1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022, 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 #include "spdk/keyring.h" 19 20 #include "spdk_internal/assert.h" 21 #include "spdk_internal/nvme_tcp.h" 22 #include "spdk_internal/sock.h" 23 24 #include "nvmf_internal.h" 25 26 #include "spdk_internal/trace_defs.h" 27 28 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 29 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 30 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 31 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 32 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 33 34 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 35 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 36 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 37 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 38 39 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 42 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 43 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 44 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 46 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 47 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 48 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 49 50 #define TCP_PSK_INVALID_PERMISSIONS 0177 51 52 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 53 static bool g_tls_log = false; 54 55 /* spdk nvmf related structure */ 56 enum spdk_nvmf_tcp_req_state { 57 58 /* The request is not currently in use */ 59 TCP_REQUEST_STATE_FREE = 0, 60 61 /* Initial state when request first received */ 62 TCP_REQUEST_STATE_NEW = 1, 63 64 /* The request is queued until a data buffer is available. */ 65 TCP_REQUEST_STATE_NEED_BUFFER = 2, 66 67 /* The request is waiting for zcopy_start to finish */ 68 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 3, 69 70 /* The request has received a zero-copy buffer */ 71 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 4, 72 73 /* The request is currently transferring data from the host to the controller. */ 74 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 5, 75 76 /* The request is waiting for the R2T send acknowledgement. */ 77 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 6, 78 79 /* The request is ready to execute at the block device */ 80 TCP_REQUEST_STATE_READY_TO_EXECUTE = 7, 81 82 /* The request is currently executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTING = 8, 84 85 /* The request is waiting for zcopy buffers to be committed */ 86 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 9, 87 88 /* The request finished executing at the block device */ 89 TCP_REQUEST_STATE_EXECUTED = 10, 90 91 /* The request is ready to send a completion */ 92 TCP_REQUEST_STATE_READY_TO_COMPLETE = 11, 93 94 /* The request is currently transferring final pdus from the controller to the host. */ 95 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 12, 96 97 /* The request is waiting for zcopy buffers to be released (without committing) */ 98 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 13, 99 100 /* The request completed and can be marked free. */ 101 TCP_REQUEST_STATE_COMPLETED = 14, 102 103 /* Terminator */ 104 TCP_REQUEST_NUM_STATES, 105 }; 106 107 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 108 "Invalid PDU Header Field", 109 "PDU Sequence Error", 110 "Header Digiest Error", 111 "Data Transfer Out of Range", 112 "R2T Limit Exceeded", 113 "Unsupported parameter", 114 }; 115 116 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 117 { 118 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 119 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 120 spdk_trace_register_description("TCP_REQ_NEW", 121 TRACE_TCP_REQUEST_STATE_NEW, 122 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 123 SPDK_TRACE_ARG_TYPE_INT, "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 spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1311 caddr, sizeof(caddr), &cport); 1312 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1313 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1314 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1315 1316 /* set low water mark */ 1317 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1318 if (rc != 0) { 1319 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1320 return rc; 1321 } 1322 1323 return 0; 1324 } 1325 1326 static void 1327 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1328 struct spdk_nvmf_tcp_port *port, 1329 struct spdk_sock *sock) 1330 { 1331 struct spdk_nvmf_tcp_qpair *tqpair; 1332 int rc; 1333 1334 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1335 port->trid->traddr, port->trid->trsvcid); 1336 1337 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1338 if (tqpair == NULL) { 1339 SPDK_ERRLOG("Could not allocate new connection.\n"); 1340 spdk_sock_close(&sock); 1341 return; 1342 } 1343 1344 tqpair->sock = sock; 1345 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1346 tqpair->port = port; 1347 tqpair->qpair.transport = transport; 1348 1349 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1350 sizeof(tqpair->target_addr), &tqpair->target_port, 1351 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1352 &tqpair->initiator_port); 1353 if (rc < 0) { 1354 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1355 nvmf_tcp_qpair_destroy(tqpair); 1356 return; 1357 } 1358 1359 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1360 } 1361 1362 static uint32_t 1363 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1364 { 1365 struct spdk_sock *sock; 1366 uint32_t count = 0; 1367 int i; 1368 1369 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1370 sock = spdk_sock_accept(port->listen_sock); 1371 if (sock == NULL) { 1372 break; 1373 } 1374 count++; 1375 nvmf_tcp_handle_connect(transport, port, sock); 1376 } 1377 1378 return count; 1379 } 1380 1381 static int 1382 nvmf_tcp_accept(void *ctx) 1383 { 1384 struct spdk_nvmf_transport *transport = ctx; 1385 struct spdk_nvmf_tcp_transport *ttransport; 1386 struct spdk_nvmf_tcp_port *port; 1387 uint32_t count = 0; 1388 1389 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1390 1391 TAILQ_FOREACH(port, &ttransport->ports, link) { 1392 count += nvmf_tcp_port_accept(transport, port); 1393 } 1394 1395 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1396 } 1397 1398 static void 1399 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1400 struct spdk_nvme_transport_id *trid, 1401 struct spdk_nvmf_discovery_log_page_entry *entry) 1402 { 1403 struct spdk_nvmf_tcp_port *port; 1404 struct spdk_nvmf_tcp_transport *ttransport; 1405 1406 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1407 entry->adrfam = trid->adrfam; 1408 1409 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1410 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1411 1412 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1413 port = nvmf_tcp_find_port(ttransport, trid); 1414 1415 assert(port != NULL); 1416 1417 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1418 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1419 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1420 } else { 1421 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1422 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1423 } 1424 } 1425 1426 static struct spdk_nvmf_tcp_control_msg_list * 1427 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1428 { 1429 struct spdk_nvmf_tcp_control_msg_list *list; 1430 struct spdk_nvmf_tcp_control_msg *msg; 1431 uint16_t i; 1432 1433 list = calloc(1, sizeof(*list)); 1434 if (!list) { 1435 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1436 return NULL; 1437 } 1438 1439 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1440 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1441 if (!list->msg_buf) { 1442 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1443 free(list); 1444 return NULL; 1445 } 1446 1447 STAILQ_INIT(&list->free_msgs); 1448 1449 for (i = 0; i < num_messages; i++) { 1450 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1451 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1452 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1453 } 1454 1455 return list; 1456 } 1457 1458 static void 1459 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1460 { 1461 if (!list) { 1462 return; 1463 } 1464 1465 spdk_free(list->msg_buf); 1466 free(list); 1467 } 1468 1469 static struct spdk_nvmf_transport_poll_group * 1470 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1471 struct spdk_nvmf_poll_group *group) 1472 { 1473 struct spdk_nvmf_tcp_transport *ttransport; 1474 struct spdk_nvmf_tcp_poll_group *tgroup; 1475 1476 tgroup = calloc(1, sizeof(*tgroup)); 1477 if (!tgroup) { 1478 return NULL; 1479 } 1480 1481 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1482 if (!tgroup->sock_group) { 1483 goto cleanup; 1484 } 1485 1486 TAILQ_INIT(&tgroup->qpairs); 1487 TAILQ_INIT(&tgroup->await_req); 1488 1489 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1490 1491 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1492 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1493 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1494 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1495 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1496 if (!tgroup->control_msg_list) { 1497 goto cleanup; 1498 } 1499 } 1500 1501 tgroup->accel_channel = spdk_accel_get_io_channel(); 1502 if (spdk_unlikely(!tgroup->accel_channel)) { 1503 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1504 goto cleanup; 1505 } 1506 1507 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1508 if (ttransport->next_pg == NULL) { 1509 ttransport->next_pg = tgroup; 1510 } 1511 1512 return &tgroup->group; 1513 1514 cleanup: 1515 nvmf_tcp_poll_group_destroy(&tgroup->group); 1516 return NULL; 1517 } 1518 1519 static struct spdk_nvmf_transport_poll_group * 1520 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1521 { 1522 struct spdk_nvmf_tcp_transport *ttransport; 1523 struct spdk_nvmf_tcp_poll_group **pg; 1524 struct spdk_nvmf_tcp_qpair *tqpair; 1525 struct spdk_sock_group *group = NULL, *hint = NULL; 1526 int rc; 1527 1528 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1529 1530 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1531 return NULL; 1532 } 1533 1534 pg = &ttransport->next_pg; 1535 assert(*pg != NULL); 1536 hint = (*pg)->sock_group; 1537 1538 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1539 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1540 if (rc != 0) { 1541 return NULL; 1542 } else if (group != NULL) { 1543 /* Optimal poll group was found */ 1544 return spdk_sock_group_get_ctx(group); 1545 } 1546 1547 /* The hint was used for optimal poll group, advance next_pg. */ 1548 *pg = TAILQ_NEXT(*pg, link); 1549 if (*pg == NULL) { 1550 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1551 } 1552 1553 return