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 #include "transport.h" 26 27 #include "spdk_internal/trace_defs.h" 28 29 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 30 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 31 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 32 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 33 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 34 35 #define SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH 2 36 #define SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH 65535 37 #define SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH 2 38 #define SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH 4096 39 40 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH 128 41 #define SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH 128 42 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 43 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 44 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 45 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 46 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 47 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE UINT32_MAX 48 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 49 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 50 51 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 52 static bool g_tls_log = false; 53 54 /* spdk nvmf related structure */ 55 enum spdk_nvmf_tcp_req_state { 56 57 /* The request is not currently in use */ 58 TCP_REQUEST_STATE_FREE = 0, 59 60 /* Initial state when request first received */ 61 TCP_REQUEST_STATE_NEW = 1, 62 63 /* The request is queued until a data buffer is available. */ 64 TCP_REQUEST_STATE_NEED_BUFFER = 2, 65 66 /* The request has the data buffer available */ 67 TCP_REQUEST_STATE_HAVE_BUFFER = 3, 68 69 /* The request is waiting for zcopy_start to finish */ 70 TCP_REQUEST_STATE_AWAITING_ZCOPY_START = 4, 71 72 /* The request has received a zero-copy buffer */ 73 TCP_REQUEST_STATE_ZCOPY_START_COMPLETED = 5, 74 75 /* The request is currently transferring data from the host to the controller. */ 76 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER = 6, 77 78 /* The request is waiting for the R2T send acknowledgement. */ 79 TCP_REQUEST_STATE_AWAITING_R2T_ACK = 7, 80 81 /* The request is ready to execute at the block device */ 82 TCP_REQUEST_STATE_READY_TO_EXECUTE = 8, 83 84 /* The request is currently executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTING = 9, 86 87 /* The request is waiting for zcopy buffers to be committed */ 88 TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT = 10, 89 90 /* The request finished executing at the block device */ 91 TCP_REQUEST_STATE_EXECUTED = 11, 92 93 /* The request is ready to send a completion */ 94 TCP_REQUEST_STATE_READY_TO_COMPLETE = 12, 95 96 /* The request is currently transferring final pdus from the controller to the host. */ 97 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST = 13, 98 99 /* The request is waiting for zcopy buffers to be released (without committing) */ 100 TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE = 14, 101 102 /* The request completed and can be marked free. */ 103 TCP_REQUEST_STATE_COMPLETED = 15, 104 105 /* Terminator */ 106 TCP_REQUEST_NUM_STATES, 107 }; 108 109 enum nvmf_tcp_qpair_state { 110 NVMF_TCP_QPAIR_STATE_INVALID = 0, 111 NVMF_TCP_QPAIR_STATE_INITIALIZING = 1, 112 NVMF_TCP_QPAIR_STATE_RUNNING = 2, 113 NVMF_TCP_QPAIR_STATE_EXITING = 3, 114 NVMF_TCP_QPAIR_STATE_EXITED = 4, 115 }; 116 117 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 118 "Invalid PDU Header Field", 119 "PDU Sequence Error", 120 "Header Digiest Error", 121 "Data Transfer Out of Range", 122 "R2T Limit Exceeded", 123 "Unsupported parameter", 124 }; 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_owner_type(OWNER_TYPE_NVMF_TCP, 't'); 129 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 130 spdk_trace_register_description("TCP_REQ_NEW", 131 TRACE_TCP_REQUEST_STATE_NEW, 132 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 1, 133 SPDK_TRACE_ARG_TYPE_INT, "qd"); 134 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 135 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 136 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 137 SPDK_TRACE_ARG_TYPE_INT, ""); 138 spdk_trace_register_description("TCP_REQ_HAVE_BUFFER", 139 TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, 140 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 141 SPDK_TRACE_ARG_TYPE_INT, ""); 142 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_START", 143 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, 144 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 145 SPDK_TRACE_ARG_TYPE_INT, ""); 146 spdk_trace_register_description("TCP_REQ_ZCPY_START_CPL", 147 TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, 148 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 149 SPDK_TRACE_ARG_TYPE_INT, ""); 150 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 151 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 152 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 153 SPDK_TRACE_ARG_TYPE_INT, ""); 154 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 155 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 156 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 157 SPDK_TRACE_ARG_TYPE_INT, ""); 158 spdk_trace_register_description("TCP_REQ_EXECUTING", 159 TRACE_TCP_REQUEST_STATE_EXECUTING, 160 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 161 SPDK_TRACE_ARG_TYPE_INT, ""); 162 spdk_trace_register_description("TCP_REQ_WAIT_ZCPY_CMT", 163 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, 164 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 165 SPDK_TRACE_ARG_TYPE_INT, ""); 166 spdk_trace_register_description("TCP_REQ_EXECUTED", 167 TRACE_TCP_REQUEST_STATE_EXECUTED, 168 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 169 SPDK_TRACE_ARG_TYPE_INT, ""); 170 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 171 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 172 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 173 SPDK_TRACE_ARG_TYPE_INT, ""); 174 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 175 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 176 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 177 SPDK_TRACE_ARG_TYPE_INT, ""); 178 spdk_trace_register_description("TCP_REQ_AWAIT_ZCPY_RLS", 179 TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, 180 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 181 SPDK_TRACE_ARG_TYPE_INT, ""); 182 spdk_trace_register_description("TCP_REQ_COMPLETED", 183 TRACE_TCP_REQUEST_STATE_COMPLETED, 184 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 185 SPDK_TRACE_ARG_TYPE_INT, "qd"); 186 spdk_trace_register_description("TCP_READ_DONE", 187 TRACE_TCP_READ_FROM_SOCKET_DONE, 188 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 189 SPDK_TRACE_ARG_TYPE_INT, ""); 190 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 191 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 192 OWNER_TYPE_NVMF_TCP, OBJECT_NVMF_TCP_IO, 0, 193 SPDK_TRACE_ARG_TYPE_INT, ""); 194 195 spdk_trace_register_description("TCP_QP_CREATE", TRACE_TCP_QP_CREATE, 196 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 197 SPDK_TRACE_ARG_TYPE_INT, ""); 198 spdk_trace_register_description("TCP_QP_SOCK_INIT", TRACE_TCP_QP_SOCK_INIT, 199 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 200 SPDK_TRACE_ARG_TYPE_INT, ""); 201 spdk_trace_register_description("TCP_QP_STATE_CHANGE", TRACE_TCP_QP_STATE_CHANGE, 202 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 203 SPDK_TRACE_ARG_TYPE_INT, "state"); 204 spdk_trace_register_description("TCP_QP_DISCONNECT", TRACE_TCP_QP_DISCONNECT, 205 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 206 SPDK_TRACE_ARG_TYPE_INT, ""); 207 spdk_trace_register_description("TCP_QP_DESTROY", TRACE_TCP_QP_DESTROY, 208 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 209 SPDK_TRACE_ARG_TYPE_INT, ""); 210 spdk_trace_register_description("TCP_QP_ABORT_REQ", TRACE_TCP_QP_ABORT_REQ, 211 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 212 SPDK_TRACE_ARG_TYPE_INT, ""); 213 spdk_trace_register_description("TCP_QP_RCV_STATE_CHANGE", TRACE_TCP_QP_RCV_STATE_CHANGE, 214 OWNER_TYPE_NVMF_TCP, OBJECT_NONE, 0, 215 SPDK_TRACE_ARG_TYPE_INT, "state"); 216 217 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_START, OBJECT_NVMF_TCP_IO, 1); 218 spdk_trace_tpoint_register_relation(TRACE_BDEV_IO_DONE, OBJECT_NVMF_TCP_IO, 0); 219 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_QUEUE, OBJECT_NVMF_TCP_IO, 0); 220 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_PEND, OBJECT_NVMF_TCP_IO, 0); 221 spdk_trace_tpoint_register_relation(TRACE_SOCK_REQ_COMPLETE, OBJECT_NVMF_TCP_IO, 0); 222 } 223 224 struct spdk_nvmf_tcp_req { 225 struct spdk_nvmf_request req; 226 struct spdk_nvme_cpl rsp; 227 struct spdk_nvme_cmd cmd; 228 229 /* A PDU that can be used for sending responses. This is 230 * not the incoming PDU! */ 231 struct nvme_tcp_pdu *pdu; 232 233 /* In-capsule data buffer */ 234 uint8_t *buf; 235 236 struct spdk_nvmf_tcp_req *fused_pair; 237 238 /* 239 * The PDU for a request may be used multiple times in serial over 240 * the request's lifetime. For example, first to send an R2T, then 241 * to send a completion. To catch mistakes where the PDU is used 242 * twice at the same time, add a debug flag here for init/fini. 243 */ 244 bool pdu_in_use; 245 bool has_in_capsule_data; 246 bool fused_failed; 247 248 /* transfer_tag */ 249 uint16_t ttag; 250 251 enum spdk_nvmf_tcp_req_state state; 252 253 /* 254 * h2c_offset is used when we receive the h2c_data PDU. 255 */ 256 uint32_t h2c_offset; 257 258 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 259 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 260 STAILQ_ENTRY(spdk_nvmf_tcp_req) control_msg_link; 261 }; 262 263 struct spdk_nvmf_tcp_qpair { 264 struct spdk_nvmf_qpair qpair; 265 struct spdk_nvmf_tcp_poll_group *group; 266 struct spdk_sock *sock; 267 268 enum nvme_tcp_pdu_recv_state recv_state; 269 enum nvmf_tcp_qpair_state state; 270 271 /* PDU being actively received */ 272 struct nvme_tcp_pdu *pdu_in_progress; 273 274 struct spdk_nvmf_tcp_req *fused_first; 275 276 /* Queues to track the requests in all states */ 277 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 278 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 279 SLIST_HEAD(, nvme_tcp_pdu) tcp_pdu_free_queue; 280 /* Number of working pdus */ 281 uint32_t tcp_pdu_working_count; 282 283 /* Number of requests in each state */ 284 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 285 286 uint8_t cpda; 287 288 bool host_hdgst_enable; 289 bool host_ddgst_enable; 290 291 /* This is a spare PDU used for sending special management 292 * operations. Primarily, this is used for the initial 293 * connection response and c2h termination request. */ 294 struct nvme_tcp_pdu *mgmt_pdu; 295 296 /* Arrays of in-capsule buffers, requests, and pdus. 297 * Each array is 'resource_count' number of elements */ 298 void *bufs; 299 struct spdk_nvmf_tcp_req *reqs; 300 struct nvme_tcp_pdu *pdus; 301 uint32_t resource_count; 302 uint32_t recv_buf_size; 303 304 struct spdk_nvmf_tcp_port *port; 305 306 /* IP address */ 307 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 308 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 309 310 /* IP port */ 311 uint16_t initiator_port; 312 uint16_t target_port; 313 314 /* Wait until the host terminates the connection (e.g. after sending C2HTermReq) */ 315 bool wait_terminate; 316 317 /* Timer used to destroy qpair after detecting transport error issue if initiator does 318 * not close the connection. 319 */ 320 struct spdk_poller *timeout_poller; 321 322 spdk_nvmf_transport_qpair_fini_cb fini_cb_fn; 323 void *fini_cb_arg; 324 325 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 326 }; 327 328 struct spdk_nvmf_tcp_control_msg { 329 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 330 }; 331 332 struct spdk_nvmf_tcp_control_msg_list { 333 void *msg_buf; 334 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 335 STAILQ_HEAD(, spdk_nvmf_tcp_req) waiting_for_msg_reqs; 336 }; 337 338 struct spdk_nvmf_tcp_poll_group { 339 struct spdk_nvmf_transport_poll_group group; 340 struct spdk_sock_group *sock_group; 341 342 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 343 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 344 345 struct spdk_io_channel *accel_channel; 346 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 347 348 TAILQ_ENTRY(spdk_nvmf_tcp_poll_group) link; 349 }; 350 351 struct spdk_nvmf_tcp_port { 352 const struct spdk_nvme_transport_id *trid; 353 struct spdk_sock *listen_sock; 354 struct spdk_nvmf_transport *transport; 355 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 356 }; 357 358 struct tcp_transport_opts { 359 bool c2h_success; 360 uint16_t control_msg_num; 361 uint32_t sock_priority; 362 }; 363 364 struct tcp_psk_entry { 365 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 366 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 367 char pskid[NVMF_PSK_IDENTITY_LEN]; 368 uint8_t psk[SPDK_TLS_PSK_MAX_LEN]; 369 struct spdk_key *key; 370 uint32_t psk_size; 371 enum nvme_tcp_cipher_suite tls_cipher_suite; 372 TAILQ_ENTRY(tcp_psk_entry) link; 373 }; 374 375 struct spdk_nvmf_tcp_transport { 376 struct spdk_nvmf_transport transport; 377 struct tcp_transport_opts tcp_opts; 378 uint32_t ack_timeout; 379 380 struct spdk_nvmf_tcp_poll_group *next_pg; 381 382 struct spdk_poller *accept_poller; 383 struct spdk_sock_group *listen_sock_group; 384 385 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 386 TAILQ_HEAD(, spdk_nvmf_tcp_poll_group) poll_groups; 387 388 TAILQ_HEAD(, tcp_psk_entry) psks; 389 }; 390 391 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 392 { 393 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 394 spdk_json_decode_bool, true 395 }, 396 { 397 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 398 spdk_json_decode_uint16, true 399 }, 400 { 401 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 402 spdk_json_decode_uint32, true 403 }, 404 }; 405 406 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 407 struct spdk_nvmf_tcp_req *tcp_req); 408 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 409 410 static void _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 411 struct spdk_nvmf_tcp_req *tcp_req); 412 413 static inline void 414 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 415 enum spdk_nvmf_tcp_req_state state) 416 { 417 struct spdk_nvmf_qpair *qpair; 418 struct spdk_nvmf_tcp_qpair *tqpair; 419 420 qpair = tcp_req->req.qpair; 421 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 422 423 assert(tqpair->state_cntr[tcp_req->state] > 0); 424 tqpair->state_cntr[tcp_req->state]--; 425 tqpair->state_cntr[state]++; 426 427 tcp_req->state = state; 428 } 429 430 static inline struct nvme_tcp_pdu * 431 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 432 { 433 assert(tcp_req->pdu_in_use == false); 434 435 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 436 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 437 438 return tcp_req->pdu; 439 } 440 441 static struct spdk_nvmf_tcp_req * 442 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 443 { 444 struct spdk_nvmf_tcp_req *tcp_req; 445 446 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 447 if (spdk_unlikely(!tcp_req)) { 448 return NULL; 449 } 450 451 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 452 tcp_req->h2c_offset = 0; 453 tcp_req->has_in_capsule_data = false; 454 tcp_req->req.dif_enabled = false; 455 tcp_req->req.zcopy_phase = NVMF_ZCOPY_PHASE_NONE; 456 457 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 458 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 459 tqpair->qpair.queue_depth++; 460 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 461 return tcp_req; 462 } 463 464 static inline void 465 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 466 { 467 assert(!tcp_req->pdu_in_use); 468 469 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 470 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 471 tqpair->qpair.queue_depth--; 472 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 473 } 474 475 static void 476 nvmf_tcp_req_get_buffers_done(struct spdk_nvmf_request *req) 477 { 478 struct spdk_nvmf_tcp_req *tcp_req; 479 struct spdk_nvmf_transport *transport; 480 struct spdk_nvmf_tcp_transport *ttransport; 481 482 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 483 transport = req->qpair->transport; 484 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 485 486 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 487 nvmf_tcp_req_process(ttransport, tcp_req); 