { 986 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 987 return -1; 988 } 989 } 990 991 for (i = 0; i < size; i++) { 992 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 993 994 tcp_req->ttag = i + 1; 995 tcp_req->req.qpair = &tqpair->qpair; 996 997 /* Set up memory to receive commands */ 998 if (tqpair->bufs) { 999 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * transport->opts.in_capsule_data_size)); 1000 } 1001 1002 /* Set the cmdn and rsp */ 1003 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1004 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1005 1006 /* Initialize request state to FREE */ 1007 tcp_req->state = TCP_REQUEST_STATE_FREE; 1008 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 1009 } 1010 1011 tqpair->pdu_pool = calloc(size, sizeof(*tqpair->pdu_pool)); 1012 if (!tqpair->pdu_pool) { 1013 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1014 return -1; 1015 } 1016 1017 for (i = 0; i < size; i++) { 1018 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu_pool[i], tailq); 1019 } 1020 } 1021 1022 return 0; 1023 } 1024 1025 static int 1026 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1027 { 1028 struct spdk_nvmf_tcp_qpair *tqpair; 1029 int i; 1030 1031 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1032 1033 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 1034 1035 TAILQ_INIT(&tqpair->send_queue); 1036 TAILQ_INIT(&tqpair->free_queue); 1037 TAILQ_INIT(&tqpair->queued_c2h_data_tcp_req); 1038 1039 /* Initialise request state queues of the qpair */ 1040 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 1041 TAILQ_INIT(&tqpair->state_queue[i]); 1042 } 1043 1044 tqpair->host_hdgst_enable = true; 1045 tqpair->host_ddgst_enable = true; 1046 1047 return 0; 1048 } 1049 1050 static int 1051 spdk_nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1052 { 1053 1054 int rc; 1055 int buf_size; 1056 1057 /* set recv buffer size */ 1058 buf_size = 2 * 1024 * 1024; 1059 rc = spdk_sock_set_recvbuf(tqpair->sock, buf_size); 1060 if (rc != 0) { 1061 SPDK_ERRLOG("spdk_sock_set_recvbuf failed\n"); 1062 return rc; 1063 } 1064 1065 /* set send buffer size */ 1066 rc = spdk_sock_set_sendbuf(tqpair->sock, buf_size); 1067 if (rc != 0) { 1068 SPDK_ERRLOG("spdk_sock_set_sendbuf failed\n"); 1069 return rc; 1070 } 1071 1072 /* set low water mark */ 1073 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr)); 1074 if (rc != 0) { 1075 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1076 return rc; 1077 } 1078 1079 return 0; 1080 } 1081 1082 static void 1083 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1084 struct spdk_nvmf_tcp_port *port, 1085 struct spdk_sock *sock, new_qpair_fn cb_fn) 1086 { 1087 struct spdk_nvmf_tcp_qpair *tqpair; 1088 int rc; 1089 1090 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 1091 port->trid.traddr, port->trid.trsvcid); 1092 1093 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1094 if (tqpair == NULL) { 1095 SPDK_ERRLOG("Could not allocate new connection.\n"); 1096 spdk_sock_close(&sock); 1097 return; 1098 } 1099 1100 tqpair->sock = sock; 1101 tqpair->max_queue_depth = 1; 1102 tqpair->free_pdu_num = tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; 1103 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = tqpair->max_queue_depth; 1104 tqpair->port = port; 1105 tqpair->qpair.transport = transport; 1106 1107 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1108 sizeof(tqpair->target_addr), &tqpair->target_port, 1109 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1110 &tqpair->initiator_port); 1111 if (rc < 0) { 1112 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1113 spdk_nvmf_tcp_qpair_destroy(tqpair); 1114 return; 1115 } 1116 1117 cb_fn(&tqpair->qpair); 1118 } 1119 1120 static void 1121 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port, 1122 new_qpair_fn cb_fn) 1123 { 1124 struct spdk_sock *sock; 1125 int i; 1126 1127 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1128 sock = spdk_sock_accept(port->listen_sock); 1129 if (sock) { 1130 _spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn); 1131 } 1132 } 1133 } 1134 1135 static void 1136 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn) 1137 { 1138 struct spdk_nvmf_tcp_transport *ttransport; 1139 struct spdk_nvmf_tcp_port *port; 1140 1141 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1142 1143 TAILQ_FOREACH(port, &ttransport->ports, link) { 1144 spdk_nvmf_tcp_port_accept(transport, port, cb_fn); 1145 } 1146 } 1147 1148 static void 1149 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1150 struct spdk_nvme_transport_id *trid, 1151 struct spdk_nvmf_discovery_log_page_entry *entry) 1152 { 1153 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1154 entry->adrfam = trid->adrfam; 1155 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_SPECIFIED; 1156 1157 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1158 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1159 1160 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1161 } 1162 1163 static struct spdk_nvmf_transport_poll_group * 1164 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1165 { 1166 struct spdk_nvmf_tcp_poll_group *tgroup; 1167 1168 tgroup = calloc(1, sizeof(*tgroup)); 1169 if (!tgroup) { 1170 return NULL; 1171 } 1172 1173 tgroup->sock_group = spdk_sock_group_create(); 1174 if (!tgroup->sock_group) { 1175 goto cleanup; 1176 } 1177 1178 TAILQ_INIT(&tgroup->qpairs); 1179 TAILQ_INIT(&tgroup->pending_data_buf_queue); 1180 1181 return &tgroup->group; 1182 1183 cleanup: 1184 free(tgroup); 1185 return NULL; 1186 } 1187 1188 static void 1189 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1190 { 1191 struct spdk_nvmf_tcp_poll_group *tgroup; 1192 1193 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1194 spdk_sock_group_close(&tgroup->sock_group); 1195 1196 if (!TAILQ_EMPTY(&tgroup->pending_data_buf_queue)) { 1197 SPDK_ERRLOG("Pending I/O list wasn't empty on poll group destruction\n"); 1198 } 1199 1200 free(tgroup); 1201 } 1202 1203 static void 1204 spdk_nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1205 enum nvme_tcp_pdu_recv_state state) 1206 { 1207 if (tqpair->recv_state == state) { 1208 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1209 tqpair, state); 1210 return; 1211 } 1212 1213 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1214 tqpair->recv_state = state; 1215 1216 switch (state) { 1217 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1218 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1219 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1220 break; 1221 case NVME_TCP_PDU_RECV_STATE_ERROR: 1222 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1223 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1224 break; 1225 default: 1226 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1227 abort(); 1228 break; 1229 } 1230 } 1231 1232 static int 1233 spdk_nvmf_tcp_qpair_handle_timeout(void *ctx) 1234 { 1235 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1236 1237 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1238 1239 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1240 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1241 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 1242 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair); 1243 spdk_poller_unregister(&tqpair->timeout_poller); 1244 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 1245 1246 return 0; 1247 } 1248 1249 static void 1250 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1251 { 1252 