1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/accel.h" 8 #include "spdk/stdinc.h" 9 #include "spdk/crc32.h" 10 #include "spdk/endian.h" 11 #include "spdk/assert.h" 12 #include "spdk/thread.h" 13 #include "spdk/nvmf_transport.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/util.h" 17 #include "spdk/log.h" 18 #include "spdk/keyring.h" 19 20 #include "spdk_internal/assert.h" 21 #include "spdk_internal/nvme_tcp.h" 22 #include "spdk_internal/sock.h" 23 24 #include "nvmf_internal.h" 25 26 #include "spdk_internal/trace_defs.h" 27 28 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 29 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 30 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 31 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 32 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 33 34 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 35 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 36 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 37 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 38 39 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 42 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 43 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 44 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 46 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 47 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 48 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 49 50 #define TCP_PSK_INVALID_PERMISSIONS 0177 51 52 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 53 static bool g_tls_log = false; 54 55 /* spdk nvmf related structure */ 56 enum spdk_nvmf_tcp_req_state { 57 58 /* The request is not currently in use */ 59 TCP_REQUEST_STATE_FREE = 0, 60 61 /* Initial state when request first received */ 62 TCP_REQUEST_STATE_NEW = 1, 63 64 /* The request is queued until a data buffer is available. */ 65 TCP_REQUEST_STATE_NEED_BUFFER = 2, 66 67 /* The request is waiting for zcopy_start to finish */ 68 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 3, 69 70 /* The request has received a zero-copy buffer */ 71 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 4, 72 73 /* The request is currently transferring data from the host to the controller. */ 74 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 5, 75 76 /* The request is waiting for the R2T send acknowledgement. */ 77 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 6, 78 79 /* The request is ready to execute at the block device */ 80 TCP_REQUEST_STATE_READY_TO_EXECUTE = 7, 81 82 /* The request is currently executing at the block device */ 83 TCP_REQUEST_STATE_EXECUTING = 8, 84 85 /* The request is waiting for zcopy buffers to be committed */ 86 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 9, 87 88 /* The request finished executing at the block device */ 89 TCP_REQUEST_STATE_EXECUTED = 10, 90 91 /* The request is ready to send a completion */ 92 TCP_REQUEST_STATE_READY_TO_COMPLETE = 11, 93 94 /* The request is currently transferring final pdus from the controller to the host. */ 95 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 12, 96 97 /* The request is waiting for zcopy buffers to be released (without committing) */ 98 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 13, 99 100 /* The request completed and can be marked free. */ 101 TCP_REQUEST_STATE_COMPLETED = 14, 102 103 /* Terminator */ 104 TCP_REQUEST_NUM_STATES, 105 }; 106 107 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 108 "Invalid PDU Header Field", 109 "PDU Sequence Error", 110 "Header Digiest Error", 111 "Data Transfer Out of Range", 112 "R2T Limit Exceeded", 113 "Unsupported parameter", 114 }; 115 116 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 117 { 118 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 119 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 120 spdk_trace_register_description("TCP_REQ_NEW", 121 TRACE_TCP_REQUEST_STATE_NEW, 122 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 123 SPDK_TRACE_ARG_TYPE_INT, "qd"); 124 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 125 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 126 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 127 SPDK_TRACE_ARG_TYPE_INT, ""); 128 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 129 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 130 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 131 SPDK_TRACE_ARG_TYPE_INT, ""); 132 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 133 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 134 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 135 SPDK_TRACE_ARG_TYPE_INT, ""); 136 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 137 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 138 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 139 SPDK_TRACE_ARG_TYPE_INT, ""); 140 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 141 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 142 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 143 SPDK_TRACE_ARG_TYPE_INT, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTING", 145 TRACE_TCP_REQUEST_STATE_EXECUTING, 146 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 147 SPDK_TRACE_ARG_TYPE_INT, ""); 148 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 149 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 150 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 151 SPDK_TRACE_ARG_TYPE_INT, ""); 152 spdk_trace_register_description("TCP_REQ_EXECUTED", 153 TRACE_TCP_REQUEST_STATE_EXECUTED, 154 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 155 SPDK_TRACE_ARG_TYPE_INT, ""); 156 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 157 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 158 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 159 SPDK_TRACE_ARG_TYPE_INT, ""); 160 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 161 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 162 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 163 SPDK_TRACE_ARG_TYPE_INT, ""); 164 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 165 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 166 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 167 SPDK_TRACE_ARG_TYPE_INT, ""); 168 spdk_trace_register_description("TCP_REQ_COMPLETED", 169 TRACE_TCP_REQUEST_STATE_COMPLETED, 170 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 171 SPDK_TRACE_ARG_TYPE_INT, "qd"); 172 spdk_trace_register_description("TCP_READ_DONE", 173 TRACE_TCP_READ_FROM_SOCKET_DONE, 174 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 175 SPDK_TRACE_ARG_TYPE_INT, ""); 176 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 177 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 178 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 179 SPDK_TRACE_ARG_TYPE_INT, ""); 180 181 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 182 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 183 SPDK_TRACE_ARG_TYPE_INT, ""); 184 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 185 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 186 SPDK_TRACE_ARG_TYPE_INT, ""); 187 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 188 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 189 SPDK_TRACE_ARG_TYPE_INT, "state"); 190 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 191 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 192 SPDK_TRACE_ARG_TYPE_INT, ""); 193 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 194 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 195 SPDK_TRACE_ARG_TYPE_INT, ""); 196 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 197 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 198 SPDK_TRACE_ARG_TYPE_INT, ""); 199 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 200 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 201 SPDK_TRACE_ARG_TYPE_INT, "state"); 202 203 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 204 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 205 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 206 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 207 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 208 } 209 210 struct spdk_nvmf_tcp_req { 211 struct spdk_nvmf_request req; 212 struct spdk_nvme_cpl rsp; 213 struct spdk_nvme_cmd cmd; 214 215 /* A PDU that can be used for sending responses. This is 216 * not the incoming PDU! */ 217 struct nvme_tcp_pdu *pdu; 218 219 /* In-capsule data buffer */ 220 uint8_t *buf; 221 222 struct spdk_nvmf_tcp_req *fused_pair; 223 224 /* 225 * The PDU for a request may be used multiple times in serial over 226 * the request's lifetime. For example, first to send an R2T, then 227 * to send a completion. To catch mistakes where the PDU is used 228 * twice at the same time, add a debug flag here for init/fini. 229 */ 230 bool pdu_in_use; 231 bool has_in_capsule_data; 232 bool fused_failed; 233 234 /* transfer_tag */ 235 uint16_t ttag; 236 237 enum spdk_nvmf_tcp_req_state state; 238 239 /* 240 * h2c_offset is used when we receive the h2c_data PDU. 241 */ 242 uint32_t h2c_offset; 243 244 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 245 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 246 }; 247 248 struct spdk_nvmf_tcp_qpair { 249 struct spdk_nvmf_qpair qpair; 250 struct spdk_nvmf_tcp_poll_group *group; 251 struct spdk_sock *sock; 252 253 enum nvme_tcp_pdu_recv_state recv_state; 254 enum nvme_tcp_qpair_state state; 255 256 /* PDU being actively received */ 257 struct nvme_tcp_pdu *pdu_in_progress; 258 259 struct spdk_nvmf_tcp_req *fused_first; 260 261 /* Queues to track the requests in all states */ 262 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 263 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 264 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 265 /* Number of working pdus */ 266 uint32_t tcp_pdu_working_count; 267 268 /* Number of requests in each state */ 269 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 270 271 uint8_t cpda; 272 273 bool host_hdgst_enable; 274 bool host_ddgst_enable; 275 276 /* This is a spare PDU used for sending special management 277 * operations. Primarily, this is used for the initial 278 * connection response and c2h termination request. */ 279 struct nvme_tcp_pdu *mgmt_pdu; 280 281 /* Arrays of in-capsule buffers, requests, and pdus. 282 * Each array is 'resource_count' number of elements */ 283 void *bufs; 284 struct spdk_nvmf_tcp_req *reqs; 285 struct nvme_tcp_pdu *pdus; 286 uint32_t resource_count; 287 uint32_t recv_buf_size; 288 289 struct spdk_nvmf_tcp_port *port; 290 291 /* IP address */ 292 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 293 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 294 295 /* IP port */ 296 uint16_t initiator_port; 297 uint16_t target_port; 298 299 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 300 bool wait_terminate; 301 302 /* Timer used to destroy qpair after detecting transport error issue if initiator does 303 * not close the connection. 304 */ 305 struct spdk_poller *timeout_poller; 306 307 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 308 void *fini_cb_arg; 309 310 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 311 }; 312 313 struct spdk_nvmf_tcp_control_msg { 314 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 315 }; 316 317 struct spdk_nvmf_tcp_control_msg_list { 318 void *msg_buf; 319 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 320 }; 321 322 struct spdk_nvmf_tcp_poll_group { 323 struct spdk_nvmf_transport_poll_group group; 324 struct spdk_sock_group *sock_group; 325 326 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 327 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 328 329 struct spdk_io_channel *accel_channel; 330 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 331 332 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 333 }; 334 335 struct spdk_nvmf_tcp_port { 336 const struct spdk_nvme_transport_id *trid; 337 struct spdk_sock *listen_sock; 338 struct spdk_nvmf_transport *transport; 339 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 340 }; 341 342 struct tcp_transport_opts { 343 bool c2h_success; 344 uint16_t control_msg_num; 345 uint32_t sock_priority; 346 }; 347 348 struct tcp_psk_entry { 349 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 350 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 351 char pskid[NVMF_PSK_IDENTITY_LEN]; 352 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 353 struct spdk_key *key; 354 355 /* Original path saved to emit SPDK configuration via "save_config". */ 356 char psk_path[PATH_MAX]; 357 uint32_t psk_size; 358 enum nvme_tcp_cipher_suite tls_cipher_suite; 359 TAILQ_ENTRY(tcp_psk_entry) link; 360 }; 361 362 struct spdk_nvmf_tcp_transport { 363 struct spdk_nvmf_transport transport; 364 struct tcp_transport_opts tcp_opts; 365 uint32_t ack_timeout; 366 367 struct spdk_nvmf_tcp_poll_group *next_pg; 368 369 struct spdk_poller *accept_poller; 370 struct spdk_sock_group *listen_sock_group; 371 372 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 373 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 374 375 TAILQ_HEAD(, tcp_psk_entry) psks; 376 }; 377 378 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 379 { 380 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 381 spdk_json_decode_bool, true 382 }, 383 { 384 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 385 spdk_json_decode_uint16, true 386 }, 387 { 388 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 389 spdk_json_decode_uint32, true 390 }, 391 }; 392 393 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 394 struct spdk_nvmf_tcp_req *tcp_req); 395 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 396 397 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 398 struct spdk_nvmf_tcp_req *tcp_req); 399 400 static inline void 401 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 402 enum spdk_nvmf_tcp_req_state state) 403 { 404 struct spdk_nvmf_qpair *qpair; 405 struct spdk_nvmf_tcp_qpair *tqpair; 406 407 qpair = tcp_req->req.qpair; 408 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 409 410 assert(tqpair->state_cntr[tcp_req->state] > 0); 411 tqpair->state_cntr[tcp_req->state]--; 412 tqpair->state_cntr[state]++; 413 414 tcp_req->state = state; 415 } 416 417 static inline struct nvme_tcp_pdu * 418 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 419 { 420 assert(tcp_req->pdu_in_use == false); 421 422 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 423 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 424 425 return tcp_req->pdu; 426 } 427 428 static struct spdk_nvmf_tcp_req * 429 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 430 { 431 struct spdk_nvmf_tcp_req *tcp_req; 432 433 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 434 if (spdk_unlikely(!tcp_req)) { 435 return NULL; 436 } 437 438 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 439 tcp_req->h2c_offset = 0; 440 tcp_req->has_in_capsule_data = false; 441 tcp_req->req.dif_enabled = false; 442 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 443 444 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 445 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 446 tqpair->qpair.queue_depth++; 447 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 448 return tcp_req; 449 } 450 451 static inline void 452 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 453 { 454 assert(!tcp_req->pdu_in_use); 455 456 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 457 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 458 