spdk_sock_group_get_ctx(hint); 1554 } 1555 1556 static void 1557 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1558 { 1559 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1560 struct spdk_nvmf_tcp_transport *ttransport; 1561 1562 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1563 spdk_sock_group_close(&tgroup->sock_group); 1564 if (tgroup->control_msg_list) { 1565 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1566 } 1567 1568 if (tgroup->accel_channel) { 1569 spdk_put_io_channel(tgroup->accel_channel); 1570 } 1571 1572 if (tgroup->group.transport == NULL) { 1573 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1574 * calls this function directly in a failure path. */ 1575 free(tgroup); 1576 return; 1577 } 1578 1579 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1580 1581 next_tgroup = TAILQ_NEXT(tgroup, link); 1582 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1583 if (next_tgroup == NULL) { 1584 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1585 } 1586 if (ttransport->next_pg == tgroup) { 1587 ttransport->next_pg = next_tgroup; 1588 } 1589 1590 free(tgroup); 1591 } 1592 1593 static void 1594 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1595 enum nvme_tcp_pdu_recv_state state) 1596 { 1597 if (tqpair->recv_state == state) { 1598 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1599 tqpair, state); 1600 return; 1601 } 1602 1603 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1604 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1605 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1606 tqpair->tcp_pdu_working_count--; 1607 } 1608 } 1609 1610 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1611 assert(tqpair->tcp_pdu_working_count == 0); 1612 } 1613 1614 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1615 /* When leaving the await req state, move the qpair to the main list */ 1616 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1617 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1618 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1619 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1620 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1621 } 1622 1623 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1624 tqpair->recv_state = state; 1625 1626 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1627 (uint64_t)tqpair->recv_state); 1628 } 1629 1630 static int 1631 nvmf_tcp_qpair_handle_timeout(void *ctx) 1632 { 1633 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1634 1635 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1636 1637 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1638 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1639 1640 nvmf_tcp_qpair_disconnect(tqpair); 1641 return SPDK_POLLER_BUSY; 1642 } 1643 1644 static void 1645 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1646 { 1647 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1648 1649 if (!tqpair->timeout_poller) { 1650 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1651 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1652 } 1653 } 1654 1655 static void 1656 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1657 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1658 { 1659 struct nvme_tcp_pdu *rsp_pdu; 1660 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1661 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1662 uint32_t copy_len; 1663 1664 rsp_pdu = tqpair->mgmt_pdu; 1665 1666 c2h_term_req = &rsp_pdu->hdr.term_req; 1667 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1668 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1669 c2h_term_req->fes = fes; 1670 1671 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1672 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1673 DSET32(&c2h_term_req->fei, error_offset); 1674 } 1675 1676 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1677 1678 /* Copy the error info into the buffer */ 1679 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1680 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1681 1682 /* Contain the header of the wrong received pdu */ 1683 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1684 tqpair->wait_terminate = true; 1685 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1686 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1687 } 1688 1689 static void 1690 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1691 struct spdk_nvmf_tcp_qpair *tqpair, 1692 struct nvme_tcp_pdu *pdu) 1693 { 1694 struct spdk_nvmf_tcp_req *tcp_req; 1695 1696 assert(pdu->psh_valid_bytes == pdu->psh_len); 1697 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1698 1699 tcp_req = nvmf_tcp_req_get(tqpair); 1700 if (!tcp_req) { 1701 /* Directly return and make the allocation retry again. This can happen if we're 1702 * using asynchronous writes to send the response to the host or when releasing 1703 * zero-copy buffers after a response has been sent. In both cases, the host might 1704 * receive the response before we've finished processing the request and is free to 1705 * send another one. 1706 */ 1707 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1708 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1709 return; 1710 } 1711 1712 /* The host sent more commands than the maximum queue depth. */ 1713 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1714 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1715 return; 1716 } 1717 1718 pdu->req = tcp_req; 1719 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1720 nvmf_tcp_req_process(ttransport, tcp_req); 1721 } 1722 1723 static void 1724 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1725 struct spdk_nvmf_tcp_qpair *tqpair, 1726 struct nvme_tcp_pdu *pdu) 1727 { 1728 struct spdk_nvmf_tcp_req *tcp_req; 1729 struct spdk_nvme_tcp_cmd *capsule_cmd; 1730 uint32_t error_offset = 0; 1731 enum spdk_nvme_tcp_term_req_fes fes; 1732 struct spdk_nvme_cpl *rsp; 1733 1734 capsule_cmd = &pdu->hdr.capsule_cmd; 1735 tcp_req = pdu->req; 1736 assert(tcp_req != NULL); 1737 1738 /* Zero-copy requests don't support ICD */ 1739 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1740 1741 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1742 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1743 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1744 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1745 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1746 goto err; 1747 } 1748 1749 rsp = &tcp_req->req.rsp->nvme_cpl; 1750 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1751 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1752 } else { 1753 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1754 } 1755 1756 nvmf_tcp_req_process(ttransport, tcp_req); 1757 1758 return; 1759 err: 1760 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1761 } 1762 1763 static void 1764 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1765 struct spdk_nvmf_tcp_qpair *tqpair, 1766 struct nvme_tcp_pdu *pdu) 1767 { 1768 struct spdk_nvmf_tcp_req *tcp_req; 1769 uint32_t error_offset = 0; 1770 enum spdk_nvme_tcp_term_req_fes fes = 0; 1771 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1772 1773 h2c_data = &pdu->hdr.h2c_data; 1774 1775 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1776 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1777 1778 if (h2c_data->ttag > tqpair->resource_count) { 1779 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1780 tqpair->resource_count); 1781 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1782 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1783 goto err; 1784 } 1785 1786 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1787 1788 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1789 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1790 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1791 tcp_req->state); 1792 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1793 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1794 goto err; 1795 } 1796 1797 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1798 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1799 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1800 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1801 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1802 goto err; 1803 } 1804 1805 if (tcp_req->h2c_offset != h2c_data->datao) { 1806 SPDK_DEBUGLOG(nvmf_tcp, 1807 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1808 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1809 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1810 goto err; 1811 } 1812 1813 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1814 SPDK_DEBUGLOG(nvmf_tcp, 1815 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1816 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1817 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1818 goto err; 1819 } 1820 1821 pdu->req = tcp_req; 1822 1823 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1824 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1825 } 1826 1827 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1828 h2c_data->datao, h2c_data->datal); 1829 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1830 return; 1831 1832 err: 1833 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1834 } 1835 1836 static void 1837 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1838 struct spdk_nvmf_tcp_qpair *tqpair) 1839 { 1840 struct nvme_tcp_pdu *rsp_pdu; 1841 struct spdk_nvme_tcp_rsp *capsule_resp; 1842 1843 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1844 1845 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1846 assert(rsp_pdu != NULL); 1847 1848 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1849 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1850 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1851 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1852 if (tqpair->host_hdgst_enable) { 1853 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1854 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1855 } 1856 1857 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1858 } 1859 1860 static void 1861 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1862 { 1863 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1864 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1865 struct spdk_nvmf_tcp_qpair, qpair); 1866 1867 assert(tqpair != NULL); 1868 1869 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1870 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1871 tcp_req->req.length); 1872 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1873 return; 1874 } 1875 1876 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1877 nvmf_tcp_request_free(tcp_req); 1878 } else { 1879 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1880 } 1881 } 1882 1883 static void 1884 nvmf_tcp_r2t_complete(void *cb_arg) 1885 { 1886 