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) 2022-2024 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, "qd"); 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, "qd"); 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 tqpair->qpair.queue_depth++; 445 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 446 return tcp_req; 447 } 448 449 static inline void 450 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 451 { 452 assert(!tcp_req->pdu_in_use); 453 454 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 455 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 456 tqpair->qpair.queue_depth--; 457 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 458 } 459 460 static void 461 nvmf_tcp_request_free(void *cb_arg) 462 { 463 struct spdk_nvmf_tcp_transport *ttransport; 464 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 465 466 assert(tcp_req != NULL); 467 468 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 469 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 470 struct spdk_nvmf_tcp_transport, transport); 471 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 472 nvmf_tcp_req_process(ttransport, tcp_req); 473 } 474 475 static int 476 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 477 { 478 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 479 480 nvmf_tcp_request_free(tcp_req); 481 482 return 0; 483 } 484 485 static void 486 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 487 enum spdk_nvmf_tcp_req_state state) 488 { 489 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 490 491 assert(state != TCP_REQUEST_STATE_FREE); 492 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 493 if (state == tcp_req->state) { 494 nvmf_tcp_request_free(tcp_req); 495 } 496 } 497 } 498 499 static void 500 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 501 { 502 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 503 504 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 505 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 506 507 /* Wipe the requests waiting for buffer from the global list */ 508 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 509 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 510 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 511 spdk_nvmf_request, buf_link); 512 } 513 } 514 515 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 516 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 517 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 518 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 519 } 520 521 static void 522 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 523 { 524 int i; 525 struct spdk_nvmf_tcp_req *tcp_req; 526 527 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 528 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 529 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 530 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 531 if ((int)tcp_req->state == i) { 532 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 533 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 534 } 535 } 536 } 537 } 538 539 static void 540 _nvmf_tcp_qpair_destroy(void *_tqpair) 541 { 542 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 543 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 544 void *cb_arg = tqpair->fini_cb_arg; 545 int err = 0; 546 547 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 548 549 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 550 551 err = spdk_sock_close(&tqpair->sock); 552 assert(err == 0); 553 nvmf_tcp_cleanup_all_states(tqpair); 554 555 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 556 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 557 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 558 tqpair->resource_count); 559 err++; 560 } 561 562 if (err > 0) { 563 nvmf_tcp_dump_qpair_req_contents(tqpair); 564 } 565 566 /* The timeout poller might still be registered here if we close the qpair before host 567 * terminates the connection. 568 */ 569 spdk_poller_unregister(&tqpair->timeout_poller); 570 spdk_dma_free(tqpair->pdus); 571 free(tqpair->reqs); 572 spdk_free(tqpair->bufs); 573 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 574 free(tqpair); 575 576 if (cb_fn != NULL) { 577 cb_fn(cb_arg); 578 } 579 580 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 581 } 582 583 static void 584 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 585 { 586 /* Delay the destruction to make sure it isn't performed from the context of a sock 587 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 588 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 589 */ 590 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 591 } 592 593 static void 594 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 595 { 596 struct spdk_nvmf_tcp_transport *ttransport; 597 assert(w != NULL); 598 599 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 600 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 601 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 602 } 603 604 static void 605 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 606 { 607 if (entry == NULL) { 608 return; 609 } 610 611 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 612 spdk_keyring_put_key(entry->key); 613 free(entry); 614 } 615 616 static int 617 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 618 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 619 { 620 struct spdk_nvmf_tcp_transport *ttransport; 621 struct tcp_psk_entry *entry, *tmp; 622 623 assert(transport != NULL); 624 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 625 626 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 627 TAILQ_REMOVE(&ttransport->psks, entry, link); 628 nvmf_tcp_free_psk_entry(entry); 629 } 630 631 spdk_poller_unregister(&ttransport->accept_poller); 632 free(ttransport); 633 634 if (cb_fn) { 635 cb_fn(cb_arg); 636 } 637 return 0; 638 } 639 640 static int nvmf_tcp_accept(void *ctx); 641 642 static struct spdk_nvmf_transport * 643 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 644 { 645 struct spdk_nvmf_tcp_transport *ttransport; 646 uint32_t sge_count; 647 uint32_t min_shared_buffers; 648 649 ttransport = calloc(1, sizeof(*ttransport)); 650 if (!ttransport) { 651 return NULL; 652 } 653 654 TAILQ_INIT(&ttransport->ports); 655 TAILQ_INIT(&ttransport->poll_groups); 656 TAILQ_INIT(&ttransport->psks); 657 658 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 659 660 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 661 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 662 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 663 if (opts->transport_specific != NULL && 664 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 665 SPDK_COUNTOF(tcp_transport_opts_decoder), 666 &ttransport->tcp_opts)) { 667 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 668 free(ttransport); 669 return NULL; 670 } 671 672 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 673 674 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 675 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 676 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 677 " in_capsule_data_size=%d, max_aq_depth=%d\n" 678 " num_shared_buffers=%d, c2h_success=%d,\n" 679 " dif_insert_or_strip=%d, sock_priority=%d\n" 680 " abort_timeout_sec=%d, control_msg_num=%hu\n" 681 " ack_timeout=%d\n", 682 opts->max_queue_depth, 683 opts->max_io_size, 684 opts->max_qpairs_per_ctrlr - 1, 685 opts->io_unit_size, 686 opts->in_capsule_data_size, 687 opts->max_aq_depth, 688 opts->num_shared_buffers, 689 ttransport->tcp_opts.c2h_success, 690 opts->dif_insert_or_strip, 691 ttransport->tcp_opts.sock_priority, 692 opts->abort_timeout_sec, 693 ttransport->tcp_opts.control_msg_num, 694 opts->ack_timeout); 695 696 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 697 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 698 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 699 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 700 free(ttransport); 701 return NULL; 702 } 703 704 if (ttransport->tcp_opts.control_msg_num == 0 && 705 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 706 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 707 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 708 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 709 } 710 711 /* I/O unit size cannot be larger than max I/O size */ 712 if (opts->io_unit_size > opts->max_io_size) { 713 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", 714 opts->io_unit_size, opts->max_io_size); 715 opts->io_unit_size = opts->max_io_size; 716 } 717 718 /* In capsule data size cannot be larger than max I/O size */ 719 if (opts->in_capsule_data_size > opts->max_io_size) { 720 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 721 opts->io_unit_size, opts->max_io_size); 722 opts->in_capsule_data_size = opts->max_io_size; 723 } 724 725 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 726 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 727 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 728 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 729 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 730 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 731 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 732 } 733 734 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 735 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 736 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 737 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 738 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 739 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 740 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 741 } 742 743 sge_count = opts->max_io_size / opts->io_unit_size; 744 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 745 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 746 free(ttransport); 747 return NULL; 748 } 749 750 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 751 if (opts->buf_cache_size < UINT32_MAX) { 752 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 753 if (min_shared_buffers > opts->num_shared_buffers) { 754 SPDK_ERRLOG("There are not enough buffers to satisfy " 755 "per-poll group caches for each thread. (%" PRIu32 ") " 756 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 757 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 758 free(ttransport); 759 return NULL; 760 } 761 } 762 763 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 764 opts->acceptor_poll_rate); 765 if (!ttransport->accept_poller) { 766 free(ttransport); 767 return NULL; 768 } 769 770 return &ttransport->transport; 771 } 772 773 static int 774 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 775 { 776 unsigned long long ull; 777 char *end = NULL; 778 779 ull = strtoull(trsvcid, &end, 10); 780 if (end == NULL || end == trsvcid || *end != '\0') { 781 return -1; 782 } 783 784 /* Valid TCP/IP port numbers are in [1, 65535] */ 785 if (ull == 0 || ull > 65535) { 786 return -1; 787 } 788 789 return (int)ull; 790 } 791 792 /** 793 * Canonicalize a listen address trid. 794 */ 795 static int 796 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 797 const struct spdk_nvme_transport_id *trid) 798 { 799 int trsvcid_int; 800 801 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 802 if (trsvcid_int < 0) { 803 return -EINVAL; 804 } 805 806 memset(canon_trid, 0, sizeof(*canon_trid)); 807 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 808 canon_trid->adrfam = trid->adrfam; 809 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 810 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 811 812 return 0; 813 } 814 815 /** 816 * Find an existing listening port. 817 */ 818 static struct spdk_nvmf_tcp_port * 819 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 820 const struct spdk_nvme_transport_id *trid) 821 { 822 struct spdk_nvme_transport_id canon_trid; 823 struct spdk_nvmf_tcp_port *port; 824 825 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 826 return NULL; 827 } 828 829 TAILQ_FOREACH(port, &ttransport->ports, link) { 830 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 831 return port; 832 } 833 } 834 835 return NULL; 836 } 837 838 static int 839 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 840 void *get_key_ctx) 841 { 842 struct tcp_psk_entry *entry; 843 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 844 size_t psk_len; 845 int rc; 846 847 TAILQ_FOREACH(entry, &ttransport->psks, link) { 848 if (strcmp(pskid, entry->pskid) != 0) { 849 continue; 850 } 851 852 psk_len = entry->psk_size; 853 if ((size_t)out_len < psk_len) { 854 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 855 out_len, psk_len); 856 return -ENOBUFS; 857 } 858 859 /* Convert PSK to the TLS PSK format. */ 860 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 861 entry->tls_cipher_suite); 862 if (rc < 0) { 863 SPDK_ERRLOG("Could not generate TLS PSK\n"); 864 } 865 866 switch (entry->tls_cipher_suite) { 867 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 868 *cipher = "TLS_AES_128_GCM_SHA256"; 869 break; 870 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 871 *cipher = "TLS_AES_256_GCM_SHA384"; 872 break; 873 default: 874 *cipher = NULL; 875 return -ENOTSUP; 876 } 877 878 return rc; 879 } 880 881 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 882 883 return -EINVAL; 884 } 885 886 static int 887 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 888 struct spdk_nvmf_listen_opts *listen_opts) 889 { 890 struct spdk_nvmf_tcp_transport *ttransport; 891 struct spdk_nvmf_tcp_port *port; 892 int trsvcid_int; 893 uint8_t adrfam; 894 const char *sock_impl_name; 895 struct spdk_sock_impl_opts impl_opts; 