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1253 1254 if (!tqpair->timeout_poller) { 1255 tqpair->timeout_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_handle_timeout, tqpair, 1256 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1257 } 1258 } 1259 1260 static void 1261 spdk_nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1262 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1263 { 1264 struct nvme_tcp_pdu *rsp_pdu; 1265 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1266 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1267 uint32_t copy_len; 1268 1269 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1270 if (!rsp_pdu) { 1271 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1272 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1273 return; 1274 } 1275 1276 c2h_term_req = &rsp_pdu->hdr.term_req; 1277 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1278 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1279 1280 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1281 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1282 DSET32(&c2h_term_req->fei, error_offset); 1283 } 1284 1285 copy_len = pdu->hdr.common.hlen; 1286 if (copy_len > SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE) { 1287 copy_len = SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE; 1288 } 1289 1290 /* Copy the error info into the buffer */ 1291 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1292 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1293 1294 /* Contain the header of the wrong received pdu */ 1295 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1296 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1297 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair); 1298 } 1299 1300 static void 1301 spdk_nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1302 struct spdk_nvmf_tcp_qpair *tqpair, 1303 struct nvme_tcp_pdu *pdu) 1304 { 1305 struct spdk_nvmf_tcp_req *tcp_req; 1306 1307 tcp_req = spdk_nvmf_tcp_req_get(tqpair); 1308 if (!tcp_req) { 1309 SPDK_ERRLOG("Cannot allocate tcp_req\n"); 1310 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1311 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1312 return; 1313 } 1314 1315 pdu->ctx = tcp_req; 1316 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 1317 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1318 return; 1319 } 1320 1321 static void 1322 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1323 struct spdk_nvmf_tcp_qpair *tqpair, 1324 struct nvme_tcp_pdu *pdu) 1325 { 1326 struct spdk_nvmf_tcp_req *tcp_req; 1327 struct spdk_nvme_tcp_cmd *capsule_cmd; 1328 uint32_t error_offset = 0; 1329 enum spdk_nvme_tcp_term_req_fes fes; 1330 1331 capsule_cmd = &pdu->hdr.capsule_cmd; 1332 tcp_req = pdu->ctx; 1333 assert(tcp_req != NULL); 1334 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1335 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1336 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1337 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1338 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1339 goto err; 1340 } 1341 1342 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1343 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1344 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1345 1346 return; 1347 err: 1348 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1349 } 1350 1351 static void 1352 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1353 struct spdk_nvmf_tcp_qpair *tqpair, 1354 struct nvme_tcp_pdu *pdu) 1355 { 1356 struct spdk_nvmf_tcp_req *tcp_req; 1357 uint32_t error_offset = 0; 1358 enum spdk_nvme_tcp_term_req_fes fes = 0; 1359 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1360 uint32_t iov_index; 1361 bool ttag_offset_error = false; 1362 1363 h2c_data = &pdu->hdr.h2c_data; 1364 1365 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1366 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1367 1368 /* According to the information in the pdu to find the req */ 1369 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER], 1370 state_link) { 1371 if ((tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid) && (tcp_req->ttag == h2c_data->ttag)) { 1372 break; 1373 } 1374 1375 if (!ttag_offset_error && (tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid)) { 1376 ttag_offset_error = true; 1377 } 1378 } 1379 1380 if (!tcp_req) { 1381 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1382 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1383 if (!ttag_offset_error) { 1384 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1385 } else { 1386 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1387 } 1388 goto err; 1389 } 1390 1391 if (tcp_req->next_expected_r2t_offset != h2c_data->datao) { 1392 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1393 "tcp_req(%p), tqpair=%p, expected_r2t_offset=%u, but data offset =%u\n", 1394 tcp_req, tqpair, tcp_req->next_expected_r2t_offset, h2c_data->datao); 1395 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1396 goto err; 1397 } 1398 1399 if (h2c_data->datal > tqpair->maxh2cdata) { 1400 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req(%p), tqpair=%p, datao=%u execeeds maxh2cdata size=%u\n", 1401 tcp_req, tqpair, h2c_data->datao, tqpair->maxh2cdata); 1402 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1403 goto err; 1404 } 1405 1406 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1407 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1408 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1409 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1410 fes = SPDK_NVME_TCP_TERM_REQ_FES_R2T_LIMIT_EXCEEDED; 1411 goto err; 1412 } 1413 1414 pdu->ctx = tcp_req; 1415 iov_index = pdu->hdr.h2c_data.datao / ttransport->transport.opts.io_unit_size; 1416 nvme_tcp_pdu_set_data(pdu, tcp_req->req.iov[iov_index].iov_base + (pdu->hdr.h2c_data.datao % 1417 ttransport->transport.opts.io_unit_size), h2c_data->datal); 1418 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1419 return; 1420 1421 err: 1422 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1423 } 1424 1425 static void 1426 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1427 { 1428 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1429 nvmf_tcp_request_free(tcp_req); 1430 } 1431 1432 static void 1433 spdk_nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1434 struct spdk_nvmf_tcp_qpair *tqpair) 1435 { 1436 struct nvme_tcp_pdu *rsp_pdu; 1437 struct spdk_nvme_tcp_rsp *capsule_resp; 1438 1439 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1440 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1441 if (!rsp_pdu) { 1442 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1443 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1444 return; 1445 } 1446 1447 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1448 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1449 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1450 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1451 if (tqpair->host_hdgst_enable) { 1452 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1453 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1454 } 1455 1456 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req); 1457 } 1458 1459 static void 1460 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1461 { 1462 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1463 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1464 struct spdk_nvmf_tcp_qpair, qpair); 1465 1466 assert(tqpair != NULL); 