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1887 struct spdk_nvmf_tcp_transport *ttransport; 1888 1889 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1890 struct spdk_nvmf_tcp_transport, transport); 1891 1892 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1893 1894 if (tcp_req->h2c_offset == tcp_req->req.length) { 1895 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1896 nvmf_tcp_req_process(ttransport, tcp_req); 1897 } 1898 } 1899 1900 static void 1901 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1902 struct spdk_nvmf_tcp_req *tcp_req) 1903 { 1904 struct nvme_tcp_pdu *rsp_pdu; 1905 struct spdk_nvme_tcp_r2t_hdr *r2t; 1906 1907 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1908 assert(rsp_pdu != NULL); 1909 1910 r2t = &rsp_pdu->hdr.r2t; 1911 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1912 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1913 1914 if (tqpair->host_hdgst_enable) { 1915 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1916 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1917 } 1918 1919 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1920 r2t->ttag = tcp_req->ttag; 1921 r2t->r2to = tcp_req->h2c_offset; 1922 r2t->r2tl = tcp_req->req.length; 1923 1924 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1925 1926 SPDK_DEBUGLOG(nvmf_tcp, 1927 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1928 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1929 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1930 } 1931 1932 static void 1933 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1934 struct spdk_nvmf_tcp_qpair *tqpair, 1935 struct nvme_tcp_pdu *pdu) 1936 { 1937 struct spdk_nvmf_tcp_req *tcp_req; 1938 struct spdk_nvme_cpl *rsp; 1939 1940 tcp_req = pdu->req; 1941 assert(tcp_req != NULL); 1942 1943 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1944 1945 tcp_req->h2c_offset += pdu->data_len; 1946 1947 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1948 * acknowledged before moving on. */ 1949 if (tcp_req->h2c_offset == tcp_req->req.length && 1950 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1951 /* After receiving all the h2c data, we need to check whether there is 1952 * transient transport error */ 1953 rsp = &tcp_req->req.rsp->nvme_cpl; 1954 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1955 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1956 } else { 1957 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1958 } 1959 nvmf_tcp_req_process(ttransport, tcp_req); 1960 } 1961 } 1962 1963 static void 1964 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1965 { 1966 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1967 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1968 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1969 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1970 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1971 DGET32(h2c_term_req->fei)); 1972 } 1973 } 1974 1975 static void 1976 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1977 struct nvme_tcp_pdu *pdu) 1978 { 1979 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1980 uint32_t error_offset = 0; 1981 enum spdk_nvme_tcp_term_req_fes fes; 1982 1983 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1984 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1985 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1986 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1987 goto end; 1988 } 1989 1990 /* set the data buffer */ 1991 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1992 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1993 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1994 return; 1995 end: 1996 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1997 } 1998 1999 static void 2000 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2001 struct nvme_tcp_pdu *pdu) 2002 { 2003 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2004 2005 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2006 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2007 } 2008 2009 static void 2010 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2011 { 2012 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2013 struct spdk_nvmf_tcp_transport, transport); 2014 2015 switch (pdu->hdr.common.pdu_type) { 2016 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2017 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2018 break; 2019 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2020 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2021 break; 2022 2023 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2024 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2025 break; 2026 2027 default: 2028 /* The code should not go to here */ 2029 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2030 break; 2031 } 2032 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2033 tqpair->tcp_pdu_working_count--; 2034 } 2035 2036 static inline void 2037 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2038 { 2039 treq->req.rsp->nvme_cpl.status.sct = sct; 2040 treq->req.rsp->nvme_cpl.status.sc = sc; 2041 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2042 } 2043 2044 static void 2045 data_crc32_calc_done(void *cb_arg, int status) 2046 { 2047 struct nvme_tcp_pdu *pdu = cb_arg; 2048 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2049 2050 /* async crc32 calculation is failed and use direct calculation to check */ 2051 if (spdk_unlikely(status)) { 2052 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2053 tqpair, pdu); 2054 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2055 } 2056 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2057 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2058 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2059 assert(pdu->req != NULL); 2060 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2061 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2062 } 2063 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2064 } 2065 2066 static void 2067 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2068 { 2069 int rc = 0; 2070 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2071 tqpair->pdu_in_progress = NULL; 2072 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2073 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2074 /* check data digest if need */ 2075 if (pdu->ddgst_enable) { 2076 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2077 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2078 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2079 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2080 if (spdk_likely(rc == 0)) { 2081 return; 2082 } 2083 } else { 2084 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2085 } 2086 data_crc32_calc_done(pdu, rc); 2087 } else { 2088 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2089 } 2090 } 2091 2092 static void 2093 nvmf_tcp_send_icresp_complete(void *cb_arg) 2094 { 2095 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2096 2097 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2098 } 2099 2100 static void 2101 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2102 struct spdk_nvmf_tcp_qpair *tqpair, 2103 struct nvme_tcp_pdu *pdu) 2104 { 2105 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2106 struct nvme_tcp_pdu *rsp_pdu; 2107 struct spdk_nvme_tcp_ic_resp *ic_resp; 2108 uint32_t error_offset = 0; 2109 enum spdk_nvme_tcp_term_req_fes fes; 2110 2111 /* Only PFV 0 is defined currently */ 2112 if (ic_req->pfv != 0) { 2113 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2114 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2115 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2116 goto end; 2117 } 2118 2119 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2120 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2121 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2122 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2123 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2124 goto end; 2125 } 2126 2127 /* MAXR2T is 0's based */ 2128 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2129 2130 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2131 if (!tqpair->host_hdgst_enable) { 2132 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2133 } 2134 2135 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2136 if (!tqpair->host_ddgst_enable) { 2137 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2138 } 2139 2140 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2141 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2142 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2143 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2144 tqpair, 2145 tqpair->recv_buf_size); 2146 /* Not fatal. */ 2147 } 2148 2149 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2150 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2151 2152 rsp_pdu = tqpair->mgmt_pdu; 2153 2154 ic_resp = &rsp_pdu->hdr.ic_resp; 2155 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2156 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2157 ic_resp->pfv = 0; 2158 ic_resp->cpda = tqpair->cpda; 2159 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2160 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2161 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2162 2163 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2164 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2165 2166 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2167 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2168 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2169 return; 2170 end: 2171 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2172 } 2173 2174 static void 2175 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2176 struct spdk_nvmf_tcp_transport *ttransport) 2177 { 2178 struct nvme_tcp_pdu *pdu; 2179 int rc; 2180 uint32_t crc32c, error_offset = 0; 2181 enum spdk_nvme_tcp_term_req_fes fes; 2182 2183 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2184 pdu = tqpair->pdu_in_progress; 2185 2186 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2187 pdu->hdr.common.pdu_type); 2188 /* check header digest if needed */ 2189 if (pdu->has_hdgst) { 2190 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2191 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2192 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2193 if (rc == 0) { 2194 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2195 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2196 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2197 return; 2198 2199 } 2200 } 2201 2202 switch (pdu->hdr.common.pdu_type) { 2203 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2204 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2205 break; 2206 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2207 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2208 break; 2209 