896 size_t impl_opts_size = sizeof(impl_opts); 897 struct spdk_sock_opts opts; 898 899 if (!strlen(trid->trsvcid)) { 900 SPDK_ERRLOG("Service id is required\n"); 901 return -EINVAL; 902 } 903 904 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 905 906 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 907 if (trsvcid_int < 0) { 908 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 909 return -EINVAL; 910 } 911 912 port = calloc(1, sizeof(*port)); 913 if (!port) { 914 SPDK_ERRLOG("Port allocation failed\n"); 915 return -ENOMEM; 916 } 917 918 port->trid = trid; 919 920 sock_impl_name = NULL; 921 922 opts.opts_size = sizeof(opts); 923 spdk_sock_get_default_opts(&opts); 924 opts.priority = ttransport->tcp_opts.sock_priority; 925 opts.ack_timeout = transport->opts.ack_timeout; 926 if (listen_opts->secure_channel) { 927 if (!g_tls_log) { 928 SPDK_NOTICELOG("TLS support is considered experimental\n"); 929 g_tls_log = true; 930 } 931 sock_impl_name = "ssl"; 932 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 933 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 934 impl_opts.get_key = tcp_sock_get_key; 935 impl_opts.get_key_ctx = ttransport; 936 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 937 opts.impl_opts = &impl_opts; 938 opts.impl_opts_size = sizeof(impl_opts); 939 } 940 941 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 942 sock_impl_name, &opts); 943 if (port->listen_sock == NULL) { 944 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 945 trid->traddr, trsvcid_int, 946 spdk_strerror(errno), errno); 947 free(port); 948 return -errno; 949 } 950 951 if (spdk_sock_is_ipv4(port->listen_sock)) { 952 adrfam = SPDK_NVMF_ADRFAM_IPV4; 953 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 954 adrfam = SPDK_NVMF_ADRFAM_IPV6; 955 } else { 956 SPDK_ERRLOG("Unhandled socket type\n"); 957 adrfam = 0; 958 } 959 960 if (adrfam != trid->adrfam) { 961 SPDK_ERRLOG("Socket address family mismatch\n"); 962 spdk_sock_close(&port->listen_sock); 963 free(port); 964 return -EINVAL; 965 } 966 967 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 968 trid->traddr, trid->trsvcid); 969 970 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 971 return 0; 972 } 973 974 static void 975 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 976 const struct spdk_nvme_transport_id *trid) 977 { 978 struct spdk_nvmf_tcp_transport *ttransport; 979 struct spdk_nvmf_tcp_port *port; 980 981 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 982 983 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 984 trid->traddr, trid->trsvcid); 985 986 port = nvmf_tcp_find_port(ttransport, trid); 987 if (port) { 988 TAILQ_REMOVE(&ttransport->ports, port, link); 989 spdk_sock_close(&port->listen_sock); 990 free(port); 991 } 992 } 993 994 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 995 enum nvme_tcp_pdu_recv_state state); 996 997 static void 998 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 999 { 1000 tqpair->state = state; 1001 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1002 (uint64_t)tqpair->state); 1003 } 1004 1005 static void 1006 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1007 { 1008 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1009 1010 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1011 1012 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 1013 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 1014 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1015 spdk_poller_unregister(&tqpair->timeout_poller); 1016 1017 /* This will end up calling nvmf_tcp_close_qpair */ 1018 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1019 } 1020 } 1021 1022 static void 1023 _mgmt_pdu_write_done(void *_tqpair, int err) 1024 { 1025 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1026 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1027 1028 if (spdk_unlikely(err != 0)) { 1029 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1030 return; 1031 } 1032 1033 assert(pdu->cb_fn != NULL); 1034 pdu->cb_fn(pdu->cb_arg); 1035 } 1036 1037 static void 1038 _req_pdu_write_done(void *req, int err) 1039 { 1040 struct spdk_nvmf_tcp_req *tcp_req = req; 1041 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1042 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1043 1044 assert(tcp_req->pdu_in_use); 1045 tcp_req->pdu_in_use = false; 1046 1047 /* If the request is in a completed state, we're waiting for write completion to free it */ 1048 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1049 nvmf_tcp_request_free(tcp_req); 1050 return; 1051 } 1052 1053 if (spdk_unlikely(err != 0)) { 1054 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1055 return; 1056 } 1057 1058 assert(pdu->cb_fn != NULL); 1059 pdu->cb_fn(pdu->cb_arg); 1060 } 1061 1062 static void 1063 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1064 { 1065 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1066 } 1067 1068 static void 1069 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1070 { 1071 int rc; 1072 uint32_t mapped_length; 1073 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1074 1075 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1076 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1077 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1078 1079 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1080 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1081 /* Try to force the send immediately. */ 1082 rc = spdk_sock_flush(tqpair->sock); 1083 if (rc > 0 && (uint32_t)rc == mapped_length) { 1084 _pdu_write_done(pdu, 0); 1085 } else { 1086 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1087 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1088 "IC_RESP" : "TERM_REQ", rc, errno); 1089 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1090 } 1091 } 1092 } 1093 1094 static void 1095 data_crc32_accel_done(void *cb_arg, int status) 1096 { 1097 struct nvme_tcp_pdu *pdu = cb_arg; 1098 1099 if (spdk_unlikely(status)) { 1100 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1101 _pdu_write_done(pdu, status); 1102 return; 1103 } 1104 1105 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1106 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1107 1108 _tcp_write_pdu(pdu); 1109 } 1110 1111 static void 1112 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1113 { 1114 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1115 int rc = 0; 1116 1117 /* Data Digest */ 1118 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1119 /* Only support this limitated case for the first step */ 1120 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1121 && tqpair->group)) { 1122 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1123 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1124 if (spdk_likely(rc == 0)) { 1125 return; 1126 } 1127 } else { 1128 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1129 } 1130 data_crc32_accel_done(pdu, rc); 1131 } else { 1132 _tcp_write_pdu(pdu); 1133 } 1134 } 1135 1136 static void 1137 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1138 struct nvme_tcp_pdu *pdu, 1139 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1140 void *cb_arg) 1141 { 1142 int hlen; 1143 uint32_t crc32c; 1144 1145 assert(tqpair->pdu_in_progress != pdu); 1146 1147 hlen = pdu->hdr.common.hlen; 1148 pdu->cb_fn = cb_fn; 1149 pdu->cb_arg = cb_arg; 1150 1151 pdu->iov[0].iov_base = &pdu->hdr.raw; 1152 pdu->iov[0].iov_len = hlen; 1153 1154 /* Header Digest */ 1155 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1156 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1157 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1158 } 1159 1160 /* Data Digest */ 1161 pdu_data_crc32_compute(pdu); 1162 } 1163 1164 static void 1165 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1166 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1167 void *cb_arg) 1168 { 1169 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1170 1171 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1172 pdu->sock_req.cb_arg = tqpair; 1173 1174 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1175 } 1176 1177 static void 1178 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1179 struct spdk_nvmf_tcp_req *tcp_req, 1180 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1181 void *cb_arg) 1182 { 1183 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1184 1185 pdu->sock_req.cb_fn = _req_pdu_write_done; 1186 pdu->sock_req.cb_arg = tcp_req; 1187 1188 assert(!tcp_req->pdu_in_use); 1189 tcp_req->pdu_in_use = true; 1190 1191 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1192 } 1193 1194 static int 1195 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1196 { 1197 uint32_t i; 1198 struct spdk_nvmf_transport_opts *opts; 1199 uint32_t in_capsule_data_size; 1200 1201 opts = &tqpair->qpair.transport->opts; 1202 1203 in_capsule_data_size = opts->in_capsule_data_size; 1204 if (opts->dif_insert_or_strip) { 1205 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1206 } 1207 1208 tqpair->resource_count = opts->max_queue_depth; 1209 1210 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1211 if (!tqpair->reqs) { 1212 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1213 return -1; 1214 } 1215 1216 if (in_capsule_data_size) { 1217 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1218 NULL, SPDK_ENV_LCORE_ID_ANY, 1219 SPDK_MALLOC_DMA); 1220 if (!tqpair->bufs) { 1221 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1222 return -1; 1223 } 1224 } 1225 /* prepare memory space for receiving pdus and tcp_req */ 1226 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1227 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1228 NULL); 1229 if (!tqpair->pdus) { 1230 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1231 return -1; 1232 } 1233 1234 for (i = 0; i < tqpair->resource_count; i++) { 1235 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1236 1237 tcp_req->ttag = i + 1; 1238 tcp_req->req.qpair = &tqpair->qpair; 1239 1240 tcp_req->pdu = &tqpair->pdus[i]; 1241 tcp_req->pdu->qpair = tqpair; 1242 1243 /* Set up memory to receive commands */ 1244 if (tqpair->bufs) { 1245 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1246 } 1247 1248 /* Set the cmdn and rsp */ 1249 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1250 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1251 1252 tcp_req->req.stripped_data = NULL; 1253 1254 /* Initialize request state to FREE */ 1255 tcp_req->state = TCP_REQUEST_STATE_FREE; 1256 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1257 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1258 } 1259 1260 for (; i < 2 * tqpair->resource_count; i++) { 1261 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1262 1263 pdu->qpair = tqpair; 1264 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1265 } 1266 1267 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1268 tqpair->mgmt_pdu->qpair = tqpair; 1269 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1270 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1271 tqpair->tcp_pdu_working_count = 1; 1272 1273 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1274 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1275 1276 return 0; 1277 } 1278 1279 static int 1280 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1281 { 1282 struct spdk_nvmf_tcp_qpair *tqpair; 1283 1284 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1285 1286 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1287 1288 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1289 1290 /* Initialise request state queues of the qpair */ 1291 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1292 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1293 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1294 tqpair->qpair.queue_depth = 0; 1295 1296 tqpair->host_hdgst_enable = true; 1297 tqpair->host_ddgst_enable = true; 1298 1299 return 0; 1300 } 1301 1302 static int 1303 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1304 { 1305 char saddr[32], caddr[32]; 1306 uint16_t sport, cport; 1307 char owner[256]; 1308 int rc; 1309 1310 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1311 caddr, sizeof(caddr), &cport); 1312 if (rc != 0) { 1313 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1314 return rc; 1315 } 1316 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1317 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1318 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1319 1320 /* set low water mark */ 1321 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1322 if (rc != 0) { 1323 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1324 return rc; 1325 } 1326 1327 return 0; 1328 } 1329 1330 static void 1331 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1332 struct spdk_nvmf_tcp_port *port, 1333 struct spdk_sock *sock) 1334 { 1335 struct spdk_nvmf_tcp_qpair *tqpair; 1336 int rc; 1337 1338 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1339 port->trid->traddr, port->trid->trsvcid); 1340 1341 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1342 if (tqpair == NULL) { 1343 SPDK_ERRLOG("Could not allocate new connection.