1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022, 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 #include "spdk/keyring.h" 19 20 #include "spdk_internal/assert.h" 21 #include "spdk_internal/nvme_tcp.h" 22 #include "spdk_internal/sock.h" 23 24 #include "nvmf_internal.h" 25 26 #include "spdk_internal/trace_defs.h" 27 28 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 29 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 30 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 31 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 32 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 33 34 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 35 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 36 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 37 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 38 39 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 42 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 43 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 44 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 46 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 47 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 48 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 49 50 #define TCP_PSK_INVALID_PERMISSIONS 0177 51 52 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 53 static bool g_tls_log = false; 54 55 /* spdk nvmf related structure */ 56 enum spdk_nvmf_tcp_req_state { 57 58 /* The request is not currently in use */ 59 TCP_REQUEST_STATE_FREE = 0, 60 61 /* Initial state when request first received */ 62 TCP_REQUEST_STATE_NEW = 1, 63 64 /* The request is queued until a data buffer is available. */ 65 TCP_REQUEST_STATE_NEED_BUFFER = 2, 66 67 /* The request is waiting for zcopy_start to finish */ 68 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 3, 69 70 /* The request has received a zero-copy buffer */ 71 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 4, 72 73 /* The request is currently transferring data from the host to the controller. */ 74 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 5, 75 76 /* The request is waiting for the R2T send acknowledgement. */ 77 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 6, 78 79 /* The request is ready to execute at the block device */ 80 TCP_REQUEST_STATE_READY_TO_EXECUTE = 7, 81 82 /* The request is currently executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTING = 8, 84 85 /* The request is waiting for zcopy buffers to be committed */ 86 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 9, 87 88 /* The request finished executing at the block device */ 89 TCP_REQUEST_STATE_EXECUTED = 10, 90 91 /* The request is ready to send a completion */ 92 TCP_REQUEST_STATE_READY_TO_COMPLETE = 11, 93 94 /* The request is currently transferring final pdus from the controller to the host. */ 95 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 12, 96 97 /* The request is waiting for zcopy buffers to be released (without committing) */ 98 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 13, 99 100 /* The request completed and can be marked free. */ 101 TCP_REQUEST_STATE_COMPLETED = 14, 102 103 /* Terminator */ 104 TCP_REQUEST_NUM_STATES, 105 }; 106 107 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 108 "Invalid PDU Header Field", 109 "PDU Sequence Error", 110 "Header Digiest Error", 111 "Data Transfer Out of Range", 112 "R2T Limit Exceeded", 113 "Unsupported parameter", 114 }; 115 116 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 117 { 118 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 119 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 120 spdk_trace_register_description("TCP_REQ_NEW", 121 TRACE_TCP_REQUEST_STATE_NEW, 122 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 123 SPDK_TRACE_ARG_TYPE_INT, ""); 124 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 125 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 126 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 127 SPDK_TRACE_ARG_TYPE_INT, ""); 128 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 129 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 130 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 131 SPDK_TRACE_ARG_TYPE_INT, ""); 132 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 133 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 134 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 135 SPDK_TRACE_ARG_TYPE_INT, ""); 136 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 137 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 138 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 139 SPDK_TRACE_ARG_TYPE_INT, ""); 140 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 141 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 142 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 143 SPDK_TRACE_ARG_TYPE_INT, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTING", 145 TRACE_TCP_REQUEST_STATE_EXECUTING, 146 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 147 SPDK_TRACE_ARG_TYPE_INT, ""); 148 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 149 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 150 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 151 SPDK_TRACE_ARG_TYPE_INT, ""); 152 spdk_trace_register_description("TCP_REQ_EXECUTED", 153 TRACE_TCP_REQUEST_STATE_EXECUTED, 154 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 155 SPDK_TRACE_ARG_TYPE_INT, ""); 156 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 157 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 158 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 159 SPDK_TRACE_ARG_TYPE_INT, ""); 160 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 161 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 162 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 163 SPDK_TRACE_ARG_TYPE_INT, ""); 164 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 165 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 166 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 167 SPDK_TRACE_ARG_TYPE_INT, ""); 168 spdk_trace_register_description("TCP_REQ_COMPLETED", 169 TRACE_TCP_REQUEST_STATE_COMPLETED, 170 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 171 SPDK_TRACE_ARG_TYPE_INT, ""); 172 spdk_trace_register_description("TCP_READ_DONE", 173 TRACE_TCP_READ_FROM_SOCKET_DONE, 174 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 175 SPDK_TRACE_ARG_TYPE_INT, ""); 176 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 177 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 178 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 179 SPDK_TRACE_ARG_TYPE_INT, ""); 180 181 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 182 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 183 SPDK_TRACE_ARG_TYPE_INT, ""); 184 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 185 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 186 SPDK_TRACE_ARG_TYPE_INT, ""); 187 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 188 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 189 SPDK_TRACE_ARG_TYPE_INT, "state"); 190 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 191 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 192 SPDK_TRACE_ARG_TYPE_INT, ""); 193 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 194 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 195 SPDK_TRACE_ARG_TYPE_INT, ""); 196 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 197 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 198 SPDK_TRACE_ARG_TYPE_INT, ""); 199 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 200 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 201 SPDK_TRACE_ARG_TYPE_INT, "state"); 202 203 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 204 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 205 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 206 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 207 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 208 } 209 210 struct spdk_nvmf_tcp_req { 211 struct spdk_nvmf_request req; 212 struct spdk_nvme_cpl rsp; 213 struct spdk_nvme_cmd cmd; 214 215 /* A PDU that can be used for sending responses. This is 216 * not the incoming PDU! */ 217 struct nvme_tcp_pdu *pdu; 218 219 /* In-capsule data buffer */ 220 uint8_t *buf; 221 222 struct spdk_nvmf_tcp_req *fused_pair; 223 224 /* 225 * The PDU for a request may be used multiple times in serial over 226 * the request's lifetime. For example, first to send an R2T, then 227 * to send a completion. To catch mistakes where the PDU is used 228 * twice at the same time, add a debug flag here for init/fini. 229 */ 230 bool pdu_in_use; 231 bool has_in_capsule_data; 232 bool fused_failed; 233 234 /* transfer_tag */ 235 uint16_t ttag; 236 237 enum spdk_nvmf_tcp_req_state state; 238 239 /* 240 * h2c_offset is used when we receive the h2c_data PDU. 241 */ 242 uint32_t h2c_offset; 243 244 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 245 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 246 }; 247 248 struct spdk_nvmf_tcp_qpair { 249 struct spdk_nvmf_qpair qpair; 250 struct spdk_nvmf_tcp_poll_group *group; 251 struct spdk_sock *sock; 252 253 enum nvme_tcp_pdu_recv_state recv_state; 254 enum nvme_tcp_qpair_state state; 255 256 /* PDU being actively received */ 257 struct nvme_tcp_pdu *pdu_in_progress; 258 259 struct spdk_nvmf_tcp_req *fused_first; 260 261 /* Queues to track the requests in all states */ 262 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 263 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 264 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 265 /* Number of working pdus */ 266 uint32_t tcp_pdu_working_count; 267 268 /* Number of requests in each state */ 269 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 270 271 uint8_t cpda; 272 273 bool host_hdgst_enable; 274 bool host_ddgst_enable; 275 276 /* This is a spare PDU used for sending special management 277 * operations. Primarily, this is used for the initial 278 * connection response and c2h termination request. */ 279 struct nvme_tcp_pdu *mgmt_pdu; 280 281 /* Arrays of in-capsule buffers, requests, and pdus. 282 * Each array is 'resource_count' number of elements */ 283 void *bufs; 284 struct spdk_nvmf_tcp_req *reqs; 285 struct nvme_tcp_pdu *pdus; 286 uint32_t resource_count; 287 uint32_t recv_buf_size; 288 289 struct spdk_nvmf_tcp_port *port; 290 291 /* IP address */ 292 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 293 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 294 295 /* IP port */ 296 uint16_t initiator_port; 297 uint16_t target_port; 298 299 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 300 bool wait_terminate; 301 302 /* Timer used to destroy qpair after detecting transport error issue if initiator does 303 * not close the connection. 304 */ 305 struct spdk_poller *timeout_poller; 306 307 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 308 void *fini_cb_arg; 309 310 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 311 }; 312 313 struct spdk_nvmf_tcp_control_msg { 314 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 315 }; 316 317 struct spdk_nvmf_tcp_control_msg_list { 318 void *msg_buf; 319 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 320 }; 321 322 struct spdk_nvmf_tcp_poll_group { 323 struct spdk_nvmf_transport_poll_group group; 324 struct spdk_sock_group *sock_group; 325 326 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 327 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 328 329 struct spdk_io_channel *accel_channel; 330 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 331 332 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 333 }; 334 335 struct spdk_nvmf_tcp_port { 336 const struct spdk_nvme_transport_id *trid; 337 struct spdk_sock *listen_sock; 338 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 339 }; 340 341 struct tcp_transport_opts { 342 bool c2h_success; 343 uint16_t control_msg_num; 344 uint32_t sock_priority; 345 }; 346 347 struct tcp_psk_entry { 348 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 349 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 350 char pskid[NVMF_PSK_IDENTITY_LEN]; 351 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 352 struct spdk_key *key; 353 354 /* Original path saved to emit SPDK configuration via "save_config". */ 355 char psk_path[PATH_MAX]; 356 uint32_t psk_size; 357 enum nvme_tcp_cipher_suite tls_cipher_suite; 358 TAILQ_ENTRY(tcp_psk_entry) link; 359 }; 360 361 struct spdk_nvmf_tcp_transport { 362 struct spdk_nvmf_transport transport; 363 struct tcp_transport_opts tcp_opts; 364 uint32_t ack_timeout; 365 366 struct spdk_nvmf_tcp_poll_group *next_pg; 367 368 struct spdk_poller *accept_poller; 369 370 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 371 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 372 373 TAILQ_HEAD(, tcp_psk_entry) psks; 374 }; 375 376 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 377 { 378 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 379 spdk_json_decode_bool, true 380 }, 381 { 382 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 383 spdk_json_decode_uint16, true 384 }, 385 { 386 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 387 spdk_json_decode_uint32, true 388 }, 389 }; 390 391 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 392 struct spdk_nvmf_tcp_req *tcp_req); 393 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 394 395 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 396 struct spdk_nvmf_tcp_req *tcp_req); 397 398 static inline void 399 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 400 enum spdk_nvmf_tcp_req_state state) 401 { 402 struct spdk_nvmf_qpair *qpair; 403 struct spdk_nvmf_tcp_qpair *tqpair; 404 405 qpair = tcp_req->req.qpair; 406 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 407 408 assert(tqpair->state_cntr[tcp_req->state] > 0); 409 tqpair->state_cntr[tcp_req->state]--; 410 tqpair->state_cntr[state]++; 411 412 tcp_req->state = state; 413 } 414 415 static inline struct nvme_tcp_pdu * 416 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 417 { 418 assert(tcp_req->pdu_in_use == false); 419 420 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 421 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 422 423 return tcp_req->pdu; 424 } 425 426 static struct spdk_nvmf_tcp_req * 427 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 428 { 429 struct spdk_nvmf_tcp_req *tcp_req; 430 431 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 432 if (spdk_unlikely(!tcp_req)) { 433 return NULL; 434 } 435 436 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 437 tcp_req->h2c_offset = 0; 438 tcp_req->has_in_capsule_data = false; 439 tcp_req->req.dif_enabled = false; 440 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 441 442 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 443 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 444 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 445 return tcp_req; 446 } 447 448 static inline void 449 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 450 { 451 assert(!tcp_req->pdu_in_use); 452 453 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 454 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 455 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 456 } 457 458 static void 459 nvmf_tcp_request_free(void *cb_arg) 460 { 461 struct spdk_nvmf_tcp_transport *ttransport; 462 