1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/accel_engine.h" 35 #include "spdk/stdinc.h" 36 #include "spdk/crc32.h" 37 #include "spdk/endian.h" 38 #include "spdk/assert.h" 39 #include "spdk/thread.h" 40 #include "spdk/nvmf_transport.h" 41 #include "spdk/string.h" 42 #include "spdk/trace.h" 43 #include "spdk/util.h" 44 #include "spdk/log.h" 45 46 #include "spdk_internal/assert.h" 47 #include "spdk_internal/nvme_tcp.h" 48 #include "spdk_internal/sock.h" 49 50 #include "nvmf_internal.h" 51 52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 54 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 55 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 56 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 57 58 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 59 60 /* spdk nvmf related structure */ 61 enum spdk_nvmf_tcp_req_state { 62 63 /* The request is not currently in use */ 64 TCP_REQUEST_STATE_FREE = 0, 65 66 /* Initial state when request first received */ 67 TCP_REQUEST_STATE_NEW, 68 69 /* The request is queued until a data buffer is available. */ 70 TCP_REQUEST_STATE_NEED_BUFFER, 71 72 /* The request is currently transferring data from the host to the controller. */ 73 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 74 75 /* The request is waiting for the R2T send acknowledgement. */ 76 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 77 78 /* The request is ready to execute at the block device */ 79 TCP_REQUEST_STATE_READY_TO_EXECUTE, 80 81 /* The request is currently executing at the block device */ 82 TCP_REQUEST_STATE_EXECUTING, 83 84 /* The request finished executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTED, 86 87 /* The request is ready to send a completion */ 88 TCP_REQUEST_STATE_READY_TO_COMPLETE, 89 90 /* The request is currently transferring final pdus from the controller to the host. */ 91 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 92 93 /* The request completed and can be marked free. */ 94 TCP_REQUEST_STATE_COMPLETED, 95 96 /* Terminator */ 97 TCP_REQUEST_NUM_STATES, 98 }; 99 100 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 101 "Invalid PDU Header Field", 102 "PDU Sequence Error", 103 "Header Digiest Error", 104 "Data Transfer Out of Range", 105 "R2T Limit Exceeded", 106 "Unsupported parameter", 107 }; 108 109 #define OBJECT_NVMF_TCP_IO 0x80 110 111 #define TRACE_GROUP_NVMF_TCP 0x5 112 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 113 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 115 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 117 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 118 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 119 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 120 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 121 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 122 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 123 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 124 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 132 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 133 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 136 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 139 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_EXECUTING", 142 TRACE_TCP_REQUEST_STATE_EXECUTING, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTED", 145 TRACE_TCP_REQUEST_STATE_EXECUTED, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 148 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 151 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_REQ_COMPLETED", 154 TRACE_TCP_REQUEST_STATE_COMPLETED, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 156 spdk_trace_register_description("TCP_WRITE_START", 157 TRACE_TCP_FLUSH_WRITEBUF_START, 158 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 159 spdk_trace_register_description("TCP_WRITE_DONE", 160 TRACE_TCP_FLUSH_WRITEBUF_DONE, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 spdk_trace_register_description("TCP_READ_DONE", 163 TRACE_TCP_READ_FROM_SOCKET_DONE, 164 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 165 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 166 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 167 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 168 } 169 170 struct spdk_nvmf_tcp_req { 171 struct spdk_nvmf_request req; 172 struct spdk_nvme_cpl rsp; 173 struct spdk_nvme_cmd cmd; 174 175 /* A PDU that can be used for sending responses. This is 176 * not the incoming PDU! */ 177 struct nvme_tcp_pdu *pdu; 178 179 /* In-capsule data buffer */ 180 uint8_t *buf; 181 /* 182 * The PDU for a request may be used multiple times in serial over 183 * the request's lifetime. For example, first to send an R2T, then 184 * to send a completion. To catch mistakes where the PDU is used 185 * twice at the same time, add a debug flag here for init/fini. 186 */ 187 bool pdu_in_use; 188 bool has_incapsule_data; 189 190 /* transfer_tag */ 191 uint16_t ttag; 192 193 enum spdk_nvmf_tcp_req_state state; 194 195 /* 196 * h2c_offset is used when we receive the h2c_data PDU. 197 */ 198 uint32_t h2c_offset; 199 200 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 201 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 202 }; 203 204 struct spdk_nvmf_tcp_qpair { 205 struct spdk_nvmf_qpair qpair; 206 struct spdk_nvmf_tcp_poll_group *group; 207 struct spdk_sock *sock; 208 209 enum nvme_tcp_pdu_recv_state recv_state; 210 enum nvme_tcp_qpair_state state; 211 212 /* PDU being actively received */ 213 struct nvme_tcp_pdu pdu_in_progress; 214 215 /* Queues to track the requests in all states */ 216 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 217 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 218 219 /* Number of requests in each state */ 220 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 221 222 uint8_t cpda; 223 224 bool host_hdgst_enable; 225 bool host_ddgst_enable; 226 227 /* This is a spare PDU used for sending special management 228 * operations. Primarily, this is used for the initial 229 * connection response and c2h termination request. */ 230 struct nvme_tcp_pdu *mgmt_pdu; 231 232 /* Arrays of in-capsule buffers, requests, and pdus. 233 * Each array is 'resource_count' number of elements */ 234 void *bufs; 235 struct spdk_nvmf_tcp_req *reqs; 236 struct nvme_tcp_pdu *pdus; 237 uint32_t resource_count; 238 uint32_t recv_buf_size; 239 240 struct spdk_nvmf_tcp_port *port; 241 242 /* IP address */ 243 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 244 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 245 246 /* IP port */ 247 uint16_t initiator_port; 248 uint16_t target_port; 249 250 /* Timer used to destroy qpair after detecting transport error issue if initiator does 251 * not close the connection. 252 */ 253 struct spdk_poller *timeout_poller; 254 255 struct spdk_io_channel *accel_channel; 256 257 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 258 }; 259 260 struct spdk_nvmf_tcp_control_msg { 261 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 262 }; 263 264 struct spdk_nvmf_tcp_control_msg_list { 265 void *msg_buf; 266 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 267 }; 268 269 struct spdk_nvmf_tcp_poll_group { 270 struct spdk_nvmf_transport_poll_group group; 271 struct spdk_sock_group *sock_group; 272 273 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 274 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 275 276 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 277 }; 278 279 struct spdk_nvmf_tcp_port { 280 const struct spdk_nvme_transport_id *trid; 281 struct spdk_sock *listen_sock; 282 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 283 }; 284 285 struct tcp_transport_opts { 286 bool c2h_success; 287 uint16_t control_msg_num; 288 uint32_t sock_priority; 289 }; 290 291 struct spdk_nvmf_tcp_transport { 292 struct spdk_nvmf_transport transport; 293 struct tcp_transport_opts tcp_opts; 294 295 pthread_mutex_t lock; 296 297 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 298 }; 299 300 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 301 { 302 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 303 spdk_json_decode_bool, true 304 }, 305 { 306 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 307 spdk_json_decode_uint16, true 308 }, 309 { 310 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 311 spdk_json_decode_uint32, true 312 }, 313 }; 314 315 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 316 struct spdk_nvmf_tcp_req *tcp_req); 317 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 318 319 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 320 struct spdk_nvmf_tcp_req *tcp_req); 321 322 static void 323 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 324 enum spdk_nvmf_tcp_req_state state) 325 { 326 struct spdk_nvmf_qpair *qpair; 327 struct spdk_nvmf_tcp_qpair *tqpair; 328 329 qpair = tcp_req->req.qpair; 330 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 331 332 assert(tqpair->state_cntr[tcp_req->state] > 0); 333 tqpair->state_cntr[tcp_req->state]--; 334 tqpair->state_cntr[state]++; 335 336 tcp_req->state = state; 337 } 338 339 static inline struct nvme_tcp_pdu * 340 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 341 { 342 assert(tcp_req->pdu_in_use == false); 343 tcp_req->pdu_in_use = true; 344 345 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 346 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 347 348 return tcp_req->pdu; 349 } 350 351 static inline void 352 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 353 { 354 tcp_req->pdu_in_use = false; 355 } 356 357 static struct spdk_nvmf_tcp_req * 358 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 359 { 360 struct spdk_nvmf_tcp_req *tcp_req; 361 362 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 363 if (!tcp_req) { 364 return NULL; 365 } 366 367 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 368 tcp_req->h2c_offset = 0; 369 tcp_req->has_incapsule_data = false; 370 tcp_req->req.dif.dif_insert_or_strip = false; 371 372 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 373 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 374 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 375 return tcp_req; 376 } 377 378 static inline void 379 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 380 { 381 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 382 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 384 } 385 386 static void 387 nvmf_tcp_request_free(void *cb_arg) 388 { 389 struct spdk_nvmf_tcp_transport *ttransport; 390 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 391 392 assert(tcp_req != NULL); 393 394 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 395 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 396 struct spdk_nvmf_tcp_transport, transport); 397 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 398 nvmf_tcp_req_process(ttransport, tcp_req); 399 } 400 401 static int 402 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 403 { 404 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 405 406 