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2210 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2211 break; 2212 2213 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2214 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2215 break; 2216 2217 default: 2218 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2219 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2220 error_offset = 1; 2221 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2222 break; 2223 } 2224 } 2225 2226 static void 2227 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2228 { 2229 struct nvme_tcp_pdu *pdu; 2230 uint32_t error_offset = 0; 2231 enum spdk_nvme_tcp_term_req_fes fes; 2232 uint8_t expected_hlen, pdo; 2233 bool plen_error = false, pdo_error = false; 2234 2235 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2236 pdu = tqpair->pdu_in_progress; 2237 assert(pdu); 2238 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2239 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2240 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2241 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2242 goto err; 2243 } 2244 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2245 if (pdu->hdr.common.plen != expected_hlen) { 2246 plen_error = true; 2247 } 2248 } else { 2249 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2250 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2251 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2252 goto err; 2253 } 2254 2255 switch (pdu->hdr.common.pdu_type) { 2256 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2257 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2258 pdo = pdu->hdr.common.pdo; 2259 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2260 pdo_error = true; 2261 break; 2262 } 2263 2264 if (pdu->hdr.common.plen < expected_hlen) { 2265 plen_error = true; 2266 } 2267 break; 2268 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2269 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2270 pdo = pdu->hdr.common.pdo; 2271 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2272 pdo_error = true; 2273 break; 2274 } 2275 if (pdu->hdr.common.plen < expected_hlen) { 2276 plen_error = true; 2277 } 2278 break; 2279 2280 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2281 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2282 if ((pdu->hdr.common.plen <= expected_hlen) || 2283 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2284 plen_error = true; 2285 } 2286 break; 2287 2288 default: 2289 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2290 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2291 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2292 goto err; 2293 } 2294 } 2295 2296 if (pdu->hdr.common.hlen != expected_hlen) { 2297 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2298 pdu->hdr.common.pdu_type, 2299 expected_hlen, pdu->hdr.common.hlen, tqpair); 2300 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2301 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2302 goto err; 2303 } else if (pdo_error) { 2304 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2305 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2306 } else if (plen_error) { 2307 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2308 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2309 goto err; 2310 } else { 2311 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2312 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2313 return; 2314 } 2315 err: 2316 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2317 } 2318 2319 static int 2320 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2321 { 2322 int rc = 0; 2323 struct nvme_tcp_pdu *pdu; 2324 enum nvme_tcp_pdu_recv_state prev_state; 2325 uint32_t data_len; 2326 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2327 struct spdk_nvmf_tcp_transport, transport); 2328 2329 /* The loop here is to allow for several back-to-back state changes. */ 2330 do { 2331 prev_state = tqpair->recv_state; 2332 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2333 2334 pdu = tqpair->pdu_in_progress; 2335 assert(pdu != NULL || 2336 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2337 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2338 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2339 2340 switch (tqpair->recv_state) { 2341 /* Wait for the common header */ 2342 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2343 if (!pdu) { 2344 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2345 if (spdk_unlikely(!pdu)) { 2346 return NVME_TCP_PDU_IN_PROGRESS; 2347 } 2348 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2349 tqpair->pdu_in_progress = pdu; 2350 tqpair->tcp_pdu_working_count++; 2351 } 2352 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2353 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2354 /* FALLTHROUGH */ 2355 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2356 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2357 return rc; 2358 } 2359 2360 rc = nvme_tcp_read_data(tqpair->sock, 2361 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2362 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2363 if (rc < 0) { 2364 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2365 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2366 break; 2367 } else if (rc > 0) { 2368 pdu->ch_valid_bytes += rc; 2369 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2370 } 2371 2372 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2373 return NVME_TCP_PDU_IN_PROGRESS; 2374 } 2375 2376 /* The command header of this PDU has now been read from the socket. */ 2377 nvmf_tcp_pdu_ch_handle(tqpair); 2378 break; 2379 /* Wait for the pdu specific header */ 2380 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2381 rc = nvme_tcp_read_data(tqpair->sock, 2382 pdu->psh_len - pdu->psh_valid_bytes, 2383 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2384 if (rc < 0) { 2385 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2386 break; 2387 } else if (rc > 0) { 2388 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2389 pdu->psh_valid_bytes += rc; 2390 } 2391 2392 if (pdu->psh_valid_bytes < pdu->psh_len) { 2393 return NVME_TCP_PDU_IN_PROGRESS; 2394 } 2395 2396 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2397 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2398 break; 2399 /* Wait for the req slot */ 2400 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2401 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2402 break; 2403 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2404 /* check whether the data is valid, if not we just return */ 2405 if (!pdu->data_len) { 2406 return NVME_TCP_PDU_IN_PROGRESS; 2407 } 2408 2409 data_len = pdu->data_len; 2410 /* data digest */ 2411 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2412 tqpair->host_ddgst_enable)) { 2413 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2414 pdu->ddgst_enable = true; 2415 } 2416 2417 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2418 if (rc < 0) { 2419 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2420 break; 2421 } 2422 pdu->rw_offset += rc; 2423 2424 if (pdu->rw_offset < data_len) { 2425 return NVME_TCP_PDU_IN_PROGRESS; 2426 } 2427 2428 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2429 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2430 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2431 pdu->dif_ctx) != 0) { 2432 SPDK_ERRLOG("DIF generate failed\n"); 2433 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2434 break; 2435 } 2436 2437 /* All of this PDU has now been read from the socket. */ 2438 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2439 break; 2440 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2441 if (tqpair->tcp_pdu_working_count != 0) { 2442 return NVME_TCP_PDU_IN_PROGRESS; 2443 } 2444 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2445 break; 2446 case NVME_TCP_PDU_RECV_STATE_ERROR: 2447 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2448 return NVME_TCP_PDU_IN_PROGRESS; 2449 } 2450 return NVME_TCP_PDU_FATAL; 2451 default: 2452 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2453 abort(); 2454 break; 2455 } 2456 } while (tqpair->recv_state != prev_state); 2457 2458 return rc; 2459 } 2460 2461 static inline void * 2462 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2463 { 2464 struct spdk_nvmf_tcp_control_msg *msg; 2465 2466 assert(list); 2467 2468 msg = STAILQ_FIRST(&list->free_msgs); 2469 if (!msg) { 2470 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2471 return NULL; 2472 } 2473 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2474 return msg; 2475 } 2476 2477 static inline void 2478 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2479 { 2480 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2481 2482 assert(list); 2483 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2484 } 2485 2486 static int 2487 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2488 struct spdk_nvmf_transport *transport, 2489 struct spdk_nvmf_transport_poll_group *group) 2490 { 2491 struct spdk_nvmf_request *req = &tcp_req->req; 2492 struct spdk_nvme_cmd *cmd; 2493 struct spdk_nvme_sgl_descriptor *sgl; 2494 struct spdk_nvmf_tcp_poll_group *tgroup; 2495 enum spdk_nvme_tcp_term_req_fes fes; 2496 struct nvme_tcp_pdu *pdu; 2497 struct spdk_nvmf_tcp_qpair *tqpair; 2498 uint32_t length, error_offset = 0; 2499 2500 cmd = &req->cmd->nvme_cmd; 2501 sgl = &cmd->dptr.sgl1; 2502 2503 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2504 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2505 /* get request length from sgl */ 2506 length = sgl->unkeyed.length; 2507 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2508 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2509 length, transport->opts.max_io_size); 2510 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2511 goto fatal_err; 2512 } 2513 2514 /* fill request length and populate iovs */ 2515 req->length = length; 2516 2517 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2518 2519 if (spdk_unlikely(req->dif_enabled)) { 2520 req->dif.orig_length = length; 2521 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2522 req->dif.elba_length = length; 2523 } 2524 2525 if (nvmf_ctrlr_use_zcopy(req)) { 2526 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2527 req->data_from_pool = false; 2528 return 0; 2529 } 2530 2531 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2532 /* No available buffers. Queue this request up. */ 2533 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2534 tcp_req); 2535 return 0; 2536 } 2537 2538 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2539 tcp_req, req->iovcnt, req->iov[0].iov_base); 2540 2541 return 0; 2542 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2543 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2544 uint64_t offset = sgl->address; 2545 