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 463 464 assert(tcp_req != NULL); 465 466 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 467 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 468 struct spdk_nvmf_tcp_transport, transport); 469 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 470 nvmf_tcp_req_process(ttransport, tcp_req); 471 } 472 473 static int 474 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 475 { 476 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 477 478 nvmf_tcp_request_free(tcp_req); 479 480 return 0; 481 } 482 483 static void 484 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 485 enum spdk_nvmf_tcp_req_state state) 486 { 487 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 488 489 assert(state != TCP_REQUEST_STATE_FREE); 490 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 491 if (state == tcp_req->state) { 492 nvmf_tcp_request_free(tcp_req); 493 } 494 } 495 } 496 497 static void 498 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 499 { 500 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 501 502 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 503 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 504 505 /* Wipe the requests waiting for buffer from the global list */ 506 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 507 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 508 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 509 spdk_nvmf_request, buf_link); 510 } 511 } 512 513 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 514 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 515 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 516 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 517 } 518 519 static void 520 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 521 { 522 int i; 523 struct spdk_nvmf_tcp_req *tcp_req; 524 525 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 526 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 527 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 528 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 529 if ((int)tcp_req->state == i) { 530 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 531 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 532 } 533 } 534 } 535 } 536 537 static void 538 _nvmf_tcp_qpair_destroy(void *_tqpair) 539 { 540 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 541 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 542 void *cb_arg = tqpair->fini_cb_arg; 543 int err = 0; 544 545 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 546 547 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 548 549 err = spdk_sock_close(&tqpair->sock); 550 assert(err == 0); 551 nvmf_tcp_cleanup_all_states(tqpair); 552 553 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 554 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 555 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 556 tqpair->resource_count); 557 err++; 558 } 559 560 if (err > 0) { 561 nvmf_tcp_dump_qpair_req_contents(tqpair); 562 } 563 564 /* The timeout poller might still be registered here if we close the qpair before host 565 * terminates the connection. 566 */ 567 spdk_poller_unregister(&tqpair->timeout_poller); 568 spdk_dma_free(tqpair->pdus); 569 free(tqpair->reqs); 570 spdk_free(tqpair->bufs); 571 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 572 free(tqpair); 573 574 if (cb_fn != NULL) { 575 cb_fn(cb_arg); 576 } 577 578 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 579 } 580 581 static void 582 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 583 { 584 /* Delay the destruction to make sure it isn't performed from the context of a sock 585 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 586 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 587 */ 588 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 589 } 590 591 static void 592 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 593 { 594 struct spdk_nvmf_tcp_transport *ttransport; 595 assert(w != NULL); 596 597 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 598 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 599 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 600 } 601 602 static void 603 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 604 { 605 if (entry == NULL) { 606 return; 607 } 608 609 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 610 spdk_keyring_put_key(entry->key); 611 free(entry); 612 } 613 614 static int 615 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 616 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 617 { 618 struct spdk_nvmf_tcp_transport *ttransport; 619 struct tcp_psk_entry *entry, *tmp; 620 621 assert(transport != NULL); 622 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 623 624 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 625 TAILQ_REMOVE(&ttransport->psks, entry, link); 626 nvmf_tcp_free_psk_entry(entry); 627 } 628 629 spdk_poller_unregister(&ttransport->accept_poller); 630 free(ttransport); 631 632 if (cb_fn) { 633 cb_fn(cb_arg); 634 } 635 return 0; 636 } 637 638 static int nvmf_tcp_accept(void *ctx); 639 640 static struct spdk_nvmf_transport * 641 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 642 { 643 struct spdk_nvmf_tcp_transport *ttransport; 644 uint32_t sge_count; 645 uint32_t min_shared_buffers; 646 647 ttransport = calloc(1, sizeof(*ttransport)); 648 if (!ttransport) { 649 return NULL; 650 } 651 652 TAILQ_INIT(&ttransport->ports); 653 TAILQ_INIT(&ttransport->poll_groups); 654 TAILQ_INIT(&ttransport->psks); 655 656 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 657 658 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 659 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 660 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 661 if (opts->transport_specific != NULL && 662 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 663 SPDK_COUNTOF(tcp_transport_opts_decoder), 664 &ttransport->tcp_opts)) { 665 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 666 free(ttransport); 667 return NULL; 668 } 669 670 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 671 672 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 673 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 674 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 675 " in_capsule_data_size=%d, max_aq_depth=%d\n" 676 " num_shared_buffers=%d, c2h_success=%d,\n" 677 " dif_insert_or_strip=%d, sock_priority=%d\n" 678 " abort_timeout_sec=%d, control_msg_num=%hu\n" 679 " ack_timeout=%d\n", 680 opts->max_queue_depth, 681 opts->max_io_size, 682 opts->max_qpairs_per_ctrlr - 1, 683 opts->io_unit_size, 684 opts->in_capsule_data_size, 685 opts->max_aq_depth, 686 opts->num_shared_buffers, 687 ttransport->tcp_opts.c2h_success, 688 opts->dif_insert_or_strip, 689 ttransport->tcp_opts.sock_priority, 690 opts->abort_timeout_sec, 691 ttransport->tcp_opts.control_msg_num, 692 opts->ack_timeout); 693 694 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 695 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 696 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 697 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 698 free(ttransport); 699 return NULL; 700 } 701 702 if (ttransport->tcp_opts.control_msg_num == 0 && 703 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 704 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 705 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 706 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 707 } 708 709 /* I/O unit size cannot be larger than max I/O size */ 710 if (opts->io_unit_size > opts->max_io_size) { 711 SPDK_WARNLOG("TCP param io_unit_size %u can't be larger than max_io_size %u. Using max_io_size as io_unit_size\n", 712 opts->io_unit_size, opts->max_io_size); 713 opts->io_unit_size = opts->max_io_size; 714 } 715 716 /* In capsule data size cannot be larger than max I/O size */ 717 if (opts->in_capsule_data_size > opts->max_io_size) { 718 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 719 opts->io_unit_size, opts->max_io_size); 720 opts->in_capsule_data_size = opts->max_io_size; 721 } 722 723 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 724 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 725 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 726 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 727 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 728 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 729 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 730 } 731 732 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 733 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 734 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 735 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 736 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 737 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 738 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 739 } 740 741 sge_count = opts->max_io_size / opts->io_unit_size; 742 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 743 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 744 free(ttransport); 745 return NULL; 746 } 747 748 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 749 if (opts->buf_cache_size < UINT32_MAX) { 750 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 751 if (min_shared_buffers > opts->num_shared_buffers) { 752 SPDK_ERRLOG("There are not enough buffers to satisfy " 753 "per-poll group caches for each thread. (%" PRIu32 ") " 754 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 755 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 756 free(ttransport); 757 return NULL; 758 } 759 } 760 761 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 762 opts->acceptor_poll_rate); 763 if (!ttransport->accept_poller) { 764 free(ttransport); 765 return NULL; 766 } 767 768 return &ttransport->transport; 769 } 770 771 static int 772 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 773 { 774 unsigned long long ull; 775 char *end = NULL; 776 777 ull = strtoull(trsvcid, &end, 10); 778 if (end == NULL || end == trsvcid || *end != '\0') { 779 return -1; 780 } 781 782 /* Valid TCP/IP port numbers are in [1, 65535] */ 783 if (ull == 0 || ull > 65535) { 784 return -1; 785 } 786 787 return (int)ull; 788 } 789 790 /** 791 * Canonicalize a listen address trid. 792 */ 793 static int 794 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 795 const struct spdk_nvme_transport_id *trid) 796 { 797 int trsvcid_int; 798 799 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 800 if (trsvcid_int < 0) { 801 return -EINVAL; 802 } 803 804 memset(canon_trid, 0, sizeof(*canon_trid)); 805 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 806 canon_trid->adrfam = trid->adrfam; 807 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 808 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 809 810 return 0; 811 } 812 813 /** 814 * Find an existing listening port. 815 */ 816 static struct spdk_nvmf_tcp_port * 817 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 818 const struct spdk_nvme_transport_id *trid) 819 { 820 struct spdk_nvme_transport_id canon_trid; 821 struct spdk_nvmf_tcp_port *port; 822 823 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 824 return NULL; 825 } 826 827 TAILQ_FOREACH(port, &ttransport->ports, link) { 828 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 829 return port; 830 } 831 } 832 833 return NULL; 834 } 835 836 static int 837 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 838 void *get_key_ctx) 839 { 840 struct tcp_psk_entry *entry; 841 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 842 size_t psk_len; 843 int rc; 844 845 TAILQ_FOREACH(entry, &ttransport->psks, link) { 846 if (strcmp(pskid, entry->pskid) != 0) { 847 continue; 848 } 849 850 psk_len = entry->psk_size; 851 if ((size_t)out_len < psk_len) { 852 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 853 out_len, psk_len); 854 return -ENOBUFS; 855 } 856 857 /* Convert PSK to the TLS PSK format. */ 858 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 859 entry->tls_cipher_suite); 860 if (rc < 0) { 861 SPDK_ERRLOG("Could not generate TLS PSK\n"); 862 } 863 864 switch (entry->tls_cipher_suite) { 865 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 866 *cipher = "TLS_AES_128_GCM_SHA256"; 867 break; 868 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 869 *cipher = "TLS_AES_256_GCM_SHA384"; 870 break; 871 default: 872 *cipher = NULL; 873 return -ENOTSUP; 874 } 875 876 return rc; 877 } 878 879 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 880 881 return -EINVAL; 882 } 883 884 static int 885 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 886 struct spdk_nvmf_listen_opts *listen_opts) 887 { 888 struct spdk_nvmf_tcp_transport *ttransport; 889 struct spdk_nvmf_tcp_port *port; 890 int trsvcid_int; 891 uint8_t adrfam; 892 const char *sock_impl_name; 893 struct spdk_sock_impl_opts impl_opts; 894 size_t impl_opts_size = sizeof(impl_opts); 895 struct spdk_sock_opts opts; 896 897 if (!strlen(trid->trsvcid)) { 898 SPDK_ERRLOG("Service id is required\n"); 899 return -EINVAL; 900 } 901 902 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 903 904 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 905 if (trsvcid_int < 0) { 906 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 907 return -EINVAL; 908 } 909 910 port = calloc(1, sizeof(*port)); 911 if (!port) { 912 SPDK_ERRLOG("Port allocation failed\n"); 913 return -ENOMEM; 914 } 915 916 port->trid = trid; 917 918 sock_impl_name = NULL; 919 920 opts.opts_size = sizeof(opts); 921 spdk_sock_get_default_opts(&opts); 922 opts.priority = ttransport->tcp_opts.sock_priority; 923 opts.ack_timeout = transport->opts.ack_timeout; 924 if (listen_opts->secure_channel) { 925 if (!g_tls_log) { 926 SPDK_NOTICELOG("TLS support is considered experimental\n"); 927 g_tls_log = true; 928 } 929 sock_impl_name = "ssl"; 930 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 931 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 932 impl_opts.get_key = tcp_sock_get_key; 933 impl_opts.get_key_ctx = ttransport; 934 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 935 opts.impl_opts = &impl_opts; 936 opts.impl_opts_size = sizeof(impl_opts); 937 } 938 939 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 940 sock_impl_name, &opts); 941 if (port->listen_sock == NULL) { 942 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 943 trid->traddr, trsvcid_int, 944 spdk_strerror(errno), errno); 945 free(port); 946 return -errno; 947 } 948 949 if (spdk_sock_is_ipv4(port->listen_sock)) { 950 adrfam = SPDK_NVMF_ADRFAM_IPV4; 951 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 952 adrfam = SPDK_NVMF_ADRFAM_IPV6; 953 } else { 954 SPDK_ERRLOG("Unhandled socket type\n"); 955 adrfam = 0; 956 } 957 958 if (adrfam != trid->adrfam) { 959 SPDK_ERRLOG("Socket address family mismatch\n"); 960 spdk_sock_close(&port->listen_sock); 961 free(port); 962 return -EINVAL; 963 } 964 965 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 966 trid->traddr, trid->trsvcid); 967 968 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 969 return 0; 970 } 971 972 static void 973 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 974 const struct spdk_nvme_transport_id *trid) 975 { 976 struct spdk_nvmf_tcp_transport *ttransport; 977 struct spdk_nvmf_tcp_port *port; 978 979 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 980 981 