nvmf_tcp_request_free(tcp_req); 407 408 return 0; 409 } 410 411 static void 412 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 413 enum spdk_nvmf_tcp_req_state state) 414 { 415 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 416 417 assert(state != TCP_REQUEST_STATE_FREE); 418 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 419 if (state == tcp_req->state) { 420 nvmf_tcp_request_free(tcp_req); 421 } 422 } 423 } 424 425 static void 426 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 427 { 428 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 429 430 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 431 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 432 433 /* Wipe the requests waiting for buffer from the global list */ 434 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 435 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 436 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 437 spdk_nvmf_request, buf_link); 438 } 439 } 440 441 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 442 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 443 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 444 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 445 } 446 447 static void 448 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 449 { 450 int i; 451 struct spdk_nvmf_tcp_req *tcp_req; 452 453 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 454 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 455 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 456 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 457 if ((int)tcp_req->state == i) { 458 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 459 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 460 } 461 } 462 } 463 } 464 465 static void 466 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 467 { 468 int err = 0; 469 470 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 471 472 err = spdk_sock_close(&tqpair->sock); 473 assert(err == 0); 474 nvmf_tcp_cleanup_all_states(tqpair); 475 476 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 477 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 478 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 479 tqpair->resource_count); 480 err++; 481 } 482 483 if (err > 0) { 484 nvmf_tcp_dump_qpair_req_contents(tqpair); 485 } 486 487 if (tqpair->accel_channel) { 488 spdk_put_io_channel(tqpair->accel_channel); 489 } 490 spdk_dma_free(tqpair->pdus); 491 free(tqpair->reqs); 492 spdk_free(tqpair->bufs); 493 free(tqpair); 494 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 495 } 496 497 static void 498 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 499 { 500 struct spdk_nvmf_tcp_transport *ttransport; 501 assert(w != NULL); 502 503 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 504 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 505 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 506 } 507 508 static int 509 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 510 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 511 { 512 struct spdk_nvmf_tcp_transport *ttransport; 513 514 assert(transport != NULL); 515 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 516 517 pthread_mutex_destroy(&ttransport->lock); 518 free(ttransport); 519 520 if (cb_fn) { 521 cb_fn(cb_arg); 522 } 523 return 0; 524 } 525 526 static struct spdk_nvmf_transport * 527 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 528 { 529 struct spdk_nvmf_tcp_transport *ttransport; 530 uint32_t sge_count; 531 uint32_t min_shared_buffers; 532 533 ttransport = calloc(1, sizeof(*ttransport)); 534 if (!ttransport) { 535 return NULL; 536 } 537 538 TAILQ_INIT(&ttransport->ports); 539 540 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 541 542 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 543 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 544 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 545 if (opts->transport_specific != NULL && 546 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 547 SPDK_COUNTOF(tcp_transport_opts_decoder), 548 &ttransport->tcp_opts)) { 549 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 550 free(ttransport); 551 return NULL; 552 } 553 554 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 555 556 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 557 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 558 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 559 " in_capsule_data_size=%d, max_aq_depth=%d\n" 560 " num_shared_buffers=%d, c2h_success=%d,\n" 561 " dif_insert_or_strip=%d, sock_priority=%d\n" 562 " abort_timeout_sec=%d, control_msg_num=%hu\n", 563 opts->max_queue_depth, 564 opts->max_io_size, 565 opts->max_qpairs_per_ctrlr - 1, 566 opts->io_unit_size, 567 opts->in_capsule_data_size, 568 opts->max_aq_depth, 569 opts->num_shared_buffers, 570 ttransport->tcp_opts.c2h_success, 571 opts->dif_insert_or_strip, 572 ttransport->tcp_opts.sock_priority, 573 opts->abort_timeout_sec, 574 ttransport->tcp_opts.control_msg_num); 575 576 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 577 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 578 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 579 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 580 free(ttransport); 581 return NULL; 582 } 583 584 if (ttransport->tcp_opts.control_msg_num == 0 && 585 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 586 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 587 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 588 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 589 } 590 591 /* I/O unit size cannot be larger than max I/O size */ 592 if (opts->io_unit_size > opts->max_io_size) { 593 opts->io_unit_size = opts->max_io_size; 594 } 595 596 sge_count = opts->max_io_size / opts->io_unit_size; 597 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 598 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 599 free(ttransport); 600 return NULL; 601 } 602 603 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 604 if (min_shared_buffers > opts->num_shared_buffers) { 605 SPDK_ERRLOG("There are not enough buffers to satisfy" 606 "per-poll group caches for each thread. (%" PRIu32 ")" 607 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 608 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 609 nvmf_tcp_destroy(&ttransport->transport, NULL, NULL); 610 return NULL; 611 } 612 613 pthread_mutex_init(&ttransport->lock, NULL); 614 615 return &ttransport->transport; 616 } 617 618 static int 619 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 620 { 621 unsigned long long ull; 622 char *end = NULL; 623 624 ull = strtoull(trsvcid, &end, 10); 625 if (end == NULL || end == trsvcid || *end != '\0') { 626 return -1; 627 } 628 629 /* Valid TCP/IP port numbers are in [0, 65535] */ 630 if (ull > 65535) { 631 return -1; 632 } 633 634 return (int)ull; 635 } 636 637 /** 638 * Canonicalize a listen address trid. 639 */ 640 static int 641 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 642 const struct spdk_nvme_transport_id *trid) 643 { 644 int trsvcid_int; 645 646 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 647 if (trsvcid_int < 0) { 648 return -EINVAL; 649 } 650 651 memset(canon_trid, 0, sizeof(*canon_trid)); 652 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 653 canon_trid->adrfam = trid->adrfam; 654 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 655 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 656 657 return 0; 658 } 659 660 /** 661 * Find an existing listening port. 662 * 663 * Caller must hold ttransport->lock. 664 */ 665 static struct spdk_nvmf_tcp_port * 666 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 667 const struct spdk_nvme_transport_id *trid) 668 { 669 struct spdk_nvme_transport_id canon_trid; 670 struct spdk_nvmf_tcp_port *port; 671 672 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 673 return NULL; 674 } 675 676 TAILQ_FOREACH(port, &ttransport->ports, link) { 677 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 678 return port; 679 } 680 } 681 682 return NULL; 683 } 684 685 static int 686 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 687 struct spdk_nvmf_listen_opts *listen_opts) 688 { 689 struct spdk_nvmf_tcp_transport *ttransport; 690 struct spdk_nvmf_tcp_port *port; 691 int trsvcid_int; 692 uint8_t adrfam; 693 struct spdk_sock_opts opts; 694 695 if (!strlen(trid->trsvcid)) { 696 SPDK_ERRLOG("Service id is required\n"); 697 return -EINVAL; 698 } 699 700 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 701 702 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 703 if (trsvcid_int < 0) { 704 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 705 return -EINVAL; 706 } 707 708 pthread_mutex_lock(&ttransport->lock); 709 port = calloc(1, sizeof(*port)); 710 if (!port) { 711 SPDK_ERRLOG("Port allocation failed\n"); 712 pthread_mutex_unlock(&ttransport->lock); 713 return -ENOMEM; 714 } 715 716 port->trid = trid; 717 opts.opts_size = sizeof(opts); 718 spdk_sock_get_default_opts(&opts); 719 opts.priority = ttransport->tcp_opts.sock_priority; 720 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 721 NULL, &opts); 722 if (port->listen_sock == NULL) { 723 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 724 trid->traddr, trsvcid_int, 725 spdk_strerror(errno), errno); 726 free(port); 727 pthread_mutex_unlock(&ttransport->lock); 728 return -errno; 729 } 730 731 if (spdk_sock_is_ipv4(port->listen_sock)) { 732 adrfam = SPDK_NVMF_ADRFAM_IPV4; 733 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 734 adrfam = SPDK_NVMF_ADRFAM_IPV6; 735 } else { 736 SPDK_ERRLOG("Unhandled socket type\n"); 737 adrfam = 0; 738 } 739 740 if (adrfam != trid->adrfam) { 741 SPDK_ERRLOG("Socket address family mismatch\n"); 742 spdk_sock_close(&port->listen_sock); 743 free(port); 744 pthread_mutex_unlock(&ttransport->lock); 745 return -EINVAL; 746 } 747 748 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 749 trid->traddr, trid->trsvcid); 750 751 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 752 pthread_mutex_unlock(&ttransport->lock); 753 return 0; 754 } 755 756 static void 757 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 758 const struct spdk_nvme_transport_id *trid) 759 { 760 struct spdk_nvmf_tcp_transport *ttransport; 761 struct spdk_nvmf_tcp_port *port; 762 763 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 764 765 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 766 trid->traddr, trid->trsvcid); 767 768 pthread_mutex_lock(&ttransport->lock); 769 port = nvmf_tcp_find_port(ttransport, trid); 770 if (port) { 771 TAILQ_REMOVE(&ttransport->ports, port, link); 772 spdk_sock_close(&port->listen_sock); 773 free(port); 774 } 775 776 pthread_mutex_unlock(&ttransport->lock); 777 } 778 779 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 780 enum nvme_tcp_pdu_recv_state state); 781 782 static void 783 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 784 { 785 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 786 787 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 788 