1467 assert(tcp_req->c2h_data_pdu_num > 0); 1468 tcp_req->c2h_data_pdu_num--; 1469 if (!tcp_req->c2h_data_pdu_num) { 1470 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 1471 nvmf_tcp_request_free(tcp_req); 1472 #else 1473 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1474 #endif 1475 } 1476 1477 tqpair->c2h_data_pdu_cnt--; 1478 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 1479 } 1480 1481 static void 1482 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1483 struct spdk_nvmf_tcp_req *tcp_req) 1484 { 1485 struct nvme_tcp_pdu *rsp_pdu; 1486 struct spdk_nvme_tcp_r2t_hdr *r2t; 1487 1488 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1489 if (!rsp_pdu) { 1490 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1491 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1492 return; 1493 } 1494 1495 r2t = &rsp_pdu->hdr.r2t; 1496 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1497 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1498 1499 if (tqpair->host_hdgst_enable) { 1500 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1501 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1502 } 1503 1504 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1505 r2t->ttag = tcp_req->ttag; 1506 r2t->r2to = tcp_req->next_expected_r2t_offset; 1507 r2t->r2tl = spdk_min(tcp_req->req.length - tcp_req->next_expected_r2t_offset, tqpair->maxh2cdata); 1508 tcp_req->r2tl_remain = r2t->r2tl; 1509 1510 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1511 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1512 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1513 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, NULL); 1514 } 1515 1516 static void 1517 spdk_nvmf_tcp_handle_queued_r2t_req(struct spdk_nvmf_tcp_qpair *tqpair) 1518 { 1519 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 1520 1521 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T], 1522 state_link, req_tmp) { 1523 if (tqpair->pending_r2t < tqpair->maxr2t) { 1524 tqpair->pending_r2t++; 1525 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1526 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 1527 } else { 1528 break; 1529 } 1530 } 1531 } 1532 1533 static void 1534 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1535 struct spdk_nvmf_tcp_qpair *tqpair, 1536 struct nvme_tcp_pdu *pdu) 1537 { 1538 struct spdk_nvmf_tcp_req *tcp_req; 1539 1540 tcp_req = pdu->ctx; 1541 assert(tcp_req != NULL); 1542 1543 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1544 1545 tcp_req->next_expected_r2t_offset += pdu->data_len; 1546 tcp_req->r2tl_remain -= pdu->data_len; 1547 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1548 1549 if (!tcp_req->r2tl_remain) { 1550 if (tcp_req->next_expected_r2t_offset == tcp_req->req.length) { 1551 assert(tqpair->pending_r2t > 0); 1552 tqpair->pending_r2t--; 1553 assert(tqpair->pending_r2t < tqpair->maxr2t); 1554 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1555 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1556 1557 spdk_nvmf_tcp_handle_queued_r2t_req(tqpair); 1558 } else { 1559 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Send r2t pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 1560 spdk_nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 1561 } 1562 } 1563 } 1564 1565 static void 1566 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1567 { 1568 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1569 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1570 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1571 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1572 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1573 DGET32(h2c_term_req->fei)); 1574 } 1575 } 1576 1577 static void 1578 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1579 struct nvme_tcp_pdu *pdu) 1580 { 1581 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1582 uint32_t error_offset = 0; 1583 enum spdk_nvme_tcp_term_req_fes fes; 1584 1585 1586 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1587 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1588 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1589 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1590 goto end; 1591 } 1592 1593 /* set the data buffer */ 1594 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1595 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1596 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1597 return; 1598 end: 1599 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1600 return; 1601 } 1602 1603 static void 1604 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1605 struct nvme_tcp_pdu *pdu) 1606 { 1607 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1608 1609 spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1610 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1611 return; 1612 } 1613 1614 static void 1615 spdk_nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1616 { 1617 int rc = 0; 1618 struct nvme_tcp_pdu *pdu; 1619 uint32_t crc32c, error_offset = 0; 1620 enum spdk_nvme_tcp_term_req_fes fes; 1621 struct spdk_nvmf_tcp_transport *ttransport; 1622 1623 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1624 pdu = &tqpair->pdu_in_progress; 1625 1626 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1627 /* check data digest if need */ 1628 if (pdu->ddgst_enable) { 1629 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1630 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1631 if (rc == 0) { 1632 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1633 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1634 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1635 return; 1636 1637 } 1638 } 1639 1640 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1641 switch (pdu->hdr.common.pdu_type) { 1642 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1643 spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1644 break; 1645 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1646 spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1647 break; 1648 1649 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1650 spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1651 break; 1652 1653 default: 1654 /* The code should not go to here */ 1655 SPDK_ERRLOG("The code should not go to here\n"); 1656 break; 1657 } 1658 } 1659 1660 static void 1661 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg) 1662 { 1663 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1664 1665 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1666 } 1667 1668 static void 1669 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1670 struct spdk_nvmf_tcp_qpair *tqpair, 1671 struct nvme_tcp_pdu *pdu) 1672 { 1673 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1674 struct nvme_tcp_pdu *rsp_pdu; 1675 struct spdk_nvme_tcp_ic_resp *ic_resp; 1676 uint32_t error_offset = 0; 1677 enum spdk_nvme_tcp_term_req_fes fes; 1678 1679 /* Only PFV 0 is defined currently */ 1680 if (ic_req->pfv != 0) { 1681 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1682 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1683 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1684 goto end; 1685 } 1686 1687 /* MAXR2T is 0's based */ 1688 tqpair->maxr2t = ic_req->maxr2t + 1ull; 1689 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", tqpair->maxr2t); 1690 1691 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1692 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1693 1694 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1695 