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 982 trid->traddr, trid->trsvcid); 983 984 port = nvmf_tcp_find_port(ttransport, trid); 985 if (port) { 986 TAILQ_REMOVE(&ttransport->ports, port, link); 987 spdk_sock_close(&port->listen_sock); 988 free(port); 989 } 990 } 991 992 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 993 enum nvme_tcp_pdu_recv_state state); 994 995 static void 996 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 997 { 998 tqpair->state = state; 999 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, tqpair->state); 1000 } 1001 1002 static void 1003 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1004 { 1005 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1006 1007 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1008 1009 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 1010 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 1011 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1012 spdk_poller_unregister(&tqpair->timeout_poller); 1013 1014 /* This will end up calling nvmf_tcp_close_qpair */ 1015 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1016 } 1017 } 1018 1019 static void 1020 _mgmt_pdu_write_done(void *_tqpair, int err) 1021 { 1022 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1023 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1024 1025 if (spdk_unlikely(err != 0)) { 1026 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1027 return; 1028 } 1029 1030 assert(pdu->cb_fn != NULL); 1031 pdu->cb_fn(pdu->cb_arg); 1032 } 1033 1034 static void 1035 _req_pdu_write_done(void *req, int err) 1036 { 1037 struct spdk_nvmf_tcp_req *tcp_req = req; 1038 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1039 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1040 1041 assert(tcp_req->pdu_in_use); 1042 tcp_req->pdu_in_use = false; 1043 1044 /* If the request is in a completed state, we're waiting for write completion to free it */ 1045 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1046 nvmf_tcp_request_free(tcp_req); 1047 return; 1048 } 1049 1050 if (spdk_unlikely(err != 0)) { 1051 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1052 return; 1053 } 1054 1055 assert(pdu->cb_fn != NULL); 1056 pdu->cb_fn(pdu->cb_arg); 1057 } 1058 1059 static void 1060 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1061 { 1062 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1063 } 1064 1065 static void 1066 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1067 { 1068 int rc; 1069 uint32_t mapped_length; 1070 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1071 1072 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1073 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1074 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1075 1076 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1077 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1078 /* Try to force the send immediately. */ 1079 rc = spdk_sock_flush(tqpair->sock); 1080 if (rc > 0 && (uint32_t)rc == mapped_length) { 1081 _pdu_write_done(pdu, 0); 1082 } else { 1083 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1084 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1085 "IC_RESP" : "TERM_REQ", rc, errno); 1086 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1087 } 1088 } 1089 } 1090 1091 static void 1092 data_crc32_accel_done(void *cb_arg, int status) 1093 { 1094 struct nvme_tcp_pdu *pdu = cb_arg; 1095 1096 if (spdk_unlikely(status)) { 1097 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1098 _pdu_write_done(pdu, status); 1099 return; 1100 } 1101 1102 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1103 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1104 1105 _tcp_write_pdu(pdu); 1106 } 1107 1108 static void 1109 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1110 { 1111 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1112 int rc = 0; 1113 1114 /* Data Digest */ 1115 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1116 /* Only support this limitated case for the first step */ 1117 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1118 && tqpair->group)) { 1119 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1120 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1121 if (spdk_likely(rc == 0)) { 1122 return; 1123 } 1124 } else { 1125 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1126 } 1127 data_crc32_accel_done(pdu, rc); 1128 } else { 1129 _tcp_write_pdu(pdu); 1130 } 1131 } 1132 1133 static void 1134 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1135 struct nvme_tcp_pdu *pdu, 1136 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1137 void *cb_arg) 1138 { 1139 int hlen; 1140 uint32_t crc32c; 1141 1142 assert(tqpair->pdu_in_progress != pdu); 1143 1144 hlen = pdu->hdr.common.hlen; 1145 pdu->cb_fn = cb_fn; 1146 pdu->cb_arg = cb_arg; 1147 1148 pdu->iov[0].iov_base = &pdu->hdr.raw; 1149 pdu->iov[0].iov_len = hlen; 1150 1151 /* Header Digest */ 1152 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1153 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1154 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1155 } 1156 1157 /* Data Digest */ 1158 pdu_data_crc32_compute(pdu); 1159 } 1160 1161 static void 1162 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1163 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1164 void *cb_arg) 1165 { 1166 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1167 1168 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1169 pdu->sock_req.cb_arg = tqpair; 1170 1171 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1172 } 1173 1174 static void 1175 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1176 struct spdk_nvmf_tcp_req *tcp_req, 1177 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1178 void *cb_arg) 1179 { 1180 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1181 1182 pdu->sock_req.cb_fn = _req_pdu_write_done; 1183 pdu->sock_req.cb_arg = tcp_req; 1184 1185 assert(!tcp_req->pdu_in_use); 1186 tcp_req->pdu_in_use = true; 1187 1188 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1189 } 1190 1191 static int 1192 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1193 { 1194 uint32_t i; 1195 struct spdk_nvmf_transport_opts *opts; 1196 uint32_t in_capsule_data_size; 1197 1198 opts = &tqpair->qpair.transport->opts; 1199 1200 in_capsule_data_size = opts->in_capsule_data_size; 1201 if (opts->dif_insert_or_strip) { 1202 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1203 } 1204 1205 tqpair->resource_count = opts->max_queue_depth; 1206 1207 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1208 if (!tqpair->reqs) { 1209 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1210 return -1; 1211 } 1212 1213 if (in_capsule_data_size) { 1214 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1215 NULL, SPDK_ENV_LCORE_ID_ANY, 1216 SPDK_MALLOC_DMA); 1217 if (!tqpair->bufs) { 1218 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1219 return -1; 1220 } 1221 } 1222 /* prepare memory space for receiving pdus and tcp_req */ 1223 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1224 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1225 NULL); 1226 if (!tqpair->pdus) { 1227 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1228 return -1; 1229 } 1230 1231 for (i = 0; i < tqpair->resource_count; i++) { 1232 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1233 1234 tcp_req->ttag = i + 1; 1235 tcp_req->req.qpair = &tqpair->qpair; 1236 1237 tcp_req->pdu = &tqpair->pdus[i]; 1238 tcp_req->pdu->qpair = tqpair; 1239 1240 /* Set up memory to receive commands */ 1241 if (tqpair->bufs) { 1242 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1243 } 1244 1245 /* Set the cmdn and rsp */ 1246 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1247 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1248 1249 tcp_req->req.stripped_data = NULL; 1250 1251 /* Initialize request state to FREE */ 1252 tcp_req->state = TCP_REQUEST_STATE_FREE; 1253 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1254 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1255 } 1256 1257 for (; i < 2 * tqpair->resource_count; i++) { 1258 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1259 1260 pdu->qpair = tqpair; 1261 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1262 } 1263 1264 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1265 tqpair->mgmt_pdu->qpair = tqpair; 1266 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1267 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1268 tqpair->tcp_pdu_working_count = 1; 1269 1270 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1271 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1272 1273 return 0; 1274 } 1275 1276 static int 1277 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1278 { 1279 struct spdk_nvmf_tcp_qpair *tqpair; 1280 1281 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1282 1283 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1284 1285 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1286 1287 /* Initialise request state queues of the qpair */ 1288 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1289 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1290 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1291 1292 tqpair->host_hdgst_enable = true; 1293 tqpair->host_ddgst_enable = true; 1294 1295 return 0; 1296 } 1297 1298 static int 1299 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1300 { 1301 char saddr[32], caddr[32]; 1302 uint16_t sport, cport; 1303 char owner[256]; 1304 int rc; 1305 1306 spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1307 caddr, sizeof(caddr), &cport); 1308 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1309 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1310 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1311 1312 /* set low water mark */ 1313 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1314 if (rc != 0) { 1315 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1316 return rc; 1317 } 1318 1319 return 0; 1320 } 1321 1322 static void 1323 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1324 struct spdk_nvmf_tcp_port *port, 1325 struct spdk_sock *sock) 1326 { 1327 struct spdk_nvmf_tcp_qpair *tqpair; 1328 int rc; 1329 1330 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1331 port->trid->traddr, port->trid->trsvcid); 1332 1333 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1334 if (tqpair == NULL) { 1335 SPDK_ERRLOG("Could not allocate new connection.\n"); 1336 spdk_sock_close(&sock); 1337 return; 1338 } 1339 1340 tqpair->sock = sock; 1341 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1342 tqpair->port = port; 1343 tqpair->qpair.transport = transport; 1344 1345 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1346 sizeof(tqpair->target_addr), &tqpair->target_port, 1347 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1348 &tqpair->initiator_port); 1349 if (rc < 0) { 1350 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1351 nvmf_tcp_qpair_destroy(tqpair); 1352 return; 1353 } 1354 1355 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1356 } 1357 1358 static uint32_t 1359 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1360 { 1361 struct spdk_sock *sock; 1362 uint32_t count = 0; 1363 int i; 1364 1365 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1366 sock = spdk_sock_accept(port->listen_sock); 1367 if (sock == NULL) { 1368 break; 1369 } 1370 count++; 1371 nvmf_tcp_handle_connect(transport, port, sock); 1372 } 1373 1374 return count; 1375 } 1376 1377 static int 1378 nvmf_tcp_accept(void *ctx) 1379 { 1380 struct spdk_nvmf_transport *transport = ctx; 1381 struct spdk_nvmf_tcp_transport *ttransport; 1382 struct spdk_nvmf_tcp_port *port; 1383 uint32_t count = 0; 1384 1385 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1386 1387 TAILQ_FOREACH(port, &ttransport->ports, link) { 1388 count += nvmf_tcp_port_accept(transport, port); 1389 } 1390 1391 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1392 } 1393 1394 static void 1395 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1396 struct spdk_nvme_transport_id *trid, 1397 struct spdk_nvmf_discovery_log_page_entry *entry) 1398 { 1399 struct spdk_nvmf_tcp_port *port; 1400 struct spdk_nvmf_tcp_transport *ttransport; 1401 1402 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1403 entry->adrfam = trid->adrfam; 1404 1405 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1406 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1407 1408 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1409 port = nvmf_tcp_find_port(ttransport, trid); 1410 1411 assert(port != NULL); 1412 1413 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1414 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1415 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1416 } else { 1417 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1418 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1419 } 1420 } 1421 1422 static struct spdk_nvmf_tcp_control_msg_list * 1423 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1424 { 1425 struct spdk_nvmf_tcp_control_msg_list *list; 1426 struct spdk_nvmf_tcp_control_msg *msg; 1427 uint16_t i; 1428 1429 list = calloc(1, sizeof(*list)); 1430 if (!list) { 1431 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1432 return NULL; 1433 } 1434 1435 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1436 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1437 if (!list->msg_buf) { 1438 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1439 free(list); 1440 return NULL; 1441 } 1442 1443 STAILQ_INIT(&list->free_msgs); 1444 1445 for (i = 0; i < num_messages; i++) { 1446 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1447 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1448 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1449 } 1450 1451 return list; 1452 } 1453 1454 static void 1455 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1456 { 1457 if (!list) { 1458 return; 1459 } 1460 1461 spdk_free(list->msg_buf); 1462 free(list); 1463 } 1464 1465 static struct spdk_nvmf_transport_poll_group * 1466 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1467 struct spdk_nvmf_poll_group *group) 1468 { 1469 struct spdk_nvmf_tcp_transport *ttransport; 1470 struct spdk_nvmf_tcp_poll_group *tgroup; 1471 1472 tgroup = calloc(1, sizeof(*tgroup)); 1473 if (!tgroup) { 1474 return NULL; 1475 } 1476 1477 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1478 if (!tgroup->sock_group) { 1479 goto cleanup; 1480 } 1481 1482 TAILQ_INIT(&tgroup->qpairs); 1483 TAILQ_INIT(&tgroup->await_req); 1484 1485 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1486 1487 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1488 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1489 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1490 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1491 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1492 if (!tgroup->control_msg_list) { 1493 goto cleanup; 1494 } 1495 } 1496 1497 tgroup->accel_channel = spdk_accel_get_io_channel(); 1498 if (spdk_unlikely(!tgroup->accel_channel)) { 1499 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1500 goto cleanup; 1501 } 1502 1503 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1504 if (ttransport->next_pg == NULL) { 1505 ttransport->next_pg = tgroup; 1506 } 1507 1508 return &tgroup->group; 1509 1510 cleanup: 1511 nvmf_tcp_poll_group_destroy(&tgroup->group); 1512 