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 789 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 790 spdk_poller_unregister(&tqpair->timeout_poller); 791 792 /* This will end up calling nvmf_tcp_close_qpair */ 793 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 794 } 795 } 796 797 static void 798 _pdu_write_done(void *_pdu, int err) 799 { 800 struct nvme_tcp_pdu *pdu = _pdu; 801 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 802 803 if (err != 0) { 804 nvmf_tcp_qpair_disconnect(tqpair); 805 return; 806 } 807 808 assert(pdu->cb_fn != NULL); 809 pdu->cb_fn(pdu->cb_arg); 810 } 811 812 static void 813 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 814 { 815 uint32_t mapped_length = 0; 816 ssize_t rc; 817 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 818 uint32_t crc32c; 819 820 /* Data Digest */ 821 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 822 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 823 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 824 } 825 826 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 827 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 828 &mapped_length); 829 pdu->sock_req.cb_fn = _pdu_write_done; 830 pdu->sock_req.cb_arg = pdu; 831 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 832 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 833 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 834 if (rc == mapped_length) { 835 _pdu_write_done(pdu, 0); 836 } else { 837 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 838 _pdu_write_done(pdu, -1); 839 } 840 } else { 841 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 842 } 843 } 844 845 static void 846 header_crc32_accel_done(void *cb_arg, int status) 847 { 848 struct nvme_tcp_pdu *pdu = cb_arg; 849 850 pdu->header_digest_crc32 = pdu->header_digest_crc32 ^ SPDK_CRC32C_XOR; 851 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, pdu->header_digest_crc32); 852 if (spdk_unlikely(status)) { 853 SPDK_ERRLOG("Failed to finish the crc32 work\n"); 854 _pdu_write_done(pdu, status); 855 return; 856 } 857 858 _tcp_write_pdu(pdu); 859 } 860 861 static void 862 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 863 struct nvme_tcp_pdu *pdu, 864 nvme_tcp_qpair_xfer_complete_cb cb_fn, 865 void *cb_arg) 866 { 867 int hlen; 868 869 assert(&tqpair->pdu_in_progress != pdu); 870 871 hlen = pdu->hdr.common.hlen; 872 pdu->cb_fn = cb_fn; 873 pdu->cb_arg = cb_arg; 874 pdu->qpair = tqpair; 875 876 /* Header Digest */ 877 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 878 spdk_accel_submit_crc32c(tqpair->accel_channel, &pdu->header_digest_crc32, 879 &pdu->hdr.raw, 0, 880 hlen, header_crc32_accel_done, pdu); 881 return; 882 } 883 884 _tcp_write_pdu(pdu); 885 } 886 887 static int 888 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 889 { 890 uint32_t i; 891 struct spdk_nvmf_transport_opts *opts; 892 uint32_t in_capsule_data_size; 893 894 opts = &tqpair->qpair.transport->opts; 895 896 in_capsule_data_size = opts->in_capsule_data_size; 897 if (opts->dif_insert_or_strip) { 898 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 899 } 900 901 tqpair->resource_count = opts->max_queue_depth; 902 903 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 904 if (!tqpair->reqs) { 905 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 906 return -1; 907 } 908 909 if (in_capsule_data_size) { 910 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 911 NULL, SPDK_ENV_LCORE_ID_ANY, 912 SPDK_MALLOC_DMA); 913 if (!tqpair->bufs) { 914 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 915 return -1; 916 } 917 } 918 919 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 920 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 921 if (!tqpair->pdus) { 922 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 923 return -1; 924 } 925 926 for (i = 0; i < tqpair->resource_count; i++) { 927 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 928 929 tcp_req->ttag = i + 1; 930 tcp_req->req.qpair = &tqpair->qpair; 931 932 tcp_req->pdu = &tqpair->pdus[i]; 933 tcp_req->pdu->qpair = tqpair; 934 935 /* Set up memory to receive commands */ 936 if (tqpair->bufs) { 937 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 938 } 939 940 /* Set the cmdn and rsp */ 941 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 942 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 943 944 /* Initialize request state to FREE */ 945 tcp_req->state = TCP_REQUEST_STATE_FREE; 946 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 947 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 948 } 949 950 tqpair->mgmt_pdu = &tqpair->pdus[i]; 951 tqpair->mgmt_pdu->qpair = tqpair; 952 953 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 954 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 955 956 return 0; 957 } 958 959 static int 960 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 961 { 962 struct spdk_nvmf_tcp_qpair *tqpair; 963 964 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 965 966 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 967 968 /* Initialise request state queues of the qpair */ 969 TAILQ_INIT(&tqpair->tcp_req_free_queue); 970 TAILQ_INIT(&tqpair->tcp_req_working_queue); 971 972 tqpair->host_hdgst_enable = true; 973 tqpair->host_ddgst_enable = true; 974 975 return 0; 976 } 977 978 static int 979 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 980 { 981 int rc; 982 983 /* set low water mark */ 984 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 985 if (rc != 0) { 986 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 987 return rc; 988 } 989 990 return 0; 991 } 992 993 static void 994 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 995 struct spdk_nvmf_tcp_port *port, 996 struct spdk_sock *sock) 997 { 998 struct spdk_nvmf_tcp_qpair *tqpair; 999 int rc; 1000 1001 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1002 port->trid->traddr, port->trid->trsvcid); 1003 1004 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1005 if (tqpair == NULL) { 1006 SPDK_ERRLOG("Could not allocate new connection.\n"); 1007 spdk_sock_close(&sock); 1008 return; 1009 } 1010 1011 tqpair->sock = sock; 1012 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1013 tqpair->port = port; 1014 tqpair->qpair.transport = transport; 1015 1016 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1017 sizeof(tqpair->target_addr), &tqpair->target_port, 1018 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1019 &tqpair->initiator_port); 1020 if (rc < 0) { 1021 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1022 nvmf_tcp_qpair_destroy(tqpair); 1023 return; 1024 } 1025 1026 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1027 } 1028 1029 static uint32_t 1030 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1031 { 1032 struct spdk_sock *sock; 1033 uint32_t count = 0; 1034 int i; 1035 1036 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1037 sock = spdk_sock_accept(port->listen_sock); 1038 if (sock == NULL) { 1039 break; 1040 } 1041 count++; 1042 nvmf_tcp_handle_connect(transport, port, sock); 1043 } 1044 1045 return count; 1046 } 1047 1048 static uint32_t 1049 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1050 { 1051 struct spdk_nvmf_tcp_transport *ttransport; 1052 struct spdk_nvmf_tcp_port *port; 1053 uint32_t count = 0; 1054 1055 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1056 1057 TAILQ_FOREACH(port, &ttransport->ports, link) { 1058 count += nvmf_tcp_port_accept(transport, port); 1059 } 1060 1061 return count; 1062 } 1063 1064 static void 1065 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1066 struct spdk_nvme_transport_id *trid, 1067 struct spdk_nvmf_discovery_log_page_entry *entry) 1068 { 1069 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1070 entry->adrfam = trid->adrfam; 1071 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1072 1073 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1074 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1075 1076 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1077 } 1078 1079 static struct spdk_nvmf_tcp_control_msg_list * 1080 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1081 { 1082 struct spdk_nvmf_tcp_control_msg_list *list; 1083 struct spdk_nvmf_tcp_control_msg *msg; 1084 uint16_t i; 1085 1086 list = calloc(1, sizeof(*list)); 1087 if (!list) { 1088 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1089 return NULL; 1090 } 1091 1092 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1093 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1094 if (!list->msg_buf) { 1095 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1096 free(list); 1097 return NULL; 1098 } 1099 1100 STAILQ_INIT(&list->free_msgs); 1101 1102 for (i = 0; i < num_messages; i++) { 1103 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1104 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1105 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1106 } 1107 1108 return list; 1109 } 1110 1111 static void 1112 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1113 { 1114 if (!list) { 1115 return; 1116 } 1117 1118 spdk_free(list->msg_buf); 1119 free(list); 1120 } 1121 1122 static struct spdk_nvmf_transport_poll_group * 1123 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1124 { 1125 struct spdk_nvmf_tcp_transport *ttransport; 1126 struct spdk_nvmf_tcp_poll_group *tgroup; 1127 1128 tgroup = calloc(1, sizeof(*tgroup)); 1129 if (!tgroup) { 1130 return NULL; 1131 } 1132 1133 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1134 if (!tgroup->sock_group) { 1135 goto cleanup; 1136 } 1137 1138 TAILQ_INIT(&tgroup->qpairs); 1139 TAILQ_INIT(&tgroup->await_req); 1140 1141 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1142 1143 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1144 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1145 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1146 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1147 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1148 if (!tgroup->control_msg_list) { 1149 goto cleanup; 1150 } 1151 } 1152 1153 return &tgroup->group; 1154 1155 cleanup: 1156 nvmf_tcp_poll_group_destroy(&tgroup->group); 1157 return NULL; 1158 } 1159 1160 static struct spdk_nvmf_transport_poll_group * 1161 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1162 { 1163 struct spdk_nvmf_tcp_qpair *tqpair; 1164 struct spdk_sock_group *group = NULL; 1165 int rc; 1166 1167 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1168 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1169 if (!rc && group != NULL) { 1170 return spdk_sock_group_get_ctx(group); 1171 } 1172 1173 return NULL; 1174 } 1175 1176 static void 1177 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1178 { 1179 struct spdk_nvmf_tcp_poll_group *tgroup; 1180 1181 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1182 spdk_sock_group_close(&tgroup->sock_group); 1183 if (tgroup->control_msg_list) { 1184 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1185 } 1186 1187 free(tgroup); 1188 } 1189 1190 static void 1191 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1192 