uint32_t max_len = transport->opts.in_capsule_data_size; 2546 2547 assert(tcp_req->has_in_capsule_data); 2548 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2549 tqpair = tcp_req->pdu->qpair; 2550 /* receiving pdu is not same with the pdu in tcp_req */ 2551 pdu = tqpair->pdu_in_progress; 2552 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2553 if (tqpair->host_ddgst_enable) { 2554 length -= SPDK_NVME_TCP_DIGEST_LEN; 2555 } 2556 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2557 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2558 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2559 length, sgl->unkeyed.length); 2560 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2561 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2562 goto fatal_err; 2563 } 2564 2565 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2566 offset, length); 2567 2568 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2569 if (spdk_unlikely(offset != 0)) { 2570 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2571 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2572 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2573 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2574 goto fatal_err; 2575 } 2576 2577 if (spdk_unlikely(length > max_len)) { 2578 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2579 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2580 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2581 2582 /* Get a buffer from dedicated list */ 2583 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2584 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2585 assert(tgroup->control_msg_list); 2586 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2587 if (!req->iov[0].iov_base) { 2588 /* No available buffers. Queue this request up. */ 2589 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2590 return 0; 2591 } 2592 } else { 2593 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2594 length, max_len); 2595 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2596 goto fatal_err; 2597 } 2598 } else { 2599 req->iov[0].iov_base = tcp_req->buf; 2600 } 2601 2602 req->length = length; 2603 req->data_from_pool = false; 2604 2605 if (spdk_unlikely(req->dif_enabled)) { 2606 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2607 req->dif.elba_length = length; 2608 } 2609 2610 req->iov[0].iov_len = length; 2611 req->iovcnt = 1; 2612 2613 return 0; 2614 } 2615 /* If we want to handle the problem here, then we can't skip the following data segment. 2616 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2617 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2618 sgl->generic.type, sgl->generic.subtype); 2619 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2620 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2621 fatal_err: 2622 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2623 return -1; 2624 } 2625 2626 static inline enum spdk_nvme_media_error_status_code 2627 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2628 enum spdk_nvme_media_error_status_code result; 2629 2630 switch (err_type) 2631 { 2632 case SPDK_DIF_REFTAG_ERROR: 2633 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2634 break; 2635 case SPDK_DIF_APPTAG_ERROR: 2636 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2637 break; 2638 case SPDK_DIF_GUARD_ERROR: 2639 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2640 break; 2641 default: 2642 SPDK_UNREACHABLE(); 2643 break; 2644 } 2645 2646 return result; 2647 } 2648 2649 static void 2650 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2651 struct spdk_nvmf_tcp_req *tcp_req) 2652 { 2653 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2654 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2655 struct nvme_tcp_pdu *rsp_pdu; 2656 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2657 uint32_t plen, pdo, alignment; 2658 int rc; 2659 2660 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2661 2662 rsp_pdu = tcp_req->pdu; 2663 assert(rsp_pdu != NULL); 2664 2665 c2h_data = &rsp_pdu->hdr.c2h_data; 2666 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2667 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2668 2669 if (tqpair->host_hdgst_enable) { 2670 plen += SPDK_NVME_TCP_DIGEST_LEN; 2671 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2672 } 2673 2674 /* set the psh */ 2675 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2676 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2677 c2h_data->datao = tcp_req->pdu->rw_offset; 2678 2679 /* set the padding */ 2680 rsp_pdu->padding_len = 0; 2681 pdo = plen; 2682 if (tqpair->cpda) { 2683 alignment = (tqpair->cpda + 1) << 2; 2684 if (plen % alignment != 0) { 2685 pdo = (plen + alignment) / alignment * alignment; 2686 rsp_pdu->padding_len = pdo - plen; 2687 plen = pdo; 2688 } 2689 } 2690 2691 c2h_data->common.pdo = pdo; 2692 plen += c2h_data->datal; 2693 if (tqpair->host_ddgst_enable) { 2694 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2695 plen += SPDK_NVME_TCP_DIGEST_LEN; 2696 } 2697 2698 c2h_data->common.plen = plen; 2699 2700 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2701 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2702 } 2703 2704 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2705 c2h_data->datao, c2h_data->datal); 2706 2707 2708 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2709 /* Need to send the capsule response if response is not all 0 */ 2710 if (ttransport->tcp_opts.c2h_success && 2711 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2712 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2713 } 2714 2715 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2716 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2717 struct spdk_dif_error err_blk = {}; 2718 uint32_t mapped_length = 0; 2719 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2720 uint32_t ddgst_len = 0; 2721 2722 if (tqpair->host_ddgst_enable) { 2723 /* Data digest consumes additional iov entry */ 2724 available_iovs--; 2725 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2726 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2727 c2h_data->common.plen -= ddgst_len; 2728 } 2729 /* Temp call to estimate if data can be described by limited number of iovs. 2730 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2731 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2732 false, &mapped_length); 2733 2734 if (mapped_length != c2h_data->common.plen) { 2735 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2736 SPDK_DEBUGLOG(nvmf_tcp, 2737 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2738 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2739 c2h_data->common.plen = mapped_length; 2740 2741 /* Rebuild pdu->data_iov since data length is changed */ 2742 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2743 c2h_data->datal); 2744 2745 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2746 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2747 } 2748 2749 c2h_data->common.plen += ddgst_len; 2750 2751 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2752 2753 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2754 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2755 if (rc != 0) { 2756 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2757 err_blk.err_type, err_blk.err_offset); 2758 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2759 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2760 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2761 return; 2762 } 2763 } 2764 2765 rsp_pdu->rw_offset += c2h_data->datal; 2766 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2767 } 2768 2769 static void 2770 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2771 struct spdk_nvmf_tcp_req *tcp_req) 2772 { 2773 nvmf_tcp_req_pdu_init(tcp_req); 2774 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2775 } 2776 2777 static int 2778 request_transfer_out(struct spdk_nvmf_request *req) 2779 { 2780 struct spdk_nvmf_tcp_req *tcp_req; 2781 struct spdk_nvmf_qpair *qpair; 2782 struct spdk_nvmf_tcp_qpair *tqpair; 2783 struct spdk_nvme_cpl *rsp; 2784 2785 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2786 2787 qpair = req->qpair; 2788 rsp = &req->rsp->nvme_cpl; 2789 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2790 2791 /* Advance our sq_head pointer */ 2792 if (qpair->sq_head == qpair->sq_head_max) { 2793 qpair->sq_head = 0; 2794 } else { 2795 qpair->sq_head++; 2796 } 2797 rsp->sqhd = qpair->sq_head; 2798 2799 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2800 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2801 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2802 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2803 } else { 2804 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2805 } 2806 2807 return 0; 2808 } 2809 2810 static void 2811 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2812 struct spdk_nvmf_tcp_qpair *tqpair, 2813 struct spdk_nvmf_tcp_req *tcp_req) 2814 { 2815 enum spdk_nvme_cmd_fuse last, next; 2816 2817 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2818 next = tcp_req->cmd.fuse; 2819 2820 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2821 2822 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2823 return; 2824 } 2825 2826 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2827 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2828 /* This is a valid pair of fused commands. Point them at each other 2829 * so they can be submitted consecutively once ready to be executed. 2830 */ 2831 tqpair->fused_first->fused_pair = tcp_req; 2832 tcp_req->fused_pair = tqpair->fused_first; 2833 tqpair->fused_first = NULL; 2834 return; 2835 } else { 2836 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2837 tqpair->fused_first->fused_failed = true; 2838 2839 /* 2840 * If the last req is in READY_TO_EXECUTE state, then call 2841 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2842 */ 2843 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2844 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2845 } 2846 2847 tqpair->fused_first = NULL; 2848 } 2849 } 2850 2851 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2852 /* Set tqpair->fused_first here so that we know to check that the next request 2853 * is a SECOND (and to fail this one if it isn't). 