return NULL; 1513 } 1514 1515 static struct spdk_nvmf_transport_poll_group * 1516 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1517 { 1518 struct spdk_nvmf_tcp_transport *ttransport; 1519 struct spdk_nvmf_tcp_poll_group **pg; 1520 struct spdk_nvmf_tcp_qpair *tqpair; 1521 struct spdk_sock_group *group = NULL, *hint = NULL; 1522 int rc; 1523 1524 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1525 1526 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1527 return NULL; 1528 } 1529 1530 pg = &ttransport->next_pg; 1531 assert(*pg != NULL); 1532 hint = (*pg)->sock_group; 1533 1534 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1535 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1536 if (rc != 0) { 1537 return NULL; 1538 } else if (group != NULL) { 1539 /* Optimal poll group was found */ 1540 return spdk_sock_group_get_ctx(group); 1541 } 1542 1543 /* The hint was used for optimal poll group, advance next_pg. */ 1544 *pg = TAILQ_NEXT(*pg, link); 1545 if (*pg == NULL) { 1546 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1547 } 1548 1549 return spdk_sock_group_get_ctx(hint); 1550 } 1551 1552 static void 1553 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1554 { 1555 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1556 struct spdk_nvmf_tcp_transport *ttransport; 1557 1558 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1559 spdk_sock_group_close(&tgroup->sock_group); 1560 if (tgroup->control_msg_list) { 1561 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1562 } 1563 1564 if (tgroup->accel_channel) { 1565 spdk_put_io_channel(tgroup->accel_channel); 1566 } 1567 1568 if (tgroup->group.transport == NULL) { 1569 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1570 * calls this function directly in a failure path. */ 1571 free(tgroup); 1572 return; 1573 } 1574 1575 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1576 1577 next_tgroup = TAILQ_NEXT(tgroup, link); 1578 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1579 if (next_tgroup == NULL) { 1580 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1581 } 1582 if (ttransport->next_pg == tgroup) { 1583 ttransport->next_pg = next_tgroup; 1584 } 1585 1586 free(tgroup); 1587 } 1588 1589 static void 1590 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1591 enum nvme_tcp_pdu_recv_state state) 1592 { 1593 if (tqpair->recv_state == state) { 1594 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1595 tqpair, state); 1596 return; 1597 } 1598 1599 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1600 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1601 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1602 tqpair->tcp_pdu_working_count--; 1603 } 1604 } 1605 1606 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1607 assert(tqpair->tcp_pdu_working_count == 0); 1608 } 1609 1610 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1611 /* When leaving the await req state, move the qpair to the main list */ 1612 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1613 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1614 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1615 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1616 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1617 } 1618 1619 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1620 tqpair->recv_state = state; 1621 1622 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, tqpair->recv_state); 1623 } 1624 1625 static int 1626 nvmf_tcp_qpair_handle_timeout(void *ctx) 1627 { 1628 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1629 1630 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1631 1632 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1633 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1634 1635 nvmf_tcp_qpair_disconnect(tqpair); 1636 return SPDK_POLLER_BUSY; 1637 } 1638 1639 static void 1640 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1641 { 1642 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1643 1644 if (!tqpair->timeout_poller) { 1645 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1646 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1647 } 1648 } 1649 1650 static void 1651 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1652 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1653 { 1654 struct nvme_tcp_pdu *rsp_pdu; 1655 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1656 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1657 uint32_t copy_len; 1658 1659 rsp_pdu = tqpair->mgmt_pdu; 1660 1661 c2h_term_req = &rsp_pdu->hdr.term_req; 1662 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1663 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1664 c2h_term_req->fes = fes; 1665 1666 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1667 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1668 DSET32(&c2h_term_req->fei, error_offset); 1669 } 1670 1671 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1672 1673 /* Copy the error info into the buffer */ 1674 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1675 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1676 1677 /* Contain the header of the wrong received pdu */ 1678 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1679 tqpair->wait_terminate = true; 1680 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1681 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1682 } 1683 1684 static void 1685 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1686 struct spdk_nvmf_tcp_qpair *tqpair, 1687 struct nvme_tcp_pdu *pdu) 1688 { 1689 struct spdk_nvmf_tcp_req *tcp_req; 1690 1691 assert(pdu->psh_valid_bytes == pdu->psh_len); 1692 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1693 1694 tcp_req = nvmf_tcp_req_get(tqpair); 1695 if (!tcp_req) { 1696 /* Directly return and make the allocation retry again. This can happen if we're 1697 * using asynchronous writes to send the response to the host or when releasing 1698 * zero-copy buffers after a response has been sent. In both cases, the host might 1699 * receive the response before we've finished processing the request and is free to 1700 * send another one. 1701 */ 1702 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1703 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1704 return; 1705 } 1706 1707 /* The host sent more commands than the maximum queue depth. */ 1708 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1709 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1710 return; 1711 } 1712 1713 pdu->req = tcp_req; 1714 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1715 nvmf_tcp_req_process(ttransport, tcp_req); 1716 } 1717 1718 static void 1719 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1720 struct spdk_nvmf_tcp_qpair *tqpair, 1721 struct nvme_tcp_pdu *pdu) 1722 { 1723 struct spdk_nvmf_tcp_req *tcp_req; 1724 struct spdk_nvme_tcp_cmd *capsule_cmd; 1725 uint32_t error_offset = 0; 1726 enum spdk_nvme_tcp_term_req_fes fes; 1727 struct spdk_nvme_cpl *rsp; 1728 1729 capsule_cmd = &pdu->hdr.capsule_cmd; 1730 tcp_req = pdu->req; 1731 assert(tcp_req != NULL); 1732 1733 /* Zero-copy requests don't support ICD */ 1734 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1735 1736 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1737 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1738 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1739 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1740 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1741 goto err; 1742 } 1743 1744 rsp = &tcp_req->req.rsp->nvme_cpl; 1745 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1746 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1747 } else { 1748 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1749 } 1750 1751 nvmf_tcp_req_process(ttransport, tcp_req); 1752 1753 return; 1754 err: 1755 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1756 } 1757 1758 static void 1759 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1760 struct spdk_nvmf_tcp_qpair *tqpair, 1761 struct nvme_tcp_pdu *pdu) 1762 { 1763 struct spdk_nvmf_tcp_req *tcp_req; 1764 uint32_t error_offset = 0; 1765 enum spdk_nvme_tcp_term_req_fes fes = 0; 1766 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1767 1768 h2c_data = &pdu->hdr.h2c_data; 1769 1770 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1771 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1772 1773 if (h2c_data->ttag > tqpair->resource_count) { 1774 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1775 tqpair->resource_count); 1776 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1777 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1778 goto err; 1779 } 1780 1781 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1782 1783 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1784 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1785 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1786 tcp_req->state); 1787 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1788 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1789 goto err; 1790 } 1791 1792 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1793 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1794 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1795 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1796 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1797 goto err; 1798 } 1799 1800 if (tcp_req->h2c_offset != h2c_data->datao) { 1801 SPDK_DEBUGLOG(nvmf_tcp, 1802 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1803 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1804 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1805 goto err; 1806 } 1807 1808 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1809 SPDK_DEBUGLOG(nvmf_tcp, 1810 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1811 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1812 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1813 goto err; 1814 } 1815 1816 pdu->req = tcp_req; 1817 1818 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1819 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1820 } 1821 1822 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1823 h2c_data->datao, h2c_data->datal); 1824 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1825 return; 1826 1827 err: 1828 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1829 } 1830 1831 static void 1832 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1833 struct spdk_nvmf_tcp_qpair *tqpair) 1834 { 1835 struct nvme_tcp_pdu *rsp_pdu; 1836 struct spdk_nvme_tcp_rsp *capsule_resp; 1837 1838 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1839 1840 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1841 assert(rsp_pdu != NULL); 1842 1843 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1844 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1845 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1846 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1847 if (tqpair->host_hdgst_enable) { 1848 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1849 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1850 } 1851 1852 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1853 } 1854 1855 static void 1856 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1857 { 1858 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1859 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1860 struct spdk_nvmf_tcp_qpair, qpair); 1861 1862 assert(tqpair != NULL); 1863 1864 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1865 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1866 tcp_req->req.length); 1867 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1868 return; 1869 } 1870 1871 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1872 nvmf_tcp_request_free(tcp_req); 1873 } else { 1874 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1875 } 1876 } 1877 1878 static void 1879 nvmf_tcp_r2t_complete(void *cb_arg) 1880 { 1881 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1882 struct spdk_nvmf_tcp_transport *ttransport; 1883 1884 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1885 struct spdk_nvmf_tcp_transport, transport); 1886 1887 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1888 1889 if (tcp_req->h2c_offset == tcp_req->req.length) { 1890 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1891 nvmf_tcp_req_process(ttransport, tcp_req); 1892 } 1893 } 1894 1895 static void 1896 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1897 struct spdk_nvmf_tcp_req *tcp_req) 1898 { 1899 struct nvme_tcp_pdu *rsp_pdu; 1900 struct spdk_nvme_tcp_r2t_hdr *r2t; 1901 1902 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1903 assert(rsp_pdu != NULL); 1904 1905 r2t = &rsp_pdu->hdr.r2t; 1906 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1907 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1908 1909 if (tqpair->host_hdgst_enable) { 1910 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1911 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1912 } 1913 1914 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1915 r2t->ttag = tcp_req->ttag; 1916 r2t->r2to = tcp_req->h2c_offset; 1917 r2t->r2tl = tcp_req->req.length; 1918 1919 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1920 1921 SPDK_DEBUGLOG(nvmf_tcp, 1922 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1923 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1924 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1925 } 1926 1927 static void 1928 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1929 struct spdk_nvmf_tcp_qpair *tqpair, 1930 struct nvme_tcp_pdu *pdu) 1931 { 1932 struct spdk_nvmf_tcp_req *tcp_req; 1933 struct spdk_nvme_cpl *rsp; 1934 1935 tcp_req = pdu->req; 1936 assert(tcp_req != NULL); 1937 1938 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1939 1940 tcp_req->h2c_offset += pdu->data_len; 1941 1942 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1943 * acknowledged before moving on. */ 1944 if (tcp_req->h2c_offset == tcp_req->req.length && 1945 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1946 /* After receiving all the h2c data, we need to check whether there is 1947 * transient transport error */ 1948 rsp = &tcp_req->req.rsp->nvme_cpl; 1949 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1950 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1951 } else { 1952 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1953 } 1954 nvmf_tcp_req_process(ttransport, tcp_req); 1955 } 1956 } 1957 1958 static void 1959 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1960 { 1961 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1962 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1963 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1964 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1965 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1966 DGET32(h2c_term_req->fei)); 1967 } 1968 } 1969 1970 static void 1971 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1972 struct nvme_tcp_pdu *pdu) 1973 { 1974 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1975 uint32_t error_offset = 0; 1976 enum spdk_nvme_tcp_term_req_fes fes; 1977 1978 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1979 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1980 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1981 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1982 goto end; 1983 } 1984 1985 /* set the data buffer */ 1986 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1987 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1988 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1989 return; 1990 end: 1991 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1992 } 1993 1994 static void 1995 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1996 struct nvme_tcp_pdu *pdu) 1997 { 1998 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1999 2000 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2001 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2002 } 2003 2004 static void 2005 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2006 { 2007 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2008 struct spdk_nvmf_tcp_transport, transport); 2009 2010 switch (pdu->hdr.common.pdu_type) { 2011 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2012 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2013 break; 2014 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2015 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2016 break; 2017 2018 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2019 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2020 break; 2021 2022 default: 2023 /* The code should not go to here */ 2024 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2025 break; 2026 } 2027 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2028 tqpair->tcp_pdu_working_count--; 2029 } 2030 2031 static inline void 2032 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2033 { 2034 treq->req.rsp->nvme_cpl.status.sct = sct; 2035 treq->req.rsp->nvme_cpl.status.sc = sc; 2036 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2037 } 2038 2039 static void 2040 data_crc32_calc_done(void *cb_arg, int status) 2041 { 2042 struct nvme_tcp_pdu *pdu = cb_arg; 2043 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2044 2045 /* async crc32 calculation is failed and use direct calculation to check */ 2046 if (spdk_unlikely(status)) { 2047 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2048 tqpair, pdu); 2049 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2050 } 2051 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2052 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2053 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2054 assert(pdu->req != NULL); 2055 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2056 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2057 } 2058 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2059 } 2060 2061 static void 2062 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2063 { 2064 int rc = 0; 2065 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2066 tqpair->pdu_in_progress = NULL; 2067 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2068 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2069 /* check data digest if need */ 2070 if (pdu->ddgst_enable) { 2071 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2072 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2073 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2074 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2075 if (spdk_likely(rc == 0)) { 2076 return; 2077 } 2078 } else { 2079 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2080 } 2081 data_crc32_calc_done(pdu, rc); 2082 } else { 2083 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2084 } 2085 } 2086 2087 static void 2088 nvmf_tcp_send_icresp_complete(void *cb_arg) 2089 { 2090 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2091 2092 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2093 } 2094 2095 static void 2096 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2097 struct spdk_nvmf_tcp_qpair *tqpair, 2098 struct nvme_tcp_pdu *pdu) 2099 { 2100 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2101 struct nvme_tcp_pdu *rsp_pdu; 2102 struct spdk_nvme_tcp_ic_resp *ic_resp; 2103 uint32_t error_offset = 0; 2104 enum spdk_nvme_tcp_term_req_fes fes; 2105 2106 /* Only PFV 0 is defined currently */ 2107 if (ic_req->pfv != 0) { 2108 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2109 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2110 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2111 goto end; 2112 } 2113 2114 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2115 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2116 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2117 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2118 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2119 goto end; 2120 } 2121 2122 /* MAXR2T is 0's based */ 2123 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2124 2125 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2126 if (!tqpair->host_hdgst_enable) { 2127 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2128 } 2129 2130 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2131 if (!tqpair->host_ddgst_enable) { 2132 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2133 } 2134 2135 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2136 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2137 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2138 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2139 tqpair, 2140 tqpair->recv_buf_size); 2141 /* Not fatal. */ 2142 } 2143 2144 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2145 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2146 2147 rsp_pdu = tqpair->mgmt_pdu; 2148 2149 ic_resp = &rsp_pdu->hdr.ic_resp; 2150 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2151 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2152 ic_resp->pfv = 0; 2153 ic_resp->cpda = tqpair->cpda; 2154 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2155 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2156 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2157 2158 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2159 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2160 2161 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2162 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2163 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2164 return; 2165 end: 2166 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2167 } 2168 2169 static void 2170 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2171 struct spdk_nvmf_tcp_transport *ttransport) 2172 { 2173 struct nvme_tcp_pdu *pdu; 2174 int rc; 2175 uint32_t crc32c, error_offset = 0; 2176 enum spdk_nvme_tcp_term_req_fes fes; 2177 2178 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2179 pdu = tqpair->pdu_in_progress; 2180 2181 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2182 pdu->hdr.common.pdu_type); 2183 /* check header digest if needed */ 2184 if (pdu->has_hdgst) { 2185 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2186 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2187 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2188 if (rc == 0) { 2189 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2190 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2191 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2192 return; 2193 2194 } 2195 } 2196 2197 switch (pdu->hdr.common.pdu_type) { 2198 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2199 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2200 break; 2201 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2202 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2203 break; 2204 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2205 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2206 break; 2207 2208 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2209 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2210 break; 2211 2212 default: 2213 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2214 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2215 error_offset = 1; 2216 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2217 break; 2218 } 2219 } 2220 2221 static void 2222 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2223 { 2224 struct nvme_tcp_pdu *pdu; 2225 uint32_t error_offset = 0; 2226 enum spdk_nvme_tcp_term_req_fes fes; 2227 uint8_t expected_hlen, pdo; 2228 bool plen_error = false, pdo_error = false; 2229 2230 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2231 pdu = tqpair->pdu_in_progress; 2232 assert(pdu); 2233 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2234 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2235 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2236 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2237 goto err; 2238 } 2239 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2240 if (pdu->hdr.common.plen != expected_hlen) { 2241 plen_error = true; 2242 } 2243 } else { 2244 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2245 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2246 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2247 goto err; 2248 } 2249 2250 switch (pdu->hdr.common.pdu_type) { 2251 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2252 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2253 pdo = pdu->hdr.common.pdo; 2254 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2255 pdo_error = true; 2256 break; 2257 } 2258 2259 if (pdu->hdr.common.plen < expected_hlen) { 2260 plen_error = true; 2261 } 2262 break; 2263 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2264 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2265 pdo = pdu->hdr.common.pdo; 2266 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2267 pdo_error = true; 2268 break; 2269 } 2270 if (pdu->hdr.common.plen < expected_hlen) { 2271 plen_error = true; 2272 } 2273 break; 2274 2275 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2276 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2277 if ((pdu->hdr.common.plen <= expected_hlen) || 2278 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2279 plen_error = true; 2280 } 2281 break; 2282 2283 default: 2284 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2285 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2286 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2287 goto err; 2288 } 2289 } 2290 2291 if (pdu->hdr.common.hlen != expected_hlen) { 2292 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2293 pdu->hdr.common.pdu_type, 2294 expected_hlen, pdu->hdr.common.hlen, tqpair); 2295 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2296 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2297 goto err; 2298 } else if (pdo_error) { 2299 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2300 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2301 } else if (plen_error) { 2302 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2303 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2304 goto err; 2305 } else { 2306 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2307 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2308 return; 2309 } 2310 err: 2311 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2312 } 2313 2314 static int 2315 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2316 { 2317 int rc = 0; 2318 struct nvme_tcp_pdu *pdu; 2319 enum nvme_tcp_pdu_recv_state prev_state; 2320 uint32_t data_len; 2321 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2322 struct spdk_nvmf_tcp_transport, transport); 2323 2324 /* The loop here is to allow for several back-to-back state changes. */ 2325 do { 2326 prev_state = tqpair->recv_state; 2327 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2328 2329 pdu = tqpair->pdu_in_progress; 2330 assert(pdu != NULL || 2331 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2332 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2333 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2334 2335 switch (tqpair->recv_state) { 2336 /* Wait for the common header */ 2337 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2338 if (!pdu) { 2339 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2340 if (spdk_unlikely(!pdu)) { 2341 return NVME_TCP_PDU_IN_PROGRESS; 2342 } 2343 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2344 tqpair->pdu_in_progress = pdu; 2345 tqpair->tcp_pdu_working_count++; 2346 } 2347 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2348 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2349 /* FALLTHROUGH */ 2350 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2351 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2352 return rc; 2353 } 2354 2355 rc = nvme_tcp_read_data(tqpair->sock, 2356 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2357 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2358 if (rc < 0) { 2359 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2360 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2361 break; 2362 } else if (rc > 0) { 2363 pdu->ch_valid_bytes += rc; 2364 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2365 } 2366 2367 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2368 return NVME_TCP_PDU_IN_PROGRESS; 2369 } 2370 2371 /* The command header of this PDU has now been read from the socket. */ 2372 nvmf_tcp_pdu_ch_handle(tqpair); 2373 break; 2374 /* Wait for the pdu specific header */ 2375 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2376 rc = nvme_tcp_read_data(tqpair->sock, 2377 pdu->psh_len - pdu->psh_valid_bytes, 2378 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2379 if (rc < 0) { 2380 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2381 break; 2382 } else if (rc > 0) { 2383 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2384 pdu->psh_valid_bytes += rc; 2385 } 2386 2387 if (pdu->psh_valid_bytes < pdu->psh_len) { 2388 return NVME_TCP_PDU_IN_PROGRESS; 2389 } 2390 2391 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2392 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2393 break; 2394 /* Wait for the req slot */ 2395 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2396 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2397 break; 2398 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2399 /* check whether the data is valid, if not we just return */ 2400 if (!pdu->data_len) { 2401 return NVME_TCP_PDU_IN_PROGRESS; 2402 } 2403 2404 data_len = pdu->data_len; 2405 /* data digest */ 2406 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2407 tqpair->host_ddgst_enable)) { 2408 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2409 pdu->ddgst_enable = true; 2410 } 2411 2412 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2413 if (rc < 0) { 2414 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2415 break; 2416 } 2417 pdu->rw_offset += rc; 2418 2419 if (pdu->rw_offset < data_len) { 2420 return NVME_TCP_PDU_IN_PROGRESS; 2421 } 2422 2423 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2424 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2425 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2426 pdu->dif_ctx) != 0) { 2427 SPDK_ERRLOG("DIF generate failed\n"); 2428 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2429 break; 2430 } 2431 2432 /* All of this PDU has now been read from the socket. */ 2433 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2434 break; 2435 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2436 if (tqpair->tcp_pdu_working_count != 0) { 2437 return NVME_TCP_PDU_IN_PROGRESS; 2438 } 2439 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2440 break; 2441 case NVME_TCP_PDU_RECV_STATE_ERROR: 2442 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2443 