enum nvme_tcp_pdu_recv_state state) 1193 { 1194 if (tqpair->recv_state == state) { 1195 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1196 tqpair, state); 1197 return; 1198 } 1199 1200 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1201 /* When leaving the await req state, move the qpair to the main list */ 1202 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1203 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1204 } 1205 1206 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1207 tqpair->recv_state = state; 1208 1209 switch (state) { 1210 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1211 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1212 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1213 break; 1214 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1215 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1216 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1217 break; 1218 case NVME_TCP_PDU_RECV_STATE_ERROR: 1219 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1220 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1221 break; 1222 default: 1223 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1224 abort(); 1225 break; 1226 } 1227 } 1228 1229 static int 1230 nvmf_tcp_qpair_handle_timeout(void *ctx) 1231 { 1232 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1233 1234 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1235 1236 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1237 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1238 1239 nvmf_tcp_qpair_disconnect(tqpair); 1240 return SPDK_POLLER_BUSY; 1241 } 1242 1243 static void 1244 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1245 { 1246 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1247 1248 if (!tqpair->timeout_poller) { 1249 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1250 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1251 } 1252 } 1253 1254 static void 1255 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1256 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1257 { 1258 struct nvme_tcp_pdu *rsp_pdu; 1259 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1260 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1261 uint32_t copy_len; 1262 1263 rsp_pdu = tqpair->mgmt_pdu; 1264 1265 c2h_term_req = &rsp_pdu->hdr.term_req; 1266 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1267 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1268 1269 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1270 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1271 DSET32(&c2h_term_req->fei, error_offset); 1272 } 1273 1274 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1275 1276 /* Copy the error info into the buffer */ 1277 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1278 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1279 1280 /* Contain the header of the wrong received pdu */ 1281 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1282 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1283 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1284 } 1285 1286 static void 1287 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1288 struct spdk_nvmf_tcp_qpair *tqpair, 1289 struct nvme_tcp_pdu *pdu) 1290 { 1291 struct spdk_nvmf_tcp_req *tcp_req; 1292 1293 assert(pdu->psh_valid_bytes == pdu->psh_len); 1294 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1295 1296 tcp_req = nvmf_tcp_req_get(tqpair); 1297 if (!tcp_req) { 1298 /* Directly return and make the allocation retry again */ 1299 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1300 return; 1301 } 1302 1303 /* The host sent more commands than the maximum queue depth. */ 1304 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1305 nvmf_tcp_qpair_disconnect(tqpair); 1306 return; 1307 } 1308 1309 pdu->req = tcp_req; 1310 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1311 nvmf_tcp_req_process(ttransport, tcp_req); 1312 } 1313 1314 static void 1315 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1316 struct spdk_nvmf_tcp_qpair *tqpair, 1317 struct nvme_tcp_pdu *pdu) 1318 { 1319 struct spdk_nvmf_tcp_req *tcp_req; 1320 struct spdk_nvme_tcp_cmd *capsule_cmd; 1321 uint32_t error_offset = 0; 1322 enum spdk_nvme_tcp_term_req_fes fes; 1323 1324 capsule_cmd = &pdu->hdr.capsule_cmd; 1325 tcp_req = pdu->req; 1326 assert(tcp_req != NULL); 1327 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1328 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1329 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1330 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1331 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1332 goto err; 1333 } 1334 1335 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1336 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1337 nvmf_tcp_req_process(ttransport, tcp_req); 1338 1339 return; 1340 err: 1341 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1342 } 1343 1344 static int 1345 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1346 enum spdk_nvmf_tcp_req_state state, 1347 uint16_t cid, uint16_t tag, 1348 struct spdk_nvmf_tcp_req **req) 1349 { 1350 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1351 1352 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1353 if (tcp_req->state != state) { 1354 continue; 1355 } 1356 1357 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1358 continue; 1359 } 1360 1361 if (tcp_req->ttag == tag) { 1362 *req = tcp_req; 1363 return 0; 1364 } 1365 1366 *req = NULL; 1367 return -1; 1368 } 1369 1370 /* Didn't find it, but not an error */ 1371 *req = NULL; 1372 return 0; 1373 } 1374 1375 static void 1376 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1377 struct spdk_nvmf_tcp_qpair *tqpair, 1378 struct nvme_tcp_pdu *pdu) 1379 { 1380 struct spdk_nvmf_tcp_req *tcp_req; 1381 uint32_t error_offset = 0; 1382 enum spdk_nvme_tcp_term_req_fes fes = 0; 1383 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1384 int rc; 1385 1386 h2c_data = &pdu->hdr.h2c_data; 1387 1388 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1389 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1390 1391 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1392 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1393 if (rc == 0 && tcp_req == NULL) { 1394 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1395 h2c_data->ttag, &tcp_req); 1396 } 1397 1398 if (!tcp_req) { 1399 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1400 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1401 if (rc == 0) { 1402 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1403 } else { 1404 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1405 } 1406 goto err; 1407 } 1408 1409 if (tcp_req->h2c_offset != h2c_data->datao) { 1410 SPDK_DEBUGLOG(nvmf_tcp, 1411 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1412 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1413 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1414 goto err; 1415 } 1416 1417 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1418 SPDK_DEBUGLOG(nvmf_tcp, 1419 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1420 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1421 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1422 goto err; 1423 } 1424 1425 pdu->req = tcp_req; 1426 1427 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1428 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1429 } 1430 1431 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1432 h2c_data->datao, h2c_data->datal); 1433 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1434 return; 1435 1436 err: 1437 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1438 } 1439 1440 static void 1441 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1442 struct spdk_nvmf_tcp_qpair *tqpair) 1443 { 1444 struct nvme_tcp_pdu *rsp_pdu; 1445 struct spdk_nvme_tcp_rsp *capsule_resp; 1446 1447 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1448 1449 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1450 assert(rsp_pdu != NULL); 1451 1452 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1453 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1454 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1455 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1456 if (tqpair->host_hdgst_enable) { 1457 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1458 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1459 } 1460 1461 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1462 } 1463 1464 static void 1465 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1466 { 1467 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1468 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1469 struct spdk_nvmf_tcp_qpair, qpair); 1470 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 1471 tcp_req->req.qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1472 1473 assert(tqpair != NULL); 1474 1475 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1476 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1477 tcp_req->req.length); 1478 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1479 return; 1480 } 1481 1482 if (ttransport->tcp_opts.c2h_success) { 1483 nvmf_tcp_request_free(tcp_req); 1484 } else { 1485 nvmf_tcp_req_pdu_fini(tcp_req); 1486 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1487 } 1488 } 1489 1490 static void 1491 nvmf_tcp_r2t_complete(void *cb_arg) 1492 { 1493 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1494 struct spdk_nvmf_tcp_transport *ttransport; 1495 1496 nvmf_tcp_req_pdu_fini(tcp_req); 1497 1498 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1499 struct spdk_nvmf_tcp_transport, transport); 1500 1501 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1502 1503 if (tcp_req->h2c_offset == tcp_req->req.length) { 1504 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1505 nvmf_tcp_req_process(ttransport, tcp_req); 1506 } 1507 } 1508 1509 static void 1510 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1511 struct spdk_nvmf_tcp_req *tcp_req) 1512 { 1513 struct nvme_tcp_pdu *rsp_pdu; 1514 struct spdk_nvme_tcp_r2t_hdr *r2t; 1515 1516 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1517 assert(rsp_pdu != NULL); 1518 1519 r2t = &rsp_pdu->hdr.r2t; 1520 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1521 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1522 1523 if (tqpair->host_hdgst_enable) { 1524 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1525 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1526 } 1527 1528 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1529 r2t->ttag = tcp_req->ttag; 1530 r2t->r2to = tcp_req->h2c_offset; 1531 r2t->r2tl = tcp_req->req.length; 1532 1533 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1534 1535 SPDK_DEBUGLOG(nvmf_tcp, 1536 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1537 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1538 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1539 } 1540 1541 static void 1542 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1543 struct spdk_nvmf_tcp_qpair *tqpair, 1544 