\n"); 1344 spdk_sock_close(&sock); 1345 return; 1346 } 1347 1348 tqpair->sock = sock; 1349 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1350 tqpair->port = port; 1351 tqpair->qpair.transport = transport; 1352 1353 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1354 sizeof(tqpair->target_addr), &tqpair->target_port, 1355 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1356 &tqpair->initiator_port); 1357 if (rc < 0) { 1358 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1359 nvmf_tcp_qpair_destroy(tqpair); 1360 return; 1361 } 1362 1363 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1364 } 1365 1366 static uint32_t 1367 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1368 { 1369 struct spdk_sock *sock; 1370 uint32_t count = 0; 1371 int i; 1372 1373 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1374 sock = spdk_sock_accept(port->listen_sock); 1375 if (sock == NULL) { 1376 break; 1377 } 1378 count++; 1379 nvmf_tcp_handle_connect(transport, port, sock); 1380 } 1381 1382 return count; 1383 } 1384 1385 static int 1386 nvmf_tcp_accept(void *ctx) 1387 { 1388 struct spdk_nvmf_transport *transport = ctx; 1389 struct spdk_nvmf_tcp_transport *ttransport; 1390 struct spdk_nvmf_tcp_port *port; 1391 uint32_t count = 0; 1392 1393 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1394 1395 TAILQ_FOREACH(port, &ttransport->ports, link) { 1396 count += nvmf_tcp_port_accept(transport, port); 1397 } 1398 1399 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1400 } 1401 1402 static void 1403 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1404 struct spdk_nvme_transport_id *trid, 1405 struct spdk_nvmf_discovery_log_page_entry *entry) 1406 { 1407 struct spdk_nvmf_tcp_port *port; 1408 struct spdk_nvmf_tcp_transport *ttransport; 1409 1410 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1411 entry->adrfam = trid->adrfam; 1412 1413 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1414 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1415 1416 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1417 port = nvmf_tcp_find_port(ttransport, trid); 1418 1419 assert(port != NULL); 1420 1421 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1422 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1423 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1424 } else { 1425 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1426 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1427 } 1428 } 1429 1430 static struct spdk_nvmf_tcp_control_msg_list * 1431 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1432 { 1433 struct spdk_nvmf_tcp_control_msg_list *list; 1434 struct spdk_nvmf_tcp_control_msg *msg; 1435 uint16_t i; 1436 1437 list = calloc(1, sizeof(*list)); 1438 if (!list) { 1439 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1440 return NULL; 1441 } 1442 1443 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1444 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1445 if (!list->msg_buf) { 1446 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1447 free(list); 1448 return NULL; 1449 } 1450 1451 STAILQ_INIT(&list->free_msgs); 1452 1453 for (i = 0; i < num_messages; i++) { 1454 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1455 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1456 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1457 } 1458 1459 return list; 1460 } 1461 1462 static void 1463 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1464 { 1465 if (!list) { 1466 return; 1467 } 1468 1469 spdk_free(list->msg_buf); 1470 free(list); 1471 } 1472 1473 static struct spdk_nvmf_transport_poll_group * 1474 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1475 struct spdk_nvmf_poll_group *group) 1476 { 1477 struct spdk_nvmf_tcp_transport *ttransport; 1478 struct spdk_nvmf_tcp_poll_group *tgroup; 1479 1480 if (spdk_interrupt_mode_is_enabled()) { 1481 SPDK_ERRLOG("TCP transport does not support interrupt mode\n"); 1482 return NULL; 1483 } 1484 1485 tgroup = calloc(1, sizeof(*tgroup)); 1486 if (!tgroup) { 1487 return NULL; 1488 } 1489 1490 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1491 if (!tgroup->sock_group) { 1492 goto cleanup; 1493 } 1494 1495 TAILQ_INIT(&tgroup->qpairs); 1496 TAILQ_INIT(&tgroup->await_req); 1497 1498 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1499 1500 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1501 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1502 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1503 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1504 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1505 if (!tgroup->control_msg_list) { 1506 goto cleanup; 1507 } 1508 } 1509 1510 tgroup->accel_channel = spdk_accel_get_io_channel(); 1511 if (spdk_unlikely(!tgroup->accel_channel)) { 1512 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1513 goto cleanup; 1514 } 1515 1516 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1517 if (ttransport->next_pg == NULL) { 1518 ttransport->next_pg = tgroup; 1519 } 1520 1521 return &tgroup->group; 1522 1523 cleanup: 1524 nvmf_tcp_poll_group_destroy(&tgroup->group); 1525 return NULL; 1526 } 1527 1528 static struct spdk_nvmf_transport_poll_group * 1529 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1530 { 1531 struct spdk_nvmf_tcp_transport *ttransport; 1532 struct spdk_nvmf_tcp_poll_group **pg; 1533 struct spdk_nvmf_tcp_qpair *tqpair; 1534 struct spdk_sock_group *group = NULL, *hint = NULL; 1535 int rc; 1536 1537 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1538 1539 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1540 return NULL; 1541 } 1542 1543 pg = &ttransport->next_pg; 1544 assert(*pg != NULL); 1545 hint = (*pg)->sock_group; 1546 1547 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1548 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1549 if (rc != 0) { 1550 return NULL; 1551 } else if (group != NULL) { 1552 /* Optimal poll group was found */ 1553 return spdk_sock_group_get_ctx(group); 1554 } 1555 1556 /* The hint was used for optimal poll group, advance next_pg. */ 1557 *pg = TAILQ_NEXT(*pg, link); 1558 if (*pg == NULL) { 1559 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1560 } 1561 1562 return spdk_sock_group_get_ctx(hint); 1563 } 1564 1565 static void 1566 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1567 { 1568 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1569 struct spdk_nvmf_tcp_transport *ttransport; 1570 1571 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1572 spdk_sock_group_close(&tgroup->sock_group); 1573 if (tgroup->control_msg_list) { 1574 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1575 } 1576 1577 if (tgroup->accel_channel) { 1578 spdk_put_io_channel(tgroup->accel_channel); 1579 } 1580 1581 if (tgroup->group.transport == NULL) { 1582 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1583 * calls this function directly in a failure path. */ 1584 free(tgroup); 1585 return; 1586 } 1587 1588 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1589 1590 next_tgroup = TAILQ_NEXT(tgroup, link); 1591 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1592 if (next_tgroup == NULL) { 1593 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1594 } 1595 if (ttransport->next_pg == tgroup) { 1596 ttransport->next_pg = next_tgroup; 1597 } 1598 1599 free(tgroup); 1600 } 1601 1602 static void 1603 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1604 enum nvme_tcp_pdu_recv_state state) 1605 { 1606 if (tqpair->recv_state == state) { 1607 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1608 tqpair, state); 1609 return; 1610 } 1611 1612 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1613 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1614 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1615 tqpair->tcp_pdu_working_count--; 1616 } 1617 } 1618 1619 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1620 assert(tqpair->tcp_pdu_working_count == 0); 1621 } 1622 1623 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1624 /* When leaving the await req state, move the qpair to the main list */ 1625 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1626 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1627 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1628 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1629 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1630 } 1631 1632 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1633 tqpair->recv_state = state; 1634 1635 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1636 (uint64_t)tqpair->recv_state); 1637 } 1638 1639 static int 1640 nvmf_tcp_qpair_handle_timeout(void *ctx) 1641 { 1642 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1643 1644 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1645 1646 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1647 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1648 1649 nvmf_tcp_qpair_disconnect(tqpair); 1650 return SPDK_POLLER_BUSY; 1651 } 1652 1653 static void 1654 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1655 { 1656 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1657 1658 if (!tqpair->timeout_poller) { 1659 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1660 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1661 } 1662 } 1663 1664 static void 1665 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1666 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1667 { 1668 struct nvme_tcp_pdu *rsp_pdu; 1669 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1670 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1671 uint32_t copy_len; 1672 1673 rsp_pdu = tqpair->mgmt_pdu; 1674 1675 c2h_term_req = &rsp_pdu->hdr.term_req; 1676 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1677 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1678 c2h_term_req->fes = fes; 1679 1680 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1681 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1682 DSET32(&c2h_term_req->fei, error_offset); 1683 } 1684 1685 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1686 1687 /* Copy the error info into the buffer */ 1688 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1689 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1690 1691 /* Contain the header of the wrong received pdu */ 1692 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1693 tqpair->wait_terminate = true; 1694 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1695 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1696 } 1697 1698 static void 1699 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1700 struct spdk_nvmf_tcp_qpair *tqpair, 1701 struct nvme_tcp_pdu *pdu) 1702 { 1703 struct spdk_nvmf_tcp_req *tcp_req; 1704 1705 assert(pdu->psh_valid_bytes == pdu->psh_len); 1706 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1707 1708 tcp_req = nvmf_tcp_req_get(tqpair); 1709 if (!tcp_req) { 1710 /* Directly return and make the allocation retry again. This can happen if we're 1711 * using asynchronous writes to send the response to the host or when releasing 1712 * zero-copy buffers after a response has been sent. In both cases, the host might 1713 * receive the response before we've finished processing the request and is free to 1714 * send another one. 1715 */ 1716 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1717 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1718 return; 1719 } 1720 1721 /* The host sent more commands than the maximum queue depth. */ 1722 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1723 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1724 return; 1725 } 1726 1727 pdu->req = tcp_req; 1728 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1729 nvmf_tcp_req_process(ttransport, tcp_req); 1730 } 1731 1732 static void 1733 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1734 struct spdk_nvmf_tcp_qpair *tqpair, 1735 struct nvme_tcp_pdu *pdu) 1736 { 1737 struct spdk_nvmf_tcp_req *tcp_req; 1738 struct spdk_nvme_tcp_cmd *capsule_cmd; 1739 uint32_t error_offset = 0; 1740 enum spdk_nvme_tcp_term_req_fes fes; 1741 struct spdk_nvme_cpl *rsp; 1742 1743 capsule_cmd = &pdu->hdr.capsule_cmd; 1744 tcp_req = pdu->req; 1745 assert(tcp_req != NULL); 1746 1747 /* Zero-copy requests don't support ICD */ 1748 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1749 1750 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1751 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1752 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1753 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1754 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1755 goto err; 1756 } 1757 1758 rsp = &tcp_req->req.rsp->nvme_cpl; 1759 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1760 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1761 } else { 1762 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1763 } 1764 1765 nvmf_tcp_req_process(ttransport, tcp_req); 1766 1767 return; 1768 err: 1769 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1770 } 1771 1772 static void 1773 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1774 struct spdk_nvmf_tcp_qpair *tqpair, 1775 struct nvme_tcp_pdu *pdu) 1776 { 1777 struct spdk_nvmf_tcp_req *tcp_req; 1778 uint32_t error_offset = 0; 1779 enum spdk_nvme_tcp_term_req_fes fes = 0; 1780 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1781 1782 h2c_data = &pdu->hdr.h2c_data; 1783 1784 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1785 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1786 1787 if (h2c_data->ttag > tqpair->resource_count) { 1788 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1789 tqpair->resource_count); 1790 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1791 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1792 goto err; 1793 } 1794 1795 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1796 1797 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1798 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1799 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1800 tcp_req->state); 1801 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1802 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1803 goto err; 1804 } 1805 1806 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1807 