tqpair->qpair.queue_depth--; 459 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 460 } 461 462 static void 463 nvmf_tcp_request_free(void *cb_arg) 464 { 465 struct spdk_nvmf_tcp_transport *ttransport; 466 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 467 468 assert(tcp_req != NULL); 469 470 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 471 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 472 struct spdk_nvmf_tcp_transport, transport); 473 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 474 nvmf_tcp_req_process(ttransport, tcp_req); 475 } 476 477 static int 478 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 479 { 480 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 481 482 nvmf_tcp_request_free(tcp_req); 483 484 return 0; 485 } 486 487 static void 488 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 489 enum spdk_nvmf_tcp_req_state state) 490 { 491 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 492 493 assert(state != TCP_REQUEST_STATE_FREE); 494 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 495 if (state == tcp_req->state) { 496 nvmf_tcp_request_free(tcp_req); 497 } 498 } 499 } 500 501 static void 502 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 503 { 504 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 505 506 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 507 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 508 509 /* Wipe the requests waiting for buffer from the global list */ 510 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 511 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 512 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 513 spdk_nvmf_request, buf_link); 514 } 515 } 516 517 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 518 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 519 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 520 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 521 } 522 523 static void 524 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 525 { 526 int i; 527 struct spdk_nvmf_tcp_req *tcp_req; 528 529 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 530 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 531 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 532 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 533 if ((int)tcp_req->state == i) { 534 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 535 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 536 } 537 } 538 } 539 } 540 541 static void 542 _nvmf_tcp_qpair_destroy(void *_tqpair) 543 { 544 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 545 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 546 void *cb_arg = tqpair->fini_cb_arg; 547 int err = 0; 548 549 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 550 551 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 552 553 err = spdk_sock_close(&tqpair->sock); 554 assert(err == 0); 555 nvmf_tcp_cleanup_all_states(tqpair); 556 557 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 558 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 559 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 560 tqpair->resource_count); 561 err++; 562 } 563 564 if (err > 0) { 565 nvmf_tcp_dump_qpair_req_contents(tqpair); 566 } 567 568 /* The timeout poller might still be registered here if we close the qpair before host 569 * terminates the connection. 570 */ 571 spdk_poller_unregister(&tqpair->timeout_poller); 572 spdk_dma_free(tqpair->pdus); 573 free(tqpair->reqs); 574 spdk_free(tqpair->bufs); 575 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 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 spdk_sock_group_close(&ttransport->listen_sock_group); 635 free(ttransport); 636 637 if (cb_fn) { 638 cb_fn(cb_arg); 639 } 640 return 0; 641 } 642 643 static int nvmf_tcp_accept(void *ctx); 644 645 static void nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock); 646 647 static struct spdk_nvmf_transport * 648 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 649 { 650 struct spdk_nvmf_tcp_transport *ttransport; 651 uint32_t sge_count; 652 uint32_t min_shared_buffers; 653 654 ttransport = calloc(1, sizeof(*ttransport)); 655 if (!ttransport) { 656 return NULL; 657 } 658 659 TAILQ_INIT(&ttransport->ports); 660 TAILQ_INIT(&ttransport->poll_groups); 661 TAILQ_INIT(&ttransport->psks); 662 663 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 664 665 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 666 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 667 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 668 if (opts->transport_specific != NULL && 669 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 670 SPDK_COUNTOF(tcp_transport_opts_decoder), 671 &ttransport->tcp_opts)) { 672 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 673 free(ttransport); 674 return NULL; 675 } 676 677 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 678 679 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 680 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 681 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 682 " in_capsule_data_size=%d, max_aq_depth=%d\n" 683 " num_shared_buffers=%d, c2h_success=%d,\n" 684 " dif_insert_or_strip=%d, sock_priority=%d\n" 685 " abort_timeout_sec=%d, control_msg_num=%hu\n" 686 " ack_timeout=%d\n", 687 opts->max_queue_depth, 688 opts->max_io_size, 689 opts->max_qpairs_per_ctrlr - 1, 690 opts->io_unit_size, 691 opts->in_capsule_data_size, 692 opts->max_aq_depth, 693 opts->num_shared_buffers, 694 ttransport->tcp_opts.c2h_success, 695 opts->dif_insert_or_strip, 696 ttransport->tcp_opts.sock_priority, 697 opts->abort_timeout_sec, 698 ttransport->tcp_opts.control_msg_num, 699 opts->ack_timeout); 700 701 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 702 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 703 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 704 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 705 free(ttransport); 706 return NULL; 707 } 708 709 if (ttransport->tcp_opts.control_msg_num == 0 && 710 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 711 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 712 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 713 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 714 } 715 716 /* I/O unit size cannot be larger than max I/O size */ 717 if (opts->io_unit_size > opts->max_io_size) { 718 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", 719 opts->io_unit_size, opts->max_io_size); 720 opts->io_unit_size = opts->max_io_size; 721 } 722 723 /* In capsule data size cannot be larger than max I/O size */ 724 if (opts->in_capsule_data_size > opts->max_io_size) { 725 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 726 opts->io_unit_size, opts->max_io_size); 727 opts->in_capsule_data_size = opts->max_io_size; 728 } 729 730 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 731 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 732 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 733 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 734 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 735 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 736 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 737 } 738 739 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 740 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 741 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 742 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 743 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 744 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 745 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 746 } 747 748 sge_count = opts->max_io_size / opts->io_unit_size; 749 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 750 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 751 free(ttransport); 752 return NULL; 753 } 754 755 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 756 if (opts->buf_cache_size < UINT32_MAX) { 757 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 758 if (min_shared_buffers > opts->num_shared_buffers) { 759 SPDK_ERRLOG("There are not enough buffers to satisfy " 760 "per-poll group caches for each thread. (%" PRIu32 ") " 761 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 762 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 763 free(ttransport); 764 return NULL; 765 } 766 } 767 768 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, 769 opts->acceptor_poll_rate); 770 if (!ttransport->accept_poller) { 771 free(ttransport); 772 return NULL; 773 } 774 775 ttransport->listen_sock_group = spdk_sock_group_create(NULL); 776 if (ttransport->listen_sock_group == NULL) { 777 SPDK_ERRLOG("Failed to create socket group for listen sockets\n"); 778 spdk_poller_unregister(&ttransport->accept_poller); 779 free(ttransport); 780 return NULL; 781 } 782 783 return &ttransport->transport; 784 } 785 786 static int 787 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 788 { 789 unsigned long long ull; 790 char *end = NULL; 791 792 ull = strtoull(trsvcid, &end, 10); 793 if (end == NULL || end == trsvcid || *end != '\0') { 794 return -1; 795 } 796 797 /* Valid TCP/IP port numbers are in [1, 65535] */ 798 if (ull == 0 || ull > 65535) { 799 return -1; 800 } 801 802 return (int)ull; 803 } 804 805 /** 806 * Canonicalize a listen address trid. 807 */ 808 static int 809 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 810 const struct spdk_nvme_transport_id *trid) 811 { 812 int trsvcid_int; 813 814 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 815 if (trsvcid_int < 0) { 816 return -EINVAL; 817 } 818 819 memset(canon_trid, 0, sizeof(*canon_trid)); 820 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 821 canon_trid->adrfam = trid->adrfam; 822 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 823 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 824 825 return 0; 826 } 827 828 /** 829 * Find an existing listening port. 830 */ 831 static struct spdk_nvmf_tcp_port * 832 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 833 const struct spdk_nvme_transport_id *trid) 834 { 835 struct spdk_nvme_transport_id canon_trid; 836 struct spdk_nvmf_tcp_port *port; 837 838 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 839 return NULL; 840 } 841 842 TAILQ_FOREACH(port, &ttransport->ports, link) { 843 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 844 return port; 845 } 846 } 847 848 return NULL; 849 } 850 851 static int 852 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 853 void *get_key_ctx) 854 { 855 struct tcp_psk_entry *entry; 856 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 857 size_t psk_len; 858 int rc; 859 860 TAILQ_FOREACH(entry, &ttransport->psks, link) { 861 if (strcmp(pskid, entry->pskid) != 0) { 862 continue; 863 } 864 865 psk_len = entry->psk_size; 866 if ((size_t)out_len < psk_len) { 867 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 868 out_len, psk_len); 869 return -ENOBUFS; 870 } 871 872 /* Convert PSK to the TLS PSK format. */ 873 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 874 entry->tls_cipher_suite); 875 if (rc < 0) { 876 SPDK_ERRLOG("Could not generate TLS PSK\n"); 877 } 878 879 switch (entry->tls_cipher_suite) { 880 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 881 *cipher = "TLS_AES_128_GCM_SHA256"; 882 break; 883 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 884 *cipher = "TLS_AES_256_GCM_SHA384"; 885 break; 886 default: 887 *cipher = NULL; 888 return -ENOTSUP; 889 } 890 891 return rc; 892 } 893 894 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 895 896 return -EINVAL; 897 } 898 899 static int 900 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 901 struct spdk_nvmf_listen_opts *listen_opts) 902 { 903 struct spdk_nvmf_tcp_transport *ttransport; 904 struct spdk_nvmf_tcp_port *port; 905 int trsvcid_int; 906 uint8_t adrfam; 907 const char *sock_impl_name; 908 struct spdk_sock_impl_opts impl_opts; 909 size_t impl_opts_size = sizeof(impl_opts); 910 struct spdk_sock_opts opts; 911 int rc; 912 913 if (!strlen(trid->trsvcid)) { 914 SPDK_ERRLOG("Service id is required\n"); 915 return -EINVAL; 916 } 917 918 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 919 920 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 921 if (trsvcid_int < 0) { 922 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 923 return -EINVAL; 924 } 925 926 port = calloc(1, sizeof(*port)); 927 if (!port) { 928 SPDK_ERRLOG("Port allocation failed\n"); 929 return -ENOMEM; 930 } 931 932 port->trid = trid; 933 934 sock_impl_name = NULL; 935 936 opts.opts_size = sizeof(opts); 937 spdk_sock_get_default_opts(&opts); 938 opts.priority = ttransport->tcp_opts.sock_priority; 939 opts.ack_timeout = transport->opts.ack_timeout; 940 if (listen_opts->secure_channel) { 941 if (listen_opts->sock_impl && 942 strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) { 943 SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported"); 944 free(port); 945 return -EINVAL; 946 } 947 listen_opts->sock_impl = "ssl"; 948 } 949 950 if (listen_opts->sock_impl) { 951 sock_impl_name = listen_opts->sock_impl; 952 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 953 954 if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) { 955 if (!g_tls_log) { 956 SPDK_NOTICELOG("TLS support is considered experimental\n"); 957 g_tls_log = true; 958 } 959 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 960 impl_opts.get_key = tcp_sock_get_key; 961 impl_opts.get_key_ctx = ttransport; 962 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 963 } 964 965 opts.impl_opts = &impl_opts; 966 opts.impl_opts_size = sizeof(impl_opts); 967 } 968 969 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 970 sock_impl_name, &opts); 971 if (port->listen_sock == NULL) { 972 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 973 trid->traddr, trsvcid_int, 974 spdk_strerror(errno), errno); 975 free(port); 976 return -errno; 977 } 978 979 if (spdk_sock_is_ipv4(port->listen_sock)) { 980 adrfam = SPDK_NVMF_ADRFAM_IPV4; 981 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 982 adrfam = SPDK_NVMF_ADRFAM_IPV6; 983 } else { 984 SPDK_ERRLOG("Unhandled socket type\n"); 985 adrfam = 0; 986 } 987 988 if (adrfam != trid->adrfam) { 989 SPDK_ERRLOG("Socket address family mismatch\n"); 990 spdk_sock_close(&port->listen_sock); 991 free(port); 992 return -EINVAL; 993 } 994 995 rc = spdk_sock_group_add_sock(ttransport->listen_sock_group, port->listen_sock, nvmf_tcp_accept_cb, 996 port); 997 if (rc < 0) { 998 SPDK_ERRLOG("Failed to add socket to the listen socket group\n"); 999 spdk_sock_close(&port->listen_sock); 1000 free(port); 1001 return -errno; 1002 } 1003 1004 port->transport = transport; 1005 1006 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 1007 trid->traddr, trid->trsvcid); 1008 1009 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 1010 return 0; 1011 } 1012 1013 static void 1014 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 1015 const struct spdk_nvme_transport_id *trid) 1016 { 1017 struct spdk_nvmf_tcp_transport *ttransport; 1018 struct spdk_nvmf_tcp_port *port; 1019 1020 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1021 1022 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 1023 trid->traddr, trid->trsvcid); 1024 1025 port = nvmf_tcp_find_port(ttransport, trid); 1026 if (port) { 1027 spdk_sock_group_remove_sock(ttransport->listen_sock_group, port->listen_sock); 1028 TAILQ_REMOVE(&ttransport->ports, port, link); 1029 spdk_sock_close(&port->listen_sock); 1030 free(port); 1031 } 1032 } 1033 1034 