1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 #include "spdk/keyring.h" 19 20 #include "spdk_internal/assert.h" 21 #include "spdk_internal/nvme_tcp.h" 22 #include "spdk_internal/sock.h" 23 24 #include "nvmf_internal.h" 25 26 #include "spdk_internal/trace_defs.h" 27 28 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 29 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 30 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 31 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 32 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 33 34 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 35 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 36 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 37 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 38 39 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 42 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 43 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 44 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 46 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 47 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 48 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 49 50 #define TCP_PSK_INVALID_PERMISSIONS 0177 51 52 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 53 static bool g_tls_log = false; 54 55 /* spdk nvmf related structure */ 56 enum spdk_nvmf_tcp_req_state { 57 58 /* The request is not currently in use */ 59 TCP_REQUEST_STATE_FREE = 0, 60 61 /* Initial state when request first received */ 62 TCP_REQUEST_STATE_NEW = 1, 63 64 /* The request is queued until a data buffer is available. */ 65 TCP_REQUEST_STATE_NEED_BUFFER = 2, 66 67 /* The request is waiting for zcopy_start to finish */ 68 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 3, 69 70 /* The request has received a zero-copy buffer */ 71 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 4, 72 73 /* The request is currently transferring data from the host to the controller. */ 74 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 5, 75 76 /* The request is waiting for the R2T send acknowledgement. */ 77 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 6, 78 79 /* The request is ready to execute at the block device */ 80 TCP_REQUEST_STATE_READY_TO_EXECUTE = 7, 81 82 /* The request is currently executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTING = 8, 84 85 /* The request is waiting for zcopy buffers to be committed */ 86 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 9, 87 88 /* The request finished executing at the block device */ 89 TCP_REQUEST_STATE_EXECUTED = 10, 90 91 /* The request is ready to send a completion */ 92 TCP_REQUEST_STATE_READY_TO_COMPLETE = 11, 93 94 /* The request is currently transferring final pdus from the controller to the host. */ 95 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 12, 96 97 /* The request is waiting for zcopy buffers to be released (without committing) */ 98 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 13, 99 100 /* The request completed and can be marked free. */ 101 TCP_REQUEST_STATE_COMPLETED = 14, 102 103 /* Terminator */ 104 TCP_REQUEST_NUM_STATES, 105 }; 106 107 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 108 "Invalid PDU Header Field", 109 "PDU Sequence Error", 110 "Header Digiest Error", 111 "Data Transfer Out of Range", 112 "R2T Limit Exceeded", 113 "Unsupported parameter", 114 }; 115 116 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 117 { 118 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 119 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 120 spdk_trace_register_description("TCP_REQ_NEW", 121 TRACE_TCP_REQUEST_STATE_NEW, 122 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 123 SPDK_TRACE_ARG_TYPE_INT, "qd"); 124 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 125 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 126 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 127 SPDK_TRACE_ARG_TYPE_INT, ""); 128 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 129 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 130 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 131 SPDK_TRACE_ARG_TYPE_INT, ""); 132 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 133 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 134 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 135 SPDK_TRACE_ARG_TYPE_INT, ""); 136 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 137 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 138 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 139 SPDK_TRACE_ARG_TYPE_INT, ""); 140 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 141 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 142 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 143 SPDK_TRACE_ARG_TYPE_INT, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTING", 145 TRACE_TCP_REQUEST_STATE_EXECUTING, 146 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 147 SPDK_TRACE_ARG_TYPE_INT, ""); 148 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 149 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 150 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 151 SPDK_TRACE_ARG_TYPE_INT, ""); 152 spdk_trace_register_description("TCP_REQ_EXECUTED", 153 TRACE_TCP_REQUEST_STATE_EXECUTED, 154 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 155 SPDK_TRACE_ARG_TYPE_INT, ""); 156 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 157 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 158 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 159 SPDK_TRACE_ARG_TYPE_INT, ""); 160 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 161 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 162 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 163 SPDK_TRACE_ARG_TYPE_INT, ""); 164 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 165 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 166 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 167 SPDK_TRACE_ARG_TYPE_INT, ""); 168 spdk_trace_register_description("TCP_REQ_COMPLETED", 169 TRACE_TCP_REQUEST_STATE_COMPLETED, 170 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 171 SPDK_TRACE_ARG_TYPE_INT, "qd"); 172 spdk_trace_register_description("TCP_READ_DONE", 173 TRACE_TCP_READ_FROM_SOCKET_DONE, 174 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 175 SPDK_TRACE_ARG_TYPE_INT, ""); 176 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 177 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 178 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 179 SPDK_TRACE_ARG_TYPE_INT, ""); 180 181 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 182 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 183 SPDK_TRACE_ARG_TYPE_INT, ""); 184 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 185 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 186 SPDK_TRACE_ARG_TYPE_INT, ""); 187 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 188 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 189 SPDK_TRACE_ARG_TYPE_INT, "state"); 190 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 191 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 192 SPDK_TRACE_ARG_TYPE_INT, ""); 193 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 194 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 195 SPDK_TRACE_ARG_TYPE_INT, ""); 196 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 197 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 198 SPDK_TRACE_ARG_TYPE_INT, ""); 199 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 200 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 201 SPDK_TRACE_ARG_TYPE_INT, "state"); 202 203 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 204 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 205 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 206 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 207 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 208 } 209 210 struct spdk_nvmf_tcp_req { 211 struct spdk_nvmf_request req; 212 struct spdk_nvme_cpl rsp; 213 struct spdk_nvme_cmd cmd; 214 215 /* A PDU that can be used for sending responses. This is 216 * not the incoming PDU! */ 217 struct nvme_tcp_pdu *pdu; 218 219 /* In-capsule data buffer */ 220 uint8_t *buf; 221 222 struct spdk_nvmf_tcp_req *fused_pair; 223 224 /* 225 * The PDU for a request may be used multiple times in serial over 226 * the request's lifetime. For example, first to send an R2T, then 227 * to send a completion. To catch mistakes where the PDU is used 228 * twice at the same time, add a debug flag here for init/fini. 229 */ 230 bool pdu_in_use; 231 bool has_in_capsule_data; 232 bool fused_failed; 233 234 /* transfer_tag */ 235 uint16_t ttag; 236 237 enum spdk_nvmf_tcp_req_state state; 238 239 /* 240 * h2c_offset is used when we receive the h2c_data PDU. 241 */ 242 uint32_t h2c_offset; 243 244 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 245 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 246 }; 247 248 struct spdk_nvmf_tcp_qpair { 249 struct spdk_nvmf_qpair qpair; 250 struct spdk_nvmf_tcp_poll_group *group; 251 struct spdk_sock *sock; 252 253 enum nvme_tcp_pdu_recv_state recv_state; 254 enum nvme_tcp_qpair_state state; 255 256 /* PDU being actively received */ 257 struct nvme_tcp_pdu *pdu_in_progress; 258 259 struct spdk_nvmf_tcp_req *fused_first; 260 261 /* Queues to track the requests in all states */ 262 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 263 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 264 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 265 /* Number of working pdus */ 266 uint32_t tcp_pdu_working_count; 267 268 /* Number of requests in each state */ 269 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 270 271 uint8_t cpda; 272 273 bool host_hdgst_enable; 274 bool host_ddgst_enable; 275 276 /* This is a spare PDU used for sending special management 277 * operations. Primarily, this is used for the initial 278 * connection response and c2h termination request. */ 279 struct nvme_tcp_pdu *mgmt_pdu; 280 281 /* Arrays of in-capsule buffers, requests, and pdus. 282 * Each array is 'resource_count' number of elements */ 283 void *bufs; 284 struct spdk_nvmf_tcp_req *reqs; 285 struct nvme_tcp_pdu *pdus; 286 uint32_t resource_count; 287 uint32_t recv_buf_size; 288 289 struct spdk_nvmf_tcp_port *port; 290 291 /* IP address */ 292 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 293 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 294 295 /* IP port */ 296 uint16_t initiator_port; 297 uint16_t target_port; 298 299 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 300 bool wait_terminate; 301 302 /* Timer used to destroy qpair after detecting transport error issue if initiator does 303 * not close the connection. 304 */ 305 struct spdk_poller *timeout_poller; 306 307 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 308 void *fini_cb_arg; 309 310 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 311 }; 312 313 struct spdk_nvmf_tcp_control_msg { 314 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 315 }; 316 317 struct spdk_nvmf_tcp_control_msg_list { 318 void *msg_buf; 319 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 320 }; 321 322 struct spdk_nvmf_tcp_poll_group { 323 struct spdk_nvmf_transport_poll_group group; 324 struct spdk_sock_group *sock_group; 325 326 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 327 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 328 329 struct spdk_io_channel *accel_channel; 330 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 331 332 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 333 }; 334 335 struct spdk_nvmf_tcp_port { 336 const struct spdk_nvme_transport_id *trid; 337 struct spdk_sock *listen_sock; 338 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 339 }; 340 341 struct tcp_transport_opts { 342 bool c2h_success; 343 uint16_t control_msg_num; 344 uint32_t sock_priority; 345 }; 346 347 struct tcp_psk_entry { 348 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 349 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 350 char pskid[NVMF_PSK_IDENTITY_LEN]; 351 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 352 struct spdk_key *key; 353 354 /* Original path saved to emit SPDK configuration via "save_config". */ 355 char psk_path[PATH_MAX]; 356 uint32_t psk_size; 357 enum nvme_tcp_cipher_suite tls_cipher_suite; 358 TAILQ_ENTRY(tcp_psk_entry) link; 359 }; 360 361 struct spdk_nvmf_tcp_transport { 362 struct spdk_nvmf_transport transport; 363 struct tcp_transport_opts tcp_opts; 364 uint32_t ack_timeout; 365 366 struct spdk_nvmf_tcp_poll_group *next_pg; 367 368 struct spdk_poller *accept_poller; 369 370 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 371 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 372 373 TAILQ_HEAD(, tcp_psk_entry) psks; 374 }; 375 376 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 377 { 378 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 379 spdk_json_decode_bool, true 380 }, 381 { 382 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 383 spdk_json_decode_uint16, true 384 }, 385 { 386 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 387 spdk_json_decode_uint32, true 388 }, 389 }; 390 391 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 392 struct spdk_nvmf_tcp_req *tcp_req); 393 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 394 395 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 396 struct spdk_nvmf_tcp_req *tcp_req); 397 398 static inline void 399 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 400 enum spdk_nvmf_tcp_req_state state) 401 { 402 struct spdk_nvmf_qpair *qpair; 403 struct spdk_nvmf_tcp_qpair *tqpair; 404 405 qpair = tcp_req->req.qpair; 406 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 407 408 assert(tqpair->state_cntr[tcp_req->state] > 0); 409 tqpair->state_cntr[tcp_req->state]--; 410 tqpair->state_cntr[state]++; 411 412 tcp_req->state = state; 413 } 414 415 static inline struct nvme_tcp_pdu * 416 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 417 { 418 assert(tcp_req->pdu_in_use == false); 419 420 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 421 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 422 423 return tcp_req->pdu; 424 } 425 426 static struct spdk_nvmf_tcp_req * 427 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 428 { 429 struct spdk_nvmf_tcp_req *tcp_req; 430 431 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 432 if (spdk_unlikely(!tcp_req)) { 433 return NULL; 434 } 435 436 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 437 tcp_req->h2c_offset = 0; 438 tcp_req->has_in_capsule_data = false; 439 tcp_req->req.dif_enabled = false; 440 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 441 442 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 443 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 444 tqpair->qpair.queue_depth++; 445 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 446 return tcp_req; 447 } 448 449 static inline void 450 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 451 { 452 assert(!tcp_req->pdu_in_use); 453 454 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 455 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 456 tqpair->qpair.queue_depth--; 457 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 458 } 459 460 static void 461 nvmf_tcp_request_free(void *cb_arg) 462 { 463 struct spdk_nvmf_tcp_transport *ttransport; 464 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 465 466 assert(tcp_req != NULL); 467 468 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 