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1808 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1809 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1810 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1811 goto err; 1812 } 1813 1814 if (tcp_req->h2c_offset != h2c_data->datao) { 1815 SPDK_DEBUGLOG(nvmf_tcp, 1816 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1817 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1818 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1819 goto err; 1820 } 1821 1822 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1823 SPDK_DEBUGLOG(nvmf_tcp, 1824 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1825 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1826 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1827 goto err; 1828 } 1829 1830 pdu->req = tcp_req; 1831 1832 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1833 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1834 } 1835 1836 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1837 h2c_data->datao, h2c_data->datal); 1838 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1839 return; 1840 1841 err: 1842 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1843 } 1844 1845 static void 1846 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1847 struct spdk_nvmf_tcp_qpair *tqpair) 1848 { 1849 struct nvme_tcp_pdu *rsp_pdu; 1850 struct spdk_nvme_tcp_rsp *capsule_resp; 1851 1852 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1853 1854 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1855 assert(rsp_pdu != NULL); 1856 1857 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1858 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1859 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1860 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1861 if (tqpair->host_hdgst_enable) { 1862 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1863 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1864 } 1865 1866 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1867 } 1868 1869 static void 1870 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1871 { 1872 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1873 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1874 struct spdk_nvmf_tcp_qpair, qpair); 1875 1876 assert(tqpair != NULL); 1877 1878 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1879 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1880 tcp_req->req.length); 1881 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1882 return; 1883 } 1884 1885 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1886 nvmf_tcp_request_free(tcp_req); 1887 } else { 1888 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1889 } 1890 } 1891 1892 static void 1893 nvmf_tcp_r2t_complete(void *cb_arg) 1894 { 1895 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1896 struct spdk_nvmf_tcp_transport *ttransport; 1897 1898 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1899 struct spdk_nvmf_tcp_transport, transport); 1900 1901 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1902 1903 if (tcp_req->h2c_offset == tcp_req->req.length) { 1904 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1905 nvmf_tcp_req_process(ttransport, tcp_req); 1906 } 1907 } 1908 1909 static void 1910 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1911 struct spdk_nvmf_tcp_req *tcp_req) 1912 { 1913 struct nvme_tcp_pdu *rsp_pdu; 1914 struct spdk_nvme_tcp_r2t_hdr *r2t; 1915 1916 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1917 assert(rsp_pdu != NULL); 1918 1919 r2t = &rsp_pdu->hdr.r2t; 1920 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1921 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1922 1923 if (tqpair->host_hdgst_enable) { 1924 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1925 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1926 } 1927 1928 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1929 r2t->ttag = tcp_req->ttag; 1930 r2t->r2to = tcp_req->h2c_offset; 1931 r2t->r2tl = tcp_req->req.length; 1932 1933 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1934 1935 SPDK_DEBUGLOG(nvmf_tcp, 1936 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1937 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1938 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1939 } 1940 1941 static void 1942 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1943 struct spdk_nvmf_tcp_qpair *tqpair, 1944 struct nvme_tcp_pdu *pdu) 1945 { 1946 struct spdk_nvmf_tcp_req *tcp_req; 1947 struct spdk_nvme_cpl *rsp; 1948 1949 tcp_req = pdu->req; 1950 assert(tcp_req != NULL); 1951 1952 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1953 1954 tcp_req->h2c_offset += pdu->data_len; 1955 1956 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1957 * acknowledged before moving on. */ 1958 if (tcp_req->h2c_offset == tcp_req->req.length && 1959 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1960 /* After receiving all the h2c data, we need to check whether there is 1961 * transient transport error */ 1962 rsp = &tcp_req->req.rsp->nvme_cpl; 1963 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1964 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1965 } else { 1966 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1967 } 1968 nvmf_tcp_req_process(ttransport, tcp_req); 1969 } 1970 } 1971 1972 static void 1973 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1974 { 1975 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1976 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1977 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1978 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1979 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1980 DGET32(h2c_term_req->fei)); 1981 } 1982 } 1983 1984 static void 1985 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1986 struct nvme_tcp_pdu *pdu) 1987 { 1988 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1989 uint32_t error_offset = 0; 1990 enum spdk_nvme_tcp_term_req_fes fes; 1991 1992 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1993 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1994 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1995 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1996 goto end; 1997 } 1998 1999 /* set the data buffer */ 2000 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2001 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2002 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2003 return; 2004 end: 2005 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2006 } 2007 2008 static void 2009 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2010 struct nvme_tcp_pdu *pdu) 2011 { 2012 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2013 2014 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2015 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2016 } 2017 2018 static void 2019 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2020 { 2021 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2022 struct spdk_nvmf_tcp_transport, transport); 2023 2024 switch (pdu->hdr.common.pdu_type) { 2025 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2026 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2027 break; 2028 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2029 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2030 break; 2031 2032 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2033 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2034 break; 2035 2036 default: 2037 /* The code should not go to here */ 2038 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2039 break; 2040 } 2041 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2042 tqpair->tcp_pdu_working_count--; 2043 } 2044 2045 static inline void 2046 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2047 { 2048 treq->req.rsp->nvme_cpl.status.sct = sct; 2049 treq->req.rsp->nvme_cpl.status.sc = sc; 2050 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2051 } 2052 2053 static void 2054 data_crc32_calc_done(void *cb_arg, int status) 2055 { 2056 struct nvme_tcp_pdu *pdu = cb_arg; 2057 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2058 2059 /* async crc32 calculation is failed and use direct calculation to check */ 2060 if (spdk_unlikely(status)) { 2061 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2062 tqpair, pdu); 2063 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2064 } 2065 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2066 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2067 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2068 assert(pdu->req != NULL); 2069 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2070 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2071 } 2072 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2073 } 2074 2075 static void 2076 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2077 { 2078 int rc = 0; 2079 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2080 tqpair->pdu_in_progress = NULL; 2081 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2082 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2083 /* check data digest if need */ 2084 if (pdu->ddgst_enable) { 2085 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2086 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2087 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2088 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2089 if (spdk_likely(rc == 0)) { 2090 return; 2091 } 2092 } else { 2093 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2094 } 2095 data_crc32_calc_done(pdu, rc); 2096 } else { 2097 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2098 } 2099 } 2100 2101 static void 2102 nvmf_tcp_send_icresp_complete(void *cb_arg) 2103 { 2104 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2105 2106 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2107 } 2108 2109 static void 2110 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2111 struct spdk_nvmf_tcp_qpair *tqpair, 2112 struct nvme_tcp_pdu *pdu) 2113 { 2114 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2115 struct nvme_tcp_pdu *rsp_pdu; 2116 struct spdk_nvme_tcp_ic_resp *ic_resp; 2117 uint32_t error_offset = 0; 2118 enum spdk_nvme_tcp_term_req_fes fes; 2119 2120 /* Only PFV 0 is defined currently */ 2121 if (ic_req->pfv != 0) { 2122 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2123 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2124 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2125 goto end; 2126 } 2127 2128 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2129 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2130 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2131 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2132 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2133 goto end; 2134 } 2135 2136 /* MAXR2T is 0's based */ 2137 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2138 2139 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2140 if (!tqpair->host_hdgst_enable) { 2141 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2142 } 2143 2144 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2145 if (!tqpair->host_ddgst_enable) { 2146 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2147 } 2148 2149 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2150 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2151 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2152 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2153 tqpair, 2154 tqpair->recv_buf_size); 2155 /* Not fatal. */ 2156 } 2157 2158 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2159 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2160 2161 rsp_pdu = tqpair->mgmt_pdu; 2162 2163 ic_resp = &rsp_pdu->hdr.ic_resp; 2164 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2165 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2166 ic_resp->pfv = 0; 2167 ic_resp->cpda = tqpair->cpda; 2168 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2169 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2170 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2171 2172 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2173 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2174 2175 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2176 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2177 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2178 return; 2179 end: 2180 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2181 } 2182 2183 static void 2184 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2185 struct spdk_nvmf_tcp_transport *ttransport) 2186 { 2187 struct nvme_tcp_pdu *pdu; 2188 int rc; 2189 uint32_t crc32c, error_offset = 0; 2190 enum spdk_nvme_tcp_term_req_fes fes; 2191 2192 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2193 pdu = tqpair->pdu_in_progress; 2194 2195 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2196 pdu->hdr.common.pdu_type); 2197 /* check header digest if needed */ 2198 if (pdu->has_hdgst) { 2199 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2200 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2201 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2202 if (rc == 0) { 2203 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2204 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2205 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2206 return; 2207 2208 } 2209 } 2210 2211 switch (pdu->hdr.common.pdu_type) { 2212 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2213 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2214 break; 2215 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2216 