488 } 489 490 static void 491 nvmf_tcp_request_free(void *cb_arg) 492 { 493 struct spdk_nvmf_tcp_transport *ttransport; 494 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 495 496 assert(tcp_req != NULL); 497 498 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 499 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 500 struct spdk_nvmf_tcp_transport, transport); 501 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 502 nvmf_tcp_req_process(ttransport, tcp_req); 503 } 504 505 static int 506 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 507 { 508 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 509 510 nvmf_tcp_request_free(tcp_req); 511 512 return 0; 513 } 514 515 static void 516 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 517 enum spdk_nvmf_tcp_req_state state) 518 { 519 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 520 521 assert(state != TCP_REQUEST_STATE_FREE); 522 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 523 if (state == tcp_req->state) { 524 nvmf_tcp_request_free(tcp_req); 525 } 526 } 527 } 528 529 static inline void 530 nvmf_tcp_request_get_buffers_abort(struct spdk_nvmf_tcp_req *tcp_req) 531 { 532 /* Request can wait either for the iobuf or control_msg */ 533 struct spdk_nvmf_poll_group *group = tcp_req->req.qpair->group; 534 struct spdk_nvmf_transport *transport = tcp_req->req.qpair->transport; 535 struct spdk_nvmf_transport_poll_group *tgroup = nvmf_get_transport_poll_group(group, transport); 536 struct spdk_nvmf_tcp_poll_group *tcp_group = SPDK_CONTAINEROF(tgroup, 537 struct spdk_nvmf_tcp_poll_group, group); 538 struct spdk_nvmf_tcp_req *tmp_req, *abort_req; 539 540 assert(tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER); 541 542 STAILQ_FOREACH_SAFE(abort_req, &tcp_group->control_msg_list->waiting_for_msg_reqs, control_msg_link, 543 tmp_req) { 544 if (abort_req == tcp_req) { 545 STAILQ_REMOVE(&tcp_group->control_msg_list->waiting_for_msg_reqs, abort_req, spdk_nvmf_tcp_req, 546 control_msg_link); 547 return; 548 } 549 } 550 551 if (!nvmf_request_get_buffers_abort(&tcp_req->req)) { 552 SPDK_ERRLOG("Failed to abort tcp_req=%p\n", tcp_req); 553 assert(0 && "Should never happen"); 554 } 555 } 556 557 static void 558 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 559 { 560 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 561 562 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 563 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 564 565 /* Wipe the requests waiting for buffer from the waiting list */ 566 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 567 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 568 nvmf_tcp_request_get_buffers_abort(tcp_req); 569 } 570 } 571 572 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 573 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 574 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 575 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 576 } 577 578 static void 579 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 580 { 581 int i; 582 struct spdk_nvmf_tcp_req *tcp_req; 583 584 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 585 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 586 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 587 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 588 if ((int)tcp_req->state == i) { 589 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 590 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 591 } 592 } 593 } 594 } 595 596 static void 597 _nvmf_tcp_qpair_destroy(void *_tqpair) 598 { 599 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 600 spdk_nvmf_transport_qpair_fini_cb cb_fn = tqpair->fini_cb_fn; 601 void *cb_arg = tqpair->fini_cb_arg; 602 int err = 0; 603 604 spdk_trace_record(TRACE_TCP_QP_DESTROY, tqpair->qpair.trace_id, 0, 0); 605 606 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 607 608 err = spdk_sock_close(&tqpair->sock); 609 assert(err == 0); 610 nvmf_tcp_cleanup_all_states(tqpair); 611 612 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 613 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 614 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 615 tqpair->resource_count); 616 err++; 617 } 618 619 if (err > 0) { 620 nvmf_tcp_dump_qpair_req_contents(tqpair); 621 } 622 623 /* The timeout poller might still be registered here if we close the qpair before host 624 * terminates the connection. 625 */ 626 spdk_poller_unregister(&tqpair->timeout_poller); 627 spdk_dma_free(tqpair->pdus); 628 free(tqpair->reqs); 629 spdk_free(tqpair->bufs); 630 spdk_trace_unregister_owner(tqpair->qpair.trace_id); 631 free(tqpair); 632 633 if (cb_fn != NULL) { 634 cb_fn(cb_arg); 635 } 636 637 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 638 } 639 640 static void 641 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 642 { 643 /* Delay the destruction to make sure it isn't performed from the context of a sock 644 * callback. Otherwise, spdk_sock_close() might not abort pending requests, causing their 645 * completions to be executed after the qpair is freed. (Note: this fixed issue #2471.) 646 */ 647 spdk_thread_send_msg(spdk_get_thread(), _nvmf_tcp_qpair_destroy, tqpair); 648 } 649 650 static void 651 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 652 { 653 struct spdk_nvmf_tcp_transport *ttransport; 654 assert(w != NULL); 655 656 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 657 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 658 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 659 } 660 661 static void 662 nvmf_tcp_free_psk_entry(struct tcp_psk_entry *entry) 663 { 664 if (entry == NULL) { 665 return; 666 } 667 668 spdk_memset_s(entry->psk, sizeof(entry->psk), 0, sizeof(entry->psk)); 669 spdk_keyring_put_key(entry->key); 670 free(entry); 671 } 672 673 static int 674 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 675 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 676 { 677 struct spdk_nvmf_tcp_transport *ttransport; 678 struct tcp_psk_entry *entry, *tmp; 679 680 assert(transport != NULL); 681 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 682 683 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 684 TAILQ_REMOVE(&ttransport->psks, entry, link); 685 nvmf_tcp_free_psk_entry(entry); 686 } 687 688 spdk_poller_unregister(&ttransport->accept_poller); 689 spdk_sock_group_unregister_interrupt(ttransport->listen_sock_group); 690 spdk_sock_group_close(&ttransport->listen_sock_group); 691 free(ttransport); 692 693 if (cb_fn) { 694 cb_fn(cb_arg); 695 } 696 return 0; 697 } 698 699 static int nvmf_tcp_accept(void *ctx); 700 701 static void nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock); 702 703 static struct spdk_nvmf_transport * 704 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 705 { 706 struct spdk_nvmf_tcp_transport *ttransport; 707 uint32_t sge_count; 708 uint32_t min_shared_buffers; 709 int rc; 710 uint64_t period; 711 712 ttransport = calloc(1, sizeof(*ttransport)); 713 if (!ttransport) { 714 return NULL; 715 } 716 717 TAILQ_INIT(&ttransport->ports); 718 TAILQ_INIT(&ttransport->poll_groups); 719 TAILQ_INIT(&ttransport->psks); 720 721 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 722 723 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 724 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 725 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 726 if (opts->transport_specific != NULL && 727 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 728 SPDK_COUNTOF(tcp_transport_opts_decoder), 729 &ttransport->tcp_opts)) { 730 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 731 free(ttransport); 732 return NULL; 733 } 734 735 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 736 737 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 738 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 739 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 740 " in_capsule_data_size=%d, max_aq_depth=%d\n" 741 " num_shared_buffers=%d, c2h_success=%d,\n" 742 " dif_insert_or_strip=%d, sock_priority=%d\n" 743 " abort_timeout_sec=%d, control_msg_num=%hu\n" 744 " ack_timeout=%d\n", 745 opts->max_queue_depth, 746 opts->max_io_size, 747 opts->max_qpairs_per_ctrlr - 1, 748 opts->io_unit_size, 749 opts->in_capsule_data_size, 750 opts->max_aq_depth, 751 opts->num_shared_buffers, 752 ttransport->tcp_opts.c2h_success, 753 opts->dif_insert_or_strip, 754 ttransport->tcp_opts.sock_priority, 755 opts->abort_timeout_sec, 756 ttransport->tcp_opts.control_msg_num, 757 opts->ack_timeout); 758 759 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 760 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 761 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 762 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 763 free(ttransport); 764 return NULL; 765 } 766 767 if (ttransport->tcp_opts.control_msg_num == 0 && 768 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 769 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 770 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 771 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 772 } 773 774 /* I/O unit size cannot be larger than max I/O size */ 775 if (opts->io_unit_size > opts->max_io_size) { 776 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", 777 opts->io_unit_size, opts->max_io_size); 778 opts->io_unit_size = opts->max_io_size; 779 } 780 781 /* In capsule data size cannot be larger than max I/O size */ 782 if (opts->in_capsule_data_size > opts->max_io_size) { 783 SPDK_WARNLOG("TCP param ICD size %u can't be larger than max_io_size %u. Using max_io_size as ICD size\n", 784 opts->io_unit_size, opts->max_io_size); 785 opts->in_capsule_data_size = opts->max_io_size; 786 } 787 788 /* max IO queue depth cannot be smaller than 2 or larger than 65535. 789 * We will not check SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, because max_queue_depth is 16bits and always not larger than 64k. */ 790 if (opts->max_queue_depth < SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH) { 791 SPDK_WARNLOG("TCP param max_queue_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 792 opts->max_queue_depth, SPDK_NVMF_TCP_MIN_IO_QUEUE_DEPTH, 793 SPDK_NVMF_TCP_MAX_IO_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH); 794 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 795 } 796 797 /* max admin queue depth cannot be smaller than 2 or larger than 4096 */ 798 if (opts->max_aq_depth < SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH || 799 opts->max_aq_depth > SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH) { 800 SPDK_WARNLOG("TCP param max_aq_depth %u can't be smaller than %u or larger than %u. Using default value %u\n", 801 opts->max_aq_depth, SPDK_NVMF_TCP_MIN_ADMIN_QUEUE_DEPTH, 802 SPDK_NVMF_TCP_MAX_ADMIN_QUEUE_DEPTH, SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH); 803 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 804 } 805 806 sge_count = opts->max_io_size / opts->io_unit_size; 807 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 808 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 809 free(ttransport); 810 return NULL; 811 } 812 813 /* If buf_cache_size == UINT32_MAX, we will dynamically pick a cache size later that we know will fit. */ 814 if (opts->buf_cache_size < UINT32_MAX) { 815 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 816 if (min_shared_buffers > opts->num_shared_buffers) { 817 SPDK_ERRLOG("There are not enough buffers to satisfy " 818 "per-poll group caches for each thread. (%" PRIu32 ") " 819 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 820 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 821 free(ttransport); 822 return NULL; 823 } 824 } 825 826 period = spdk_interrupt_mode_is_enabled() ? 0 : opts->acceptor_poll_rate; 827 ttransport->accept_poller = SPDK_POLLER_REGISTER(nvmf_tcp_accept, &ttransport->transport, period); 828 if (!ttransport->accept_poller) { 829 free(ttransport); 830 return NULL; 831 } 832 833 spdk_poller_register_interrupt(ttransport->accept_poller, NULL, NULL); 834 835 ttransport->listen_sock_group = spdk_sock_group_create(NULL); 836 if (ttransport->listen_sock_group == NULL) { 837 SPDK_ERRLOG("Failed to create socket group for listen sockets\n"); 838 spdk_poller_unregister(&ttransport->accept_poller); 839 free(ttransport); 840 return NULL; 841 } 842 843 if (spdk_interrupt_mode_is_enabled()) { 844 rc = SPDK_SOCK_GROUP_REGISTER_INTERRUPT(ttransport->listen_sock_group, 845 SPDK_INTERRUPT_EVENT_IN | SPDK_INTERRUPT_EVENT_OUT, nvmf_tcp_accept, &ttransport->transport); 846 if (rc != 0) { 847 SPDK_ERRLOG("Failed to register interrupt for listen socker sock group\n"); 848 spdk_sock_group_close(&ttransport->listen_sock_group); 849 spdk_poller_unregister(&ttransport->accept_poller); 850 free(ttransport); 851 return NULL; 852 } 853 } 854 855 return &ttransport->transport; 856 } 857 858 static int 859 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 860 { 861 unsigned long long ull; 862 char *end = NULL; 863 864 ull = strtoull(trsvcid, &end, 10); 865 if (end == NULL || end == trsvcid || *end != '\0') { 866 return -1; 867 } 868 869 /* Valid TCP/IP port numbers are in [1, 65535] */ 870 if (ull == 0 || ull > 65535) { 871 return -1; 872 } 873 874 return (int)ull; 875 } 876 877 /** 878 * Canonicalize a listen address trid. 879 */ 880 static int 881 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 882 const struct spdk_nvme_transport_id *trid) 883 { 884 int trsvcid_int; 885 886 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 887 if (trsvcid_int < 0) { 888 return -EINVAL; 889 } 890 891 memset(canon_trid, 0, sizeof(*canon_trid)); 892 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 893 canon_trid->adrfam = trid->adrfam; 894 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 895 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 896 897 return 0; 898 } 899 900 /** 901 * Find an existing listening port. 902 */ 903 static struct spdk_nvmf_tcp_port * 904 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 905 const struct spdk_nvme_transport_id *trid) 906 { 907 struct spdk_nvme_transport_id canon_trid; 908 struct spdk_nvmf_tcp_port *port; 909 910 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 911 return NULL; 912 } 913 914 TAILQ_FOREACH(port, &ttransport->ports, link) { 915 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 916 return port; 917 } 918 } 919 920 return NULL; 921 } 922 923 static int 924 tcp_sock_get_key(uint8_t *out, int out_len, const char **cipher, const char *pskid, 925 void *get_key_ctx) 926 { 927 struct tcp_psk_entry *entry; 928 struct spdk_nvmf_tcp_transport *ttransport = get_key_ctx; 929 size_t psk_len; 930 int rc; 931 932 TAILQ_FOREACH(entry, &ttransport->psks, link) { 933 if (strcmp(pskid, entry->pskid) != 0) { 934 continue; 935 } 936 937 psk_len = entry->psk_size; 938 if ((size_t)out_len < psk_len) { 939 SPDK_ERRLOG("Out buffer of size: %" PRIu32 " cannot fit PSK of len: %lu\n", 940 out_len, psk_len); 941 return -ENOBUFS; 942 } 943 944 /* Convert PSK to the TLS PSK format. */ 945 rc = nvme_tcp_derive_tls_psk(entry->psk, psk_len, pskid, out, out_len, 946 entry->tls_cipher_suite); 947 if (rc < 0) { 948 SPDK_ERRLOG("Could not generate TLS PSK\n"); 949 } 950 951 switch (entry->tls_cipher_suite) { 952 case NVME_TCP_CIPHER_AES_128_GCM_SHA256: 953 *cipher = "TLS_AES_128_GCM_SHA256"; 954 break; 955 case NVME_TCP_CIPHER_AES_256_GCM_SHA384: 956 *cipher = "TLS_AES_256_GCM_SHA384"; 957 break; 958 default: 959 *cipher = NULL; 960 return -ENOTSUP; 961 } 962 963 return rc; 964 } 965 966 SPDK_ERRLOG("Could not find PSK for identity: %s\n", pskid); 967 968 return -EINVAL; 969 } 970 971 static int 972 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 973 struct spdk_nvmf_listen_opts *listen_opts) 974 { 975 struct spdk_nvmf_tcp_transport *ttransport; 976 struct spdk_nvmf_tcp_port *port; 977 int trsvcid_int; 978 uint8_t adrfam; 979 const char *sock_impl_name; 980 struct spdk_sock_impl_opts impl_opts; 981 size_t impl_opts_size = sizeof(impl_opts); 982 struct spdk_sock_opts opts; 983 int rc; 984 985 if (!strlen(trid->trsvcid)) { 986 SPDK_ERRLOG("Service id is required\n"); 987 return -EINVAL; 988 } 989 990 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 991 992 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 993 if (trsvcid_int < 0) { 994 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 995 return -EINVAL; 996 } 997 998 port = calloc(1, sizeof(*port)); 999 if (!port) { 1000 SPDK_ERRLOG("Port allocation failed\n"); 1001 return -ENOMEM; 1002 } 1003 1004 port->trid = trid; 1005 1006 sock_impl_name = NULL; 1007 1008 opts.opts_size = sizeof(opts); 1009 spdk_sock_get_default_opts(&opts); 1010 opts.priority = ttransport->tcp_opts.sock_priority; 1011 opts.ack_timeout = transport->opts.ack_timeout; 