469 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 470 struct spdk_nvmf_tcp_transport, transport); 471 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 472 nvmf_tcp_req_process(ttransport, tcp_req); 473 } 474 475 static int 476 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 477 { 478 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 479 480 nvmf_tcp_request_free(tcp_req); 481 482 return 0; 483 } 484 485 static void 486 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 487 enum spdk_nvmf_tcp_req_state state) 488 { 489 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 490 491 assert(state != TCP_REQUEST_STATE_FREE); 492 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 493 if (state == tcp_req->state) { 494 nvmf_tcp_request_free(tcp_req); 495 } 496 } 497 } 498 499 static void 500 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 501 { 502 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 503 504 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 505 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 506 507 /* Wipe the requests waiting for buffer from the global list */ 508 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 509 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 510 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 511 spdk_nvmf_request, buf_link); 512 } 513 } 514 515 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 516 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 517 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 518 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 519 } 520 521 static void 522 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 523 { 524 int i; 525 struct spdk_nvmf_tcp_req *tcp_req; 526 527 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 528 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 529 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 530 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 531 if ((int)tcp_req->state == i) { 532 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 533 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 534 } 535 } 536 } 537 } 538 539 static void 540 _nvmf_tcp_qpair_destroy(void *_tqpair) 541 { 542 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 543 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 544 void *cb_arg = tqpair->fini_cb_arg; 545 int err = 0; 546 547 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 548 549 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 550 551 err = spdk_sock_close(&tqpair->sock); 552 assert(err == 0); 553 nvmf_tcp_cleanup_all_states(tqpair); 554 555 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 556 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 557 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 558 tqpair->resource_count); 559 err++; 560 } 561 562 if (err > 0) { 563 nvmf_tcp_dump_qpair_req_contents(tqpair); 564 } 565 566 /* The timeout poller might still be registered here if we close the qpair before host 567 * terminates the connection. 568 */ 569 spdk_poller_unregister(&tqpair->timeout_poller); 570 spdk_dma_free(tqpair->pdus); 571 free(tqpair->reqs); 572 spdk_free(tqpair->bufs); 573 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 574 free(tqpair); 575 576 if (cb_fn != NULL) { 577 cb_fn(cb_arg); 578 } 579 580 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 581 } 582 583 static void 584 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 585 { 586 /* Delay the destruction to make sure it isn't performed from the context of a sock 587 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 588 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 589 */ 590 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 591 } 592 593 static void 594 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 595 { 596 struct spdk_nvmf_tcp_transport *ttransport; 597 assert(w != NULL); 598 599 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 600 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 601 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 602 } 603 604 static void 605 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 606 { 607 if (entry == NULL) { 608 return; 609 } 610 611 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 612 spdk_keyring_put_key(entry->key); 613 free(entry); 614 } 615 616 static int 617 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 618 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 619 { 620 struct spdk_nvmf_tcp_transport *ttransport; 621 struct tcp_psk_entry *entry, *tmp; 622 623 assert(transport != NULL); 624 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 625 626 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 627 TAILQ_REMOVE(&ttransport->psks, entry, link); 628 nvmf_tcp_free_psk_entry(entry); 629 } 630 631 spdk_poller_unregister(&ttransport->accept_poller); 632 free(ttransport); 633 634 if (cb_fn) { 635 cb_fn(cb_arg); 636 } 637 return 0; 638 } 639 640 static int nvmf_tcp_accept(void *ctx); 641 642 static struct spdk_nvmf_transport * 643 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 644 { 645 struct spdk_nvmf_tcp_transport *ttransport; 646 uint32_t sge_count; 647 uint32_t min_shared_buffers; 648 649 ttransport = calloc(1, sizeof(*ttransport)); 650 if (!ttransport) { 651 return NULL; 652 } 653 654 TAILQ_INIT(&ttransport->ports); 655 TAILQ_INIT(&ttransport->poll_groups); 656 TAILQ_INIT(&ttransport->psks); 657 658 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 659 660 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 661 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 662 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 663 if (opts->transport_specific != NULL && 664 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 665 SPDK_COUNTOF(tcp_transport_opts_decoder), 666 &ttransport->tcp_opts)) { 667 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 668 free(ttransport); 669 return NULL; 670 } 671 672 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 673 674 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 675 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 676 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 677 " in_capsule_data_size=%d, max_aq_depth=%d\n" 678 " num_shared_buffers=%d, c2h_success=%d,\n" 679 " dif_insert_or_strip=%d, sock_priority=%d\n" 680 " abort_timeout_sec=%d, control_msg_num=%hu\n" 681 " ack_timeout=%d\n", 682 opts->max_queue_depth, 683 opts->max_io_size, 684 opts->max_qpairs_per_ctrlr - 1, 685 opts->io_unit_size, 686 opts->in_capsule_data_size, 687 opts->max_aq_depth, 688 opts->num_shared_buffers, 689 ttransport->tcp_opts.c2h_success, 690 opts->dif_insert_or_strip, 691 ttransport->tcp_opts.sock_priority, 692 opts->abort_timeout_sec, 693 ttransport->tcp_opts.control_msg_num, 694 opts->ack_timeout); 695 696 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 697 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 698 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 699 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 700 free(ttransport); 701 return NULL; 702 } 703 704 if (ttransport->tcp_opts.control_msg_num == 0 && 705 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 706 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 707 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 708 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 709 } 710 711 /* I/O unit size cannot be larger than max I/O size */ 712 if (opts->io_unit_size > opts->max_io_size) { 713 SPDK_WARNLOG("TCP param io_unit_size %u can't be larger than max_io_size %u. Using max_io_size as io_unit_size\n", 714 opts->io_unit_size, opts->max_io_size); 715 opts->io_unit_size = opts->max_io_size; 716 } 717 718 /* In capsule data size cannot be larger than max I/O size */ 719 if (opts->in_capsule_data_size > opts->max_io_size) { 720 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 721 opts->io_unit_size, opts->max_io_size); 722 opts->in_capsule_data_size = opts->max_io_size; 723 } 724 725 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 726 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 727 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 728 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 729 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 730 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 731 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 732 } 733 734 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 735 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 736 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 737 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 738 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 739 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 740 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 741 } 742 743 sge_count = opts->max_io_size / opts->io_unit_size; 744 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 745 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 746 free(ttransport); 747 return NULL; 748 } 749 750 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 751 if (opts->buf_cache_size < UINT32_MAX) { 752 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 753 if (min_shared_buffers > opts->num_shared_buffers) { 754 SPDK_ERRLOG("There are not enough buffers to satisfy " 755 "per-poll group caches for each thread. (%" PRIu32 ") " 756 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 757 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 758 free(ttransport); 759 return NULL; 760 } 761 } 762 763 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 764 opts->acceptor_poll_rate); 765 if (!ttransport->accept_poller) { 766 free(ttransport); 767 return NULL; 768 } 769 770 return &ttransport->transport; 771 } 772 773 static int 774 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 775 { 776 unsigned long long ull; 777 char *end = NULL; 778 779 ull = strtoull(trsvcid, &end, 10); 780 if (end == NULL || end == trsvcid || *end != '\0') { 781 return -1; 782 } 783 784 /* Valid TCP/IP port numbers are in [1, 65535] */ 785 if (ull == 0 || ull > 65535) { 786 return -1; 787 } 788 789 return (int)ull; 790 } 791 792 /** 793 * Canonicalize a listen address trid. 794 */ 795 static int 796 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 797 const struct spdk_nvme_transport_id *trid) 798 { 799 int trsvcid_int; 800 801 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 802 if (trsvcid_int < 0) { 803 return -EINVAL; 804 } 805 806 memset(canon_trid, 0, sizeof(*canon_trid)); 807 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 808 canon_trid->adrfam = trid->adrfam; 809 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 810 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 811 812 return 0; 813 } 814 815 /** 816 * Find an existing listening port. 817 */ 818 static struct spdk_nvmf_tcp_port * 819 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 820 const struct spdk_nvme_transport_id *trid) 821 { 822 struct spdk_nvme_transport_id canon_trid; 823 struct spdk_nvmf_tcp_port *port; 824 825 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 826 return NULL; 827 } 828 829 TAILQ_FOREACH(port, &ttransport->ports, link) { 830 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 831 return port; 832 } 833 } 834 835 return NULL; 836 } 837 838 static int 839 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 840 void *get_key_ctx) 841 { 842 struct tcp_psk_entry *entry; 843 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 844 size_t psk_len; 845 int rc; 846 847 TAILQ_FOREACH(entry, &ttransport->psks, link) { 848 if (strcmp(pskid, entry->pskid) != 0) { 849 continue; 850 } 851 852 psk_len = entry->psk_size; 853 if ((size_t)out_len < psk_len) { 854 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 855 out_len, psk_len); 856 return -ENOBUFS; 857 } 858 859 /* Convert PSK to the TLS PSK format. */ 860 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 861 entry->tls_cipher_suite); 862 if (rc < 0) { 863 SPDK_ERRLOG("Could not generate TLS PSK\n"); 864 } 865 866 switch (entry->tls_cipher_suite) { 867 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 868 *cipher = "TLS_AES_128_GCM_SHA256"; 869 break; 870 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 871 *cipher = "TLS_AES_256_GCM_SHA384"; 872 break; 873 default: 874 *cipher = NULL; 875 return -ENOTSUP; 876 } 877 878 return rc; 879 } 880 881 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 882 883 return -EINVAL; 884 } 885 886 static int 887 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 888 struct spdk_nvmf_listen_opts *listen_opts) 889 { 890 struct spdk_nvmf_tcp_transport *ttransport; 891 struct spdk_nvmf_tcp_port *port; 892 int trsvcid_int; 893 uint8_t adrfam; 894 const char *sock_impl_name; 895 struct spdk_sock_impl_opts impl_opts; 896 size_t impl_opts_size = sizeof(impl_opts); 897 struct spdk_sock_opts opts; 898 899 if (!strlen(trid->trsvcid)) { 900 SPDK_ERRLOG("Service id is required\n"); 901 return -EINVAL; 902 } 903 904 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 905 906 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 907 if (trsvcid_int < 0) { 908 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 909 return -EINVAL; 910 } 911 912 port = calloc(1, sizeof(*port)); 913 if (!port) { 914 SPDK_ERRLOG("Port allocation failed\n"); 915 return -ENOMEM; 916 } 917 918 port->trid = trid; 919 920 sock_impl_name = NULL; 921 922 opts.opts_size = sizeof(opts); 923 spdk_sock_get_default_opts(&opts); 924 opts.priority = ttransport->tcp_opts.sock_priority; 925 opts.ack_timeout = transport->opts.ack_timeout; 926 if (listen_opts->secure_channel) { 927 if (listen_opts->sock_impl && 928 strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) { 929 SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported"); 930 free(port); 931 return -EINVAL; 932 } 933 listen_opts->sock_impl = "ssl"; 934 } 935 936 if (listen_opts->sock_impl) { 937 sock_impl_name = listen_opts->sock_impl; 938 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 939 940 if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) { 941 if (!g_tls_log) { 942 SPDK_NOTICELOG("TLS support is considered experimental\n"); 943 g_tls_log = true; 944 } 945 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 946 impl_opts.get_key = tcp_sock_get_key; 947 impl_opts.get_key_ctx = ttransport; 948 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 949 } 950 951 opts.impl_opts = &impl_opts; 952 opts.impl_opts_size = sizeof(impl_opts); 953 } 954 955 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 956 sock_impl_name, &opts); 957 if (port->listen_sock == NULL) { 958 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 959 trid->traddr, trsvcid_int, 960 spdk_strerror(errno), errno); 961 free(port); 962 return -errno; 963 } 964 965 if (spdk_sock_is_ipv4(port->listen_sock)) { 966 adrfam = SPDK_NVMF_ADRFAM_IPV4; 967 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 968 adrfam = SPDK_NVMF_ADRFAM_IPV6; 969 } else { 970 SPDK_ERRLOG("Unhandled socket type\n"); 971 adrfam = 0; 972 } 973 974 if (adrfam != trid->adrfam) { 975 SPDK_ERRLOG("Socket address family mismatch\n"); 976 spdk_sock_close(&port->listen_sock); 977 free(port); 978 return -EINVAL; 979 } 980 981 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 982 trid->traddr, trid->trsvcid); 983 984 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 985 return 0; 986 } 987 988 static void 989 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 990 const struct spdk_nvme_transport_id *trid) 991 { 992 struct spdk_nvmf_tcp_transport *ttransport; 993 struct