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2217 break; 2218 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2219 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2220 break; 2221 2222 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2223 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2224 break; 2225 2226 default: 2227 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2228 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2229 error_offset = 1; 2230 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2231 break; 2232 } 2233 } 2234 2235 static void 2236 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2237 { 2238 struct nvme_tcp_pdu *pdu; 2239 uint32_t error_offset = 0; 2240 enum spdk_nvme_tcp_term_req_fes fes; 2241 uint8_t expected_hlen, pdo; 2242 bool plen_error = false, pdo_error = false; 2243 2244 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2245 pdu = tqpair->pdu_in_progress; 2246 assert(pdu); 2247 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2248 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2249 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2250 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2251 goto err; 2252 } 2253 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2254 if (pdu->hdr.common.plen != expected_hlen) { 2255 plen_error = true; 2256 } 2257 } else { 2258 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2259 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2260 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2261 goto err; 2262 } 2263 2264 switch (pdu->hdr.common.pdu_type) { 2265 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2266 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2267 pdo = pdu->hdr.common.pdo; 2268 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2269 pdo_error = true; 2270 break; 2271 } 2272 2273 if (pdu->hdr.common.plen < expected_hlen) { 2274 plen_error = true; 2275 } 2276 break; 2277 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2278 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2279 pdo = pdu->hdr.common.pdo; 2280 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2281 pdo_error = true; 2282 break; 2283 } 2284 if (pdu->hdr.common.plen < expected_hlen) { 2285 plen_error = true; 2286 } 2287 break; 2288 2289 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2290 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2291 if ((pdu->hdr.common.plen <= expected_hlen) || 2292 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2293 plen_error = true; 2294 } 2295 break; 2296 2297 default: 2298 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2299 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2300 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2301 goto err; 2302 } 2303 } 2304 2305 if (pdu->hdr.common.hlen != expected_hlen) { 2306 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2307 pdu->hdr.common.pdu_type, 2308 expected_hlen, pdu->hdr.common.hlen, tqpair); 2309 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2310 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2311 goto err; 2312 } else if (pdo_error) { 2313 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2314 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2315 } else if (plen_error) { 2316 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2317 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2318 goto err; 2319 } else { 2320 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2321 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2322 return; 2323 } 2324 err: 2325 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2326 } 2327 2328 static int 2329 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2330 { 2331 int rc = 0; 2332 struct nvme_tcp_pdu *pdu; 2333 enum nvme_tcp_pdu_recv_state prev_state; 2334 uint32_t data_len; 2335 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2336 struct spdk_nvmf_tcp_transport, transport); 2337 2338 /* The loop here is to allow for several back-to-back state changes. */ 2339 do { 2340 prev_state = tqpair->recv_state; 2341 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2342 2343 pdu = tqpair->pdu_in_progress; 2344 assert(pdu != NULL || 2345 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2346 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2347 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2348 2349 switch (tqpair->recv_state) { 2350 /* Wait for the common header */ 2351 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2352 if (!pdu) { 2353 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2354 if (spdk_unlikely(!pdu)) { 2355 return NVME_TCP_PDU_IN_PROGRESS; 2356 } 2357 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2358 tqpair->pdu_in_progress = pdu; 2359 tqpair->tcp_pdu_working_count++; 2360 } 2361 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2362 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2363 /* FALLTHROUGH */ 2364 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2365 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2366 return rc; 2367 } 2368 2369 rc = nvme_tcp_read_data(tqpair->sock, 2370 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2371 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2372 if (rc < 0) { 2373 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2374 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2375 break; 2376 } else if (rc > 0) { 2377 pdu->ch_valid_bytes += rc; 2378 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2379 } 2380 2381 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2382 return NVME_TCP_PDU_IN_PROGRESS; 2383 } 2384 2385 /* The command header of this PDU has now been read from the socket. */ 2386 nvmf_tcp_pdu_ch_handle(tqpair); 2387 break; 2388 /* Wait for the pdu specific header */ 2389 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2390 rc = nvme_tcp_read_data(tqpair->sock, 2391 pdu->psh_len - pdu->psh_valid_bytes, 2392 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2393 if (rc < 0) { 2394 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2395 break; 2396 } else if (rc > 0) { 2397 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2398 pdu->psh_valid_bytes += rc; 2399 } 2400 2401 if (pdu->psh_valid_bytes < pdu->psh_len) { 2402 return NVME_TCP_PDU_IN_PROGRESS; 2403 } 2404 2405 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2406 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2407 break; 2408 /* Wait for the req slot */ 2409 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2410 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2411 break; 2412 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2413 /* check whether the data is valid, if not we just return */ 2414 if (!pdu->data_len) { 2415 return NVME_TCP_PDU_IN_PROGRESS; 2416 } 2417 2418 data_len = pdu->data_len; 2419 /* data digest */ 2420 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2421 tqpair->host_ddgst_enable)) { 2422 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2423 pdu->ddgst_enable = true; 2424 } 2425 2426 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2427 if (rc < 0) { 2428 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2429 break; 2430 } 2431 pdu->rw_offset += rc; 2432 2433 if (pdu->rw_offset < data_len) { 2434 return NVME_TCP_PDU_IN_PROGRESS; 2435 } 2436 2437 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2438 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2439 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2440 pdu->dif_ctx) != 0) { 2441 SPDK_ERRLOG("DIF generate failed\n"); 2442 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2443 break; 2444 } 2445 2446 /* All of this PDU has now been read from the socket. */ 2447 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2448 break; 2449 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2450 if (tqpair->tcp_pdu_working_count != 0) { 2451 return NVME_TCP_PDU_IN_PROGRESS; 2452 } 2453 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2454 break; 2455 case NVME_TCP_PDU_RECV_STATE_ERROR: 2456 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2457 return NVME_TCP_PDU_IN_PROGRESS; 2458 } 2459 return NVME_TCP_PDU_FATAL; 2460 default: 2461 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2462 abort(); 2463 break; 2464 } 2465 } while (tqpair->recv_state != prev_state); 2466 2467 return rc; 2468 } 2469 2470 static inline void * 2471 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2472 { 2473 struct spdk_nvmf_tcp_control_msg *msg; 2474 2475 assert(list); 2476 2477 msg = STAILQ_FIRST(&list->free_msgs); 2478 if (!msg) { 2479 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2480 return NULL; 2481 } 2482 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2483 return msg; 2484 } 2485 2486 static inline void 2487 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2488 { 2489 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2490 2491 assert(list); 2492 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2493 } 2494 2495 static int 2496 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2497 struct spdk_nvmf_transport *transport, 2498 struct spdk_nvmf_transport_poll_group *group) 2499 { 2500 struct spdk_nvmf_request *req = &tcp_req->req; 2501 struct spdk_nvme_cmd *cmd; 2502 struct spdk_nvme_sgl_descriptor *sgl; 2503 struct spdk_nvmf_tcp_poll_group *tgroup; 2504 enum spdk_nvme_tcp_term_req_fes fes; 2505 struct nvme_tcp_pdu *pdu; 2506 struct spdk_nvmf_tcp_qpair *tqpair; 2507 uint32_t length, error_offset = 0; 2508 2509 cmd = &req->cmd->nvme_cmd; 2510 sgl = &cmd->dptr.sgl1; 2511 2512 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2513 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2514 /* get request length from sgl */ 2515 length = sgl->unkeyed.length; 2516 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2517 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2518 length, transport->opts.max_io_size); 2519 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2520 goto fatal_err; 2521 } 2522 2523 /* fill request length and populate iovs */ 2524 req->length = length; 2525 2526 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2527 2528 if (spdk_unlikely(req->dif_enabled)) { 2529 req->dif.orig_length = length; 2530 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2531 req->dif.elba_length = length; 2532 } 2533 2534 if (nvmf_ctrlr_use_zcopy(req)) { 2535 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2536 req->data_from_pool = false; 2537 return 0; 2538 } 2539 2540 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2541 /* No available buffers. Queue this request up. */ 2542 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2543 tcp_req); 2544 return 0; 2545 } 2546 2547 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2548 tcp_req, req->iovcnt, req->iov[0].iov_base); 2549 2550 return 0; 2551 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2552 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2553 uint64_t offset = sgl->address; 2554 uint32_t max_len = transport->opts.in_capsule_data_size; 2555 2556 assert(tcp_req->has_in_capsule_data); 2557 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2558 tqpair = tcp_req->pdu->qpair; 2559 /* receiving pdu is not same with the pdu in tcp_req */ 2560 pdu = tqpair->pdu_in_progress; 2561 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2562 if (tqpair->host_ddgst_enable) { 2563 length -= SPDK_NVME_TCP_DIGEST_LEN; 2564 } 2565 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2566 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2567 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2568 length, sgl->unkeyed.length); 2569 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2570 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2571 goto fatal_err; 2572 } 2573 2574 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2575 offset, length); 2576 2577 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2578 if (spdk_unlikely(offset != 0)) { 2579 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2580 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2581 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2582 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2583 goto fatal_err; 2584 } 2585 2586 if (spdk_unlikely(length > max_len)) { 2587 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2588 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2589 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2590 2591 /* Get a buffer from dedicated list */ 2592 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2593 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2594 assert(tgroup->control_msg_list); 2595 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2596 if (!req->iov[0].iov_base) { 2597 /* No available buffers. Queue this request up. */ 2598 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2599 return 0; 2600 } 2601 } else { 2602 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2603 length, max_len); 2604 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2605 goto fatal_err; 2606 } 2607 } else { 2608 req->iov[0].iov_base = tcp_req->buf; 2609 } 2610 2611 req->length = length; 2612 req->data_from_pool = false; 2613 2614 if (spdk_unlikely(req->dif_enabled)) { 2615 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2616 req->dif.elba_length = length; 2617 } 2618 2619 req->iov[0].iov_len = length; 2620 req->iovcnt = 1; 2621 2622 return 0; 2623 } 2624 /* If we want to handle the problem here, then we can't skip the following data segment. 