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1035 enum nvme_tcp_pdu_recv_state state); 1036 1037 static void 1038 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvme_tcp_qpair_state state) 1039 { 1040 tqpair->state = state; 1041 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1042 (uint64_t)tqpair->state); 1043 } 1044 1045 static void 1046 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1047 { 1048 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1049 1050 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1051 1052 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 1053 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITING); 1054 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1055 spdk_poller_unregister(&tqpair->timeout_poller); 1056 1057 /* This will end up calling nvmf_tcp_close_qpair */ 1058 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1059 } 1060 } 1061 1062 static void 1063 _mgmt_pdu_write_done(void *_tqpair, int err) 1064 { 1065 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1066 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1067 1068 if (spdk_unlikely(err != 0)) { 1069 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1070 return; 1071 } 1072 1073 assert(pdu->cb_fn != NULL); 1074 pdu->cb_fn(pdu->cb_arg); 1075 } 1076 1077 static void 1078 _req_pdu_write_done(void *req, int err) 1079 { 1080 struct spdk_nvmf_tcp_req *tcp_req = req; 1081 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1082 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1083 1084 assert(tcp_req->pdu_in_use); 1085 tcp_req->pdu_in_use = false; 1086 1087 /* If the request is in a completed state, we're waiting for write completion to free it */ 1088 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1089 nvmf_tcp_request_free(tcp_req); 1090 return; 1091 } 1092 1093 if (spdk_unlikely(err != 0)) { 1094 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1095 return; 1096 } 1097 1098 assert(pdu->cb_fn != NULL); 1099 pdu->cb_fn(pdu->cb_arg); 1100 } 1101 1102 static void 1103 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1104 { 1105 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1106 } 1107 1108 static void 1109 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1110 { 1111 int rc; 1112 uint32_t mapped_length; 1113 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1114 1115 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1116 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1117 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1118 1119 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1120 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1121 /* Try to force the send immediately. */ 1122 rc = spdk_sock_flush(tqpair->sock); 1123 if (rc > 0 && (uint32_t)rc == mapped_length) { 1124 _pdu_write_done(pdu, 0); 1125 } else { 1126 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1127 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1128 "IC_RESP" : "TERM_REQ", rc, errno); 1129 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1130 } 1131 } 1132 } 1133 1134 static void 1135 data_crc32_accel_done(void *cb_arg, int status) 1136 { 1137 struct nvme_tcp_pdu *pdu = cb_arg; 1138 1139 if (spdk_unlikely(status)) { 1140 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1141 _pdu_write_done(pdu, status); 1142 return; 1143 } 1144 1145 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1146 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1147 1148 _tcp_write_pdu(pdu); 1149 } 1150 1151 static void 1152 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1153 { 1154 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1155 int rc = 0; 1156 1157 /* Data Digest */ 1158 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1159 /* Only support this limitated case for the first step */ 1160 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1161 && tqpair->group)) { 1162 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1163 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1164 if (spdk_likely(rc == 0)) { 1165 return; 1166 } 1167 } else { 1168 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1169 } 1170 data_crc32_accel_done(pdu, rc); 1171 } else { 1172 _tcp_write_pdu(pdu); 1173 } 1174 } 1175 1176 static void 1177 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1178 struct nvme_tcp_pdu *pdu, 1179 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1180 void *cb_arg) 1181 { 1182 int hlen; 1183 uint32_t crc32c; 1184 1185 assert(tqpair->pdu_in_progress != pdu); 1186 1187 hlen = pdu->hdr.common.hlen; 1188 pdu->cb_fn = cb_fn; 1189 pdu->cb_arg = cb_arg; 1190 1191 pdu->iov[0].iov_base = &pdu->hdr.raw; 1192 pdu->iov[0].iov_len = hlen; 1193 1194 /* Header Digest */ 1195 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1196 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1197 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1198 } 1199 1200 /* Data Digest */ 1201 pdu_data_crc32_compute(pdu); 1202 } 1203 1204 static void 1205 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1206 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1207 void *cb_arg) 1208 { 1209 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1210 1211 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1212 pdu->sock_req.cb_arg = tqpair; 1213 1214 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1215 } 1216 1217 static void 1218 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1219 struct spdk_nvmf_tcp_req *tcp_req, 1220 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1221 void *cb_arg) 1222 { 1223 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1224 1225 pdu->sock_req.cb_fn = _req_pdu_write_done; 1226 pdu->sock_req.cb_arg = tcp_req; 1227 1228 assert(!tcp_req->pdu_in_use); 1229 tcp_req->pdu_in_use = true; 1230 1231 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1232 } 1233 1234 static int 1235 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1236 { 1237 uint32_t i; 1238 struct spdk_nvmf_transport_opts *opts; 1239 uint32_t in_capsule_data_size; 1240 1241 opts = &tqpair->qpair.transport->opts; 1242 1243 in_capsule_data_size = opts->in_capsule_data_size; 1244 if (opts->dif_insert_or_strip) { 1245 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1246 } 1247 1248 tqpair->resource_count = opts->max_queue_depth; 1249 1250 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1251 if (!tqpair->reqs) { 1252 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1253 return -1; 1254 } 1255 1256 if (in_capsule_data_size) { 1257 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1258 NULL, SPDK_ENV_LCORE_ID_ANY, 1259 SPDK_MALLOC_DMA); 1260 if (!tqpair->bufs) { 1261 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1262 return -1; 1263 } 1264 } 1265 /* prepare memory space for receiving pdus and tcp_req */ 1266 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1267 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1268 NULL); 1269 if (!tqpair->pdus) { 1270 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1271 return -1; 1272 } 1273 1274 for (i = 0; i < tqpair->resource_count; i++) { 1275 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1276 1277 tcp_req->ttag = i + 1; 1278 tcp_req->req.qpair = &tqpair->qpair; 1279 1280 tcp_req->pdu = &tqpair->pdus[i]; 1281 tcp_req->pdu->qpair = tqpair; 1282 1283 /* Set up memory to receive commands */ 1284 if (tqpair->bufs) { 1285 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1286 } 1287 1288 /* Set the cmdn and rsp */ 1289 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1290 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1291 1292 tcp_req->req.stripped_data = NULL; 1293 1294 /* Initialize request state to FREE */ 1295 tcp_req->state = TCP_REQUEST_STATE_FREE; 1296 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1297 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1298 } 1299 1300 for (; i < 2 * tqpair->resource_count; i++) { 1301 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1302 1303 pdu->qpair = tqpair; 1304 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1305 } 1306 1307 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1308 tqpair->mgmt_pdu->qpair = tqpair; 1309 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1310 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1311 tqpair->tcp_pdu_working_count = 1; 1312 1313 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1314 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1315 1316 return 0; 1317 } 1318 1319 static int 1320 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1321 { 1322 struct spdk_nvmf_tcp_qpair *tqpair; 1323 1324 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1325 1326 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1327 1328 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1329 1330 /* Initialise request state queues of the qpair */ 1331 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1332 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1333 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1334 tqpair->qpair.queue_depth = 0; 1335 1336 tqpair->host_hdgst_enable = true; 1337 tqpair->host_ddgst_enable = true; 1338 1339 return 0; 1340 } 1341 1342 static int 1343 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1344 { 1345 char saddr[32], caddr[32]; 1346 uint16_t sport, cport; 1347 char owner[256]; 1348 int rc; 1349 1350 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1351 caddr, sizeof(caddr), &cport); 1352 if (rc != 0) { 1353 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1354 return rc; 1355 } 1356 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1357 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1358 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1359 1360 /* set low water mark */ 1361 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1362 if (rc != 0) { 1363 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1364 return rc; 1365 } 1366 1367 return 0; 1368 } 1369 1370 static void 1371 nvmf_tcp_handle_connect(struct spdk_nvmf_tcp_port *port, struct spdk_sock *sock) 1372 { 1373 struct spdk_nvmf_tcp_qpair *tqpair; 1374 int rc; 1375 1376 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1377 port->trid->traddr, port->trid->trsvcid); 1378 1379 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1380 if (tqpair == NULL) { 1381 SPDK_ERRLOG("Could not allocate new connection.\n"); 1382 spdk_sock_close(&sock); 1383 return; 1384 } 1385 1386 tqpair->sock = sock; 1387 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1388 tqpair->port = port; 1389 tqpair->qpair.transport = port->transport; 1390 1391 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1392 sizeof(tqpair->target_addr), &tqpair->target_port, 1393 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1394 &tqpair->initiator_port); 1395 if (rc < 0) { 1396 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1397 nvmf_tcp_qpair_destroy(tqpair); 1398 return; 1399 } 1400 1401 spdk_nvmf_tgt_new_qpair(port->transport->tgt, &tqpair->qpair); 1402 } 1403 1404 static uint32_t 1405 nvmf_tcp_port_accept(struct spdk_nvmf_tcp_port *port) 1406 { 1407 struct spdk_sock *sock; 1408 uint32_t count = 0; 1409 int i; 1410 1411 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1412 sock = spdk_sock_accept(port->listen_sock); 1413 if (sock == NULL) { 1414 break; 1415 } 1416 count++; 1417 nvmf_tcp_handle_connect(port, sock); 1418 } 1419 1420 return count; 1421 } 1422 1423 static int 1424 nvmf_tcp_accept(void *ctx) 1425 { 1426 struct spdk_nvmf_transport *transport = ctx; 1427 struct spdk_nvmf_tcp_transport *ttransport; 1428 int count; 1429 1430 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1431 1432 count = spdk_sock_group_poll(ttransport->listen_sock_group); 1433 if (count < 0) { 1434 SPDK_ERRLOG("Fail in TCP listen socket group poll\n"); 1435 } 1436 1437 return count != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1438 } 1439 1440 static void 1441 nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock) 1442 { 1443 struct spdk_nvmf_tcp_port *port = ctx; 1444 1445 nvmf_tcp_port_accept(port); 1446 } 1447 1448 static void 1449 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1450 struct spdk_nvme_transport_id *trid, 1451 struct spdk_nvmf_discovery_log_page_entry *entry) 1452 { 1453 struct spdk_nvmf_tcp_port *port; 1454 struct spdk_nvmf_tcp_transport *ttransport; 1455 1456 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1457 entry->adrfam = trid->adrfam; 1458 1459 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1460 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1461 1462 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1463 port = nvmf_tcp_find_port(ttransport, trid); 1464 1465 assert(port != NULL); 1466 1467 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1468 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1469 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1470 } else { 1471 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1472 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1473 } 1474 } 1475 1476 static struct spdk_nvmf_tcp_control_msg_list * 1477 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1478 { 1479 struct spdk_nvmf_tcp_control_msg_list *list; 1480 struct spdk_nvmf_tcp_control_msg *msg; 1481 uint16_t i; 1482 1483 list = calloc(1, sizeof(*list)); 1484 if (!list) { 1485 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1486 return NULL; 1487 } 1488 1489 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1490 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1491 if (!list->msg_buf) { 1492 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1493 free(list); 1494 return NULL; 1495 } 1496 1497 STAILQ_INIT(&list->free_msgs); 1498 1499 for (i = 0; i < num_messages; i++) { 1500 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1501 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1502 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1503 } 1504 1505 return list; 1506 } 1507 1508 static void 1509 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1510 { 1511 if (!list) { 1512 return; 1513 } 1514 1515 spdk_free(list->msg_buf); 1516 free(list); 1517 } 1518 1519 static struct spdk_nvmf_transport_poll_group * 1520 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1521 struct spdk_nvmf_poll_group *group) 1522 { 1523 struct spdk_nvmf_tcp_transport *ttransport; 1524 struct spdk_nvmf_tcp_poll_group *tgroup; 1525 1526 if (spdk_interrupt_mode_is_enabled()) { 1527 SPDK_ERRLOG("TCP transport does not support interrupt mode\n"); 1528 return NULL; 1529 } 1530 1531 tgroup = calloc(1, sizeof(*tgroup)); 1532 if (!tgroup) { 1533 return NULL; 1534 } 1535 1536 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1537 if (!tgroup->sock_group) { 1538 goto cleanup; 1539 } 1540 1541 TAILQ_INIT(&tgroup->qpairs); 1542 TAILQ_INIT(&tgroup->await_req); 1543 1544 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1545 1546 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1547 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1548 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1549 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1550 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1551 if (!tgroup->control_msg_list) { 1552 goto cleanup; 1553 } 1554 } 1555 1556 tgroup->accel_channel = spdk_accel_get_io_channel(); 1557 if (spdk_unlikely(!tgroup->accel_channel)) { 1558 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1559 goto cleanup; 1560 } 1561 1562 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1563 if (ttransport->next_pg == NULL) { 1564 ttransport->next_pg = tgroup; 1565 } 1566 1567 return &tgroup->group; 