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 /* 53 * AIO backend requires block size aligned data buffers, 54 * extra 4KiB aligned data buffer should work for most devices. 55 */ 56 #define SHIFT_4KB 12u 57 #define NVMF_DATA_BUFFER_ALIGNMENT (1u << SHIFT_4KB) 58 #define NVMF_DATA_BUFFER_MASK (NVMF_DATA_BUFFER_ALIGNMENT - 1LL) 59 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 60 61 #define NVMF_TCP_PDU_MAX_H2C_DATA_SIZE 131072 62 #define NVMF_TCP_PDU_MAX_C2H_DATA_SIZE 131072 63 #define NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM 64 /* Maximal c2h_data pdu number for ecah tqpair */ 64 65 /* This is used to support the Linux kernel NVMe-oF initiator */ 66 #define LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 0 67 68 /* spdk nvmf related structure */ 69 enum spdk_nvmf_tcp_req_state { 70 71 /* The request is not currently in use */ 72 TCP_REQUEST_STATE_FREE = 0, 73 74 /* Initial state when request first received */ 75 TCP_REQUEST_STATE_NEW, 76 77 /* The request is queued until a data buffer is available. */ 78 TCP_REQUEST_STATE_NEED_BUFFER, 79 80 /* The request is pending on r2t slots */ 81 TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 82 83 /* The request is currently transferring data from the host to the controller. */ 84 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 85 86 /* The request is ready to execute at the block device */ 87 TCP_REQUEST_STATE_READY_TO_EXECUTE, 88 89 /* The request is currently executing at the block device */ 90 TCP_REQUEST_STATE_EXECUTING, 91 92 /* The request finished executing at the block device */ 93 TCP_REQUEST_STATE_EXECUTED, 94 95 /* The request is ready to send a completion */ 96 TCP_REQUEST_STATE_READY_TO_COMPLETE, 97 98 /* The request is currently transferring final pdus from the controller to the host. */ 99 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 100 101 /* The request completed and can be marked free. */ 102 TCP_REQUEST_STATE_COMPLETED, 103 104 /* Terminator */ 105 TCP_REQUEST_NUM_STATES, 106 }; 107 108 static const char *spdk_nvme_tcp_term_req_fes_str[] = { 109 "Invalid PDU Header Field", 110 "PDU Sequence Error", 111 "Header Digiest Error", 112 "Data Transfer Out of Range", 113 "R2T Limit Exceeded", 114 "Unsupported parameter", 115 }; 116 117 #define OBJECT_NVMF_TCP_IO 0x80 118 119 #define TRACE_GROUP_NVMF_TCP 0x5 120 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 121 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 122 #define TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 123 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 124 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 125 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 126 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 127 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 128 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 129 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 130 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 131 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 132 #define TRACE_TCP_FLUSH_WRITEBUF_PDU_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 133 134 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 135 { 136 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 137 spdk_trace_register_description("TCP_REQ_NEW", "", 138 TRACE_TCP_REQUEST_STATE_NEW, 139 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 140 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", "", 141 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 142 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 143 spdk_trace_register_description("TCP_REQ_TX_PENDING_R2T", "", 144 TRACE_TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 145 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 146 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", "", 147 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 148 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 149 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", "", 150 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 151 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 152 spdk_trace_register_description("TCP_REQ_EXECUTING", "", 153 TRACE_TCP_REQUEST_STATE_EXECUTING, 154 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 155 spdk_trace_register_description("TCP_REQ_EXECUTED", "", 156 TRACE_TCP_REQUEST_STATE_EXECUTED, 157 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 158 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", "", 159 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 160 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 161 spdk_trace_register_description("TCP_REQ_COMPLETING_INCAPSULE", "", 162 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 163 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 164 spdk_trace_register_description("TCP_REQ_COMPLETED", "", 165 TRACE_TCP_REQUEST_STATE_COMPLETED, 166 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 167 spdk_trace_register_description("TCP_FLUSH_WRITEBUF_START", "", 168 TRACE_TCP_FLUSH_WRITEBUF_START, 169 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 170 spdk_trace_register_description("TCP_FLUSH_WRITEBUF_DONE", "", 171 TRACE_TCP_FLUSH_WRITEBUF_DONE, 172 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 173 spdk_trace_register_description("TCP_FLUSH_WRITEBUF_PDU_DONE", "", 174 TRACE_TCP_FLUSH_WRITEBUF_PDU_DONE, 175 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 176 } 177 178 struct nvme_tcp_req { 179 struct spdk_nvmf_request req; 180 struct spdk_nvme_cpl rsp; 181 struct spdk_nvme_cmd cmd; 182 183 /* In-capsule data buffer */ 184 uint8_t *buf; 185 186 bool data_from_pool; 187 void *buffers[SPDK_NVMF_MAX_SGL_ENTRIES]; 188 189 /* transfer_tag */ 190 uint16_t ttag; 191 192 /* 193 * next_expected_r2t_offset is used when we receive the h2c_data PDU. 194 */ 195 uint32_t next_expected_r2t_offset; 196 uint32_t r2tl_remain; 197 198 /* 199 * c2h_data_offset is used when we send the c2h_data PDU. 200 */ 201 uint32_t c2h_data_offset; 202 uint32_t c2h_data_pdu_num; 203 204 enum spdk_nvmf_tcp_req_state state; 205 206 TAILQ_ENTRY(nvme_tcp_req) link; 207 TAILQ_ENTRY(nvme_tcp_req) state_link; 208 }; 209 210 struct nvme_tcp_qpair { 211 struct spdk_nvmf_qpair qpair; 212 struct spdk_nvmf_tcp_port *port; 213 struct spdk_sock *sock; 214 struct spdk_poller *flush_poller; 215 216 enum nvme_tcp_pdu_recv_state recv_state; 217 enum nvme_tcp_qpair_state state; 218 219 struct nvme_tcp_pdu pdu_in_progress; 220 221 TAILQ_HEAD(, nvme_tcp_pdu) send_queue; 222 TAILQ_HEAD(, nvme_tcp_pdu) free_queue; 223 224 struct nvme_tcp_pdu *pdu; 225 struct nvme_tcp_pdu *pdu_pool; 226 uint16_t free_pdu_num; 227 228 /* Queues to track the requests in all states */ 229 TAILQ_HEAD(, nvme_tcp_req) state_queue[TCP_REQUEST_NUM_STATES]; 230 /* Number of requests in each state */ 231 int32_t state_cntr[TCP_REQUEST_NUM_STATES]; 232 233 uint32_t maxr2t; 234 uint32_t pending_r2t; 235 TAILQ_HEAD(, nvme_tcp_req) queued_r2t_tcp_req; 236 TAILQ_HEAD(, nvme_tcp_req) queued_c2h_data_tcp_req; 237 238 uint8_t cpda; 239 240 /* Array of size "max_queue_depth * InCapsuleDataSize" containing 241 * buffers to be used for in capsule data. 242 */ 243 void *buf; 244 void *bufs; 245 struct nvme_tcp_req *req; 246 struct nvme_tcp_req *reqs; 247 248 bool host_hdgst_enable; 249 bool host_ddgst_enable; 250 251 252 /* The maximum number of I/O outstanding on this connection at one time */ 253 uint16_t max_queue_depth; 254 255 256 /** Specifies the maximum number of PDU-Data bytes per H2C Data Transfer PDU */ 257 uint32_t maxh2cdata; 258 259 /* Timer used to destroy qpair after detecting transport error issue if initiator does 260 * not close the connection. 261 */ 262 uint64_t last_pdu_time; 263 int timeout; 264 265 /* Mgmt channel */ 266 struct spdk_io_channel *mgmt_channel; 267 struct spdk_nvmf_tcp_mgmt_channel *ch; 268 269 uint32_t c2h_data_pdu_cnt; 270 271 /* IP address */ 272 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 273 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 274 275 /* IP port */ 276 uint16_t initiator_port; 277 uint16_t target_port; 278 279 /* qpair->group is freed early, this should be a temporal fix */ 280 struct spdk_nvmf_tcp_poll_group *tgroup; 281 TAILQ_ENTRY(nvme_tcp_qpair) link; 282 }; 283 284 struct spdk_nvmf_tcp_poll_group { 285 struct spdk_nvmf_transport_poll_group group; 286 struct spdk_sock_group *sock_group; 287 TAILQ_HEAD(, nvme_tcp_qpair) qpairs; 288 }; 289 290 struct spdk_nvmf_tcp_port { 291 struct spdk_nvme_transport_id trid; 292 struct spdk_sock *listen_sock; 293 uint32_t ref; 294 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 295 }; 296 297 struct spdk_nvmf_tcp_transport { 298 struct spdk_nvmf_transport transport; 299 300 pthread_mutex_t lock; 301 302 struct spdk_mempool *data_buf_pool; 303 304 uint16_t max_queue_depth; 305 uint32_t max_io_size; 306 uint32_t io_unit_size; 307 uint32_t in_capsule_data_size; 308 309 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 310 }; 311 312 struct spdk_nvmf_tcp_mgmt_channel { 313 /* Requests that are waiting to obtain a data buffer */ 314 TAILQ_HEAD(, nvme_tcp_req) pending_data_buf_queue; 315 }; 316 317 static bool spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 318 struct nvme_tcp_req *tcp_req); 319 static void spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct nvme_tcp_qpair *tqpair); 320 321 static void 322 spdk_nvmf_tcp_req_set_state(struct nvme_tcp_req *tcp_req, 323 enum spdk_nvmf_tcp_req_state state) 324 { 325 struct spdk_nvmf_qpair *qpair; 326 struct nvme_tcp_qpair *tqpair; 327 328 qpair = tcp_req->req.qpair; 329 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 330 331 TAILQ_REMOVE(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 332 tqpair->state_cntr[tcp_req->state]--; 333 assert(tqpair->state_cntr[tcp_req->state] >= 0); 334 335 TAILQ_INSERT_TAIL(&tqpair->state_queue[state], tcp_req, state_link); 336 tqpair->state_cntr[state]++; 337 338 tcp_req->state = state; 339 } 340 341 static struct nvme_tcp_pdu * 342 spdk_nvmf_tcp_pdu_get(struct nvme_tcp_qpair *tqpair) 343 { 344 struct nvme_tcp_pdu *pdu; 345 346 pdu = TAILQ_FIRST(&tqpair->free_queue); 347 if (!pdu) { 348 SPDK_ERRLOG("Unable to get PDU for tqpair=%p\n", tqpair); 349 abort(); 350 return NULL; 351 } 352 353 tqpair->free_pdu_num--; 354 TAILQ_REMOVE(&tqpair->free_queue, pdu, tailq); 355 memset(pdu, 0, sizeof(*pdu)); 356 pdu->ref = 1; 357 pdu->tqpair = tqpair; 358 359 return pdu; 360 } 361 362 static void 363 spdk_nvmf_tcp_pdu_put(struct nvme_tcp_pdu *pdu) 364 { 365 struct nvme_tcp_qpair *tqpair; 366 if (!pdu) { 367 return; 368 } 369 370 assert(pdu->ref > 0); 371 assert(pdu->tqpair != NULL); 372 tqpair = pdu->tqpair; 373 374 pdu->ref--; 375 if (pdu->ref == 0) { 376 tqpair->free_pdu_num++; 377 TAILQ_INSERT_HEAD(&tqpair->free_queue, pdu, tailq); 378 } 379 } 380 381 static struct nvme_tcp_req * 382 spdk_nvmf_tcp_req_get(struct nvme_tcp_qpair *tqpair) 383 { 384 struct nvme_tcp_req *tcp_req; 385 386 tcp_req = TAILQ_FIRST(&tqpair->state_queue[TCP_REQUEST_STATE_FREE]); 387 if (!tcp_req) { 388 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 389 return NULL; 390 } 391 392 memset(&tcp_req->cmd, 0, sizeof(tcp_req->cmd)); 393 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 394 tcp_req->next_expected_r2t_offset = 0; 395 tcp_req->r2tl_remain = 0; 396 tcp_req->c2h_data_offset = 0; 397 398 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 399 return tcp_req; 400 } 401 402 static void 403 nvmf_tcp_request_free(struct nvme_tcp_req *tcp_req) 404 { 405 struct spdk_nvmf_tcp_transport *ttransport; 406 407 if (!tcp_req) { 408 return; 409 } 410 411 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req=%p will be freed\n", tcp_req); 412 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 413 struct spdk_nvmf_tcp_transport, transport); 414 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 415 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 416 } 417 418 static int 419 spdk_nvmf_tcp_req_free(struct spdk_nvmf_request *req) 420 { 421 struct nvme_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct nvme_tcp_req, req); 422 423 nvmf_tcp_request_free(tcp_req); 424 425 return 0; 426 } 427 428 static int 429 spdk_nvmf_tcp_mgmt_channel_create(void *io_device, void *ctx_buf) 430 { 431 struct spdk_nvmf_tcp_mgmt_channel *ch = ctx_buf; 432 433 TAILQ_INIT(&ch->pending_data_buf_queue); 434 return 0; 435 } 436 437 static void 438 spdk_nvmf_tcp_mgmt_channel_destroy(void *io_device, void *ctx_buf) 439 { 440 struct spdk_nvmf_tcp_mgmt_channel *ch = ctx_buf; 441 442 if (!TAILQ_EMPTY(&ch->pending_data_buf_queue)) { 443 SPDK_ERRLOG("Pending I/O list wasn't empty on channel destruction\n"); 444 } 445 } 446 447 static void 448 spdk_nvmf_tcp_drain_state_queue(struct nvme_tcp_qpair *tqpair, 449 enum spdk_nvmf_tcp_req_state state) 450 { 451 struct nvme_tcp_req *tcp_req, *req_tmp; 452 453 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[state], state_link, req_tmp) { 454 nvmf_tcp_request_free(tcp_req); 455 } 456 } 457 458 static void 459 spdk_nvmf_tcp_cleanup_all_states(struct nvme_tcp_qpair *tqpair) 460 { 461 struct nvme_tcp_req *tcp_req, *req_tmp; 462 struct nvme_tcp_pdu *pdu, *tmp_pdu; 463 464 /* For the requests in TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 465 * they can be refreed via the pdu in the send_queue 466 */ 467 TAILQ_FOREACH_SAFE(pdu, &tqpair->send_queue, tailq, tmp_pdu) { 468 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 469 assert(pdu->cb_fn != NULL); 470 pdu->cb_fn(pdu->cb_arg); 471 spdk_nvmf_tcp_pdu_put(pdu); 472 } 473 474 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 475 476 /* Wipe the requests waiting for R2t */ 477 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T], link, 478 req_tmp) { 479 TAILQ_REMOVE(&tqpair->queued_r2t_tcp_req, tcp_req, link); 480 } 481 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T); 482 483 /* Wipe the requests waiting for buffer from the global list */ 484 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_NEED_BUFFER], link, req_tmp) { 485 TAILQ_REMOVE(&tqpair->ch->pending_data_buf_queue, tcp_req, link); 486 } 487 488 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 489 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 490 491 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->queued_c2h_data_tcp_req, link, req_tmp) { 492 TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 493 } 494 495 spdk_nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 496 } 497 498 static void 499 spdk_nvmf_tcp_qpair_destroy(struct nvme_tcp_qpair *tqpair) 500 { 501 struct spdk_nvmf_tcp_poll_group *tgroup; 502 int rc; 503 504 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 505 tgroup = tqpair->tgroup; 506 507 if (tgroup) { 508 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 509 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 510 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 511 if (rc != 0) { 512 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 513 spdk_strerror(errno), errno); 514 } 515 } 516 517 spdk_poller_unregister(&tqpair->flush_poller); 518 spdk_sock_close(&tqpair->sock); 519 spdk_nvmf_tcp_cleanup_all_states(tqpair); 520 if (tqpair->mgmt_channel) { 521 spdk_put_io_channel(tqpair->mgmt_channel); 522 } 523 524 if (tqpair->free_pdu_num != (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 525 SPDK_ERRLOG("tqpair(%p) free pdu pool num is %u but should be %u\n", tqpair, 526 tqpair->free_pdu_num, 527 (tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)); 528 } 529 530 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->max_queue_depth) { 531 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 532 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 533 tqpair->max_queue_depth); 534 } 535 536 if (tqpair->c2h_data_pdu_cnt != 0) { 537 SPDK_ERRLOG("tqpair(%p) free c2h_data_pdu cnt is %u but should be 0\n", tqpair, 538 tqpair->c2h_data_pdu_cnt); 539 } 540 541 free(tqpair->pdu); 542 free(tqpair->pdu_pool); 543 free(tqpair->req); 544 free(tqpair->reqs); 545 spdk_dma_free(tqpair->buf); 546 spdk_dma_free(tqpair->bufs); 547 free(tqpair); 548 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Leave\n"); 549 } 550 551 static struct spdk_nvmf_transport * 552 spdk_nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 553 { 554 struct spdk_nvmf_tcp_transport *ttransport; 555 uint32_t sge_count; 556 557 ttransport = calloc(1, sizeof(*ttransport)); 558 if (!ttransport) { 559 return NULL; 560 } 561 562 TAILQ_INIT(&ttransport->ports); 563 564 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 565 566 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 567 568 ttransport->max_queue_depth = opts->max_queue_depth; 569 ttransport->max_io_size = opts->max_io_size; 570 ttransport->in_capsule_data_size = opts->in_capsule_data_size; 571 ttransport->io_unit_size = opts->io_unit_size; 572 573 /* I/O unit size cannot be larger than max I/O size */ 574 if (ttransport->io_unit_size > ttransport->max_io_size) { 575 ttransport->io_unit_size = ttransport->max_io_size; 576 } 577 578 sge_count = ttransport->max_io_size / ttransport->io_unit_size; 579 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 580 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", ttransport->io_unit_size); 581 free(ttransport); 582 return NULL; 583 } 584 585 ttransport->data_buf_pool = spdk_mempool_create("spdk_nvmf_tcp_data", 586 ttransport->max_queue_depth * 4, /* The 4 is arbitrarily chosen. Needs to be configurable. */ 587 ttransport->max_io_size + NVMF_DATA_BUFFER_ALIGNMENT, 588 SPDK_MEMPOOL_DEFAULT_CACHE_SIZE, 589 SPDK_ENV_SOCKET_ID_ANY); 590 591 if (!ttransport->data_buf_pool) { 592 SPDK_ERRLOG("Unable to allocate buffer pool for poll group\n"); 593 free(ttransport); 594 return NULL; 595 } 596 597 pthread_mutex_init(&ttransport->lock, NULL); 598 599 spdk_io_device_register(ttransport, spdk_nvmf_tcp_mgmt_channel_create, 600 spdk_nvmf_tcp_mgmt_channel_destroy, 601 sizeof(struct spdk_nvmf_tcp_mgmt_channel), "tcp_transport"); 602 603 return &ttransport->transport; 604 } 605 606 static int 607 spdk_nvmf_tcp_destroy(struct spdk_nvmf_transport *transport) 608 { 609 struct spdk_nvmf_tcp_transport *ttransport; 610 611 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 612 613 if (spdk_mempool_count(ttransport->data_buf_pool) != (ttransport->max_queue_depth * 4)) { 614 SPDK_ERRLOG("transport buffer pool count is %zu but should be %u\n", 615 spdk_mempool_count(ttransport->data_buf_pool), 616 ttransport->max_queue_depth * 4); 617 } 618 619 spdk_mempool_free(ttransport->data_buf_pool); 620 spdk_io_device_unregister(ttransport, NULL); 621 pthread_mutex_destroy(&ttransport->lock); 622 free(ttransport); 623 return 0; 624 } 625 626 static int 627 _spdk_nvmf_tcp_trsvcid_to_int(const char *trsvcid) 628 { 629 unsigned long long ull; 630 char *end = NULL; 631 632 ull = strtoull(trsvcid, &end, 10); 633 if (end == NULL || end == trsvcid || *end != '\0') { 634 return -1; 635 } 636 637 /* Valid TCP/IP port numbers are in [0, 65535] */ 638 if (ull > 65535) { 639 return -1; 640 } 641 642 return (int)ull; 643 } 644 645 /** 646 * Canonicalize a listen address trid. 647 */ 648 static int 649 _spdk_nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 650 const struct spdk_nvme_transport_id *trid) 651 { 652 int trsvcid_int; 653 654 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 655 if (trsvcid_int < 0) { 656 return -EINVAL; 657 } 658 659 memset(canon_trid, 0, sizeof(*canon_trid)); 660 canon_trid->trtype = SPDK_NVME_TRANSPORT_TCP; 661 canon_trid->adrfam = trid->adrfam; 662 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 663 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 664 665 return 0; 666 } 667 668 /** 669 * Find an existing listening port. 670 * 671 * Caller must hold ttransport->lock. 672 */ 673 static struct spdk_nvmf_tcp_port * 674 _spdk_nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 675 const struct spdk_nvme_transport_id *trid) 676 { 677 struct spdk_nvme_transport_id canon_trid; 678 struct spdk_nvmf_tcp_port *port; 679 680 if (_spdk_nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 681 return NULL; 682 } 683 684 TAILQ_FOREACH(port, &ttransport->ports, link) { 685 if (spdk_nvme_transport_id_compare(&canon_trid, &port->trid) == 0) { 686 return port; 687 } 688 } 689 690 return NULL; 691 } 692 693 static int 694 spdk_nvmf_tcp_listen(struct spdk_nvmf_transport *transport, 695 const struct spdk_nvme_transport_id *trid) 696 { 697 struct spdk_nvmf_tcp_transport *ttransport; 698 struct spdk_nvmf_tcp_port *port; 699 int trsvcid_int; 700 uint8_t adrfam; 701 702 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 703 704 trsvcid_int = _spdk_nvmf_tcp_trsvcid_to_int(trid->trsvcid); 705 if (trsvcid_int < 0) { 706 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 707 return -EINVAL; 708 } 709 710 pthread_mutex_lock(&ttransport->lock); 711 712 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 713 if (port) { 714 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Already listening on %s port %s\n", 715 trid->traddr, trid->trsvcid); 716 port->ref++; 717 pthread_mutex_unlock(&ttransport->lock); 718 return 0; 719 } 720 721 port = calloc(1, sizeof(*port)); 722 if (!port) { 723 SPDK_ERRLOG("Port allocation failed\n"); 724 free(port); 725 pthread_mutex_unlock(&ttransport->lock); 726 return -ENOMEM; 727 } 728 729 port->ref = 1; 730 731 if (_spdk_nvmf_tcp_canon_listen_trid(&port->trid, trid) != 0) { 732 SPDK_ERRLOG("Invalid traddr %s / trsvcid %s\n", 733 trid->traddr, trid->trsvcid); 734 free(port); 735 pthread_mutex_unlock(&ttransport->lock); 736 return -ENOMEM; 737 } 738 739 port->listen_sock = spdk_sock_listen(trid->traddr, trsvcid_int); 740 if (port->listen_sock == NULL) { 741 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 742 trid->traddr, trsvcid_int, 743 spdk_strerror(errno), errno); 744 free(port); 745 pthread_mutex_unlock(&ttransport->lock); 746 return -errno; 747 } 748 749 if (spdk_sock_is_ipv4(port->listen_sock)) { 750 adrfam = SPDK_NVMF_ADRFAM_IPV4; 751 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 752 adrfam = SPDK_NVMF_ADRFAM_IPV6; 753 } else { 754 SPDK_ERRLOG("Unhandled socket type\n"); 755 adrfam = 0; 756 } 757 758 if (adrfam != trid->adrfam) { 759 SPDK_ERRLOG("Socket address family mismatch\n"); 760 spdk_sock_close(&port->listen_sock); 761 free(port); 762 pthread_mutex_unlock(&ttransport->lock); 763 return -EINVAL; 764 } 765 766 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %d ***\n", 767 trid->traddr, trsvcid_int); 768 769 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 770 pthread_mutex_unlock(&ttransport->lock); 771 772 return 0; 773 } 774 775 static int 776 spdk_nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 777 const struct spdk_nvme_transport_id *trid) 778 { 779 struct spdk_nvmf_tcp_transport *ttransport; 780 struct spdk_nvmf_tcp_port *port; 781 int rc; 782 783 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 784 785 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Removing listen address %s port %s\n", 786 trid->traddr, trid->trsvcid); 787 788 pthread_mutex_lock(&ttransport->lock); 789 port = _spdk_nvmf_tcp_find_port(ttransport, trid); 790 if (port) { 791 assert(port->ref > 0); 792 port->ref--; 793 if (port->ref == 0) { 794 TAILQ_REMOVE(&ttransport->ports, port, link); 795 spdk_sock_close(&port->listen_sock); 796 free(port); 797 } 798 rc = 0; 799 } else { 800 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Port not found\n"); 801 rc = -ENOENT; 802 } 803 pthread_mutex_unlock(&ttransport->lock); 804 805 return rc; 806 } 807 808 static int 809 spdk_nvmf_tcp_qpair_flush_pdus_internal(struct nvme_tcp_qpair *tqpair) 810 { 811 const int array_size = 32; 812 struct iovec iovec_array[array_size]; 813 struct iovec *iov = iovec_array; 814 int iovec_cnt = 0; 815 int bytes = 0; 816 int total_length = 0; 817 uint32_t writev_offset; 818 struct nvme_tcp_pdu *pdu; 819 int pdu_length; 820 TAILQ_HEAD(, nvme_tcp_pdu) completed_pdus_list; 821 822 pdu = TAILQ_FIRST(&tqpair->send_queue); 823 824 if (pdu == NULL) { 825 return 0; 826 } 827 828 /* 829 * Build up a list of iovecs for the first few PDUs in the 830 * tqpair 's send_queue. 