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 tgroup = calloc(1, sizeof(*tgroup)); 1481 if (!tgroup) { 1482 return NULL; 1483 } 1484 1485 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1486 if (!tgroup->sock_group) { 1487 goto cleanup; 1488 } 1489 1490 TAILQ_INIT(&tgroup->qpairs); 1491 TAILQ_INIT(&tgroup->await_req); 1492 1493 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1494 1495 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1496 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1497 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1498 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1499 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1500 if (!tgroup->control_msg_list) { 1501 goto cleanup; 1502 } 1503 } 1504 1505 tgroup->accel_channel = spdk_accel_get_io_channel(); 1506 if (spdk_unlikely(!tgroup->accel_channel)) { 1507 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1508 goto cleanup; 1509 } 1510 1511 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1512 if (ttransport->next_pg == NULL) { 1513 ttransport->next_pg = tgroup; 1514 } 1515 1516 return &tgroup->group; 1517 1518 cleanup: 1519 nvmf_tcp_poll_group_destroy(&tgroup->group); 1520 return NULL; 1521 } 1522 1523 static struct spdk_nvmf_transport_poll_group * 1524 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1525 { 1526 struct spdk_nvmf_tcp_transport *ttransport; 1527 struct spdk_nvmf_tcp_poll_group **pg; 1528 struct spdk_nvmf_tcp_qpair *tqpair; 1529 struct spdk_sock_group *group = NULL, *hint = NULL; 1530 int rc; 1531 1532 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1533 1534 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1535 return NULL; 1536 } 1537 1538 pg = &ttransport->next_pg; 1539 assert(*pg != NULL); 1540 hint = (*pg)->sock_group; 1541 1542 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1543 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1544 if (rc != 0) { 1545 return NULL; 1546 } else if (group != NULL) { 1547 /* Optimal poll group was found */ 1548 return spdk_sock_group_get_ctx(group); 1549 } 1550 1551 /* The hint was used for optimal poll group, advance next_pg. */ 1552 *pg = TAILQ_NEXT(*pg, link); 1553 if (*pg == NULL) { 1554 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1555 } 1556 1557 return spdk_sock_group_get_ctx(hint); 1558 } 1559 1560 static void 1561 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1562 { 1563 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1564 struct spdk_nvmf_tcp_transport *ttransport; 1565 1566 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1567 spdk_sock_group_close(&tgroup->sock_group); 1568 if (tgroup->control_msg_list) { 1569 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1570 } 1571 1572 if (tgroup->accel_channel) { 1573 spdk_put_io_channel(tgroup->accel_channel); 1574 } 1575 1576 if (tgroup->group.transport == NULL) { 1577 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1578 * calls this function directly in a failure path. */ 1579 free(tgroup); 1580 return; 1581 } 1582 1583 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1584 1585 next_tgroup = TAILQ_NEXT(tgroup, link); 1586 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1587 if (next_tgroup == NULL) { 1588 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1589 } 1590 if (ttransport->next_pg == tgroup) { 1591 ttransport->next_pg = next_tgroup; 1592 } 1593 1594 free(tgroup); 1595 } 1596 1597 static void 1598 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1599 enum nvme_tcp_pdu_recv_state state) 1600 { 1601 if (tqpair->recv_state == state) { 1602 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1603 tqpair, state); 1604 return; 1605 } 1606 1607 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1608 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1609 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1610 tqpair->tcp_pdu_working_count--; 1611 } 1612 } 1613 1614 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1615 assert(tqpair->tcp_pdu_working_count == 0); 1616 } 1617 1618 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1619 /* When leaving the await req state, move the qpair to the main list */ 1620 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1621 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1622 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1623 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1624 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1625 } 1626 1627 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1628 tqpair->recv_state = state; 1629 1630 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1631 (uint64_t)tqpair->recv_state); 1632 } 1633 1634 static int 1635 nvmf_tcp_qpair_handle_timeout(void *ctx) 1636 { 1637 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1638 1639 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1640 1641 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1642 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1643 1644 nvmf_tcp_qpair_disconnect(tqpair); 1645 return SPDK_POLLER_BUSY; 1646 } 1647 1648 static void 1649 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1650 { 1651 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1652 1653 if (!tqpair->timeout_poller) { 1654 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1655 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1656 } 1657 } 1658 1659 static void 1660 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1661 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1662 { 1663 struct nvme_tcp_pdu *rsp_pdu; 1664 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1665 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1666 uint32_t copy_len; 1667 1668 rsp_pdu = tqpair->mgmt_pdu; 1669 1670 c2h_term_req = &rsp_pdu->hdr.term_req; 1671 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1672 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1673 c2h_term_req->fes = fes; 1674 1675 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1676 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1677 DSET32(&c2h_term_req->fei, error_offset); 1678 } 1679 1680 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1681 1682 /* Copy the error info into the buffer */ 1683 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1684 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1685 1686 /* Contain the header of the wrong received pdu */ 1687 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1688 tqpair->wait_terminate = true; 1689 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1690 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1691 } 1692 1693 static void 1694 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1695 struct spdk_nvmf_tcp_qpair *tqpair, 1696 struct nvme_tcp_pdu *pdu) 1697 { 1698 struct spdk_nvmf_tcp_req *tcp_req; 1699 1700 assert(pdu->psh_valid_bytes == pdu->psh_len); 1701 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1702 1703 tcp_req = nvmf_tcp_req_get(tqpair); 1704 if (!tcp_req) { 1705 /* Directly return and make the allocation retry again. This can happen if we're 1706 * using asynchronous writes to send the response to the host or when releasing 1707 * zero-copy buffers after a response has been sent. In both cases, the host might 1708 * receive the response before we've finished processing the request and is free to 1709 * send another one. 1710 */ 1711 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1712 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1713 return; 1714 } 1715 1716 /* The host sent more commands than the maximum queue depth. */ 1717 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1718 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1719 return; 1720 } 1721 1722 pdu->req = tcp_req; 1723 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1724 nvmf_tcp_req_process(ttransport, tcp_req); 1725 } 1726 1727 static void 1728 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1729 struct spdk_nvmf_tcp_qpair *tqpair, 1730 struct nvme_tcp_pdu *pdu) 1731 { 1732 struct spdk_nvmf_tcp_req *tcp_req; 1733 struct spdk_nvme_tcp_cmd *capsule_cmd; 1734 uint32_t error_offset = 0; 1735 enum spdk_nvme_tcp_term_req_fes fes; 1736 struct spdk_nvme_cpl *rsp; 1737 1738 capsule_cmd = &pdu->hdr.capsule_cmd; 1739 tcp_req = pdu->req; 1740 assert(tcp_req != NULL); 1741 1742 /* Zero-copy requests don't support ICD */ 1743 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1744 1745 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1746 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1747 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1748 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1749 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1750 goto err; 1751 } 1752 1753 rsp = &tcp_req->req.rsp->nvme_cpl; 1754 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1755 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1756 } else { 1757 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1758 } 1759 1760 nvmf_tcp_req_process(ttransport, tcp_req); 1761 1762 return; 1763 err: 1764 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1765 } 1766 1767 static void 1768 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1769 struct spdk_nvmf_tcp_qpair *tqpair, 1770 struct nvme_tcp_pdu *pdu) 1771 { 1772 struct spdk_nvmf_tcp_req *tcp_req; 1773 uint32_t error_offset = 0; 1774 enum spdk_nvme_tcp_term_req_fes fes = 0; 1775 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1776 1777 h2c_data = &pdu->hdr.h2c_data; 1778 1779 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1780 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1781 1782 if (h2c_data->ttag > tqpair->resource_count) { 1783 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1784 tqpair->resource_count); 1785 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1786 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1787 