1568 1569 cleanup: 1570 nvmf_tcp_poll_group_destroy(&tgroup->group); 1571 return NULL; 1572 } 1573 1574 static struct spdk_nvmf_transport_poll_group * 1575 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1576 { 1577 struct spdk_nvmf_tcp_transport *ttransport; 1578 struct spdk_nvmf_tcp_poll_group **pg; 1579 struct spdk_nvmf_tcp_qpair *tqpair; 1580 struct spdk_sock_group *group = NULL, *hint = NULL; 1581 int rc; 1582 1583 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1584 1585 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1586 return NULL; 1587 } 1588 1589 pg = &ttransport->next_pg; 1590 assert(*pg != NULL); 1591 hint = (*pg)->sock_group; 1592 1593 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1594 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1595 if (rc != 0) { 1596 return NULL; 1597 } else if (group != NULL) { 1598 /* Optimal poll group was found */ 1599 return spdk_sock_group_get_ctx(group); 1600 } 1601 1602 /* The hint was used for optimal poll group, advance next_pg. */ 1603 *pg = TAILQ_NEXT(*pg, link); 1604 if (*pg == NULL) { 1605 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1606 } 1607 1608 return spdk_sock_group_get_ctx(hint); 1609 } 1610 1611 static void 1612 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1613 { 1614 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1615 struct spdk_nvmf_tcp_transport *ttransport; 1616 1617 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1618 spdk_sock_group_close(&tgroup->sock_group); 1619 if (tgroup->control_msg_list) { 1620 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1621 } 1622 1623 if (tgroup->accel_channel) { 1624 spdk_put_io_channel(tgroup->accel_channel); 1625 } 1626 1627 if (tgroup->group.transport == NULL) { 1628 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1629 * calls this function directly in a failure path. */ 1630 free(tgroup); 1631 return; 1632 } 1633 1634 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1635 1636 next_tgroup = TAILQ_NEXT(tgroup, link); 1637 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1638 if (next_tgroup == NULL) { 1639 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1640 } 1641 if (ttransport->next_pg == tgroup) { 1642 ttransport->next_pg = next_tgroup; 1643 } 1644 1645 free(tgroup); 1646 } 1647 1648 static void 1649 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1650 enum nvme_tcp_pdu_recv_state state) 1651 { 1652 if (tqpair->recv_state == state) { 1653 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1654 tqpair, state); 1655 return; 1656 } 1657 1658 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1659 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1660 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1661 tqpair->tcp_pdu_working_count--; 1662 } 1663 } 1664 1665 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1666 assert(tqpair->tcp_pdu_working_count == 0); 1667 } 1668 1669 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1670 /* When leaving the await req state, move the qpair to the main list */ 1671 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1672 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1673 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1674 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1675 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1676 } 1677 1678 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1679 tqpair->recv_state = state; 1680 1681 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1682 (uint64_t)tqpair->recv_state); 1683 } 1684 1685 static int 1686 nvmf_tcp_qpair_handle_timeout(void *ctx) 1687 { 1688 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1689 1690 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1691 1692 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1693 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1694 1695 nvmf_tcp_qpair_disconnect(tqpair); 1696 return SPDK_POLLER_BUSY; 1697 } 1698 1699 static void 1700 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1701 { 1702 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1703 1704 if (!tqpair->timeout_poller) { 1705 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1706 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1707 } 1708 } 1709 1710 static void 1711 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1712 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1713 { 1714 struct nvme_tcp_pdu *rsp_pdu; 1715 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1716 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1717 uint32_t copy_len; 1718 1719 rsp_pdu = tqpair->mgmt_pdu; 1720 1721 c2h_term_req = &rsp_pdu->hdr.term_req; 1722 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1723 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1724 c2h_term_req->fes = fes; 1725 1726 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1727 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1728 DSET32(&c2h_term_req->fei, error_offset); 1729 } 1730 1731 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1732 1733 /* Copy the error info into the buffer */ 1734 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1735 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1736 1737 /* Contain the header of the wrong received pdu */ 1738 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1739 tqpair->wait_terminate = true; 1740 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1741 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1742 } 1743 1744 static void 1745 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1746 struct spdk_nvmf_tcp_qpair *tqpair, 1747 struct nvme_tcp_pdu *pdu) 1748 { 1749 struct spdk_nvmf_tcp_req *tcp_req; 1750 1751 assert(pdu->psh_valid_bytes == pdu->psh_len); 1752 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1753 1754 tcp_req = nvmf_tcp_req_get(tqpair); 1755 if (!tcp_req) { 1756 /* Directly return and make the allocation retry again. This can happen if we're 1757 * using asynchronous writes to send the response to the host or when releasing 1758 * zero-copy buffers after a response has been sent. In both cases, the host might 1759 * receive the response before we've finished processing the request and is free to 1760 * send another one. 1761 */ 1762 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1763 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1764 return; 1765 } 1766 1767 /* The host sent more commands than the maximum queue depth. */ 1768 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1769 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1770 return; 1771 } 1772 1773 pdu->req = tcp_req; 1774 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1775 nvmf_tcp_req_process(ttransport, tcp_req); 1776 } 1777 1778 static void 1779 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1780 struct spdk_nvmf_tcp_qpair *tqpair, 1781 struct nvme_tcp_pdu *pdu) 1782 { 1783 struct spdk_nvmf_tcp_req *tcp_req; 1784 struct spdk_nvme_tcp_cmd *capsule_cmd; 1785 uint32_t error_offset = 0; 1786 enum spdk_nvme_tcp_term_req_fes fes; 1787 struct spdk_nvme_cpl *rsp; 1788 1789 capsule_cmd = &pdu->hdr.capsule_cmd; 1790 tcp_req = pdu->req; 1791 assert(tcp_req != NULL); 1792 1793 /* Zero-copy requests don't support ICD */ 1794 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1795 1796 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1797 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1798 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1799 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1800 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1801 goto err; 1802 } 1803 1804 rsp = &tcp_req->req.rsp->nvme_cpl; 1805 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1806 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1807 } else { 1808 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1809 } 1810 1811 nvmf_tcp_req_process(ttransport, tcp_req); 1812 1813 return; 1814 err: 1815 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1816 } 1817 1818 static void 1819 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1820 struct spdk_nvmf_tcp_qpair *tqpair, 1821 struct nvme_tcp_pdu *pdu) 1822 { 1823 struct spdk_nvmf_tcp_req *tcp_req; 1824 uint32_t error_offset = 0; 1825 enum spdk_nvme_tcp_term_req_fes fes = 0; 1826 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1827 1828 h2c_data = &pdu->hdr.h2c_data; 1829 1830 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1831 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1832 1833 if (h2c_data->ttag > tqpair->resource_count) { 1834 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1835 tqpair->resource_count); 1836 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1837 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1838 goto err; 1839 } 1840 1841 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1842 1843 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1844 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1845 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1846 tcp_req->state); 1847 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1848 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1849 goto err; 1850 } 1851 1852 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1853 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1854 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1855 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1856 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1857 goto err; 1858 } 1859 1860 if (tcp_req->h2c_offset != h2c_data->datao) { 1861 SPDK_DEBUGLOG(nvmf_tcp, 1862 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1863 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1864 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1865 goto err; 1866 } 1867 1868 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1869 SPDK_DEBUGLOG(nvmf_tcp, 1870 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1871 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1872 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1873 goto err; 1874 } 1875 1876 pdu->req = tcp_req; 1877 1878 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1879 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1880 } 1881 1882 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1883 h2c_data->datao, h2c_data->datal); 1884 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1885 return; 1886 1887 err: 1888 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1889 } 1890 1891 static void 1892 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1893 struct spdk_nvmf_tcp_qpair *tqpair) 1894 { 1895 struct nvme_tcp_pdu *rsp_pdu; 1896 struct spdk_nvme_tcp_rsp *capsule_resp; 1897 1898 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1899 1900 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1901 assert(rsp_pdu != NULL); 1902 1903 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1904 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1905 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1906 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1907 if (tqpair->host_hdgst_enable) { 1908 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1909 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1910 } 1911 1912 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 1913 } 1914 1915 static void 1916 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1917 { 1918 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1919 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1920 struct spdk_nvmf_tcp_qpair, qpair); 1921 1922 assert(tqpair != NULL); 1923 1924 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 1925 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 1926 tcp_req->req.length); 1927 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 1928 return; 1929 } 1930 1931 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 1932 nvmf_tcp_request_free(tcp_req); 1933 } else { 1934 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1935 } 1936 } 1937 1938 static void 1939 nvmf_tcp_r2t_complete(void *cb_arg) 1940 { 1941 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1942 struct spdk_nvmf_tcp_transport *ttransport; 1943 1944 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1945 struct spdk_nvmf_tcp_transport, transport); 1946 1947 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1948 1949 if (tcp_req->h2c_offset == tcp_req->req.length) { 1950 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1951 nvmf_tcp_req_process(ttransport, tcp_req); 1952 } 1953 } 1954 1955 static void 1956 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1957 struct spdk_nvmf_tcp_req *tcp_req) 1958 { 1959 struct nvme_tcp_pdu *rsp_pdu; 1960 struct spdk_nvme_tcp_r2t_hdr *r2t; 1961 1962 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1963 assert(rsp_pdu != NULL); 1964 1965 r2t = &rsp_pdu->hdr.r2t; 1966 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1967 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1968 1969 if (tqpair->host_hdgst_enable) { 1970 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1971 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1972 } 1973 1974 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1975 r2t->ttag = tcp_req->ttag; 1976 r2t->r2to = tcp_req->h2c_offset; 1977 r2t->r2tl = tcp_req->req.length; 1978 1979 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1980 1981 SPDK_DEBUGLOG(nvmf_tcp, 1982 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1983 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1984 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 1985 } 1986 1987 static void 1988 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1989 struct spdk_nvmf_tcp_qpair *tqpair, 1990 struct nvme_tcp_pdu *pdu) 1991 { 1992 struct spdk_nvmf_tcp_req *tcp_req; 1993 struct spdk_nvme_cpl *rsp; 1994 1995 tcp_req = pdu->req; 1996 assert(tcp_req != NULL); 1997 1998 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1999 2000 tcp_req->h2c_offset += pdu->data_len; 2001 2002 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 2003 * acknowledged before moving on. */ 2004 if (tcp_req->h2c_offset == tcp_req->req.length && 2005 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 2006 /* After receiving all the h2c data, we need to check whether there is 2007 * transient transport error */ 2008 rsp = &tcp_req->req.rsp->nvme_cpl; 2009 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 2010 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2011 } else { 2012 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2013 } 2014 nvmf_tcp_req_process(ttransport, tcp_req); 2015 } 2016 } 2017 2018 static void 2019 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 2020 { 2021 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 2022 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 2023 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 2024 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 2025 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 2026 DGET32(h2c_term_req->fei)); 2027 } 2028 } 2029 2030 static void 2031 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2032 struct nvme_tcp_pdu *pdu) 2033 { 2034 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2035 uint32_t error_offset = 0; 2036 enum spdk_nvme_tcp_term_req_fes fes; 2037 2038 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 2039 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 2040 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2041 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 2042 goto end; 2043 } 2044 2045 /* set the data buffer */ 2046 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2047 