spdk_nvmf_tcp_port *port; 994 995 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 996 997 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 998 trid->traddr, trid->trsvcid); 999 1000 port = nvmf_tcp_find_port(ttransport, trid); 1001 if (port) { 1002 TAILQ_REMOVE(&ttransport->ports, port, link); 1003 spdk_sock_close(&port->listen_sock); 1004 free(port); 1005 } 1006 } 1007 1008 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1009 enum nvme_tcp_pdu_recv_state state); 1010 1011 static void 1012 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 1013 { 1014 tqpair->state = state; 1015 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1016 (uint64_t)tqpair->state); 1017 } 1018 1019 static void 1020 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1021 { 1022 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1023 1024 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1025 1026 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 1027 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 1028 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1029 spdk_poller_unregister(&tqpair->timeout_poller); 1030 1031 /* This will end up calling nvmf_tcp_close_qpair */ 1032 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1033 } 1034 } 1035 1036 static void 1037 _mgmt_pdu_write_done(void *_tqpair, int err) 1038 { 1039 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1040 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1041 1042 if (spdk_unlikely(err != 0)) { 1043 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1044 return; 1045 } 1046 1047 assert(pdu->cb_fn != NULL); 1048 pdu->cb_fn(pdu->cb_arg); 1049 } 1050 1051 static void 1052 _req_pdu_write_done(void *req, int err) 1053 { 1054 struct spdk_nvmf_tcp_req *tcp_req = req; 1055 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1056 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1057 1058 assert(tcp_req->pdu_in_use); 1059 tcp_req->pdu_in_use = false; 1060 1061 /* If the request is in a completed state, we're waiting for write completion to free it */ 1062 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1063 nvmf_tcp_request_free(tcp_req); 1064 return; 1065 } 1066 1067 if (spdk_unlikely(err != 0)) { 1068 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1069 return; 1070 } 1071 1072 assert(pdu->cb_fn != NULL); 1073 pdu->cb_fn(pdu->cb_arg); 1074 } 1075 1076 static void 1077 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1078 { 1079 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1080 } 1081 1082 static void 1083 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1084 { 1085 int rc; 1086 uint32_t mapped_length; 1087 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1088 1089 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1090 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1091 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1092 1093 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1094 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1095 /* Try to force the send immediately. */ 1096 rc = spdk_sock_flush(tqpair->sock); 1097 if (rc > 0 && (uint32_t)rc == mapped_length) { 1098 _pdu_write_done(pdu, 0); 1099 } else { 1100 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1101 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1102 "IC_RESP" : "TERM_REQ", rc, errno); 1103 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1104 } 1105 } 1106 } 1107 1108 static void 1109 data_crc32_accel_done(void *cb_arg, int status) 1110 { 1111 struct nvme_tcp_pdu *pdu = cb_arg; 1112 1113 if (spdk_unlikely(status)) { 1114 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1115 _pdu_write_done(pdu, status); 1116 return; 1117 } 1118 1119 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1120 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1121 1122 _tcp_write_pdu(pdu); 1123 } 1124 1125 static void 1126 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1127 { 1128 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1129 int rc = 0; 1130 1131 /* Data Digest */ 1132 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1133 /* Only support this limitated case for the first step */ 1134 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1135 && tqpair->group)) { 1136 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1137 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1138 if (spdk_likely(rc == 0)) { 1139 return; 1140 } 1141 } else { 1142 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1143 } 1144 data_crc32_accel_done(pdu, rc); 1145 } else { 1146 _tcp_write_pdu(pdu); 1147 } 1148 } 1149 1150 static void 1151 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1152 struct nvme_tcp_pdu *pdu, 1153 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1154 void *cb_arg) 1155 { 1156 int hlen; 1157 uint32_t crc32c; 1158 1159 assert(tqpair->pdu_in_progress != pdu); 1160 1161 hlen = pdu->hdr.common.hlen; 1162 pdu->cb_fn = cb_fn; 1163 pdu->cb_arg = cb_arg; 1164 1165 pdu->iov[0].iov_base = &pdu->hdr.raw; 1166 pdu->iov[0].iov_len = hlen; 1167 1168 /* Header Digest */ 1169 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1170 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1171 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1172 } 1173 1174 /* Data Digest */ 1175 pdu_data_crc32_compute(pdu); 1176 } 1177 1178 static void 1179 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1180 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1181 void *cb_arg) 1182 { 1183 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1184 1185 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1186 pdu->sock_req.cb_arg = tqpair; 1187 1188 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1189 } 1190 1191 static void 1192 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1193 struct spdk_nvmf_tcp_req *tcp_req, 1194 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1195 void *cb_arg) 1196 { 1197 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1198 1199 pdu->sock_req.cb_fn = _req_pdu_write_done; 1200 pdu->sock_req.cb_arg = tcp_req; 1201 1202 assert(!tcp_req->pdu_in_use); 1203 tcp_req->pdu_in_use = true; 1204 1205 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1206 } 1207 1208 static int 1209 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1210 { 1211 uint32_t i; 1212 struct spdk_nvmf_transport_opts *opts; 1213 uint32_t in_capsule_data_size; 1214 1215 opts = &tqpair->qpair.transport->opts; 1216 1217 in_capsule_data_size = opts->in_capsule_data_size; 1218 if (opts->dif_insert_or_strip) { 1219 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1220 } 1221 1222 tqpair->resource_count = opts->max_queue_depth; 1223 1224 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1225 if (!tqpair->reqs) { 1226 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1227 return -1; 1228 } 1229 1230 if (in_capsule_data_size) { 1231 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1232 NULL, SPDK_ENV_LCORE_ID_ANY, 1233 SPDK_MALLOC_DMA); 1234 if (!tqpair->bufs) { 1235 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1236 return -1; 1237 } 1238 } 1239 /* prepare memory space for receiving pdus and tcp_req */ 1240 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1241 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1242 NULL); 1243 if (!tqpair->pdus) { 1244 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1245 return -1; 1246 } 1247 1248 for (i = 0; i < tqpair->resource_count; i++) { 1249 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1250 1251 tcp_req->ttag = i + 1; 1252 tcp_req->req.qpair = &tqpair->qpair; 1253 1254 tcp_req->pdu = &tqpair->pdus[i]; 1255 tcp_req->pdu->qpair = tqpair; 1256 1257 /* Set up memory to receive commands */ 1258 if (tqpair->bufs) { 1259 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1260 } 1261 1262 /* Set the cmdn and rsp */ 1263 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1264 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1265 1266 tcp_req->req.stripped_data = NULL; 1267 1268 /* Initialize request state to FREE */ 1269 tcp_req->state = TCP_REQUEST_STATE_FREE; 1270 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1271 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1272 } 1273 1274 for (; i < 2 * tqpair->resource_count; i++) { 1275 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1276 1277 pdu->qpair = tqpair; 1278 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1279 } 1280 1281 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1282 tqpair->mgmt_pdu->qpair = tqpair; 1283 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1284 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1285 tqpair->tcp_pdu_working_count = 1; 1286 1287 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1288 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1289 1290 return 0; 1291 } 1292 1293 static int 1294 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1295 { 1296 struct spdk_nvmf_tcp_qpair *tqpair; 1297 1298 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1299 1300 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1301 1302 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1303 1304 /* Initialise request state queues of the qpair */ 1305 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1306 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1307 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1308 tqpair->qpair.queue_depth = 0; 1309 1310 tqpair->host_hdgst_enable = true; 1311 tqpair->host_ddgst_enable = true; 1312 1313 return 0; 1314 } 1315 1316 static int 1317 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1318 { 1319 char saddr[32], caddr[32]; 1320 uint16_t sport, cport; 1321 char owner[256]; 1322 int rc; 1323 1324 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1325 caddr, sizeof(caddr), &cport); 1326 if (rc != 0) { 1327 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1328 return rc; 1329 } 1330 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1331 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1332 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1333 1334 /* set low water mark */ 1335 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1336 if (rc != 0) { 1337 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1338 return rc; 1339 } 1340 1341 return 0; 1342 } 1343 1344 static void 1345 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 1346 struct spdk_nvmf_tcp_port *port, 1347 struct spdk_sock *sock) 1348 { 1349 struct spdk_nvmf_tcp_qpair *tqpair; 1350 int rc; 1351 1352 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1353 port->trid->traddr, port->trid->trsvcid); 1354 1355 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1356 if (tqpair == NULL) { 1357 SPDK_ERRLOG("Could not allocate new connection.\n"); 1358 spdk_sock_close(&sock); 1359 return; 1360 } 1361 1362 tqpair->sock = sock; 1363 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1364 tqpair->port = port; 1365 tqpair->qpair.transport = transport; 1366 1367 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1368 sizeof(tqpair->target_addr), &tqpair->target_port, 1369 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1370 &tqpair->initiator_port); 1371 if (rc < 0) { 1372 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1373 nvmf_tcp_qpair_destroy(tqpair); 1374 return; 1375 } 1376 1377 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1378 } 1379 1380 static uint32_t 1381 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1382 { 1383 struct spdk_sock *sock; 1384 uint32_t count = 0; 1385 int i; 1386 1387 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1388 sock = spdk_sock_accept(port->listen_sock); 1389 if (sock == NULL) { 1390 break; 1391 } 1392 count++; 1393 nvmf_tcp_handle_connect(transport, port, sock); 1394 } 1395 1396 return count; 1397 } 1398 1399 static int 1400 nvmf_tcp_accept(void *ctx) 1401 { 1402 struct spdk_nvmf_transport *transport = ctx; 1403 struct spdk_nvmf_tcp_transport *ttransport; 1404 struct spdk_nvmf_tcp_port *port; 1405 uint32_t count = 0; 1406 1407 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1408 1409 TAILQ_FOREACH(port, &ttransport->ports, link) { 1410 count += nvmf_tcp_port_accept(transport, port); 1411 } 1412 1413 return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1414 } 1415 1416 static void 1417 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1418 struct spdk_nvme_transport_id *trid, 1419 struct spdk_nvmf_discovery_log_page_entry *entry) 1420 { 1421 struct spdk_nvmf_tcp_port *port; 1422 struct spdk_nvmf_tcp_transport *ttransport; 1423 1424 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1425 entry->adrfam = trid->adrfam; 1426 1427 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1428 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1429 1430 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1431 port = nvmf_tcp_find_port(ttransport, trid); 1432 1433 assert(port != NULL); 1434 1435 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1436 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1437 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1438 } else { 1439 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1440 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1441 } 1442 } 1443 1444 static struct spdk_nvmf_tcp_control_msg_list * 1445 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1446 { 1447 struct spdk_nvmf_tcp_control_msg_list *list; 1448 struct spdk_nvmf_tcp_control_msg *msg; 1449 uint16_t i; 1450 1451 list = calloc(1, sizeof(*list)); 1452 if (!list) { 1453 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1454 return NULL; 1455 } 1456 1457 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1458 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1459 if (!list->msg_buf) { 1460 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1461 free(list); 1462 return NULL; 1463 } 1464 1465 STAILQ_INIT(&list->free_msgs); 1466 1467 for (i = 0; i < num_messages; i++) { 1468 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1469 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1470 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1471 } 1472 1473 return list; 1474 } 1475 1476 static void 1477 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1478 { 1479 if (!list) { 1480 return; 1481 } 1482 1483 spdk_free(list->msg_buf); 1484 free(list); 1485 } 1486 1487 static struct spdk_nvmf_transport_poll_group * 1488 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1489 struct spdk_nvmf_poll_group *group) 1490 { 1491 struct spdk_nvmf_tcp_transport *ttransport; 1492 struct spdk_nvmf_tcp_poll_group *tgroup; 1493 1494 if (spdk_interrupt_mode_is_enabled()) { 1495 SPDK_ERRLOG("TCP transport does not support interrupt mode\n"); 1496 return NULL; 1497 } 1498 1499 tgroup = calloc(1, sizeof(*tgroup)); 1500 if (!tgroup) { 1501 return NULL; 1502 } 1503 1504 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1505 if (!tgroup->sock_group) { 1506 goto cleanup; 1507 } 1508 1509 TAILQ_INIT(&tgroup->qpairs); 1510 TAILQ_INIT(&tgroup->await_req); 1511 1512 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1513 1514 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1515 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1516 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1517 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1518 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1519 if (!tgroup->control_msg_list) { 1520 goto cleanup; 1521 } 1522 } 1523 1524 tgroup->accel_channel = spdk_accel_get_io_channel(); 1525 if (spdk_unlikely(!tgroup->accel_channel)) { 1526 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1527 