goto err; 1788 } 1789 1790 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1791 1792 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1793 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1794 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1795 tcp_req->state); 1796 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1797 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1798 goto err; 1799 } 1800 1801 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1802 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1803 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1804 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1805 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1806 goto err; 1807 } 1808 1809 if (tcp_req->h2c_offset != h2c_data->datao) { 1810 SPDK_DEBUGLOG(nvmf_tcp, 1811 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1812 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1813 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1814 goto err; 1815 } 1816 1817 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1818 SPDK_DEBUGLOG(nvmf_tcp, 1819 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1820 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1821 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1822 goto err; 1823 } 1824 1825 pdu->req = tcp_req; 1826 1827 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1828 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1829 } 1830 1831 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1832 h2c_data->datao, h2c_data->datal); 1833 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1834 return; 1835 1836 err: 1837 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1838 } 1839 1840 static void 1841 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1842 struct spdk_nvmf_tcp_qpair *tqpair) 1843 { 1844 struct nvme_tcp_pdu *rsp_pdu; 1845 struct spdk_nvme_tcp_rsp *capsule_resp; 1846 1847 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1848 1849 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1850 assert(rsp_pdu != NULL); 1851 1852 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1853 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1854 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1855 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1856 if (tqpair->host_hdgst_enable) { 1857 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1858 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1859 } 1860 1861 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1862 } 1863 1864 static void 1865 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1866 { 1867 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1868 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1869 struct spdk_nvmf_tcp_qpair, qpair); 1870 1871 assert(tqpair != NULL); 1872 1873 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1874 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1875 tcp_req->req.length); 1876 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1877 return; 1878 } 1879 1880 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1881 nvmf_tcp_request_free(tcp_req); 1882 } else { 1883 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1884 } 1885 } 1886 1887 static void 1888 nvmf_tcp_r2t_complete(void *cb_arg) 1889 { 1890 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1891 struct spdk_nvmf_tcp_transport *ttransport; 1892 1893 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1894 struct spdk_nvmf_tcp_transport, transport); 1895 1896 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1897 1898 if (tcp_req->h2c_offset == tcp_req->req.length) { 1899 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1900 nvmf_tcp_req_process(ttransport, tcp_req); 1901 } 1902 } 1903 1904 static void 1905 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1906 struct spdk_nvmf_tcp_req *tcp_req) 1907 { 1908 struct nvme_tcp_pdu *rsp_pdu; 1909 struct spdk_nvme_tcp_r2t_hdr *r2t; 1910 1911 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1912 assert(rsp_pdu != NULL); 1913 1914 r2t = &rsp_pdu->hdr.r2t; 1915 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1916 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1917 1918 if (tqpair->host_hdgst_enable) { 1919 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1920 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1921 } 1922 1923 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1924 r2t->ttag = tcp_req->ttag; 1925 r2t->r2to = tcp_req->h2c_offset; 1926 r2t->r2tl = tcp_req->req.length; 1927 1928 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1929 1930 SPDK_DEBUGLOG(nvmf_tcp, 1931 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1932 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1933 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1934 } 1935 1936 static void 1937 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1938 struct spdk_nvmf_tcp_qpair *tqpair, 1939 struct nvme_tcp_pdu *pdu) 1940 { 1941 struct spdk_nvmf_tcp_req *tcp_req; 1942 struct spdk_nvme_cpl *rsp; 1943 1944 tcp_req = pdu->req; 1945 assert(tcp_req != NULL); 1946 1947 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1948 1949 tcp_req->h2c_offset += pdu->data_len; 1950 1951 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1952 * acknowledged before moving on. */ 1953 if (tcp_req->h2c_offset == tcp_req->req.length && 1954 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1955 /* After receiving all the h2c data, we need to check whether there is 1956 * transient transport error */ 1957 rsp = &tcp_req->req.rsp->nvme_cpl; 1958 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1959 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1960 } else { 1961 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1962 } 1963 nvmf_tcp_req_process(ttransport, tcp_req); 1964 } 1965 } 1966 1967 static void 1968 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1969 { 1970 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1971 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1972 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1973 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1974 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1975 DGET32(h2c_term_req->fei)); 1976 } 1977 } 1978 1979 static void 1980 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1981 struct nvme_tcp_pdu *pdu) 1982 { 1983 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1984 uint32_t error_offset = 0; 1985 enum spdk_nvme_tcp_term_req_fes fes; 1986 1987 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1988 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1989 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1990 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1991 goto end; 1992 } 1993 1994 /* set the data buffer */ 1995 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1996 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1997 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1998 return; 1999 end: 2000 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2001 } 2002 2003 static void 2004 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2005 struct nvme_tcp_pdu *pdu) 2006 { 2007 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2008 2009 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2010 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2011 } 2012 2013 static void 2014 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2015 { 2016 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2017 struct spdk_nvmf_tcp_transport, transport); 2018 2019 switch (pdu->hdr.common.pdu_type) { 2020 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2021 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2022 break; 2023 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2024 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2025 break; 2026 2027 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2028 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2029 break; 2030 2031 default: 2032 /* The code should not go to here */ 2033 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2034 break; 2035 } 2036 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2037 tqpair->tcp_pdu_working_count--; 2038 } 2039 2040 static inline void 2041 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2042 { 2043 treq->req.rsp->nvme_cpl.status.sct = sct; 2044 treq->req.rsp->nvme_cpl.status.sc = sc; 2045 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2046 } 2047 2048 static void 2049 data_crc32_calc_done(void *cb_arg, int status) 2050 { 2051 struct nvme_tcp_pdu *pdu = cb_arg; 2052 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2053 2054 /* async crc32 calculation is failed and use direct calculation to check */ 2055 if (spdk_unlikely(status)) { 2056 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2057 tqpair, pdu); 2058 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2059 } 2060 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2061 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2062 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2063 assert(pdu->req != NULL); 2064 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2065 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2066 } 2067 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2068 } 2069 2070 static void 2071 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2072 { 2073 int rc = 0; 2074 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2075 tqpair->pdu_in_progress = NULL; 2076 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2077 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2078 /* check data digest if need */ 2079 if (pdu->ddgst_enable) { 2080 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2081 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2082 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2083 