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2048 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2049 return; 2050 end: 2051 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2052 } 2053 2054 static void 2055 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2056 struct nvme_tcp_pdu *pdu) 2057 { 2058 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2059 2060 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2061 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2062 } 2063 2064 static void 2065 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2066 { 2067 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2068 struct spdk_nvmf_tcp_transport, transport); 2069 2070 switch (pdu->hdr.common.pdu_type) { 2071 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2072 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2073 break; 2074 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2075 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2076 break; 2077 2078 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2079 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2080 break; 2081 2082 default: 2083 /* The code should not go to here */ 2084 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2085 break; 2086 } 2087 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2088 tqpair->tcp_pdu_working_count--; 2089 } 2090 2091 static inline void 2092 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2093 { 2094 treq->req.rsp->nvme_cpl.status.sct = sct; 2095 treq->req.rsp->nvme_cpl.status.sc = sc; 2096 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2097 } 2098 2099 static void 2100 data_crc32_calc_done(void *cb_arg, int status) 2101 { 2102 struct nvme_tcp_pdu *pdu = cb_arg; 2103 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2104 2105 /* async crc32 calculation is failed and use direct calculation to check */ 2106 if (spdk_unlikely(status)) { 2107 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2108 tqpair, pdu); 2109 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2110 } 2111 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2112 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2113 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2114 assert(pdu->req != NULL); 2115 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2116 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2117 } 2118 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2119 } 2120 2121 static void 2122 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2123 { 2124 int rc = 0; 2125 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2126 tqpair->pdu_in_progress = NULL; 2127 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2128 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2129 /* check data digest if need */ 2130 if (pdu->ddgst_enable) { 2131 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2132 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2133 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2134 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2135 if (spdk_likely(rc == 0)) { 2136 return; 2137 } 2138 } else { 2139 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2140 } 2141 data_crc32_calc_done(pdu, rc); 2142 } else { 2143 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2144 } 2145 } 2146 2147 static void 2148 nvmf_tcp_send_icresp_complete(void *cb_arg) 2149 { 2150 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2151 2152 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_RUNNING); 2153 } 2154 2155 static void 2156 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2157 struct spdk_nvmf_tcp_qpair *tqpair, 2158 struct nvme_tcp_pdu *pdu) 2159 { 2160 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2161 struct nvme_tcp_pdu *rsp_pdu; 2162 struct spdk_nvme_tcp_ic_resp *ic_resp; 2163 uint32_t error_offset = 0; 2164 enum spdk_nvme_tcp_term_req_fes fes; 2165 2166 /* Only PFV 0 is defined currently */ 2167 if (ic_req->pfv != 0) { 2168 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2169 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2170 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2171 goto end; 2172 } 2173 2174 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2175 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2176 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2177 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2178 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2179 goto end; 2180 } 2181 2182 /* MAXR2T is 0's based */ 2183 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2184 2185 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2186 if (!tqpair->host_hdgst_enable) { 2187 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2188 } 2189 2190 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2191 if (!tqpair->host_ddgst_enable) { 2192 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2193 } 2194 2195 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2196 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2197 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2198 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2199 tqpair, 2200 tqpair->recv_buf_size); 2201 /* Not fatal. */ 2202 } 2203 2204 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2205 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2206 2207 rsp_pdu = tqpair->mgmt_pdu; 2208 2209 ic_resp = &rsp_pdu->hdr.ic_resp; 2210 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2211 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2212 ic_resp->pfv = 0; 2213 ic_resp->cpda = tqpair->cpda; 2214 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2215 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2216 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2217 2218 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2219 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2220 2221 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INITIALIZING); 2222 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2223 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2224 return; 2225 end: 2226 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2227 } 2228 2229 static void 2230 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2231 struct spdk_nvmf_tcp_transport *ttransport) 2232 { 2233 struct nvme_tcp_pdu *pdu; 2234 int rc; 2235 uint32_t crc32c, error_offset = 0; 2236 enum spdk_nvme_tcp_term_req_fes fes; 2237 2238 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2239 pdu = tqpair->pdu_in_progress; 2240 2241 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2242 pdu->hdr.common.pdu_type); 2243 /* check header digest if needed */ 2244 if (pdu->has_hdgst) { 2245 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2246 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2247 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2248 if (rc == 0) { 2249 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2250 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2251 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2252 return; 2253 2254 } 2255 } 2256 2257 switch (pdu->hdr.common.pdu_type) { 2258 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2259 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2260 break; 2261 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2262 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2263 break; 2264 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2265 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2266 break; 2267 2268 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2269 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2270 break; 2271 2272 default: 2273 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2274 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2275 error_offset = 1; 2276 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2277 break; 2278 } 2279 } 2280 2281 static void 2282 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2283 { 2284 struct nvme_tcp_pdu *pdu; 2285 uint32_t error_offset = 0; 2286 enum spdk_nvme_tcp_term_req_fes fes; 2287 uint8_t expected_hlen, pdo; 2288 bool plen_error = false, pdo_error = false; 2289 2290 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2291 pdu = tqpair->pdu_in_progress; 2292 assert(pdu); 2293 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2294 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 2295 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2296 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2297 goto err; 2298 } 2299 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2300 if (pdu->hdr.common.plen != expected_hlen) { 2301 plen_error = true; 2302 } 2303 } else { 2304 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 2305 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2306 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2307 goto err; 2308 } 2309 2310 switch (pdu->hdr.common.pdu_type) { 2311 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2312 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2313 pdo = pdu->hdr.common.pdo; 2314 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2315 pdo_error = true; 2316 break; 2317 } 2318 2319 if (pdu->hdr.common.plen < expected_hlen) { 2320 plen_error = true; 2321 } 2322 break; 2323 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2324 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2325 pdo = pdu->hdr.common.pdo; 2326 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2327 pdo_error = true; 2328 break; 2329 } 2330 if (pdu->hdr.common.plen < expected_hlen) { 2331 plen_error = true; 2332 } 2333 break; 2334 2335 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2336 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2337 if ((pdu->hdr.common.plen <= expected_hlen) || 2338 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2339 plen_error = true; 2340 } 2341 break; 2342 2343 default: 2344 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2345 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2346 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2347 goto err; 2348 } 2349 } 2350 2351 if (pdu->hdr.common.hlen != expected_hlen) { 2352 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2353 pdu->hdr.common.pdu_type, 2354 expected_hlen, pdu->hdr.common.hlen, tqpair); 2355 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2356 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2357 goto err; 2358 } else if (pdo_error) { 2359 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2360 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2361 } else if (plen_error) { 2362 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2363 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2364 goto err; 2365 } else { 2366 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2367 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2368 return; 2369 } 2370 err: 2371 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2372 } 2373 2374 static int 2375 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2376 { 2377 int rc = 0; 2378 struct nvme_tcp_pdu *pdu; 2379 enum nvme_tcp_pdu_recv_state prev_state; 2380 uint32_t data_len; 2381 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2382 struct spdk_nvmf_tcp_transport, transport); 2383 2384 /* The loop here is to allow for several back-to-back state changes. */ 2385 do { 2386 prev_state = tqpair->recv_state; 2387 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2388 2389 pdu = tqpair->pdu_in_progress; 2390 assert(pdu != NULL || 2391 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2392 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2393 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2394 2395 switch (tqpair->recv_state) { 2396 /* Wait for the common header */ 2397 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2398 if (!pdu) { 2399 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2400 if (spdk_unlikely(!pdu)) { 2401 return NVME_TCP_PDU_IN_PROGRESS; 2402 } 2403 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2404 tqpair->pdu_in_progress = pdu; 2405 tqpair->tcp_pdu_working_count++; 2406 } 2407 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2408 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2409 /* FALLTHROUGH */ 2410 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2411 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 2412 return rc; 2413 } 2414 2415 rc = nvme_tcp_read_data(tqpair->sock, 2416 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2417 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2418 if (rc < 0) { 2419 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2420 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2421 break; 2422 } else if (rc > 0) { 2423 pdu->ch_valid_bytes += rc; 2424 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2425 } 2426 2427 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2428 return NVME_TCP_PDU_IN_PROGRESS; 2429 } 2430 2431 /* The command header of this PDU has now been read from the socket. */ 2432 nvmf_tcp_pdu_ch_handle(tqpair); 2433 break; 2434 /* Wait for the pdu specific header */ 2435 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2436 rc = nvme_tcp_read_data(tqpair->sock, 2437 pdu->psh_len - pdu->psh_valid_bytes, 2438 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2439 if (rc < 0) { 2440 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2441 break; 2442 } else if (rc > 0) { 2443 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2444 pdu->psh_valid_bytes += rc; 2445 } 2446 2447 if (pdu->psh_valid_bytes < pdu->psh_len) { 2448 return NVME_TCP_PDU_IN_PROGRESS; 2449 } 2450 2451 /* All header(ch, psh, head digits) of this PDU has now been read from the socket. */ 2452 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2453 break; 2454 /* Wait for the req slot */ 2455 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2456 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2457 break; 2458 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2459 /* check whether the data is valid, if not we just return */ 2460 if (!pdu->data_len) { 2461 return NVME_TCP_PDU_IN_PROGRESS; 2462 } 2463 2464 data_len = pdu->data_len; 2465 /* data digest */ 2466 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2467 tqpair->host_ddgst_enable)) { 2468 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2469 pdu->ddgst_enable = true; 2470 } 2471 2472 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2473 if (rc < 0) { 2474 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2475 break; 2476 } 2477 pdu->rw_offset += rc; 2478 2479 if (pdu->rw_offset < data_len) { 2480 return NVME_TCP_PDU_IN_PROGRESS; 2481 } 2482 2483 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2484 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2485 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2486 pdu->dif_ctx) != 0) { 2487 SPDK_ERRLOG("DIF generate failed\n"); 2488 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2489 break; 2490 } 2491 2492 /* All of this PDU has now been read from the socket. */ 2493 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2494 break; 2495 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2496 if (tqpair->tcp_pdu_working_count != 0) { 2497 return NVME_TCP_PDU_IN_PROGRESS; 2498 } 2499 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2500 break; 2501 case NVME_TCP_PDU_RECV_STATE_ERROR: 2502 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2503 return NVME_TCP_PDU_IN_PROGRESS; 2504 } 2505 return NVME_TCP_PDU_FATAL; 2506 default: 2507 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2508 