2625 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2626 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2627 sgl->generic.type, sgl->generic.subtype); 2628 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2629 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2630 fatal_err: 2631 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2632 return -1; 2633 } 2634 2635 static inline enum spdk_nvme_media_error_status_code 2636 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2637 enum spdk_nvme_media_error_status_code result; 2638 2639 switch (err_type) 2640 { 2641 case SPDK_DIF_REFTAG_ERROR: 2642 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2643 break; 2644 case SPDK_DIF_APPTAG_ERROR: 2645 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2646 break; 2647 case SPDK_DIF_GUARD_ERROR: 2648 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2649 break; 2650 default: 2651 SPDK_UNREACHABLE(); 2652 break; 2653 } 2654 2655 return result; 2656 } 2657 2658 static void 2659 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2660 struct spdk_nvmf_tcp_req *tcp_req) 2661 { 2662 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2663 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2664 struct nvme_tcp_pdu *rsp_pdu; 2665 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2666 uint32_t plen, pdo, alignment; 2667 int rc; 2668 2669 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2670 2671 rsp_pdu = tcp_req->pdu; 2672 assert(rsp_pdu != NULL); 2673 2674 c2h_data = &rsp_pdu->hdr.c2h_data; 2675 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2676 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2677 2678 if (tqpair->host_hdgst_enable) { 2679 plen += SPDK_NVME_TCP_DIGEST_LEN; 2680 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2681 } 2682 2683 /* set the psh */ 2684 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2685 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2686 c2h_data->datao = tcp_req->pdu->rw_offset; 2687 2688 /* set the padding */ 2689 rsp_pdu->padding_len = 0; 2690 pdo = plen; 2691 if (tqpair->cpda) { 2692 alignment = (tqpair->cpda + 1) << 2; 2693 if (plen % alignment != 0) { 2694 pdo = (plen + alignment) / alignment * alignment; 2695 rsp_pdu->padding_len = pdo - plen; 2696 plen = pdo; 2697 } 2698 } 2699 2700 c2h_data->common.pdo = pdo; 2701 plen += c2h_data->datal; 2702 if (tqpair->host_ddgst_enable) { 2703 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2704 plen += SPDK_NVME_TCP_DIGEST_LEN; 2705 } 2706 2707 c2h_data->common.plen = plen; 2708 2709 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2710 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2711 } 2712 2713 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2714 c2h_data->datao, c2h_data->datal); 2715 2716 2717 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2718 /* Need to send the capsule response if response is not all 0 */ 2719 if (ttransport->tcp_opts.c2h_success && 2720 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2721 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2722 } 2723 2724 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2725 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2726 struct spdk_dif_error err_blk = {}; 2727 uint32_t mapped_length = 0; 2728 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2729 uint32_t ddgst_len = 0; 2730 2731 if (tqpair->host_ddgst_enable) { 2732 /* Data digest consumes additional iov entry */ 2733 available_iovs--; 2734 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2735 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2736 c2h_data->common.plen -= ddgst_len; 2737 } 2738 /* Temp call to estimate if data can be described by limited number of iovs. 2739 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2740 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2741 false, &mapped_length); 2742 2743 if (mapped_length != c2h_data->common.plen) { 2744 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2745 SPDK_DEBUGLOG(nvmf_tcp, 2746 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2747 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2748 c2h_data->common.plen = mapped_length; 2749 2750 /* Rebuild pdu->data_iov since data length is changed */ 2751 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2752 c2h_data->datal); 2753 2754 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2755 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2756 } 2757 2758 c2h_data->common.plen += ddgst_len; 2759 2760 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2761 2762 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2763 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2764 if (rc != 0) { 2765 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2766 err_blk.err_type, err_blk.err_offset); 2767 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2768 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2769 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2770 return; 2771 } 2772 } 2773 2774 rsp_pdu->rw_offset += c2h_data->datal; 2775 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2776 } 2777 2778 static void 2779 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2780 struct spdk_nvmf_tcp_req *tcp_req) 2781 { 2782 nvmf_tcp_req_pdu_init(tcp_req); 2783 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2784 } 2785 2786 static int 2787 request_transfer_out(struct spdk_nvmf_request *req) 2788 { 2789 struct spdk_nvmf_tcp_req *tcp_req; 2790 struct spdk_nvmf_qpair *qpair; 2791 struct spdk_nvmf_tcp_qpair *tqpair; 2792 struct spdk_nvme_cpl *rsp; 2793 2794 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2795 2796 qpair = req->qpair; 2797 rsp = &req->rsp->nvme_cpl; 2798 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2799 2800 /* Advance our sq_head pointer */ 2801 if (qpair->sq_head == qpair->sq_head_max) { 2802 qpair->sq_head = 0; 2803 } else { 2804 qpair->sq_head++; 2805 } 2806 rsp->sqhd = qpair->sq_head; 2807 2808 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2809 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2810 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2811 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2812 } else { 2813 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2814 } 2815 2816 return 0; 2817 } 2818 2819 static void 2820 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2821 struct spdk_nvmf_tcp_qpair *tqpair, 2822 struct spdk_nvmf_tcp_req *tcp_req) 2823 { 2824 enum spdk_nvme_cmd_fuse last, next; 2825 2826 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2827 next = tcp_req->cmd.fuse; 2828 2829 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2830 2831 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2832 return; 2833 } 2834 2835 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2836 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2837 /* This is a valid pair of fused commands. Point them at each other 2838 * so they can be submitted consecutively once ready to be executed. 2839 */ 2840 tqpair->fused_first->fused_pair = tcp_req; 2841 tcp_req->fused_pair = tqpair->fused_first; 2842 tqpair->fused_first = NULL; 2843 return; 2844 } else { 2845 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2846 tqpair->fused_first->fused_failed = true; 2847 2848 /* 2849 * If the last req is in READY_TO_EXECUTE state, then call 2850 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2851 */ 2852 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2853 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2854 } 2855 2856 tqpair->fused_first = NULL; 2857 } 2858 } 2859 2860 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2861 /* Set tqpair->fused_first here so that we know to check that the next request 2862 * is a SECOND (and to fail this one if it isn't). 2863 */ 2864 tqpair->fused_first = tcp_req; 2865 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2866 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2867 tcp_req->fused_failed = true; 2868 } 2869 } 2870 2871 static bool 2872 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2873 struct spdk_nvmf_tcp_req *tcp_req) 2874 { 2875 struct spdk_nvmf_tcp_qpair *tqpair; 2876 uint32_t plen; 2877 struct nvme_tcp_pdu *pdu; 2878 enum spdk_nvmf_tcp_req_state prev_state; 2879 bool progress = false; 2880 struct spdk_nvmf_transport *transport = &ttransport->transport; 2881 struct spdk_nvmf_transport_poll_group *group; 2882 struct spdk_nvmf_tcp_poll_group *tgroup; 2883 2884 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2885 group = &tqpair->group->group; 2886 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2887 2888 /* If the qpair is not active, we need to abort the outstanding requests. */ 2889 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 2890 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2891 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2892 } 2893 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2894 } 2895 2896 /* The loop here is to allow for several back-to-back state changes. */ 2897 do { 2898 prev_state = tcp_req->state; 2899 2900 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2901 tqpair); 2902 2903 switch (tcp_req->state) { 2904 case TCP_REQUEST_STATE_FREE: 2905 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2906 * to escape this state. */ 2907 break; 2908 case TCP_REQUEST_STATE_NEW: 2909 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 2910 tqpair->qpair.queue_depth); 2911 2912 /* copy the cmd from the receive pdu */ 2913 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2914 2915 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2916 tcp_req->req.dif_enabled = true; 2917 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2918 } 2919 2920 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2921 2922 /* The next state transition depends on the data transfer needs of this request. */ 2923 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2924 2925 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2926 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2927 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2928 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2929 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2930 break; 2931 } 2932 2933 /* If no data to transfer, ready to execute. */ 2934 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2935 /* Reset the tqpair receiving pdu state */ 2936 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2937 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2938 break; 2939 } 2940 2941 pdu = tqpair->pdu_in_progress; 2942 plen = pdu->hdr.common.hlen; 2943 if (tqpair->host_hdgst_enable) { 2944 plen += SPDK_NVME_TCP_DIGEST_LEN; 2945 } 2946 if (pdu->hdr.common.plen != plen) { 2947 tcp_req->has_in_capsule_data = true; 2948 } else { 2949 /* Data is transmitted by C2H PDUs */ 2950 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2951 } 2952 2953 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2954 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2955 break; 2956 case TCP_REQUEST_STATE_NEED_BUFFER: 2957 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 2958 (uintptr_t)tcp_req); 2959 2960 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2961 2962 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2963 SPDK_DEBUGLOG(nvmf_tcp, 2964 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2965 tcp_req, tqpair); 2966 /* This request needs to wait in line to obtain a buffer */ 2967 break; 2968 } 2969 2970 /* Try to get a data buffer */ 2971 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 2972 break; 2973 } 2974 2975 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2976 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2977 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2978 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2979 tcp_req->req.length = tcp_req->req.dif.elba_length; 2980 } 2981 2982 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2983 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2984 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2985 break; 2986 } 2987 2988 if (tcp_req->req.iovcnt < 1) { 2989 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2990 tcp_req, tqpair); 2991 /* No buffers available. */ 2992 break; 2993 } 2994 2995 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2996 2997 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2998 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2999 if (tcp_req->req.data_from_pool) { 3000 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3001 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3002 } else { 3003 struct nvme_tcp_pdu *pdu; 3004 3005 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3006 3007 pdu = tqpair->pdu_in_progress; 3008 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3009 tqpair); 3010 /* No need to send r2t, contained in the capsuled data */ 3011 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3012 0, tcp_req->req.length); 3013 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3014 } 3015 break; 3016 } 3017 3018 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3019 break; 3020 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3021 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3022 (uintptr_t)tcp_req); 3023 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3024 * to escape this state. */ 3025 break; 3026 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3027 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3028 (uintptr_t)tcp_req); 3029 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3030 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3031 tcp_req, tqpair); 3032 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3033 break; 3034 } 3035 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3036 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3037 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3038 } else { 3039 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3040 } 3041 break; 3042 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3043 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3044 (uintptr_t)tcp_req); 3045 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3046 break; 3047 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3048 3049 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3050 0, (uintptr_t)tcp_req); 3051 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3052 * to escape this state. */ 3053 break; 3054 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3055 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3056 (uintptr_t)tcp_req); 3057 3058 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3059 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3060 tcp_req->req.length = tcp_req->req.dif.elba_length; 3061 } 3062 3063 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3064 if (tcp_req->fused_failed) { 3065 /* This request failed FUSED semantics. Fail it immediately, without 3066 * even sending it to the target layer. 