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2084 if (spdk_likely(rc == 0)) { 2085 return; 2086 } 2087 } else { 2088 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2089 } 2090 data_crc32_calc_done(pdu, rc); 2091 } else { 2092 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2093 } 2094 } 2095 2096 static void 2097 nvmf_tcp_send_icresp_complete(void *cb_arg) 2098 { 2099 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2100 2101 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2102 } 2103 2104 static void 2105 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2106 struct spdk_nvmf_tcp_qpair *tqpair, 2107 struct nvme_tcp_pdu *pdu) 2108 { 2109 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2110 struct nvme_tcp_pdu *rsp_pdu; 2111 struct spdk_nvme_tcp_ic_resp *ic_resp; 2112 uint32_t error_offset = 0; 2113 enum spdk_nvme_tcp_term_req_fes fes; 2114 2115 /* Only PFV 0 is defined currently */ 2116 if (ic_req->pfv != 0) { 2117 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2118 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2119 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2120 goto end; 2121 } 2122 2123 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2124 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2125 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2126 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2127 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2128 goto end; 2129 } 2130 2131 /* MAXR2T is 0's based */ 2132 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2133 2134 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2135 if (!tqpair->host_hdgst_enable) { 2136 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2137 } 2138 2139 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2140 if (!tqpair->host_ddgst_enable) { 2141 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2142 } 2143 2144 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2145 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2146 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2147 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2148 tqpair, 2149 tqpair->recv_buf_size); 2150 /* Not fatal. */ 2151 } 2152 2153 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2154 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2155 2156 rsp_pdu = tqpair->mgmt_pdu; 2157 2158 ic_resp = &rsp_pdu->hdr.ic_resp; 2159 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2160 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2161 ic_resp->pfv = 0; 2162 ic_resp->cpda = tqpair->cpda; 2163 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2164 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2165 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2166 2167 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2168 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2169 2170 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2171 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2172 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2173 return; 2174 end: 2175 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2176 } 2177 2178 static void 2179 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2180 struct spdk_nvmf_tcp_transport *ttransport) 2181 { 2182 struct nvme_tcp_pdu *pdu; 2183 int rc; 2184 uint32_t crc32c, error_offset = 0; 2185 enum spdk_nvme_tcp_term_req_fes fes; 2186 2187 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2188 pdu = tqpair->pdu_in_progress; 2189 2190 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2191 pdu->hdr.common.pdu_type); 2192 /* check header digest if needed */ 2193 if (pdu->has_hdgst) { 2194 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2195 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2196 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2197 if (rc == 0) { 2198 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2199 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2200 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2201 return; 2202 2203 } 2204 } 2205 2206 switch (pdu->hdr.common.pdu_type) { 2207 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2208 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2209 break; 2210 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2211 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2212 break; 2213 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2214 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2215 break; 2216 2217 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2218 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2219 break; 2220 2221 default: 2222 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2223 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2224 error_offset = 1; 2225 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2226 break; 2227 } 2228 } 2229 2230 static void 2231 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2232 { 2233 struct nvme_tcp_pdu *pdu; 2234 uint32_t error_offset = 0; 2235 enum spdk_nvme_tcp_term_req_fes fes; 2236 uint8_t expected_hlen, pdo; 2237 bool plen_error = false, pdo_error = false; 2238 2239 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2240 pdu = tqpair->pdu_in_progress; 2241 assert(pdu); 2242 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2243 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2244 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2245 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2246 goto err; 2247 } 2248 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2249 if (pdu->hdr.common.plen != expected_hlen) { 2250 plen_error = true; 2251 } 2252 } else { 2253 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2254 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2255 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2256 goto err; 2257 } 2258 2259 switch (pdu->hdr.common.pdu_type) { 2260 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2261 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2262 pdo = pdu->hdr.common.pdo; 2263 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2264 pdo_error = true; 2265 break; 2266 } 2267 2268 if (pdu->hdr.common.plen < expected_hlen) { 2269 plen_error = true; 2270 } 2271 break; 2272 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2273 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2274 pdo = pdu->hdr.common.pdo; 2275 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2276 pdo_error = true; 2277 break; 2278 } 2279 if (pdu->hdr.common.plen < expected_hlen) { 2280 plen_error = true; 2281 } 2282 break; 2283 2284 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2285 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2286 if ((pdu->hdr.common.plen <= expected_hlen) || 2287 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2288 plen_error = true; 2289 } 2290 break; 2291 2292 default: 2293 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2294 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2295 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2296 goto err; 2297 } 2298 } 2299 2300 if (pdu->hdr.common.hlen != expected_hlen) { 2301 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2302 pdu->hdr.common.pdu_type, 2303 expected_hlen, pdu->hdr.common.hlen, tqpair); 2304 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2305 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2306 goto err; 2307 } else if (pdo_error) { 2308 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2309 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2310 } else if (plen_error) { 2311 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2312 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2313 goto err; 2314 } else { 2315 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2316 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2317 return; 2318 } 2319 err: 2320 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2321 } 2322 2323 static int 2324 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2325 { 2326 int rc = 0; 2327 struct nvme_tcp_pdu *pdu; 2328 enum nvme_tcp_pdu_recv_state prev_state; 2329 uint32_t data_len; 2330 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2331 struct spdk_nvmf_tcp_transport, transport); 2332 2333 /* The loop here is to allow for several back-to-back state changes. */ 2334 do { 2335 prev_state = tqpair->recv_state; 2336 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2337 2338 pdu = tqpair->pdu_in_progress; 2339 assert(pdu != NULL || 2340 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2341 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2342 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2343 2344 switch (tqpair->recv_state) { 2345 /* Wait for the common header */ 2346 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2347 if (!pdu) { 2348 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2349 if (spdk_unlikely(!pdu)) { 2350 return NVME_TCP_PDU_IN_PROGRESS; 2351 } 2352 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2353 tqpair->pdu_in_progress = pdu; 2354 tqpair->tcp_pdu_working_count++; 2355 } 2356 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2357 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2358 /* FALLTHROUGH */ 2359 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2360 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2361 return rc; 2362 } 2363 2364 rc = nvme_tcp_read_data(tqpair->sock, 2365 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2366 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2367 if (rc < 0) { 2368 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2369 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2370 break; 2371 } else if (rc > 0) { 2372 pdu->ch_valid_bytes += rc; 2373 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2374 } 2375 2376 