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1696 1697 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1698 if (!rsp_pdu) { 1699 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1700 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1701 return; 1702 } 1703 1704 ic_resp = &rsp_pdu->hdr.ic_resp; 1705 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1706 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1707 ic_resp->pfv = 0; 1708 ic_resp->cpda = tqpair->cpda; 1709 tqpair->maxh2cdata = spdk_min(NVMF_TCP_PDU_MAX_H2C_DATA_SIZE, 1710 ttransport->transport.opts.io_unit_size); 1711 ic_resp->maxh2cdata = tqpair->maxh2cdata; 1712 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1713 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1714 1715 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1716 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1717 1718 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair); 1719 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1720 return; 1721 end: 1722 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1723 return; 1724 } 1725 1726 static void 1727 spdk_nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1728 { 1729 struct nvme_tcp_pdu *pdu; 1730 int rc; 1731 uint32_t crc32c, error_offset = 0; 1732 enum spdk_nvme_tcp_term_req_fes fes; 1733 struct spdk_nvmf_tcp_transport *ttransport; 1734 1735 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1736 pdu = &tqpair->pdu_in_progress; 1737 1738 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1739 pdu->hdr.common.pdu_type); 1740 /* check header digest if needed */ 1741 if (pdu->has_hdgst) { 1742 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1743 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1744 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1745 if (rc == 0) { 1746 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1747 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1748 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1749 return; 1750 1751 } 1752 } 1753 1754 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1755 switch (pdu->hdr.common.pdu_type) { 1756 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1757 spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1758 break; 1759 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1760 spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1761 break; 1762 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1763 spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1764 break; 1765 1766 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1767 spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1768 break; 1769 1770 default: 1771 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1772 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1773 error_offset = 1; 1774 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1775 break; 1776 } 1777 } 1778 1779 static void 1780 spdk_nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1781 { 1782 struct nvme_tcp_pdu *pdu; 1783 uint32_t error_offset = 0; 1784 enum spdk_nvme_tcp_term_req_fes fes; 1785 uint8_t expected_hlen, pdo; 1786 bool plen_error = false, pdo_error = false; 1787 1788 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1789 pdu = &tqpair->pdu_in_progress; 1790 1791 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1792 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1793 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1794 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1795 goto err; 1796 } 1797 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1798 if (pdu->hdr.common.plen != expected_hlen) { 1799 plen_error = true; 1800 } 1801 } else { 1802 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1803 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1804 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1805 goto err; 1806 } 1807 1808 switch (pdu->hdr.common.pdu_type) { 1809 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1810 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1811 pdo = pdu->hdr.common.pdo; 1812 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1813 pdo_error = true; 1814 break; 1815 } 1816 1817 if (pdu->hdr.common.plen < expected_hlen) { 1818 plen_error = true; 1819 } 1820 break; 1821 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1822 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1823 pdo = pdu->hdr.common.pdo; 1824 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1825 pdo_error = true; 1826 break; 1827 } 1828 if (pdu->hdr.common.plen < expected_hlen) { 1829 plen_error = true; 1830 } 1831 break; 1832 1833 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1834 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1835 if ((pdu->hdr.common.plen <= expected_hlen) || 1836 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1837 plen_error = true; 1838 } 1839 break; 1840 1841 default: 1842 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1843 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1844 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1845 goto err; 1846 } 1847 } 1848 1849 if (pdu->hdr.common.hlen != expected_hlen) { 1850 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1851 pdu->hdr.common.pdu_type, 1852 expected_hlen, pdu->hdr.common.hlen, tqpair); 1853 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1854 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1855 goto err; 1856 } else if (pdo_error) { 1857 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1858 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1859 } else if (plen_error) { 1860 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1861 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1862 goto err; 1863 } else { 1864 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1865 return; 1866 } 1867 err: 1868 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1869 } 1870 1871 static int 1872 spdk_nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1873 { 1874 int rc = 0; 1875 struct nvme_tcp_pdu *pdu; 1876 enum nvme_tcp_pdu_recv_state prev_state; 1877 uint32_t data_len; 1878 uint8_t psh_len, pdo, hlen; 1879 int8_t padding_len; 1880 1881 /* The loop here is to allow for several back-to-back state changes. */ 1882 do { 1883 prev_state = tqpair->recv_state; 1884 1885 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1886 1887 switch (tqpair->recv_state) { 1888 /* Wait for the common header */ 1889 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1890 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1891 pdu = &tqpair->pdu_in_progress; 1892 1893 rc = nvme_tcp_read_data(tqpair->sock, 1894 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1895 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1896 if (rc < 0) { 1897 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair); 1898 return NVME_TCP_PDU_FATAL; 1899 } else if (rc > 0) { 1900 pdu->ch_valid_bytes += rc; 1901 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1902 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1903 } 1904 } 1905 1906 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1907 return NVME_TCP_PDU_IN_PROGRESS; 1908 } 1909 1910 /* The command header of this PDU has now been read from the socket. */ 1911 spdk_nvmf_tcp_pdu_ch_handle(tqpair); 1912 break; 1913 /* Wait for the pdu specific header */ 1914 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1915 pdu = &tqpair->pdu_in_progress; 1916 psh_len = hlen = pdu->hdr.common.hlen; 1917 /* Only capsule_cmd and h2c_data has header