3067 */ 3068 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3069 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3070 break; 3071 } 3072 3073 if (tcp_req->fused_pair == NULL || 3074 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3075 /* This request is ready to execute, but either we don't know yet if it's 3076 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3077 * or the other request of this fused pair isn't ready to execute. So 3078 * break here and this request will get processed later either when the 3079 * other request is ready or we find that this request isn't valid. 3080 */ 3081 break; 3082 } 3083 } 3084 3085 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3086 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3087 /* If we get to this point, and this request is a fused command, we know that 3088 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3089 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3090 * request, and the other request of the fused pair, in the correct order. 3091 * Also clear the ->fused_pair pointers on both requests, since after this point 3092 * we no longer need to maintain the relationship between these two requests. 3093 */ 3094 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3095 assert(tcp_req->fused_pair != NULL); 3096 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3097 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3098 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3099 tcp_req->fused_pair->fused_pair = NULL; 3100 tcp_req->fused_pair = NULL; 3101 } 3102 spdk_nvmf_request_exec(&tcp_req->req); 3103 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3104 assert(tcp_req->fused_pair != NULL); 3105 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3106 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3107 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3108 tcp_req->fused_pair->fused_pair = NULL; 3109 tcp_req->fused_pair = NULL; 3110 } 3111 } else { 3112 /* For zero-copy, only requests with data coming from host to the 3113 * controller can end up here. */ 3114 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3115 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3116 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3117 } 3118 3119 break; 3120 case TCP_REQUEST_STATE_EXECUTING: 3121 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3122 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3123 * to escape this state. */ 3124 break; 3125 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3126 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3127 (uintptr_t)tcp_req); 3128 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3129 * to escape this state. */ 3130 break; 3131 case TCP_REQUEST_STATE_EXECUTED: 3132 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3133 3134 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3135 tcp_req->req.length = tcp_req->req.dif.orig_length; 3136 } 3137 3138 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3139 break; 3140 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3141 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3142 (uintptr_t)tcp_req); 3143 if (request_transfer_out(&tcp_req->req) != 0) { 3144 assert(0); /* No good way to handle this currently */ 3145 } 3146 break; 3147 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3148 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3149 0, (uintptr_t)tcp_req); 3150 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3151 * to escape this state. */ 3152 break; 3153 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3154 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3155 (uintptr_t)tcp_req); 3156 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3157 * to escape this state. */ 3158 break; 3159 case TCP_REQUEST_STATE_COMPLETED: 3160 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3161 tqpair->qpair.queue_depth); 3162 /* If there's an outstanding PDU sent to the host, the request is completed 3163 * due to the qpair being disconnected. We must delay the completion until 3164 * that write is done to avoid freeing the request twice. */ 3165 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3166 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3167 "write on req=%p\n", tcp_req); 3168 /* This can only happen for zcopy requests */ 3169 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3170 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3171 break; 3172 } 3173 3174 if (tcp_req->req.data_from_pool) { 3175 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3176 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3177 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3178 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3179 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3180 assert(tgroup->control_msg_list); 3181 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3182 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3183 tcp_req->req.iov[0].iov_base); 3184 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3185 /* If the request has an unreleased zcopy bdev_io, it's either a 3186 * read, a failed write, or the qpair is being disconnected */ 3187 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3188 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3189 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3190 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3191 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3192 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3193 break; 3194 } 3195 tcp_req->req.length = 0; 3196 tcp_req->req.iovcnt = 0; 3197 tcp_req->fused_failed = false; 3198 if (tcp_req->fused_pair) { 3199 /* This req was part of a valid fused pair, but failed before it got to 3200 * READ_TO_EXECUTE state. This means we need to fail the other request 3201 * in the pair, because it is no longer part of a valid pair. If the pair 3202 * already reached READY_TO_EXECUTE state, we need to kick it. 3203 */ 3204 tcp_req->fused_pair->fused_failed = true; 3205 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3206 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3207 } 3208 tcp_req->fused_pair = NULL; 3209 } 3210 3211 nvmf_tcp_req_put(tqpair, tcp_req); 3212 break; 3213 case TCP_REQUEST_NUM_STATES: 3214 default: 3215 assert(0); 3216 break; 3217 } 3218 3219 if (tcp_req->state != prev_state) { 3220 progress = true; 3221 } 3222 } while (tcp_req->state != prev_state); 3223 3224 return progress; 3225 } 3226 3227 static void 3228 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3229 { 3230 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3231 int rc; 3232 3233 assert(tqpair != NULL); 3234 rc = nvmf_tcp_sock_process(tqpair); 3235 3236 /* If there was a new socket error, disconnect */ 3237 if (rc < 0) { 3238 nvmf_tcp_qpair_disconnect(tqpair); 3239 } 3240 } 3241 3242 static int 3243 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3244 struct spdk_nvmf_qpair *qpair) 3245 { 3246 struct spdk_nvmf_tcp_poll_group *tgroup; 3247 struct spdk_nvmf_tcp_qpair *tqpair; 3248 int rc; 3249 3250 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3251 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3252 3253 rc = nvmf_tcp_qpair_sock_init(tqpair); 3254 if (rc != 0) { 3255 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3256 return -1; 3257 } 3258 3259 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3260 if (rc < 0) { 3261 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3262 return -1; 3263 } 3264 3265 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3266 if (rc < 0) { 3267 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3268 return -1; 3269 } 3270 3271 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3272 nvmf_tcp_sock_cb, tqpair); 3273 if (rc != 0) { 3274 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3275 spdk_strerror(errno), errno); 3276 return -1; 3277 } 3278 3279 tqpair->group = tgroup; 3280 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3281 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3282 3283 return 0; 3284 } 3285 3286 static int 3287 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3288 struct spdk_nvmf_qpair *qpair) 3289 { 3290 struct spdk_nvmf_tcp_poll_group *tgroup; 3291 struct spdk_nvmf_tcp_qpair *tqpair; 3292 int rc; 3293 3294 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3295 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3296 3297 assert(tqpair->group == tgroup); 3298 3299 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3300 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3301 /* Change the state to move the qpair from the await_req list to the main list 3302 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3303 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3304 } 3305 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3306 3307 /* Try to force out any pending writes */ 3308 spdk_sock_flush(tqpair->sock); 3309 3310 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3311 if (rc != 0) { 3312 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3313 spdk_strerror(errno), errno); 3314 } 3315 3316 return rc; 3317 } 3318 3319 static int 3320 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3321 { 3322 struct spdk_nvmf_tcp_transport *ttransport; 3323 struct spdk_nvmf_tcp_req *tcp_req; 3324 3325 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3326 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3327 3328 switch (tcp_req->state) { 3329 case TCP_REQUEST_STATE_EXECUTING: 3330 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3331 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3332 break; 3333 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3334 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3335 break; 3336 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3337 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3338 break; 3339 default: 3340 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3341 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3342 assert(0 && "Unexpected request state"); 3343 break; 3344 } 3345 3346 nvmf_tcp_req_process(ttransport, tcp_req); 3347 3348 return 0; 3349 } 3350 3351 static void 3352 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3353 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3354 { 3355 struct spdk_nvmf_tcp_qpair *tqpair; 3356 3357 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3358 3359 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3360 3361 assert(tqpair->fini_cb_fn == NULL); 3362 tqpair->fini_cb_fn = cb_fn; 3363 tqpair->fini_cb_arg = cb_arg; 3364 3365 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3366 nvmf_tcp_qpair_destroy(tqpair); 3367 } 3368 3369 static int 3370 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3371 { 3372 struct spdk_nvmf_tcp_poll_group *tgroup; 3373 int num_events, rc = 0, rc2; 3374 struct spdk_nvmf_request *req, *req_tmp; 3375 struct spdk_nvmf_tcp_req *tcp_req; 3376 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3377 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3378 struct spdk_nvmf_tcp_transport, transport); 3379 3380 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3381 3382 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3383 return 0; 3384 } 3385 3386 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3387 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3388 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3389 break; 3390 } 3391 } 3392 3393 num_events = spdk_sock_group_poll(tgroup->sock_group); 3394 if (spdk_unlikely(num_events < 0)) { 3395 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3396 } 3397 3398 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3399 rc2 = nvmf_tcp_sock_process(tqpair); 3400 3401 /* If there was a new socket error, disconnect */ 3402 if (spdk_unlikely(rc2 < 0)) { 3403 nvmf_tcp_qpair_disconnect(tqpair); 3404 if (rc == 0) { 3405 rc = rc2; 3406 } 3407 } 3408 } 3409 3410 return rc == 0 ? num_events : rc; 3411 } 3412 3413 static int 3414 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3415 struct spdk_nvme_transport_id *trid, bool peer) 3416 { 3417 struct spdk_nvmf_tcp_qpair *tqpair; 3418 uint16_t port; 3419 3420 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3421 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3422 3423 if (peer) { 3424 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3425 port = tqpair->initiator_port; 3426 } else { 3427 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3428 port = tqpair->target_port; 3429 } 3430 3431 if (spdk_sock_is_ipv4(tqpair->sock)) { 3432 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3433 