1012 if (listen_opts->secure_channel) { 1013 if (listen_opts->sock_impl && 1014 strncmp("ssl", listen_opts->sock_impl, strlen(listen_opts->sock_impl))) { 1015 SPDK_ERRLOG("Enabling secure_channel while specifying a sock_impl different from 'ssl' is unsupported"); 1016 free(port); 1017 return -EINVAL; 1018 } 1019 listen_opts->sock_impl = "ssl"; 1020 } 1021 1022 if (listen_opts->sock_impl) { 1023 sock_impl_name = listen_opts->sock_impl; 1024 spdk_sock_impl_get_opts(sock_impl_name, &impl_opts, &impl_opts_size); 1025 1026 if (!strncmp("ssl", sock_impl_name, strlen(sock_impl_name))) { 1027 if (!g_tls_log) { 1028 SPDK_NOTICELOG("TLS support is considered experimental\n"); 1029 g_tls_log = true; 1030 } 1031 impl_opts.tls_version = SPDK_TLS_VERSION_1_3; 1032 impl_opts.get_key = tcp_sock_get_key; 1033 impl_opts.get_key_ctx = ttransport; 1034 impl_opts.tls_cipher_suites = "TLS_AES_256_GCM_SHA384:TLS_AES_128_GCM_SHA256"; 1035 } 1036 1037 opts.impl_opts = &impl_opts; 1038 opts.impl_opts_size = sizeof(impl_opts); 1039 } 1040 1041 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 1042 sock_impl_name, &opts); 1043 if (port->listen_sock == NULL) { 1044 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 1045 trid->traddr, trsvcid_int, 1046 spdk_strerror(errno), errno); 1047 free(port); 1048 return -errno; 1049 } 1050 1051 if (spdk_sock_is_ipv4(port->listen_sock)) { 1052 adrfam = SPDK_NVMF_ADRFAM_IPV4; 1053 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 1054 adrfam = SPDK_NVMF_ADRFAM_IPV6; 1055 } else { 1056 SPDK_ERRLOG("Unhandled socket type\n"); 1057 adrfam = 0; 1058 } 1059 1060 if (adrfam != trid->adrfam) { 1061 SPDK_ERRLOG("Socket address family mismatch\n"); 1062 spdk_sock_close(&port->listen_sock); 1063 free(port); 1064 return -EINVAL; 1065 } 1066 1067 rc = spdk_sock_group_add_sock(ttransport->listen_sock_group, port->listen_sock, nvmf_tcp_accept_cb, 1068 port); 1069 if (rc < 0) { 1070 SPDK_ERRLOG("Failed to add socket to the listen socket group\n"); 1071 spdk_sock_close(&port->listen_sock); 1072 free(port); 1073 return -errno; 1074 } 1075 1076 port->transport = transport; 1077 1078 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 1079 trid->traddr, trid->trsvcid); 1080 1081 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 1082 return 0; 1083 } 1084 1085 static void 1086 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 1087 const struct spdk_nvme_transport_id *trid) 1088 { 1089 struct spdk_nvmf_tcp_transport *ttransport; 1090 struct spdk_nvmf_tcp_port *port; 1091 1092 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1093 1094 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 1095 trid->traddr, trid->trsvcid); 1096 1097 port = nvmf_tcp_find_port(ttransport, trid); 1098 if (port) { 1099 spdk_sock_group_remove_sock(ttransport->listen_sock_group, port->listen_sock); 1100 TAILQ_REMOVE(&ttransport->ports, port, link); 1101 spdk_sock_close(&port->listen_sock); 1102 free(port); 1103 } 1104 } 1105 1106 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1107 enum nvme_tcp_pdu_recv_state state); 1108 1109 static void 1110 nvmf_tcp_qpair_set_state(struct spdk_nvmf_tcp_qpair *tqpair, enum nvmf_tcp_qpair_state state) 1111 { 1112 tqpair->state = state; 1113 spdk_trace_record(TRACE_TCP_QP_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1114 (uint64_t)tqpair->state); 1115 } 1116 1117 static void 1118 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 1119 { 1120 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 1121 1122 spdk_trace_record(TRACE_TCP_QP_DISCONNECT, tqpair->qpair.trace_id, 0, 0); 1123 1124 if (tqpair->state <= NVMF_TCP_QPAIR_STATE_RUNNING) { 1125 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_EXITING); 1126 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1127 spdk_poller_unregister(&tqpair->timeout_poller); 1128 1129 /* This will end up calling nvmf_tcp_close_qpair */ 1130 spdk_nvmf_qpair_disconnect(&tqpair->qpair); 1131 } 1132 } 1133 1134 static void 1135 _mgmt_pdu_write_done(void *_tqpair, int err) 1136 { 1137 struct spdk_nvmf_tcp_qpair *tqpair = _tqpair; 1138 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1139 1140 if (spdk_unlikely(err != 0)) { 1141 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1142 return; 1143 } 1144 1145 assert(pdu->cb_fn != NULL); 1146 pdu->cb_fn(pdu->cb_arg); 1147 } 1148 1149 static void 1150 _req_pdu_write_done(void *req, int err) 1151 { 1152 struct spdk_nvmf_tcp_req *tcp_req = req; 1153 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1154 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1155 1156 assert(tcp_req->pdu_in_use); 1157 tcp_req->pdu_in_use = false; 1158 1159 /* If the request is in a completed state, we're waiting for write completion to free it */ 1160 if (spdk_unlikely(tcp_req->state == TCP_REQUEST_STATE_COMPLETED)) { 1161 nvmf_tcp_request_free(tcp_req); 1162 return; 1163 } 1164 1165 if (spdk_unlikely(err != 0)) { 1166 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1167 return; 1168 } 1169 1170 assert(pdu->cb_fn != NULL); 1171 pdu->cb_fn(pdu->cb_arg); 1172 } 1173 1174 static void 1175 _pdu_write_done(struct nvme_tcp_pdu *pdu, int err) 1176 { 1177 pdu->sock_req.cb_fn(pdu->sock_req.cb_arg, err); 1178 } 1179 1180 static void 1181 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 1182 { 1183 int rc; 1184 uint32_t mapped_length; 1185 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1186 1187 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 1188 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, &mapped_length); 1189 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 1190 1191 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 1192 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 1193 /* Try to force the send immediately. */ 1194 rc = spdk_sock_flush(tqpair->sock); 1195 if (rc > 0 && (uint32_t)rc == mapped_length) { 1196 _pdu_write_done(pdu, 0); 1197 } else { 1198 SPDK_ERRLOG("Could not write %s to socket: rc=%d, errno=%d\n", 1199 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP ? 1200 "IC_RESP" : "TERM_REQ", rc, errno); 1201 _pdu_write_done(pdu, rc >= 0 ? -EAGAIN : -errno); 1202 } 1203 } 1204 } 1205 1206 static void 1207 data_crc32_accel_done(void *cb_arg, int status) 1208 { 1209 struct nvme_tcp_pdu *pdu = cb_arg; 1210 1211 if (spdk_unlikely(status)) { 1212 SPDK_ERRLOG("Failed to compute the data digest for pdu =%p\n", pdu); 1213 _pdu_write_done(pdu, status); 1214 return; 1215 } 1216 1217 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 1218 MAKE_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32); 1219 1220 _tcp_write_pdu(pdu); 1221 } 1222 1223 static void 1224 pdu_data_crc32_compute(struct nvme_tcp_pdu *pdu) 1225 { 1226 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 1227 int rc = 0; 1228 1229 /* Data Digest */ 1230 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 1231 /* Only support this limitated case for the first step */ 1232 if (spdk_likely(!pdu->dif_ctx && (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0) 1233 && tqpair->group)) { 1234 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 1235 pdu->data_iovcnt, 0, data_crc32_accel_done, pdu); 1236 if (spdk_likely(rc == 0)) { 1237 return; 1238 } 1239 } else { 1240 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 1241 } 1242 data_crc32_accel_done(pdu, rc); 1243 } else { 1244 _tcp_write_pdu(pdu); 1245 } 1246 } 1247 1248 static void 1249 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1250 struct nvme_tcp_pdu *pdu, 1251 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1252 void *cb_arg) 1253 { 1254 int hlen; 1255 uint32_t crc32c; 1256 1257 assert(tqpair->pdu_in_progress != pdu); 1258 1259 hlen = pdu->hdr.common.hlen; 1260 pdu->cb_fn = cb_fn; 1261 pdu->cb_arg = cb_arg; 1262 1263 pdu->iov[0].iov_base = &pdu->hdr.raw; 1264 pdu->iov[0].iov_len = hlen; 1265 1266 /* Header Digest */ 1267 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 1268 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1269 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + hlen, crc32c); 1270 } 1271 1272 /* Data Digest */ 1273 pdu_data_crc32_compute(pdu); 1274 } 1275 1276 static void 1277 nvmf_tcp_qpair_write_mgmt_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1278 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1279 void *cb_arg) 1280 { 1281 struct nvme_tcp_pdu *pdu = tqpair->mgmt_pdu; 1282 1283 pdu->sock_req.cb_fn = _mgmt_pdu_write_done; 1284 pdu->sock_req.cb_arg = tqpair; 1285 1286 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1287 } 1288 1289 static void 1290 nvmf_tcp_qpair_write_req_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1291 struct spdk_nvmf_tcp_req *tcp_req, 1292 nvme_tcp_qpair_xfer_complete_cb cb_fn, 1293 void *cb_arg) 1294 { 1295 struct nvme_tcp_pdu *pdu = tcp_req->pdu; 1296 1297 pdu->sock_req.cb_fn = _req_pdu_write_done; 1298 pdu->sock_req.cb_arg = tcp_req; 1299 1300 assert(!tcp_req->pdu_in_use); 1301 tcp_req->pdu_in_use = true; 1302 1303 nvmf_tcp_qpair_write_pdu(tqpair, pdu, cb_fn, cb_arg); 1304 } 1305 1306 static int 1307 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 1308 { 1309 uint32_t i; 1310 struct spdk_nvmf_transport_opts *opts; 1311 uint32_t in_capsule_data_size; 1312 1313 opts = &tqpair->qpair.transport->opts; 1314 1315 in_capsule_data_size = opts->in_capsule_data_size; 1316 if (opts->dif_insert_or_strip) { 1317 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 1318 } 1319 1320 tqpair->resource_count = opts->max_queue_depth; 1321 1322 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 1323 if (!tqpair->reqs) { 1324 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 1325 return -1; 1326 } 1327 1328 if (in_capsule_data_size) { 1329 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 1330 NULL, SPDK_ENV_LCORE_ID_ANY, 1331 SPDK_MALLOC_DMA); 1332 if (!tqpair->bufs) { 1333 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 1334 return -1; 1335 } 1336 } 1337 /* prepare memory space for receiving pdus and tcp_req */ 1338 /* Add additional 1 member, which will be used for mgmt_pdu owned by the tqpair */ 1339 tqpair->pdus = spdk_dma_zmalloc((2 * tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, 1340 NULL); 1341 if (!tqpair->pdus) { 1342 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 1343 return -1; 1344 } 1345 1346 for (i = 0; i < tqpair->resource_count; i++) { 1347 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 1348 1349 tcp_req->ttag = i + 1; 1350 tcp_req->req.qpair = &tqpair->qpair; 1351 1352 tcp_req->pdu = &tqpair->pdus[i]; 1353 tcp_req->pdu->qpair = tqpair; 1354 1355 /* Set up memory to receive commands */ 1356 if (tqpair->bufs) { 1357 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 1358 } 1359 1360 /* Set the cmdn and rsp */ 1361 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 1362 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 1363 1364 tcp_req->req.stripped_data = NULL; 1365 1366 /* Initialize request state to FREE */ 1367 tcp_req->state = TCP_REQUEST_STATE_FREE; 1368 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 1369 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 1370 } 1371 1372 for (; i < 2 * tqpair->resource_count; i++) { 1373 struct nvme_tcp_pdu *pdu = &tqpair->pdus[i]; 1374 1375 pdu->qpair = tqpair; 1376 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 1377 } 1378 1379 tqpair->mgmt_pdu = &tqpair->pdus[i]; 1380 tqpair->mgmt_pdu->qpair = tqpair; 1381 tqpair->pdu_in_progress = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 1382 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 1383 tqpair->tcp_pdu_working_count = 1; 1384 1385 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 1386 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1387 1388 return 0; 1389 } 1390 1391 static int 1392 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 1393 { 1394 struct spdk_nvmf_tcp_qpair *tqpair; 1395 1396 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1397 1398 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 1399 1400 spdk_trace_record(TRACE_TCP_QP_CREATE, tqpair->qpair.trace_id, 0, 0); 1401 1402 /* Initialise request state queues of the qpair */ 1403 TAILQ_INIT(&tqpair->tcp_req_free_queue); 1404 TAILQ_INIT(&tqpair->tcp_req_working_queue); 1405 SLIST_INIT(&tqpair->tcp_pdu_free_queue); 1406 tqpair->qpair.queue_depth = 0; 1407 1408 tqpair->host_hdgst_enable = true; 1409 tqpair->host_ddgst_enable = true; 1410 1411 return 0; 1412 } 1413 1414 static int 1415 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 1416 { 1417 char saddr[32], caddr[32]; 1418 uint16_t sport, cport; 1419 char owner[256]; 1420 int rc; 1421 1422 rc = spdk_sock_getaddr(tqpair->sock, saddr, sizeof(saddr), &sport, 1423 caddr, sizeof(caddr), &cport); 1424 if (rc != 0) { 1425 SPDK_ERRLOG("spdk_sock_getaddr() failed\n"); 1426 return rc; 1427 } 1428 snprintf(owner, sizeof(owner), "%s:%d", caddr, cport); 1429 tqpair->qpair.trace_id = spdk_trace_register_owner(OWNER_TYPE_NVMF_TCP, owner); 1430 spdk_trace_record(TRACE_TCP_QP_SOCK_INIT, tqpair->qpair.trace_id, 0, 0); 1431 1432 /* set low water mark */ 1433 rc = spdk_sock_set_recvlowat(tqpair->sock, 1); 1434 if (rc != 0) { 1435 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 1436 return rc; 1437 } 1438 1439 return 0; 1440 } 1441 1442 static void 1443 nvmf_tcp_handle_connect(struct spdk_nvmf_tcp_port *port, struct spdk_sock *sock) 1444 { 1445 struct spdk_nvmf_tcp_qpair *tqpair; 1446 int rc; 1447 1448 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 1449 port->trid->traddr, port->trid->trsvcid); 1450 1451 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1452 if (tqpair == NULL) { 1453 SPDK_ERRLOG("Could not allocate new connection.\n"); 1454 spdk_sock_close(&sock); 1455 return; 1456 } 1457 1458 tqpair->sock = sock; 1459 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1460 tqpair->port = port; 1461 tqpair->qpair.transport = port->transport; 1462 tqpair->qpair.numa.id_valid = 1; 1463 tqpair->qpair.numa.id = spdk_sock_get_numa_id(sock); 1464 1465 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1466 sizeof(tqpair->target_addr), &tqpair->target_port, 1467 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1468 &tqpair->initiator_port); 1469 if (rc < 0) { 1470 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1471 nvmf_tcp_qpair_destroy(tqpair); 1472 return; 1473 } 1474 1475 spdk_nvmf_tgt_new_qpair(port->transport->tgt, &tqpair->qpair); 1476 } 1477 1478 static uint32_t 1479 nvmf_tcp_port_accept(struct spdk_nvmf_tcp_port *port) 1480 { 1481 struct spdk_sock *sock; 1482 uint32_t count = 0; 1483 int i; 1484 1485 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1486 sock = spdk_sock_accept(port->listen_sock); 1487 if (sock == NULL) { 1488 break; 1489 } 1490 count++; 1491 nvmf_tcp_handle_connect(port, sock); 1492 } 1493 1494 return count; 1495 } 1496 1497 static int 1498 nvmf_tcp_accept(void *ctx) 1499 { 1500 struct spdk_nvmf_transport *transport = ctx; 1501 struct spdk_nvmf_tcp_transport *ttransport; 1502 int count; 1503 1504 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1505 1506 count = spdk_sock_group_poll(ttransport->listen_sock_group); 1507 if (count < 0) { 1508 SPDK_ERRLOG("Fail in TCP listen socket group poll\n"); 1509 } 1510 1511 return count != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1512 } 1513 1514 static void 1515 nvmf_tcp_accept_cb(void *ctx, struct spdk_sock_group *group, struct spdk_sock *sock) 1516 { 1517 struct spdk_nvmf_tcp_port *port = ctx; 1518 1519 nvmf_tcp_port_accept(port); 1520 } 1521 1522 static void 1523 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1524 struct spdk_nvme_transport_id *trid, 1525 struct spdk_nvmf_discovery_log_page_entry *entry) 1526 { 1527 struct spdk_nvmf_tcp_port *port; 1528 struct spdk_nvmf_tcp_transport *ttransport; 1529 1530 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1531 entry->adrfam = trid->adrfam; 1532 1533 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1534 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1535 1536 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1537 port = nvmf_tcp_find_port(ttransport, trid); 1538 1539 assert(port != NULL); 1540 1541 if (strcmp(spdk_sock_get_impl_name(port->listen_sock), "ssl") == 0) { 1542 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_REQUIRED; 1543 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_TLS_1_3; 1544 } else { 1545 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1546 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1547 } 1548 } 1549 1550 static struct spdk_nvmf_tcp_control_msg_list * 1551 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1552 { 1553 struct spdk_nvmf_tcp_control_msg_list *list; 1554 struct spdk_nvmf_tcp_control_msg *msg; 1555 uint16_t i; 1556 1557 list = calloc(1, sizeof(*list)); 1558 if (!list) { 1559 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1560 return NULL; 1561 } 1562 1563 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1564 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA); 1565 if (!list->msg_buf) { 1566 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1567 free(list); 1568 return NULL; 1569 } 1570 1571 STAILQ_INIT(&list->free_msgs); 1572 STAILQ_INIT(&list->waiting_for_msg_reqs); 1573 1574 for (i = 0; i < num_messages; i++) { 1575 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1576 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1577 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1578 } 1579 1580 return list; 1581 } 1582 1583 static void 1584 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1585 { 1586 if (!list) { 1587 return; 1588 } 1589 1590 spdk_free(list->msg_buf); 1591 free(list); 1592 } 1593 1594 static int nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group); 1595 1596 static int 1597 nvmf_tcp_poll_group_intr(void *ctx) 1598 { 1599 struct spdk_nvmf_transport_poll_group *group = ctx; 1600 int ret = 0; 1601 1602 ret = nvmf_tcp_poll_group_poll(group); 1603 1604 return ret != 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 1605 } 1606 1607 static struct spdk_nvmf_transport_poll_group * 1608 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport, 1609 struct spdk_nvmf_poll_group *group) 1610 { 1611 struct spdk_nvmf_tcp_transport *ttransport; 1612 struct spdk_nvmf_tcp_poll_group *tgroup; 1613 int rc; 1614 1615 tgroup = calloc(1, sizeof(*tgroup)); 1616 if (!tgroup) { 1617 return NULL; 1618 } 1619 1620 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1621 if (!tgroup->sock_group) { 1622 goto cleanup; 1623 } 1624 1625 TAILQ_INIT(&tgroup->qpairs); 1626 TAILQ_INIT(&tgroup->await_req); 1627 1628 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1629 1630 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1631 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1632 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1633 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1634 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1635 if (!tgroup->control_msg_list) { 1636 goto cleanup; 1637 } 1638 } 1639 1640 tgroup->accel_channel = spdk_accel_get_io_channel(); 1641 if (spdk_unlikely(!tgroup->accel_channel)) { 1642 SPDK_ERRLOG("Cannot create accel_channel for tgroup=%p\n", tgroup); 1643 goto cleanup; 1644 } 1645 1646 TAILQ_INSERT_TAIL(&ttransport->poll_groups, tgroup, link); 1647 if (ttransport->next_pg == NULL) { 1648 ttransport->next_pg = tgroup; 1649 } 1650 1651 if (spdk_interrupt_mode_is_enabled()) { 1652 rc = SPDK_SOCK_GROUP_REGISTER_INTERRUPT(tgroup->sock_group, 1653 SPDK_INTERRUPT_EVENT_IN | SPDK_INTERRUPT_EVENT_OUT, nvmf_tcp_poll_group_intr, &tgroup->group); 1654 if (rc != 0) { 1655 SPDK_ERRLOG("Failed to register interrupt for sock group\n"); 1656 goto cleanup; 1657 } 1658 } 1659 1660 return &tgroup->group; 1661 1662 cleanup: 1663 nvmf_tcp_poll_group_destroy(&tgroup->group); 1664 return NULL; 1665 } 1666 1667 static struct spdk_nvmf_transport_poll_group * 1668 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1669 { 1670 struct spdk_nvmf_tcp_transport *ttransport; 1671 struct spdk_nvmf_tcp_poll_group **pg; 1672 struct spdk_nvmf_tcp_qpair *tqpair; 1673 struct spdk_sock_group *group = NULL, *hint = NULL; 1674 int rc; 1675 1676 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1677 1678 if (TAILQ_EMPTY(&ttransport->poll_groups)) { 1679 return NULL; 1680 } 1681 1682 pg = &ttransport->next_pg; 1683 assert(*pg != NULL); 1684 hint = (*pg)->sock_group; 1685 1686 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1687 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group, hint); 1688 if (rc != 0) { 1689 return NULL; 1690 } else if (group != NULL) { 1691 /* Optimal poll group was found */ 1692 return spdk_sock_group_get_ctx(group); 1693 } 1694 1695 /* The hint was used for optimal poll group, advance next_pg. */ 1696 *pg = TAILQ_NEXT(*pg, link); 1697 if (*pg == NULL) { 1698 *pg = TAILQ_FIRST(&ttransport->poll_groups); 1699 } 1700 1701 return spdk_sock_group_get_ctx(hint); 1702 } 1703 1704 static void 1705 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1706 { 1707 struct spdk_nvmf_tcp_poll_group *tgroup, *next_tgroup; 1708 struct spdk_nvmf_tcp_transport *ttransport; 1709 1710 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1711 spdk_sock_group_unregister_interrupt(tgroup->sock_group); 1712 spdk_sock_group_close(&tgroup->sock_group); 1713 if (tgroup->control_msg_list) { 1714 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1715 } 1716 1717 if (tgroup->accel_channel) { 1718 spdk_put_io_channel(tgroup->accel_channel); 1719 } 1720 1721 if (tgroup->group.transport == NULL) { 1722 /* Transport can be NULL when nvmf_tcp_poll_group_create() 1723 * calls this function directly in a failure path. */ 1724 free(tgroup); 1725 return; 1726 } 1727 1728 ttransport = SPDK_CONTAINEROF(tgroup->group.transport, struct spdk_nvmf_tcp_transport, transport); 1729 1730 next_tgroup = TAILQ_NEXT(tgroup, link); 1731 TAILQ_REMOVE(&ttransport->poll_groups, tgroup, link); 1732 if (next_tgroup == NULL) { 1733 next_tgroup = TAILQ_FIRST(&ttransport->poll_groups); 1734 } 1735 if (ttransport->next_pg == tgroup) { 1736 ttransport->next_pg = next_tgroup; 1737 } 1738 1739 free(tgroup); 1740 } 1741 1742 static void 1743 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1744 enum nvme_tcp_pdu_recv_state state) 1745 { 1746 if (tqpair->recv_state == state) { 1747 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1748 tqpair, state); 1749 return; 1750 } 1751 1752 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_QUIESCING)) { 1753 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH && tqpair->pdu_in_progress) { 1754 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, tqpair->pdu_in_progress, slist); 1755 tqpair->tcp_pdu_working_count--; 1756 } 1757 } 1758 1759 if (spdk_unlikely(state == NVME_TCP_PDU_RECV_STATE_ERROR)) { 1760 assert(tqpair->tcp_pdu_working_count == 0); 1761 } 1762 1763 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1764 /* When leaving the await req state, move the qpair to the main list */ 1765 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1766 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1767 } else if (state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1768 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1769 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1770 } 1771 1772 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1773 tqpair->recv_state = state; 1774 1775 spdk_trace_record(TRACE_TCP_QP_RCV_STATE_CHANGE, tqpair->qpair.trace_id, 0, 0, 1776 (uint64_t)tqpair->recv_state); 1777 } 1778 1779 static int 1780 nvmf_tcp_qpair_handle_timeout(void *ctx) 1781 { 1782 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1783 1784 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1785 1786 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1787 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1788 1789 nvmf_tcp_qpair_disconnect(tqpair); 1790 return SPDK_POLLER_BUSY; 1791 } 1792 1793 static void 1794 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1795 { 1796 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1797 1798 if (!tqpair->timeout_poller) { 1799 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1800 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1801 } 1802 } 1803 1804 static void 1805 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1806 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1807 { 1808 struct nvme_tcp_pdu *rsp_pdu; 1809 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1810 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1811 uint32_t copy_len; 1812 1813 rsp_pdu = tqpair->mgmt_pdu; 1814 1815 c2h_term_req = &rsp_pdu->hdr.term_req; 1816 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1817 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1818 c2h_term_req->fes = fes; 1819 1820 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1821 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1822 DSET32(&c2h_term_req->fei, error_offset); 1823 } 1824 1825 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1826 1827 /* Copy the error info into the buffer */ 1828 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1829 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1830 1831 /* Contain the header of the wrong received pdu */ 1832 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1833 tqpair->wait_terminate = true; 1834 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1835 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1836 } 1837 1838 static void 1839 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1840 struct spdk_nvmf_tcp_qpair *tqpair, 1841 struct nvme_tcp_pdu *pdu) 1842 { 1843 struct spdk_nvmf_tcp_req *tcp_req; 1844 1845 assert(pdu->psh_valid_bytes == pdu->psh_len); 1846 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1847 1848 tcp_req = nvmf_tcp_req_get(tqpair); 1849 if (!tcp_req) { 1850 /* Directly return and make the allocation retry again. This can happen if we're 1851 * using asynchronous writes to send the response to the host or when releasing 1852 * zero-copy buffers after a response has been sent. In both cases, the host might 1853 * receive the response before we've finished processing the request and is free to 1854 * send another one. 1855 */ 1856 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0 || 1857 tqpair->state_cntr[TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE] > 0) { 1858 return; 1859 } 1860 1861 /* The host sent more commands than the maximum queue depth. */ 1862 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1863 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 1864 return; 1865 } 1866 1867 pdu->req = tcp_req; 1868 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1869 nvmf_tcp_req_process(ttransport, tcp_req); 1870 } 1871 1872 static void 1873 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1874 struct spdk_nvmf_tcp_qpair *tqpair, 1875 struct nvme_tcp_pdu *pdu) 1876 { 1877 struct spdk_nvmf_tcp_req *tcp_req; 1878 struct spdk_nvme_tcp_cmd *capsule_cmd; 1879 uint32_t error_offset = 0; 1880 enum spdk_nvme_tcp_term_req_fes fes; 1881 struct spdk_nvme_cpl *rsp; 1882 1883 capsule_cmd = &pdu->hdr.capsule_cmd; 1884 tcp_req = pdu->req; 1885 assert(tcp_req != NULL); 1886 1887 /* Zero-copy requests don't support ICD */ 1888 assert(!spdk_nvmf_request_using_zcopy(&tcp_req->req)); 1889 1890 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1891 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1892 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1893 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1894 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1895 goto err; 1896 } 1897 1898 rsp = &tcp_req->req.rsp->nvme_cpl; 1899 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 1900 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 1901 } else { 1902 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1903 } 1904 1905 nvmf_tcp_req_process(ttransport, tcp_req); 1906 1907 return; 1908 err: 1909 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1910 } 1911 1912 static void 1913 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1914 struct spdk_nvmf_tcp_qpair *tqpair, 1915 struct nvme_tcp_pdu *pdu) 1916 { 1917 struct spdk_nvmf_tcp_req *tcp_req; 1918 uint32_t error_offset = 0; 1919 enum spdk_nvme_tcp_term_req_fes fes = 0; 1920 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1921 1922 h2c_data = &pdu->hdr.h2c_data; 1923 1924 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1925 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1926 1927 if (h2c_data->ttag > tqpair->resource_count) { 1928 SPDK_DEBUGLOG(nvmf_tcp, "ttag %u is larger than allowed %u.\n", h2c_data->ttag, 1929 tqpair->resource_count); 1930 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1931 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1932 goto err; 1933 } 1934 1935 tcp_req = &tqpair->reqs[h2c_data->ttag - 1]; 1936 1937 if (spdk_unlikely(tcp_req->state != TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER && 1938 tcp_req->state != TCP_REQUEST_STATE_AWAITING_R2T_ACK)) { 1939 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, has error state in %d\n", tcp_req, tqpair, 1940 tcp_req->state); 1941 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1942 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1943 goto err; 1944 } 1945 1946 if (spdk_unlikely(tcp_req->req.cmd->nvme_cmd.cid != h2c_data->cccid)) { 1947 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req(%p), tqpair=%p, expected %u but %u for cccid.\n", tcp_req, tqpair, 1948 tcp_req->req.cmd->nvme_cmd.cid, h2c_data->cccid); 1949 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1950 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1951 goto err; 1952 } 1953 1954 if (tcp_req->h2c_offset != h2c_data->datao) { 1955 SPDK_DEBUGLOG(nvmf_tcp, 1956 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1957 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1958 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1959 goto err; 1960 } 1961 1962 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1963 SPDK_DEBUGLOG(nvmf_tcp, 1964 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) exceeds requested length=%u\n", 1965 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1966 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1967 goto err; 1968 } 1969 1970 pdu->req = tcp_req; 1971 1972 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 1973 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1974 } 1975 1976 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1977 h2c_data->datao, h2c_data->datal); 1978 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1979 return; 1980 1981 err: 1982 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1983 } 1984 1985 static void 1986 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1987 struct spdk_nvmf_tcp_qpair *tqpair) 1988 { 1989 struct nvme_tcp_pdu *rsp_pdu; 1990 struct spdk_nvme_tcp_rsp *capsule_resp; 1991 1992 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1993 1994 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1995 assert(rsp_pdu != NULL); 1996 1997 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1998 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1999 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 2000 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 2001 if (tqpair->host_hdgst_enable) { 2002 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2003 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 2004 } 2005 2006 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_request_free, tcp_req); 2007 } 2008 2009 static void 2010 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 2011 { 2012 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 2013 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 2014 struct spdk_nvmf_tcp_qpair, qpair); 2015 2016 assert(tqpair != NULL); 2017 2018 if (spdk_unlikely(tcp_req->pdu->rw_offset < tcp_req->req.length)) { 2019 SPDK_DEBUGLOG(nvmf_tcp, "sending another C2H part, offset %u length %u\n", tcp_req->pdu->rw_offset, 2020 tcp_req->req.length); 2021 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2022 return; 2023 } 2024 2025 if (tcp_req->pdu->hdr.c2h_data.common.flags & SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS) { 2026 nvmf_tcp_request_free(tcp_req); 2027 } else { 2028 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2029 } 2030 } 2031 2032 static void 2033 nvmf_tcp_r2t_complete(void *cb_arg) 2034 { 2035 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 2036 struct spdk_nvmf_tcp_transport *ttransport; 2037 2038 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 2039 struct spdk_nvmf_tcp_transport, transport); 2040 2041 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2042 2043 if (tcp_req->h2c_offset == tcp_req->req.length) { 2044 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2045 nvmf_tcp_req_process(ttransport, tcp_req); 2046 } 2047 } 2048 2049 static void 2050 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 2051 struct spdk_nvmf_tcp_req *tcp_req) 2052 { 2053 struct nvme_tcp_pdu *rsp_pdu; 2054 struct spdk_nvme_tcp_r2t_hdr *r2t; 2055 2056 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2057 assert(rsp_pdu != NULL); 2058 2059 r2t = &rsp_pdu->hdr.r2t; 2060 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 2061 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 2062 2063 if (tqpair->host_hdgst_enable) { 2064 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2065 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 2066 } 2067 2068 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2069 