2854 */ 2855 tqpair->fused_first = tcp_req; 2856 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2857 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2858 tcp_req->fused_failed = true; 2859 } 2860 } 2861 2862 static bool 2863 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2864 struct spdk_nvmf_tcp_req *tcp_req) 2865 { 2866 struct spdk_nvmf_tcp_qpair *tqpair; 2867 uint32_t plen; 2868 struct nvme_tcp_pdu *pdu; 2869 enum spdk_nvmf_tcp_req_state prev_state; 2870 bool progress = false; 2871 struct spdk_nvmf_transport *transport = &ttransport->transport; 2872 struct spdk_nvmf_transport_poll_group *group; 2873 struct spdk_nvmf_tcp_poll_group *tgroup; 2874 2875 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2876 group = &tqpair->group->group; 2877 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2878 2879 /* If the qpair is not active, we need to abort the outstanding requests. */ 2880 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 2881 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2882 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2883 } 2884 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2885 } 2886 2887 /* The loop here is to allow for several back-to-back state changes. */ 2888 do { 2889 prev_state = tcp_req->state; 2890 2891 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2892 tqpair); 2893 2894 switch (tcp_req->state) { 2895 case TCP_REQUEST_STATE_FREE: 2896 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2897 * to escape this state. */ 2898 break; 2899 case TCP_REQUEST_STATE_NEW: 2900 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 2901 tqpair->qpair.queue_depth); 2902 2903 /* copy the cmd from the receive pdu */ 2904 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2905 2906 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2907 tcp_req->req.dif_enabled = true; 2908 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2909 } 2910 2911 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2912 2913 /* The next state transition depends on the data transfer needs of this request. */ 2914 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2915 2916 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2917 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2918 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2919 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2920 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2921 break; 2922 } 2923 2924 /* If no data to transfer, ready to execute. */ 2925 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2926 /* Reset the tqpair receiving pdu state */ 2927 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2928 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2929 break; 2930 } 2931 2932 pdu = tqpair->pdu_in_progress; 2933 plen = pdu->hdr.common.hlen; 2934 if (tqpair->host_hdgst_enable) { 2935 plen += SPDK_NVME_TCP_DIGEST_LEN; 2936 } 2937 if (pdu->hdr.common.plen != plen) { 2938 tcp_req->has_in_capsule_data = true; 2939 } else { 2940 /* Data is transmitted by C2H PDUs */ 2941 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2942 } 2943 2944 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2945 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2946 break; 2947 case TCP_REQUEST_STATE_NEED_BUFFER: 2948 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 2949 (uintptr_t)tcp_req); 2950 2951 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2952 2953 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2954 SPDK_DEBUGLOG(nvmf_tcp, 2955 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2956 tcp_req, tqpair); 2957 /* This request needs to wait in line to obtain a buffer */ 2958 break; 2959 } 2960 2961 /* Try to get a data buffer */ 2962 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 2963 break; 2964 } 2965 2966 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2967 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2968 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2969 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2970 tcp_req->req.length = tcp_req->req.dif.elba_length; 2971 } 2972 2973 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2974 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2975 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2976 break; 2977 } 2978 2979 if (tcp_req->req.iovcnt < 1) { 2980 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2981 tcp_req, tqpair); 2982 /* No buffers available. */ 2983 break; 2984 } 2985 2986 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2987 2988 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2989 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2990 if (tcp_req->req.data_from_pool) { 2991 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2992 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2993 } else { 2994 struct nvme_tcp_pdu *pdu; 2995 2996 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2997 2998 pdu = tqpair->pdu_in_progress; 2999 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3000 tqpair); 3001 /* No need to send r2t, contained in the capsuled data */ 3002 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3003 0, tcp_req->req.length); 3004 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3005 } 3006 break; 3007 } 3008 3009 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3010 break; 3011 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3012 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3013 (uintptr_t)tcp_req); 3014 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3015 * to escape this state. */ 3016 break; 3017 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3018 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3019 (uintptr_t)tcp_req); 3020 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3021 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3022 tcp_req, tqpair); 3023 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3024 break; 3025 } 3026 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3027 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3028 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3029 } else { 3030 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3031 } 3032 break; 3033 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3034 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3035 (uintptr_t)tcp_req); 3036 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3037 break; 3038 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3039 3040 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3041 0, (uintptr_t)tcp_req); 3042 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3043 * to escape this state. */ 3044 break; 3045 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3046 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3047 (uintptr_t)tcp_req); 3048 3049 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3050 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3051 tcp_req->req.length = tcp_req->req.dif.elba_length; 3052 } 3053 3054 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3055 if (tcp_req->fused_failed) { 3056 /* This request failed FUSED semantics. Fail it immediately, without 3057 * even sending it to the target layer. 3058 */ 3059 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3060 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3061 break; 3062 } 3063 3064 if (tcp_req->fused_pair == NULL || 3065 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3066 /* This request is ready to execute, but either we don't know yet if it's 3067 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3068 * or the other request of this fused pair isn't ready to execute. So 3069 * break here and this request will get processed later either when the 3070 * other request is ready or we find that this request isn't valid. 3071 */ 3072 break; 3073 } 3074 } 3075 3076 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3077 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3078 /* If we get to this point, and this request is a fused command, we know that 3079 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3080 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3081 * request, and the other request of the fused pair, in the correct order. 3082 * Also clear the ->fused_pair pointers on both requests, since after this point 3083 * we no longer need to maintain the relationship between these two requests. 3084 */ 3085 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3086 assert(tcp_req->fused_pair != NULL); 3087 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3088 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3089 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3090 tcp_req->fused_pair->fused_pair = NULL; 3091 tcp_req->fused_pair = NULL; 3092 } 3093 spdk_nvmf_request_exec(&tcp_req->req); 3094 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3095 assert(tcp_req->fused_pair != NULL); 3096 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3097 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3098 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3099 tcp_req->fused_pair->fused_pair = NULL; 3100 tcp_req->fused_pair = NULL; 3101 } 3102 } else { 3103 /* For zero-copy, only requests with data coming from host to the 3104 * controller can end up here. */ 3105 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3106 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3107 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3108 } 3109 3110 break; 3111 case TCP_REQUEST_STATE_EXECUTING: 3112 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3113 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3114 * to escape this state. */ 3115 break; 3116 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3117 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3118 (uintptr_t)tcp_req); 3119 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3120 * to escape this state. */ 3121 break; 3122 case TCP_REQUEST_STATE_EXECUTED: 3123 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3124 3125 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3126 tcp_req->req.length = tcp_req->req.dif.orig_length; 3127 } 3128 3129 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3130 break; 3131 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3132 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3133 (uintptr_t)tcp_req); 3134 if (request_transfer_out(&tcp_req->req) != 0) { 3135 assert(0); /* No good way to handle this currently */ 3136 } 3137 break; 3138 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3139 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3140 0, (uintptr_t)tcp_req); 3141 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3142 * to escape this state. */ 3143 break; 3144 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3145 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3146 (uintptr_t)tcp_req); 3147 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3148 * to escape this state. */ 3149 break; 3150 case TCP_REQUEST_STATE_COMPLETED: 3151 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3152 tqpair->qpair.queue_depth); 3153 /* If there's an outstanding PDU sent to the host, the request is completed 3154 * due to the qpair being disconnected. We must delay the completion until 3155 * that write is done to avoid freeing the request twice. */ 3156 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3157 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3158 "write on req=%p\n", tcp_req); 