abort(); 2509 break; 2510 } 2511 } while (tqpair->recv_state != prev_state); 2512 2513 return rc; 2514 } 2515 2516 static inline void * 2517 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2518 { 2519 struct spdk_nvmf_tcp_control_msg *msg; 2520 2521 assert(list); 2522 2523 msg = STAILQ_FIRST(&list->free_msgs); 2524 if (!msg) { 2525 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2526 return NULL; 2527 } 2528 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2529 return msg; 2530 } 2531 2532 static inline void 2533 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2534 { 2535 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2536 2537 assert(list); 2538 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2539 } 2540 2541 static int 2542 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2543 struct spdk_nvmf_transport *transport, 2544 struct spdk_nvmf_transport_poll_group *group) 2545 { 2546 struct spdk_nvmf_request *req = &tcp_req->req; 2547 struct spdk_nvme_cmd *cmd; 2548 struct spdk_nvme_sgl_descriptor *sgl; 2549 struct spdk_nvmf_tcp_poll_group *tgroup; 2550 enum spdk_nvme_tcp_term_req_fes fes; 2551 struct nvme_tcp_pdu *pdu; 2552 struct spdk_nvmf_tcp_qpair *tqpair; 2553 uint32_t length, error_offset = 0; 2554 2555 cmd = &req->cmd->nvme_cmd; 2556 sgl = &cmd->dptr.sgl1; 2557 2558 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2559 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2560 /* get request length from sgl */ 2561 length = sgl->unkeyed.length; 2562 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2563 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2564 length, transport->opts.max_io_size); 2565 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2566 goto fatal_err; 2567 } 2568 2569 /* fill request length and populate iovs */ 2570 req->length = length; 2571 2572 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2573 2574 if (spdk_unlikely(req->dif_enabled)) { 2575 req->dif.orig_length = length; 2576 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2577 req->dif.elba_length = length; 2578 } 2579 2580 if (nvmf_ctrlr_use_zcopy(req)) { 2581 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2582 req->data_from_pool = false; 2583 return 0; 2584 } 2585 2586 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2587 /* No available buffers. Queue this request up. */ 2588 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2589 tcp_req); 2590 return 0; 2591 } 2592 2593 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2594 tcp_req, req->iovcnt, req->iov[0].iov_base); 2595 2596 return 0; 2597 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2598 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2599 uint64_t offset = sgl->address; 2600 uint32_t max_len = transport->opts.in_capsule_data_size; 2601 2602 assert(tcp_req->has_in_capsule_data); 2603 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2604 tqpair = tcp_req->pdu->qpair; 2605 /* receiving pdu is not same with the pdu in tcp_req */ 2606 pdu = tqpair->pdu_in_progress; 2607 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2608 if (tqpair->host_ddgst_enable) { 2609 length -= SPDK_NVME_TCP_DIGEST_LEN; 2610 } 2611 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2612 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2613 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2614 length, sgl->unkeyed.length); 2615 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2616 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2617 goto fatal_err; 2618 } 2619 2620 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2621 offset, length); 2622 2623 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2624 if (spdk_unlikely(offset != 0)) { 2625 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2626 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2627 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2628 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2629 goto fatal_err; 2630 } 2631 2632 if (spdk_unlikely(length > max_len)) { 2633 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2634 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2635 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2636 2637 /* Get a buffer from dedicated list */ 2638 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2639 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2640 assert(tgroup->control_msg_list); 2641 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2642 if (!req->iov[0].iov_base) { 2643 /* No available buffers. Queue this request up. */ 2644 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2645 return 0; 2646 } 2647 } else { 2648 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2649 length, max_len); 2650 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2651 goto fatal_err; 2652 } 2653 } else { 2654 req->iov[0].iov_base = tcp_req->buf; 2655 } 2656 2657 req->length = length; 2658 req->data_from_pool = false; 2659 2660 if (spdk_unlikely(req->dif_enabled)) { 2661 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2662 req->dif.elba_length = length; 2663 } 2664 2665 req->iov[0].iov_len = length; 2666 req->iovcnt = 1; 2667 2668 return 0; 2669 } 2670 /* If we want to handle the problem here, then we can't skip the following data segment. 2671 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2672 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2673 sgl->generic.type, sgl->generic.subtype); 2674 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2675 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2676 fatal_err: 2677 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2678 return -1; 2679 } 2680 2681 static inline enum spdk_nvme_media_error_status_code 2682 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2683 enum spdk_nvme_media_error_status_code result; 2684 2685 switch (err_type) 2686 { 2687 case SPDK_DIF_REFTAG_ERROR: 2688 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2689 break; 2690 case SPDK_DIF_APPTAG_ERROR: 2691 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2692 break; 2693 case SPDK_DIF_GUARD_ERROR: 2694 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2695 break; 2696 default: 2697 SPDK_UNREACHABLE(); 2698 break; 2699 } 2700 2701 return result; 2702 } 2703 2704 static void 2705 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2706 struct spdk_nvmf_tcp_req *tcp_req) 2707 { 2708 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2709 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2710 struct nvme_tcp_pdu *rsp_pdu; 2711 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2712 uint32_t plen, pdo, alignment; 2713 int rc; 2714 2715 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2716 2717 rsp_pdu = tcp_req->pdu; 2718 assert(rsp_pdu != NULL); 2719 2720 c2h_data = &rsp_pdu->hdr.c2h_data; 2721 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2722 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2723 2724 if (tqpair->host_hdgst_enable) { 2725 plen += SPDK_NVME_TCP_DIGEST_LEN; 2726 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2727 } 2728 2729 /* set the psh */ 2730 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2731 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2732 c2h_data->datao = tcp_req->pdu->rw_offset; 2733 2734 /* set the padding */ 2735 rsp_pdu->padding_len = 0; 2736 pdo = plen; 2737 if (tqpair->cpda) { 2738 alignment = (tqpair->cpda + 1) << 2; 2739 if (plen % alignment != 0) { 2740 pdo = (plen + alignment) / alignment * alignment; 2741 rsp_pdu->padding_len = pdo - plen; 2742 plen = pdo; 2743 } 2744 } 2745 2746 c2h_data->common.pdo = pdo; 2747 plen += c2h_data->datal; 2748 if (tqpair->host_ddgst_enable) { 2749 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2750 plen += SPDK_NVME_TCP_DIGEST_LEN; 2751 } 2752 2753 c2h_data->common.plen = plen; 2754 2755 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2756 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2757 } 2758 2759 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2760 c2h_data->datao, c2h_data->datal); 2761 2762 2763 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2764 /* Need to send the capsule response if response is not all 0 */ 2765 if (ttransport->tcp_opts.c2h_success && 2766 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2767 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2768 } 2769 2770 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2771 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2772 struct spdk_dif_error err_blk = {}; 2773 uint32_t mapped_length = 0; 2774 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2775 uint32_t ddgst_len = 0; 2776 2777 if (tqpair->host_ddgst_enable) { 2778 /* Data digest consumes additional iov entry */ 2779 available_iovs--; 2780 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2781 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2782 c2h_data->common.plen -= ddgst_len; 2783 } 2784 /* Temp call to estimate if data can be described by limited number of iovs. 2785 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2786 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2787 false, &mapped_length); 2788 2789 if (mapped_length != c2h_data->common.plen) { 2790 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2791 SPDK_DEBUGLOG(nvmf_tcp, 2792 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2793 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2794 c2h_data->common.plen = mapped_length; 2795 2796 /* Rebuild pdu->data_iov since data length is changed */ 2797 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2798 c2h_data->datal); 2799 2800 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2801 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2802 } 2803 2804 c2h_data->common.plen += ddgst_len; 2805 2806 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2807 2808 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2809 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2810 if (rc != 0) { 2811 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2812 err_blk.err_type, err_blk.err_offset); 2813 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2814 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2815 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2816 return; 2817 } 2818 } 2819 2820 rsp_pdu->rw_offset += c2h_data->datal; 2821 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2822 } 2823 2824 static void 2825 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2826 struct spdk_nvmf_tcp_req *tcp_req) 2827 { 2828 nvmf_tcp_req_pdu_init(tcp_req); 2829 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2830 } 2831 2832 static int 2833 request_transfer_out(struct spdk_nvmf_request *req) 2834 { 2835 struct spdk_nvmf_tcp_req *tcp_req; 2836 struct spdk_nvmf_qpair *qpair; 2837 struct spdk_nvmf_tcp_qpair *tqpair; 2838 struct spdk_nvme_cpl *rsp; 2839 2840 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2841 2842 qpair = req->qpair; 2843 rsp = &req->rsp->nvme_cpl; 2844 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2845 2846 /* Advance our sq_head pointer */ 2847 if (qpair->sq_head == qpair->sq_head_max) { 2848 qpair->sq_head = 0; 2849 } else { 2850 qpair->sq_head++; 2851 } 2852 rsp->sqhd = qpair->sq_head; 2853 2854 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2855 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2856 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2857 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2858 } else { 2859 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2860 } 2861 2862 return 0; 2863 } 2864 2865 static void 2866 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2867 struct spdk_nvmf_tcp_qpair *tqpair, 2868 struct spdk_nvmf_tcp_req *tcp_req) 2869 { 2870 enum spdk_nvme_cmd_fuse last, next; 2871 2872 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2873 next = tcp_req->cmd.fuse; 2874 2875 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2876 2877 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2878 return; 2879 } 2880 2881 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2882 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2883 /* This is a valid pair of fused commands. Point them at each other 2884 * so they can be submitted consecutively once ready to be executed. 2885 */ 2886 tqpair->fused_first->fused_pair = tcp_req; 2887 tcp_req->fused_pair = tqpair->fused_first; 2888 tqpair->fused_first = NULL; 2889 return; 2890 } else { 2891 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 2892 tqpair->fused_first->fused_failed = true; 2893 2894 /* 2895 * If the last req is in READY_TO_EXECUTE state, then call 2896 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 2897 */ 2898 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 2899 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 2900 } 2901 2902 tqpair->fused_first = NULL; 2903 } 2904 } 2905 2906 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 2907 /* Set tqpair->fused_first here so that we know to check that the next request 2908 * is a SECOND (and to fail this one if it isn't). 2909 */ 2910 tqpair->fused_first = tcp_req; 2911 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2912 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 2913 tcp_req->fused_failed = true; 2914 } 2915 } 2916 2917 static bool 2918 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2919 struct spdk_nvmf_tcp_req *tcp_req) 2920 { 2921 struct spdk_nvmf_tcp_qpair *tqpair; 2922 uint32_t plen; 2923 struct nvme_tcp_pdu *pdu; 2924 enum spdk_nvmf_tcp_req_state prev_state; 2925 bool progress = false; 2926 struct spdk_nvmf_transport *transport = &ttransport->transport; 2927 struct spdk_nvmf_transport_poll_group *group; 2928 struct spdk_nvmf_tcp_poll_group *tgroup; 2929 2930 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2931 group = &tqpair->group->group; 2932 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2933 2934 /* If the qpair is not active, we need to abort the outstanding requests. */ 2935 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 2936 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2937 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2938 } 2939 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2940 } 2941 2942 /* The loop here is to allow for several back-to-back state changes. */ 2943 do { 2944 prev_state = tcp_req->state; 2945 2946 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2947 tqpair); 2948 2949 switch (tcp_req->state) { 2950 case TCP_REQUEST_STATE_FREE: 2951 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2952 * to escape this state. */ 2953 break; 2954 case TCP_REQUEST_STATE_NEW: 2955 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 2956 tqpair->qpair.queue_depth); 2957 2958 /* copy the cmd from the receive pdu */ 2959 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 2960 2961 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2962 tcp_req->req.dif_enabled = true; 2963 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 2964 } 2965 2966 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 2967 2968 /* The next state transition depends on the data transfer needs of this request. */ 2969 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2970 2971 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2972 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 2973 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2974 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2975 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2976 break; 2977 } 2978 2979 /* If no data to transfer, ready to execute. */ 