struct nvme_tcp_pdu *pdu) 1545 { 1546 struct spdk_nvmf_tcp_req *tcp_req; 1547 1548 tcp_req = pdu->req; 1549 assert(tcp_req != NULL); 1550 1551 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1552 1553 tcp_req->h2c_offset += pdu->data_len; 1554 1555 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1556 1557 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1558 * acknowledged before moving on. */ 1559 if (tcp_req->h2c_offset == tcp_req->req.length && 1560 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1561 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1562 nvmf_tcp_req_process(ttransport, tcp_req); 1563 } 1564 } 1565 1566 static void 1567 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1568 { 1569 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1570 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1571 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1572 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1573 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1574 DGET32(h2c_term_req->fei)); 1575 } 1576 } 1577 1578 static void 1579 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1580 struct nvme_tcp_pdu *pdu) 1581 { 1582 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1583 uint32_t error_offset = 0; 1584 enum spdk_nvme_tcp_term_req_fes fes; 1585 1586 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1587 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1588 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1589 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1590 goto end; 1591 } 1592 1593 /* set the data buffer */ 1594 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1595 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1596 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1597 return; 1598 end: 1599 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1600 } 1601 1602 static void 1603 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1604 struct nvme_tcp_pdu *pdu) 1605 { 1606 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1607 1608 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1609 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1610 } 1611 1612 static void 1613 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1614 struct spdk_nvmf_tcp_transport *ttransport) 1615 { 1616 int rc = 0; 1617 struct nvme_tcp_pdu *pdu; 1618 uint32_t crc32c, error_offset = 0; 1619 enum spdk_nvme_tcp_term_req_fes fes; 1620 1621 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1622 pdu = &tqpair->pdu_in_progress; 1623 1624 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1625 /* check data digest if need */ 1626 if (pdu->ddgst_enable) { 1627 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1628 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1629 if (rc == 0) { 1630 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1631 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1632 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1633 return; 1634 1635 } 1636 } 1637 1638 switch (pdu->hdr.common.pdu_type) { 1639 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1640 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1641 break; 1642 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1643 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1644 break; 1645 1646 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1647 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1648 break; 1649 1650 default: 1651 /* The code should not go to here */ 1652 SPDK_ERRLOG("The code should not go to here\n"); 1653 break; 1654 } 1655 } 1656 1657 static void 1658 nvmf_tcp_send_icresp_complete(void *cb_arg) 1659 { 1660 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1661 1662 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1663 } 1664 1665 static void 1666 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1667 struct spdk_nvmf_tcp_qpair *tqpair, 1668 struct nvme_tcp_pdu *pdu) 1669 { 1670 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1671 struct nvme_tcp_pdu *rsp_pdu; 1672 struct spdk_nvme_tcp_ic_resp *ic_resp; 1673 uint32_t error_offset = 0; 1674 enum spdk_nvme_tcp_term_req_fes fes; 1675 1676 /* Only PFV 0 is defined currently */ 1677 if (ic_req->pfv != 0) { 1678 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1679 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1680 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1681 goto end; 1682 } 1683 1684 /* MAXR2T is 0's based */ 1685 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1686 1687 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1688 if (!tqpair->host_hdgst_enable) { 1689 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1690 } 1691 1692 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1693 if (!tqpair->host_ddgst_enable) { 1694 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1695 } 1696 1697 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1698 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1699 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1700 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1701 tqpair, 1702 tqpair->recv_buf_size); 1703 /* Not fatal. */ 1704 } 1705 1706 if (tqpair->host_hdgst_enable) { 1707 tqpair->accel_channel = spdk_accel_engine_get_io_channel(); 1708 if (spdk_unlikely(!tqpair->accel_channel)) { 1709 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1710 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, dgst); 1711 SPDK_ERRLOG("Unabled to get accel_channel for tqpair=%p, failed to enable header digest\n", 1712 tqpair); 1713 goto end; 1714 } 1715 } 1716 1717 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1718 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1719 1720 rsp_pdu = tqpair->mgmt_pdu; 1721 1722 ic_resp = &rsp_pdu->hdr.ic_resp; 1723 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1724 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1725 ic_resp->pfv = 0; 1726 ic_resp->cpda = tqpair->cpda; 1727 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1728 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1729 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1730 1731 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1732 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1733 1734 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1735 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1736 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1737 return; 1738 end: 1739 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1740 } 1741 1742 static void 1743 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1744 struct spdk_nvmf_tcp_transport *ttransport) 1745 { 1746 struct nvme_tcp_pdu *pdu; 1747 int rc; 1748 uint32_t crc32c, error_offset = 0; 1749 enum spdk_nvme_tcp_term_req_fes fes; 1750 1751 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1752 pdu = &tqpair->pdu_in_progress; 1753 1754 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1755 pdu->hdr.common.pdu_type); 1756 /* check header digest if needed */ 1757 if (pdu->has_hdgst) { 1758 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1759 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1760 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1761 if (rc == 0) { 1762 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1763 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1764 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1765 return; 1766 1767 } 1768 } 1769 1770 switch (pdu->hdr.common.pdu_type) { 1771 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1772 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1773 break; 1774 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1775 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1776 break; 1777 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1778 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1779 break; 1780 1781 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1782 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1783 break; 1784 1785 default: 1786 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1787 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1788 error_offset = 1; 1789 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1790 break; 1791 } 1792 } 1793 1794 static void 1795 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1796 { 1797 struct nvme_tcp_pdu *pdu; 1798 uint32_t error_offset = 0; 1799 enum spdk_nvme_tcp_term_req_fes fes; 1800 uint8_t expected_hlen, pdo; 1801 bool plen_error = false, pdo_error = false; 1802 1803 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1804 pdu = &tqpair->pdu_in_progress; 1805 1806 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1807 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1808 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1809 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1810 goto err; 1811 } 1812 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1813 if (pdu->hdr.common.plen != expected_hlen) { 1814 plen_error = true; 1815 } 1816 } else { 1817 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1818 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1819 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1820 goto err; 1821 } 1822 1823 switch (pdu->hdr.common.pdu_type) { 1824 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1825 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1826 pdo = pdu->hdr.common.pdo; 1827 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1828 pdo_error = true; 1829 break; 1830 } 1831 1832 if (pdu->hdr.common.plen < expected_hlen) { 1833 plen_error = true; 1834 } 1835 break; 1836 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1837 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1838 pdo = pdu->hdr.common.pdo; 1839 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1840 pdo_error = true; 1841 break; 1842 } 1843 if (pdu->hdr.common.plen < expected_hlen) { 1844 plen_error = true; 1845 } 1846 break; 1847 1848 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1849 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1850 if ((pdu->hdr.common.plen <= expected_hlen) || 1851 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1852 plen_error = true; 1853 } 1854 break; 1855 1856 default: 1857 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1858 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1859 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1860 goto err; 1861 } 1862 } 1863 1864 if (pdu->hdr.common.hlen != expected_hlen) { 1865 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1866 pdu->hdr.common.pdu_type, 1867 expected_hlen, pdu->hdr.common.hlen, tqpair); 1868 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1869 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1870 goto err; 1871 } else if (pdo_error) { 1872 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1873 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1874 } else if (plen_error) { 1875 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1876 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1877 goto err; 1878 } else { 1879 