goto cleanup; 1528 } 1529 1530 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1531 if (ttransport->next_pg == NULL) { 1532 ttransport->next_pg = tgroup; 1533 } 1534 1535 return &tgroup->group; 1536 1537 cleanup: 1538 nvmf_tcp_poll_group_destroy(&tgroup->group); 1539 return NULL; 1540 } 1541 1542 static struct spdk_nvmf_transport_poll_group * 1543 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1544 { 1545 struct spdk_nvmf_tcp_transport *ttransport; 1546 struct spdk_nvmf_tcp_poll_group **pg; 1547 struct spdk_nvmf_tcp_qpair *tqpair; 1548 struct spdk_sock_group *group = NULL, *hint = NULL; 1549 int rc; 1550 1551 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1552 1553 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1554 return NULL; 1555 } 1556 1557 pg = &ttransport->next_pg; 1558 assert(*pg != NULL); 1559 hint = (*pg)->sock_group; 1560 1561 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1562 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1563 if (rc != 0) { 1564 return NULL; 1565 } else if (group != NULL) { 1566 /* Optimal poll group was found */ 1567 return spdk_sock_group_get_ctx(group); 1568 } 1569 1570 /* The hint was used for optimal poll group, advance next_pg. */ 1571 *pg = TAILQ_NEXT(*pg, link); 1572 if (*pg == NULL) { 1573 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1574 } 1575 1576 return spdk_sock_group_get_ctx(hint); 1577 } 1578 1579 static void 1580 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1581 { 1582 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1583 struct spdk_nvmf_tcp_transport *ttransport; 1584 1585 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1586 spdk_sock_group_close(&tgroup->sock_group); 1587 if (tgroup->control_msg_list) { 1588 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1589 } 1590 1591 if (tgroup->accel_channel) { 1592 spdk_put_io_channel(tgroup->accel_channel); 1593 } 1594 1595 if (tgroup->group.transport == NULL) { 1596 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1597 * calls this function directly in a failure path. */ 1598 free(tgroup); 1599 return; 1600 } 1601 1602 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1603 1604 next_tgroup = TAILQ_NEXT(tgroup, link); 1605 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1606 if (next_tgroup == NULL) { 1607 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1608 } 1609 if (ttransport->next_pg == tgroup) { 1610 ttransport->next_pg = next_tgroup; 1611 } 1612 1613 free(tgroup); 1614 } 1615 1616 static void 1617 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1618 enum nvme_tcp_pdu_recv_state state) 1619 { 1620 if (tqpair->recv_state == state) { 1621 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1622 tqpair, state); 1623 return; 1624 } 1625 1626 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1627 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1628 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1629 tqpair->tcp_pdu_working_count--; 1630 } 1631 } 1632 1633 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1634 assert(tqpair->tcp_pdu_working_count == 0); 1635 } 1636 1637 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1638 /* When leaving the await req state, move the qpair to the main list */ 1639 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1640 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1641 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1642 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1643 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1644 } 1645 1646 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1647 tqpair->recv_state = state; 1648 1649 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1650 (uint64_t)tqpair->recv_state); 1651 } 1652 1653 static int 1654 nvmf_tcp_qpair_handle_timeout(void *ctx) 1655 { 1656 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1657 1658 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1659 1660 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1661 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1662 1663 nvmf_tcp_qpair_disconnect(tqpair); 1664 return SPDK_POLLER_BUSY; 1665 } 1666 1667 static void 1668 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1669 { 1670 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1671 1672 if (!tqpair->timeout_poller) { 1673 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1674 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1675 } 1676 } 1677 1678 static void 1679 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1680 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1681 { 1682 struct nvme_tcp_pdu *rsp_pdu; 1683 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1684 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1685 uint32_t copy_len; 1686 1687 rsp_pdu = tqpair->mgmt_pdu; 1688 1689 c2h_term_req = &rsp_pdu->hdr.term_req; 1690 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1691 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1692 c2h_term_req->fes = fes; 1693 1694 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1695 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1696 DSET32(&c2h_term_req->fei, error_offset); 1697 } 1698 1699 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1700 1701 /* Copy the error info into the buffer */ 1702 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1703 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1704 1705 /* Contain the header of the wrong received pdu */ 1706 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1707 tqpair->wait_terminate = true; 1708 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1709 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1710 } 1711 1712 static void 1713 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1714 struct spdk_nvmf_tcp_qpair *tqpair, 1715 struct nvme_tcp_pdu *pdu) 1716 { 1717 struct spdk_nvmf_tcp_req *tcp_req; 1718 1719 assert(pdu->psh_valid_bytes == pdu->psh_len); 1720 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1721 1722 tcp_req = nvmf_tcp_req_get(tqpair); 1723 if (!tcp_req) { 1724 /* Directly return and make the allocation retry again. This can happen if we're 1725 * using asynchronous writes to send the response to the host or when releasing 1726 * zero-copy buffers after a response has been sent. In both cases, the host might 1727 * receive the response before we've finished processing the request and is free to 1728 * send another one. 1729 */ 1730 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1731 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1732 return; 1733 } 1734 1735 /* The host sent more commands than the maximum queue depth. */ 1736 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1737 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1738 return; 1739 } 1740 1741 pdu->req = tcp_req; 1742 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1743 nvmf_tcp_req_process(ttransport, tcp_req); 1744 } 1745 1746 static void 1747 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1748 struct spdk_nvmf_tcp_qpair *tqpair, 1749 struct nvme_tcp_pdu *pdu) 1750 { 1751 struct spdk_nvmf_tcp_req *tcp_req; 1752 struct spdk_nvme_tcp_cmd *capsule_cmd; 1753 uint32_t error_offset = 0; 1754 enum spdk_nvme_tcp_term_req_fes fes; 1755 struct spdk_nvme_cpl *rsp; 1756 1757 capsule_cmd = &pdu->hdr.capsule_cmd; 1758 tcp_req = pdu->req; 1759 assert(tcp_req != NULL); 1760 1761 /* Zero-copy requests don't support ICD */ 1762 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1763 1764 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1765 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1766 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1767 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1768 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1769 goto err; 1770 } 1771 1772 rsp = &tcp_req->req.rsp->nvme_cpl; 1773 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1774 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1775 } else { 1776 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1777 } 1778 1779 nvmf_tcp_req_process(ttransport, tcp_req); 1780 1781 return; 1782 err: 1783 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1784 } 1785 1786 static void 1787 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1788 struct spdk_nvmf_tcp_qpair *tqpair, 1789 struct nvme_tcp_pdu *pdu) 1790 { 1791 struct spdk_nvmf_tcp_req *tcp_req; 1792 uint32_t error_offset = 0; 1793 enum spdk_nvme_tcp_term_req_fes fes = 0; 1794 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1795 1796 h2c_data = &pdu->hdr.h2c_data; 1797 1798 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1799 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1800 1801 if (h2c_data->ttag > tqpair->resource_count) { 1802 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1803 tqpair->resource_count); 1804 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1805 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1806 goto err; 1807 } 1808 1809 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1810 1811 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1812 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1813 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1814 tcp_req->state); 1815 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1816 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1817 goto err; 1818 } 1819 1820 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1821 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1822 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1823 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1824 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1825 goto err; 1826 } 1827 1828 if (tcp_req->h2c_offset != h2c_data->datao) { 1829 SPDK_DEBUGLOG(nvmf_tcp, 1830 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1831 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1832 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1833 goto err; 1834 } 1835 1836 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1837 SPDK_DEBUGLOG(nvmf_tcp, 1838 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1839 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1840 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1841 goto err; 1842 } 1843 1844 pdu->req = tcp_req; 1845 1846 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1847 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1848 } 1849 1850 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1851 h2c_data->datao, h2c_data->datal); 1852 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1853 return; 1854 1855 err: 1856 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1857 } 1858 1859 static void 1860 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1861 struct spdk_nvmf_tcp_qpair *tqpair) 1862 { 1863 struct nvme_tcp_pdu *rsp_pdu; 1864 struct spdk_nvme_tcp_rsp *capsule_resp; 1865 1866 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1867 1868 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1869 assert(rsp_pdu != NULL); 1870 1871 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1872 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1873 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1874 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1875 if (tqpair->host_hdgst_enable) { 1876 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1877 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1878 } 1879 1880 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1881 } 1882 1883 static void 1884 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1885 { 1886 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1887 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1888 struct spdk_nvmf_tcp_qpair, qpair); 1889 1890 assert(tqpair != NULL); 1891 1892 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1893 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1894 tcp_req->req.length); 1895 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1896 return; 1897 } 1898 1899 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1900 nvmf_tcp_request_free(tcp_req); 1901 } else { 1902 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1903 } 1904 } 1905 1906 static void 1907 nvmf_tcp_r2t_complete(void *cb_arg) 1908 { 1909 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1910 struct spdk_nvmf_tcp_transport *ttransport; 1911 1912 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1913 struct spdk_nvmf_tcp_transport, transport); 1914 1915 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1916 1917 if (tcp_req->h2c_offset == tcp_req->req.length) { 1918 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1919 nvmf_tcp_req_process(ttransport, tcp_req); 1920 } 1921 } 1922 1923 static void 1924 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1925 struct spdk_nvmf_tcp_req *tcp_req) 1926 { 1927 struct nvme_tcp_pdu *rsp_pdu; 1928 struct spdk_nvme_tcp_r2t_hdr *r2t; 1929 1930 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1931 assert(rsp_pdu != NULL); 1932 1933 r2t = &rsp_pdu->hdr.r2t; 1934 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1935 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1936 1937 if (tqpair->host_hdgst_enable) { 1938 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1939 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1940 } 1941 1942 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1943 r2t->ttag = tcp_req->ttag; 1944 r2t->r2to = tcp_req->h2c_offset; 1945 r2t->r2tl = tcp_req->req.length; 1946 1947 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1948 1949 SPDK_DEBUGLOG(nvmf_tcp, 1950 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1951 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1952 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1953 } 1954 1955 static void 1956 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1957 struct spdk_nvmf_tcp_qpair *tqpair, 1958 struct nvme_tcp_pdu *pdu) 1959 { 1960 struct spdk_nvmf_tcp_req *tcp_req; 1961 struct spdk_nvme_cpl *rsp; 1962 1963 tcp_req = pdu->req; 1964 assert(tcp_req != NULL); 1965 1966 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1967 1968 tcp_req->h2c_offset += pdu->data_len; 1969 1970 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1971 * acknowledged before moving on. */ 1972 if (tcp_req->h2c_offset == tcp_req->req.length && 1973 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1974 /* After receiving all the h2c data, we need to check whether there is 1975 * transient transport error */ 1976 rsp = &tcp_req->req.rsp->nvme_cpl; 1977 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1978 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1979 } else { 1980 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1981 } 1982 nvmf_tcp_req_process(ttransport, tcp_req); 1983 } 1984 } 1985 1986 static void 1987 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1988 { 1989 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1990 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1991 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1992 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1993 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1994 DGET32(h2c_term_req->fei)); 1995 } 1996 } 1997 1998 static void 1999 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2000 struct nvme_tcp_pdu *pdu) 2001 { 2002 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2003 uint32_t error_offset = 0; 2004 enum spdk_nvme_tcp_term_req_fes fes; 2005 2006 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 2007 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 2008 