831 */ 832 while (pdu != NULL && ((array_size - iovec_cnt) >= 3)) { 833 iovec_cnt += nvme_tcp_build_iovecs(&iovec_array[iovec_cnt], 834 pdu, 835 tqpair->host_hdgst_enable, 836 tqpair->host_ddgst_enable); 837 total_length += pdu->hdr.common.plen; 838 pdu = TAILQ_NEXT(pdu, tailq); 839 } 840 841 /* 842 * Check if the first PDU was partially written out the last time 843 * this function was called, and if so adjust the iovec array 844 * accordingly. 845 */ 846 writev_offset = TAILQ_FIRST(&tqpair->send_queue)->writev_offset; 847 total_length -= writev_offset; 848 while ((writev_offset > 0) && (iovec_cnt > 0)) { 849 if (writev_offset >= iov->iov_len) { 850 writev_offset -= iov->iov_len; 851 iov++; 852 iovec_cnt--; 853 } else { 854 iov->iov_len -= writev_offset; 855 iov->iov_base = (char *)iov->iov_base + writev_offset; 856 writev_offset = 0; 857 } 858 } 859 860 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_START, 0, total_length, 0, iovec_cnt); 861 862 bytes = spdk_sock_writev(tqpair->sock, iov, iovec_cnt); 863 if (bytes == -1) { 864 if (errno == EWOULDBLOCK || errno == EAGAIN) { 865 return 1; 866 } else { 867 SPDK_ERRLOG("spdk_sock_writev() failed, errno %d: %s\n", 868 errno, spdk_strerror(errno)); 869 return -1; 870 } 871 } 872 873 spdk_trace_record(TRACE_TCP_FLUSH_WRITEBUF_DONE, 0, bytes, 0, 0); 874 875 pdu = TAILQ_FIRST(&tqpair->send_queue); 876 877 /* 878 * Free any PDUs that were fully written. If a PDU was only 879 * partially written, update its writev_offset so that next 880 * time only the unwritten portion will be sent to writev(). 881 */ 882 TAILQ_INIT(&completed_pdus_list); 883 while (bytes > 0) { 884 pdu_length = pdu->hdr.common.plen - pdu->writev_offset; 885 if (bytes >= pdu_length) { 886 bytes -= pdu_length; 887 TAILQ_REMOVE(&tqpair->send_queue, pdu, tailq); 888 TAILQ_INSERT_TAIL(&completed_pdus_list, pdu, tailq); 889 pdu = TAILQ_FIRST(&tqpair->send_queue); 890 891 } else { 892 pdu->writev_offset += bytes; 893 bytes = 0; 894 } 895 } 896 897 while (!TAILQ_EMPTY(&completed_pdus_list)) { 898 pdu = TAILQ_FIRST(&completed_pdus_list); 899 TAILQ_REMOVE(&completed_pdus_list, pdu, tailq); 900 assert(pdu->cb_fn != NULL); 901 pdu->cb_fn(pdu->cb_arg); 902 spdk_nvmf_tcp_pdu_put(pdu); 903 } 904 905 return TAILQ_EMPTY(&tqpair->send_queue) ? 0 : 1; 906 } 907 908 static int 909 spdk_nvmf_tcp_qpair_flush_pdus(void *_tqpair) 910 { 911 struct nvme_tcp_qpair *tqpair = _tqpair; 912 int rc; 913 914 if (tqpair->state == NVME_TCP_QPAIR_STATE_RUNNING) { 915 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 916 if (rc == 0 && tqpair->flush_poller != NULL) { 917 spdk_poller_unregister(&tqpair->flush_poller); 918 } else if (rc == 1 && tqpair->flush_poller == NULL) { 919 tqpair->flush_poller = spdk_poller_register(spdk_nvmf_tcp_qpair_flush_pdus, 920 tqpair, 50); 921 } 922 } else { 923 /* 924 * If the tqpair state is not RUNNING, then 925 * keep trying to flush PDUs until our list is 926 * empty - to make sure all data is sent before 927 * closing the connection. 928 */ 929 do { 930 rc = spdk_nvmf_tcp_qpair_flush_pdus_internal(tqpair); 931 } while (rc == 1); 932 } 933 934 if (rc < 0 && tqpair->state < NVME_TCP_QPAIR_STATE_EXITING) { 935 /* 936 * If the poller has already started destruction of the tqpair, 937 * i.e. the socket read failed, then the connection state may already 938 * be EXITED. We don't want to set it back to EXITING in that case. 939 */ 940 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 941 } 942 943 return -1; 944 } 945 946 static void 947 spdk_nvmf_tcp_qpair_write_pdu(struct nvme_tcp_qpair *tqpair, 948 struct nvme_tcp_pdu *pdu, 949 nvme_tcp_qpair_xfer_complete_cb cb_fn, 950 void *cb_arg) 951 { 952 int enable_digest; 953 int hlen; 954 uint32_t crc32c; 955 956 hlen = pdu->hdr.common.hlen; 957 enable_digest = 1; 958 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 959 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 960 /* this PDU should be sent without digest */ 961 enable_digest = 0; 962 } 963 964 /* Header Digest */ 965 if (enable_digest && tqpair->host_hdgst_enable) { 966 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 967 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 968 } 969 970 /* Data Digest */ 971 if (pdu->data_len > 0 && enable_digest && tqpair->host_ddgst_enable) { 972 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 973 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 974 } 975 976 pdu->cb_fn = cb_fn; 977 pdu->cb_arg = cb_arg; 978 TAILQ_INSERT_TAIL(&tqpair->send_queue, pdu, tailq); 979 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 980 } 981 982 static int 983 spdk_nvmf_tcp_qpair_init_mem_resource(struct nvme_tcp_qpair *tqpair, uint16_t size) 984 { 985 int i; 986 struct nvme_tcp_req *tcp_req; 987 struct spdk_nvmf_tcp_transport *ttransport; 988 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 989 990 if (!tqpair->qpair.sq_head_max) { 991 tqpair->req = calloc(1, sizeof(*tqpair->req)); 992 if (!tqpair->req) { 993 SPDK_ERRLOG("Unable to allocate req on tqpair=%p.\n", tqpair); 994 return -1; 995 } 996 997 if (ttransport->in_capsule_data_size) { 998 tqpair->buf = spdk_dma_zmalloc(ttransport->in_capsule_data_size, 0x1000, NULL); 999 if (!tqpair->buf) { 1000 SPDK_ERRLOG("Unable to allocate buf on tqpair=%p.\n", tqpair); 1001 return -1; 1002 } 1003 } 1004 1005 tcp_req = tqpair->req; 1006 tcp_req->ttag = 0; 1007 tcp_req->req.qpair = &tqpair->qpair; 1008 1009 /* Set up memory to receive commands */ 1010 if (tqpair->buf) { 1011 tcp_req->buf = tqpair->buf; 1012 } 1013 1014 /* Set the cmdn and rsp */ 1015 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1016 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1017 1018 /* Initialize request state to FREE */ 1019 tcp_req->state = TCP_REQUEST_STATE_FREE; 1020 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 1021 1022 tqpair->pdu = calloc(NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM + 1, sizeof(*tqpair->pdu)); 1023 if (!tqpair->pdu) { 1024 SPDK_ERRLOG("Unable to allocate pdu on tqpair=%p.\n", tqpair); 1025 return -1; 1026 } 1027 1028 for (i = 0; i < 1 + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; i++) { 1029 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu[i], tailq); 1030 } 1031 1032 } else { 1033 tqpair->reqs = calloc(size, sizeof(*tqpair->reqs)); 1034 if (!tqpair->reqs) { 1035 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1036 return -1; 1037 } 1038 1039 if (ttransport->in_capsule_data_size) { 1040 tqpair->bufs = spdk_dma_zmalloc(size * ttransport->in_capsule_data_size, 1041 0x1000, NULL); 1042 if (!tqpair->bufs) { 1043 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1044 return -1; 1045 } 1046 } 1047 1048 for (i = 0; i < size; i++) { 1049 struct nvme_tcp_req *tcp_req = &tqpair->reqs[i]; 1050 1051 tcp_req->ttag = i + 1; 1052 tcp_req->req.qpair = &tqpair->qpair; 1053 1054 /* Set up memory to receive commands */ 1055 if (tqpair->bufs) { 1056 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * ttransport->in_capsule_data_size)); 1057 } 1058 1059 /* Set the cmdn and rsp */ 1060 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1061 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1062 1063 /* Initialize request state to FREE */ 1064 tcp_req->state = TCP_REQUEST_STATE_FREE; 1065 TAILQ_INSERT_TAIL(&tqpair->state_queue[tcp_req->state], tcp_req, state_link); 1066 } 1067 1068 tqpair->pdu_pool = calloc(size, sizeof(*tqpair->pdu_pool)); 1069 if (!tqpair->pdu_pool) { 1070 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1071 return -1; 1072 } 1073 1074 for (i = 0; i < size; i++) { 1075 TAILQ_INSERT_TAIL(&tqpair->free_queue, &tqpair->pdu_pool[i], tailq); 1076 } 1077 } 1078 1079 return 0; 1080 } 1081 1082 static int 1083 spdk_nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1084 { 1085 struct spdk_nvmf_tcp_transport *ttransport; 1086 struct nvme_tcp_qpair *tqpair; 1087 int i; 1088 1089 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1090 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 1091 1092 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New TCP Connection: %p\n", qpair); 1093 1094 TAILQ_INIT(&tqpair->send_queue); 1095 TAILQ_INIT(&tqpair->free_queue); 1096 TAILQ_INIT(&tqpair->queued_r2t_tcp_req); 1097 TAILQ_INIT(&tqpair->queued_c2h_data_tcp_req); 1098 1099 /* Initialise request state queues of the qpair */ 1100 for (i = TCP_REQUEST_STATE_FREE; i < TCP_REQUEST_NUM_STATES; i++) { 1101 TAILQ_INIT(&tqpair->state_queue[i]); 1102 } 1103 1104 tqpair->host_hdgst_enable = true; 1105 tqpair->host_ddgst_enable = true; 1106 1107 tqpair->mgmt_channel = spdk_get_io_channel(ttransport); 1108 if (!tqpair->mgmt_channel) { 1109 return -1; 1110 } 1111 tqpair->ch = spdk_io_channel_get_ctx(tqpair->mgmt_channel); 1112 assert(tqpair->ch != NULL); 1113 1114 return 0; 1115 } 1116 1117 static int 1118 spdk_nvmf_tcp_qpair_sock_init(struct nvme_tcp_qpair *tqpair) 1119 { 1120 1121 int rc; 1122 int buf_size; 1123 1124 /* set recv buffer size */ 1125 buf_size = 2 * 1024 * 1024; 1126 rc = spdk_sock_set_recvbuf(tqpair->sock, buf_size); 1127 if (rc != 0) { 1128 SPDK_ERRLOG("spdk_sock_set_recvbuf failed\n"); 1129 return rc; 1130 } 1131 1132 /* set send buffer size */ 1133 rc = spdk_sock_set_sendbuf(tqpair->sock, buf_size); 1134 if (rc != 0) { 1135 SPDK_ERRLOG("spdk_sock_set_sendbuf failed\n"); 1136 return rc; 1137 } 1138 1139 /* set low water mark */ 1140 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_c2h_data_hdr)); 1141 if (rc != 0) { 1142 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1143 return rc; 1144 } 1145 1146 return 0; 1147 } 1148 1149 static void 1150 _spdk_nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1151 struct spdk_nvmf_tcp_port *port, 1152 struct spdk_sock *sock, new_qpair_fn cb_fn) 1153 { 1154 struct nvme_tcp_qpair *tqpair; 1155 int rc; 1156 1157 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "New connection accepted on %s port %s\n", 1158 port->trid.traddr, port->trid.trsvcid); 1159 1160 tqpair = calloc(1, sizeof(struct nvme_tcp_qpair)); 1161 if (tqpair == NULL) { 1162 SPDK_ERRLOG("Could not allocate new connection.