digest */ 1918 if (((pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD) || 1919 (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_H2C_DATA)) && 1920 tqpair->host_hdgst_enable) { 1921 pdu->has_hdgst = true; 1922 psh_len += SPDK_NVME_TCP_DIGEST_LEN; 1923 if (pdu->hdr.common.plen > psh_len) { 1924 pdo = pdu->hdr.common.pdo; 1925 padding_len = pdo - psh_len; 1926 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "padding length is =%d for pdu=%p on tqpair=%p\n", padding_len, 1927 pdu, tqpair); 1928 if (padding_len > 0) { 1929 psh_len = pdo; 1930 } 1931 } 1932 } 1933 1934 psh_len -= sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 1935 /* The following will read psh + hdgest (if possbile) + padding (if posssible) */ 1936 if (pdu->psh_valid_bytes < psh_len) { 1937 rc = nvme_tcp_read_data(tqpair->sock, 1938 psh_len - pdu->psh_valid_bytes, 1939 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1940 if (rc < 0) { 1941 return NVME_TCP_PDU_FATAL; 1942 } 1943 1944 pdu->psh_valid_bytes += rc; 1945 if (pdu->psh_valid_bytes < psh_len) { 1946 return NVME_TCP_PDU_IN_PROGRESS; 1947 } 1948 } 1949 1950 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1951 spdk_nvmf_tcp_pdu_psh_handle(tqpair); 1952 break; 1953 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1954 pdu = &tqpair->pdu_in_progress; 1955 1956 /* check whether the data is valid, if not we just return */ 1957 if (!pdu->data_len) { 1958 return NVME_TCP_PDU_IN_PROGRESS; 1959 } 1960 1961 data_len = pdu->data_len; 1962 /* data digest */ 1963 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1964 tqpair->host_ddgst_enable)) { 1965 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1966 pdu->ddgst_enable = true; 1967 } 1968 1969 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1970 if (rc < 0) { 1971 return NVME_TCP_PDU_IN_PROGRESS; 1972 } 1973 1974 pdu->readv_offset += rc; 1975 if (pdu->readv_offset < data_len) { 1976 return NVME_TCP_PDU_IN_PROGRESS; 1977 } 1978 1979 /* All of this PDU has now been read from the socket. */ 1980 spdk_nvmf_tcp_pdu_payload_handle(tqpair); 1981 break; 1982 case NVME_TCP_PDU_RECV_STATE_ERROR: 1983 pdu = &tqpair->pdu_in_progress; 1984 /* Check whether the connection is closed. Each time, we only read 1 byte every time */ 1985 rc = nvme_tcp_read_data(tqpair->sock, 1, (void *)&pdu->hdr.common); 1986 if (rc < 0) { 1987 return NVME_TCP_PDU_FATAL; 1988 } 1989 break; 1990 default: 1991 assert(0); 1992 SPDK_ERRLOG("code should not come to here"); 1993 break; 1994 } 1995 } while (tqpair->recv_state != prev_state); 1996 1997 return rc; 1998 } 1999 2000 static enum spdk_nvme_data_transfer 2001 spdk_nvmf_tcp_req_get_xfer(struct spdk_nvmf_tcp_req *tcp_req) { 2002 enum spdk_nvme_data_transfer xfer; 2003 struct spdk_nvme_cmd *cmd = &tcp_req->req.cmd->nvme_cmd; 2004 struct spdk_nvme_sgl_descriptor *sgl = &cmd->dptr.sgl1; 2005 2006 /* Figure out data transfer direction */ 2007 if (cmd->opc == SPDK_NVME_OPC_FABRIC) 2008 { 2009 xfer = spdk_nvme_opc_get_data_transfer(tcp_req->req.cmd->nvmf_cmd.fctype); 2010 } else 2011 { 2012 xfer = spdk_nvme_opc_get_data_transfer(cmd->opc); 2013 2014 /* Some admin commands are special cases */ 2015 if ((tcp_req->req.qpair->qid == 0) && 2016 ((cmd->opc == SPDK_NVME_OPC_GET_FEATURES) || 2017 (cmd->opc == SPDK_NVME_OPC_SET_FEATURES))) { 2018 switch (cmd->cdw10 & 0xff) { 2019 case SPDK_NVME_FEAT_LBA_RANGE_TYPE: 2020 case SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION: 2021 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 2022 break; 2023 default: 2024 xfer = SPDK_NVME_DATA_NONE; 2025 } 2026 } 2027 } 2028 2029 if (xfer == SPDK_NVME_DATA_NONE) 2030 { 2031 return xfer; 2032 } 2033 2034 /* Even for commands that may transfer data, they could have specified 0 length. 2035 * We want those to show up with xfer SPDK_NVME_DATA_NONE. 2036 */ 2037 switch (sgl->generic.type) 2038 { 2039 case SPDK_NVME_SGL_TYPE_DATA_BLOCK: 2040 case SPDK_NVME_SGL_TYPE_BIT_BUCKET: 2041 case SPDK_NVME_SGL_TYPE_SEGMENT: 2042 case SPDK_NVME_SGL_TYPE_LAST_SEGMENT: 2043 case SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK: 2044 if (sgl->unkeyed.length == 0) { 2045 xfer = SPDK_NVME_DATA_NONE; 2046 } 2047 break; 2048 case SPDK_NVME_SGL_TYPE_KEYED_DATA_BLOCK: 2049 if (sgl->keyed.length == 0) { 2050 xfer = SPDK_NVME_DATA_NONE; 2051 } 2052 break; 2053 } 2054 2055 return xfer; 2056 } 2057 2058 static void 2059 spdk_nvmf_tcp_request_free_buffers(struct spdk_nvmf_tcp_req *tcp_req, 2060 struct spdk_nvmf_transport_poll_group *group, struct spdk_nvmf_transport *transport) 2061 { 2062 for (uint32_t i = 0; i < tcp_req->req.iovcnt; i++) { 2063 assert(tcp_req->buffers[i] != NULL); 2064 if (group->buf_cache_count < group->buf_cache_size) { 2065 STAILQ_INSERT_HEAD(&group->buf_cache, 2066 (struct spdk_nvmf_transport_pg_cache_buf *)tcp_req->buffers[i], link); 2067 group->buf_cache_count++; 2068 } else { 2069 spdk_mempool_put(transport->data_buf_pool, tcp_req->buffers[i]); 2070 } 2071 tcp_req->req.iov[i].iov_base = NULL; 2072 tcp_req->buffers[i] = NULL; 2073 tcp_req->req.iov[i].iov_len = 0; 2074 } 2075 tcp_req->data_from_pool = false; 2076 } 2077 2078 static int 2079 spdk_nvmf_tcp_req_fill_iovs(struct spdk_nvmf_tcp_transport *ttransport, 2080 struct spdk_nvmf_tcp_req *tcp_req) 2081 { 2082 void *buf = NULL; 2083 uint32_t length = tcp_req->req.length; 2084 uint32_t i = 0; 2085 struct spdk_nvmf_tcp_qpair *tqpair; 2086 struct spdk_nvmf_transport_poll_group *group; 2087 2088 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2089 group = &tqpair->group->group; 2090 2091 tcp_req->req.iovcnt = 0; 2092 while (length) { 2093 if (!(STAILQ_EMPTY(&group->buf_cache))) { 2094 group->buf_cache_count--; 2095 buf = STAILQ_FIRST(&group->buf_cache); 2096 STAILQ_REMOVE_HEAD(&group->buf_cache, link); 2097 } else { 2098 buf = spdk_mempool_get(ttransport->transport.data_buf_pool); 2099 if (!buf) { 2100 goto nomem; 2101 } 2102 } 2103 2104 tcp_req->req.iov[i].iov_base = (void *)((uintptr_t)(buf + NVMF_DATA_BUFFER_MASK) & 2105 ~NVMF_DATA_BUFFER_MASK); 2106 tcp_req->req.iov[i].iov_len = spdk_min(length, ttransport->transport.opts.io_unit_size); 2107 tcp_req->req.iovcnt++; 2108 tcp_req->buffers[i] = buf; 2109 length -= tcp_req->req.iov[i].iov_len; 2110 i++; 2111 } 2112 2113 assert(tcp_req->req.iovcnt < SPDK_NVMF_MAX_SGL_ENTRIES); 2114 tcp_req->data_from_pool = true; 2115 return 0; 2116 2117 nomem: 2118 spdk_nvmf_tcp_request_free_buffers(tcp_req, group, &ttransport->transport); 2119 tcp_req->req.iovcnt = 0; 2120 return -ENOMEM; 2121 } 2122 2123 static int 2124 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_transport *ttransport, 2125 struct spdk_nvmf_tcp_req *tcp_req) 2126 { 2127 struct spdk_nvme_cmd *cmd; 2128 struct spdk_nvme_cpl *rsp; 2129 struct spdk_nvme_sgl_descriptor *sgl; 2130 2131 cmd = &tcp_req->req.cmd->nvme_cmd; 2132 rsp = &tcp_req->req.rsp->nvme_cpl; 2133 sgl = &cmd->dptr.sgl1; 2134 2135 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2136 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2137 if (sgl->unkeyed.length > ttransport->transport.opts.max_io_size) { 2138 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2139 sgl->unkeyed.length, ttransport->transport.opts.max_io_size); 2140 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2141 return -1; 2142 } 2143 2144 /* fill request length and populate iovs */ 2145 tcp_req->req.length = sgl->unkeyed.length; 2146 2147 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", 2148 sgl->unkeyed.length); 2149 2150 if (spdk_nvmf_tcp_req_fill_iovs(ttransport, tcp_req) < 0) { 2151 /* No