3159 /* This can only happen for zcopy requests */ 3160 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3161 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3162 break; 3163 } 3164 3165 if (tcp_req->req.data_from_pool) { 3166 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3167 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3168 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3169 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3170 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3171 assert(tgroup->control_msg_list); 3172 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3173 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3174 tcp_req->req.iov[0].iov_base); 3175 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3176 /* If the request has an unreleased zcopy bdev_io, it's either a 3177 * read, a failed write, or the qpair is being disconnected */ 3178 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3179 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3180 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3181 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3182 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3183 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3184 break; 3185 } 3186 tcp_req->req.length = 0; 3187 tcp_req->req.iovcnt = 0; 3188 tcp_req->fused_failed = false; 3189 if (tcp_req->fused_pair) { 3190 /* This req was part of a valid fused pair, but failed before it got to 3191 * READ_TO_EXECUTE state. This means we need to fail the other request 3192 * in the pair, because it is no longer part of a valid pair. If the pair 3193 * already reached READY_TO_EXECUTE state, we need to kick it. 3194 */ 3195 tcp_req->fused_pair->fused_failed = true; 3196 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3197 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3198 } 3199 tcp_req->fused_pair = NULL; 3200 } 3201 3202 nvmf_tcp_req_put(tqpair, tcp_req); 3203 break; 3204 case TCP_REQUEST_NUM_STATES: 3205 default: 3206 assert(0); 3207 break; 3208 } 3209 3210 if (tcp_req->state != prev_state) { 3211 progress = true; 3212 } 3213 } while (tcp_req->state != prev_state); 3214 3215 return progress; 3216 } 3217 3218 static void 3219 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3220 { 3221 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3222 int rc; 3223 3224 assert(tqpair != NULL); 3225 rc = nvmf_tcp_sock_process(tqpair); 3226 3227 /* If there was a new socket error, disconnect */ 3228 if (rc < 0) { 3229 nvmf_tcp_qpair_disconnect(tqpair); 3230 } 3231 } 3232 3233 static int 3234 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3235 struct spdk_nvmf_qpair *qpair) 3236 { 3237 struct spdk_nvmf_tcp_poll_group *tgroup; 3238 struct spdk_nvmf_tcp_qpair *tqpair; 3239 int rc; 3240 3241 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3242 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3243 3244 rc = nvmf_tcp_qpair_sock_init(tqpair); 3245 if (rc != 0) { 3246 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3247 return -1; 3248 } 3249 3250 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3251 if (rc < 0) { 3252 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3253 return -1; 3254 } 3255 3256 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3257 if (rc < 0) { 3258 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3259 return -1; 3260 } 3261 3262 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3263 nvmf_tcp_sock_cb, tqpair); 3264 if (rc != 0) { 3265 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3266 spdk_strerror(errno), errno); 3267 return -1; 3268 } 3269 3270 tqpair->group = tgroup; 3271 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3272 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3273 3274 return 0; 3275 } 3276 3277 static int 3278 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3279 struct spdk_nvmf_qpair *qpair) 3280 { 3281 struct spdk_nvmf_tcp_poll_group *tgroup; 3282 struct spdk_nvmf_tcp_qpair *tqpair; 3283 int rc; 3284 3285 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3286 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3287 3288 assert(tqpair->group == tgroup); 3289 3290 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3291 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3292 /* Change the state to move the qpair from the await_req list to the main list 3293 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3294 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3295 } 3296 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3297 3298 /* Try to force out any pending writes */ 3299 spdk_sock_flush(tqpair->sock); 3300 3301 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3302 if (rc != 0) { 3303 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3304 spdk_strerror(errno), errno); 3305 } 3306 3307 return rc; 3308 } 3309 3310 static int 3311 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3312 { 3313 struct spdk_nvmf_tcp_transport *ttransport; 3314 struct spdk_nvmf_tcp_req *tcp_req; 3315 3316 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3317 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3318 3319 switch (tcp_req->state) { 3320 case TCP_REQUEST_STATE_EXECUTING: 3321 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3322 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3323 break; 3324 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3325 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3326 break; 3327 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3328 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3329 break; 3330 default: 3331 assert(0 && "Unexpected request state"); 3332 break; 3333 } 3334 3335 nvmf_tcp_req_process(ttransport, tcp_req); 3336 3337 return 0; 3338 } 3339 3340 static void 3341 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3342 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3343 { 3344 struct spdk_nvmf_tcp_qpair *tqpair; 3345 3346 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3347 3348 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3349 3350 assert(tqpair->fini_cb_fn == NULL); 3351 tqpair->fini_cb_fn = cb_fn; 3352 tqpair->fini_cb_arg = cb_arg; 3353 3354 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3355 nvmf_tcp_qpair_destroy(tqpair); 3356 } 3357 3358 static int 3359 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3360 { 3361 struct spdk_nvmf_tcp_poll_group *tgroup; 3362 int rc; 3363 struct spdk_nvmf_request *req, *req_tmp; 3364 struct spdk_nvmf_tcp_req *tcp_req; 3365 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3366 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3367 struct spdk_nvmf_tcp_transport, transport); 3368 3369 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3370 3371 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3372 return 0; 3373 } 3374 3375 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3376 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3377 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3378 break; 3379 } 3380 } 3381 3382 rc = spdk_sock_group_poll(tgroup->sock_group); 3383 if (rc < 0) { 3384 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3385 } 3386 3387 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3388 rc = nvmf_tcp_sock_process(tqpair); 3389 3390 /* If there was a new socket error, disconnect */ 3391 if (rc < 0) { 3392 nvmf_tcp_qpair_disconnect(tqpair); 3393 } 3394 } 3395 3396 return rc; 3397 } 3398 3399 static int 3400 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3401 struct spdk_nvme_transport_id *trid, bool peer) 3402 { 3403 struct spdk_nvmf_tcp_qpair *tqpair; 3404 uint16_t port; 3405 3406 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3407 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3408 3409 if (peer) { 3410 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3411 port = tqpair->initiator_port; 3412 } else { 3413 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3414 port = tqpair->target_port; 3415 } 3416 3417 if (spdk_sock_is_ipv4(tqpair->sock)) { 3418 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3419 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3420 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3421 } else { 3422 return -1; 3423 } 3424 3425 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3426 return 0; 3427 } 3428 3429 static int 3430 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3431 struct spdk_nvme_transport_id *trid) 3432 { 3433 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3434 } 3435 3436 static int 3437 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3438 struct spdk_nvme_transport_id *trid) 3439 { 3440 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3441 } 3442 3443 static int 3444 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3445 struct spdk_nvme_transport_id *trid) 3446 { 3447 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3448 } 3449 3450 static void 3451 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3452 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3453 { 3454 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3455 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3456 3457 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3458 } 3459 3460 static int 3461 _nvmf_tcp_qpair_abort_request(void *ctx) 3462 { 3463 struct spdk_nvmf_request *req = ctx; 3464 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3465 struct spdk_nvmf_tcp_req, req); 3466 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3467 struct spdk_nvmf_tcp_qpair, qpair); 3468 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3469 struct spdk_nvmf_tcp_transport, transport); 3470 int rc; 3471 3472 spdk_poller_unregister(&req->poller); 3473 3474 switch (tcp_req_to_abort->state) { 3475 case TCP_REQUEST_STATE_EXECUTING: 3476 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3477 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3478 rc = nvmf_ctrlr_abort_request(req); 3479 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3480 return SPDK_POLLER_BUSY; 3481 } 3482 break; 3483 3484 case TCP_REQUEST_STATE_NEED_BUFFER: 3485 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3486 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3487 3488 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3489 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3490 break; 3491 3492 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3493 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3494 if (spdk_get_ticks() < req->timeout_tsc) { 3495 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3496 return SPDK_POLLER_BUSY; 3497 } 3498 break; 3499 3500 default: 3501 /* Requests in other states are either un-abortable (e.g. 3502 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3503 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3504 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3505 * abort request with the bit0 of dword0 set (command not aborted). 