\n"); 1163 spdk_sock_close(&sock); 1164 return; 1165 } 1166 1167 tqpair->sock = sock; 1168 tqpair->max_queue_depth = 1; 1169 tqpair->free_pdu_num = tqpair->max_queue_depth + NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM; 1170 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = tqpair->max_queue_depth; 1171 tqpair->port = port; 1172 tqpair->qpair.transport = transport; 1173 1174 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1175 sizeof(tqpair->target_addr), &tqpair->target_port, 1176 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1177 &tqpair->initiator_port); 1178 if (rc < 0) { 1179 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1180 spdk_nvmf_tcp_qpair_destroy(tqpair); 1181 return; 1182 } 1183 1184 cb_fn(&tqpair->qpair); 1185 } 1186 1187 static void 1188 spdk_nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port, 1189 new_qpair_fn cb_fn) 1190 { 1191 struct spdk_sock *sock; 1192 int i; 1193 1194 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1195 sock = spdk_sock_accept(port->listen_sock); 1196 if (sock) { 1197 _spdk_nvmf_tcp_handle_connect(transport, port, sock, cb_fn); 1198 } 1199 } 1200 } 1201 1202 static void 1203 spdk_nvmf_tcp_accept(struct spdk_nvmf_transport *transport, new_qpair_fn cb_fn) 1204 { 1205 struct spdk_nvmf_tcp_transport *ttransport; 1206 struct spdk_nvmf_tcp_port *port; 1207 1208 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1209 1210 TAILQ_FOREACH(port, &ttransport->ports, link) { 1211 spdk_nvmf_tcp_port_accept(transport, port, cb_fn); 1212 } 1213 } 1214 1215 static void 1216 spdk_nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1217 struct spdk_nvme_transport_id *trid, 1218 struct spdk_nvmf_discovery_log_page_entry *entry) 1219 { 1220 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1221 entry->adrfam = trid->adrfam; 1222 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_SPECIFIED; 1223 1224 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1225 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1226 1227 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1228 } 1229 1230 static struct spdk_nvmf_transport_poll_group * 1231 spdk_nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1232 { 1233 struct spdk_nvmf_tcp_poll_group *tgroup; 1234 1235 tgroup = calloc(1, sizeof(*tgroup)); 1236 if (!tgroup) { 1237 return NULL; 1238 } 1239 1240 tgroup->sock_group = spdk_sock_group_create(); 1241 if (!tgroup->sock_group) { 1242 goto cleanup; 1243 } 1244 1245 TAILQ_INIT(&tgroup->qpairs); 1246 return &tgroup->group; 1247 1248 cleanup: 1249 free(tgroup); 1250 return NULL; 1251 } 1252 1253 static void 1254 spdk_nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1255 { 1256 struct spdk_nvmf_tcp_poll_group *tgroup; 1257 1258 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1259 spdk_sock_group_close(&tgroup->sock_group); 1260 free(tgroup); 1261 } 1262 1263 static void 1264 spdk_nvmf_tcp_qpair_set_recv_state(struct nvme_tcp_qpair *tqpair, 1265 enum nvme_tcp_pdu_recv_state state) 1266 { 1267 if (tqpair->recv_state == state) { 1268 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1269 tqpair, state); 1270 return; 1271 } 1272 1273 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv state=%d\n", tqpair, state); 1274 tqpair->recv_state = state; 1275 switch (state) { 1276 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1277 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1278 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1279 break; 1280 case NVME_TCP_PDU_RECV_STATE_ERROR: 1281 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1282 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1283 break; 1284 default: 1285 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1286 abort(); 1287 break; 1288 } 1289 } 1290 1291 static void 1292 spdk_nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1293 { 1294 } 1295 1296 static void 1297 spdk_nvmf_tcp_send_c2h_term_req(struct nvme_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1298 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1299 { 1300 struct nvme_tcp_pdu *rsp_pdu; 1301 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1302 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1303 1304 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1305 if (!rsp_pdu) { 1306 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1307 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1308 return; 1309 } 1310 1311 c2h_term_req = &rsp_pdu->hdr.term_req; 1312 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1313 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1314 /* It should contain the header the received pdu */ 1315 c2h_term_req->common.plen = c2h_term_req->common.hlen + pdu->hdr.common.hlen; 1316 1317 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1318 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1319 DSET32(&c2h_term_req->fei, error_offset); 1320 } 1321 1322 /* copy the error code into the buffer */ 1323 rsp_pdu->data = (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len; 1324 if (pdu->hdr.common.plen <= SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE) { 1325 memcpy((uint8_t *)rsp_pdu->data, pdu->hdr.raw, pdu->hdr.common.plen); 1326 } else { 1327 memcpy((uint8_t *)rsp_pdu->data, pdu->hdr.raw, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1328 } 1329 1330 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1331 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_c2h_term_req_complete, tqpair); 1332 } 1333 1334 static void 1335 spdk_nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1336 struct nvme_tcp_qpair *tqpair, 1337 struct nvme_tcp_pdu *pdu) 1338 { 1339 struct nvme_tcp_req *tcp_req; 1340 1341 tcp_req = spdk_nvmf_tcp_req_get(tqpair); 1342 if (!tcp_req) { 1343 SPDK_ERRLOG("Cannot allocate tcp_req\n"); 1344 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1345 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1346 return; 1347 } 1348 1349 pdu->tcp_req = tcp_req; 1350 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 1351 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1352 return; 1353 } 1354 1355 static void 1356 spdk_nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1357 struct nvme_tcp_qpair *tqpair, 1358 struct nvme_tcp_pdu *pdu) 1359 { 1360 struct nvme_tcp_req *tcp_req; 1361 struct spdk_nvme_tcp_cmd *capsule_cmd; 1362 uint32_t error_offset = 0; 1363 enum spdk_nvme_tcp_term_req_fes fes; 1364 1365 capsule_cmd = &pdu->hdr.capsule_cmd; 1366 tcp_req = pdu->tcp_req; 1367 assert(tcp_req != NULL); 1368 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1369 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1370 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1371 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1372 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1373 goto err; 1374 } 1375 1376 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1377 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1378 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1379 1380 return; 1381 err: 1382 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1383 } 1384 1385 static void 1386 spdk_nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1387 struct nvme_tcp_qpair *tqpair, 1388 struct nvme_tcp_pdu *pdu) 1389 { 1390 struct nvme_tcp_req *tcp_req; 1391 uint32_t error_offset = 0; 1392 enum spdk_nvme_tcp_term_req_fes fes = 0; 1393 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1394 uint32_t iov_index; 1395 bool ttag_offset_error = false; 1396 1397 h2c_data = &pdu->hdr.h2c_data; 1398 1399 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1400 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1401 1402 /* According to the information in the pdu to find the req */ 1403 TAILQ_FOREACH(tcp_req, &tqpair->state_queue[TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER], 1404 state_link) { 1405 if ((tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid) && (tcp_req->ttag == h2c_data->ttag)) { 1406 break; 1407 } 1408 1409 if (!ttag_offset_error && (tcp_req->req.cmd->nvme_cmd.cid == h2c_data->cccid)) { 1410 ttag_offset_error = true; 1411 } 1412 } 1413 1414 if (!tcp_req) { 1415 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req is not found for tqpair=%p\n", tqpair); 1416 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1417 if (!ttag_offset_error) { 1418 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1419 } else { 1420 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1421 } 1422 goto err; 1423 } 1424 1425 if (tcp_req->next_expected_r2t_offset != h2c_data->datao) { 1426 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1427 "tcp_req(%p), tqpair=%p, expected_r2t_offset=%u, but data offset =%u\n", 1428 tcp_req, tqpair, tcp_req->next_expected_r2t_offset, h2c_data->datao); 1429 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1430 goto err; 1431 } 1432 1433 if (h2c_data->datal > tqpair->maxh2cdata) { 1434 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tcp_req(%p), tqpair=%p, datao=%u execeeds maxh2cdata size=%u\n", 1435 tcp_req, tqpair, h2c_data->datao, tqpair->maxh2cdata); 1436 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1437 goto err; 1438 } 1439 1440 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1441 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1442 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1443 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1444 fes = SPDK_NVME_TCP_TERM_REQ_FES_R2T_LIMIT_EXCEEDED; 1445 goto err; 1446 } 1447 1448 pdu->tcp_req = tcp_req; 1449 pdu->data_len = h2c_data->datal; 1450 iov_index = pdu->hdr.h2c_data.datao / ttransport->io_unit_size; 1451 pdu->data = tcp_req->req.iov[iov_index].iov_base + (pdu->hdr.h2c_data.datao % 1452 ttransport->io_unit_size); 1453 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1454 return; 1455 1456 err: 1457 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1458 } 1459 1460 static void 1461 spdk_nvmf_tcp_pdu_cmd_complete(void *cb_arg) 1462 { 1463 struct nvme_tcp_req *tcp_req = cb_arg; 1464 nvmf_tcp_request_free(tcp_req); 1465 } 1466 1467 static void 1468 spdk_nvmf_tcp_send_capsule_resp_pdu(struct nvme_tcp_req *tcp_req, 1469 struct nvme_tcp_qpair *tqpair) 1470 { 1471 struct nvme_tcp_pdu *rsp_pdu; 1472 struct spdk_nvme_tcp_rsp *capsule_resp; 1473 1474 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter, tqpair=%p\n", tqpair); 1475 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1476 if (!rsp_pdu) { 1477 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1478 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1479 return; 1480 } 1481 1482 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1483 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1484 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1485 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1486 if (tqpair->host_hdgst_enable) { 1487 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1488 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1489 } 1490 1491 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, tcp_req); 1492 } 1493 1494 static void 1495 spdk_nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1496 { 1497 struct nvme_tcp_req *tcp_req = cb_arg; 1498 struct nvme_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1499 