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1880 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1881 return; 1882 } 1883 err: 1884 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1885 } 1886 1887 static int 1888 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1889 int read_len) 1890 { 1891 int rc; 1892 1893 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1894 read_offset, read_len, pdu->dif_ctx); 1895 if (rc != 0) { 1896 SPDK_ERRLOG("DIF generate failed\n"); 1897 } 1898 1899 return rc; 1900 } 1901 1902 static int 1903 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1904 { 1905 int rc = 0; 1906 struct nvme_tcp_pdu *pdu; 1907 enum nvme_tcp_pdu_recv_state prev_state; 1908 uint32_t data_len; 1909 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1910 struct spdk_nvmf_tcp_transport, transport); 1911 1912 /* The loop here is to allow for several back-to-back state changes. */ 1913 do { 1914 prev_state = tqpair->recv_state; 1915 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1916 1917 pdu = &tqpair->pdu_in_progress; 1918 switch (tqpair->recv_state) { 1919 /* Wait for the common header */ 1920 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1921 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1922 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1923 return rc; 1924 } 1925 1926 rc = nvme_tcp_read_data(tqpair->sock, 1927 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1928 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1929 if (rc < 0) { 1930 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1931 return NVME_TCP_PDU_FATAL; 1932 } else if (rc > 0) { 1933 pdu->ch_valid_bytes += rc; 1934 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1935 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1936 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1937 } 1938 } 1939 1940 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1941 return NVME_TCP_PDU_IN_PROGRESS; 1942 } 1943 1944 /* The command header of this PDU has now been read from the socket. */ 1945 nvmf_tcp_pdu_ch_handle(tqpair); 1946 break; 1947 /* Wait for the pdu specific header */ 1948 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1949 rc = nvme_tcp_read_data(tqpair->sock, 1950 pdu->psh_len - pdu->psh_valid_bytes, 1951 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1952 if (rc < 0) { 1953 return NVME_TCP_PDU_FATAL; 1954 } else if (rc > 0) { 1955 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1956 0, rc, 0, 0); 1957 pdu->psh_valid_bytes += rc; 1958 } 1959 1960 if (pdu->psh_valid_bytes < pdu->psh_len) { 1961 return NVME_TCP_PDU_IN_PROGRESS; 1962 } 1963 1964 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1965 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1966 break; 1967 /* Wait for the req slot */ 1968 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1969 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1970 break; 1971 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1972 /* check whether the data is valid, if not we just return */ 1973 if (!pdu->data_len) { 1974 return NVME_TCP_PDU_IN_PROGRESS; 1975 } 1976 1977 data_len = pdu->data_len; 1978 /* data digest */ 1979 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1980 tqpair->host_ddgst_enable)) { 1981 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1982 pdu->ddgst_enable = true; 1983 } 1984 1985 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1986 if (rc < 0) { 1987 return NVME_TCP_PDU_FATAL; 1988 } 1989 pdu->rw_offset += rc; 1990 1991 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1992 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->rw_offset - rc, rc); 1993 if (rc != 0) { 1994 return NVME_TCP_PDU_FATAL; 1995 } 1996 } 1997 1998 if (pdu->rw_offset < data_len) { 1999 return NVME_TCP_PDU_IN_PROGRESS; 2000 } 2001 2002 /* All of this PDU has now been read from the socket. */ 2003 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 2004 break; 2005 case NVME_TCP_PDU_RECV_STATE_ERROR: 2006 if (!spdk_sock_is_connected(tqpair->sock)) { 2007 return NVME_TCP_PDU_FATAL; 2008 } 2009 break; 2010 default: 2011 assert(0); 2012 SPDK_ERRLOG("code should not come to here"); 2013 break; 2014 } 2015 } while (tqpair->recv_state != prev_state); 2016 2017 return rc; 2018 } 2019 2020 static inline void * 2021 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2022 { 2023 struct spdk_nvmf_tcp_control_msg *msg; 2024 2025 assert(list); 2026 2027 msg = STAILQ_FIRST(&list->free_msgs); 2028 if (!msg) { 2029 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2030 return NULL; 2031 } 2032 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2033 return msg; 2034 } 2035 2036 static inline void 2037 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2038 { 2039 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2040 2041 assert(list); 2042 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2043 } 2044 2045 static int 2046 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2047 struct spdk_nvmf_transport *transport, 2048 struct spdk_nvmf_transport_poll_group *group) 2049 { 2050 struct spdk_nvmf_request *req = &tcp_req->req; 2051 struct spdk_nvme_cmd *cmd; 2052 struct spdk_nvme_cpl *rsp; 2053 struct spdk_nvme_sgl_descriptor *sgl; 2054 struct spdk_nvmf_tcp_poll_group *tgroup; 2055 uint32_t length; 2056 2057 cmd = &req->cmd->nvme_cmd; 2058 rsp = &req->rsp->nvme_cpl; 2059 sgl = &cmd->dptr.sgl1; 2060 2061 length = sgl->unkeyed.length; 2062 2063 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2064 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2065 if (length > transport->opts.max_io_size) { 2066 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2067 length, transport->opts.max_io_size); 2068 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2069 return -1; 2070 } 2071 2072 /* fill request length and populate iovs */ 2073 req->length = length; 2074 2075 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2076 2077 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2078 req->dif.orig_length = length; 2079 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2080 req->dif.elba_length = length; 2081 } 2082 2083 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2084 /* No available buffers. Queue this request up. */ 2085 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2086 tcp_req); 2087 return 0; 2088 } 2089 2090 /* backward compatible */ 2091 req->data = req->iov[0].iov_base; 2092 2093 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2094 tcp_req, req->iovcnt, req->data); 2095 2096 return 0; 2097 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2098 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2099 uint64_t offset = sgl->address; 2100 uint32_t max_len = transport->opts.in_capsule_data_size; 2101 assert(tcp_req->has_incapsule_data); 2102 2103 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2104 offset, length); 2105 2106 if (offset > max_len) { 2107 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2108 offset, max_len); 2109 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2110 return -1; 2111 } 2112 max_len -= (uint32_t)offset; 2113 2114 if (spdk_unlikely(length > max_len)) { 2115 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2116 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2117 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2118 2119 /* Get a buffer from dedicated list */ 2120 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2121 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2122 assert(tgroup->control_msg_list); 2123 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2124 if (!req->data) { 2125 /* No available buffers. Queue this request up. */ 2126 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2127 return 0; 2128 } 2129 } else { 2130 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2131 length, max_len); 2132 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2133 return -1; 2134 } 2135 } else { 2136 req->data = tcp_req->buf; 2137 } 2138 2139 req->length = length; 2140 req->data_from_pool = false; 2141 2142 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2143 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2144 req->dif.elba_length = length; 2145 } 2146 2147 req->iov[0].iov_base = req->data; 2148 req->iov[0].iov_len = length; 2149 req->iovcnt = 1; 2150 2151 return 0; 2152 } 2153 2154 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2155 sgl->generic.type, sgl->generic.subtype); 2156 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2157 return -1; 2158 } 2159 2160 static inline enum spdk_nvme_media_error_status_code 2161 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2162 enum spdk_nvme_media_error_status_code result; 2163 2164 switch (err_type) 2165 { 2166 case SPDK_DIF_REFTAG_ERROR: 2167 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2168 break; 2169 case SPDK_DIF_APPTAG_ERROR: 2170 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2171 break; 2172 case SPDK_DIF_GUARD_ERROR: 2173 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2174 break; 2175 default: 2176 SPDK_UNREACHABLE(); 2177 break; 2178 } 2179 2180 return result; 2181 } 2182 2183 static void 2184 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2185 struct spdk_nvmf_tcp_req *tcp_req) 2186 { 2187 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2188 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2189 struct nvme_tcp_pdu *rsp_pdu; 2190 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2191 uint32_t plen, pdo, alignment; 2192 int rc; 2193 2194 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2195 2196 rsp_pdu = tcp_req->pdu; 2197 assert(rsp_pdu != NULL); 2198 assert(tcp_req->pdu_in_use); 2199 2200 c2h_data = &rsp_pdu->hdr.c2h_data; 2201 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2202 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2203 2204 if (tqpair->host_hdgst_enable) { 2205 plen += SPDK_NVME_TCP_DIGEST_LEN; 2206 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2207 } 2208 2209 /* set the psh */ 2210 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2211 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2212 c2h_data->datao = tcp_req->pdu->rw_offset; 2213 2214 /* set the padding */ 2215 rsp_pdu->padding_len = 0; 2216 pdo = plen; 2217 if (tqpair->cpda) { 2218 alignment = (tqpair->cpda + 1) << 2; 2219 if (alignment > plen) { 2220 rsp_pdu->padding_len = alignment - plen; 2221 pdo = plen = alignment; 2222 } 2223 } 2224 2225 c2h_data->common.pdo = pdo; 2226 plen += c2h_data->datal; 2227 if (tqpair->host_ddgst_enable) { 2228 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2229 plen += SPDK_NVME_TCP_DIGEST_LEN; 2230 } 2231 2232 c2h_data->common.plen = plen; 2233 2234 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2235 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2236 } 2237 2238 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2239 c2h_data->datao, c2h_data->datal); 2240 2241 2242 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2243 if (ttransport->tcp_opts.c2h_success) { 2244 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2245 } 2246 2247 