3506 */ 3507 break; 3508 } 3509 3510 spdk_nvmf_request_complete(req); 3511 return SPDK_POLLER_BUSY; 3512 } 3513 3514 static void 3515 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3516 struct spdk_nvmf_request *req) 3517 { 3518 struct spdk_nvmf_tcp_qpair *tqpair; 3519 struct spdk_nvmf_tcp_transport *ttransport; 3520 struct spdk_nvmf_transport *transport; 3521 uint16_t cid; 3522 uint32_t i; 3523 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3524 3525 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3526 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3527 transport = &ttransport->transport; 3528 3529 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3530 3531 for (i = 0; i < tqpair->resource_count; i++) { 3532 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3533 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3534 tcp_req_to_abort = &tqpair->reqs[i]; 3535 break; 3536 } 3537 } 3538 3539 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3540 3541 if (tcp_req_to_abort == NULL) { 3542 spdk_nvmf_request_complete(req); 3543 return; 3544 } 3545 3546 req->req_to_abort = &tcp_req_to_abort->req; 3547 req->timeout_tsc = spdk_get_ticks() + 3548 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3549 req->poller = NULL; 3550 3551 _nvmf_tcp_qpair_abort_request(req); 3552 } 3553 3554 struct tcp_subsystem_add_host_opts { 3555 char *psk; 3556 }; 3557 3558 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3559 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3560 }; 3561 3562 static int 3563 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3564 { 3565 FILE *psk_file; 3566 struct stat statbuf; 3567 int rc; 3568 3569 if (stat(fname, &statbuf) != 0) { 3570 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3571 return -EACCES; 3572 } 3573 3574 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3575 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3576 return -EPERM; 3577 } 3578 if ((size_t)statbuf.st_size > bufsz) { 3579 SPDK_ERRLOG("Invalid PSK: too long\n"); 3580 return -EINVAL; 3581 } 3582 psk_file = fopen(fname, "r"); 3583 if (psk_file == NULL) { 3584 SPDK_ERRLOG("Could not open PSK file\n"); 3585 return -EINVAL; 3586 } 3587 3588 rc = fread(buf, 1, statbuf.st_size, psk_file); 3589 if (rc != statbuf.st_size) { 3590 SPDK_ERRLOG("Failed to read PSK\n"); 3591 fclose(psk_file); 3592 return -EINVAL; 3593 } 3594 3595 fclose(psk_file); 3596 return 0; 3597 } 3598 3599 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3600 3601 static int 3602 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3603 const struct spdk_nvmf_subsystem *subsystem, 3604 const char *hostnqn, 3605 const struct spdk_json_val *transport_specific) 3606 { 3607 struct tcp_subsystem_add_host_opts opts; 3608 struct spdk_nvmf_tcp_transport *ttransport; 3609 struct tcp_psk_entry *tmp, *entry = NULL; 3610 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3611 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3612 uint8_t tls_cipher_suite; 3613 int rc = 0; 3614 uint8_t psk_retained_hash; 3615 uint64_t psk_configured_size; 3616 3617 if (transport_specific == NULL) { 3618 return 0; 3619 } 3620 3621 assert(transport != NULL); 3622 assert(subsystem != NULL); 3623 3624 memset(&opts, 0, sizeof(opts)); 3625 3626 /* Decode PSK (either name of a key or file path) */ 3627 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3628 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3629 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3630 return -EINVAL; 3631 } 3632 3633 if (opts.psk == NULL) { 3634 return 0; 3635 } 3636 3637 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3638 if (entry == NULL) { 3639 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3640 rc = -ENOMEM; 3641 goto end; 3642 } 3643 3644 entry->key = spdk_keyring_get_key(opts.psk); 3645 if (entry->key != NULL) { 3646 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3647 if (rc < 0) { 3648 SPDK_ERRLOG("Failed to retreive PSK '%s'\n", opts.psk); 3649 rc = -EINVAL; 3650 goto end; 3651 } 3652 } else { 3653 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3654 SPDK_ERRLOG("PSK path too long\n"); 3655 rc = -EINVAL; 3656 goto end; 3657 } 3658 3659 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3660 if (rc) { 3661 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3662 goto end; 3663 } 3664 3665 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3666 } 3667 3668 /* Parse PSK interchange to get length of base64 encoded data. 3669 * This is then used to decide which cipher suite should be used 3670 * to generate PSK identity and TLS PSK later on. */ 3671 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3672 &psk_configured_size, &psk_retained_hash); 3673 if (rc < 0) { 3674 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3675 goto end; 3676 } 3677 3678 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3679 * This check also ensures that psk_configured_size is smaller than 3680 * psk_retained buffer size. */ 3681 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3682 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3683 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3684 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3685 } else { 3686 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3687 rc = -EINVAL; 3688 goto end; 3689 } 3690 3691 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3692 /* Generate PSK identity. */ 3693 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3694 subsystem->subnqn, tls_cipher_suite); 3695 if (rc) { 3696 rc = -EINVAL; 3697 goto end; 3698 } 3699 /* Check if PSK identity entry already exists. */ 3700 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3701 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3702 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3703 rc = -EEXIST; 3704 goto end; 3705 } 3706 } 3707 3708 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3709 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3710 rc = -EINVAL; 3711 goto end; 3712 } 3713 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3714 SPDK_ERRLOG("Could not write subnqn string!\n"); 3715 rc = -EINVAL; 3716 goto end; 3717 } 3718 3719 entry->tls_cipher_suite = tls_cipher_suite; 3720 3721 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3722 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3723 /* Psk configured is either 32 or 48 bytes long. */ 3724 memcpy(entry->psk, psk_configured, psk_configured_size); 3725 entry->psk_size = psk_configured_size; 3726 } else { 3727 /* Derive retained PSK. */ 3728 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3729 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3730 if (rc < 0) { 3731 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3732 goto end; 3733 } 3734 entry->psk_size = rc; 3735 } 3736 3737 if (entry->key == NULL) { 3738 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3739 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3740 SPDK_ERRLOG("Could not save PSK path!\n"); 3741 rc = -ENAMETOOLONG; 3742 goto end; 3743 } 3744 } 3745 3746 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3747 rc = 0; 3748 3749 end: 3750 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3751 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3752 3753 free(opts.psk); 3754 if (rc != 0) { 3755 nvmf_tcp_free_psk_entry(entry); 3756 } 3757 3758 return rc; 3759 } 3760 3761 static void 3762 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3763 const struct spdk_nvmf_subsystem *subsystem, 3764 const char *hostnqn) 3765 { 3766 struct spdk_nvmf_tcp_transport *ttransport; 3767 struct tcp_psk_entry *entry, *tmp; 3768 3769 assert(transport != NULL); 3770 assert(subsystem != NULL); 3771 3772 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3773 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3774 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3775 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3776 TAILQ_REMOVE(&ttransport->psks, entry, link); 3777 nvmf_tcp_free_psk_entry(entry); 3778 break; 3779 } 3780 } 3781 } 3782 3783 static void 3784 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3785 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3786 struct spdk_json_write_ctx *w) 3787 { 3788 struct spdk_nvmf_tcp_transport *ttransport; 3789 struct tcp_psk_entry *entry; 3790 3791 assert(transport != NULL); 3792 assert(subsystem != NULL); 3793 3794 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3795 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3796 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3797 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3798 spdk_json_write_named_string(w, "psk", entry->key ? 3799 spdk_key_get_name(entry->key) : entry->psk_path); 3800 break; 3801 } 3802 } 3803 } 3804 3805 static void 3806 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3807 { 3808 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3809 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3810 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3811 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3812 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3813 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3814 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3815 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3816 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3817 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3818 opts->transport_specific = NULL; 3819 } 3820 3821 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3822 .name = "TCP", 3823 .type = SPDK_NVME_TRANSPORT_TCP, 3824 .opts_init = nvmf_tcp_opts_init, 3825 .create = nvmf_tcp_create, 3826 .dump_opts = nvmf_tcp_dump_opts, 3827 .destroy = nvmf_tcp_destroy, 3828 3829 .listen = nvmf_tcp_listen, 3830 .stop_listen = nvmf_tcp_stop_listen, 3831 3832 .listener_discover = nvmf_tcp_discover, 3833 3834 .poll_group_create = nvmf_tcp_poll_group_create, 3835 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3836 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3837 .poll_group_add = nvmf_tcp_poll_group_add, 3838 .poll_group_remove = nvmf_tcp_poll_group_remove, 3839 .poll_group_poll = nvmf_tcp_poll_group_poll, 3840 3841 .req_free = nvmf_tcp_req_free, 3842 .req_complete = nvmf_tcp_req_complete, 3843 3844 .qpair_fini = nvmf_tcp_close_qpair, 3845 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3846 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3847 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3848 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3849 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3850 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3851 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3852 }; 3853 3854 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3855 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3856