struct nvme_tcp_qpair, qpair); 1500 1501 assert(tqpair != NULL); 1502 assert(tcp_req->c2h_data_pdu_num > 0); 1503 tcp_req->c2h_data_pdu_num--; 1504 if (!tcp_req->c2h_data_pdu_num) { 1505 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 1506 nvmf_tcp_request_free(tcp_req); 1507 #else 1508 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1509 #endif 1510 } 1511 1512 tqpair->c2h_data_pdu_cnt--; 1513 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 1514 } 1515 1516 static void 1517 spdk_nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_transport *ttransport, 1518 struct nvme_tcp_qpair *tqpair, 1519 struct nvme_tcp_req *tcp_req) 1520 { 1521 struct nvme_tcp_pdu *rsp_pdu; 1522 struct spdk_nvme_tcp_r2t_hdr *r2t; 1523 1524 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1525 if (!rsp_pdu) { 1526 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1527 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1528 return; 1529 } 1530 1531 r2t = &rsp_pdu->hdr.r2t; 1532 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1533 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1534 1535 if (tqpair->host_hdgst_enable) { 1536 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1537 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1538 } 1539 1540 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1541 r2t->ttag = tcp_req->ttag; 1542 r2t->r2to = tcp_req->next_expected_r2t_offset; 1543 r2t->r2tl = spdk_min(tcp_req->req.length - tcp_req->next_expected_r2t_offset, tqpair->maxh2cdata); 1544 tcp_req->r2tl_remain = r2t->r2tl; 1545 1546 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 1547 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1548 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1549 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_cmd_complete, NULL); 1550 } 1551 1552 static void 1553 spdk_nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1554 struct nvme_tcp_qpair *tqpair, 1555 struct nvme_tcp_pdu *pdu) 1556 { 1557 struct nvme_tcp_req *tcp_req; 1558 1559 tcp_req = pdu->tcp_req; 1560 assert(tcp_req != NULL); 1561 1562 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1563 1564 tcp_req->next_expected_r2t_offset += pdu->data_len; 1565 tcp_req->r2tl_remain -= pdu->data_len; 1566 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1567 1568 if (!tcp_req->r2tl_remain) { 1569 if (tcp_req->next_expected_r2t_offset == tcp_req->req.length) { 1570 assert(tqpair->pending_r2t > 0); 1571 tqpair->pending_r2t--; 1572 assert(tqpair->pending_r2t < tqpair->maxr2t); 1573 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1574 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1575 1576 /* fetch next */ 1577 tcp_req = TAILQ_FIRST(&tqpair->queued_r2t_tcp_req); 1578 if (tcp_req) { 1579 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 1580 } 1581 } else { 1582 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Send r2t pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 1583 spdk_nvmf_tcp_send_r2t_pdu(ttransport, tqpair, tcp_req); 1584 } 1585 } 1586 } 1587 1588 static void 1589 spdk_nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1590 { 1591 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1592 spdk_nvme_tcp_term_req_fes_str[h2c_term_req->fes]); 1593 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1594 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1595 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The offset from the start of the PDU header is %u\n", 1596 DGET32(h2c_term_req->fei)); 1597 } 1598 } 1599 1600 static void 1601 spdk_nvmf_tcp_h2c_term_req_hdr_handle(struct nvme_tcp_qpair *tqpair, 1602 struct nvme_tcp_pdu *pdu) 1603 { 1604 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1605 uint32_t error_offset = 0; 1606 enum spdk_nvme_tcp_term_req_fes fes; 1607 1608 1609 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1610 SPDK_ERRLOG("Fatal Error Stauts(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1611 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1612 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1613 goto end; 1614 } 1615 1616 /* set the data buffer */ 1617 pdu->data = (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen; 1618 pdu->data_len = h2c_term_req->common.plen - h2c_term_req->common.hlen; 1619 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1620 return; 1621 end: 1622 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1623 return; 1624 } 1625 1626 static void 1627 spdk_nvmf_tcp_h2c_term_req_payload_handle(struct nvme_tcp_qpair *tqpair, 1628 struct nvme_tcp_pdu *pdu) 1629 { 1630 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1631 1632 spdk_nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1633 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1634 return; 1635 } 1636 1637 static void 1638 spdk_nvmf_tcp_pdu_payload_handle(struct nvme_tcp_qpair *tqpair) 1639 { 1640 int rc = 0; 1641 struct nvme_tcp_pdu *pdu; 1642 uint32_t crc32c, error_offset = 0; 1643 enum spdk_nvme_tcp_term_req_fes fes; 1644 struct spdk_nvmf_tcp_transport *ttransport; 1645 1646 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1647 pdu = &tqpair->pdu_in_progress; 1648 1649 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 1650 /* check data digest if need */ 1651 if (pdu->ddigest_valid_bytes) { 1652 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1653 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1654 if (rc == 0) { 1655 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1656 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1657 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1658 return; 1659 1660 } 1661 } 1662 1663 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1664 switch (pdu->hdr.common.pdu_type) { 1665 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1666 spdk_nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1667 break; 1668 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1669 spdk_nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1670 break; 1671 1672 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1673 spdk_nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1674 break; 1675 1676 default: 1677 /* The code should not go to here */ 1678 SPDK_ERRLOG("The code should not go to here\n"); 1679 break; 1680 } 1681 } 1682 1683 static void 1684 spdk_nvmf_tcp_send_icresp_complete(void *cb_arg) 1685 { 1686 struct nvme_tcp_qpair *tqpair = cb_arg; 1687 1688 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1689 } 1690 1691 static void 1692 spdk_nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1693 struct nvme_tcp_qpair *tqpair, 1694 struct nvme_tcp_pdu *pdu) 1695 { 1696 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1697 struct nvme_tcp_pdu *rsp_pdu; 1698 struct spdk_nvme_tcp_ic_resp *ic_resp; 1699 uint32_t error_offset = 0; 1700 enum spdk_nvme_tcp_term_req_fes fes; 1701 1702 /* Only PFV 0 is defined currently */ 1703 if (ic_req->pfv != 0) { 1704 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1705 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1706 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1707 goto end; 1708 } 1709 1710 /* MAXR2T is 0's based */ 1711 tqpair->maxr2t = ic_req->maxr2t + 1ull; 1712 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "maxr2t =%u\n", tqpair->maxr2t); 1713 1714 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1715 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1716 1717 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1718 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1719 1720 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 1721 if (!rsp_pdu) { 1722 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 1723 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1724 return; 1725 } 1726 1727 ic_resp = &rsp_pdu->hdr.ic_resp; 1728 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1729 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1730 ic_resp->pfv = 0; 1731 ic_resp->cpda = tqpair->cpda; 1732 tqpair->maxh2cdata = spdk_min(NVMF_TCP_PDU_MAX_H2C_DATA_SIZE, ttransport->io_unit_size); 1733 ic_resp->maxh2cdata = tqpair->maxh2cdata; 1734 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1735 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1736 1737 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1738 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1739 1740 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_send_icresp_complete, tqpair); 1741 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1742 return; 1743 end: 1744 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1745 return; 1746 } 1747 1748 static void 1749 spdk_nvmf_tcp_pdu_psh_handle(struct nvme_tcp_qpair *tqpair) 1750 { 1751 struct nvme_tcp_pdu *pdu; 1752 int rc; 1753 uint32_t crc32c, error_offset = 0; 1754 enum spdk_nvme_tcp_term_req_fes fes; 1755 struct spdk_nvmf_tcp_transport *ttransport; 1756 1757 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1758 pdu = &tqpair->pdu_in_progress; 1759 1760 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "pdu type of tqpair(%p) is %d\n", tqpair, 1761 pdu->hdr.common.pdu_type); 1762 /* check header digest if needed */ 1763 if (pdu->has_hdgst) { 1764 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1765 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1766 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1767 if (rc == 0) { 1768 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1769 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1770 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1771 return; 1772 1773 } 1774 } 1775 1776 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 1777 switch (pdu->hdr.common.pdu_type) { 1778 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1779 spdk_nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1780 break; 1781 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1782 spdk_nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1783 break; 1784 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1785 spdk_nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1786 break; 1787 1788 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1789 spdk_nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1790 break; 1791 1792 default: 1793 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1794 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1795 error_offset = 1; 1796 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1797 break; 1798 } 1799 } 1800 1801 static void 1802 spdk_nvmf_tcp_pdu_ch_handle(struct nvme_tcp_qpair *tqpair) 1803 { 1804 struct nvme_tcp_pdu *pdu; 1805 uint32_t error_offset = 0; 1806 enum spdk_nvme_tcp_term_req_fes fes; 1807 uint8_t expected_hlen, pdo; 1808 bool plen_error = false, pdo_error = false; 1809 1810 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1811 pdu = &tqpair->pdu_in_progress; 1812 1813 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1814 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1815 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1816 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1817 goto err; 1818 } 1819 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1820 if (pdu->hdr.common.plen != expected_hlen) { 1821 plen_error = true; 1822 } 1823 } else { 1824 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1825 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1826 