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2009 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 2010 goto end; 2011 } 2012 2013 /* set the data buffer */ 2014 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2015 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2016 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2017 return; 2018 end: 2019 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2020 } 2021 2022 static void 2023 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2024 struct nvme_tcp_pdu *pdu) 2025 { 2026 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2027 2028 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2029 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2030 } 2031 2032 static void 2033 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2034 { 2035 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2036 struct spdk_nvmf_tcp_transport, transport); 2037 2038 switch (pdu->hdr.common.pdu_type) { 2039 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2040 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2041 break; 2042 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2043 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2044 break; 2045 2046 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2047 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2048 break; 2049 2050 default: 2051 /* The code should not go to here */ 2052 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2053 break; 2054 } 2055 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2056 tqpair->tcp_pdu_working_count--; 2057 } 2058 2059 static inline void 2060 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2061 { 2062 treq->req.rsp->nvme_cpl.status.sct = sct; 2063 treq->req.rsp->nvme_cpl.status.sc = sc; 2064 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2065 } 2066 2067 static void 2068 data_crc32_calc_done(void *cb_arg, int status) 2069 { 2070 struct nvme_tcp_pdu *pdu = cb_arg; 2071 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2072 2073 /* async crc32 calculation is failed and use direct calculation to check */ 2074 if (spdk_unlikely(status)) { 2075 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2076 tqpair, pdu); 2077 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2078 } 2079 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2080 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2081 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2082 assert(pdu->req != NULL); 2083 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2084 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2085 } 2086 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2087 } 2088 2089 static void 2090 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2091 { 2092 int rc = 0; 2093 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2094 tqpair->pdu_in_progress = NULL; 2095 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2096 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2097 /* check data digest if need */ 2098 if (pdu->ddgst_enable) { 2099 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2100 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2101 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2102 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2103 if (spdk_likely(rc == 0)) { 2104 return; 2105 } 2106 } else { 2107 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2108 } 2109 data_crc32_calc_done(pdu, rc); 2110 } else { 2111 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2112 } 2113 } 2114 2115 static void 2116 nvmf_tcp_send_icresp_complete(void *cb_arg) 2117 { 2118 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2119 2120 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2121 } 2122 2123 static void 2124 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2125 struct spdk_nvmf_tcp_qpair *tqpair, 2126 struct nvme_tcp_pdu *pdu) 2127 { 2128 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2129 struct nvme_tcp_pdu *rsp_pdu; 2130 struct spdk_nvme_tcp_ic_resp *ic_resp; 2131 uint32_t error_offset = 0; 2132 enum spdk_nvme_tcp_term_req_fes fes; 2133 2134 /* Only PFV 0 is defined currently */ 2135 if (ic_req->pfv != 0) { 2136 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2137 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2138 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2139 goto end; 2140 } 2141 2142 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2143 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2144 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2145 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2146 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2147 goto end; 2148 } 2149 2150 /* MAXR2T is 0's based */ 2151 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2152 2153 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2154 if (!tqpair->host_hdgst_enable) { 2155 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2156 } 2157 2158 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2159 if (!tqpair->host_ddgst_enable) { 2160 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2161 } 2162 2163 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2164 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2165 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2166 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2167 tqpair, 2168 tqpair->recv_buf_size); 2169 /* Not fatal. */ 2170 } 2171 2172 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2173 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2174 2175 rsp_pdu = tqpair->mgmt_pdu; 2176 2177 ic_resp = &rsp_pdu->hdr.ic_resp; 2178 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2179 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2180 ic_resp->pfv = 0; 2181 ic_resp->cpda = tqpair->cpda; 2182 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2183 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2184 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2185 2186 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2187 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2188 2189 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2190 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2191 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2192 return; 2193 end: 2194 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2195 } 2196 2197 static void 2198 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2199 struct spdk_nvmf_tcp_transport *ttransport) 2200 { 2201 struct nvme_tcp_pdu *pdu; 2202 int rc; 2203 uint32_t crc32c, error_offset = 0; 2204 enum spdk_nvme_tcp_term_req_fes fes; 2205 2206 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2207 pdu = tqpair->pdu_in_progress; 2208 2209 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2210 pdu->hdr.common.pdu_type); 2211 /* check header digest if needed */ 2212 if (pdu->has_hdgst) { 2213 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2214 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2215 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2216 if (rc == 0) { 2217 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2218 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2219 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2220 return; 2221 2222 } 2223 } 2224 2225 switch (pdu->hdr.common.pdu_type) { 2226 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2227 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2228 break; 2229 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2230 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2231 break; 2232 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2233 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2234 break; 2235 2236 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2237 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2238 break; 2239 2240 default: 2241 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2242 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2243 error_offset = 1; 2244 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2245 break; 2246 } 2247 } 2248 2249 static void 2250 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2251 { 2252 struct nvme_tcp_pdu *pdu; 2253 uint32_t error_offset = 0; 2254 enum spdk_nvme_tcp_term_req_fes fes; 2255 uint8_t expected_hlen, pdo; 2256 bool plen_error = false, pdo_error = false; 2257 2258 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2259 pdu = tqpair->pdu_in_progress; 2260 assert(pdu); 2261 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2262 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2263 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2264 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2265 goto err; 2266 } 2267 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2268 if (pdu->hdr.common.plen != expected_hlen) { 2269 plen_error = true; 2270 } 2271 } else { 2272 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2273 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2274 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2275 goto err; 2276 } 2277 2278 switch (pdu->hdr.common.pdu_type) { 2279 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2280 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2281 pdo = pdu->hdr.common.pdo; 2282 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2283 pdo_error = true; 2284 break; 2285 } 2286 2287 if (pdu->hdr.common.plen < expected_hlen) { 2288 plen_error = true; 2289 } 2290 break; 2291 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2292 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2293 pdo = pdu->hdr.common.pdo; 2294 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2295 pdo_error = true; 2296 break; 2297 } 2298 if (pdu->hdr.common.plen < expected_hlen) { 2299 plen_error = true; 2300 } 2301 break; 2302 2303 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2304 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2305 if ((pdu->hdr.common.plen <= expected_hlen) || 2306 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2307 plen_error = true; 2308 } 2309 break; 2310 2311 default: 2312 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2313 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2314 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2315 goto err; 2316 } 2317 } 2318 2319 if (pdu->hdr.common.hlen != expected_hlen) { 2320 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2321 pdu->hdr.common.pdu_type, 2322 expected_hlen, pdu->hdr.common.hlen, tqpair); 2323 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2324 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2325 goto err; 2326 } else if (pdo_error) { 2327 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2328 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2329 } else if (plen_error) { 2330 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2331 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2332 goto err; 2333 } else { 2334 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2335 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2336 return; 2337 } 2338 err: 2339 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2340 } 2341 2342 static int 2343 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2344 { 2345 int rc = 0; 2346 struct nvme_tcp_pdu *pdu; 2347 enum nvme_tcp_pdu_recv_state prev_state; 2348 uint32_t data_len; 2349 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2350 struct spdk_nvmf_tcp_transport, transport); 2351 2352 /* The loop here is to allow for several back-to-back state changes. */ 2353 do { 2354 prev_state = tqpair->recv_state; 2355 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2356 2357 pdu = tqpair->pdu_in_progress; 2358 assert(pdu != NULL || 2359 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2360 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2361 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2362 2363 switch (tqpair->recv_state) { 2364 /* Wait for the common header */ 2365 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2366 if (!pdu) { 2367 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2368 if (spdk_unlikely(!pdu)) { 2369 return NVME_TCP_PDU_IN_PROGRESS; 2370 } 2371 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2372 tqpair->pdu_in_progress = pdu; 2373 tqpair->tcp_pdu_working_count++; 2374 } 2375 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2376 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2377 /* FALLTHROUGH */ 2378 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2379 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2380 return rc; 2381 } 2382 2383 rc = nvme_tcp_read_data(tqpair->sock, 2384 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2385 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2386 if (rc < 0) { 2387 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2388 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2389 break; 2390 } else if (rc > 0) { 2391 pdu->ch_valid_bytes += rc; 2392 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2393 } 2394 2395 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2396 return NVME_TCP_PDU_IN_PROGRESS; 2397 } 2398 2399 /* The command header of this PDU has now been read from the socket. */ 2400 nvmf_tcp_pdu_ch_handle(tqpair); 2401 break; 2402 /* Wait for the pdu specific header */ 2403 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2404 rc = nvme_tcp_read_data(tqpair->sock, 2405 pdu->psh_len - pdu->psh_valid_bytes, 2406 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2407 if (rc < 0) { 2408 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2409 break; 2410 } else if (rc > 0) { 2411 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2412 pdu->psh_valid_bytes += rc; 2413 } 2414 2415 if (pdu->psh_valid_bytes < pdu->psh_len) { 2416 return NVME_TCP_PDU_IN_PROGRESS; 2417 } 2418 2419 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 2420 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2421 break; 2422 /* Wait for the req slot */ 2423 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2424 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2425 break; 2426 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2427 /* check whether the data is valid, if not we just return */ 2428 if (!pdu->data_len) { 2429 return NVME_TCP_PDU_IN_PROGRESS; 2430 } 2431 2432 data_len = pdu->data_len; 2433 /* data digest */ 2434 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2435 tqpair->host_ddgst_enable)) { 2436 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2437 pdu->ddgst_enable = true; 2438 } 2439 2440 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2441 if (rc < 0) { 2442 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2443 break; 2444 } 2445 pdu->rw_offset += rc; 2446 2447 if (pdu->rw_offset < data_len) { 2448 return NVME_TCP_PDU_IN_PROGRESS; 2449 } 2450 2451 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2452 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2453 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2454 pdu->dif_ctx) != 0) { 2455 SPDK_ERRLOG("DIF generate failed\n"); 2456 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2457 break; 2458 } 2459 2460 /* All of this PDU has now been read from the socket. */ 2461 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2462 break; 2463 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2464 if (tqpair->tcp_pdu_working_count != 0) { 2465 return NVME_TCP_PDU_IN_PROGRESS; 2466 } 2467 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2468 break; 2469 case NVME_TCP_PDU_RECV_STATE_ERROR: 2470 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2471 