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2377 return NVME_TCP_PDU_IN_PROGRESS; 2378 } 2379 2380 /* The command header of this PDU has now been read from the socket. */ 2381 nvmf_tcp_pdu_ch_handle(tqpair); 2382 break; 2383 /* Wait for the pdu specific header */ 2384 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2385 rc = nvme_tcp_read_data(tqpair->sock, 2386 pdu->psh_len - pdu->psh_valid_bytes, 2387 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2388 if (rc < 0) { 2389 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2390 break; 2391 } else if (rc > 0) { 2392 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2393 pdu->psh_valid_bytes += rc; 2394 } 2395 2396 if (pdu->psh_valid_bytes < pdu->psh_len) { 2397 return NVME_TCP_PDU_IN_PROGRESS; 2398 } 2399 2400 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2401 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2402 break; 2403 /* Wait for the req slot */ 2404 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2405 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2406 break; 2407 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2408 /* check whether the data is valid, if not we just return */ 2409 if (!pdu->data_len) { 2410 return NVME_TCP_PDU_IN_PROGRESS; 2411 } 2412 2413 data_len = pdu->data_len; 2414 /* data digest */ 2415 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2416 tqpair->host_ddgst_enable)) { 2417 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2418 pdu->ddgst_enable = true; 2419 } 2420 2421 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2422 if (rc < 0) { 2423 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2424 break; 2425 } 2426 pdu->rw_offset += rc; 2427 2428 if (pdu->rw_offset < data_len) { 2429 return NVME_TCP_PDU_IN_PROGRESS; 2430 } 2431 2432 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2433 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2434 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2435 pdu->dif_ctx) != 0) { 2436 SPDK_ERRLOG("DIF generate failed\n"); 2437 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2438 break; 2439 } 2440 2441 /* All of this PDU has now been read from the socket. */ 2442 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2443 break; 2444 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2445 if (tqpair->tcp_pdu_working_count != 0) { 2446 return NVME_TCP_PDU_IN_PROGRESS; 2447 } 2448 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2449 break; 2450 case NVME_TCP_PDU_RECV_STATE_ERROR: 2451 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2452 return NVME_TCP_PDU_IN_PROGRESS; 2453 } 2454 return NVME_TCP_PDU_FATAL; 2455 default: 2456 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2457 abort(); 2458 break; 2459 } 2460 } while (tqpair->recv_state != prev_state); 2461 2462 return rc; 2463 } 2464 2465 static inline void * 2466 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2467 { 2468 struct spdk_nvmf_tcp_control_msg *msg; 2469 2470 assert(list); 2471 2472 msg = STAILQ_FIRST(&list->free_msgs); 2473 if (!msg) { 2474 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2475 return NULL; 2476 } 2477 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2478 return msg; 2479 } 2480 2481 static inline void 2482 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2483 { 2484 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2485 2486 assert(list); 2487 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2488 } 2489 2490 static int 2491 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2492 struct spdk_nvmf_transport *transport, 2493 struct spdk_nvmf_transport_poll_group *group) 2494 { 2495 struct spdk_nvmf_request *req = &tcp_req->req; 2496 struct spdk_nvme_cmd *cmd; 2497 struct spdk_nvme_sgl_descriptor *sgl; 2498 struct spdk_nvmf_tcp_poll_group *tgroup; 2499 enum spdk_nvme_tcp_term_req_fes fes; 2500 struct nvme_tcp_pdu *pdu; 2501 struct spdk_nvmf_tcp_qpair *tqpair; 2502 uint32_t length, error_offset = 0; 2503 2504 cmd = &req->cmd->nvme_cmd; 2505 sgl = &cmd->dptr.sgl1; 2506 2507 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2508 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2509 /* get request length from sgl */ 2510 length = sgl->unkeyed.length; 2511 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2512 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2513 length, transport->opts.max_io_size); 2514 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2515 goto fatal_err; 2516 } 2517 2518 /* fill request length and populate iovs */ 2519 req->length = length; 2520 2521 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2522 2523 if (spdk_unlikely(req->dif_enabled)) { 2524 req->dif.orig_length = length; 2525 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2526 req->dif.elba_length = length; 2527 } 2528 2529 if (nvmf_ctrlr_use_zcopy(req)) { 2530 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2531 req->data_from_pool = false; 2532 return 0; 2533 } 2534 2535 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2536 /* No available buffers. Queue this request up. */ 2537 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2538 tcp_req); 2539 return 0; 2540 } 2541 2542 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2543 tcp_req, req->iovcnt, req->iov[0].iov_base); 2544 2545 return 0; 2546 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2547 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2548 uint64_t offset = sgl->address; 2549 uint32_t max_len = transport->opts.in_capsule_data_size; 2550 2551 assert(tcp_req->has_in_capsule_data); 2552 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2553 tqpair = tcp_req->pdu->qpair; 2554 /* receiving pdu is not same with the pdu in tcp_req */ 2555 pdu = tqpair->pdu_in_progress; 2556 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2557 if (tqpair->host_ddgst_enable) { 2558 length -= SPDK_NVME_TCP_DIGEST_LEN; 2559 } 2560 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2561 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2562 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2563 length, sgl->unkeyed.length); 2564 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2565 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2566 goto fatal_err; 2567 } 2568 2569 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2570 offset, length); 2571 2572 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2573 if (spdk_unlikely(offset != 0)) { 2574 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2575 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2576 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2577 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2578 goto fatal_err; 2579 } 2580 2581 if (spdk_unlikely(length > max_len)) { 2582 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2583 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2584 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2585 2586 /* Get a buffer from dedicated list */ 2587 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2588 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2589 assert(tgroup->control_msg_list); 2590 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2591 if (!req->iov[0].iov_base) { 2592 /* No available buffers. Queue this request up. */ 2593 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2594 return 0; 2595 } 2596 } else { 2597 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2598 length, max_len); 2599 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2600 goto fatal_err; 2601 } 2602 } else { 2603 req->iov[0].iov_base = tcp_req->buf; 2604 } 2605 2606 req->length = length; 2607 req->data_from_pool = false; 2608 2609 if (spdk_unlikely(req->dif_enabled)) { 2610 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2611 req->dif.elba_length = length; 2612 } 2613 2614 req->iov[0].iov_len = length; 2615 req->iovcnt = 1; 2616 2617 return 0; 2618 } 2619 /* If we want to handle the problem here, then we can't skip the following data segment. 2620 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2621 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2622 sgl->generic.type, sgl->generic.subtype); 2623 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2624 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2625 fatal_err: 2626 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2627 return -1; 2628 } 2629 2630 static inline enum spdk_nvme_media_error_status_code 2631 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2632 enum spdk_nvme_media_error_status_code result; 2633 2634 switch (err_type) 2635 { 2636 case SPDK_DIF_REFTAG_ERROR: 2637 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2638 break; 2639 case SPDK_DIF_APPTAG_ERROR: 2640 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2641 break; 2642 case SPDK_DIF_GUARD_ERROR: 2643 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2644 break; 2645 default: 2646 SPDK_UNREACHABLE(); 2647 break; 2648 } 2649 2650 return result; 2651 } 2652 2653 static void 2654 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2655 struct spdk_nvmf_tcp_req *tcp_req) 2656 { 2657 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2658 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2659 struct nvme_tcp_pdu *rsp_pdu; 2660 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2661 uint32_t plen, pdo, alignment; 2662 int rc; 2663 2664 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2665 2666 rsp_pdu = tcp_req->pdu; 2667 assert(rsp_pdu != NULL); 2668 2669 c2h_data = &rsp_pdu->hdr.c2h_data; 2670 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2671 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2672 2673 if (tqpair->host_hdgst_enable) { 2674 plen += SPDK_NVME_TCP_DIGEST_LEN; 2675 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2676 } 2677 2678 /* set the psh */ 2679 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2680 