2980 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2981 /* Reset the tqpair receiving pdu state */ 2982 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2983 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2984 break; 2985 } 2986 2987 pdu = tqpair->pdu_in_progress; 2988 plen = pdu->hdr.common.hlen; 2989 if (tqpair->host_hdgst_enable) { 2990 plen += SPDK_NVME_TCP_DIGEST_LEN; 2991 } 2992 if (pdu->hdr.common.plen != plen) { 2993 tcp_req->has_in_capsule_data = true; 2994 } else { 2995 /* Data is transmitted by C2H PDUs */ 2996 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2997 } 2998 2999 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 3000 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 3001 break; 3002 case TCP_REQUEST_STATE_NEED_BUFFER: 3003 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 3004 (uintptr_t)tcp_req); 3005 3006 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 3007 3008 if (!tcp_req->has_in_capsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 3009 SPDK_DEBUGLOG(nvmf_tcp, 3010 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 3011 tcp_req, tqpair); 3012 /* This request needs to wait in line to obtain a buffer */ 3013 break; 3014 } 3015 3016 /* Try to get a data buffer */ 3017 if (nvmf_tcp_req_parse_sgl(tcp_req, transport, group) < 0) { 3018 break; 3019 } 3020 3021 /* Get a zcopy buffer if the request can be serviced through zcopy */ 3022 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3023 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3024 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3025 tcp_req->req.length = tcp_req->req.dif.elba_length; 3026 } 3027 3028 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 3029 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 3030 spdk_nvmf_request_zcopy_start(&tcp_req->req); 3031 break; 3032 } 3033 3034 if (tcp_req->req.iovcnt < 1) { 3035 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 3036 tcp_req, tqpair); 3037 /* No buffers available. */ 3038 break; 3039 } 3040 3041 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 3042 3043 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 3044 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3045 if (tcp_req->req.data_from_pool) { 3046 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3047 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3048 } else { 3049 struct nvme_tcp_pdu *pdu; 3050 3051 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3052 3053 pdu = tqpair->pdu_in_progress; 3054 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3055 tqpair); 3056 /* No need to send r2t, contained in the capsuled data */ 3057 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3058 0, tcp_req->req.length); 3059 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3060 } 3061 break; 3062 } 3063 3064 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3065 break; 3066 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3067 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3068 (uintptr_t)tcp_req); 3069 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3070 * to escape this state. */ 3071 break; 3072 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3073 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3074 (uintptr_t)tcp_req); 3075 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3076 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3077 tcp_req, tqpair); 3078 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3079 break; 3080 } 3081 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3082 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3083 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3084 } else { 3085 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3086 } 3087 break; 3088 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3089 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3090 (uintptr_t)tcp_req); 3091 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3092 break; 3093 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3094 3095 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3096 0, (uintptr_t)tcp_req); 3097 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3098 * to escape this state. */ 3099 break; 3100 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3101 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3102 (uintptr_t)tcp_req); 3103 3104 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3105 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3106 tcp_req->req.length = tcp_req->req.dif.elba_length; 3107 } 3108 3109 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3110 if (tcp_req->fused_failed) { 3111 /* This request failed FUSED semantics. Fail it immediately, without 3112 * even sending it to the target layer. 3113 */ 3114 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3115 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3116 break; 3117 } 3118 3119 if (tcp_req->fused_pair == NULL || 3120 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3121 /* This request is ready to execute, but either we don't know yet if it's 3122 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3123 * or the other request of this fused pair isn't ready to execute. So 3124 * break here and this request will get processed later either when the 3125 * other request is ready or we find that this request isn't valid. 3126 */ 3127 break; 3128 } 3129 } 3130 3131 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3132 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3133 /* If we get to this point, and this request is a fused command, we know that 3134 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3135 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3136 * request, and the other request of the fused pair, in the correct order. 3137 * Also clear the ->fused_pair pointers on both requests, since after this point 3138 * we no longer need to maintain the relationship between these two requests. 3139 */ 3140 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3141 assert(tcp_req->fused_pair != NULL); 3142 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3143 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3144 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3145 tcp_req->fused_pair->fused_pair = NULL; 3146 tcp_req->fused_pair = NULL; 3147 } 3148 spdk_nvmf_request_exec(&tcp_req->req); 3149 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3150 assert(tcp_req->fused_pair != NULL); 3151 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3152 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3153 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3154 tcp_req->fused_pair->fused_pair = NULL; 3155 tcp_req->fused_pair = NULL; 3156 } 3157 } else { 3158 /* For zero-copy, only requests with data coming from host to the 3159 * controller can end up here. */ 3160 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3161 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3162 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3163 } 3164 3165 break; 3166 case TCP_REQUEST_STATE_EXECUTING: 3167 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3168 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3169 * to escape this state. */ 3170 break; 3171 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3172 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3173 (uintptr_t)tcp_req); 3174 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3175 * to escape this state. */ 3176 break; 3177 case TCP_REQUEST_STATE_EXECUTED: 3178 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3179 3180 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3181 tcp_req->req.length = tcp_req->req.dif.orig_length; 3182 } 3183 3184 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3185 break; 3186 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3187 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3188 (uintptr_t)tcp_req); 3189 if (request_transfer_out(&tcp_req->req) != 0) { 3190 assert(0); /* No good way to handle this currently */ 3191 } 3192 break; 3193 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3194 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3195 0, (uintptr_t)tcp_req); 3196 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3197 * to escape this state. */ 3198 break; 3199 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3200 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3201 (uintptr_t)tcp_req); 3202 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3203 * to escape this state. */ 3204 break; 3205 case TCP_REQUEST_STATE_COMPLETED: 3206 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3207 tqpair->qpair.queue_depth); 3208 /* If there's an outstanding PDU sent to the host, the request is completed 3209 * due to the qpair being disconnected. We must delay the completion until 3210 * that write is done to avoid freeing the request twice. */ 3211 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3212 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3213 "write on req=%p\n", tcp_req); 3214 /* This can only happen for zcopy requests */ 3215 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3216 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3217 break; 3218 } 3219 3220 if (tcp_req->req.data_from_pool) { 3221 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3222 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3223 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3224 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3225 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3226 assert(tgroup->control_msg_list); 3227 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3228 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3229 tcp_req->req.iov[0].iov_base); 3230 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3231 /* If the request has an unreleased zcopy bdev_io, it's either a 3232 * read, a failed write, or the qpair is being disconnected */ 3233 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3234 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3235 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3236 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3237 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3238 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3239 break; 3240 } 3241 tcp_req->req.length = 0; 3242 tcp_req->req.iovcnt = 0; 3243 tcp_req->fused_failed = false; 3244 if (tcp_req->fused_pair) { 3245 /* This req was part of a valid fused pair, but failed before it got to 3246 * READ_TO_EXECUTE state. This means we need to fail the other request 3247 * in the pair, because it is no longer part of a valid pair. If the pair 3248 * already reached READY_TO_EXECUTE state, we need to kick it. 3249 */ 3250 tcp_req->fused_pair->fused_failed = true; 3251 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3252 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3253 } 3254 tcp_req->fused_pair = NULL; 3255 } 3256 3257 nvmf_tcp_req_put(tqpair, tcp_req); 3258 break; 3259 case TCP_REQUEST_NUM_STATES: 3260 default: 3261 assert(0); 3262 break; 3263 } 3264 3265 if (tcp_req->state != prev_state) { 3266 progress = true; 3267 } 3268 } while (tcp_req->state != prev_state); 3269 3270 return progress; 3271 } 3272 3273 static void 3274 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3275 { 3276 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3277 int rc; 3278 3279 assert(tqpair != NULL); 3280 rc = nvmf_tcp_sock_process(tqpair); 3281 3282 /* If there was a new socket error, disconnect */ 3283 if (rc < 0) { 3284 nvmf_tcp_qpair_disconnect(tqpair); 3285 } 3286 } 3287 3288 static int 3289 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3290 struct spdk_nvmf_qpair *qpair) 3291 { 3292 struct spdk_nvmf_tcp_poll_group *tgroup; 3293 struct spdk_nvmf_tcp_qpair *tqpair; 3294 int rc; 3295 3296 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3297 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3298 3299 rc = nvmf_tcp_qpair_sock_init(tqpair); 3300 if (rc != 0) { 3301 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3302 return -1; 3303 } 3304 3305 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3306 if (rc < 0) { 3307 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3308 return -1; 3309 } 3310 3311 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3312 if (rc < 0) { 3313 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3314 return -1; 3315 } 3316 3317 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3318 nvmf_tcp_sock_cb, tqpair); 3319 if (rc != 0) { 3320 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3321 spdk_strerror(errno), errno); 3322 return -1; 3323 } 3324 3325 tqpair->group = tgroup; 3326 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_INVALID); 3327 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3328 3329 return 0; 3330 } 3331 3332 static int 3333 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3334 struct spdk_nvmf_qpair *qpair) 3335 { 3336 struct spdk_nvmf_tcp_poll_group *tgroup; 3337 struct spdk_nvmf_tcp_qpair *tqpair; 3338 int rc; 3339 3340 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3341 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3342 3343 assert(tqpair->group == tgroup); 3344 3345 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3346 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3347 /* Change the state to move the qpair from the await_req list to the main list 3348 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3349 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3350 } 3351 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3352 3353 /* Try to force out any pending writes */ 3354 spdk_sock_flush(tqpair->sock); 3355 3356 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3357 if (rc != 0) { 3358 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3359 spdk_strerror(errno), errno); 3360 } 3361 3362 return rc; 3363 } 3364 3365 static int 3366 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3367 { 3368 struct spdk_nvmf_tcp_transport *ttransport; 3369 struct spdk_nvmf_tcp_req *tcp_req; 3370 3371 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3372 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3373 3374 switch (tcp_req->state) { 3375 case TCP_REQUEST_STATE_EXECUTING: 3376 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3377 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3378 break; 3379 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3380 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3381 break; 3382 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3383 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3384 break; 3385 default: 3386 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3387 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3388 assert(0 && "Unexpected request state"); 3389 break; 3390 } 3391 3392 nvmf_tcp_req_process(ttransport, tcp_req); 3393 3394 return 0; 3395 } 3396 3397 static void 3398 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3399 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3400 { 3401 struct spdk_nvmf_tcp_qpair *tqpair; 3402 3403 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3404 3405 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3406 3407 assert(tqpair->fini_cb_fn == NULL); 3408 tqpair->fini_cb_fn = cb_fn; 3409 tqpair->fini_cb_arg = cb_arg; 3410 3411 nvmf_tcp_qpair_set_state(tqpair, NVME_TCP_QPAIR_STATE_EXITED); 3412 nvmf_tcp_qpair_destroy(tqpair); 3413 } 3414 3415 static int 3416 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3417 { 3418 struct spdk_nvmf_tcp_poll_group *tgroup; 3419 int num_events, rc = 0, rc2; 3420 struct spdk_nvmf_request *req, *req_tmp; 3421 struct spdk_nvmf_tcp_req *tcp_req; 3422 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3423 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 3424 struct spdk_nvmf_tcp_transport, transport); 3425 3426 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3427 3428 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3429 return 0; 3430 } 3431 3432 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 3433 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3434 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 3435 break; 3436 } 3437 } 3438 3439 num_events = spdk_sock_group_poll(tgroup->sock_group); 3440 if (spdk_unlikely(num_events < 0)) { 3441 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3442 } 3443 3444 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3445 rc2 = nvmf_tcp_sock_process(tqpair); 3446 3447 /* If there was a new socket error, disconnect */ 3448 if (spdk_unlikely(rc2 < 0)) { 3449 nvmf_tcp_qpair_disconnect(tqpair); 3450 if (rc == 0) { 3451 rc = rc2; 3452 } 3453 } 3454 } 3455 3456 return rc == 0 ? num_events : rc; 3457 } 3458 3459 static int 3460 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3461 struct spdk_nvme_transport_id *trid, bool peer) 3462 { 3463 struct spdk_nvmf_tcp_qpair *tqpair; 3464 uint16_t port; 3465 3466 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3467 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3468 3469 if (peer) { 3470 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3471 port = tqpair->initiator_port; 3472 } else { 3473 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3474 port = tqpair->target_port; 3475 } 3476 3477 if (spdk_sock_is_ipv4(tqpair->sock)) { 3478 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3479 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3480 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3481 } else { 3482 return -1; 3483 } 3484 3485 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3486 return 0; 3487 } 3488 3489 static int 3490 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3491 struct spdk_nvme_transport_id *trid) 3492 { 3493 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3494 } 3495 3496 static int 3497 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3498 struct spdk_nvme_transport_id *trid) 3499 { 3500 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3501 } 3502 3503 static int 3504 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3505 struct spdk_nvme_transport_id *trid) 3506 { 3507 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3508 } 3509 3510 static void 3511 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3512 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3513 { 3514 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3515 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3516 3517 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3518 } 3519 3520 static int 3521 _nvmf_tcp_qpair_abort_request(void *ctx) 3522 { 3523 struct spdk_nvmf_request *req = ctx; 3524 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3525 struct spdk_nvmf_tcp_req, req); 3526 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3527 struct spdk_nvmf_tcp_qpair, qpair); 3528 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3529 struct spdk_nvmf_tcp_transport, transport); 3530 int rc; 3531 3532 spdk_poller_unregister(&req->poller); 3533 3534 switch (tcp_req_to_abort->state) { 3535 case TCP_REQUEST_STATE_EXECUTING: 3536 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3537 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3538 rc = nvmf_ctrlr_abort_request(req); 3539 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3540 return SPDK_POLLER_BUSY; 3541 } 3542 break; 3543 3544 case TCP_REQUEST_STATE_NEED_BUFFER: 3545 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 3546 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 3547 3548 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3549 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3550 break; 3551 3552 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3553 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3554 if (spdk_get_ticks() < req->timeout_tsc) { 3555 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3556 return SPDK_POLLER_BUSY; 3557 } 3558 break; 3559 3560 default: 3561 /* Requests in other states are either un-abortable (e.g. 3562 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3563 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3564 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3565 * abort request with the bit0 of dword0 set (command not aborted). 3566 */ 3567 break; 3568 } 3569 3570 spdk_nvmf_request_complete(req); 3571 return SPDK_POLLER_BUSY; 3572 } 3573 3574 static void 3575 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3576 struct spdk_nvmf_request *req) 3577 { 3578 struct spdk_nvmf_tcp_qpair *tqpair; 3579 struct spdk_nvmf_tcp_transport *ttransport; 3580 struct spdk_nvmf_transport *transport; 3581 uint16_t cid; 3582 uint32_t i; 3583 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3584 3585 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3586 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3587 transport = &ttransport->transport; 3588 3589 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3590 3591 for (i = 0; i < tqpair->resource_count; i++) { 3592 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3593 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3594 tcp_req_to_abort = &tqpair->reqs[i]; 3595 break; 3596 } 3597 } 3598 3599 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3600 3601 if (tcp_req_to_abort == NULL) { 3602 spdk_nvmf_request_complete(req); 3603 return; 3604 } 3605 3606 req->req_to_abort = &tcp_req_to_abort->req; 3607 req->timeout_tsc = spdk_get_ticks() + 3608 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3609 req->poller = NULL; 3610 3611 _nvmf_tcp_qpair_abort_request(req); 3612 } 3613 3614 struct tcp_subsystem_add_host_opts { 3615 char *psk; 3616 }; 3617 3618 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3619 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3620 }; 3621 3622 static int 3623 tcp_load_psk(const char *fname, char *buf, size_t bufsz) 3624 { 3625 FILE *psk_file; 3626 struct stat statbuf; 3627 int rc; 3628 3629 if (stat(fname, &statbuf) != 0) { 3630 SPDK_ERRLOG("Could not read permissions for PSK file\n"); 3631 return -EACCES; 3632 } 3633 3634 if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) { 3635 SPDK_ERRLOG("Incorrect permissions for PSK file\n"); 3636 return -EPERM; 3637 } 3638 if ((size_t)statbuf.st_size > bufsz) { 3639 SPDK_ERRLOG("Invalid PSK: too long\n"); 3640 return -EINVAL; 3641 } 3642 psk_file = fopen(fname, "r"); 3643 if (psk_file == NULL) { 3644 SPDK_ERRLOG("Could not open PSK file\n"); 3645 return -EINVAL; 3646 } 3647 3648 rc = fread(buf, 1, statbuf.st_size, psk_file); 3649 if (rc != statbuf.st_size) { 3650 SPDK_ERRLOG("Failed to read PSK\n"); 3651 fclose(psk_file); 3652 return -EINVAL; 3653 } 3654 3655 fclose(psk_file); 3656 return 0; 3657 } 3658 3659 SPDK_LOG_DEPRECATION_REGISTER(nvmf_tcp_psk_path, "PSK path", "v24.09", 0); 3660 3661 static int 3662 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3663 const struct spdk_nvmf_subsystem *subsystem, 3664 const char *hostnqn, 3665 const struct spdk_json_val *transport_specific) 3666 { 3667 struct tcp_subsystem_add_host_opts opts; 3668 struct spdk_nvmf_tcp_transport *ttransport; 3669 struct tcp_psk_entry *tmp, *entry = NULL; 3670 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3671 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3672 uint8_t tls_cipher_suite; 3673 int rc = 0; 3674 uint8_t psk_retained_hash; 3675 uint64_t psk_configured_size; 3676 3677 if (transport_specific == NULL) { 3678 return 0; 3679 } 3680 3681 assert(transport != NULL); 3682 assert(subsystem != NULL); 3683 3684 memset(&opts, 0, sizeof(opts)); 3685 3686 /* Decode PSK (either name of a key or file path) */ 3687 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3688 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3689 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3690 return -EINVAL; 3691 } 3692 3693 if (opts.psk == NULL) { 3694 return 0; 3695 } 3696 3697 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3698 if (entry == NULL) { 3699 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3700 rc = -ENOMEM; 3701 goto end; 3702 } 3703 3704 entry->key = spdk_keyring_get_key(opts.psk); 3705 if (entry->key != NULL) { 3706 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3707 if (rc < 0) { 3708 SPDK_ERRLOG("Failed to retrieve PSK '%s'\n", opts.psk); 3709 rc = -EINVAL; 3710 goto end; 3711 } 3712 } else { 3713 if (strlen(opts.psk) >= sizeof(entry->psk)) { 3714 SPDK_ERRLOG("PSK path too long\n"); 3715 rc = -EINVAL; 3716 goto end; 3717 } 3718 3719 rc = tcp_load_psk(opts.psk, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3720 if (rc) { 3721 SPDK_ERRLOG("Could not retrieve PSK from file\n"); 3722 goto end; 3723 } 3724 3725 SPDK_LOG_DEPRECATED(nvmf_tcp_psk_path); 3726 } 3727 3728 /* Parse PSK interchange to get length of base64 encoded data. 3729 * This is then used to decide which cipher suite should be used 3730 * to generate PSK identity and TLS PSK later on. */ 3731 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3732 &psk_configured_size, &psk_retained_hash); 3733 if (rc < 0) { 3734 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3735 goto end; 3736 } 3737 3738 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3739 * This check also ensures that psk_configured_size is smaller than 3740 * psk_retained buffer size. */ 3741 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3742 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3743 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3744 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3745 } else { 3746 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3747 rc = -EINVAL; 3748 goto end; 3749 } 3750 3751 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3752 /* Generate PSK identity. */ 3753 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3754 subsystem->subnqn, tls_cipher_suite); 3755 if (rc) { 3756 rc = -EINVAL; 3757 goto end; 3758 } 3759 /* Check if PSK identity entry already exists. */ 3760 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3761 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3762 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3763 rc = -EEXIST; 3764 goto end; 3765 } 3766 } 3767 3768 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3769 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3770 rc = -EINVAL; 3771 goto end; 3772 } 3773 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3774 SPDK_ERRLOG("Could not write subnqn string!\n"); 3775 rc = -EINVAL; 3776 goto end; 3777 } 3778 3779 entry->tls_cipher_suite = tls_cipher_suite; 3780 3781 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3782 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3783 /* Psk configured is either 32 or 48 bytes long. */ 3784 memcpy(entry->psk, psk_configured, psk_configured_size); 3785 entry->psk_size = psk_configured_size; 3786 } else { 3787 /* Derive retained PSK. */ 3788 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3789 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3790 if (rc < 0) { 3791 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3792 goto end; 3793 } 3794 entry->psk_size = rc; 3795 } 3796 3797 if (entry->key == NULL) { 3798 rc = snprintf(entry->psk_path, sizeof(entry->psk_path), "%s", opts.psk); 3799 if (rc < 0 || (size_t)rc >= sizeof(entry->psk_path)) { 3800 SPDK_ERRLOG("Could not save PSK path!\n"); 3801 rc = -ENAMETOOLONG; 3802 goto end; 3803 } 3804 } 3805 3806 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3807 rc = 0; 3808 3809 end: 3810 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3811 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3812 3813 free(opts.psk); 3814 if (rc != 0) { 3815 nvmf_tcp_free_psk_entry(entry); 3816 } 3817 3818 return rc; 3819 } 3820 3821 static void 3822 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3823 const struct spdk_nvmf_subsystem *subsystem, 3824 const char *hostnqn) 3825 { 3826 struct spdk_nvmf_tcp_transport *ttransport; 3827 struct tcp_psk_entry *entry, *tmp; 3828 3829 assert(transport != NULL); 3830 assert(subsystem != NULL); 3831 3832 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3833 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3834 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3835 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3836 TAILQ_REMOVE(&ttransport->psks, entry, link); 3837 nvmf_tcp_free_psk_entry(entry); 3838 break; 3839 } 3840 } 3841 } 3842 3843 static void 3844 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3845 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3846 struct spdk_json_write_ctx *w) 3847 { 3848 struct spdk_nvmf_tcp_transport *ttransport; 3849 struct tcp_psk_entry *entry; 3850 3851 assert(transport != NULL); 3852 assert(subsystem != NULL); 3853 3854 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3855 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3856 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3857 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3858 spdk_json_write_named_string(w, "psk", entry->key ? 3859 spdk_key_get_name(entry->key) : entry->psk_path); 3860 break; 3861 } 3862 } 3863 } 3864 3865 static void 3866 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3867 { 3868 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3869 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3870 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3871 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3872 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3873 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3874 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3875 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3876 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3877 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3878 opts->transport_specific = NULL; 3879 } 3880 3881 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3882 .name = "TCP", 3883 .type = SPDK_NVME_TRANSPORT_TCP, 3884 .opts_init = nvmf_tcp_opts_init, 3885 .create = nvmf_tcp_create, 3886 .dump_opts = nvmf_tcp_dump_opts, 3887 .destroy = nvmf_tcp_destroy, 3888 3889 .listen = nvmf_tcp_listen, 3890 .stop_listen = nvmf_tcp_stop_listen, 3891 3892 .listener_discover = nvmf_tcp_discover, 3893 3894 .poll_group_create = nvmf_tcp_poll_group_create, 3895 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3896 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3897 .poll_group_add = nvmf_tcp_poll_group_add, 3898 .poll_group_remove = nvmf_tcp_poll_group_remove, 3899 .poll_group_poll = nvmf_tcp_poll_group_poll, 3900 3901 .req_free = nvmf_tcp_req_free, 3902 .req_complete = nvmf_tcp_req_complete, 3903 3904 .qpair_fini = nvmf_tcp_close_qpair, 3905 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3906 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3907 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3908 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3909 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3910 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3911 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3912 }; 3913 3914 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3915 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3916