return NVME_TCP_PDU_IN_PROGRESS; 2472 } 2473 return NVME_TCP_PDU_FATAL; 2474 default: 2475 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2476 abort(); 2477 break; 2478 } 2479 } while (tqpair->recv_state != prev_state); 2480 2481 return rc; 2482 } 2483 2484 static inline void * 2485 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2486 { 2487 struct spdk_nvmf_tcp_control_msg *msg; 2488 2489 assert(list); 2490 2491 msg = STAILQ_FIRST(&list->free_msgs); 2492 if (!msg) { 2493 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2494 return NULL; 2495 } 2496 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2497 return msg; 2498 } 2499 2500 static inline void 2501 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2502 { 2503 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2504 2505 assert(list); 2506 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2507 } 2508 2509 static int 2510 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2511 struct spdk_nvmf_transport *transport, 2512 struct spdk_nvmf_transport_poll_group *group) 2513 { 2514 struct spdk_nvmf_request *req = &tcp_req->req; 2515 struct spdk_nvme_cmd *cmd; 2516 struct spdk_nvme_sgl_descriptor *sgl; 2517 struct spdk_nvmf_tcp_poll_group *tgroup; 2518 enum spdk_nvme_tcp_term_req_fes fes; 2519 struct nvme_tcp_pdu *pdu; 2520 struct spdk_nvmf_tcp_qpair *tqpair; 2521 uint32_t length, error_offset = 0; 2522 2523 cmd = &req->cmd->nvme_cmd; 2524 sgl = &cmd->dptr.sgl1; 2525 2526 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2527 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2528 /* get request length from sgl */ 2529 length = sgl->unkeyed.length; 2530 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2531 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2532 length, transport->opts.max_io_size); 2533 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2534 goto fatal_err; 2535 } 2536 2537 /* fill request length and populate iovs */ 2538 req->length = length; 2539 2540 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2541 2542 if (spdk_unlikely(req->dif_enabled)) { 2543 req->dif.orig_length = length; 2544 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2545 req->dif.elba_length = length; 2546 } 2547 2548 if (nvmf_ctrlr_use_zcopy(req)) { 2549 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2550 req->data_from_pool = false; 2551 return 0; 2552 } 2553 2554 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2555 /* No available buffers. Queue this request up. */ 2556 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2557 tcp_req); 2558 return 0; 2559 } 2560 2561 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2562 tcp_req, req->iovcnt, req->iov[0].iov_base); 2563 2564 return 0; 2565 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2566 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2567 uint64_t offset = sgl->address; 2568 uint32_t max_len = transport->opts.in_capsule_data_size; 2569 2570 assert(tcp_req->has_in_capsule_data); 2571 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2572 tqpair = tcp_req->pdu->qpair; 2573 /* receiving pdu is not same with the pdu in tcp_req */ 2574 pdu = tqpair->pdu_in_progress; 2575 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2576 if (tqpair->host_ddgst_enable) { 2577 length -= SPDK_NVME_TCP_DIGEST_LEN; 2578 } 2579 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2580 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2581 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2582 length, sgl->unkeyed.length); 2583 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2584 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2585 goto fatal_err; 2586 } 2587 2588 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2589 offset, length); 2590 2591 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2592 if (spdk_unlikely(offset != 0)) { 2593 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2594 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2595 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2596 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2597 goto fatal_err; 2598 } 2599 2600 if (spdk_unlikely(length > max_len)) { 2601 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2602 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2603 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2604 2605 /* Get a buffer from dedicated list */ 2606 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2607 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2608 assert(tgroup->control_msg_list); 2609 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2610 if (!req->iov[0].iov_base) { 2611 /* No available buffers. Queue this request up. */ 2612 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2613 return 0; 2614 } 2615 } else { 2616 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2617 length, max_len); 2618 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2619 goto fatal_err; 2620 } 2621 } else { 2622 req->iov[0].iov_base = tcp_req->buf; 2623 } 2624 2625 req->length = length; 2626 req->data_from_pool = false; 2627 2628 if (spdk_unlikely(req->dif_enabled)) { 2629 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2630 req->dif.elba_length = length; 2631 } 2632 2633 req->iov[0].iov_len = length; 2634 req->iovcnt = 1; 2635 2636 return 0; 2637 } 2638 /* If we want to handle the problem here, then we can't skip the following data segment. 2639 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2640 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2641 sgl->generic.type, sgl->generic.subtype); 2642 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2643 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2644 fatal_err: 2645 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2646 return -1; 2647 } 2648 2649 static inline enum spdk_nvme_media_error_status_code 2650 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2651 enum spdk_nvme_media_error_status_code result; 2652 2653 switch (err_type) 2654 { 2655 case SPDK_DIF_REFTAG_ERROR: 2656 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2657 break; 2658 case SPDK_DIF_APPTAG_ERROR: 2659 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2660 break; 2661 case SPDK_DIF_GUARD_ERROR: 2662 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2663 break; 2664 default: 2665 SPDK_UNREACHABLE(); 2666 break; 2667 } 2668 2669 return result; 2670 } 2671 2672 static void 2673 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2674 struct spdk_nvmf_tcp_req *tcp_req) 2675 { 2676 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2677 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2678 struct nvme_tcp_pdu *rsp_pdu; 2679 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2680 uint32_t plen, pdo, alignment; 2681 int rc; 2682 2683 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2684 2685 rsp_pdu = tcp_req->pdu; 2686 assert(rsp_pdu != NULL); 2687 2688 c2h_data = &rsp_pdu->hdr.c2h_data; 2689 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2690 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2691 2692 if (tqpair->host_hdgst_enable) { 2693 plen += SPDK_NVME_TCP_DIGEST_LEN; 2694 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2695 } 2696 2697 /* set the psh */ 2698 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2699 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2700 c2h_data->datao = tcp_req->pdu->rw_offset; 2701 2702 /* set the padding */ 2703 rsp_pdu->padding_len = 0; 2704 pdo = plen; 2705 if (tqpair->cpda) { 2706 alignment = (tqpair->cpda + 1) << 2; 2707 if (plen % alignment != 0) { 2708 pdo = (plen + alignment) / alignment * alignment; 2709 rsp_pdu->padding_len = pdo - plen; 2710 plen = pdo; 2711 } 2712 } 2713 2714 c2h_data->common.pdo = pdo; 2715 plen += c2h_data->datal; 2716 if (tqpair->host_ddgst_enable) { 2717 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2718 plen += SPDK_NVME_TCP_DIGEST_LEN; 2719 } 2720 2721 c2h_data->common.plen = plen; 2722 2723 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2724 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2725 } 2726 2727 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2728 c2h_data->datao, c2h_data->datal); 2729 2730 2731 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2732 /* Need to send the capsule response if response is not all 0 */ 2733 if (ttransport->tcp_opts.c2h_success && 2734 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2735 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2736 } 2737 2738 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2739 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2740 struct spdk_dif_error err_blk = {}; 2741 uint32_t mapped_length = 0; 2742 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2743 uint32_t ddgst_len = 0; 2744 2745 if (tqpair->host_ddgst_enable) { 2746 /* Data digest consumes additional iov entry */ 2747 available_iovs--; 2748 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2749 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2750 c2h_data->common.plen -= ddgst_len; 2751 } 2752 /* Temp call to estimate if data can be described by limited number of iovs. 2753 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2754 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2755 false, &mapped_length); 2756 2757 if (mapped_length != c2h_data->common.plen) { 2758 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2759 SPDK_DEBUGLOG(nvmf_tcp, 2760 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2761 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2762 c2h_data->common.plen = mapped_length; 2763 2764 /* Rebuild pdu->data_iov since data length is changed */ 2765 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2766 c2h_data->datal); 2767 2768 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2769 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2770 } 2771 2772 c2h_data->common.plen += ddgst_len; 2773 2774 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2775 2776 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2777 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2778 if (rc != 0) { 2779 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2780 err_blk.err_type, err_blk.err_offset); 2781 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2782 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2783 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2784 return; 2785 } 2786 } 2787 2788 rsp_pdu->rw_offset += c2h_data->datal; 2789 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2790 } 2791 2792 static void 2793 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2794 struct spdk_nvmf_tcp_req *tcp_req) 2795 { 2796 nvmf_tcp_req_pdu_init(tcp_req); 2797 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2798 } 2799 2800 static int 2801 request_transfer_out(struct spdk_nvmf_request *req) 2802 { 2803 struct spdk_nvmf_tcp_req *tcp_req; 2804 struct spdk_nvmf_qpair *qpair; 2805 struct spdk_nvmf_tcp_qpair *tqpair; 2806 struct spdk_nvme_cpl *rsp; 2807 2808 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2809 2810 qpair = req->qpair; 2811 rsp = &req->rsp->nvme_cpl; 2812 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2813 2814 /* Advance our sq_head pointer */ 2815 if (qpair->sq_head == qpair->sq_head_max) { 2816 qpair->sq_head = 0; 2817 } else { 2818 qpair->sq_head++; 2819 } 2820 rsp->sqhd = qpair->sq_head; 2821 2822 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2823 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2824 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2825 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2826 } else { 2827 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2828 } 2829 2830 return 0; 2831 } 2832 2833 static void 2834 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2835 struct spdk_nvmf_tcp_qpair *tqpair, 2836 struct spdk_nvmf_tcp_req *tcp_req) 2837 { 2838 enum spdk_nvme_cmd_fuse last, next; 2839 2840 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2841 next = tcp_req->cmd.fuse; 2842 2843 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2844 2845 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2846 return; 2847 } 2848 2849 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2850 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2851 /* This is a valid pair of fused commands. Point them at each other 2852 * so they can be submitted consecutively once ready to be executed. 2853 */ 2854 tqpair->fused_first->fused_pair = tcp_req; 2855 tcp_req->fused_pair = tqpair->fused_first; 2856 tqpair->fused_first = NULL; 2857 return; 2858 } else { 2859 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2860 tqpair->fused_first->fused_failed = true; 2861 2862 /* 2863 * If the last req is in READY_TO_EXECUTE state, then call 2864 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2865 */ 2866 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2867 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2868 } 2869 2870 tqpair->fused_first = NULL; 2871 } 2872 } 2873 2874 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2875 /* Set tqpair->fused_first here so that we know to check that the next request 2876 * is a SECOND (and to fail this one if it isn't). 2877 */ 2878 tqpair->fused_first = tcp_req; 2879 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2880 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2881 tcp_req->fused_failed = true; 2882 } 2883 } 2884 2885 static bool 2886 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2887 struct spdk_nvmf_tcp_req *tcp_req) 2888 { 2889 struct spdk_nvmf_tcp_qpair *tqpair; 2890 uint32_t plen; 2891 struct nvme_tcp_pdu *pdu; 2892 enum spdk_nvmf_tcp_req_state prev_state; 2893 bool progress = false; 2894 struct spdk_nvmf_transport *transport = &ttransport->transport; 2895 struct spdk_nvmf_transport_poll_group *group; 2896 struct spdk_nvmf_tcp_poll_group *tgroup; 2897 2898 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2899 group = &tqpair->group->group; 2900 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2901 2902 /* If the qpair is not active, we need to abort the outstanding requests. */ 2903 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 2904 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2905 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2906 } 2907 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2908 } 2909 2910 /* The loop here is to allow for several back-to-back state changes. */ 2911 do { 2912 prev_state = tcp_req->state; 2913 2914 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2915 tqpair); 2916 2917 switch (tcp_req->state) { 2918 case TCP_REQUEST_STATE_FREE: 2919 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2920 * to escape this state. */ 2921 break; 2922 case TCP_REQUEST_STATE_NEW: 2923 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 2924 tqpair->qpair.queue_depth); 2925 2926 /* copy the cmd from the receive pdu */ 2927 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2928 2929 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2930 tcp_req->req.dif_enabled = true; 2931 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2932 } 2933 2934 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2935 2936 /* The next state transition depends on the data transfer needs of this request. */ 2937 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2938 2939 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2940 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2941 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2942 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2943 