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3434 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3435 } else { 3436 return -1; 3437 } 3438 3439 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3440 return 0; 3441 } 3442 3443 static int 3444 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3445 struct spdk_nvme_transport_id *trid) 3446 { 3447 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3448 } 3449 3450 static int 3451 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3452 struct spdk_nvme_transport_id *trid) 3453 { 3454 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3455 } 3456 3457 static int 3458 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3459 struct spdk_nvme_transport_id *trid) 3460 { 3461 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3462 } 3463 3464 static void 3465 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3466 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3467 { 3468 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3469 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3470 3471 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3472 } 3473 3474 static int 3475 _nvmf_tcp_qpair_abort_request(void *ctx) 3476 { 3477 struct spdk_nvmf_request *req = ctx; 3478 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3479 struct spdk_nvmf_tcp_req, req); 3480 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3481 struct spdk_nvmf_tcp_qpair, qpair); 3482 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3483 struct spdk_nvmf_tcp_transport, transport); 3484 int rc; 3485 3486 spdk_poller_unregister(&req->poller); 3487 3488 switch (tcp_req_to_abort->state) { 3489 case TCP_REQUEST_STATE_EXECUTING: 3490 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3491 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3492 rc = nvmf_ctrlr_abort_request(req); 3493 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3494 return SPDK_POLLER_BUSY; 3495 } 3496 break; 3497 3498 case TCP_REQUEST_STATE_NEED_BUFFER: 3499 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3500 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3501 3502 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3503 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3504 break; 3505 3506 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3507 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3508 if (spdk_get_ticks() < req->timeout_tsc) { 3509 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3510 return SPDK_POLLER_BUSY; 3511 } 3512 break; 3513 3514 default: 3515 /* Requests in other states are either un-abortable (e.g. 3516 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3517 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3518 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3519 * abort request with the bit0 of dword0 set (command not aborted). 3520 */ 3521 break; 3522 } 3523 3524 spdk_nvmf_request_complete(req); 3525 return SPDK_POLLER_BUSY; 3526 } 3527 3528 static void 3529 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3530 struct spdk_nvmf_request *req) 3531 { 3532 struct spdk_nvmf_tcp_qpair *tqpair; 3533 struct spdk_nvmf_tcp_transport *ttransport; 3534 struct spdk_nvmf_transport *transport; 3535 uint16_t cid; 3536 uint32_t i; 3537 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3538 3539 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3540 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3541 transport = &ttransport->transport; 3542 3543 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3544 3545 for (i = 0; i < tqpair->resource_count; i++) { 3546 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3547 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3548 tcp_req_to_abort = &tqpair->reqs[i]; 3549 break; 3550 } 3551 } 3552 3553 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3554 3555 if (tcp_req_to_abort == NULL) { 3556 spdk_nvmf_request_complete(req); 3557 return; 3558 } 3559 3560 req->req_to_abort = &tcp_req_to_abort->req; 3561 req->timeout_tsc = spdk_get_ticks() + 3562 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3563 req->poller = NULL; 3564 3565 _nvmf_tcp_qpair_abort_request(req); 3566 } 3567 3568 struct tcp_subsystem_add_host_opts { 3569 char *psk; 3570 }; 3571 3572 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3573 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3574 }; 3575 3576 static int 3577 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3578 { 3579 FILE *psk_file; 3580 struct stat statbuf; 3581 int rc; 3582 3583 if (stat(fname, &statbuf) != 0) { 3584 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3585 return -EACCES; 3586 } 3587 3588 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3589 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3590 return -EPERM; 3591 } 3592 if ((size_t)statbuf.st_size > bufsz) { 3593 SPDK_ERRLOG("Invalid PSK: too long\n"); 3594 return -EINVAL; 3595 } 3596 psk_file = fopen(fname, "r"); 3597 if (psk_file == NULL) { 3598 SPDK_ERRLOG("Could not open PSK file\n"); 3599 return -EINVAL; 3600 } 3601 3602 rc = fread(buf, 1, statbuf.st_size, psk_file); 3603 if (rc != statbuf.st_size) { 3604 SPDK_ERRLOG("Failed to read PSK\n"); 3605 fclose(psk_file); 3606 return -EINVAL; 3607 } 3608 3609 fclose(psk_file); 3610 return 0; 3611 } 3612 3613 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3614 3615 static int 3616 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3617 const struct spdk_nvmf_subsystem *subsystem, 3618 const char *hostnqn, 3619 const struct spdk_json_val *transport_specific) 3620 { 3621 struct tcp_subsystem_add_host_opts opts; 3622 struct spdk_nvmf_tcp_transport *ttransport; 3623 struct tcp_psk_entry *tmp, *entry = NULL; 3624 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3625 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3626 uint8_t tls_cipher_suite; 3627 int rc = 0; 3628 uint8_t psk_retained_hash; 3629 uint64_t psk_configured_size; 3630 3631 if (transport_specific == NULL) { 3632 return 0; 3633 } 3634 3635 assert(transport != NULL); 3636 assert(subsystem != NULL); 3637 3638 memset(&opts, 0, sizeof(opts)); 3639 3640 /* Decode PSK (either name of a key or file path) */ 3641 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3642 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3643 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3644 return -EINVAL; 3645 } 3646 3647 if (opts.psk == NULL) { 3648 return 0; 3649 } 3650 3651 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3652 if (entry == NULL) { 3653 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3654 rc = -ENOMEM; 3655 goto end; 3656 } 3657 3658 entry->key = spdk_keyring_get_key(opts.psk); 3659 if (entry->key != NULL) { 3660 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3661 if (rc < 0) { 3662 SPDK_ERRLOG("Failed to retreive PSK '%s'\n", opts.psk); 3663 rc = -EINVAL; 3664 goto end; 3665 } 3666 } else { 3667 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3668 SPDK_ERRLOG("PSK path too long\n"); 3669 rc = -EINVAL; 3670 goto end; 3671 } 3672 3673 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3674 if (rc) { 3675 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3676 goto end; 3677 } 3678 3679 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3680 } 3681 3682 /* Parse PSK interchange to get length of base64 encoded data. 3683 * This is then used to decide which cipher suite should be used 3684 * to generate PSK identity and TLS PSK later on. */ 3685 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3686 &psk_configured_size, &psk_retained_hash); 3687 if (rc < 0) { 3688 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3689 goto end; 3690 } 3691 3692 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3693 * This check also ensures that psk_configured_size is smaller than 3694 * psk_retained buffer size. */ 3695 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3696 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3697 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3698 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3699 } else { 3700 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3701 rc = -EINVAL; 3702 goto end; 3703 } 3704 3705 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3706 /* Generate PSK identity. */ 3707 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3708 subsystem->subnqn, tls_cipher_suite); 3709 if (rc) { 3710 rc = -EINVAL; 3711 goto end; 3712 } 3713 /* Check if PSK identity entry already exists. */ 3714 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3715 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3716 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3717 rc = -EEXIST; 3718 goto end; 3719 } 3720 } 3721 3722 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3723 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3724 rc = -EINVAL; 3725 goto end; 3726 } 3727 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3728 SPDK_ERRLOG("Could not write subnqn string!\n"); 3729 rc = -EINVAL; 3730 goto end; 3731 } 3732 3733 entry->tls_cipher_suite = tls_cipher_suite; 3734 3735 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3736 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3737 /* Psk configured is either 32 or 48 bytes long. */ 3738 memcpy(entry->psk, psk_configured, psk_configured_size); 3739 entry->psk_size = psk_configured_size; 3740 } else { 3741 /* Derive retained PSK. */ 3742 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3743 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3744 if (rc < 0) { 3745 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3746 goto end; 3747 } 3748 entry->psk_size = rc; 3749 } 3750 3751 if (entry->key == NULL) { 3752 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3753 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3754 SPDK_ERRLOG("Could not save PSK path!\n"); 3755 rc = -ENAMETOOLONG; 3756 goto end; 3757 } 3758 } 3759 3760 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3761 rc = 0; 3762 3763 end: 3764 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3765 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3766 3767 free(opts.psk); 3768 if (rc != 0) { 3769 nvmf_tcp_free_psk_entry(entry); 3770 } 3771 3772 return rc; 3773 } 3774 3775 static void 3776 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3777 const struct spdk_nvmf_subsystem *subsystem, 3778 const char *hostnqn) 3779 { 3780 struct spdk_nvmf_tcp_transport *ttransport; 3781 struct tcp_psk_entry *entry, *tmp; 3782 3783 assert(transport != NULL); 3784 assert(subsystem != NULL); 3785 3786 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3787 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3788 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3789 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3790 TAILQ_REMOVE(&ttransport->psks, entry, link); 3791 nvmf_tcp_free_psk_entry(entry); 3792 break; 3793 } 3794 } 3795 } 3796 3797 static void 3798 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3799 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3800 struct spdk_json_write_ctx *w) 3801 { 3802 struct spdk_nvmf_tcp_transport *ttransport; 3803 struct tcp_psk_entry *entry; 3804 3805 assert(transport != NULL); 3806 assert(subsystem != NULL); 3807 3808 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3809 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3810 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3811 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3812 spdk_json_write_named_string(w, "psk", entry->key ? 3813 spdk_key_get_name(entry->key) : entry->psk_path); 3814 break; 3815 } 3816 } 3817 } 3818 3819 static void 3820 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3821 { 3822 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3823 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3824 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3825 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3826 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3827 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3828 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3829 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3830 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3831 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3832 opts->transport_specific = NULL; 3833 } 3834 3835 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3836 .name = "TCP", 3837 .type = SPDK_NVME_TRANSPORT_TCP, 3838 .opts_init = nvmf_tcp_opts_init, 3839 .create = nvmf_tcp_create, 3840 .dump_opts = nvmf_tcp_dump_opts, 3841 .destroy = nvmf_tcp_destroy, 3842 3843 .listen = nvmf_tcp_listen, 3844 .stop_listen = nvmf_tcp_stop_listen, 3845 3846 .listener_discover = nvmf_tcp_discover, 3847 3848 .poll_group_create = nvmf_tcp_poll_group_create, 3849 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3850 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3851 .poll_group_add = nvmf_tcp_poll_group_add, 3852 .poll_group_remove = nvmf_tcp_poll_group_remove, 3853 .poll_group_poll = nvmf_tcp_poll_group_poll, 3854 3855 .req_free = nvmf_tcp_req_free, 3856 .req_complete = nvmf_tcp_req_complete, 3857 3858 .qpair_fini = nvmf_tcp_close_qpair, 3859 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3860 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3861 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3862 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3863 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3864 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3865 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3866 }; 3867 3868 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3869 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3870