return NVME_TCP_PDU_IN_PROGRESS; 2444 } 2445 return NVME_TCP_PDU_FATAL; 2446 default: 2447 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2448 abort(); 2449 break; 2450 } 2451 } while (tqpair->recv_state != prev_state); 2452 2453 return rc; 2454 } 2455 2456 static inline void * 2457 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2458 { 2459 struct spdk_nvmf_tcp_control_msg *msg; 2460 2461 assert(list); 2462 2463 msg = STAILQ_FIRST(&list->free_msgs); 2464 if (!msg) { 2465 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2466 return NULL; 2467 } 2468 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2469 return msg; 2470 } 2471 2472 static inline void 2473 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2474 { 2475 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2476 2477 assert(list); 2478 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2479 } 2480 2481 static int 2482 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2483 struct spdk_nvmf_transport *transport, 2484 struct spdk_nvmf_transport_poll_group *group) 2485 { 2486 struct spdk_nvmf_request *req = &tcp_req->req; 2487 struct spdk_nvme_cmd *cmd; 2488 struct spdk_nvme_sgl_descriptor *sgl; 2489 struct spdk_nvmf_tcp_poll_group *tgroup; 2490 enum spdk_nvme_tcp_term_req_fes fes; 2491 struct nvme_tcp_pdu *pdu; 2492 struct spdk_nvmf_tcp_qpair *tqpair; 2493 uint32_t length, error_offset = 0; 2494 2495 cmd = &req->cmd->nvme_cmd; 2496 sgl = &cmd->dptr.sgl1; 2497 2498 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2499 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2500 /* get request length from sgl */ 2501 length = sgl->unkeyed.length; 2502 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2503 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2504 length, transport->opts.max_io_size); 2505 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2506 goto fatal_err; 2507 } 2508 2509 /* fill request length and populate iovs */ 2510 req->length = length; 2511 2512 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2513 2514 if (spdk_unlikely(req->dif_enabled)) { 2515 req->dif.orig_length = length; 2516 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2517 req->dif.elba_length = length; 2518 } 2519 2520 if (nvmf_ctrlr_use_zcopy(req)) { 2521 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2522 req->data_from_pool = false; 2523 return 0; 2524 } 2525 2526 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2527 /* No available buffers. Queue this request up. */ 2528 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2529 tcp_req); 2530 return 0; 2531 } 2532 2533 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2534 tcp_req, req->iovcnt, req->iov[0].iov_base); 2535 2536 return 0; 2537 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2538 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2539 uint64_t offset = sgl->address; 2540 uint32_t max_len = transport->opts.in_capsule_data_size; 2541 2542 assert(tcp_req->has_in_capsule_data); 2543 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2544 tqpair = tcp_req->pdu->qpair; 2545 /* receiving pdu is not same with the pdu in tcp_req */ 2546 pdu = tqpair->pdu_in_progress; 2547 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2548 if (tqpair->host_ddgst_enable) { 2549 length -= SPDK_NVME_TCP_DIGEST_LEN; 2550 } 2551 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2552 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2553 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2554 length, sgl->unkeyed.length); 2555 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2556 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2557 goto fatal_err; 2558 } 2559 2560 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2561 offset, length); 2562 2563 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2564 if (spdk_unlikely(offset != 0)) { 2565 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2566 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2567 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2568 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2569 goto fatal_err; 2570 } 2571 2572 if (spdk_unlikely(length > max_len)) { 2573 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2574 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2575 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2576 2577 /* Get a buffer from dedicated list */ 2578 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2579 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2580 assert(tgroup->control_msg_list); 2581 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2582 if (!req->iov[0].iov_base) { 2583 /* No available buffers. Queue this request up. */ 2584 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2585 return 0; 2586 } 2587 } else { 2588 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2589 length, max_len); 2590 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2591 goto fatal_err; 2592 } 2593 } else { 2594 req->iov[0].iov_base = tcp_req->buf; 2595 } 2596 2597 req->length = length; 2598 req->data_from_pool = false; 2599 2600 if (spdk_unlikely(req->dif_enabled)) { 2601 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2602 req->dif.elba_length = length; 2603 } 2604 2605 req->iov[0].iov_len = length; 2606 req->iovcnt = 1; 2607 2608 return 0; 2609 } 2610 /* If we want to handle the problem here, then we can't skip the following data segment. 2611 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2612 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2613 sgl->generic.type, sgl->generic.subtype); 2614 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2615 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2616 fatal_err: 2617 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2618 return -1; 2619 } 2620 2621 static inline enum spdk_nvme_media_error_status_code 2622 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2623 enum spdk_nvme_media_error_status_code result; 2624 2625 switch (err_type) 2626 { 2627 case SPDK_DIF_REFTAG_ERROR: 2628 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2629 break; 2630 case SPDK_DIF_APPTAG_ERROR: 2631 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2632 break; 2633 case SPDK_DIF_GUARD_ERROR: 2634 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2635 break; 2636 default: 2637 SPDK_UNREACHABLE(); 2638 break; 2639 } 2640 2641 return result; 2642 } 2643 2644 static void 2645 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2646 struct spdk_nvmf_tcp_req *tcp_req) 2647 { 2648 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2649 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2650 struct nvme_tcp_pdu *rsp_pdu; 2651 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2652 uint32_t plen, pdo, alignment; 2653 int rc; 2654 2655 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2656 2657 rsp_pdu = tcp_req->pdu; 2658 assert(rsp_pdu != NULL); 2659 2660 c2h_data = &rsp_pdu->hdr.c2h_data; 2661 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2662 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2663 2664 if (tqpair->host_hdgst_enable) { 2665 plen += SPDK_NVME_TCP_DIGEST_LEN; 2666 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2667 } 2668 2669 /* set the psh */ 2670 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2671 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2672 c2h_data->datao = tcp_req->pdu->rw_offset; 2673 2674 /* set the padding */ 2675 rsp_pdu->padding_len = 0; 2676 pdo = plen; 2677 if (tqpair->cpda) { 2678 alignment = (tqpair->cpda + 1) << 2; 2679 if (plen % alignment != 0) { 2680 pdo = (plen + alignment) / alignment * alignment; 2681 rsp_pdu->padding_len = pdo - plen; 2682 plen = pdo; 2683 } 2684 } 2685 2686 c2h_data->common.pdo = pdo; 2687 plen += c2h_data->datal; 2688 if (tqpair->host_ddgst_enable) { 2689 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2690 plen += SPDK_NVME_TCP_DIGEST_LEN; 2691 } 2692 2693 c2h_data->common.plen = plen; 2694 2695 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2696 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2697 } 2698 2699 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2700 c2h_data->datao, c2h_data->datal); 2701 2702 2703 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2704 /* Need to send the capsule response if response is not all 0 */ 2705 if (ttransport->tcp_opts.c2h_success && 2706 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2707 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2708 } 2709 2710 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2711 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2712 struct spdk_dif_error err_blk = {}; 2713 uint32_t mapped_length = 0; 2714 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2715 uint32_t ddgst_len = 0; 2716 2717 if (tqpair->host_ddgst_enable) { 2718 /* Data digest consumes additional iov entry */ 2719 available_iovs--; 2720 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2721 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2722 c2h_data->common.plen -= ddgst_len; 2723 } 2724 /* Temp call to estimate if data can be described by limited number of iovs. 2725 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2726 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2727 false, &mapped_length); 2728 2729 if (mapped_length != c2h_data->common.plen) { 2730 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2731 SPDK_DEBUGLOG(nvmf_tcp, 2732 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2733 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2734 c2h_data->common.plen = mapped_length; 2735 2736 /* Rebuild pdu->data_iov since data length is changed */ 2737 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2738 c2h_data->datal); 2739 2740 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2741 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2742 } 2743 2744 c2h_data->common.plen += ddgst_len; 2745 2746 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2747 2748 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2749 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2750 if (rc != 0) { 2751 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2752 err_blk.err_type, err_blk.err_offset); 2753 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2754 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2755 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2756 return; 2757 } 2758 } 2759 2760 rsp_pdu->rw_offset += c2h_data->datal; 2761 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2762 } 2763 2764 static void 2765 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2766 struct spdk_nvmf_tcp_req *tcp_req) 2767 { 2768 nvmf_tcp_req_pdu_init(tcp_req); 2769 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2770 } 2771 2772 static int 2773 request_transfer_out(struct spdk_nvmf_request *req) 2774 { 2775 struct spdk_nvmf_tcp_req *tcp_req; 2776 struct spdk_nvmf_qpair *qpair; 2777 struct spdk_nvmf_tcp_qpair *tqpair; 2778 struct spdk_nvme_cpl *rsp; 2779 2780 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2781 2782 qpair = req->qpair; 2783 rsp = &req->rsp->nvme_cpl; 2784 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2785 2786 /* Advance our sq_head pointer */ 2787 if (qpair->sq_head == qpair->sq_head_max) { 2788 qpair->sq_head = 0; 2789 } else { 2790 qpair->sq_head++; 2791 } 2792 rsp->sqhd = qpair->sq_head; 2793 2794 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2795 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2796 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2797 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2798 } else { 2799 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2800 } 2801 2802 return 0; 2803 } 2804 2805 static void 2806 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2807 struct spdk_nvmf_tcp_qpair *tqpair, 2808 struct spdk_nvmf_tcp_req *tcp_req) 2809 { 2810 enum spdk_nvme_cmd_fuse last, next; 2811 2812 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2813 next = tcp_req->cmd.fuse; 2814 2815 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2816 2817 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2818 return; 2819 } 2820 2821 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2822 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2823 /* This is a valid pair of fused commands. Point them at each other 2824 * so they can be submitted consecutively once ready to be executed. 2825 */ 2826 tqpair->fused_first->fused_pair = tcp_req; 2827 tcp_req->fused_pair = tqpair->fused_first; 2828 tqpair->fused_first = NULL; 2829 return; 2830 } else { 2831 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2832 tqpair->fused_first->fused_failed = true; 2833 2834 /* 2835 * If the last req is in READY_TO_EXECUTE state, then call 2836 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2837 */ 2838 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2839 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2840 } 2841 2842 tqpair->fused_first = NULL; 2843 } 2844 } 2845 2846 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2847 /* Set tqpair->fused_first here so that we know to check that the next request 2848 * is a SECOND (and to fail this one if it isn't). 