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2681 c2h_data->datao = tcp_req->pdu->rw_offset; 2682 2683 /* set the padding */ 2684 rsp_pdu->padding_len = 0; 2685 pdo = plen; 2686 if (tqpair->cpda) { 2687 alignment = (tqpair->cpda + 1) << 2; 2688 if (plen % alignment != 0) { 2689 pdo = (plen + alignment) / alignment * alignment; 2690 rsp_pdu->padding_len = pdo - plen; 2691 plen = pdo; 2692 } 2693 } 2694 2695 c2h_data->common.pdo = pdo; 2696 plen += c2h_data->datal; 2697 if (tqpair->host_ddgst_enable) { 2698 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2699 plen += SPDK_NVME_TCP_DIGEST_LEN; 2700 } 2701 2702 c2h_data->common.plen = plen; 2703 2704 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2705 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2706 } 2707 2708 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2709 c2h_data->datao, c2h_data->datal); 2710 2711 2712 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2713 /* Need to send the capsule response if response is not all 0 */ 2714 if (ttransport->tcp_opts.c2h_success && 2715 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2716 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2717 } 2718 2719 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2720 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2721 struct spdk_dif_error err_blk = {}; 2722 uint32_t mapped_length = 0; 2723 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2724 uint32_t ddgst_len = 0; 2725 2726 if (tqpair->host_ddgst_enable) { 2727 /* Data digest consumes additional iov entry */ 2728 available_iovs--; 2729 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2730 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2731 c2h_data->common.plen -= ddgst_len; 2732 } 2733 /* Temp call to estimate if data can be described by limited number of iovs. 2734 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2735 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2736 false, &mapped_length); 2737 2738 if (mapped_length != c2h_data->common.plen) { 2739 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2740 SPDK_DEBUGLOG(nvmf_tcp, 2741 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2742 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2743 c2h_data->common.plen = mapped_length; 2744 2745 /* Rebuild pdu->data_iov since data length is changed */ 2746 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2747 c2h_data->datal); 2748 2749 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2750 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2751 } 2752 2753 c2h_data->common.plen += ddgst_len; 2754 2755 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2756 2757 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2758 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2759 if (rc != 0) { 2760 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2761 err_blk.err_type, err_blk.err_offset); 2762 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2763 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2764 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2765 return; 2766 } 2767 } 2768 2769 rsp_pdu->rw_offset += c2h_data->datal; 2770 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2771 } 2772 2773 static void 2774 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2775 struct spdk_nvmf_tcp_req *tcp_req) 2776 { 2777 nvmf_tcp_req_pdu_init(tcp_req); 2778 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2779 } 2780 2781 static int 2782 request_transfer_out(struct spdk_nvmf_request *req) 2783 { 2784 struct spdk_nvmf_tcp_req *tcp_req; 2785 struct spdk_nvmf_qpair *qpair; 2786 struct spdk_nvmf_tcp_qpair *tqpair; 2787 struct spdk_nvme_cpl *rsp; 2788 2789 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2790 2791 qpair = req->qpair; 2792 rsp = &req->rsp->nvme_cpl; 2793 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2794 2795 /* Advance our sq_head pointer */ 2796 if (qpair->sq_head == qpair->sq_head_max) { 2797 qpair->sq_head = 0; 2798 } else { 2799 qpair->sq_head++; 2800 } 2801 rsp->sqhd = qpair->sq_head; 2802 2803 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2804 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2805 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2806 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2807 } else { 2808 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2809 } 2810 2811 return 0; 2812 } 2813 2814 static void 2815 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2816 struct spdk_nvmf_tcp_qpair *tqpair, 2817 struct spdk_nvmf_tcp_req *tcp_req) 2818 { 2819 enum spdk_nvme_cmd_fuse last, next; 2820 2821 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2822 next = tcp_req->cmd.fuse; 2823 2824 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2825 2826 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2827 return; 2828 } 2829 2830 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2831 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2832 /* This is a valid pair of fused commands. Point them at each other 2833 * so they can be submitted consecutively once ready to be executed. 2834 */ 2835 tqpair->fused_first->fused_pair = tcp_req; 2836 tcp_req->fused_pair = tqpair->fused_first; 2837 tqpair->fused_first = NULL; 2838 return; 2839 } else { 2840 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2841 tqpair->fused_first->fused_failed = true; 2842 2843 /* 2844 * If the last req is in READY_TO_EXECUTE state, then call 2845 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2846 */ 2847 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2848 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2849 } 2850 2851 tqpair->fused_first = NULL; 2852 } 2853 } 2854 2855 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2856 /* Set tqpair->fused_first here so that we know to check that the next request 2857 * is a SECOND (and to fail this one if it isn't). 2858 */ 2859 tqpair->fused_first = tcp_req; 2860 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2861 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2862 tcp_req->fused_failed = true; 2863 } 2864 } 2865 2866 static bool 2867 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2868 struct spdk_nvmf_tcp_req *tcp_req) 2869 { 2870 struct spdk_nvmf_tcp_qpair *tqpair; 2871 uint32_t plen; 2872 struct nvme_tcp_pdu *pdu; 2873 enum spdk_nvmf_tcp_req_state prev_state; 2874 bool progress = false; 2875 struct spdk_nvmf_transport *transport = &ttransport->transport; 2876 struct spdk_nvmf_transport_poll_group *group; 2877 struct spdk_nvmf_tcp_poll_group *tgroup; 2878 2879 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2880 group = &tqpair->group->group; 2881 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2882 2883 /* If the qpair is not active, we need to abort the outstanding requests. */ 2884 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 2885 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2886 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2887 } 2888 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2889 } 2890 2891 /* The loop here is to allow for several back-to-back state changes. */ 2892 do { 2893 prev_state = tcp_req->state; 2894 2895 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2896 tqpair); 2897 2898 switch (tcp_req->state) { 2899 case TCP_REQUEST_STATE_FREE: 2900 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2901 * to escape this state. */ 2902 break; 2903 case TCP_REQUEST_STATE_NEW: 2904 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 2905 tqpair->qpair.queue_depth); 2906 2907 /* copy the cmd from the receive pdu */ 2908 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2909 2910 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2911 tcp_req->req.dif_enabled = true; 2912 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2913 } 2914 2915 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2916 2917 /* The next state transition depends on the data transfer needs of this request. */ 2918 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2919 2920 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2921 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2922 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2923 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2924 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2925 break; 2926 } 2927 2928 /* If no data to transfer, ready to execute. */ 2929 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2930 /* Reset the tqpair receiving pdu state */ 2931 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2932 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2933 break; 2934 } 2935 2936 pdu = tqpair->pdu_in_progress; 2937 plen = pdu->hdr.common.hlen; 2938 if (tqpair->host_hdgst_enable) { 2939 plen += SPDK_NVME_TCP_DIGEST_LEN; 2940 } 2941 if (pdu->hdr.common.plen != plen) { 2942 tcp_req->has_in_capsule_data = true; 2943 } else { 2944 /* Data is transmitted by C2H PDUs */ 2945 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2946 } 2947 2948 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2949 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2950 break; 2951 case TCP_REQUEST_STATE_NEED_BUFFER: 2952 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 2953 (uintptr_t)tcp_req); 2954 2955 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2956 2957 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2958 SPDK_DEBUGLOG(nvmf_tcp, 2959 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2960 tcp_req, tqpair); 2961 /* This request needs to wait in line to obtain a buffer */ 2962 break; 2963 } 2964 2965 /* Try to get a data buffer */ 2966 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 2967 break; 2968 } 2969 2970 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2971 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2972 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2973 