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1827 goto err; 1828 } 1829 1830 switch (pdu->hdr.common.pdu_type) { 1831 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1832 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1833 pdo = pdu->hdr.common.pdo; 1834 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1835 pdo_error = true; 1836 break; 1837 } 1838 1839 if (pdu->hdr.common.plen < expected_hlen) { 1840 plen_error = true; 1841 } 1842 break; 1843 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1844 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1845 pdo = pdu->hdr.common.pdo; 1846 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1847 pdo_error = true; 1848 break; 1849 } 1850 if (pdu->hdr.common.plen < expected_hlen) { 1851 plen_error = true; 1852 } 1853 break; 1854 1855 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1856 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1857 if ((pdu->hdr.common.plen <= expected_hlen) || 1858 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1859 plen_error = true; 1860 } 1861 break; 1862 1863 default: 1864 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1865 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1866 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1867 goto err; 1868 } 1869 } 1870 1871 if (pdu->hdr.common.hlen != expected_hlen) { 1872 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1873 pdu->hdr.common.pdu_type, 1874 expected_hlen, pdu->hdr.common.hlen, tqpair); 1875 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1876 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1877 goto err; 1878 } else if (pdo_error) { 1879 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1880 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1881 } else if (plen_error) { 1882 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1883 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1884 goto err; 1885 } else { 1886 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1887 return; 1888 } 1889 err: 1890 spdk_nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1891 } 1892 1893 static int 1894 spdk_nvmf_tcp_sock_process(struct nvme_tcp_qpair *tqpair) 1895 { 1896 int rc = 0; 1897 struct nvme_tcp_pdu *pdu; 1898 enum nvme_tcp_pdu_recv_state prev_state; 1899 uint32_t data_len; 1900 uint8_t psh_len, pdo, hlen; 1901 int8_t padding_len; 1902 1903 /* The loop here is to allow for several back-to-back state changes. */ 1904 do { 1905 prev_state = tqpair->recv_state; 1906 1907 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1908 1909 switch (tqpair->recv_state) { 1910 /* If in a new state */ 1911 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1912 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1913 break; 1914 /* Wait for the common header */ 1915 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1916 pdu = &tqpair->pdu_in_progress; 1917 /* common header */ 1918 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1919 rc = nvme_tcp_read_data(tqpair->sock, 1920 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1921 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1922 if (rc < 0) { 1923 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconnect tqpair=%p\n", tqpair); 1924 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1925 break; 1926 } 1927 pdu->ch_valid_bytes += rc; 1928 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1929 return NVME_TCP_PDU_IN_PROGRESS; 1930 } 1931 } 1932 1933 tqpair->last_pdu_time = spdk_get_ticks(); 1934 /* The command header of this PDU has now been read from the socket. */ 1935 spdk_nvmf_tcp_pdu_ch_handle(tqpair); 1936 break; 1937 /* Wait for the pdu specific header */ 1938 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1939 pdu = &tqpair->pdu_in_progress; 1940 if (pdu->hd_is_read) { 1941 break; 1942 } 1943 1944 psh_len = hlen = pdu->hdr.common.hlen; 1945 /* Only capsule_cmd and h2c_data has header digest */ 1946 if (((pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD) || 1947 (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_H2C_DATA)) && 1948 tqpair->host_hdgst_enable) { 1949 pdu->has_hdgst = true; 1950 psh_len += SPDK_NVME_TCP_DIGEST_LEN; 1951 if (pdu->hdr.common.plen > psh_len) { 1952 pdo = pdu->hdr.common.pdo; 1953 padding_len = pdo - psh_len; 1954 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "padding length is =%d for pdu=%p on tqpair=%p\n", padding_len, 1955 pdu, tqpair); 1956 if (padding_len > 0) { 1957 psh_len = pdo; 1958 } 1959 } 1960 } 1961 1962 psh_len -= sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 1963 /* The following will read psh + hdgest (if possbile) + padding (if posssible) */ 1964 if (pdu->psh_valid_bytes < psh_len) { 1965 rc = nvme_tcp_read_data(tqpair->sock, 1966 psh_len - pdu->psh_valid_bytes, 1967 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1968 if (rc < 0) { 1969 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1970 break; 1971 } 1972 1973 pdu->psh_valid_bytes += rc; 1974 if (pdu->psh_valid_bytes < psh_len) { 1975 return NVME_TCP_PDU_IN_PROGRESS; 1976 } 1977 } 1978 1979 pdu->hd_is_read = true; 1980 1981 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1982 spdk_nvmf_tcp_pdu_psh_handle(tqpair); 1983 break; 1984 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1985 pdu = &tqpair->pdu_in_progress; 1986 1987 /* check whether the data is valid, if not we just return */ 1988 if (!pdu->data) { 1989 return NVME_TCP_PDU_IN_PROGRESS; 1990 } 1991 1992 data_len = pdu->data_len; 1993 /* data len */ 1994 if (pdu->data_valid_bytes < data_len) { 1995 rc = nvme_tcp_read_data(tqpair->sock, data_len - pdu->data_valid_bytes, 1996 (void *)pdu->data + pdu->data_valid_bytes); 1997 if (rc < 0) { 1998 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1999 break; 2000 } 2001 2002 pdu->data_valid_bytes += rc; 2003 if (pdu->data_valid_bytes < data_len) { 2004 return NVME_TCP_PDU_IN_PROGRESS; 2005 } 2006 } 2007 2008 /* data digest */ 2009 if ((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2010 tqpair->host_ddgst_enable && (pdu->ddigest_valid_bytes < SPDK_NVME_TCP_DIGEST_LEN)) { 2011 rc = nvme_tcp_read_data(tqpair->sock, 2012 SPDK_NVME_TCP_DIGEST_LEN - pdu->ddigest_valid_bytes, 2013 pdu->data_digest + pdu->ddigest_valid_bytes); 2014 if (rc < 0) { 2015 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2016 break; 2017 } 2018 2019 pdu->ddigest_valid_bytes += rc; 2020 if (pdu->ddigest_valid_bytes < SPDK_NVME_TCP_DIGEST_LEN) { 2021 return NVME_TCP_PDU_IN_PROGRESS; 2022 } 2023 } 2024 2025 /* All of this PDU has now been read from the socket. */ 2026 spdk_nvmf_tcp_pdu_payload_handle(tqpair); 2027 break; 2028 case NVME_TCP_PDU_RECV_STATE_ERROR: 2029 rc = NVME_TCP_PDU_FATAL; 2030 break; 2031 default: 2032 assert(0); 2033 SPDK_ERRLOG("code should not come to here"); 2034 break; 2035 } 2036 } while (tqpair->recv_state != prev_state); 2037 2038 return rc; 2039 } 2040 2041 static enum spdk_nvme_data_transfer 2042 spdk_nvmf_tcp_req_get_xfer(struct nvme_tcp_req *tcp_req) { 2043 enum spdk_nvme_data_transfer xfer; 2044 struct spdk_nvme_cmd *cmd = &tcp_req->req.cmd->nvme_cmd; 2045 struct spdk_nvme_sgl_descriptor *sgl = &cmd->dptr.sgl1; 2046 2047 /* Figure out data transfer direction */ 2048 if (cmd->opc == SPDK_NVME_OPC_FABRIC) 2049 { 2050 xfer = spdk_nvme_opc_get_data_transfer(tcp_req->req.cmd->nvmf_cmd.fctype); 2051 } else 2052 { 2053 xfer = spdk_nvme_opc_get_data_transfer(cmd->opc); 2054 2055 /* Some admin commands are special cases */ 2056 if ((tcp_req->req.qpair->qid == 0) && 2057 ((cmd->opc == SPDK_NVME_OPC_GET_FEATURES) || 2058 (cmd->opc == SPDK_NVME_OPC_SET_FEATURES))) { 2059 switch (cmd->cdw10 & 0xff) { 2060 case SPDK_NVME_FEAT_LBA_RANGE_TYPE: 2061 case SPDK_NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION: 2062 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 2063 break; 2064 default: 2065 xfer = SPDK_NVME_DATA_NONE; 2066 } 2067 } 2068 } 2069 2070 if (xfer == SPDK_NVME_DATA_NONE) 2071 { 2072 return xfer; 2073 } 2074 2075 /* Even for commands that may transfer data, they could have specified 0 length. 2076 * We want those to show up with xfer SPDK_NVME_DATA_NONE. 2077 */ 2078 switch (sgl->generic.type) 2079 { 2080 case SPDK_NVME_SGL_TYPE_DATA_BLOCK: 2081 case SPDK_NVME_SGL_TYPE_BIT_BUCKET: 2082 case SPDK_NVME_SGL_TYPE_SEGMENT: 2083 case SPDK_NVME_SGL_TYPE_LAST_SEGMENT: 2084 case SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK: 2085 if (sgl->unkeyed.length == 0) { 2086 xfer = SPDK_NVME_DATA_NONE; 2087 } 2088 break; 2089 case SPDK_NVME_SGL_TYPE_KEYED_DATA_BLOCK: 2090 if (sgl->keyed.length == 0) { 2091 xfer = SPDK_NVME_DATA_NONE; 2092 } 2093 break; 2094 } 2095 2096 return xfer; 2097 } 2098 2099 static int 2100 spdk_nvmf_tcp_req_fill_iovs(struct spdk_nvmf_tcp_transport *ttransport, 2101 struct nvme_tcp_req *tcp_req) 2102 { 2103 void *buf = NULL; 2104 uint32_t length = tcp_req->req.length; 2105 uint32_t i = 0; 2106 2107 tcp_req->req.iovcnt = 0; 2108 while (length) { 2109 buf = spdk_mempool_get(ttransport->data_buf_pool); 2110 if (!buf) { 2111 goto nomem; 2112 } 2113 2114 tcp_req->req.iov[i].iov_base = (void *)((uintptr_t)(buf + NVMF_DATA_BUFFER_MASK) & 2115 ~NVMF_DATA_BUFFER_MASK); 2116 tcp_req->req.iov[i].iov_len = spdk_min(length, ttransport->io_unit_size); 2117 tcp_req->req.iovcnt++; 2118 tcp_req->buffers[i] = buf; 2119 length -= tcp_req->req.iov[i].iov_len; 2120 i++; 2121 } 2122 2123 tcp_req->data_from_pool = true; 2124 2125 return 0; 2126 2127 nomem: 2128 while (i) { 2129 i--; 2130 spdk_mempool_put(ttransport->data_buf_pool, tcp_req->req.iov[i].iov_base); 2131 tcp_req->req.iov[i].iov_base = NULL; 2132 tcp_req->req.iov[i].iov_len = 0; 2133 2134 } 2135 tcp_req->req.iovcnt = 0; 2136 return -ENOMEM; 2137 } 2138 2139 static int 2140 spdk_nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_transport *ttransport, 2141 struct nvme_tcp_req *tcp_req) 2142 { 2143 struct spdk_nvme_cmd *cmd; 2144 struct spdk_nvme_cpl *rsp; 2145 struct spdk_nvme_sgl_descriptor *sgl; 2146 2147 cmd = &tcp_req->req.cmd->nvme_cmd; 2148 rsp = &tcp_req->req.rsp->nvme_cpl; 2149 sgl = &cmd->dptr.sgl1; 2150 2151 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2152 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2153 if (sgl->unkeyed.length > ttransport->max_io_size) { 2154 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2155 sgl->unkeyed.length, ttransport->max_io_size); 2156 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2157 return -1; 2158 } 2159 2160 /* fill request length and populate iovs */ 2161 tcp_req->req.length = sgl->unkeyed.length; 2162 2163 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Data requested length= 0x%x\n", 2164 sgl->unkeyed.length); 2165 2166 if (spdk_nvmf_tcp_req_fill_iovs(ttransport, tcp_req) < 0) { 2167 /* No available buffers. Queue this request up. */ 2168 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No available large data buffers. Queueing request %p\n", tcp_req); 2169 return 0; 2170 } 2171 2172 /* backward compatible */ 2173 tcp_req->req.data = tcp_req->req.iov[0].iov_base; 2174 2175 2176 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p took %d buffer/s from central pool, and data=%p\n", 2177 tcp_req, 2178 tcp_req->req.iovcnt, tcp_req->req.data); 2179 2180 return 0; 2181 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2182 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2183 uint64_t offset = sgl->address; 2184 uint32_t max_len = ttransport->in_capsule_data_size; 2185 2186 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2187 offset, sgl->unkeyed.length); 2188 2189 if (offset > max_len) { 2190 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2191 offset, max_len); 2192 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2193 return -1; 2194 } 2195 max_len -= (uint32_t)offset; 2196 2197 if (sgl->unkeyed.length > max_len) { 2198 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2199 sgl->unkeyed.length, max_len); 2200 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2201 return -1; 2202 } 2203 2204 tcp_req->req.data = tcp_req->buf + offset; 2205 tcp_req->data_from_pool = false; 2206 tcp_req->req.length = sgl->unkeyed.length; 2207 2208 tcp_req->req.iov[0].iov_base = tcp_req->req.data; 2209 tcp_req->req.iov[0].iov_len = tcp_req->req.length; 2210 tcp_req->req.iovcnt = 1; 2211 2212 return 0; 2213 } 2214 2215 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2216 sgl->generic.type, sgl->generic.subtype); 2217 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2218 return -1; 2219 } 2220 2221 static void 2222 spdk_nvmf_tcp_send_c2h_data(struct nvme_tcp_qpair *tqpair, 2223 struct nvme_tcp_req *tcp_req) 2224 { 2225 struct nvme_tcp_pdu *rsp_pdu; 2226 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2227 uint32_t plen, pdo, alignment, offset, iov_index; 2228 2229 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2230 2231 /* always use the first iov_len, which is correct */ 2232 iov_index = tcp_req->c2h_data_offset / tcp_req->req.iov[0].iov_len; 2233 offset = tcp_req->c2h_data_offset % tcp_req->req.iov[0].iov_len; 2234 2235 rsp_pdu = spdk_nvmf_tcp_pdu_get(tqpair); 2236 assert(rsp_pdu != NULL); 2237 2238 c2h_data = &rsp_pdu->hdr.c2h_data; 2239 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2240 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2241 2242 if (tqpair->host_hdgst_enable) { 2243 plen += SPDK_NVME_TCP_DIGEST_LEN; 2244 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2245 } 2246 2247 /* set the psh */ 2248 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2249 c2h_data->datal = spdk_min(NVMF_TCP_PDU_MAX_C2H_DATA_SIZE, 2250 (tcp_req->req.iov[iov_index].iov_len - offset)); 2251 c2h_data->datao = tcp_req->c2h_data_offset; 2252 2253 /* set the padding */ 2254 rsp_pdu->padding_len = 0; 2255 pdo = plen; 2256 if (tqpair->cpda) { 2257 alignment = (tqpair->cpda + 1) << 2; 2258 if (alignment > plen) { 2259 rsp_pdu->padding_len = alignment - plen; 2260 pdo = plen = alignment; 2261 } 2262 } 2263 2264 c2h_data->common.pdo = pdo; 2265 plen += c2h_data->datal; 2266 if (tqpair->host_ddgst_enable) { 2267 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2268 plen += SPDK_NVME_TCP_DIGEST_LEN; 2269 } 2270 2271 c2h_data->common.plen = plen; 2272 2273 rsp_pdu->data = tcp_req->req.iov[iov_index].iov_base + offset; 2274 rsp_pdu->data_len = c2h_data->datal; 2275 2276 tcp_req->c2h_data_offset += c2h_data->datal; 2277 if (iov_index == (tcp_req->req.iovcnt - 1) && (tcp_req->c2h_data_offset == tcp_req->req.length)) { 2278 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Last pdu for tcp_req=%p on tqpair=%p\n", tcp_req, tqpair); 2279 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2280 /* The linux kernel does not support this yet */ 2281 #if LINUX_KERNEL_SUPPORT_NOT_SENDING_RESP_FOR_C2H 2282 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2283 #endif 2284 TAILQ_REMOVE(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 2285 } 2286 2287 tqpair->c2h_data_pdu_cnt += 1; 2288 spdk_nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, spdk_nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2289 } 2290 2291 static int 2292 spdk_nvmf_tcp_calc_c2h_data_pdu_num(struct nvme_tcp_req *tcp_req) 2293 { 2294 uint32_t i, iov_cnt, pdu_num = 0; 2295 2296 iov_cnt = tcp_req->req.iovcnt; 2297 for (i = 0; i < iov_cnt; i++) { 2298 pdu_num += (tcp_req->req.iov[i].iov_len + NVMF_TCP_PDU_MAX_C2H_DATA_SIZE - 1) / 2299 NVMF_TCP_PDU_MAX_C2H_DATA_SIZE; 2300 } 2301 2302 return pdu_num; 2303 } 2304 2305 static void 2306 spdk_nvmf_tcp_handle_pending_c2h_data_queue(struct nvme_tcp_qpair *tqpair) 2307 { 2308 struct nvme_tcp_req *tcp_req; 2309 2310 while (!TAILQ_EMPTY(&tqpair->queued_c2h_data_tcp_req) && 2311 (tqpair->c2h_data_pdu_cnt < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM)) { 2312 tcp_req = TAILQ_FIRST(&tqpair->queued_c2h_data_tcp_req); 2313 spdk_nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2314 } 2315 } 2316 2317 static void 2318 spdk_nvmf_tcp_queue_c2h_data(struct nvme_tcp_req *tcp_req, 2319 struct nvme_tcp_qpair *tqpair) 2320 { 2321 tcp_req->c2h_data_pdu_num = spdk_nvmf_tcp_calc_c2h_data_pdu_num(tcp_req); 2322 2323 assert(tcp_req->c2h_data_pdu_num < NVMF_TCP_QPAIR_MAX_C2H_PDU_NUM); 2324 2325 TAILQ_INSERT_TAIL(&tqpair->queued_c2h_data_tcp_req, tcp_req, link); 2326 spdk_nvmf_tcp_handle_pending_c2h_data_queue(tqpair); 2327 } 2328 2329 static int 2330 request_transfer_out(struct spdk_nvmf_request *req) 2331 { 2332 struct nvme_tcp_req *tcp_req; 2333 struct spdk_nvmf_qpair *qpair; 2334 struct nvme_tcp_qpair *tqpair; 2335 struct spdk_nvme_cpl *rsp; 2336 2337 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2338 2339 qpair = req->qpair; 2340 rsp = &req->rsp->nvme_cpl; 2341 tcp_req = SPDK_CONTAINEROF(req, struct nvme_tcp_req, req); 2342 2343 /* Advance our sq_head pointer */ 2344 if (qpair->sq_head == qpair->sq_head_max) { 2345 qpair->sq_head = 0; 2346 } else { 2347 qpair->sq_head++; 2348 } 2349 rsp->sqhd = qpair->sq_head; 2350 2351 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct nvme_tcp_qpair, qpair); 2352 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2353 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && 2354 req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2355 spdk_nvmf_tcp_queue_c2h_data(tcp_req, tqpair); 2356 } else { 2357 spdk_nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2358 } 2359 2360 return 0; 2361 } 2362 2363 static void 2364 spdk_nvmf_tcp_pdu_set_buf_from_req(struct spdk_nvmf_tcp_transport *ttransport, 2365 struct nvme_tcp_qpair *tqpair, 2366 struct nvme_tcp_req *tcp_req) 2367 { 2368 struct nvme_tcp_pdu *pdu; 2369 uint32_t plen = 0; 2370 2371 pdu = &tqpair->pdu_in_progress; 2372 2373 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "tqpair=%p\n", tqpair); 2374 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 2375 2376 plen = pdu->hdr.common.hlen; 2377 if (tqpair->host_hdgst_enable) { 2378 plen += SPDK_NVME_TCP_DIGEST_LEN; 2379 } 2380 2381 /* need to send r2t for data */ 2382 if (pdu->hdr.common.plen == plen) { 2383 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2384 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Will send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2385 tcp_req->next_expected_r2t_offset = 0; 2386 TAILQ_INSERT_TAIL(&tqpair->queued_r2t_tcp_req, tcp_req, link); 2387 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T); 2388 2389 } else { 2390 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2391 tqpair); 2392 /* No need to send r2t, contained in the capsuled data */ 2393 pdu->data = tcp_req->req.data; 2394 pdu->data_len = tcp_req->req.length; 2395 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2396 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2397 } 2398 } 2399 2400 static bool 2401 spdk_nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2402 struct nvme_tcp_req *tcp_req) 2403 { 2404 struct nvme_tcp_qpair *tqpair; 2405 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2406 int rc; 2407 enum spdk_nvmf_tcp_req_state prev_state; 2408 bool progress = false; 2409 2410 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct nvme_tcp_qpair, qpair); 2411 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2412 2413 /* The loop here is to allow for several back-to-back state changes. */ 2414 do { 2415 prev_state = tcp_req->state; 2416 2417 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2418 tqpair); 2419 2420 switch (tcp_req->state) { 2421 case TCP_REQUEST_STATE_FREE: 2422 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2423 * to escape this state. */ 2424 break; 2425 case TCP_REQUEST_STATE_NEW: 2426 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2427 2428 /* copy the cmd from the receive pdu */ 2429 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2430 2431 /* The next state transition depends on the data transfer needs of this request. */ 2432 tcp_req->req.xfer = spdk_nvmf_tcp_req_get_xfer(tcp_req); 2433 2434 /* If no data to transfer, ready to execute. */ 2435 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2436 /* Reset the tqpair receving pdu state */ 2437 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2438 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2439 break; 2440 } 2441 2442 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2443 TAILQ_INSERT_TAIL(&tqpair->ch->pending_data_buf_queue, tcp_req, link); 2444 break; 2445 case TCP_REQUEST_STATE_NEED_BUFFER: 2446 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2447 2448 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2449 2450 if (tcp_req != TAILQ_FIRST(&tqpair->ch->pending_data_buf_queue)) { 2451 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, 2452 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2453 tcp_req, tqpair); 2454 /* This request needs to wait in line to obtain a buffer */ 2455 break; 2456 } 2457 2458 /* Try to get a data buffer */ 2459 rc = spdk_nvmf_tcp_req_parse_sgl(ttransport, tcp_req); 2460 if (rc < 0) { 2461 TAILQ_REMOVE(&tqpair->ch->pending_data_buf_queue, tcp_req, link); 2462 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2463 /* Reset the tqpair receving pdu state */ 2464 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2465 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2466 break; 2467 } 2468 2469 if (!tcp_req->req.data) { 2470 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2471 tcp_req, tqpair); 2472 /* No buffers available. */ 2473 break; 2474 } 2475 2476 TAILQ_REMOVE(&tqpair->ch->pending_data_buf_queue, tcp_req, link); 2477 2478 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2479 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2480 spdk_nvmf_tcp_pdu_set_buf_from_req(ttransport, tqpair, tcp_req); 2481 break; 2482 } 2483 2484 spdk_nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2485 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2486 break; 2487 case TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T: 2488 spdk_trace_record(TCP_REQUEST_STATE_DATA_PENDING_FOR_R2T, 0, 0, 2489 (uintptr_t)tcp_req, 0); 2490 2491 if (tqpair->pending_r2t < tqpair->maxr2t) { 2492 tqpair->pending_r2t++; 2493 TAILQ_REMOVE(&tqpair->queued_r2t_tcp_req, tcp_req, link); 2494 spdk_nvmf_tcp_send_r2t_pdu(ttransport, tqpair, tcp_req); 2495 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2496 } 2497 2498 break; 2499 2500 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2501 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2502 (uintptr_t)tcp_req, 0); 2503 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2504 * to escape this state. */ 2505 break; 2506 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2507 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2508 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2509 spdk_nvmf_request_exec(&tcp_req->req); 2510 break; 2511 case TCP_REQUEST_STATE_EXECUTING: 2512 