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2248 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2249 struct spdk_dif_error err_blk = {}; 2250 uint32_t mapped_length = 0; 2251 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2252 uint32_t ddgst_len = 0; 2253 2254 if (tqpair->host_ddgst_enable) { 2255 /* Data digest consumes additional iov entry */ 2256 available_iovs--; 2257 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2258 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2259 c2h_data->common.plen -= ddgst_len; 2260 } 2261 /* Temp call to estimate if data can be described by limited number of iovs. 2262 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2263 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2264 false, &mapped_length); 2265 2266 if (mapped_length != c2h_data->common.plen) { 2267 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2268 SPDK_DEBUGLOG(nvmf_tcp, 2269 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2270 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2271 c2h_data->common.plen = mapped_length; 2272 2273 /* Rebuild pdu->data_iov since data length is changed */ 2274 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2275 c2h_data->datal); 2276 2277 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2278 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2279 } 2280 2281 c2h_data->common.plen += ddgst_len; 2282 2283 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2284 2285 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2286 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2287 if (rc != 0) { 2288 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2289 err_blk.err_type, err_blk.err_offset); 2290 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2291 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2292 nvmf_tcp_req_pdu_fini(tcp_req); 2293 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2294 return; 2295 } 2296 } 2297 2298 rsp_pdu->rw_offset += c2h_data->datal; 2299 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2300 } 2301 2302 static void 2303 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2304 struct spdk_nvmf_tcp_req *tcp_req) 2305 { 2306 nvmf_tcp_req_pdu_init(tcp_req); 2307 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2308 } 2309 2310 static int 2311 request_transfer_out(struct spdk_nvmf_request *req) 2312 { 2313 struct spdk_nvmf_tcp_req *tcp_req; 2314 struct spdk_nvmf_qpair *qpair; 2315 struct spdk_nvmf_tcp_qpair *tqpair; 2316 struct spdk_nvme_cpl *rsp; 2317 2318 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2319 2320 qpair = req->qpair; 2321 rsp = &req->rsp->nvme_cpl; 2322 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2323 2324 /* Advance our sq_head pointer */ 2325 if (qpair->sq_head == qpair->sq_head_max) { 2326 qpair->sq_head = 0; 2327 } else { 2328 qpair->sq_head++; 2329 } 2330 rsp->sqhd = qpair->sq_head; 2331 2332 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2333 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2334 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2335 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2336 } else { 2337 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2338 } 2339 2340 return 0; 2341 } 2342 2343 static void 2344 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2345 struct spdk_nvmf_tcp_req *tcp_req) 2346 { 2347 struct nvme_tcp_pdu *pdu; 2348 uint32_t plen = 0; 2349 2350 pdu = &tqpair->pdu_in_progress; 2351 plen = pdu->hdr.common.hlen; 2352 2353 if (tqpair->host_hdgst_enable) { 2354 plen += SPDK_NVME_TCP_DIGEST_LEN; 2355 } 2356 2357 if (pdu->hdr.common.plen != plen) { 2358 tcp_req->has_incapsule_data = true; 2359 } 2360 } 2361 2362 static bool 2363 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2364 struct spdk_nvmf_tcp_req *tcp_req) 2365 { 2366 struct spdk_nvmf_tcp_qpair *tqpair; 2367 int rc; 2368 enum spdk_nvmf_tcp_req_state prev_state; 2369 bool progress = false; 2370 struct spdk_nvmf_transport *transport = &ttransport->transport; 2371 struct spdk_nvmf_transport_poll_group *group; 2372 struct spdk_nvmf_tcp_poll_group *tgroup; 2373 2374 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2375 group = &tqpair->group->group; 2376 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2377 2378 /* If the qpair is not active, we need to abort the outstanding requests. */ 2379 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2380 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2381 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2382 } 2383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2384 } 2385 2386 /* The loop here is to allow for several back-to-back state changes. */ 2387 do { 2388 prev_state = tcp_req->state; 2389 2390 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2391 tqpair); 2392 2393 switch (tcp_req->state) { 2394 case TCP_REQUEST_STATE_FREE: 2395 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2396 * to escape this state. */ 2397 break; 2398 case TCP_REQUEST_STATE_NEW: 2399 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2400 2401 /* copy the cmd from the receive pdu */ 2402 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2403 2404 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2405 tcp_req->req.dif.dif_insert_or_strip = true; 2406 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2407 } 2408 2409 /* The next state transition depends on the data transfer needs of this request. */ 2410 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2411 2412 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2413 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2414 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2415 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2416 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2417 break; 2418 } 2419 2420 /* If no data to transfer, ready to execute. */ 2421 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2422 /* Reset the tqpair receving pdu state */ 2423 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2424 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2425 break; 2426 } 2427 2428 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2429 2430 if (!tcp_req->has_incapsule_data) { 2431 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2432 } 2433 2434 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2435 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2436 break; 2437 case TCP_REQUEST_STATE_NEED_BUFFER: 2438 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2439 2440 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2441 2442 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2443 SPDK_DEBUGLOG(nvmf_tcp, 2444 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2445 tcp_req, tqpair); 2446 /* This request needs to wait in line to obtain a buffer */ 2447 break; 2448 } 2449 2450 /* Try to get a data buffer */ 2451 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2452 if (rc < 0) { 2453 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2454 /* Reset the tqpair receving pdu state */ 2455 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2456 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2457 break; 2458 } 2459 2460 if (!tcp_req->req.data) { 2461 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2462 tcp_req, tqpair); 2463 /* No buffers available. */ 2464 break; 2465 } 2466 2467 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2468 2469 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2470 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2471 if (tcp_req->req.data_from_pool) { 2472 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2473 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2474 } else { 2475 struct nvme_tcp_pdu *pdu; 2476 2477 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2478 2479 pdu = &tqpair->pdu_in_progress; 2480 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2481 tqpair); 2482 /* No need to send r2t, contained in the capsuled data */ 2483 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2484 0, tcp_req->req.length); 2485 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2486 } 2487 break; 2488 } 2489 2490 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2491 break; 2492 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2493 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2494 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2495 break; 2496 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2497 2498 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2499 (uintptr_t)tcp_req, 0); 2500 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2501 * to escape this state. */ 2502 break; 2503 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2504 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2505 2506 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2507 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2508 tcp_req->req.length = tcp_req->req.dif.elba_length; 2509 } 2510 2511 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2512 spdk_nvmf_request_exec(&tcp_req->req); 2513 break; 2514 case TCP_REQUEST_STATE_EXECUTING: 2515 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2516 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2517 * to escape this state. */ 2518 break; 2519 case TCP_REQUEST_STATE_EXECUTED: 2520 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2521 2522 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2523 tcp_req->req.length = tcp_req->req.dif.orig_length; 2524 } 2525 2526 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2527 break; 2528 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2529 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2530 rc = request_transfer_out(&tcp_req->req); 2531 assert(rc == 0); /* No good way to handle this currently */ 2532 break; 2533 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2534 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2535 (uintptr_t)tcp_req, 2536 0); 2537 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2538 * to escape this state. */ 2539 break; 2540 case TCP_REQUEST_STATE_COMPLETED: 2541 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2542 if (tcp_req->req.data_from_pool) { 2543 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2544 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2545 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2546 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2547 assert(tgroup->control_msg_list); 2548 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2549 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2550 } 2551 tcp_req->req.length = 0; 2552 tcp_req->req.iovcnt = 0; 2553 tcp_req->req.data = NULL; 2554 2555 nvmf_tcp_req_pdu_fini(tcp_req); 2556 2557 nvmf_tcp_req_put(tqpair, tcp_req); 2558 break; 2559 case TCP_REQUEST_NUM_STATES: 2560 default: 2561 assert(0); 2562 break; 2563 } 2564 2565 if (tcp_req->state != prev_state) { 2566 progress = true; 2567 } 2568 } while (tcp_req->state != prev_state); 2569 2570 return progress; 2571 } 2572 2573 static void 2574 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2575 { 2576 