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2974 tcp_req->req.length = tcp_req->req.dif.elba_length; 2975 } 2976 2977 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2978 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2979 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2980 break; 2981 } 2982 2983 if (tcp_req->req.iovcnt < 1) { 2984 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2985 tcp_req, tqpair); 2986 /* No buffers available. */ 2987 break; 2988 } 2989 2990 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2991 2992 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2993 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2994 if (tcp_req->req.data_from_pool) { 2995 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2996 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2997 } else { 2998 struct nvme_tcp_pdu *pdu; 2999 3000 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3001 3002 pdu = tqpair->pdu_in_progress; 3003 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3004 tqpair); 3005 /* No need to send r2t, contained in the capsuled data */ 3006 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3007 0, tcp_req->req.length); 3008 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3009 } 3010 break; 3011 } 3012 3013 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3014 break; 3015 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3016 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3017 (uintptr_t)tcp_req); 3018 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3019 * to escape this state. */ 3020 break; 3021 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3022 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3023 (uintptr_t)tcp_req); 3024 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3025 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3026 tcp_req, tqpair); 3027 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3028 break; 3029 } 3030 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3031 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3032 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3033 } else { 3034 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3035 } 3036 break; 3037 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3038 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3039 (uintptr_t)tcp_req); 3040 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3041 break; 3042 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3043 3044 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3045 0, (uintptr_t)tcp_req); 3046 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3047 * to escape this state. */ 3048 break; 3049 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3050 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3051 (uintptr_t)tcp_req); 3052 3053 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3054 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3055 tcp_req->req.length = tcp_req->req.dif.elba_length; 3056 } 3057 3058 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3059 if (tcp_req->fused_failed) { 3060 /* This request failed FUSED semantics. Fail it immediately, without 3061 * even sending it to the target layer. 3062 */ 3063 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3064 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3065 break; 3066 } 3067 3068 if (tcp_req->fused_pair == NULL || 3069 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3070 /* This request is ready to execute, but either we don't know yet if it's 3071 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3072 * or the other request of this fused pair isn't ready to execute. So 3073 * break here and this request will get processed later either when the 3074 * other request is ready or we find that this request isn't valid. 3075 */ 3076 break; 3077 } 3078 } 3079 3080 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3081 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3082 /* If we get to this point, and this request is a fused command, we know that 3083 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3084 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3085 * request, and the other request of the fused pair, in the correct order. 3086 * Also clear the ->fused_pair pointers on both requests, since after this point 3087 * we no longer need to maintain the relationship between these two requests. 3088 */ 3089 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3090 assert(tcp_req->fused_pair != NULL); 3091 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3092 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3093 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3094 tcp_req->fused_pair->fused_pair = NULL; 3095 tcp_req->fused_pair = NULL; 3096 } 3097 spdk_nvmf_request_exec(&tcp_req->req); 3098 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3099 assert(tcp_req->fused_pair != NULL); 3100 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3101 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3102 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3103 tcp_req->fused_pair->fused_pair = NULL; 3104 tcp_req->fused_pair = NULL; 3105 } 3106 } else { 3107 /* For zero-copy, only requests with data coming from host to the 3108 * controller can end up here. */ 3109 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3110 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3111 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3112 } 3113 3114 break; 3115 case TCP_REQUEST_STATE_EXECUTING: 3116 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3117 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3118 * to escape this state. */ 3119 break; 3120 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3121 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3122 (uintptr_t)tcp_req); 3123 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3124 * to escape this state. */ 3125 break; 3126 case TCP_REQUEST_STATE_EXECUTED: 3127 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3128 3129 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3130 tcp_req->req.length = tcp_req->req.dif.orig_length; 3131 } 3132 3133 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3134 break; 3135 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3136 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3137 (uintptr_t)tcp_req); 3138 if (request_transfer_out(&tcp_req->req) != 0) { 3139 assert(0); /* No good way to handle this currently */ 3140 } 3141 break; 3142 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3143 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3144 0, (uintptr_t)tcp_req); 3145 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3146 * to escape this state. */ 3147 break; 3148 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3149 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3150 (uintptr_t)tcp_req); 3151 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3152 * to escape this state. */ 3153 break; 3154 case TCP_REQUEST_STATE_COMPLETED: 3155 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3156 tqpair->qpair.queue_depth); 3157 /* If there's an outstanding PDU sent to the host, the request is completed 3158 * due to the qpair being disconnected. We must delay the completion until 3159 * that write is done to avoid freeing the request twice. */ 3160 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3161 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3162 "write on req=%p\n", tcp_req); 3163 /* This can only happen for zcopy requests */ 3164 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3165 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3166 break; 3167 } 3168 3169 if (tcp_req->req.data_from_pool) { 3170 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3171 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3172 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3173 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3174 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3175 assert(tgroup->control_msg_list); 3176 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3177 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3178 tcp_req->req.iov[0].iov_base); 3179 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3180 /* If the request has an unreleased zcopy bdev_io, it's either a 3181 * read, a failed write, or the qpair is being disconnected */ 3182 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3183 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3184 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3185 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3186 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3187 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3188 break; 3189 } 3190 tcp_req->req.length = 0; 3191 tcp_req->req.iovcnt = 0; 3192 tcp_req->fused_failed = false; 3193 if (tcp_req->fused_pair) { 3194 /* This req was part of a valid fused pair, but failed before it got to 3195 * READ_TO_EXECUTE state. This means we need to fail the other request 3196 * in the pair, because it is no longer part of a valid pair. If the pair 3197 * already reached READY_TO_EXECUTE state, we need to kick it. 3198 */ 3199 tcp_req->fused_pair->fused_failed = true; 3200 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3201 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3202 } 3203 tcp_req->fused_pair = NULL; 3204 } 3205 3206 nvmf_tcp_req_put(tqpair, tcp_req); 3207 break; 3208 case TCP_REQUEST_NUM_STATES: 3209 default: 3210 assert(0); 3211 break; 3212 } 3213 3214 if (tcp_req->state != prev_state) { 3215 progress = true; 3216 } 3217 } while (tcp_req->state != prev_state); 3218 3219 return progress; 3220 } 3221 3222 static void 3223 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3224 { 3225 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3226 