2849 */ 2850 tqpair->fused_first = tcp_req; 2851 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2852 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2853 tcp_req->fused_failed = true; 2854 } 2855 } 2856 2857 static bool 2858 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2859 struct spdk_nvmf_tcp_req *tcp_req) 2860 { 2861 struct spdk_nvmf_tcp_qpair *tqpair; 2862 uint32_t plen; 2863 struct nvme_tcp_pdu *pdu; 2864 enum spdk_nvmf_tcp_req_state prev_state; 2865 bool progress = false; 2866 struct spdk_nvmf_transport *transport = &ttransport->transport; 2867 struct spdk_nvmf_transport_poll_group *group; 2868 struct spdk_nvmf_tcp_poll_group *tgroup; 2869 2870 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2871 group = &tqpair->group->group; 2872 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2873 2874 /* If the qpair is not active, we need to abort the outstanding requests. */ 2875 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2876 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2877 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2878 } 2879 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2880 } 2881 2882 /* The loop here is to allow for several back-to-back state changes. */ 2883 do { 2884 prev_state = tcp_req->state; 2885 2886 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2887 tqpair); 2888 2889 switch (tcp_req->state) { 2890 case TCP_REQUEST_STATE_FREE: 2891 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2892 * to escape this state. */ 2893 break; 2894 case TCP_REQUEST_STATE_NEW: 2895 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 2896 2897 /* copy the cmd from the receive pdu */ 2898 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2899 2900 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2901 tcp_req->req.dif_enabled = true; 2902 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2903 } 2904 2905 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2906 2907 /* The next state transition depends on the data transfer needs of this request. */ 2908 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2909 2910 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2911 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2912 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2913 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2914 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2915 break; 2916 } 2917 2918 /* If no data to transfer, ready to execute. */ 2919 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2920 /* Reset the tqpair receiving pdu state */ 2921 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2922 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2923 break; 2924 } 2925 2926 pdu = tqpair->pdu_in_progress; 2927 plen = pdu->hdr.common.hlen; 2928 if (tqpair->host_hdgst_enable) { 2929 plen += SPDK_NVME_TCP_DIGEST_LEN; 2930 } 2931 if (pdu->hdr.common.plen != plen) { 2932 tcp_req->has_in_capsule_data = true; 2933 } else { 2934 /* Data is transmitted by C2H PDUs */ 2935 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2936 } 2937 2938 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2939 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2940 break; 2941 case TCP_REQUEST_STATE_NEED_BUFFER: 2942 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 2943 (uintptr_t)tcp_req); 2944 2945 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2946 2947 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2948 SPDK_DEBUGLOG(nvmf_tcp, 2949 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2950 tcp_req, tqpair); 2951 /* This request needs to wait in line to obtain a buffer */ 2952 break; 2953 } 2954 2955 /* Try to get a data buffer */ 2956 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 2957 break; 2958 } 2959 2960 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2961 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2962 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2963 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2964 tcp_req->req.length = tcp_req->req.dif.elba_length; 2965 } 2966 2967 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2968 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2969 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2970 break; 2971 } 2972 2973 if (tcp_req->req.iovcnt < 1) { 2974 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2975 tcp_req, tqpair); 2976 /* No buffers available. */ 2977 break; 2978 } 2979 2980 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2981 2982 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2983 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2984 if (tcp_req->req.data_from_pool) { 2985 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2986 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2987 } else { 2988 struct nvme_tcp_pdu *pdu; 2989 2990 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2991 2992 pdu = tqpair->pdu_in_progress; 2993 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2994 tqpair); 2995 /* No need to send r2t, contained in the capsuled data */ 2996 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2997 0, tcp_req->req.length); 2998 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2999 } 3000 break; 3001 } 3002 3003 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3004 break; 3005 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3006 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3007 (uintptr_t)tcp_req); 3008 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3009 * to escape this state. */ 3010 break; 3011 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3012 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3013 (uintptr_t)tcp_req); 3014 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3015 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3016 tcp_req, tqpair); 3017 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3018 break; 3019 } 3020 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3021 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3022 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3023 } else { 3024 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3025 } 3026 break; 3027 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3028 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3029 (uintptr_t)tcp_req); 3030 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3031 break; 3032 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3033 3034 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3035 0, (uintptr_t)tcp_req); 3036 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3037 * to escape this state. */ 3038 break; 3039 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3040 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3041 (uintptr_t)tcp_req); 3042 3043 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3044 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3045 tcp_req->req.length = tcp_req->req.dif.elba_length; 3046 } 3047 3048 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3049 if (tcp_req->fused_failed) { 3050 /* This request failed FUSED semantics. Fail it immediately, without 3051 * even sending it to the target layer. 3052 */ 3053 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3054 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3055 break; 3056 } 3057 3058 if (tcp_req->fused_pair == NULL || 3059 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3060 /* This request is ready to execute, but either we don't know yet if it's 3061 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3062 * or the other request of this fused pair isn't ready to execute. So 3063 * break here and this request will get processed later either when the 3064 * other request is ready or we find that this request isn't valid. 3065 */ 3066 break; 3067 } 3068 } 3069 3070 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3071 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3072 /* If we get to this point, and this request is a fused command, we know that 3073 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3074 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3075 * request, and the other request of the fused pair, in the correct order. 3076 * Also clear the ->fused_pair pointers on both requests, since after this point 3077 * we no longer need to maintain the relationship between these two requests. 3078 */ 3079 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3080 assert(tcp_req->fused_pair != NULL); 3081 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3082 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3083 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3084 tcp_req->fused_pair->fused_pair = NULL; 3085 tcp_req->fused_pair = NULL; 3086 } 3087 spdk_nvmf_request_exec(&tcp_req->req); 3088 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3089 assert(tcp_req->fused_pair != NULL); 3090 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3091 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3092 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3093 tcp_req->fused_pair->fused_pair = NULL; 3094 tcp_req->fused_pair = NULL; 3095 } 3096 } else { 3097 /* For zero-copy, only requests with data coming from host to the 3098 * controller can end up here. */ 3099 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3100 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3101 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3102 } 3103 3104 break; 3105 case TCP_REQUEST_STATE_EXECUTING: 3106 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3107 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3108 * to escape this state. */ 3109 break; 3110 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3111 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3112 (uintptr_t)tcp_req); 3113 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3114 * to escape this state. */ 3115 break; 3116 case TCP_REQUEST_STATE_EXECUTED: 3117 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3118 3119 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3120 tcp_req->req.length = tcp_req->req.dif.orig_length; 3121 } 3122 3123 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3124 break; 3125 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3126 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3127 (uintptr_t)tcp_req); 3128 if (request_transfer_out(&tcp_req->req) != 0) { 3129 assert(0); /* No good way to handle this currently */ 3130 } 3131 break; 3132 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3133 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3134 0, (uintptr_t)tcp_req); 3135 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3136 * to escape this state. */ 3137 break; 3138 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3139 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3140 (uintptr_t)tcp_req); 3141 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3142 * to escape this state. */ 3143 break; 3144 case TCP_REQUEST_STATE_COMPLETED: 3145 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3146 /* If there's an outstanding PDU sent to the host, the request is completed 3147 * due to the qpair being disconnected. We must delay the completion until 3148 * that write is done to avoid freeing the request twice. */ 3149 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3150 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3151 "write on req=%p\n", tcp_req); 3152 /* This can only happen for zcopy requests */ 3153 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3154 assert(tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 3155 break; 3156 } 3157 3158 if (tcp_req->req.data_from_pool) { 3159 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3160 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3161 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3162 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3163 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3164 assert(tgroup->control_msg_list); 3165 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3166 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3167 tcp_req->req.iov[0].iov_base); 3168 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3169 /* If the request has an unreleased zcopy bdev_io, it's either a 3170 * read, a failed write, or the qpair is being disconnected */ 3171 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3172 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3173 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3174 tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE); 3175 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3176 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3177 break; 3178 } 3179 tcp_req->req.length = 0; 3180 tcp_req->req.iovcnt = 0; 3181 tcp_req->fused_failed = false; 3182 if (tcp_req->fused_pair) { 3183 /* This req was part of a valid fused pair, but failed before it got to 3184 * READ_TO_EXECUTE state. This means we need to fail the other request 3185 * in the pair, because it is no longer part of a valid pair. If the pair 3186 * already reached READY_TO_EXECUTE state, we need to kick it. 3187 */ 3188 tcp_req->fused_pair->fused_failed = true; 3189 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3190 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3191 } 3192 tcp_req->fused_pair = NULL; 3193 } 3194 3195 nvmf_tcp_req_put(tqpair, tcp_req); 3196 break; 3197 case TCP_REQUEST_NUM_STATES: 3198 default: 3199 assert(0); 3200 break; 3201 } 3202 3203 if (tcp_req->state != prev_state) { 3204 progress = true; 3205 } 3206 } while (tcp_req->state != prev_state); 3207 3208 return progress; 3209 } 3210 3211 static void 3212 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3213 { 3214 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3215 int rc; 3216 3217 assert(tqpair != NULL); 3218 rc = nvmf_tcp_sock_process(tqpair); 3219 3220 /* If there was a new socket error, disconnect */ 3221 if (rc < 0) { 3222 nvmf_tcp_qpair_disconnect(tqpair); 3223 } 3224 } 3225 3226 static int 3227 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3228 struct spdk_nvmf_qpair *qpair) 3229 { 3230 struct spdk_nvmf_tcp_poll_group *tgroup; 3231 struct spdk_nvmf_tcp_qpair *tqpair; 3232 int rc; 3233 3234 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3235 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3236 3237 rc = nvmf_tcp_qpair_sock_init(tqpair); 3238 if (rc != 0) { 3239 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3240 return -1; 3241 } 3242 3243 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3244 if (rc < 0) { 3245 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3246 return -1; 3247 } 3248 3249 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3250 if (rc < 0) { 3251 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3252 return -1; 3253 } 3254 3255 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3256 nvmf_tcp_sock_cb, tqpair); 3257 if (rc != 0) { 3258 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3259 spdk_strerror(errno), errno); 3260 return -1; 3261 } 3262 3263 tqpair->group = tgroup; 3264 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3265 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3266 3267 return 0; 3268 } 3269 3270 static int 3271 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3272 struct spdk_nvmf_qpair *qpair) 3273 { 3274 struct spdk_nvmf_tcp_poll_group *tgroup; 3275 struct spdk_nvmf_tcp_qpair *tqpair; 3276 int rc; 3277 3278 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3279 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3280 3281 assert(tqpair->group == tgroup); 3282 3283 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3284 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3285 /* Change the state to move the qpair from the await_req list to the main list 3286 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3287 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3288 } 3289 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3290 3291 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3292 if (rc != 0) { 3293 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3294 spdk_strerror(errno), errno); 3295 } 3296 3297 return rc; 3298 } 3299 3300 static int 3301 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3302 { 3303 struct spdk_nvmf_tcp_transport *ttransport; 3304 struct spdk_nvmf_tcp_req *tcp_req; 3305 3306 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3307 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3308 3309 switch (tcp_req->state) { 3310 case TCP_REQUEST_STATE_EXECUTING: 3311 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3312 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3313 break; 3314 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3315 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3316 break; 3317 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3318 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3319 break; 3320 default: 3321 assert(0 && "Unexpected request state"); 3322 break; 3323 } 3324 3325 nvmf_tcp_req_process(ttransport, tcp_req); 3326 3327 return 0; 3328 } 3329 3330 static void 3331 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3332 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3333 { 3334 struct spdk_nvmf_tcp_qpair *tqpair; 3335 3336 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3337 3338 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3339 3340 assert(tqpair->fini_cb_fn == NULL); 3341 tqpair->fini_cb_fn = cb_fn; 3342 tqpair->fini_cb_arg = cb_arg; 3343 3344 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3345 nvmf_tcp_qpair_destroy(tqpair); 3346 } 3347 3348 static int 3349 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3350 { 3351 struct spdk_nvmf_tcp_poll_group *tgroup; 3352 int rc; 3353 struct spdk_nvmf_request *req, *req_tmp; 3354 struct spdk_nvmf_tcp_req *tcp_req; 3355 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3356 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3357 struct spdk_nvmf_tcp_transport, transport); 3358 3359 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3360 3361 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3362 return 0; 3363 } 3364 3365 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3366 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3367 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3368 break; 3369 } 3370 } 3371 3372 rc = spdk_sock_group_poll(tgroup->sock_group); 3373 if (rc < 0) { 3374 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3375 } 3376 3377 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3378 rc = nvmf_tcp_sock_process(tqpair); 3379 3380 /* If there was a new socket error, disconnect */ 3381 if (rc < 0) { 3382 nvmf_tcp_qpair_disconnect(tqpair); 3383 } 3384 } 3385 3386 return rc; 3387 } 3388 3389 static int 3390 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3391 struct spdk_nvme_transport_id *trid, bool peer) 3392 { 3393 struct spdk_nvmf_tcp_qpair *tqpair; 3394 uint16_t port; 3395 3396 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3397 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3398 3399 if (peer) { 3400 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3401 port = tqpair->initiator_port; 3402 } else { 3403 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3404 port = tqpair->target_port; 3405 } 3406 3407 if (spdk_sock_is_ipv4(tqpair->sock)) { 3408 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3409 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3410 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3411 } else { 3412 return -1; 3413 } 3414 3415 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3416 return 0; 3417 } 3418 3419 static int 3420 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3421 struct spdk_nvme_transport_id *trid) 3422 { 3423 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3424 } 3425 3426 static int 3427 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3428 struct spdk_nvme_transport_id *trid) 3429 { 3430 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3431 } 3432 3433 static int 3434 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3435 struct spdk_nvme_transport_id *trid) 3436 { 3437 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3438 } 3439 3440 static void 3441 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3442 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3443 { 3444 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3445 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3446 3447 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3448 } 3449 3450 static int 3451 _nvmf_tcp_qpair_abort_request(void *ctx) 3452 { 3453 struct spdk_nvmf_request *req = ctx; 3454 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3455 struct spdk_nvmf_tcp_req, req); 3456 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3457 struct spdk_nvmf_tcp_qpair, qpair); 3458 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3459 struct spdk_nvmf_tcp_transport, transport); 3460 int rc; 3461 3462 spdk_poller_unregister(&req->poller); 3463 3464 switch (tcp_req_to_abort->state) { 3465 case TCP_REQUEST_STATE_EXECUTING: 3466 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3467 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3468 rc = nvmf_ctrlr_abort_request(req); 3469 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3470 return SPDK_POLLER_BUSY; 3471 } 3472 break; 3473 3474 case TCP_REQUEST_STATE_NEED_BUFFER: 3475 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3476 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3477 3478 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3479 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3480 break; 3481 3482 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3483 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3484 if (spdk_get_ticks() < req->timeout_tsc) { 3485 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3486 return SPDK_POLLER_BUSY; 3487 } 3488 break; 3489 3490 default: 3491 /* Requests in other states are either un-abortable (e.g. 3492 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3493 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3494 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3495 * abort request with the bit0 of dword0 set (command not aborted). 3496 */ 3497 break; 3498 } 3499 3500 spdk_nvmf_request_complete(req); 3501 return SPDK_POLLER_BUSY; 3502 } 3503 3504 static void 3505 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3506 struct spdk_nvmf_request *req) 3507 { 3508 struct spdk_nvmf_tcp_qpair *tqpair; 3509 struct spdk_nvmf_tcp_transport *ttransport; 3510 struct spdk_nvmf_transport *transport; 3511 uint16_t cid; 3512 uint32_t i; 3513 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3514 3515 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3516 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3517 transport = &ttransport->transport; 3518 3519 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3520 3521 for (i = 0; i < tqpair->resource_count; i++) { 3522 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3523 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3524 tcp_req_to_abort = &tqpair->reqs[i]; 3525 break; 3526 } 3527 } 3528 3529 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3530 3531 if (tcp_req_to_abort == NULL) { 3532 spdk_nvmf_request_complete(req); 3533 return; 3534 } 3535 3536 req->req_to_abort = &tcp_req_to_abort->req; 3537 req->timeout_tsc = spdk_get_ticks() + 3538 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3539 req->poller = NULL; 3540 3541 _nvmf_tcp_qpair_abort_request(req); 3542 } 3543 3544 struct tcp_subsystem_add_host_opts { 3545 char *psk; 3546 }; 3547 3548 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3549 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3550 }; 3551 3552 static int 3553 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3554 { 3555 FILE *psk_file; 3556 struct stat statbuf; 3557 int rc; 3558 3559 if (stat(fname, &statbuf) != 0) { 3560 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3561 return -EACCES; 3562 } 3563 3564 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3565 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3566 return -EPERM; 3567 } 3568 if ((size_t)statbuf.st_size > bufsz) { 3569 SPDK_ERRLOG("Invalid PSK: too long\n"); 3570 return -EINVAL; 3571 } 3572 psk_file = fopen(fname, "r"); 3573 if (psk_file == NULL) { 3574 SPDK_ERRLOG("Could not open PSK file\n"); 3575 return -EINVAL; 3576 } 3577 3578 rc = fread(buf, 1, statbuf.st_size, psk_file); 3579 if (rc != statbuf.st_size) { 3580 SPDK_ERRLOG("Failed to read PSK\n"); 3581 fclose(psk_file); 3582 return -EINVAL; 3583 } 3584 3585 fclose(psk_file); 3586 return 0; 3587 } 3588 3589 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3590 3591 static int 3592 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3593 const struct spdk_nvmf_subsystem *subsystem, 3594 const char *hostnqn, 3595 const struct spdk_json_val *transport_specific) 3596 { 3597 struct tcp_subsystem_add_host_opts opts; 3598 struct spdk_nvmf_tcp_transport *ttransport; 3599 struct tcp_psk_entry *tmp, *entry = NULL; 3600 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3601 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3602 uint8_t tls_cipher_suite; 3603 int rc = 0; 3604 uint8_t psk_retained_hash; 3605 uint64_t psk_configured_size; 3606 3607 if (transport_specific == NULL) { 3608 return 0; 3609 } 3610 3611 assert(transport != NULL); 3612 assert(subsystem != NULL); 3613 3614 memset(&opts, 0, sizeof(opts)); 3615 3616 /* Decode PSK (either name of a key or file path) */ 3617 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3618 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3619 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3620 return -EINVAL; 3621 } 3622 3623 if (opts.psk == NULL) { 3624 return 0; 3625 } 3626 3627 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3628 if (entry == NULL) { 3629 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3630 rc = -ENOMEM; 3631 goto end; 3632 } 3633 3634 entry->key = spdk_keyring_get_key(opts.psk); 3635 if (entry->key != NULL) { 3636 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3637 if (rc < 0) { 3638 SPDK_ERRLOG("Failed to retreive PSK '%s'\n", opts.psk); 3639 rc = -EINVAL; 3640 goto end; 3641 } 3642 } else { 3643 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3644 SPDK_ERRLOG("PSK path too long\n"); 3645 rc = -EINVAL; 3646 goto end; 3647 } 3648 3649 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3650 if (rc) { 3651 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3652 goto end; 3653 } 3654 3655 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3656 } 3657 3658 /* Parse PSK interchange to get length of base64 encoded data. 3659 * This is then used to decide which cipher suite should be used 3660 * to generate PSK identity and TLS PSK later on. */ 3661 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3662 &psk_configured_size, &psk_retained_hash); 3663 if (rc < 0) { 3664 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3665 goto end; 3666 } 3667 3668 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3669 * This check also ensures that psk_configured_size is smaller than 3670 * psk_retained buffer size. */ 3671 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3672 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3673 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3674 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3675 } else { 3676 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3677 rc = -EINVAL; 3678 goto end; 3679 } 3680 3681 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3682 /* Generate PSK identity. */ 3683 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3684 subsystem->subnqn, tls_cipher_suite); 3685 if (rc) { 3686 rc = -EINVAL; 3687 goto end; 3688 } 3689 /* Check if PSK identity entry already exists. */ 3690 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3691 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3692 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3693 rc = -EEXIST; 3694 goto end; 3695 } 3696 } 3697 3698 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3699 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3700 rc = -EINVAL; 3701 goto end; 3702 } 3703 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3704 SPDK_ERRLOG("Could not write subnqn string!\n"); 3705 rc = -EINVAL; 3706 goto end; 3707 } 3708 3709 entry->tls_cipher_suite = tls_cipher_suite; 3710 3711 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3712 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3713 /* Psk configured is either 32 or 48 bytes long. */ 3714 memcpy(entry->psk, psk_configured, psk_configured_size); 3715 entry->psk_size = psk_configured_size; 3716 } else { 3717 /* Derive retained PSK. */ 3718 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3719 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3720 if (rc < 0) { 3721 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3722 goto end; 3723 } 3724 entry->psk_size = rc; 3725 } 3726 3727 if (entry->key == NULL) { 3728 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3729 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3730 SPDK_ERRLOG("Could not save PSK path!\n"); 3731 rc = -ENAMETOOLONG; 3732 goto end; 3733 } 3734 } 3735 3736 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3737 rc = 0; 3738 3739 end: 3740 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3741 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3742 3743 free(opts.psk); 3744 if (rc != 0) { 3745 nvmf_tcp_free_psk_entry(entry); 3746 } 3747 3748 return rc; 3749 } 3750 3751 static void 3752 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3753 const struct spdk_nvmf_subsystem *subsystem, 3754 const char *hostnqn) 3755 { 3756 struct spdk_nvmf_tcp_transport *ttransport; 3757 struct tcp_psk_entry *entry, *tmp; 3758 3759 assert(transport != NULL); 3760 assert(subsystem != NULL); 3761 3762 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3763 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3764 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3765 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3766 TAILQ_REMOVE(&ttransport->psks, entry, link); 3767 nvmf_tcp_free_psk_entry(entry); 3768 break; 3769 } 3770 } 3771 } 3772 3773 static void 3774 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3775 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3776 struct spdk_json_write_ctx *w) 3777 { 3778 struct spdk_nvmf_tcp_transport *ttransport; 3779 struct tcp_psk_entry *entry; 3780 3781 assert(transport != NULL); 3782 assert(subsystem != NULL); 3783 3784 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3785 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3786 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3787 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3788 spdk_json_write_named_string(w, "psk", entry->key ? 3789 spdk_key_get_name(entry->key) : entry->psk_path); 3790 break; 3791 } 3792 } 3793 } 3794 3795 static void 3796 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3797 { 3798 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3799 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3800 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3801 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3802 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3803 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3804 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3805 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3806 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3807 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3808 opts->transport_specific = NULL; 3809 } 3810 3811 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3812 .name = "TCP", 3813 .type = SPDK_NVME_TRANSPORT_TCP, 3814 .opts_init = nvmf_tcp_opts_init, 3815 .create = nvmf_tcp_create, 3816 .dump_opts = nvmf_tcp_dump_opts, 3817 .destroy = nvmf_tcp_destroy, 3818 3819 .listen = nvmf_tcp_listen, 3820 .stop_listen = nvmf_tcp_stop_listen, 3821 3822 .listener_discover = nvmf_tcp_discover, 3823 3824 .poll_group_create = nvmf_tcp_poll_group_create, 3825 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3826 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3827 .poll_group_add = nvmf_tcp_poll_group_add, 3828 .poll_group_remove = nvmf_tcp_poll_group_remove, 3829 .poll_group_poll = nvmf_tcp_poll_group_poll, 3830 3831 .req_free = nvmf_tcp_req_free, 3832 .req_complete = nvmf_tcp_req_complete, 3833 3834 .qpair_fini = nvmf_tcp_close_qpair, 3835 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3836 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3837 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3838 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3839 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3840 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3841 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3842 }; 3843 3844 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3845 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3846