r2t->ttag = tcp_req->ttag; 2070 r2t->r2to = tcp_req->h2c_offset; 2071 r2t->r2tl = tcp_req->req.length; 2072 2073 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 2074 2075 SPDK_DEBUGLOG(nvmf_tcp, 2076 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 2077 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 2078 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_r2t_complete, tcp_req); 2079 } 2080 2081 static void 2082 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 2083 struct spdk_nvmf_tcp_qpair *tqpair, 2084 struct nvme_tcp_pdu *pdu) 2085 { 2086 struct spdk_nvmf_tcp_req *tcp_req; 2087 struct spdk_nvme_cpl *rsp; 2088 2089 tcp_req = pdu->req; 2090 assert(tcp_req != NULL); 2091 2092 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2093 2094 tcp_req->h2c_offset += pdu->data_len; 2095 2096 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 2097 * acknowledged before moving on. */ 2098 if (tcp_req->h2c_offset == tcp_req->req.length && 2099 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 2100 /* After receiving all the h2c data, we need to check whether there is 2101 * transient transport error */ 2102 rsp = &tcp_req->req.rsp->nvme_cpl; 2103 if (spdk_unlikely(rsp->status.sc == SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR)) { 2104 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2105 } else { 2106 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2107 } 2108 nvmf_tcp_req_process(ttransport, tcp_req); 2109 } 2110 } 2111 2112 static void 2113 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 2114 { 2115 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 2116 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 2117 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 2118 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 2119 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 2120 DGET32(h2c_term_req->fei)); 2121 } 2122 } 2123 2124 static void 2125 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2126 struct nvme_tcp_pdu *pdu) 2127 { 2128 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2129 uint32_t error_offset = 0; 2130 enum spdk_nvme_tcp_term_req_fes fes; 2131 2132 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 2133 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 2134 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2135 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 2136 goto end; 2137 } 2138 2139 /* set the data buffer */ 2140 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 2141 h2c_term_req->common.plen - h2c_term_req->common.hlen); 2142 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2143 return; 2144 end: 2145 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2146 } 2147 2148 static void 2149 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2150 struct nvme_tcp_pdu *pdu) 2151 { 2152 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 2153 2154 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 2155 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2156 } 2157 2158 static void 2159 _nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2160 { 2161 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2162 struct spdk_nvmf_tcp_transport, transport); 2163 2164 switch (pdu->hdr.common.pdu_type) { 2165 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2166 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 2167 break; 2168 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2169 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 2170 break; 2171 2172 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2173 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 2174 break; 2175 2176 default: 2177 /* The code should not go to here */ 2178 SPDK_ERRLOG("ERROR pdu type %d\n", pdu->hdr.common.pdu_type); 2179 break; 2180 } 2181 SLIST_INSERT_HEAD(&tqpair->tcp_pdu_free_queue, pdu, slist); 2182 tqpair->tcp_pdu_working_count--; 2183 } 2184 2185 static inline void 2186 nvmf_tcp_req_set_cpl(struct spdk_nvmf_tcp_req *treq, int sct, int sc) 2187 { 2188 treq->req.rsp->nvme_cpl.status.sct = sct; 2189 treq->req.rsp->nvme_cpl.status.sc = sc; 2190 treq->req.rsp->nvme_cpl.cid = treq->req.cmd->nvme_cmd.cid; 2191 } 2192 2193 static void 2194 data_crc32_calc_done(void *cb_arg, int status) 2195 { 2196 struct nvme_tcp_pdu *pdu = cb_arg; 2197 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 2198 2199 /* async crc32 calculation is failed and use direct calculation to check */ 2200 if (spdk_unlikely(status)) { 2201 SPDK_ERRLOG("Data digest on tqpair=(%p) with pdu=%p failed to be calculated asynchronously\n", 2202 tqpair, pdu); 2203 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2204 } 2205 pdu->data_digest_crc32 ^= SPDK_CRC32C_XOR; 2206 if (!MATCH_DIGEST_WORD(pdu->data_digest, pdu->data_digest_crc32)) { 2207 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2208 assert(pdu->req != NULL); 2209 nvmf_tcp_req_set_cpl(pdu->req, SPDK_NVME_SCT_GENERIC, 2210 SPDK_NVME_SC_COMMAND_TRANSIENT_TRANSPORT_ERROR); 2211 } 2212 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2213 } 2214 2215 static void 2216 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu) 2217 { 2218 int rc = 0; 2219 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2220 tqpair->pdu_in_progress = NULL; 2221 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2222 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2223 /* check data digest if need */ 2224 if (pdu->ddgst_enable) { 2225 if (tqpair->qpair.qid != 0 && !pdu->dif_ctx && tqpair->group && 2226 (pdu->data_len % SPDK_NVME_TCP_DIGEST_ALIGNMENT == 0)) { 2227 rc = spdk_accel_submit_crc32cv(tqpair->group->accel_channel, &pdu->data_digest_crc32, pdu->data_iov, 2228 pdu->data_iovcnt, 0, data_crc32_calc_done, pdu); 2229 if (spdk_likely(rc == 0)) { 2230 return; 2231 } 2232 } else { 2233 pdu->data_digest_crc32 = nvme_tcp_pdu_calc_data_digest(pdu); 2234 } 2235 data_crc32_calc_done(pdu, rc); 2236 } else { 2237 _nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2238 } 2239 } 2240 2241 static void 2242 nvmf_tcp_send_icresp_complete(void *cb_arg) 2243 { 2244 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 2245 2246 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_RUNNING); 2247 } 2248 2249 static void 2250 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 2251 struct spdk_nvmf_tcp_qpair *tqpair, 2252 struct nvme_tcp_pdu *pdu) 2253 { 2254 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 2255 struct nvme_tcp_pdu *rsp_pdu; 2256 struct spdk_nvme_tcp_ic_resp *ic_resp; 2257 uint32_t error_offset = 0; 2258 enum spdk_nvme_tcp_term_req_fes fes; 2259 2260 /* Only PFV 0 is defined currently */ 2261 if (ic_req->pfv != 0) { 2262 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 2263 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2264 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 2265 goto end; 2266 } 2267 2268 /* This value is 0’s based value in units of dwords should not be larger than SPDK_NVME_TCP_HPDA_MAX */ 2269 if (ic_req->hpda > SPDK_NVME_TCP_HPDA_MAX) { 2270 SPDK_ERRLOG("ICReq HPDA out of range 0 to 31, got %u\n", ic_req->hpda); 2271 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2272 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, hpda); 2273 goto end; 2274 } 2275 2276 /* MAXR2T is 0's based */ 2277 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 2278 2279 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 2280 if (!tqpair->host_hdgst_enable) { 2281 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2282 } 2283 2284 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 2285 if (!tqpair->host_ddgst_enable) { 2286 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 2287 } 2288 2289 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 2290 /* Now that we know whether digests are enabled, properly size the receive buffer */ 2291 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 2292 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 2293 tqpair, 2294 tqpair->recv_buf_size); 2295 /* Not fatal. */ 2296 } 2297 2298 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 2299 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 2300 2301 rsp_pdu = tqpair->mgmt_pdu; 2302 2303 ic_resp = &rsp_pdu->hdr.ic_resp; 2304 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 2305 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 2306 ic_resp->pfv = 0; 2307 ic_resp->cpda = tqpair->cpda; 2308 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 2309 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 2310 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 2311 2312 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 2313 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 2314 2315 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_INITIALIZING); 2316 nvmf_tcp_qpair_write_mgmt_pdu(tqpair, nvmf_tcp_send_icresp_complete, tqpair); 2317 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2318 return; 2319 end: 2320 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2321 } 2322 2323 static void 2324 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 2325 struct spdk_nvmf_tcp_transport *ttransport) 2326 { 2327 struct nvme_tcp_pdu *pdu; 2328 int rc; 2329 uint32_t crc32c, error_offset = 0; 2330 enum spdk_nvme_tcp_term_req_fes fes; 2331 2332 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2333 pdu = tqpair->pdu_in_progress; 2334 2335 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 2336 pdu->hdr.common.pdu_type); 2337 /* check header digest if needed */ 2338 if (pdu->has_hdgst) { 2339 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 2340 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 2341 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 2342 if (rc == 0) { 2343 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 2344 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 2345 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2346 return; 2347 2348 } 2349 } 2350 2351 switch (pdu->hdr.common.pdu_type) { 2352 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 2353 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 2354 break; 2355 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2356 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 2357 break; 2358 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2359 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 2360 break; 2361 2362 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2363 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 2364 break; 2365 2366 default: 2367 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress->hdr.common.pdu_type); 2368 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2369 error_offset = 1; 2370 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2371 break; 2372 } 2373 } 2374 2375 static void 2376 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 2377 { 2378 struct nvme_tcp_pdu *pdu; 2379 uint32_t error_offset = 0; 2380 enum spdk_nvme_tcp_term_req_fes fes; 2381 uint8_t expected_hlen, pdo; 2382 bool plen_error = false, pdo_error = false; 2383 2384 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2385 pdu = tqpair->pdu_in_progress; 2386 assert(pdu); 2387 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 2388 if (tqpair->state != NVMF_TCP_QPAIR_STATE_INVALID) { 2389 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 2390 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2391 goto err; 2392 } 2393 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 2394 if (pdu->hdr.common.plen != expected_hlen) { 2395 plen_error = true; 2396 } 2397 } else { 2398 if (tqpair->state != NVMF_TCP_QPAIR_STATE_RUNNING) { 2399 SPDK_ERRLOG("The TCP/IP connection is not negotiated\n"); 2400 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 2401 goto err; 2402 } 2403 2404 switch (pdu->hdr.common.pdu_type) { 2405 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 2406 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 2407 pdo = pdu->hdr.common.pdo; 2408 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2409 pdo_error = true; 2410 break; 2411 } 2412 2413 if (pdu->hdr.common.plen < expected_hlen) { 2414 plen_error = true; 2415 } 2416 break; 2417 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 2418 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 2419 pdo = pdu->hdr.common.pdo; 2420 if ((tqpair->cpda != 0) && (pdo % ((tqpair->cpda + 1) << 2) != 0)) { 2421 pdo_error = true; 2422 break; 2423 } 2424 if (pdu->hdr.common.plen < expected_hlen) { 2425 plen_error = true; 2426 } 2427 break; 2428 2429 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 2430 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 2431 if ((pdu->hdr.common.plen <= expected_hlen) || 2432 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 2433 plen_error = true; 2434 } 2435 break; 2436 2437 default: 2438 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 2439 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2440 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 2441 goto err; 2442 } 2443 } 2444 2445 if (pdu->hdr.common.hlen != expected_hlen) { 2446 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 2447 pdu->hdr.common.pdu_type, 2448 expected_hlen, pdu->hdr.common.hlen, tqpair); 2449 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2450 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 2451 goto err; 2452 } else if (pdo_error) { 2453 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2454 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 2455 } else if (plen_error) { 2456 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2457 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2458 goto err; 2459 } else { 2460 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 2461 nvme_tcp_pdu_calc_psh_len(tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 2462 return; 2463 } 2464 err: 2465 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 2466 } 2467 2468 static int 2469 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 2470 { 2471 int rc = 0; 2472 struct nvme_tcp_pdu *pdu; 2473 enum nvme_tcp_pdu_recv_state prev_state; 2474 uint32_t data_len; 2475 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 2476 struct spdk_nvmf_tcp_transport, transport); 2477 2478 /* The loop here is to allow for several back-to-back state changes. */ 2479 do { 2480 prev_state = tqpair->recv_state; 2481 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 2482 2483 pdu = tqpair->pdu_in_progress; 2484 assert(pdu != NULL || 2485 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY || 2486 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_QUIESCING || 2487 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 2488 2489 switch (tqpair->recv_state) { 2490 /* Wait for the common header */ 2491 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 2492 if (!pdu) { 2493 pdu = SLIST_FIRST(&tqpair->tcp_pdu_free_queue); 2494 if (spdk_unlikely(!pdu)) { 2495 return NVME_TCP_PDU_IN_PROGRESS; 2496 } 2497 SLIST_REMOVE_HEAD(&tqpair->tcp_pdu_free_queue, slist); 2498 tqpair->pdu_in_progress = pdu; 2499 tqpair->tcp_pdu_working_count++; 2500 } 2501 memset(pdu, 0, offsetof(struct nvme_tcp_pdu, qpair)); 2502 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 2503 /* FALLTHROUGH */ 2504 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 2505 if (spdk_unlikely(tqpair->state == NVMF_TCP_QPAIR_STATE_INITIALIZING)) { 2506 return rc; 2507 } 2508 2509 rc = nvme_tcp_read_data(tqpair->sock, 2510 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 2511 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 2512 if (rc < 0) { 2513 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 2514 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2515 break; 2516 } else if (rc > 0) { 2517 pdu->ch_valid_bytes += rc; 2518 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2519 } 2520 2521 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 2522 return NVME_TCP_PDU_IN_PROGRESS; 2523 } 2524 2525 /* The command header of this PDU has now been read from the socket. */ 2526 nvmf_tcp_pdu_ch_handle(tqpair); 2527 break; 2528 /* Wait for the pdu specific header */ 2529 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 2530 rc = nvme_tcp_read_data(tqpair->sock, 2531 pdu->psh_len - pdu->psh_valid_bytes, 2532 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 2533 if (rc < 0) { 2534 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2535 break; 2536 } else if (rc > 0) { 2537 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, tqpair->qpair.trace_id, rc, 0); 2538 pdu->psh_valid_bytes += rc; 2539 } 2540 2541 if (pdu->psh_valid_bytes < pdu->psh_len) { 2542 return NVME_TCP_PDU_IN_PROGRESS; 2543 } 2544 2545 /* All header(ch, psh, head digits) of this PDU has now been read from the socket. */ 2546 