int rc; 3227 3228 assert(tqpair != NULL); 3229 rc = nvmf_tcp_sock_process(tqpair); 3230 3231 /* If there was a new socket error, disconnect */ 3232 if (rc < 0) { 3233 nvmf_tcp_qpair_disconnect(tqpair); 3234 } 3235 } 3236 3237 static int 3238 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3239 struct spdk_nvmf_qpair *qpair) 3240 { 3241 struct spdk_nvmf_tcp_poll_group *tgroup; 3242 struct spdk_nvmf_tcp_qpair *tqpair; 3243 int rc; 3244 3245 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3246 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3247 3248 rc = nvmf_tcp_qpair_sock_init(tqpair); 3249 if (rc != 0) { 3250 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3251 return -1; 3252 } 3253 3254 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3255 if (rc < 0) { 3256 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3257 return -1; 3258 } 3259 3260 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3261 if (rc < 0) { 3262 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3263 return -1; 3264 } 3265 3266 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3267 nvmf_tcp_sock_cb, tqpair); 3268 if (rc != 0) { 3269 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3270 spdk_strerror(errno), errno); 3271 return -1; 3272 } 3273 3274 tqpair->group = tgroup; 3275 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3276 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3277 3278 return 0; 3279 } 3280 3281 static int 3282 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3283 struct spdk_nvmf_qpair *qpair) 3284 { 3285 struct spdk_nvmf_tcp_poll_group *tgroup; 3286 struct spdk_nvmf_tcp_qpair *tqpair; 3287 int rc; 3288 3289 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3290 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3291 3292 assert(tqpair->group == tgroup); 3293 3294 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3295 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3296 /* Change the state to move the qpair from the await_req list to the main list 3297 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3298 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3299 } 3300 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3301 3302 /* Try to force out any pending writes */ 3303 spdk_sock_flush(tqpair->sock); 3304 3305 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3306 if (rc != 0) { 3307 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3308 spdk_strerror(errno), errno); 3309 } 3310 3311 return rc; 3312 } 3313 3314 static int 3315 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3316 { 3317 struct spdk_nvmf_tcp_transport *ttransport; 3318 struct spdk_nvmf_tcp_req *tcp_req; 3319 3320 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3321 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3322 3323 switch (tcp_req->state) { 3324 case TCP_REQUEST_STATE_EXECUTING: 3325 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3326 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3327 break; 3328 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3329 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3330 break; 3331 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3332 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3333 break; 3334 default: 3335 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3336 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3337 assert(0 && "Unexpected request state"); 3338 break; 3339 } 3340 3341 nvmf_tcp_req_process(ttransport, tcp_req); 3342 3343 return 0; 3344 } 3345 3346 static void 3347 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3348 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3349 { 3350 struct spdk_nvmf_tcp_qpair *tqpair; 3351 3352 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3353 3354 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3355 3356 assert(tqpair->fini_cb_fn == NULL); 3357 tqpair->fini_cb_fn = cb_fn; 3358 tqpair->fini_cb_arg = cb_arg; 3359 3360 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3361 nvmf_tcp_qpair_destroy(tqpair); 3362 } 3363 3364 static int 3365 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3366 { 3367 struct spdk_nvmf_tcp_poll_group *tgroup; 3368 int num_events, rc = 0, rc2; 3369 struct spdk_nvmf_request *req, *req_tmp; 3370 struct spdk_nvmf_tcp_req *tcp_req; 3371 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3372 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3373 struct spdk_nvmf_tcp_transport, transport); 3374 3375 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3376 3377 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3378 return 0; 3379 } 3380 3381 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3382 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3383 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3384 break; 3385 } 3386 } 3387 3388 num_events = spdk_sock_group_poll(tgroup->sock_group); 3389 if (spdk_unlikely(num_events < 0)) { 3390 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3391 } 3392 3393 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3394 rc2 = nvmf_tcp_sock_process(tqpair); 3395 3396 /* If there was a new socket error, disconnect */ 3397 if (spdk_unlikely(rc2 < 0)) { 3398 nvmf_tcp_qpair_disconnect(tqpair); 3399 if (rc == 0) { 3400 rc = rc2; 3401 } 3402 } 3403 } 3404 3405 return rc == 0 ? num_events : rc; 3406 } 3407 3408 static int 3409 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3410 struct spdk_nvme_transport_id *trid, bool peer) 3411 { 3412 struct spdk_nvmf_tcp_qpair *tqpair; 3413 uint16_t port; 3414 3415 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3416 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3417 3418 if (peer) { 3419 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3420 port = tqpair->initiator_port; 3421 } else { 3422 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3423 port = tqpair->target_port; 3424 } 3425 3426 if (spdk_sock_is_ipv4(tqpair->sock)) { 3427 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3428 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3429 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3430 } else { 3431 return -1; 3432 } 3433 3434 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3435 return 0; 3436 } 3437 3438 static int 3439 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3440 struct spdk_nvme_transport_id *trid) 3441 { 3442 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3443 } 3444 3445 static int 3446 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3447 struct spdk_nvme_transport_id *trid) 3448 { 3449 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3450 } 3451 3452 static int 3453 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3454 struct spdk_nvme_transport_id *trid) 3455 { 3456 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3457 } 3458 3459 static void 3460 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3461 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3462 { 3463 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3464 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3465 3466 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3467 } 3468 3469 static int 3470 _nvmf_tcp_qpair_abort_request(void *ctx) 3471 { 3472 struct spdk_nvmf_request *req = ctx; 3473 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3474 struct spdk_nvmf_tcp_req, req); 3475 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3476 struct spdk_nvmf_tcp_qpair, qpair); 3477 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3478 struct spdk_nvmf_tcp_transport, transport); 3479 int rc; 3480 3481 spdk_poller_unregister(&req->poller); 3482 3483 switch (tcp_req_to_abort->state) { 3484 case TCP_REQUEST_STATE_EXECUTING: 3485 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3486 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3487 rc = nvmf_ctrlr_abort_request(req); 3488 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3489 return SPDK_POLLER_BUSY; 3490 } 3491 break; 3492 3493 case TCP_REQUEST_STATE_NEED_BUFFER: 3494 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3495 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3496 3497 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3498 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3499 break; 3500 3501 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3502 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3503 if (spdk_get_ticks() < req->timeout_tsc) { 3504 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3505 return SPDK_POLLER_BUSY; 3506 } 3507 break; 3508 3509 default: 3510 /* Requests in other states are either un-abortable (e.g. 3511 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3512 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3513 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3514 * abort request with the bit0 of dword0 set (command not aborted). 