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