available buffers. Queue this request up. */ 2152 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", tcp_req); 2153 return 0; 2154 } 2155 2156 /* backward compatible */ 2157 tcp_req->req.data = tcp_req->req.iov[0].iov_base; 2158 2159 2160 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 2161 tcp_req, 2162 tcp_req->req.iovcnt, tcp_req->req.data); 2163 2164 return 0; 2165 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2166 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2167 uint64_t offset = sgl->address; 2168 uint32_t max_len = ttransport->transport.opts.in_capsule_data_size; 2169 2170 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2171 offset, sgl->unkeyed.length); 2172 2173 if (offset > max_len) { 2174 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2175 offset, max_len); 2176 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2177 return -1; 2178 } 2179 max_len -= (uint32_t)offset; 2180 2181 if (sgl->unkeyed.length > max_len) { 2182 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2183 sgl->unkeyed.length, max_len); 2184 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2185 return -1; 2186 } 2187 2188 tcp_req->req.data = tcp_req->buf + offset; 2189 tcp_req->data_from_pool = false; 2190 tcp_req->req.length = sgl->unkeyed.length; 2191 2192 tcp_req->req.iov[0].iov_base = tcp_req->req.data; 2193 tcp_req->req.iov[0].iov_len = tcp_req->req.length; 2194 tcp_req->req.iovcnt = 1; 2195 2196 return 0; 2197 } 2198 2199 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2200 sgl->generic.type, sgl->generic.subtype); 2201 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2202 return -1; 2203 } 2204 2205 static void 2206 spdk_nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2207 struct spdk_nvmf_tcp_req *tcp_req) 2208 { 2209 struct nvme_tcp_pdu *rsp_pdu; 2210 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2211 uint32_t plen, pdo, alignment, offset, iov_index; 2212 2213 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2214 2215 /* always use the first iov_len, which is correct */ 2216 iov_index = tcp_req->c2h_data_offset / tcp_req->req.iov[0].iov_len; 2217 offset = tcp_req->c2h_data_offset % tcp_req->req.iov[0].iov_len; 2218 2219 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 2220 assert(rsp_pdu != NULL); 2221 2222 c2h_data = &rsp_pdu->hdr.c2h_data; 2223 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2224 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2225 2226 if (tqpair->host_hdgst_enable) { 2227 plen += SPDK_NVME_TCP_DIGEST_LEN; 2228 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2229 } 2230 2231 /* set the psh */ 2232 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2233 c2h_data->datal = spdk_min(NVMF_TCP_PDU_MAX_C2H_DATA_SIZE, 2234 (tcp_req->req.iov[iov_index].iov_len - offset)); 2235 c2h_data->datao = tcp_req->c2h_data_offset; 2236 2237 /* set the padding */ 2238 rsp_pdu->padding_len = 0; 2239 pdo = plen; 2240 if (tqpair->cpda) { 2241 alignment = (tqpair->cpda + 1) << 2; 2242 if (alignment > plen) { 2243 rsp_pdu->padding_len = alignment - plen; 2244 pdo = plen = alignment; 2245 } 2246 } 2247 2248 c2h_data->common.pdo = pdo; 2249 plen += c2h_data->datal; 2250 if (tqpair->host_ddgst_enable) { 2251 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2252 plen += SPDK_NVME_TCP_DIGEST_LEN; 2253 } 2254 2255 c2h_data->common.plen = plen; 2256 2257 nvme_tcp_pdu_set_data(rsp_pdu, tcp_req->req.iov[iov_index].iov_base + offset, c2h_data->datal); 2258 2259 tcp_req->c2h_data_offset += c2h_data->datal; 2260 if (iov_index == (tcp_req->req.iovcnt - 1) && (tcp_req->c2h_data_offset == tcp_req->req.length)) { 2261 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Last pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 2262 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2263 /* The linux kernel does not support this yet */ 2264 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 2265 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2266 #endif 2267 TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 2268 } 2269 2270 tqpair->c2h_data_pdu_cnt += 1; 2271 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2272 } 2273 2274 static int 2275 spdk_nvmf_tcp_calc_c2h_data_pdu_num(struct spdk_nvmf_tcp_req *tcp_req) 2276 { 2277 uint32_t i, iov_cnt, pdu_num = 0; 2278 2279 iov_cnt = tcp_req->req.iovcnt; 2280 for (i = 0; i < iov_cnt; i++) { 2281 pdu_num += (tcp_req->req.iov[i].iov_len + NVMF_TCP_PDU_MAX_C2H_DATA_SIZE - 1) / 2282 NVMF_TCP_PDU_MAX_C2H_DATA_SIZE; 2283 } 2284 2285 return pdu_num; 2286 } 2287 2288 static void 2289 spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct spdk_nvmf_tcp_qpair *tqpair) 2290 { 2291 struct spdk_nvmf_tcp_req *tcp_req; 2292 2293 while (!TAILQ_EMPTY(&tqpair->queued_c2h_data_tcp_req) && 2294 (tqpair->c2h_data_pdu_cnt < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 2295 tcp_req = TAILQ_FIRST(&tqpair->queued_c2h_data_tcp_req); 2296 spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2297 } 2298 } 2299 2300 static void 2301 spdk_nvmf_tcp_queue_c2h_data(struct spdk_nvmf_tcp_req *tcp_req, 2302 struct spdk_nvmf_tcp_qpair *tqpair) 2303 { 2304 tcp_req->c2h_data_pdu_num = spdk_nvmf_tcp_calc_c2h_data_pdu_num(tcp_req); 2305 2306 assert(tcp_req->c2h_data_pdu_num < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM); 2307 2308 TAILQ_INSERT_TAIL(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 2309 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 2310 } 2311 2312 static int 2313 request_transfer_out(struct spdk_nvmf_request *req) 2314 { 2315 struct spdk_nvmf_tcp_req *tcp_req; 2316 struct spdk_nvmf_qpair *qpair; 2317 struct spdk_nvmf_tcp_qpair *tqpair; 2318 struct spdk_nvme_cpl *rsp; 2319 2320 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2321 2322 qpair = req->qpair; 2323 rsp = &req->rsp->nvme_cpl; 2324 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2325 2326 /* Advance our sq_head pointer */ 2327 if (qpair->sq_head == qpair->sq_head_max) { 2328 qpair->sq_head = 0; 2329 } else { 2330 qpair->sq_head++; 2331 } 2332 rsp->sqhd = qpair->sq_head; 2333 2334 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2335 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2336 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && 2337 req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2338 spdk_nvmf_tcp_queue_c2h_data(tcp_req, tqpair); 2339 } else { 2340 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2341 } 2342 2343 return 0; 2344 } 2345 2346 static void 2347 spdk_nvmf_tcp_pdu_set_buf_from_req(struct spdk_nvmf_tcp_qpair *tqpair, 2348 struct spdk_nvmf_tcp_req *tcp_req) 2349 { 2350 struct nvme_tcp_pdu *pdu; 2351 2352 if (tcp_req->data_from_pool) { 2353 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Will send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2354 tcp_req->next_expected_r2t_offset = 0; 2355 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T); 2356 spdk_nvmf_tcp_handle_queued_r2t_req(tqpair); 2357 } else { 2358 pdu = &tqpair->pdu_in_progress; 2359 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2360 tqpair); 2361 /* No need to send r2t, contained in the capsuled data */ 2362 nvme_tcp_pdu_set_data(pdu, tcp_req->req.data, tcp_req->req.length); 2363 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2364 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2365 } 2366 } 2367 2368 static void 2369 spdk_nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2370 struct spdk_nvmf_tcp_req *tcp_req) 2371 { 2372 struct nvme_tcp_pdu *pdu; 2373 uint32_t plen = 0; 