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 2547 break; 2548 /* Wait for the req slot */ 2549 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 2550 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 2551 break; 2552 /* Wait for the request processing loop to acquire a buffer for the PDU */ 2553 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF: 2554 break; 2555 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 2556 /* check whether the data is valid, if not we just return */ 2557 if (!pdu->data_len) { 2558 return NVME_TCP_PDU_IN_PROGRESS; 2559 } 2560 2561 data_len = pdu->data_len; 2562 /* data digest */ 2563 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 2564 tqpair->host_ddgst_enable)) { 2565 data_len += SPDK_NVME_TCP_DIGEST_LEN; 2566 pdu->ddgst_enable = true; 2567 } 2568 2569 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 2570 if (rc < 0) { 2571 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2572 break; 2573 } 2574 pdu->rw_offset += rc; 2575 2576 if (pdu->rw_offset < data_len) { 2577 return NVME_TCP_PDU_IN_PROGRESS; 2578 } 2579 2580 /* Generate and insert DIF to whole data block received if DIF is enabled */ 2581 if (spdk_unlikely(pdu->dif_ctx != NULL) && 2582 spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 0, data_len, 2583 pdu->dif_ctx) != 0) { 2584 SPDK_ERRLOG("DIF generate failed\n"); 2585 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 2586 break; 2587 } 2588 2589 /* All of this PDU has now been read from the socket. */ 2590 nvmf_tcp_pdu_payload_handle(tqpair, pdu); 2591 break; 2592 case NVME_TCP_PDU_RECV_STATE_QUIESCING: 2593 if (tqpair->tcp_pdu_working_count != 0) { 2594 return NVME_TCP_PDU_IN_PROGRESS; 2595 } 2596 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2597 break; 2598 case NVME_TCP_PDU_RECV_STATE_ERROR: 2599 if (spdk_sock_is_connected(tqpair->sock) && tqpair->wait_terminate) { 2600 return NVME_TCP_PDU_IN_PROGRESS; 2601 } 2602 return NVME_TCP_PDU_FATAL; 2603 default: 2604 SPDK_ERRLOG("The state(%d) is invalid\n", tqpair->recv_state); 2605 abort(); 2606 break; 2607 } 2608 } while (tqpair->recv_state != prev_state); 2609 2610 return rc; 2611 } 2612 2613 static inline void * 2614 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list, 2615 struct spdk_nvmf_tcp_req *tcp_req) 2616 { 2617 struct spdk_nvmf_tcp_control_msg *msg; 2618 2619 assert(list); 2620 2621 msg = STAILQ_FIRST(&list->free_msgs); 2622 if (!msg) { 2623 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2624 STAILQ_INSERT_TAIL(&list->waiting_for_msg_reqs, tcp_req, control_msg_link); 2625 return NULL; 2626 } 2627 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2628 return msg; 2629 } 2630 2631 static inline void 2632 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2633 { 2634 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2635 struct spdk_nvmf_tcp_req *tcp_req; 2636 struct spdk_nvmf_tcp_transport *ttransport; 2637 2638 assert(list); 2639 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2640 if (!STAILQ_EMPTY(&list->waiting_for_msg_reqs)) { 2641 tcp_req = STAILQ_FIRST(&list->waiting_for_msg_reqs); 2642 STAILQ_REMOVE_HEAD(&list->waiting_for_msg_reqs, control_msg_link); 2643 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 2644 struct spdk_nvmf_tcp_transport, transport); 2645 nvmf_tcp_req_process(ttransport, tcp_req); 2646 } 2647 } 2648 2649 static void 2650 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2651 struct spdk_nvmf_transport *transport, 2652 struct spdk_nvmf_transport_poll_group *group) 2653 { 2654 struct spdk_nvmf_request *req = &tcp_req->req; 2655 struct spdk_nvme_cmd *cmd; 2656 struct spdk_nvme_sgl_descriptor *sgl; 2657 struct spdk_nvmf_tcp_poll_group *tgroup; 2658 enum spdk_nvme_tcp_term_req_fes fes; 2659 struct nvme_tcp_pdu *pdu; 2660 struct spdk_nvmf_tcp_qpair *tqpair; 2661 uint32_t length, error_offset = 0; 2662 2663 cmd = &req->cmd->nvme_cmd; 2664 sgl = &cmd->dptr.sgl1; 2665 2666 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2667 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2668 /* get request length from sgl */ 2669 length = sgl->unkeyed.length; 2670 if (spdk_unlikely(length > transport->opts.max_io_size)) { 2671 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2672 length, transport->opts.max_io_size); 2673 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2674 goto fatal_err; 2675 } 2676 2677 /* fill request length and populate iovs */ 2678 req->length = length; 2679 2680 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2681 2682 if (spdk_unlikely(req->dif_enabled)) { 2683 req->dif.orig_length = length; 2684 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2685 req->dif.elba_length = length; 2686 } 2687 2688 if (nvmf_ctrlr_use_zcopy(req)) { 2689 SPDK_DEBUGLOG(nvmf_tcp, "Using zero-copy to execute request %p\n", tcp_req); 2690 req->data_from_pool = false; 2691 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2692 return; 2693 } 2694 2695 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2696 /* No available buffers. Queue this request up. */ 2697 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2698 tcp_req); 2699 return; 2700 } 2701 2702 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2703 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2704 tcp_req, req->iovcnt, req->iov[0].iov_base); 2705 2706 return; 2707 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2708 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2709 uint64_t offset = sgl->address; 2710 uint32_t max_len = transport->opts.in_capsule_data_size; 2711 2712 assert(tcp_req->has_in_capsule_data); 2713 /* Capsule Cmd with In-capsule Data should get data length from pdu header */ 2714 tqpair = tcp_req->pdu->qpair; 2715 /* receiving pdu is not same with the pdu in tcp_req */ 2716 pdu = tqpair->pdu_in_progress; 2717 length = pdu->hdr.common.plen - pdu->psh_len - sizeof(struct spdk_nvme_tcp_common_pdu_hdr); 2718 if (tqpair->host_ddgst_enable) { 2719 length -= SPDK_NVME_TCP_DIGEST_LEN; 2720 } 2721 /* This error is not defined in NVMe/TCP spec, take this error as fatal error */ 2722 if (spdk_unlikely(length != sgl->unkeyed.length)) { 2723 SPDK_ERRLOG("In-Capsule Data length 0x%x is not equal to SGL data length 0x%x\n", 2724 length, sgl->unkeyed.length); 2725 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 2726 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 2727 goto fatal_err; 2728 } 2729 2730 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2731 offset, length); 2732 2733 /* The NVMe/TCP transport does not use ICDOFF to control the in-capsule data offset. ICDOFF should be '0' */ 2734 if (spdk_unlikely(offset != 0)) { 2735 /* Not defined fatal error in NVMe/TCP spec, handle this error as a fatal error */ 2736 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " should be ZERO in NVMe/TCP\n", offset); 2737 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2738 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.address); 2739 goto fatal_err; 2740 } 2741 2742 if (spdk_unlikely(length > max_len)) { 2743 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2744 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2745 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2746 2747 /* Get a buffer from dedicated list */ 2748 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2749 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2750 assert(tgroup->control_msg_list); 2751 req->iov[0].iov_base = nvmf_tcp_control_msg_get(tgroup->control_msg_list, tcp_req); 2752 if (!req->iov[0].iov_base) { 2753 /* No available buffers. Queue this request up. */ 2754 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2755 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_BUF); 2756 return; 2757 } 2758 } else { 2759 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2760 length, max_len); 2761 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_LIMIT_EXCEEDED; 2762 goto fatal_err; 2763 } 2764 } else { 2765 req->iov[0].iov_base = tcp_req->buf; 2766 } 2767 2768 req->length = length; 2769 req->data_from_pool = false; 2770 2771 if (spdk_unlikely(req->dif_enabled)) { 2772 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2773 req->dif.elba_length = length; 2774 } 2775 2776 req->iov[0].iov_len = length; 2777 req->iovcnt = 1; 2778 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_HAVE_BUFFER); 2779 2780 return; 2781 } 2782 /* If we want to handle the problem here, then we can't skip the following data segment. 2783 * Because this function runs before reading data part, now handle all errors as fatal errors. */ 2784 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2785 sgl->generic.type, sgl->generic.subtype); 2786 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 2787 error_offset = offsetof(struct spdk_nvme_tcp_cmd, ccsqe.dptr.sgl1.generic); 2788 fatal_err: 2789 nvmf_tcp_send_c2h_term_req(tcp_req->pdu->qpair, tcp_req->pdu, fes, error_offset); 2790 } 2791 2792 static inline enum spdk_nvme_media_error_status_code 2793 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2794 enum spdk_nvme_media_error_status_code result; 2795 2796 switch (err_type) 2797 { 2798 case SPDK_DIF_REFTAG_ERROR: 2799 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2800 break; 2801 case SPDK_DIF_APPTAG_ERROR: 2802 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2803 break; 2804 case SPDK_DIF_GUARD_ERROR: 2805 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2806 break; 2807 default: 2808 SPDK_UNREACHABLE(); 2809 break; 2810 } 2811 2812 return result; 2813 } 2814 2815 static void 2816 _nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2817 struct spdk_nvmf_tcp_req *tcp_req) 2818 { 2819 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2820 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2821 struct nvme_tcp_pdu *rsp_pdu; 2822 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2823 uint32_t plen, pdo, alignment; 2824 int rc; 2825 2826 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2827 2828 rsp_pdu = tcp_req->pdu; 2829 assert(rsp_pdu != NULL); 2830 2831 c2h_data = &rsp_pdu->hdr.c2h_data; 2832 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2833 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2834 2835 if (tqpair->host_hdgst_enable) { 2836 plen += SPDK_NVME_TCP_DIGEST_LEN; 2837 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2838 } 2839 2840 /* set the psh */ 2841 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2842 c2h_data->datal = tcp_req->req.length - tcp_req->pdu->rw_offset; 2843 c2h_data->datao = tcp_req->pdu->rw_offset; 2844 2845 /* set the padding */ 2846 rsp_pdu->padding_len = 0; 2847 pdo = plen; 2848 if (tqpair->cpda) { 2849 alignment = (tqpair->cpda + 1) << 2; 2850 if (plen % alignment != 0) { 2851 pdo = (plen + alignment) / alignment * alignment; 2852 rsp_pdu->padding_len = pdo - plen; 2853 plen = pdo; 2854 } 2855 } 2856 2857 c2h_data->common.pdo = pdo; 2858 plen += c2h_data->datal; 2859 if (tqpair->host_ddgst_enable) { 2860 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2861 plen += SPDK_NVME_TCP_DIGEST_LEN; 2862 } 2863 2864 c2h_data->common.plen = plen; 2865 2866 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2867 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2868 } 2869 2870 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2871 c2h_data->datao, c2h_data->datal); 2872 2873 2874 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2875 /* Need to send the capsule response if response is not all 0 */ 2876 if (ttransport->tcp_opts.c2h_success && 2877 tcp_req->rsp.cdw0 == 0 && tcp_req->rsp.cdw1 == 0) { 2878 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2879 } 2880 2881 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 2882 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2883 struct spdk_dif_error err_blk = {}; 2884 uint32_t mapped_length = 0; 2885 uint32_t available_iovs = SPDK_COUNTOF(rsp_pdu->iov); 2886 uint32_t ddgst_len = 0; 2887 2888 if (tqpair->host_ddgst_enable) { 2889 /* Data digest consumes additional iov entry */ 2890 available_iovs--; 2891 /* plen needs to be updated since nvme_tcp_build_iovs compares expected and actual plen */ 2892 ddgst_len = SPDK_NVME_TCP_DIGEST_LEN; 2893 c2h_data->common.plen -= ddgst_len; 2894 } 2895 /* Temp call to estimate if data can be described by limited number of iovs. 2896 * iov vector will be rebuilt in nvmf_tcp_qpair_write_pdu */ 2897 nvme_tcp_build_iovs(rsp_pdu->iov, available_iovs, rsp_pdu, tqpair->host_hdgst_enable, 2898 false, &mapped_length); 2899 2900 if (mapped_length != c2h_data->common.plen) { 2901 c2h_data->datal = mapped_length - (c2h_data->common.plen - c2h_data->datal); 2902 SPDK_DEBUGLOG(nvmf_tcp, 2903 "Part C2H, data_len %u (of %u), PDU len %u, updated PDU len %u, offset %u\n", 2904 c2h_data->datal, tcp_req->req.length, c2h_data->common.plen, mapped_length, rsp_pdu->rw_offset); 2905 c2h_data->common.plen = mapped_length; 2906 2907 /* Rebuild pdu->data_iov since data length is changed */ 2908 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, c2h_data->datao, 2909 c2h_data->datal); 2910 2911 c2h_data->common.flags &= ~(SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU | 2912 SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS); 2913 } 2914 2915 c2h_data->common.plen += ddgst_len; 2916 2917 assert(rsp_pdu->rw_offset <= tcp_req->req.length); 2918 2919 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2920 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2921 if (rc != 0) { 2922 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2923 err_blk.err_type, err_blk.err_offset); 2924 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2925 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2926 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2927 return; 2928 } 2929 } 2930 2931 rsp_pdu->rw_offset += c2h_data->datal; 2932 nvmf_tcp_qpair_write_req_pdu(tqpair, tcp_req, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2933 } 2934 2935 static void 2936 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2937 struct spdk_nvmf_tcp_req *tcp_req) 2938 { 2939 nvmf_tcp_req_pdu_init(tcp_req); 2940 _nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2941 } 2942 2943 static int 2944 request_transfer_out(struct spdk_nvmf_request *req) 2945 { 2946 struct spdk_nvmf_tcp_req *tcp_req; 2947 struct spdk_nvmf_qpair *qpair; 2948 struct spdk_nvmf_tcp_qpair *tqpair; 2949 struct spdk_nvme_cpl *rsp; 2950 2951 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2952 2953 qpair = req->qpair; 2954 rsp = &req->rsp->nvme_cpl; 2955 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2956 2957 /* Advance our sq_head pointer */ 2958 if (qpair->sq_head == qpair->sq_head_max) { 2959 qpair->sq_head = 0; 2960 } else { 2961 qpair->sq_head++; 2962 } 2963 rsp->sqhd = qpair->sq_head; 2964 2965 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2966 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2967 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2968 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2969 } else { 2970 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2971 } 2972 2973 return 0; 2974 } 2975 2976 static void 2977 nvmf_tcp_check_fused_ordering(struct spdk_nvmf_tcp_transport *ttransport, 2978 struct spdk_nvmf_tcp_qpair *tqpair, 2979 struct spdk_nvmf_tcp_req *tcp_req) 2980 { 2981 enum spdk_nvme_cmd_fuse last, next; 2982 2983 last = tqpair->fused_first ? tqpair->fused_first->cmd.fuse : SPDK_NVME_CMD_FUSE_NONE; 2984 next = tcp_req->cmd.fuse; 2985 2986 assert(last != SPDK_NVME_CMD_FUSE_SECOND); 2987 2988 if (spdk_likely(last == SPDK_NVME_CMD_FUSE_NONE && next == SPDK_NVME_CMD_FUSE_NONE)) { 2989 return; 2990 } 2991 2992 if (last == SPDK_NVME_CMD_FUSE_FIRST) { 2993 if (next == SPDK_NVME_CMD_FUSE_SECOND) { 2994 /* This is a valid pair of fused commands. Point them at each other 2995 * so they can be submitted consecutively once ready to be executed. 2996 */ 2997 tqpair->fused_first->fused_pair = tcp_req; 2998 tcp_req->fused_pair = tqpair->fused_first; 2999 tqpair->fused_first = NULL; 3000 return; 3001 } else { 3002 /* Mark the last req as failed since it wasn't followed by a SECOND. */ 3003 tqpair->fused_first->fused_failed = true; 3004 3005 /* 3006 * If the last req is in READY_TO_EXECUTE state, then call 3007 * nvmf_tcp_req_process(), otherwise nothing else will kick it. 3008 */ 3009 if (tqpair->fused_first->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3010 nvmf_tcp_req_process(ttransport, tqpair->fused_first); 3011 } 3012 3013 tqpair->fused_first = NULL; 3014 } 3015 } 3016 3017 if (next == SPDK_NVME_CMD_FUSE_FIRST) { 3018 /* Set tqpair->fused_first here so that we know to check that the next request 3019 * is a SECOND (and to fail this one if it isn't). 