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2513 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2514 * to escape this state. */ 2515 break; 2516 case TCP_REQUEST_STATE_EXECUTED: 2517 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2518 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2519 break; 2520 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2521 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2522 rc = request_transfer_out(&tcp_req->req); 2523 assert(rc == 0); /* No good way to handle this currently */ 2524 break; 2525 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2526 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2527 (uintptr_t)tcp_req, 2528 0); 2529 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2530 * to escape this state. */ 2531 break; 2532 case TCP_REQUEST_STATE_COMPLETED: 2533 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2534 if (tcp_req->data_from_pool) { 2535 /* Put the buffer/s back in the pool */ 2536 for (uint32_t i = 0; i < tcp_req->req.iovcnt; i++) { 2537 spdk_mempool_put(ttransport->data_buf_pool, tcp_req->buffers[i]); 2538 tcp_req->req.iov[i].iov_base = NULL; 2539 tcp_req->buffers[i] = NULL; 2540 } 2541 tcp_req->data_from_pool = false; 2542 } 2543 tcp_req->req.length = 0; 2544 tcp_req->req.iovcnt = 0; 2545 tcp_req->req.data = NULL; 2546 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 2547 break; 2548 case TCP_REQUEST_NUM_STATES: 2549 default: 2550 assert(0); 2551 break; 2552 } 2553 2554 if (tcp_req->state != prev_state) { 2555 progress = true; 2556 } 2557 } while (tcp_req->state != prev_state); 2558 2559 return progress; 2560 } 2561 2562 static void 2563 spdk_nvmf_tcp_qpair_process_pending(struct spdk_nvmf_tcp_transport *ttransport, 2564 struct nvme_tcp_qpair *tqpair) 2565 { 2566 struct nvme_tcp_req *tcp_req, *req_tmp; 2567 2568 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->queued_r2t_tcp_req, link, req_tmp) { 2569 if (tqpair->pending_r2t < tqpair->maxr2t) { 2570 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2571 break; 2572 } 2573 } 2574 } 2575 2576 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->ch->pending_data_buf_queue, link, req_tmp) { 2577 if (spdk_nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2578 break; 2579 } 2580 } 2581 } 2582 2583 static void 2584 spdk_nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2585 { 2586 struct nvme_tcp_qpair *tqpair = arg; 2587 struct spdk_nvmf_tcp_transport *ttransport; 2588 int rc; 2589 2590 assert(tqpair != NULL); 2591 2592 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR) { 2593 return; 2594 } 2595 2596 ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2597 spdk_nvmf_tcp_qpair_process_pending(ttransport, tqpair); 2598 2599 rc = spdk_nvmf_tcp_sock_process(tqpair); 2600 if (rc < 0) { 2601 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 2602 spdk_nvmf_tcp_qpair_flush_pdus(tqpair); 2603 } 2604 } 2605 2606 static int 2607 spdk_nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2608 struct spdk_nvmf_qpair *qpair) 2609 { 2610 struct spdk_nvmf_tcp_poll_group *tgroup; 2611 struct nvme_tcp_qpair *tqpair; 2612 int rc; 2613 2614 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2615 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 2616 2617 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2618 spdk_nvmf_tcp_sock_cb, tqpair); 2619 if (rc != 0) { 2620 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2621 spdk_strerror(errno), errno); 2622 spdk_nvmf_tcp_qpair_destroy(tqpair); 2623 return -1; 2624 } 2625 2626 rc = spdk_nvmf_tcp_qpair_sock_init(tqpair); 2627 if (rc != 0) { 2628 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2629 spdk_nvmf_tcp_qpair_destroy(tqpair); 2630 return -1; 2631 } 2632 2633 rc = spdk_nvmf_tcp_qpair_init(&tqpair->qpair); 2634 if (rc < 0) { 2635 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2636 spdk_nvmf_tcp_qpair_destroy(tqpair); 2637 return -1; 2638 } 2639 2640 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, 1); 2641 if (rc < 0) { 2642 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2643 spdk_nvmf_tcp_qpair_destroy(tqpair); 2644 return -1; 2645 } 2646 2647 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2648 tqpair->timeout = SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT; 2649 tqpair->last_pdu_time = spdk_get_ticks(); 2650 2651 tqpair->tgroup = tgroup; 2652 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2653 2654 return 0; 2655 } 2656 2657 static int 2658 spdk_nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2659 { 2660 struct spdk_nvmf_tcp_transport *ttransport; 2661 struct nvme_tcp_req *tcp_req; 2662 2663 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2664 tcp_req = SPDK_CONTAINEROF(req, struct nvme_tcp_req, req); 2665 2666 spdk_nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2667 spdk_nvmf_tcp_req_process(ttransport, tcp_req); 2668 2669 return 0; 2670 } 2671 2672 static void 2673 spdk_nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair) 2674 { 2675 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "enter\n"); 2676 2677 spdk_nvmf_tcp_qpair_destroy(SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair)); 2678 } 2679 2680 static void 2681 spdk_nvmf_tcp_qpair_handle_timout(struct nvme_tcp_qpair *tqpair) 2682 { 2683 uint64_t tsc; 2684 2685 if ((tqpair->state == NVME_TCP_QPAIR_STATE_EXITING) || 2686 (tqpair->state == NVME_TCP_QPAIR_STATE_EXITED)) { 2687 return; 2688 } 2689 2690 /* Currently, we did not have keep alive support, so make sure that we should have the generic support later */ 2691 if (tqpair->recv_state != NVME_TCP_PDU_RECV_STATE_ERROR) { 2692 return; 2693 } 2694 2695 /* Check for interval expiration */ 2696 tsc = spdk_get_ticks(); 2697 if ((tsc - tqpair->last_pdu_time) > (tqpair->timeout * spdk_get_ticks_hz())) { 2698 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, tqpair->timeout); 2699 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 2700 } 2701 } 2702 2703 static int 2704 spdk_nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2705 { 2706 struct spdk_nvmf_tcp_poll_group *tgroup; 2707 struct nvme_tcp_qpair *tqpair, *tmp; 2708 int rc; 2709 2710 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2711 2712 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs))) { 2713 return 0; 2714 } 2715 2716 rc = spdk_sock_group_poll(tgroup->sock_group); 2717 if (rc < 0) { 2718 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2719 return rc; 2720 } 2721 2722 2723 TAILQ_FOREACH_SAFE(tqpair, &tgroup->qpairs, link, tmp) { 2724 spdk_nvmf_tcp_qpair_handle_timout(tqpair); 2725 if (tqpair->state == NVME_TCP_QPAIR_STATE_EXITING) { 2726 /* to prevent the state is set again */ 2727 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2728 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "will disconect the tqpair=%p\n", tqpair); 2729 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 2730 } 2731 } 2732 2733 return 0; 2734 } 2735 2736 static bool 2737 spdk_nvmf_tcp_qpair_is_idle(struct spdk_nvmf_qpair *qpair) 2738 { 2739 struct nvme_tcp_qpair *tqpair; 2740 2741 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 2742 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] == tqpair->max_queue_depth) { 2743 return true; 2744 } 2745 2746 return false; 2747 } 2748 2749 static int 2750 spdk_nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2751 struct spdk_nvme_transport_id *trid, bool peer) 2752 { 2753 struct nvme_tcp_qpair *tqpair; 2754 uint16_t port; 2755 2756 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 2757 trid->trtype = SPDK_NVME_TRANSPORT_TCP; 2758 2759 if (peer) { 2760 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2761 port = tqpair->initiator_port; 2762 } else { 2763 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2764 port = tqpair->target_port; 2765 } 2766 2767 if (spdk_sock_is_ipv4(tqpair->sock)) { 2768 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2769 } else if (spdk_sock_is_ipv4(tqpair->sock)) { 2770 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2771 } else { 2772 return -1; 2773 } 2774 2775 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2776 return 0; 2777 } 2778 2779 static int 2780 spdk_nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2781 struct spdk_nvme_transport_id *trid) 2782 { 2783 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2784 } 2785 2786 static int 2787 spdk_nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2788 struct spdk_nvme_transport_id *trid) 2789 { 2790 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2791 } 2792 2793 static int 2794 spdk_nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2795 struct spdk_nvme_transport_id *trid) 2796 { 2797 return spdk_nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2798 } 2799 2800 static int 2801 spdk_nvmf_tcp_qpair_set_sq_size(struct spdk_nvmf_qpair *qpair) 2802 { 2803 struct nvme_tcp_qpair *tqpair; 2804 int rc; 2805 tqpair = SPDK_CONTAINEROF(qpair, struct nvme_tcp_qpair, qpair); 2806 2807 rc = spdk_nvmf_tcp_qpair_init_mem_resource(tqpair, tqpair->qpair.sq_head_max); 2808 if (!rc) { 2809 tqpair->max_queue_depth += tqpair->qpair.sq_head_max; 2810 tqpair->free_pdu_num += tqpair->qpair.sq_head_max; 2811 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] += tqpair->qpair.sq_head_max; 2812 SPDK_DEBUGLOG(SPDK_LOG_NVMF_TCP, "The queue depth=%u for tqpair=%p\n", 2813 tqpair->max_queue_depth, tqpair); 2814 } 2815 2816 return rc; 2817 2818 } 2819 2820 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2821 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2822 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 64 2823 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2824 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2825 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2826 2827 static void 2828 spdk_nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2829 { 2830 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2831 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2832 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2833 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2834 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2835 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2836 } 2837 2838 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2839 .type = SPDK_NVME_TRANSPORT_TCP, 2840 .opts_init = spdk_nvmf_tcp_opts_init, 2841 .create = spdk_nvmf_tcp_create, 2842 .destroy = spdk_nvmf_tcp_destroy, 2843 2844 .listen = spdk_nvmf_tcp_listen, 2845 .stop_listen = spdk_nvmf_tcp_stop_listen, 2846 .accept = spdk_nvmf_tcp_accept, 2847 2848 .listener_discover = spdk_nvmf_tcp_discover, 2849 2850 .poll_group_create = spdk_nvmf_tcp_poll_group_create, 2851 .poll_group_destroy = spdk_nvmf_tcp_poll_group_destroy, 2852 .poll_group_add = spdk_nvmf_tcp_poll_group_add, 2853 .poll_group_poll = spdk_nvmf_tcp_poll_group_poll, 2854 2855 .req_free = spdk_nvmf_tcp_req_free, 2856 .req_complete = spdk_nvmf_tcp_req_complete, 2857 2858 .qpair_fini = spdk_nvmf_tcp_close_qpair, 2859 .qpair_is_idle = spdk_nvmf_tcp_qpair_is_idle, 2860 .qpair_get_local_trid = spdk_nvmf_tcp_qpair_get_local_trid, 2861 .qpair_get_peer_trid = spdk_nvmf_tcp_qpair_get_peer_trid, 2862 .qpair_get_listen_trid = spdk_nvmf_tcp_qpair_get_listen_trid, 2863 .qpair_set_sqsize = spdk_nvmf_tcp_qpair_set_sq_size, 2864 }; 2865 2866 SPDK_LOG_REGISTER_COMPONENT("nvmf_tcp", SPDK_LOG_NVMF_TCP) 2867