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2577 int rc; 2578 2579 assert(tqpair != NULL); 2580 rc = nvmf_tcp_sock_process(tqpair); 2581 2582 /* If there was a new socket error, disconnect */ 2583 if (rc < 0) { 2584 nvmf_tcp_qpair_disconnect(tqpair); 2585 } 2586 } 2587 2588 static int 2589 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2590 struct spdk_nvmf_qpair *qpair) 2591 { 2592 struct spdk_nvmf_tcp_poll_group *tgroup; 2593 struct spdk_nvmf_tcp_qpair *tqpair; 2594 int rc; 2595 2596 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2597 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2598 2599 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2600 nvmf_tcp_sock_cb, tqpair); 2601 if (rc != 0) { 2602 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2603 spdk_strerror(errno), errno); 2604 return -1; 2605 } 2606 2607 rc = nvmf_tcp_qpair_sock_init(tqpair); 2608 if (rc != 0) { 2609 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2610 return -1; 2611 } 2612 2613 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2614 if (rc < 0) { 2615 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2616 return -1; 2617 } 2618 2619 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2620 if (rc < 0) { 2621 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2622 return -1; 2623 } 2624 2625 tqpair->group = tgroup; 2626 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2627 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2628 2629 return 0; 2630 } 2631 2632 static int 2633 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2634 struct spdk_nvmf_qpair *qpair) 2635 { 2636 struct spdk_nvmf_tcp_poll_group *tgroup; 2637 struct spdk_nvmf_tcp_qpair *tqpair; 2638 int rc; 2639 2640 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2641 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2642 2643 assert(tqpair->group == tgroup); 2644 2645 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2646 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2647 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2648 } else { 2649 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2650 } 2651 2652 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2653 if (rc != 0) { 2654 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2655 spdk_strerror(errno), errno); 2656 } 2657 2658 return rc; 2659 } 2660 2661 static int 2662 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2663 { 2664 struct spdk_nvmf_tcp_transport *ttransport; 2665 struct spdk_nvmf_tcp_req *tcp_req; 2666 2667 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2668 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2669 2670 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2671 nvmf_tcp_req_process(ttransport, tcp_req); 2672 2673 return 0; 2674 } 2675 2676 static void 2677 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2678 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2679 { 2680 struct spdk_nvmf_tcp_qpair *tqpair; 2681 2682 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2683 2684 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2685 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2686 nvmf_tcp_qpair_destroy(tqpair); 2687 2688 if (cb_fn) { 2689 cb_fn(cb_arg); 2690 } 2691 } 2692 2693 static int 2694 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2695 { 2696 struct spdk_nvmf_tcp_poll_group *tgroup; 2697 int rc; 2698 struct spdk_nvmf_request *req, *req_tmp; 2699 struct spdk_nvmf_tcp_req *tcp_req; 2700 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2701 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2702 struct spdk_nvmf_tcp_transport, transport); 2703 2704 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2705 2706 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2707 return 0; 2708 } 2709 2710 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2711 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2712 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2713 break; 2714 } 2715 } 2716 2717 rc = spdk_sock_group_poll(tgroup->sock_group); 2718 if (rc < 0) { 2719 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2720 } 2721 2722 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2723 nvmf_tcp_sock_process(tqpair); 2724 } 2725 2726 return rc; 2727 } 2728 2729 static int 2730 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2731 struct spdk_nvme_transport_id *trid, bool peer) 2732 { 2733 struct spdk_nvmf_tcp_qpair *tqpair; 2734 uint16_t port; 2735 2736 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2737 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2738 2739 if (peer) { 2740 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2741 port = tqpair->initiator_port; 2742 } else { 2743 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2744 port = tqpair->target_port; 2745 } 2746 2747 if (spdk_sock_is_ipv4(tqpair->sock)) { 2748 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2749 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2750 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2751 } else { 2752 return -1; 2753 } 2754 2755 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2756 return 0; 2757 } 2758 2759 static int 2760 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2761 struct spdk_nvme_transport_id *trid) 2762 { 2763 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2764 } 2765 2766 static int 2767 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2768 struct spdk_nvme_transport_id *trid) 2769 { 2770 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2771 } 2772 2773 static int 2774 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2775 struct spdk_nvme_transport_id *trid) 2776 { 2777 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2778 } 2779 2780 static void 2781 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2782 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2783 { 2784 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2785 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2786 2787 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2788 2789 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2790 } 2791 2792 static int 2793 _nvmf_tcp_qpair_abort_request(void *ctx) 2794 { 2795 struct spdk_nvmf_request *req = ctx; 2796 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2797 struct spdk_nvmf_tcp_req, req); 2798 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2799 struct spdk_nvmf_tcp_qpair, qpair); 2800 int rc; 2801 2802 spdk_poller_unregister(&req->poller); 2803 2804 switch (tcp_req_to_abort->state) { 2805 case TCP_REQUEST_STATE_EXECUTING: 2806 rc = nvmf_ctrlr_abort_request(req); 2807 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2808 return SPDK_POLLER_BUSY; 2809 } 2810 break; 2811 2812 case TCP_REQUEST_STATE_NEED_BUFFER: 2813 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2814 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2815 2816 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2817 break; 2818 2819 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2820 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2821 break; 2822 2823 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2824 if (spdk_get_ticks() < req->timeout_tsc) { 2825 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2826 return SPDK_POLLER_BUSY; 2827 } 2828 break; 2829 2830 default: 2831 break; 2832 } 2833 2834 spdk_nvmf_request_complete(req); 2835 return SPDK_POLLER_BUSY; 2836 } 2837 2838 static void 2839 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2840 struct spdk_nvmf_request *req) 2841 { 2842 struct spdk_nvmf_tcp_qpair *tqpair; 2843 struct spdk_nvmf_tcp_transport *ttransport; 2844 struct spdk_nvmf_transport *transport; 2845 uint16_t cid; 2846 uint32_t i; 2847 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2848 2849 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2850 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2851 transport = &ttransport->transport; 2852 2853 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2854 2855 for (i = 0; i < tqpair->resource_count; i++) { 2856 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2857 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2858 tcp_req_to_abort = &tqpair->reqs[i]; 2859 break; 2860 } 2861 } 2862 2863 if (tcp_req_to_abort == NULL) { 2864 spdk_nvmf_request_complete(req); 2865 return; 2866 } 2867 2868 req->req_to_abort = &tcp_req_to_abort->req; 2869 req->timeout_tsc = spdk_get_ticks() + 2870 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2871 req->poller = NULL; 2872 2873 _nvmf_tcp_qpair_abort_request(req); 2874 } 2875 2876 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2877 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2878 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2879 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2880 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2881 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2882 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2883 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2884 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2885 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2886 2887 static void 2888 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2889 { 2890 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2891 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2892 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2893 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2894 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2895 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2896 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2897 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2898 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2899 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2900 opts->transport_specific = NULL; 2901 } 2902 2903 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2904 .name = "TCP", 2905 .type = SPDK_NVME_TRANSPORT_TCP, 2906 .opts_init = nvmf_tcp_opts_init, 2907 .create = nvmf_tcp_create, 2908 .dump_opts = nvmf_tcp_dump_opts, 2909 .destroy = nvmf_tcp_destroy, 2910 2911 .listen = nvmf_tcp_listen, 2912 .stop_listen = nvmf_tcp_stop_listen, 2913 .accept = nvmf_tcp_accept, 2914 2915 .listener_discover = nvmf_tcp_discover, 2916 2917 .poll_group_create = nvmf_tcp_poll_group_create, 2918 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2919 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2920 .poll_group_add = nvmf_tcp_poll_group_add, 2921 .poll_group_remove = nvmf_tcp_poll_group_remove, 2922 .poll_group_poll = nvmf_tcp_poll_group_poll, 2923 2924 .req_free = nvmf_tcp_req_free, 2925 .req_complete = nvmf_tcp_req_complete, 2926 2927 .qpair_fini = nvmf_tcp_close_qpair, 2928 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2929 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2930 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2931 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2932 }; 2933 2934 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2935 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2936