3020 */ 3021 tqpair->fused_first = tcp_req; 3022 } else if (next == SPDK_NVME_CMD_FUSE_SECOND) { 3023 /* Mark this req failed since it is a SECOND and the last one was not a FIRST. */ 3024 tcp_req->fused_failed = true; 3025 } 3026 } 3027 3028 static bool 3029 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 3030 struct spdk_nvmf_tcp_req *tcp_req) 3031 { 3032 struct spdk_nvmf_tcp_qpair *tqpair; 3033 uint32_t plen; 3034 struct nvme_tcp_pdu *pdu; 3035 enum spdk_nvmf_tcp_req_state prev_state; 3036 bool progress = false; 3037 struct spdk_nvmf_transport *transport = &ttransport->transport; 3038 struct spdk_nvmf_transport_poll_group *group; 3039 struct spdk_nvmf_tcp_poll_group *tgroup; 3040 3041 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 3042 group = &tqpair->group->group; 3043 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 3044 3045 /* If the qpair is not active, we need to abort the outstanding requests. */ 3046 if (!spdk_nvmf_qpair_is_active(&tqpair->qpair)) { 3047 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 3048 nvmf_tcp_request_get_buffers_abort(tcp_req); 3049 } 3050 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3051 } 3052 3053 /* The loop here is to allow for several back-to-back state changes. */ 3054 do { 3055 prev_state = tcp_req->state; 3056 3057 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 3058 tqpair); 3059 3060 switch (tcp_req->state) { 3061 case TCP_REQUEST_STATE_FREE: 3062 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 3063 * to escape this state. */ 3064 break; 3065 case TCP_REQUEST_STATE_NEW: 3066 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3067 tqpair->qpair.queue_depth); 3068 3069 /* copy the cmd from the receive pdu */ 3070 tcp_req->cmd = tqpair->pdu_in_progress->hdr.capsule_cmd.ccsqe; 3071 3072 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 3073 tcp_req->req.dif_enabled = true; 3074 tqpair->pdu_in_progress->dif_ctx = &tcp_req->req.dif.dif_ctx; 3075 } 3076 3077 nvmf_tcp_check_fused_ordering(ttransport, tqpair, tcp_req); 3078 3079 /* The next state transition depends on the data transfer needs of this request. */ 3080 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 3081 3082 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 3083 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_INVALID_OPCODE); 3084 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3085 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3086 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 3087 break; 3088 } 3089 3090 /* If no data to transfer, ready to execute. */ 3091 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 3092 /* Reset the tqpair receiving pdu state */ 3093 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3094 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3095 break; 3096 } 3097 3098 pdu = tqpair->pdu_in_progress; 3099 plen = pdu->hdr.common.hlen; 3100 if (tqpair->host_hdgst_enable) { 3101 plen += SPDK_NVME_TCP_DIGEST_LEN; 3102 } 3103 if (pdu->hdr.common.plen != plen) { 3104 tcp_req->has_in_capsule_data = true; 3105 } else { 3106 /* Data is transmitted by C2H PDUs */ 3107 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 3108 } 3109 3110 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 3111 break; 3112 case TCP_REQUEST_STATE_NEED_BUFFER: 3113 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, tqpair->qpair.trace_id, 0, 3114 (uintptr_t)tcp_req); 3115 3116 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 3117 3118 /* Try to get a data buffer */ 3119 nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 3120 break; 3121 case TCP_REQUEST_STATE_HAVE_BUFFER: 3122 spdk_trace_record(TRACE_TCP_REQUEST_STATE_HAVE_BUFFER, tqpair->qpair.trace_id, 0, 3123 (uintptr_t)tcp_req); 3124 /* Get a zcopy buffer if the request can be serviced through zcopy */ 3125 if (spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3126 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3127 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3128 tcp_req->req.length = tcp_req->req.dif.elba_length; 3129 } 3130 3131 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_START); 3132 spdk_nvmf_request_zcopy_start(&tcp_req->req); 3133 break; 3134 } 3135 3136 assert(tcp_req->req.iovcnt > 0); 3137 3138 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 3139 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3140 if (tcp_req->req.data_from_pool) { 3141 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3142 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3143 } else { 3144 struct nvme_tcp_pdu *pdu; 3145 3146 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 3147 3148 pdu = tqpair->pdu_in_progress; 3149 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 3150 tqpair); 3151 /* No need to send r2t, contained in the capsuled data */ 3152 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 3153 0, tcp_req->req.length); 3154 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 3155 } 3156 break; 3157 } 3158 3159 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 3160 break; 3161 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3162 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_START, tqpair->qpair.trace_id, 0, 3163 (uintptr_t)tcp_req); 3164 /* Some external code must kick a request into TCP_REQUEST_STATE_ZCOPY_START_COMPLETED 3165 * to escape this state. */ 3166 break; 3167 case TCP_REQUEST_STATE_ZCOPY_START_COMPLETED: 3168 spdk_trace_record(TRACE_TCP_REQUEST_STATE_ZCOPY_START_COMPLETED, tqpair->qpair.trace_id, 0, 3169 (uintptr_t)tcp_req); 3170 if (spdk_unlikely(spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl))) { 3171 SPDK_DEBUGLOG(nvmf_tcp, "Zero-copy start failed for tcp_req(%p) on tqpair=%p\n", 3172 tcp_req, tqpair); 3173 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3174 break; 3175 } 3176 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 3177 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 3178 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 3179 } else { 3180 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3181 } 3182 break; 3183 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3184 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, tqpair->qpair.trace_id, 0, 3185 (uintptr_t)tcp_req); 3186 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 3187 break; 3188 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3189 3190 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, tqpair->qpair.trace_id, 3191 0, (uintptr_t)tcp_req); 3192 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 3193 * to escape this state. */ 3194 break; 3195 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 3196 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, tqpair->qpair.trace_id, 0, 3197 (uintptr_t)tcp_req); 3198 3199 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3200 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 3201 tcp_req->req.length = tcp_req->req.dif.elba_length; 3202 } 3203 3204 if (tcp_req->cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 3205 if (tcp_req->fused_failed) { 3206 /* This request failed FUSED semantics. Fail it immediately, without 3207 * even sending it to the target layer. 3208 */ 3209 nvmf_tcp_req_set_cpl(tcp_req, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_MISSING_FUSED); 3210 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3211 break; 3212 } 3213 3214 if (tcp_req->fused_pair == NULL || 3215 tcp_req->fused_pair->state != TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3216 /* This request is ready to execute, but either we don't know yet if it's 3217 * valid - i.e. this is a FIRST but we haven't received the next request yet), 3218 * or the other request of this fused pair isn't ready to execute. So 3219 * break here and this request will get processed later either when the 3220 * other request is ready or we find that this request isn't valid. 3221 */ 3222 break; 3223 } 3224 } 3225 3226 if (!spdk_nvmf_request_using_zcopy(&tcp_req->req)) { 3227 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 3228 /* If we get to this point, and this request is a fused command, we know that 3229 * it is part of a valid sequence (FIRST followed by a SECOND) and that both 3230 * requests are READY_TO_EXECUTE. So call spdk_nvmf_request_exec() both on this 3231 * request, and the other request of the fused pair, in the correct order. 3232 * Also clear the ->fused_pair pointers on both requests, since after this point 3233 * we no longer need to maintain the relationship between these two requests. 3234 */ 3235 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_SECOND) { 3236 assert(tcp_req->fused_pair != NULL); 3237 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3238 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3239 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3240 tcp_req->fused_pair->fused_pair = NULL; 3241 tcp_req->fused_pair = NULL; 3242 } 3243 spdk_nvmf_request_exec(&tcp_req->req); 3244 if (tcp_req->cmd.fuse == SPDK_NVME_CMD_FUSE_FIRST) { 3245 assert(tcp_req->fused_pair != NULL); 3246 assert(tcp_req->fused_pair->fused_pair == tcp_req); 3247 nvmf_tcp_req_set_state(tcp_req->fused_pair, TCP_REQUEST_STATE_EXECUTING); 3248 spdk_nvmf_request_exec(&tcp_req->fused_pair->req); 3249 tcp_req->fused_pair->fused_pair = NULL; 3250 tcp_req->fused_pair = NULL; 3251 } 3252 } else { 3253 /* For zero-copy, only requests with data coming from host to the 3254 * controller can end up here. */ 3255 assert(tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER); 3256 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT); 3257 spdk_nvmf_request_zcopy_end(&tcp_req->req, true); 3258 } 3259 3260 break; 3261 case TCP_REQUEST_STATE_EXECUTING: 3262 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3263 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3264 * to escape this state. */ 3265 break; 3266 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3267 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_COMMIT, tqpair->qpair.trace_id, 0, 3268 (uintptr_t)tcp_req); 3269 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 3270 * to escape this state. */ 3271 break; 3272 case TCP_REQUEST_STATE_EXECUTED: 3273 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req); 3274 3275 if (spdk_unlikely(tcp_req->req.dif_enabled)) { 3276 tcp_req->req.length = tcp_req->req.dif.orig_length; 3277 } 3278 3279 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3280 break; 3281 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 3282 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, tqpair->qpair.trace_id, 0, 3283 (uintptr_t)tcp_req); 3284 if (request_transfer_out(&tcp_req->req) != 0) { 3285 assert(0); /* No good way to handle this currently */ 3286 } 3287 break; 3288 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 3289 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, tqpair->qpair.trace_id, 3290 0, (uintptr_t)tcp_req); 3291 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3292 * to escape this state. */ 3293 break; 3294 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3295 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_ZCOPY_RELEASE, tqpair->qpair.trace_id, 0, 3296 (uintptr_t)tcp_req); 3297 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 3298 * to escape this state. */ 3299 break; 3300 case TCP_REQUEST_STATE_COMPLETED: 3301 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, tqpair->qpair.trace_id, 0, (uintptr_t)tcp_req, 3302 tqpair->qpair.queue_depth); 3303 /* If there's an outstanding PDU sent to the host, the request is completed 3304 * due to the qpair being disconnected. We must delay the completion until 3305 * that write is done to avoid freeing the request twice. */ 3306 if (spdk_unlikely(tcp_req->pdu_in_use)) { 3307 SPDK_DEBUGLOG(nvmf_tcp, "Delaying completion due to outstanding " 3308 "write on req=%p\n", tcp_req); 3309 /* This can only happen for zcopy requests */ 3310 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3311 assert(!spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3312 break; 3313 } 3314 3315 if (tcp_req->req.data_from_pool) { 3316 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 3317 } else if (spdk_unlikely(tcp_req->has_in_capsule_data && 3318 (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 3319 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 3320 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3321 assert(tgroup->control_msg_list); 3322 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 3323 nvmf_tcp_control_msg_put(tgroup->control_msg_list, 3324 tcp_req->req.iov[0].iov_base); 3325 } else if (tcp_req->req.zcopy_bdev_io != NULL) { 3326 /* If the request has an unreleased zcopy bdev_io, it's either a 3327 * read, a failed write, or the qpair is being disconnected */ 3328 assert(spdk_nvmf_request_using_zcopy(&tcp_req->req)); 3329 assert(tcp_req->req.xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST || 3330 spdk_nvme_cpl_is_error(&tcp_req->req.rsp->nvme_cpl) || 3331 !spdk_nvmf_qpair_is_active(&tqpair->qpair)); 3332 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE); 3333 spdk_nvmf_request_zcopy_end(&tcp_req->req, false); 3334 break; 3335 } 3336 tcp_req->req.length = 0; 3337 tcp_req->req.iovcnt = 0; 3338 tcp_req->fused_failed = false; 3339 if (tcp_req->fused_pair) { 3340 /* This req was part of a valid fused pair, but failed before it got to 3341 * READ_TO_EXECUTE state. This means we need to fail the other request 3342 * in the pair, because it is no longer part of a valid pair. If the pair 3343 * already reached READY_TO_EXECUTE state, we need to kick it. 3344 */ 3345 tcp_req->fused_pair->fused_failed = true; 3346 if (tcp_req->fused_pair->state == TCP_REQUEST_STATE_READY_TO_EXECUTE) { 3347 nvmf_tcp_req_process(ttransport, tcp_req->fused_pair); 3348 } 3349 tcp_req->fused_pair = NULL; 3350 } 3351 3352 nvmf_tcp_req_put(tqpair, tcp_req); 3353 break; 3354 case TCP_REQUEST_NUM_STATES: 3355 default: 3356 assert(0); 3357 break; 3358 } 3359 3360 if (tcp_req->state != prev_state) { 3361 progress = true; 3362 } 3363 } while (tcp_req->state != prev_state); 3364 3365 return progress; 3366 } 3367 3368 static void 3369 tcp_sock_cb(void *arg) 3370 { 3371 struct spdk_nvmf_tcp_qpair *tqpair = arg; 3372 int rc; 3373 3374 assert(tqpair != NULL); 3375 rc = nvmf_tcp_sock_process(tqpair); 3376 3377 /* If there was a new socket error, disconnect */ 3378 if (rc < 0) { 3379 nvmf_tcp_qpair_disconnect(tqpair); 3380 } 3381 } 3382 3383 static void 3384 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 3385 { 3386 tcp_sock_cb(arg); 3387 } 3388 3389 static int 3390 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 3391 struct spdk_nvmf_qpair *qpair) 3392 { 3393 struct spdk_nvmf_tcp_poll_group *tgroup; 3394 struct spdk_nvmf_tcp_qpair *tqpair; 3395 int rc; 3396 3397 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3398 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3399 3400 rc = nvmf_tcp_qpair_sock_init(tqpair); 3401 if (rc != 0) { 3402 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 3403 return -1; 3404 } 3405 3406 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 3407 if (rc < 0) { 3408 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 3409 return -1; 3410 } 3411 3412 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 3413 if (rc < 0) { 3414 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 3415 return -1; 3416 } 3417 3418 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 3419 nvmf_tcp_sock_cb, tqpair); 3420 if (rc != 0) { 3421 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 3422 spdk_strerror(errno), errno); 3423 return -1; 3424 } 3425 3426 tqpair->group = tgroup; 3427 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_INVALID); 3428 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 3429 3430 return 0; 3431 } 3432 3433 static int 3434 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 3435 struct spdk_nvmf_qpair *qpair) 3436 { 3437 struct spdk_nvmf_tcp_poll_group *tgroup; 3438 struct spdk_nvmf_tcp_qpair *tqpair; 3439 int rc; 3440 3441 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3442 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3443 3444 assert(tqpair->group == tgroup); 3445 3446 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 3447 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 3448 /* Change the state to move the qpair from the await_req list to the main list 3449 * and prevent adding it again later by nvmf_tcp_qpair_set_recv_state() */ 3450 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_QUIESCING); 3451 } 3452 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 3453 3454 /* Try to force out any pending writes */ 3455 spdk_sock_flush(tqpair->sock); 3456 3457 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 3458 if (rc != 0) { 3459 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 3460 spdk_strerror(errno), errno); 3461 } 3462 3463 return rc; 3464 } 3465 3466 static int 3467 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 3468 { 3469 struct spdk_nvmf_tcp_transport *ttransport; 3470 struct spdk_nvmf_tcp_req *tcp_req; 3471 struct spdk_nvmf_tcp_qpair *tqpair; 3472 3473 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3474 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 3475 tqpair = SPDK_CONTAINEROF(req->qpair, struct spdk_nvmf_tcp_qpair, qpair); 3476 3477 switch (tcp_req->state) { 3478 case TCP_REQUEST_STATE_EXECUTING: 3479 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3480 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 3481 break; 3482 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3483 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_ZCOPY_START_COMPLETED); 3484 break; 3485 case TCP_REQUEST_STATE_AWAITING_ZCOPY_RELEASE: 3486 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 3487 /* In the interrupt mode it's possible that all responses are already written out over 3488 * the socket but zero-copy buffers are still not released. In that case there won't be 3489 * any event to trigger further socket processing. Send msg to a thread to avoid deadlock. 