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