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2944 break; 2945 } 2946 2947 /* If no data to transfer, ready to execute. */ 2948 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2949 /* Reset the tqpair receiving pdu state */ 2950 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2951 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2952 break; 2953 } 2954 2955 pdu = tqpair->pdu_in_progress; 2956 plen = pdu->hdr.common.hlen; 2957 if (tqpair->host_hdgst_enable) { 2958 plen += SPDK_NVME_TCP_DIGEST_LEN; 2959 } 2960 if (pdu->hdr.common.plen != plen) { 2961 tcp_req->has_in_capsule_data = true; 2962 } else { 2963 /* Data is transmitted by C2H PDUs */ 2964 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2965 } 2966 2967 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2968 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2969 break; 2970 case TCP_REQUEST_STATE_NEED_BUFFER: 2971 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 2972 (uintptr_t)tcp_req); 2973 2974 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2975 2976 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2977 SPDK_DEBUGLOG(nvmf_tcp, 2978 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2979 tcp_req, tqpair); 2980 /* This request needs to wait in line to obtain a buffer */ 2981 break; 2982 } 2983 2984 /* Try to get a data buffer */ 2985 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 2986 break; 2987 } 2988 2989 /* Get a zcopy buffer if the request can be serviced through zcopy */ 2990 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 2991 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2992 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2993 tcp_req->req.length = tcp_req->req.dif.elba_length; 2994 } 2995 2996 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2997 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 2998 spdk_nvmf_request_zcopy_start(&tcp_req->req); 2999 break; 3000 } 3001 3002 if (tcp_req->req.iovcnt < 1) { 3003 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 3004 tcp_req, tqpair); 3005 /* No buffers available. */ 3006 break; 3007 } 3008 3009 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 3010 3011 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 3012 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3013 if (tcp_req->req.data_from_pool) { 3014 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3015 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3016 } else { 3017 struct nvme_tcp_pdu *pdu; 3018 3019 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3020 3021 pdu = tqpair->pdu_in_progress; 3022 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3023 tqpair); 3024 /* No need to send r2t, contained in the capsuled data */ 3025 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3026 0, tcp_req->req.length); 3027 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3028 } 3029 break; 3030 } 3031 3032 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3033 break; 3034 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3035 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3036 (uintptr_t)tcp_req); 3037 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3038 * to escape this state. */ 3039 break; 3040 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3041 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3042 (uintptr_t)tcp_req); 3043 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3044 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3045 tcp_req, tqpair); 3046 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3047 break; 3048 } 3049 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3050 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3051 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3052 } else { 3053 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3054 } 3055 break; 3056 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3057 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3058 (uintptr_t)tcp_req); 3059 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3060 break; 3061 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3062 3063 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3064 0, (uintptr_t)tcp_req); 3065 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3066 * to escape this state. */ 3067 break; 3068 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3069 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3070 (uintptr_t)tcp_req); 3071 3072 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3073 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3074 tcp_req->req.length = tcp_req->req.dif.elba_length; 3075 } 3076 3077 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3078 if (tcp_req->fused_failed) { 3079 /* This request failed FUSED semantics. Fail it immediately, without 3080 * even sending it to the target layer. 3081 */ 3082 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3083 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3084 break; 3085 } 3086 3087 if (tcp_req->fused_pair == NULL || 3088 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3089 /* This request is ready to execute, but either we don't know yet if it's 3090 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3091 * or the other request of this fused pair isn't ready to execute. So 3092 * break here and this request will get processed later either when the 3093 * other request is ready or we find that this request isn't valid. 3094 */ 3095 break; 3096 } 3097 } 3098 3099 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3100 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3101 /* If we get to this point, and this request is a fused command, we know that 3102 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3103 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3104 * request, and the other request of the fused pair, in the correct order. 3105 * Also clear the ->fused_pair pointers on both requests, since after this point 3106 * we no longer need to maintain the relationship between these two requests. 3107 */ 3108 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3109 assert(tcp_req->fused_pair != NULL); 3110 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3111 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3112 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3113 tcp_req->fused_pair->fused_pair = NULL; 3114 tcp_req->fused_pair = NULL; 3115 } 3116 spdk_nvmf_request_exec(&tcp_req->req); 3117 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3118 assert(tcp_req->fused_pair != NULL); 3119 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3120 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3121 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3122 tcp_req->fused_pair->fused_pair = NULL; 3123 tcp_req->fused_pair = NULL; 3124 } 3125 } else { 3126 /* For zero-copy, only requests with data coming from host to the 3127 * controller can end up here. */ 3128 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3129 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3130 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3131 } 3132 3133 break; 3134 case TCP_REQUEST_STATE_EXECUTING: 3135 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3136 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3137 * to escape this state. */ 3138 break; 3139 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3140 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3141 (uintptr_t)tcp_req); 3142 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3143 * to escape this state. */ 3144 break; 3145 case TCP_REQUEST_STATE_EXECUTED: 3146 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3147 3148 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3149 tcp_req->req.length = tcp_req->req.dif.orig_length; 3150 } 3151 3152 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3153 break; 3154 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3155 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3156 (uintptr_t)tcp_req); 3157 if (request_transfer_out(&tcp_req->req) != 0) { 3158 assert(0); /* No good way to handle this currently */ 3159 } 3160 break; 3161 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3162 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3163 0, (uintptr_t)tcp_req); 3164 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3165 * to escape this state. */ 3166 break; 3167 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3168 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3169 (uintptr_t)tcp_req); 3170 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3171 * to escape this state. */ 3172 break; 3173 case TCP_REQUEST_STATE_COMPLETED: 3174 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3175 tqpair->qpair.queue_depth); 3176 /* If there's an outstanding PDU sent to the host, the request is completed 3177 * due to the qpair being disconnected. We must delay the completion until 3178 * that write is done to avoid freeing the request twice. */ 3179 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3180 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3181 "write on req=%p\n", tcp_req); 3182 /* This can only happen for zcopy requests */ 3183 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3184 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3185 break; 3186 } 3187 3188 if (tcp_req->req.data_from_pool) { 3189 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3190 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3191 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3192 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3193 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3194 assert(tgroup->control_msg_list); 3195 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3196 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3197 tcp_req->req.iov[0].iov_base); 3198 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3199 /* If the request has an unreleased zcopy bdev_io, it's either a 3200 * read, a failed write, or the qpair is being disconnected */ 3201 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3202 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3203 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3204 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3205 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3206 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3207 break; 3208 } 3209 tcp_req->req.length = 0; 3210 tcp_req->req.iovcnt = 0; 3211 tcp_req->fused_failed = false; 3212 if (tcp_req->fused_pair) { 3213 /* This req was part of a valid fused pair, but failed before it got to 3214 * READ_TO_EXECUTE state. This means we need to fail the other request 3215 * in the pair, because it is no longer part of a valid pair. If the pair 3216 * already reached READY_TO_EXECUTE state, we need to kick it. 3217 */ 3218 tcp_req->fused_pair->fused_failed = true; 3219 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3220 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3221 } 3222 tcp_req->fused_pair = NULL; 3223 } 3224 3225 nvmf_tcp_req_put(tqpair, tcp_req); 3226 break; 3227 case TCP_REQUEST_NUM_STATES: 3228 default: 3229 assert(0); 3230 break; 3231 } 3232 3233 if (tcp_req->state != prev_state) { 3234 progress = true; 3235 } 3236 } while (tcp_req->state != prev_state); 3237 3238 return progress; 3239 } 3240 3241 static void 3242 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3243 { 3244 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3245 int rc; 3246 3247 assert(tqpair != NULL); 3248 rc = nvmf_tcp_sock_process(tqpair); 3249 3250 /* If there was a new socket error, disconnect */ 3251 if (rc < 0) { 3252 nvmf_tcp_qpair_disconnect(tqpair); 3253 } 3254 } 3255 3256 static int 3257 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3258 struct spdk_nvmf_qpair *qpair) 3259 { 3260 struct spdk_nvmf_tcp_poll_group *tgroup; 3261 struct spdk_nvmf_tcp_qpair *tqpair; 3262 int rc; 3263 3264 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3265 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3266 3267 rc = nvmf_tcp_qpair_sock_init(tqpair); 3268 if (rc != 0) { 3269 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3270 return -1; 3271 } 3272 3273 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3274 if (rc < 0) { 3275 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3276 return -1; 3277 } 3278 3279 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3280 if (rc < 0) { 3281 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3282 return -1; 3283 } 3284 3285 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3286 nvmf_tcp_sock_cb, tqpair); 3287 if (rc != 0) { 3288 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3289 spdk_strerror(errno), errno); 3290 return -1; 3291 } 3292 3293 tqpair->group = tgroup; 3294 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3295 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3296 3297 return 0; 3298 } 3299 3300 static int 3301 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3302 struct spdk_nvmf_qpair *qpair) 3303 { 3304 struct spdk_nvmf_tcp_poll_group *tgroup; 3305 struct spdk_nvmf_tcp_qpair *tqpair; 3306 int rc; 3307 3308 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3309 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3310 3311 assert(tqpair->group == tgroup); 3312 3313 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3314 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3315 /* Change the state to move the qpair from the await_req list to the main list 3316 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3317 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3318 } 3319 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3320 3321 /* Try to force out any pending writes */ 3322 spdk_sock_flush(tqpair->sock); 3323 3324 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3325 if (rc != 0) { 3326 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3327 spdk_strerror(errno), errno); 3328 } 3329 3330 return rc; 3331 } 3332 3333 static int 3334 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3335 { 3336 struct spdk_nvmf_tcp_transport *ttransport; 3337 struct spdk_nvmf_tcp_req *tcp_req; 3338 3339 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3340 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3341 3342 switch (tcp_req->state) { 3343 case TCP_REQUEST_STATE_EXECUTING: 3344 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3345 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3346 break; 3347 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3348 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3349 break; 3350 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3351 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3352 break; 3353 default: 3354 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3355 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3356 assert(0 && "Unexpected request state"); 3357 break; 3358 } 3359 3360 nvmf_tcp_req_process(ttransport, tcp_req); 3361 3362 return 0; 3363 } 3364 3365 static void 3366 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3367 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3368 { 3369 struct spdk_nvmf_tcp_qpair *tqpair; 3370 3371 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3372 3373 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3374 3375 assert(tqpair->fini_cb_fn == NULL); 3376 tqpair->fini_cb_fn = cb_fn; 3377 tqpair->fini_cb_arg = cb_arg; 3378 3379 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3380 nvmf_tcp_qpair_destroy(tqpair); 3381 } 3382 3383 static int 3384 