3515 */ 3516 break; 3517 } 3518 3519 spdk_nvmf_request_complete(req); 3520 return SPDK_POLLER_BUSY; 3521 } 3522 3523 static void 3524 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3525 struct spdk_nvmf_request *req) 3526 { 3527 struct spdk_nvmf_tcp_qpair *tqpair; 3528 struct spdk_nvmf_tcp_transport *ttransport; 3529 struct spdk_nvmf_transport *transport; 3530 uint16_t cid; 3531 uint32_t i; 3532 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3533 3534 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3535 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3536 transport = &ttransport->transport; 3537 3538 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3539 3540 for (i = 0; i < tqpair->resource_count; i++) { 3541 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3542 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3543 tcp_req_to_abort = &tqpair->reqs[i]; 3544 break; 3545 } 3546 } 3547 3548 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3549 3550 if (tcp_req_to_abort == NULL) { 3551 spdk_nvmf_request_complete(req); 3552 return; 3553 } 3554 3555 req->req_to_abort = &tcp_req_to_abort->req; 3556 req->timeout_tsc = spdk_get_ticks() + 3557 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3558 req->poller = NULL; 3559 3560 _nvmf_tcp_qpair_abort_request(req); 3561 } 3562 3563 struct tcp_subsystem_add_host_opts { 3564 char *psk; 3565 }; 3566 3567 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3568 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3569 }; 3570 3571 static int 3572 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3573 { 3574 FILE *psk_file; 3575 struct stat statbuf; 3576 int rc; 3577 3578 if (stat(fname, &statbuf) != 0) { 3579 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3580 return -EACCES; 3581 } 3582 3583 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3584 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3585 return -EPERM; 3586 } 3587 if ((size_t)statbuf.st_size > bufsz) { 3588 SPDK_ERRLOG("Invalid PSK: too long\n"); 3589 return -EINVAL; 3590 } 3591 psk_file = fopen(fname, "r"); 3592 if (psk_file == NULL) { 3593 SPDK_ERRLOG("Could not open PSK file\n"); 3594 return -EINVAL; 3595 } 3596 3597 rc = fread(buf, 1, statbuf.st_size, psk_file); 3598 if (rc != statbuf.st_size) { 3599 SPDK_ERRLOG("Failed to read PSK\n"); 3600 fclose(psk_file); 3601 return -EINVAL; 3602 } 3603 3604 fclose(psk_file); 3605 return 0; 3606 } 3607 3608 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3609 3610 static int 3611 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3612 const struct spdk_nvmf_subsystem *subsystem, 3613 const char *hostnqn, 3614 const struct spdk_json_val *transport_specific) 3615 { 3616 struct tcp_subsystem_add_host_opts opts; 3617 struct spdk_nvmf_tcp_transport *ttransport; 3618 struct tcp_psk_entry *tmp, *entry = NULL; 3619 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3620 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3621 uint8_t tls_cipher_suite; 3622 int rc = 0; 3623 uint8_t psk_retained_hash; 3624 uint64_t psk_configured_size; 3625 3626 if (transport_specific == NULL) { 3627 return 0; 3628 } 3629 3630 assert(transport != NULL); 3631 assert(subsystem != NULL); 3632 3633 memset(&opts, 0, sizeof(opts)); 3634 3635 /* Decode PSK (either name of a key or file path) */ 3636 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3637 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3638 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3639 return -EINVAL; 3640 } 3641 3642 if (opts.psk == NULL) { 3643 return 0; 3644 } 3645 3646 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3647 if (entry == NULL) { 3648 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3649 rc = -ENOMEM; 3650 goto end; 3651 } 3652 3653 entry->key = spdk_keyring_get_key(opts.psk); 3654 if (entry->key != NULL) { 3655 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3656 if (rc < 0) { 3657 SPDK_ERRLOG("Failed to retreive PSK '%s'\n", opts.psk); 3658 rc = -EINVAL; 3659 goto end; 3660 } 3661 } else { 3662 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3663 SPDK_ERRLOG("PSK path too long\n"); 3664 rc = -EINVAL; 3665 goto end; 3666 } 3667 3668 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3669 if (rc) { 3670 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3671 goto end; 3672 } 3673 3674 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3675 } 3676 3677 /* Parse PSK interchange to get length of base64 encoded data. 3678 * This is then used to decide which cipher suite should be used 3679 * to generate PSK identity and TLS PSK later on. */ 3680 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3681 &psk_configured_size, &psk_retained_hash); 3682 if (rc < 0) { 3683 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3684 goto end; 3685 } 3686 3687 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3688 * This check also ensures that psk_configured_size is smaller than 3689 * psk_retained buffer size. */ 3690 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3691 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3692 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3693 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3694 } else { 3695 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3696 rc = -EINVAL; 3697 goto end; 3698 } 3699 3700 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3701 /* Generate PSK identity. */ 3702 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3703 subsystem->subnqn, tls_cipher_suite); 3704 if (rc) { 3705 rc = -EINVAL; 3706 goto end; 3707 } 3708 /* Check if PSK identity entry already exists. */ 3709 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3710 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3711 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3712 rc = -EEXIST; 3713 goto end; 3714 } 3715 } 3716 3717 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3718 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3719 rc = -EINVAL; 3720 goto end; 3721 } 3722 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3723 SPDK_ERRLOG("Could not write subnqn string!\n"); 3724 rc = -EINVAL; 3725 goto end; 3726 } 3727 3728 entry->tls_cipher_suite = tls_cipher_suite; 3729 3730 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3731 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3732 /* Psk configured is either 32 or 48 bytes long. */ 3733 memcpy(entry->psk, psk_configured, psk_configured_size); 3734 entry->psk_size = psk_configured_size; 3735 } else { 3736 /* Derive retained PSK. */ 3737 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3738 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3739 if (rc < 0) { 3740 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3741 goto end; 3742 } 3743 entry->psk_size = rc; 3744 } 3745 3746 if (entry->key == NULL) { 3747 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3748 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3749 SPDK_ERRLOG("Could not save PSK path!\n"); 3750 rc = -ENAMETOOLONG; 3751 goto end; 3752 } 3753 } 3754 3755 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3756 rc = 0; 3757 3758 end: 3759 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3760 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3761 3762 free(opts.psk); 3763 if (rc != 0) { 3764 nvmf_tcp_free_psk_entry(entry); 3765 } 3766 3767 return rc; 3768 } 3769 3770 static void 3771 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3772 const struct spdk_nvmf_subsystem *subsystem, 3773 const char *hostnqn) 3774 { 3775 struct spdk_nvmf_tcp_transport *ttransport; 3776 struct tcp_psk_entry *entry, *tmp; 3777 3778 assert(transport != NULL); 3779 assert(subsystem != NULL); 3780 3781 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3782 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3783 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3784 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3785 TAILQ_REMOVE(&ttransport->psks, entry, link); 3786 nvmf_tcp_free_psk_entry(entry); 3787 break; 3788 } 3789 } 3790 } 3791 3792 static void 3793 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3794 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3795 struct spdk_json_write_ctx *w) 3796 { 3797 struct spdk_nvmf_tcp_transport *ttransport; 3798 struct tcp_psk_entry *entry; 3799 3800 assert(transport != NULL); 3801 assert(subsystem != NULL); 3802 3803 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3804 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3805 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3806 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3807 spdk_json_write_named_string(w, "psk", entry->key ? 3808 spdk_key_get_name(entry->key) : entry->psk_path); 3809 break; 3810 } 3811 } 3812 } 3813 3814 static void 3815 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3816 { 3817 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3818 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3819 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3820 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3821 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3822 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3823 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3824 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3825 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3826 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3827 opts->transport_specific = NULL; 3828 } 3829 3830 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3831 .name = "TCP", 3832 .type = SPDK_NVME_TRANSPORT_TCP, 3833 .opts_init = nvmf_tcp_opts_init, 3834 .create = nvmf_tcp_create, 3835 .dump_opts = nvmf_tcp_dump_opts, 3836 .destroy = nvmf_tcp_destroy, 3837 3838 .listen = nvmf_tcp_listen, 3839 .stop_listen = nvmf_tcp_stop_listen, 3840 3841 .listener_discover = nvmf_tcp_discover, 3842 3843 .poll_group_create = nvmf_tcp_poll_group_create, 3844 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3845 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3846 .poll_group_add = nvmf_tcp_poll_group_add, 3847 .poll_group_remove = nvmf_tcp_poll_group_remove, 3848 .poll_group_poll = nvmf_tcp_poll_group_poll, 3849 3850 .req_free = nvmf_tcp_req_free, 3851 .req_complete = nvmf_tcp_req_complete, 3852 3853 .qpair_fini = nvmf_tcp_close_qpair, 3854 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3855 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3856 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3857 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3858 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3859 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3860 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3861 }; 3862 3863 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3864 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3865