2374 2375 pdu = &tqpair->pdu_in_progress; 2376 plen = pdu->hdr.common.hlen; 2377 2378 if (tqpair->host_hdgst_enable) { 2379 plen += SPDK_NVME_TCP_DIGEST_LEN; 2380 } 2381 2382 if (pdu->hdr.common.plen != plen) { 2383 tcp_req->has_incapsule_data = true; 2384 } 2385 } 2386 2387 static bool 2388 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2389 struct spdk_nvmf_tcp_req *tcp_req) 2390 { 2391 struct spdk_nvmf_tcp_qpair *tqpair; 2392 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2393 int rc; 2394 enum spdk_nvmf_tcp_req_state prev_state; 2395 bool progress = false; 2396 struct spdk_nvmf_transport_poll_group *group; 2397 2398 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2399 group = &tqpair->group->group; 2400 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2401 2402 /* The loop here is to allow for several back-to-back state changes. */ 2403 do { 2404 prev_state = tcp_req->state; 2405 2406 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2407 tqpair); 2408 2409 switch (tcp_req->state) { 2410 case TCP_REQUEST_STATE_FREE: 2411 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2412 * to escape this state. */ 2413 break; 2414 case TCP_REQUEST_STATE_NEW: 2415 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2416 2417 /* copy the cmd from the receive pdu */ 2418 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2419 2420 /* The next state transition depends on the data transfer needs of this request. */ 2421 tcp_req->req.xfer = spdk_nvmf_tcp_req_get_xfer(tcp_req); 2422 2423 /* If no data to transfer, ready to execute. */ 2424 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2425 /* Reset the tqpair receving pdu state */ 2426 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2427 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2428 break; 2429 } 2430 2431 spdk_nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2432 2433 if (!tcp_req->has_incapsule_data) { 2434 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2435 } 2436 2437 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2438 TAILQ_INSERT_TAIL(&tqpair->group->pending_data_buf_queue, tcp_req, link); 2439 break; 2440 case TCP_REQUEST_STATE_NEED_BUFFER: 2441 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2442 2443 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2444 2445 if (!tcp_req->has_incapsule_data && 2446 (tcp_req != TAILQ_FIRST(&tqpair->group->pending_data_buf_queue))) { 2447 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2448 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2449 tcp_req, tqpair); 2450 /* This request needs to wait in line to obtain a buffer */ 2451 break; 2452 } 2453 2454 /* Try to get a data buffer */ 2455 rc = spdk_nvmf_tcp_req_parse_sgl(ttransport, tcp_req); 2456 if (rc < 0) { 2457 TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link); 2458 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2459 /* Reset the tqpair receving pdu state */ 2460 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2461 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2462 break; 2463 } 2464 2465 if (!tcp_req->req.data) { 2466 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2467 tcp_req, tqpair); 2468 /* No buffers available. */ 2469 break; 2470 } 2471 2472 TAILQ_REMOVE(&tqpair->group->pending_data_buf_queue, tcp_req, link); 2473 2474 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2475 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2476 spdk_nvmf_tcp_pdu_set_buf_from_req(tqpair, tcp_req); 2477 break; 2478 } 2479 2480 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2481 break; 2482 case TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T: 2483 spdk_trace_record(TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 0, 0, 2484 (uintptr_t)tcp_req, 0); 2485 /* Some external code must kick a request into TCP_REQUEST_STATE_DATA_PENDING_R2T 2486 * to escape this state. */ 2487 break; 2488 2489 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2490 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2491 (uintptr_t)tcp_req, 0); 2492 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2493 * to escape this state. */ 2494 break; 2495 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2496 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2497 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2498 spdk_nvmf_request_exec(&tcp_req->req); 2499 break; 2500 case TCP_REQUEST_STATE_EXECUTING: 2501 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2502 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2503 * to escape this state. */ 2504 break; 2505 case TCP_REQUEST_STATE_EXECUTED: 2506 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2507 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2508 break; 2509 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2510 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2511 rc = request_transfer_out(&tcp_req->req); 2512 assert(rc == 0); /* No good way to handle this currently */ 2513 break; 2514 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2515 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2516 (uintptr_t)tcp_req, 2517 0); 2518 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2519 * to escape this state. */ 2520 break; 2521 case TCP_REQUEST_STATE_COMPLETED: 2522 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2523 if (tcp_req->data_from_pool) { 2524 spdk_nvmf_tcp_request_free_buffers(tcp_req, group, &ttransport->transport); 2525 } 2526 tcp_req->req.length = 0; 2527 tcp_req->req.iovcnt = 0; 2528 tcp_req->req.data = NULL; 2529 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2530 break; 2531 case TCP_REQUEST_NUM_STATES: 2532 default: 2533 assert(0); 2534 break; 2535 } 2536 2537 if (tcp_req->state != prev_state) { 2538 progress = true; 2539 } 2540 } while (tcp_req->state != prev_state); 2541 2542 return progress; 2543 } 2544 2545 static void 2546 spdk_nvmf_tcp_qpair_process_pending(struct spdk_nvmf_tcp_transport *ttransport, 2547 struct spdk_nvmf_tcp_qpair *tqpair) 2548 { 2549 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 2550 2551 /* Tqpair is not in a good state, so return it */ 2552 if (spdk_unlikely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 2553 return; 2554 } 2555 2556 spdk_nvmf_tcp_handle_queued_r2t_req(tqpair); 2557 2558 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->group->pending_data_buf_queue, link, req_tmp) { 2559 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2560 break; 2561 } 2562 } 2563 } 2564 2565 static void 2566 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2567 { 2568 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2569 struct spdk_nvmf_tcp_transport *ttransport; 2570 int rc; 2571 2572 assert(tqpair != NULL); 2573 2574 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2575 spdk_nvmf_tcp_qpair_process_pending(ttransport, tqpair); 2576 rc = spdk_nvmf_tcp_sock_process(tqpair); 2577 2578 /* check the following two factors: 2579 * rc: The socket is closed 2580 * State of tqpair: The tqpair is in EXITING state due to internal error 2581 */ 2582 if ((rc < 0) || (tqpair->state == NVME_TCP_QPAIR_STATE_EXITING)) { 2583 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2584 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 2585 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair); 2586 spdk_poller_unregister(&tqpair->timeout_poller); 2587 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 2588 } 2589 } 2590 2591 static int 2592 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2593 struct spdk_nvmf_qpair *qpair) 2594 { 2595 struct spdk_nvmf_tcp_poll_group *tgroup; 2596 struct spdk_nvmf_tcp_qpair *tqpair; 2597 int rc; 2598 2599 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2600 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2601 2602 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2603 spdk_nvmf_tcp_sock_cb, tqpair); 2604 if (rc != 0) { 2605 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2606 spdk_strerror(errno), errno); 2607 spdk_nvmf_tcp_qpair_destroy(tqpair); 2608 return -1; 2609 } 2610 2611 rc = spdk_nvmf_tcp_qpair_sock_init(tqpair); 2612 if (rc != 0) { 2613 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2614 spdk_nvmf_tcp_qpair_destroy(tqpair); 2615 return -1; 2616 } 2617 2618 rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair); 2619 if (rc < 0) { 2620 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2621 spdk_nvmf_tcp_qpair_destroy(tqpair); 2622 return -1; 2623 } 2624 2625 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, 1); 2626 if (rc < 0) { 2627 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2628 spdk_nvmf_tcp_qpair_destroy(tqpair); 2629 return -1; 2630 } 2631 2632 tqpair->group = tgroup; 2633 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2634 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2635 2636 return 0; 2637 } 2638 2639 static int 2640 spdk_nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2641 struct spdk_nvmf_qpair *qpair) 2642 { 2643 struct spdk_nvmf_tcp_poll_group *tgroup; 2644 struct spdk_nvmf_tcp_qpair *tqpair; 2645 int rc; 2646 2647 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2648 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2649 2650 assert(tqpair->group == tgroup); 2651 2652 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2653 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2654 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2655 if (rc != 0) { 2656 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2657 spdk_strerror(errno), errno); 2658 } 2659 2660 return rc; 2661 } 2662 2663 static int 2664 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2665 { 2666 struct spdk_nvmf_tcp_transport *ttransport; 2667 struct spdk_nvmf_tcp_req *tcp_req; 2668 2669 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2670 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2671 2672 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2673 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 2674 2675 return 0; 2676 } 2677 2678 static void 2679 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2680 { 2681 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2682 2683 spdk_nvmf_tcp_qpair_destroy(SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair)); 2684 } 2685 2686 static int 2687 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2688 { 2689 struct spdk_nvmf_tcp_poll_group *tgroup; 2690 int rc; 2691 2692 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2693 2694 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs))) { 2695 return 0; 2696 } 2697 2698 rc = spdk_sock_group_poll(tgroup->sock_group); 2699 if (rc < 0) { 2700 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2701 return rc; 2702 } 2703 2704 return 0; 2705 } 2706 2707 static int 2708 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2709 struct spdk_nvme_transport_id *trid, bool peer) 2710 { 2711 struct spdk_nvmf_tcp_qpair *tqpair; 2712 uint16_t port; 2713 2714 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2715 trid->trtype = SPDK_NVME_TRANSPORT_TCP; 2716 2717 if (peer) { 2718 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2719 port = tqpair->initiator_port; 2720 } else { 2721 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2722 port = tqpair->target_port; 2723 } 2724 2725 if (spdk_sock_is_ipv4(tqpair->sock)) { 2726 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2727 } else if (spdk_sock_is_ipv4(tqpair->sock)) { 2728 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2729 } else { 2730 return -1; 2731 } 2732 2733 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2734 return 0; 2735 } 2736 2737 static int 2738 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2739 struct spdk_nvme_transport_id *trid) 2740 { 2741 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2742 } 2743 2744 static int 2745 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2746 struct spdk_nvme_transport_id *trid) 2747 { 2748 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2749 } 2750 2751 static int 2752 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2753 struct spdk_nvme_transport_id *trid) 2754 { 2755 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2756 } 2757 2758 static int 2759 spdk_nvmf_tcp_qpair_set_sq_size(struct spdk_nvmf_qpair *qpair) 2760 { 2761 struct spdk_nvmf_tcp_qpair *tqpair; 2762 int rc; 2763 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2764 2765 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, tqpair->qpair.sq_head_max); 2766 if (!rc) { 2767 tqpair->max_queue_depth += tqpair->qpair.sq_head_max; 2768 tqpair->free_pdu_num += tqpair->qpair.sq_head_max; 2769 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] += tqpair->qpair.sq_head_max; 2770 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The queue depth=%u for tqpair=%p\n", 2771 tqpair->max_queue_depth, tqpair); 2772 } 2773 2774 return rc; 2775 2776 } 2777 2778 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2779 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2780 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 64 2781 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2782 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2783 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2784 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 512 2785 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2786 2787 static void 2788 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2789 { 2790 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2791 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2792 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2793 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2794 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2795 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2796 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2797 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2798 } 2799 2800 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2801 .type = SPDK_NVME_TRANSPORT_TCP, 2802 .opts_init = spdk_nvmf_tcp_opts_init, 2803 .create = spdk_nvmf_tcp_create, 2804 .destroy = spdk_nvmf_tcp_destroy, 2805 2806 .listen = spdk_nvmf_tcp_listen, 2807 .stop_listen = spdk_nvmf_tcp_stop_listen, 2808 .accept = spdk_nvmf_tcp_accept, 2809 2810 .listener_discover = spdk_nvmf_tcp_discover, 2811 2812 .poll_group_create = spdk_nvmf_tcp_poll_group_create, 2813 .poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy, 2814 .poll_group_add = spdk_nvmf_tcp_poll_group_add, 2815 .poll_group_remove = spdk_nvmf_tcp_poll_group_remove, 2816 .poll_group_poll = spdk_nvmf_tcp_poll_group_poll, 2817 2818 .req_free = spdk_nvmf_tcp_req_free, 2819 .req_complete = spdk_nvmf_tcp_req_complete, 2820 2821 .qpair_fini = spdk_nvmf_tcp_close_qpair, 2822 .qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid, 2823 .qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid, 2824 .qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid, 2825 .qpair_set_sqsize = spdk_nvmf_tcp_qpair_set_sq_size, 2826 }; 2827 2828 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2829