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3385 { 3386 struct spdk_nvmf_tcp_poll_group *tgroup; 3387 int num_events, rc = 0, rc2; 3388 struct spdk_nvmf_request *req, *req_tmp; 3389 struct spdk_nvmf_tcp_req *tcp_req; 3390 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3391 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3392 struct spdk_nvmf_tcp_transport, transport); 3393 3394 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3395 3396 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3397 return 0; 3398 } 3399 3400 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3401 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3402 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3403 break; 3404 } 3405 } 3406 3407 num_events = spdk_sock_group_poll(tgroup->sock_group); 3408 if (spdk_unlikely(num_events < 0)) { 3409 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3410 } 3411 3412 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3413 rc2 = nvmf_tcp_sock_process(tqpair); 3414 3415 /* If there was a new socket error, disconnect */ 3416 if (spdk_unlikely(rc2 < 0)) { 3417 nvmf_tcp_qpair_disconnect(tqpair); 3418 if (rc == 0) { 3419 rc = rc2; 3420 } 3421 } 3422 } 3423 3424 return rc == 0 ? num_events : rc; 3425 } 3426 3427 static int 3428 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3429 struct spdk_nvme_transport_id *trid, bool peer) 3430 { 3431 struct spdk_nvmf_tcp_qpair *tqpair; 3432 uint16_t port; 3433 3434 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3435 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3436 3437 if (peer) { 3438 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3439 port = tqpair->initiator_port; 3440 } else { 3441 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3442 port = tqpair->target_port; 3443 } 3444 3445 if (spdk_sock_is_ipv4(tqpair->sock)) { 3446 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3447 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3448 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3449 } else { 3450 return -1; 3451 } 3452 3453 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3454 return 0; 3455 } 3456 3457 static int 3458 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3459 struct spdk_nvme_transport_id *trid) 3460 { 3461 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3462 } 3463 3464 static int 3465 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3466 struct spdk_nvme_transport_id *trid) 3467 { 3468 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3469 } 3470 3471 static int 3472 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3473 struct spdk_nvme_transport_id *trid) 3474 { 3475 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3476 } 3477 3478 static void 3479 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3480 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3481 { 3482 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3483 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3484 3485 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3486 } 3487 3488 static int 3489 _nvmf_tcp_qpair_abort_request(void *ctx) 3490 { 3491 struct spdk_nvmf_request *req = ctx; 3492 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3493 struct spdk_nvmf_tcp_req, req); 3494 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3495 struct spdk_nvmf_tcp_qpair, qpair); 3496 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3497 struct spdk_nvmf_tcp_transport, transport); 3498 int rc; 3499 3500 spdk_poller_unregister(&req->poller); 3501 3502 switch (tcp_req_to_abort->state) { 3503 case TCP_REQUEST_STATE_EXECUTING: 3504 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3505 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3506 rc = nvmf_ctrlr_abort_request(req); 3507 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3508 return SPDK_POLLER_BUSY; 3509 } 3510 break; 3511 3512 case TCP_REQUEST_STATE_NEED_BUFFER: 3513 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3514 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3515 3516 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3517 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3518 break; 3519 3520 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3521 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3522 if (spdk_get_ticks() < req->timeout_tsc) { 3523 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3524 return SPDK_POLLER_BUSY; 3525 } 3526 break; 3527 3528 default: 3529 /* Requests in other states are either un-abortable (e.g. 3530 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3531 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3532 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3533 * abort request with the bit0 of dword0 set (command not aborted). 3534 */ 3535 break; 3536 } 3537 3538 spdk_nvmf_request_complete(req); 3539 return SPDK_POLLER_BUSY; 3540 } 3541 3542 static void 3543 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3544 struct spdk_nvmf_request *req) 3545 { 3546 struct spdk_nvmf_tcp_qpair *tqpair; 3547 struct spdk_nvmf_tcp_transport *ttransport; 3548 struct spdk_nvmf_transport *transport; 3549 uint16_t cid; 3550 uint32_t i; 3551 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3552 3553 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3554 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3555 transport = &ttransport->transport; 3556 3557 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3558 3559 for (i = 0; i < tqpair->resource_count; i++) { 3560 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3561 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3562 tcp_req_to_abort = &tqpair->reqs[i]; 3563 break; 3564 } 3565 } 3566 3567 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3568 3569 if (tcp_req_to_abort == NULL) { 3570 spdk_nvmf_request_complete(req); 3571 return; 3572 } 3573 3574 req->req_to_abort = &tcp_req_to_abort->req; 3575 req->timeout_tsc = spdk_get_ticks() + 3576 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3577 req->poller = NULL; 3578 3579 _nvmf_tcp_qpair_abort_request(req); 3580 } 3581 3582 struct tcp_subsystem_add_host_opts { 3583 char *psk; 3584 }; 3585 3586 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3587 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3588 }; 3589 3590 static int 3591 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3592 { 3593 FILE *psk_file; 3594 struct stat statbuf; 3595 int rc; 3596 3597 if (stat(fname, &statbuf) != 0) { 3598 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3599 return -EACCES; 3600 } 3601 3602 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3603 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3604 return -EPERM; 3605 } 3606 if ((size_t)statbuf.st_size > bufsz) { 3607 SPDK_ERRLOG("Invalid PSK: too long\n"); 3608 return -EINVAL; 3609 } 3610 psk_file = fopen(fname, "r"); 3611 if (psk_file == NULL) { 3612 SPDK_ERRLOG("Could not open PSK file\n"); 3613 return -EINVAL; 3614 } 3615 3616 rc = fread(buf, 1, statbuf.st_size, psk_file); 3617 if (rc != statbuf.st_size) { 3618 SPDK_ERRLOG("Failed to read PSK\n"); 3619 fclose(psk_file); 3620 return -EINVAL; 3621 } 3622 3623 fclose(psk_file); 3624 return 0; 3625 } 3626 3627 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3628 3629 static int 3630 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3631 const struct spdk_nvmf_subsystem *subsystem, 3632 const char *hostnqn, 3633 const struct spdk_json_val *transport_specific) 3634 { 3635 struct tcp_subsystem_add_host_opts opts; 3636 struct spdk_nvmf_tcp_transport *ttransport; 3637 struct tcp_psk_entry *tmp, *entry = NULL; 3638 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3639 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3640 uint8_t tls_cipher_suite; 3641 int rc = 0; 3642 uint8_t psk_retained_hash; 3643 uint64_t psk_configured_size; 3644 3645 if (transport_specific == NULL) { 3646 return 0; 3647 } 3648 3649 assert(transport != NULL); 3650 assert(subsystem != NULL); 3651 3652 memset(&opts, 0, sizeof(opts)); 3653 3654 /* Decode PSK (either name of a key or file path) */ 3655 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3656 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3657 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3658 return -EINVAL; 3659 } 3660 3661 if (opts.psk == NULL) { 3662 return 0; 3663 } 3664 3665 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3666 if (entry == NULL) { 3667 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3668 rc = -ENOMEM; 3669 goto end; 3670 } 3671 3672 entry->key = spdk_keyring_get_key(opts.psk); 3673 if (entry->key != NULL) { 3674 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3675 if (rc < 0) { 3676 SPDK_ERRLOG("Failed to retreive PSK '%s'\n", opts.psk); 3677 rc = -EINVAL; 3678 goto end; 3679 } 3680 } else { 3681 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3682 SPDK_ERRLOG("PSK path too long\n"); 3683 rc = -EINVAL; 3684 goto end; 3685 } 3686 3687 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3688 if (rc) { 3689 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3690 goto end; 3691 } 3692 3693 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3694 } 3695 3696 /* Parse PSK interchange to get length of base64 encoded data. 3697 * This is then used to decide which cipher suite should be used 3698 * to generate PSK identity and TLS PSK later on. */ 3699 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3700 &psk_configured_size, &psk_retained_hash); 3701 if (rc < 0) { 3702 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3703 goto end; 3704 } 3705 3706 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3707 * This check also ensures that psk_configured_size is smaller than 3708 * psk_retained buffer size. */ 3709 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3710 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3711 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3712 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3713 } else { 3714 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3715 rc = -EINVAL; 3716 goto end; 3717 } 3718 3719 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3720 /* Generate PSK identity. */ 3721 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3722 subsystem->subnqn, tls_cipher_suite); 3723 if (rc) { 3724 rc = -EINVAL; 3725 goto end; 3726 } 3727 /* Check if PSK identity entry already exists. */ 3728 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3729 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3730 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3731 rc = -EEXIST; 3732 goto end; 3733 } 3734 } 3735 3736 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3737 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3738 rc = -EINVAL; 3739 goto end; 3740 } 3741 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3742 SPDK_ERRLOG("Could not write subnqn string!\n"); 3743 rc = -EINVAL; 3744 goto end; 3745 } 3746 3747 entry->tls_cipher_suite = tls_cipher_suite; 3748 3749 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3750 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3751 /* Psk configured is either 32 or 48 bytes long. */ 3752 memcpy(entry->psk, psk_configured, psk_configured_size); 3753 entry->psk_size = psk_configured_size; 3754 } else { 3755 /* Derive retained PSK. */ 3756 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3757 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3758 if (rc < 0) { 3759 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3760 goto end; 3761 } 3762 entry->psk_size = rc; 3763 } 3764 3765 if (entry->key == NULL) { 3766 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3767 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3768 SPDK_ERRLOG("Could not save PSK path!\n"); 3769 rc = -ENAMETOOLONG; 3770 goto end; 3771 } 3772 } 3773 3774 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3775 rc = 0; 3776 3777 end: 3778 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3779 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3780 3781 free(opts.psk); 3782 if (rc != 0) { 3783 nvmf_tcp_free_psk_entry(entry); 3784 } 3785 3786 return rc; 3787 } 3788 3789 static void 3790 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3791 const struct spdk_nvmf_subsystem *subsystem, 3792 const char *hostnqn) 3793 { 3794 struct spdk_nvmf_tcp_transport *ttransport; 3795 struct tcp_psk_entry *entry, *tmp; 3796 3797 assert(transport != NULL); 3798 assert(subsystem != NULL); 3799 3800 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3801 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3802 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3803 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3804 TAILQ_REMOVE(&ttransport->psks, entry, link); 3805 nvmf_tcp_free_psk_entry(entry); 3806 break; 3807 } 3808 } 3809 } 3810 3811 static void 3812 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3813 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3814 struct spdk_json_write_ctx *w) 3815 { 3816 struct spdk_nvmf_tcp_transport *ttransport; 3817 struct tcp_psk_entry *entry; 3818 3819 assert(transport != NULL); 3820 assert(subsystem != NULL); 3821 3822 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3823 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3824 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3825 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3826 spdk_json_write_named_string(w, "psk", entry->key ? 3827 spdk_key_get_name(entry->key) : entry->psk_path); 3828 break; 3829 } 3830 } 3831 } 3832 3833 static void 3834 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3835 { 3836 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3837 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3838 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3839 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3840 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3841 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3842 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3843 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3844 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3845 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3846 opts->transport_specific = NULL; 3847 } 3848 3849 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3850 .name = "TCP", 3851 .type = SPDK_NVME_TRANSPORT_TCP, 3852 .opts_init = nvmf_tcp_opts_init, 3853 .create = nvmf_tcp_create, 3854 .dump_opts = nvmf_tcp_dump_opts, 3855 .destroy = nvmf_tcp_destroy, 3856 3857 .listen = nvmf_tcp_listen, 3858 .stop_listen = nvmf_tcp_stop_listen, 3859 3860 .listener_discover = nvmf_tcp_discover, 3861 3862 .poll_group_create = nvmf_tcp_poll_group_create, 3863 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3864 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3865 .poll_group_add = nvmf_tcp_poll_group_add, 3866 .poll_group_remove = nvmf_tcp_poll_group_remove, 3867 .poll_group_poll = nvmf_tcp_poll_group_poll, 3868 3869 .req_free = nvmf_tcp_req_free, 3870 .req_complete = nvmf_tcp_req_complete, 3871 3872 .qpair_fini = nvmf_tcp_close_qpair, 3873 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3874 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3875 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3876 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3877 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3878 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3879 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3880 }; 3881 3882 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3883 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3884