3490 */ 3491 if (spdk_unlikely(spdk_interrupt_mode_is_enabled() && 3492 tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ && 3493 spdk_nvmf_qpair_is_active(&tqpair->qpair))) { 3494 spdk_thread_send_msg(spdk_get_thread(), tcp_sock_cb, tqpair); 3495 } 3496 break; 3497 default: 3498 SPDK_ERRLOG("Unexpected request state %d (cntlid:%d, qid:%d)\n", 3499 tcp_req->state, req->qpair->ctrlr->cntlid, req->qpair->qid); 3500 assert(0 && "Unexpected request state"); 3501 break; 3502 } 3503 3504 nvmf_tcp_req_process(ttransport, tcp_req); 3505 3506 return 0; 3507 } 3508 3509 static void 3510 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 3511 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 3512 { 3513 struct spdk_nvmf_tcp_qpair *tqpair; 3514 3515 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 3516 3517 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3518 3519 assert(tqpair->fini_cb_fn == NULL); 3520 tqpair->fini_cb_fn = cb_fn; 3521 tqpair->fini_cb_arg = cb_arg; 3522 3523 nvmf_tcp_qpair_set_state(tqpair, NVMF_TCP_QPAIR_STATE_EXITED); 3524 nvmf_tcp_qpair_destroy(tqpair); 3525 } 3526 3527 static int 3528 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 3529 { 3530 struct spdk_nvmf_tcp_poll_group *tgroup; 3531 int num_events, rc = 0, rc2; 3532 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 3533 3534 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 3535 3536 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 3537 return 0; 3538 } 3539 3540 num_events = spdk_sock_group_poll(tgroup->sock_group); 3541 if (spdk_unlikely(num_events < 0)) { 3542 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 3543 } 3544 3545 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 3546 rc2 = nvmf_tcp_sock_process(tqpair); 3547 3548 /* If there was a new socket error, disconnect */ 3549 if (spdk_unlikely(rc2 < 0)) { 3550 nvmf_tcp_qpair_disconnect(tqpair); 3551 if (rc == 0) { 3552 rc = rc2; 3553 } 3554 } 3555 } 3556 3557 return rc == 0 ? num_events : rc; 3558 } 3559 3560 static int 3561 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 3562 struct spdk_nvme_transport_id *trid, bool peer) 3563 { 3564 struct spdk_nvmf_tcp_qpair *tqpair; 3565 uint16_t port; 3566 3567 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3568 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 3569 3570 if (peer) { 3571 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 3572 port = tqpair->initiator_port; 3573 } else { 3574 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 3575 port = tqpair->target_port; 3576 } 3577 3578 if (spdk_sock_is_ipv4(tqpair->sock)) { 3579 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 3580 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 3581 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 3582 } else { 3583 return -1; 3584 } 3585 3586 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 3587 return 0; 3588 } 3589 3590 static int 3591 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 3592 struct spdk_nvme_transport_id *trid) 3593 { 3594 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3595 } 3596 3597 static int 3598 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 3599 struct spdk_nvme_transport_id *trid) 3600 { 3601 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 3602 } 3603 3604 static int 3605 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 3606 struct spdk_nvme_transport_id *trid) 3607 { 3608 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 3609 } 3610 3611 static void 3612 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 3613 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 3614 { 3615 nvmf_tcp_req_set_cpl(tcp_req_to_abort, SPDK_NVME_SCT_GENERIC, SPDK_NVME_SC_ABORTED_BY_REQUEST); 3616 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 3617 3618 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 3619 } 3620 3621 static int 3622 _nvmf_tcp_qpair_abort_request(void *ctx) 3623 { 3624 struct spdk_nvmf_request *req = ctx; 3625 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 3626 struct spdk_nvmf_tcp_req, req); 3627 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 3628 struct spdk_nvmf_tcp_qpair, qpair); 3629 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 3630 struct spdk_nvmf_tcp_transport, transport); 3631 int rc; 3632 3633 spdk_poller_unregister(&req->poller); 3634 3635 switch (tcp_req_to_abort->state) { 3636 case TCP_REQUEST_STATE_EXECUTING: 3637 case TCP_REQUEST_STATE_AWAITING_ZCOPY_START: 3638 case TCP_REQUEST_STATE_AWAITING_ZCOPY_COMMIT: 3639 rc = nvmf_ctrlr_abort_request(req); 3640 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 3641 return SPDK_POLLER_BUSY; 3642 } 3643 break; 3644 3645 case TCP_REQUEST_STATE_NEED_BUFFER: 3646 nvmf_tcp_request_get_buffers_abort(tcp_req_to_abort); 3647 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 3648 nvmf_tcp_req_process(ttransport, tcp_req_to_abort); 3649 break; 3650 3651 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 3652 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 3653 if (spdk_get_ticks() < req->timeout_tsc) { 3654 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 3655 return SPDK_POLLER_BUSY; 3656 } 3657 break; 3658 3659 default: 3660 /* Requests in other states are either un-abortable (e.g. 3661 * TRANSFERRING_CONTROLLER_TO_HOST) or should never end up here, as they're 3662 * immediately transitioned to other states in nvmf_tcp_req_process() (e.g. 3663 * READY_TO_EXECUTE). But it is fine to end up here, as we'll simply complete the 3664 * abort request with the bit0 of dword0 set (command not aborted). 3665 */ 3666 break; 3667 } 3668 3669 spdk_nvmf_request_complete(req); 3670 return SPDK_POLLER_BUSY; 3671 } 3672 3673 static void 3674 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 3675 struct spdk_nvmf_request *req) 3676 { 3677 struct spdk_nvmf_tcp_qpair *tqpair; 3678 struct spdk_nvmf_tcp_transport *ttransport; 3679 struct spdk_nvmf_transport *transport; 3680 uint16_t cid; 3681 uint32_t i; 3682 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 3683 3684 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 3685 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 3686 transport = &ttransport->transport; 3687 3688 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3689 3690 for (i = 0; i < tqpair->resource_count; i++) { 3691 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 3692 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 3693 tcp_req_to_abort = &tqpair->reqs[i]; 3694 break; 3695 } 3696 } 3697 3698 spdk_trace_record(TRACE_TCP_QP_ABORT_REQ, tqpair->qpair.trace_id, 0, (uintptr_t)req); 3699 3700 if (tcp_req_to_abort == NULL) { 3701 spdk_nvmf_request_complete(req); 3702 return; 3703 } 3704 3705 req->req_to_abort = &tcp_req_to_abort->req; 3706 req->timeout_tsc = spdk_get_ticks() + 3707 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 3708 req->poller = NULL; 3709 3710 _nvmf_tcp_qpair_abort_request(req); 3711 } 3712 3713 struct tcp_subsystem_add_host_opts { 3714 char *psk; 3715 }; 3716 3717 static const struct spdk_json_object_decoder tcp_subsystem_add_host_opts_decoder[] = { 3718 {"psk", offsetof(struct tcp_subsystem_add_host_opts, psk), spdk_json_decode_string, true}, 3719 }; 3720 3721 static int 3722 nvmf_tcp_subsystem_add_host(struct spdk_nvmf_transport *transport, 3723 const struct spdk_nvmf_subsystem *subsystem, 3724 const char *hostnqn, 3725 const struct spdk_json_val *transport_specific) 3726 { 3727 struct tcp_subsystem_add_host_opts opts; 3728 struct spdk_nvmf_tcp_transport *ttransport; 3729 struct tcp_psk_entry *tmp, *entry = NULL; 3730 uint8_t psk_configured[SPDK_TLS_PSK_MAX_LEN] = {}; 3731 char psk_interchange[SPDK_TLS_PSK_MAX_LEN + 1] = {}; 3732 uint8_t tls_cipher_suite; 3733 int rc = 0; 3734 uint8_t psk_retained_hash; 3735 uint64_t psk_configured_size; 3736 3737 if (transport_specific == NULL) { 3738 return 0; 3739 } 3740 3741 assert(transport != NULL); 3742 assert(subsystem != NULL); 3743 3744 memset(&opts, 0, sizeof(opts)); 3745 3746 /* Decode PSK (either name of a key or file path) */ 3747 if (spdk_json_decode_object_relaxed(transport_specific, tcp_subsystem_add_host_opts_decoder, 3748 SPDK_COUNTOF(tcp_subsystem_add_host_opts_decoder), &opts)) { 3749 SPDK_ERRLOG("spdk_json_decode_object failed\n"); 3750 return -EINVAL; 3751 } 3752 3753 if (opts.psk == NULL) { 3754 return 0; 3755 } 3756 3757 entry = calloc(1, sizeof(struct tcp_psk_entry)); 3758 if (entry == NULL) { 3759 SPDK_ERRLOG("Unable to allocate memory for PSK entry!\n"); 3760 rc = -ENOMEM; 3761 goto end; 3762 } 3763 3764 entry->key = spdk_keyring_get_key(opts.psk); 3765 if (entry->key == NULL) { 3766 SPDK_ERRLOG("Key '%s' does not exist\n", opts.psk); 3767 rc = -EINVAL; 3768 goto end; 3769 } 3770 3771 rc = spdk_key_get_key(entry->key, psk_interchange, SPDK_TLS_PSK_MAX_LEN); 3772 if (rc < 0) { 3773 SPDK_ERRLOG("Failed to retrieve PSK '%s'\n", opts.psk); 3774 rc = -EINVAL; 3775 goto end; 3776 } 3777 3778 /* Parse PSK interchange to get length of base64 encoded data. 3779 * This is then used to decide which cipher suite should be used 3780 * to generate PSK identity and TLS PSK later on. */ 3781 rc = nvme_tcp_parse_interchange_psk(psk_interchange, psk_configured, sizeof(psk_configured), 3782 &psk_configured_size, &psk_retained_hash); 3783 if (rc < 0) { 3784 SPDK_ERRLOG("Failed to parse PSK interchange!\n"); 3785 goto end; 3786 } 3787 3788 /* The Base64 string encodes the configured PSK (32 or 48 bytes binary). 3789 * This check also ensures that psk_configured_size is smaller than 3790 * psk_retained buffer size. */ 3791 if (psk_configured_size == SHA256_DIGEST_LENGTH) { 3792 tls_cipher_suite = NVME_TCP_CIPHER_AES_128_GCM_SHA256; 3793 } else if (psk_configured_size == SHA384_DIGEST_LENGTH) { 3794 tls_cipher_suite = NVME_TCP_CIPHER_AES_256_GCM_SHA384; 3795 } else { 3796 SPDK_ERRLOG("Unrecognized cipher suite!\n"); 3797 rc = -EINVAL; 3798 goto end; 3799 } 3800 3801 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3802 /* Generate PSK identity. */ 3803 rc = nvme_tcp_generate_psk_identity(entry->pskid, sizeof(entry->pskid), hostnqn, 3804 subsystem->subnqn, tls_cipher_suite); 3805 if (rc) { 3806 rc = -EINVAL; 3807 goto end; 3808 } 3809 /* Check if PSK identity entry already exists. */ 3810 TAILQ_FOREACH(tmp, &ttransport->psks, link) { 3811 if (strncmp(tmp->pskid, entry->pskid, NVMF_PSK_IDENTITY_LEN) == 0) { 3812 SPDK_ERRLOG("Given PSK identity: %s entry already exists!\n", entry->pskid); 3813 rc = -EEXIST; 3814 goto end; 3815 } 3816 } 3817 3818 if (snprintf(entry->hostnqn, sizeof(entry->hostnqn), "%s", hostnqn) < 0) { 3819 SPDK_ERRLOG("Could not write hostnqn string!\n"); 3820 rc = -EINVAL; 3821 goto end; 3822 } 3823 if (snprintf(entry->subnqn, sizeof(entry->subnqn), "%s", subsystem->subnqn) < 0) { 3824 SPDK_ERRLOG("Could not write subnqn string!\n"); 3825 rc = -EINVAL; 3826 goto end; 3827 } 3828 3829 entry->tls_cipher_suite = tls_cipher_suite; 3830 3831 /* No hash indicates that Configured PSK must be used as Retained PSK. */ 3832 if (psk_retained_hash == NVME_TCP_HASH_ALGORITHM_NONE) { 3833 /* Psk configured is either 32 or 48 bytes long. */ 3834 memcpy(entry->psk, psk_configured, psk_configured_size); 3835 entry->psk_size = psk_configured_size; 3836 } else { 3837 /* Derive retained PSK. */ 3838 rc = nvme_tcp_derive_retained_psk(psk_configured, psk_configured_size, hostnqn, entry->psk, 3839 SPDK_TLS_PSK_MAX_LEN, psk_retained_hash); 3840 if (rc < 0) { 3841 SPDK_ERRLOG("Unable to derive retained PSK!\n"); 3842 goto end; 3843 } 3844 entry->psk_size = rc; 3845 } 3846 3847 TAILQ_INSERT_TAIL(&ttransport->psks, entry, link); 3848 rc = 0; 3849 3850 end: 3851 spdk_memset_s(psk_configured, sizeof(psk_configured), 0, sizeof(psk_configured)); 3852 spdk_memset_s(psk_interchange, sizeof(psk_interchange), 0, sizeof(psk_interchange)); 3853 3854 free(opts.psk); 3855 if (rc != 0) { 3856 nvmf_tcp_free_psk_entry(entry); 3857 } 3858 3859 return rc; 3860 } 3861 3862 static void 3863 nvmf_tcp_subsystem_remove_host(struct spdk_nvmf_transport *transport, 3864 const struct spdk_nvmf_subsystem *subsystem, 3865 const char *hostnqn) 3866 { 3867 struct spdk_nvmf_tcp_transport *ttransport; 3868 struct tcp_psk_entry *entry, *tmp; 3869 3870 assert(transport != NULL); 3871 assert(subsystem != NULL); 3872 3873 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3874 TAILQ_FOREACH_SAFE(entry, &ttransport->psks, link, tmp) { 3875 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3876 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3877 TAILQ_REMOVE(&ttransport->psks, entry, link); 3878 nvmf_tcp_free_psk_entry(entry); 3879 break; 3880 } 3881 } 3882 } 3883 3884 static void 3885 nvmf_tcp_subsystem_dump_host(struct spdk_nvmf_transport *transport, 3886 const struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 3887 struct spdk_json_write_ctx *w) 3888 { 3889 struct spdk_nvmf_tcp_transport *ttransport; 3890 struct tcp_psk_entry *entry; 3891 3892 assert(transport != NULL); 3893 assert(subsystem != NULL); 3894 3895 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 3896 TAILQ_FOREACH(entry, &ttransport->psks, link) { 3897 if ((strncmp(entry->hostnqn, hostnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0 && 3898 (strncmp(entry->subnqn, subsystem->subnqn, SPDK_NVMF_NQN_MAX_LEN)) == 0) { 3899 spdk_json_write_named_string(w, "psk", spdk_key_get_name(entry->key)); 3900 break; 3901 } 3902 } 3903 } 3904 3905 static void 3906 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 3907 { 3908 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_IO_QUEUE_DEPTH; 3909 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 3910 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 3911 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 3912 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 3913 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_MAX_ADMIN_QUEUE_DEPTH; 3914 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 3915 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 3916 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 3917 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 3918 opts->transport_specific = NULL; 3919 } 3920 3921 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 3922 .name = "TCP", 3923 .type = SPDK_NVME_TRANSPORT_TCP, 3924 .opts_init = nvmf_tcp_opts_init, 3925 .create = nvmf_tcp_create, 3926 .dump_opts = nvmf_tcp_dump_opts, 3927 .destroy = nvmf_tcp_destroy, 3928 3929 .listen = nvmf_tcp_listen, 3930 .stop_listen = nvmf_tcp_stop_listen, 3931 3932 .listener_discover = nvmf_tcp_discover, 3933 3934 .poll_group_create = nvmf_tcp_poll_group_create, 3935 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 3936 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 3937 .poll_group_add = nvmf_tcp_poll_group_add, 3938 .poll_group_remove = nvmf_tcp_poll_group_remove, 3939 .poll_group_poll = nvmf_tcp_poll_group_poll, 3940 3941 .req_free = nvmf_tcp_req_free, 3942 .req_complete = nvmf_tcp_req_complete, 3943 .req_get_buffers_done = nvmf_tcp_req_get_buffers_done, 3944 3945 .qpair_fini = nvmf_tcp_close_qpair, 3946 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 3947 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 3948 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 3949 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 3950 .subsystem_add_host = nvmf_tcp_subsystem_add_host, 3951 .subsystem_remove_host = nvmf_tcp_subsystem_remove_host, 3952 .subsystem_dump_host = nvmf_tcp_subsystem_dump_host, 3953 }; 3954 3955 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 3956 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 3957